diff --git a/AUTHORS b/AUTHORS
new file mode 100644
index 0000000000..540821528b
--- /dev/null
+++ b/AUTHORS
@@ -0,0 +1 @@
+See README.txt
diff --git a/COPYING b/COPYING
new file mode 100644
index 0000000000..805d7df96b
--- /dev/null
+++ b/COPYING
@@ -0,0 +1 @@
+See LICENSE.txt
diff --git a/ChangeLog b/ChangeLog
new file mode 100644
index 0000000000..540821528b
--- /dev/null
+++ b/ChangeLog
@@ -0,0 +1 @@
+See README.txt
diff --git a/INSTALL b/INSTALL
new file mode 100644
index 0000000000..2ee1151e63
--- /dev/null
+++ b/INSTALL
@@ -0,0 +1,9 @@
+
+It is advised to run ./autogen.sh before./configure (autoconf and automake
+need to be installed on your system for autogen.sh to work)
+
+./configure has an option named --with-cuda that allows you to specify
+where your CUDA 5.5 toolkit is installed (usually /usr/local/cuda-5.5,
+but some distros may have a different default location)
+
+See README.txt
diff --git a/LICENSE b/LICENSE
index 70566f2d0e..2d7f3b98d8 100644
--- a/LICENSE
+++ b/LICENSE
@@ -1,674 +1,3 @@
-GNU GENERAL PUBLIC LICENSE
-                       Version 3, 29 June 2007
-
- Copyright (C) 2007 Free Software Foundation, Inc. <http://fsf.org/>
- Everyone is permitted to copy and distribute verbatim copies
- of this license document, but changing it is not allowed.
-
-                            Preamble
-
-  The GNU General Public License is a free, copyleft license for
-software and other kinds of works.
-
-  The licenses for most software and other practical works are designed
-to take away your freedom to share and change the works.  By contrast,
-the GNU General Public License is intended to guarantee your freedom to
-share and change all versions of a program--to make sure it remains free
-software for all its users.  We, the Free Software Foundation, use the
-GNU General Public License for most of our software; it applies also to
-any other work released this way by its authors.  You can apply it to
-your programs, too.
-
-  When we speak of free software, we are referring to freedom, not
-price.  Our General Public Licenses are designed to make sure that you
-have the freedom to distribute copies of free software (and charge for
-them if you wish), that you receive source code or can get it if you
-want it, that you can change the software or use pieces of it in new
-free programs, and that you know you can do these things.
-
-  To protect your rights, we need to prevent others from denying you
-these rights or asking you to surrender the rights.  Therefore, you have
-certain responsibilities if you distribute copies of the software, or if
-you modify it: responsibilities to respect the freedom of others.
-
-  For example, if you distribute copies of such a program, whether
-gratis or for a fee, you must pass on to the recipients the same
-freedoms that you received.  You must make sure that they, too, receive
-or can get the source code.  And you must show them these terms so they
-know their rights.
-
-  Developers that use the GNU GPL protect your rights with two steps:
-(1) assert copyright on the software, and (2) offer you this License
-giving you legal permission to copy, distribute and/or modify it.
-
-  For the developers' and authors' protection, the GPL clearly explains
-that there is no warranty for this free software.  For both users' and
-authors' sake, the GPL requires that modified versions be marked as
-changed, so that their problems will not be attributed erroneously to
-authors of previous versions.
-
-  Some devices are designed to deny users access to install or run
-modified versions of the software inside them, although the manufacturer
-can do so.  This is fundamentally incompatible with the aim of
-protecting users' freedom to change the software.  The systematic
-pattern of such abuse occurs in the area of products for individuals to
-use, which is precisely where it is most unacceptable.  Therefore, we
-have designed this version of the GPL to prohibit the practice for those
-products.  If such problems arise substantially in other domains, we
-stand ready to extend this provision to those domains in future versions
-of the GPL, as needed to protect the freedom of users.
-
-  Finally, every program is threatened constantly by software patents.
-States should not allow patents to restrict development and use of
-software on general-purpose computers, but in those that do, we wish to
-avoid the special danger that patents applied to a free program could
-make it effectively proprietary.  To prevent this, the GPL assures that
-patents cannot be used to render the program non-free.
-
-  The precise terms and conditions for copying, distribution and
-modification follow.
-
-                       TERMS AND CONDITIONS
-
-  0. Definitions.
-
-  "This License" refers to version 3 of the GNU General Public License.
-
-  "Copyright" also means copyright-like laws that apply to other kinds of
-works, such as semiconductor masks.
-
-  "The Program" refers to any copyrightable work licensed under this
-License.  Each licensee is addressed as "you".  "Licensees" and
-"recipients" may be individuals or organizations.
-
-  To "modify" a work means to copy from or adapt all or part of the work
-in a fashion requiring copyright permission, other than the making of an
-exact copy.  The resulting work is called a "modified version" of the
-earlier work or a work "based on" the earlier work.
-
-  A "covered work" means either the unmodified Program or a work based
-on the Program.
-
-  To "propagate" a work means to do anything with it that, without
-permission, would make you directly or secondarily liable for
-infringement under applicable copyright law, except executing it on a
-computer or modifying a private copy.  Propagation includes copying,
-distribution (with or without modification), making available to the
-public, and in some countries other activities as well.
-
-  To "convey" a work means any kind of propagation that enables other
-parties to make or receive copies.  Mere interaction with a user through
-a computer network, with no transfer of a copy, is not conveying.
-
-  An interactive user interface displays "Appropriate Legal Notices"
-to the extent that it includes a convenient and prominently visible
-feature that (1) displays an appropriate copyright notice, and (2)
-tells the user that there is no warranty for the work (except to the
-extent that warranties are provided), that licensees may convey the
-work under this License, and how to view a copy of this License.  If
-the interface presents a list of user commands or options, such as a
-menu, a prominent item in the list meets this criterion.
-
-  1. Source Code.
-
-  The "source code" for a work means the preferred form of the work
-for making modifications to it.  "Object code" means any non-source
-form of a work.
-
-  A "Standard Interface" means an interface that either is an official
-standard defined by a recognized standards body, or, in the case of
-interfaces specified for a particular programming language, one that
-is widely used among developers working in that language.
-
-  The "System Libraries" of an executable work include anything, other
-than the work as a whole, that (a) is included in the normal form of
-packaging a Major Component, but which is not part of that Major
-Component, and (b) serves only to enable use of the work with that
-Major Component, or to implement a Standard Interface for which an
-implementation is available to the public in source code form.  A
-"Major Component", in this context, means a major essential component
-(kernel, window system, and so on) of the specific operating system
-(if any) on which the executable work runs, or a compiler used to
-produce the work, or an object code interpreter used to run it.
-
-  The "Corresponding Source" for a work in object code form means all
-the source code needed to generate, install, and (for an executable
-work) run the object code and to modify the work, including scripts to
-control those activities.  However, it does not include the work's
-System Libraries, or general-purpose tools or generally available free
-programs which are used unmodified in performing those activities but
-which are not part of the work.  For example, Corresponding Source
-includes interface definition files associated with source files for
-the work, and the source code for shared libraries and dynamically
-linked subprograms that the work is specifically designed to require,
-such as by intimate data communication or control flow between those
-subprograms and other parts of the work.
-
-  The Corresponding Source need not include anything that users
-can regenerate automatically from other parts of the Corresponding
-Source.
-
-  The Corresponding Source for a work in source code form is that
-same work.
-
-  2. Basic Permissions.
-
-  All rights granted under this License are granted for the term of
-copyright on the Program, and are irrevocable provided the stated
-conditions are met.  This License explicitly affirms your unlimited
-permission to run the unmodified Program.  The output from running a
-covered work is covered by this License only if the output, given its
-content, constitutes a covered work.  This License acknowledges your
-rights of fair use or other equivalent, as provided by copyright law.
-
-  You may make, run and propagate covered works that you do not
-convey, without conditions so long as your license otherwise remains
-in force.  You may convey covered works to others for the sole purpose
-of having them make modifications exclusively for you, or provide you
-with facilities for running those works, provided that you comply with
-the terms of this License in conveying all material for which you do
-not control copyright.  Those thus making or running the covered works
-for you must do so exclusively on your behalf, under your direction
-and control, on terms that prohibit them from making any copies of
-your copyrighted material outside their relationship with you.
-
-  Conveying under any other circumstances is permitted solely under
-the conditions stated below.  Sublicensing is not allowed; section 10
-makes it unnecessary.
-
-  3. Protecting Users' Legal Rights From Anti-Circumvention Law.
-
-  No covered work shall be deemed part of an effective technological
-measure under any applicable law fulfilling obligations under article
-11 of the WIPO copyright treaty adopted on 20 December 1996, or
-similar laws prohibiting or restricting circumvention of such
-measures.
-
-  When you convey a covered work, you waive any legal power to forbid
-circumvention of technological measures to the extent such circumvention
-is effected by exercising rights under this License with respect to
-the covered work, and you disclaim any intention to limit operation or
-modification of the work as a means of enforcing, against the work's
-users, your or third parties' legal rights to forbid circumvention of
-technological measures.
-
-  4. Conveying Verbatim Copies.
-
-  You may convey verbatim copies of the Program's source code as you
-receive it, in any medium, provided that you conspicuously and
-appropriately publish on each copy an appropriate copyright notice;
-keep intact all notices stating that this License and any
-non-permissive terms added in accord with section 7 apply to the code;
-keep intact all notices of the absence of any warranty; and give all
-recipients a copy of this License along with the Program.
-
-  You may charge any price or no price for each copy that you convey,
-and you may offer support or warranty protection for a fee.
-
-  5. Conveying Modified Source Versions.
-
-  You may convey a work based on the Program, or the modifications to
-produce it from the Program, in the form of source code under the
-terms of section 4, provided that you also meet all of these conditions:
-
-    a) The work must carry prominent notices stating that you modified
-    it, and giving a relevant date.
-
-    b) The work must carry prominent notices stating that it is
-    released under this License and any conditions added under section
-    7.  This requirement modifies the requirement in section 4 to
-    "keep intact all notices".
-
-    c) You must license the entire work, as a whole, under this
-    License to anyone who comes into possession of a copy.  This
-    License will therefore apply, along with any applicable section 7
-    additional terms, to the whole of the work, and all its parts,
-    regardless of how they are packaged.  This License gives no
-    permission to license the work in any other way, but it does not
-    invalidate such permission if you have separately received it.
-
-    d) If the work has interactive user interfaces, each must display
-    Appropriate Legal Notices; however, if the Program has interactive
-    interfaces that do not display Appropriate Legal Notices, your
-    work need not make them do so.
-
-  A compilation of a covered work with other separate and independent
-works, which are not by their nature extensions of the covered work,
-and which are not combined with it such as to form a larger program,
-in or on a volume of a storage or distribution medium, is called an
-"aggregate" if the compilation and its resulting copyright are not
-used to limit the access or legal rights of the compilation's users
-beyond what the individual works permit.  Inclusion of a covered work
-in an aggregate does not cause this License to apply to the other
-parts of the aggregate.
-
-  6. Conveying Non-Source Forms.
-
-  You may convey a covered work in object code form under the terms
-of sections 4 and 5, provided that you also convey the
-machine-readable Corresponding Source under the terms of this License,
-in one of these ways:
-
-    a) Convey the object code in, or embodied in, a physical product
-    (including a physical distribution medium), accompanied by the
-    Corresponding Source fixed on a durable physical medium
-    customarily used for software interchange.
-
-    b) Convey the object code in, or embodied in, a physical product
-    (including a physical distribution medium), accompanied by a
-    written offer, valid for at least three years and valid for as
-    long as you offer spare parts or customer support for that product
-    model, to give anyone who possesses the object code either (1) a
-    copy of the Corresponding Source for all the software in the
-    product that is covered by this License, on a durable physical
-    medium customarily used for software interchange, for a price no
-    more than your reasonable cost of physically performing this
-    conveying of source, or (2) access to copy the
-    Corresponding Source from a network server at no charge.
-
-    c) Convey individual copies of the object code with a copy of the
-    written offer to provide the Corresponding Source.  This
-    alternative is allowed only occasionally and noncommercially, and
-    only if you received the object code with such an offer, in accord
-    with subsection 6b.
-
-    d) Convey the object code by offering access from a designated
-    place (gratis or for a charge), and offer equivalent access to the
-    Corresponding Source in the same way through the same place at no
-    further charge.  You need not require recipients to copy the
-    Corresponding Source along with the object code.  If the place to
-    copy the object code is a network server, the Corresponding Source
-    may be on a different server (operated by you or a third party)
-    that supports equivalent copying facilities, provided you maintain
-    clear directions next to the object code saying where to find the
-    Corresponding Source.  Regardless of what server hosts the
-    Corresponding Source, you remain obligated to ensure that it is
-    available for as long as needed to satisfy these requirements.
-
-    e) Convey the object code using peer-to-peer transmission, provided
-    you inform other peers where the object code and Corresponding
-    Source of the work are being offered to the general public at no
-    charge under subsection 6d.
-
-  A separable portion of the object code, whose source code is excluded
-from the Corresponding Source as a System Library, need not be
-included in conveying the object code work.
-
-  A "User Product" is either (1) a "consumer product", which means any
-tangible personal property which is normally used for personal, family,
-or household purposes, or (2) anything designed or sold for incorporation
-into a dwelling.  In determining whether a product is a consumer product,
-doubtful cases shall be resolved in favor of coverage.  For a particular
-product received by a particular user, "normally used" refers to a
-typical or common use of that class of product, regardless of the status
-of the particular user or of the way in which the particular user
-actually uses, or expects or is expected to use, the product.  A product
-is a consumer product regardless of whether the product has substantial
-commercial, industrial or non-consumer uses, unless such uses represent
-the only significant mode of use of the product.
-
-  "Installation Information" for a User Product means any methods,
-procedures, authorization keys, or other information required to install
-and execute modified versions of a covered work in that User Product from
-a modified version of its Corresponding Source.  The information must
-suffice to ensure that the continued functioning of the modified object
-code is in no case prevented or interfered with solely because
-modification has been made.
-
-  If you convey an object code work under this section in, or with, or
-specifically for use in, a User Product, and the conveying occurs as
-part of a transaction in which the right of possession and use of the
-User Product is transferred to the recipient in perpetuity or for a
-fixed term (regardless of how the transaction is characterized), the
-Corresponding Source conveyed under this section must be accompanied
-by the Installation Information.  But this requirement does not apply
-if neither you nor any third party retains the ability to install
-modified object code on the User Product (for example, the work has
-been installed in ROM).
-
-  The requirement to provide Installation Information does not include a
-requirement to continue to provide support service, warranty, or updates
-for a work that has been modified or installed by the recipient, or for
-the User Product in which it has been modified or installed.  Access to a
-network may be denied when the modification itself materially and
-adversely affects the operation of the network or violates the rules and
-protocols for communication across the network.
-
-  Corresponding Source conveyed, and Installation Information provided,
-in accord with this section must be in a format that is publicly
-documented (and with an implementation available to the public in
-source code form), and must require no special password or key for
-unpacking, reading or copying.
-
-  7. Additional Terms.
-
-  "Additional permissions" are terms that supplement the terms of this
-License by making exceptions from one or more of its conditions.
-Additional permissions that are applicable to the entire Program shall
-be treated as though they were included in this License, to the extent
-that they are valid under applicable law.  If additional permissions
-apply only to part of the Program, that part may be used separately
-under those permissions, but the entire Program remains governed by
-this License without regard to the additional permissions.
-
-  When you convey a copy of a covered work, you may at your option
-remove any additional permissions from that copy, or from any part of
-it.  (Additional permissions may be written to require their own
-removal in certain cases when you modify the work.)  You may place
-additional permissions on material, added by you to a covered work,
-for which you have or can give appropriate copyright permission.
-
-  Notwithstanding any other provision of this License, for material you
-add to a covered work, you may (if authorized by the copyright holders of
-that material) supplement the terms of this License with terms:
-
-    a) Disclaiming warranty or limiting liability differently from the
-    terms of sections 15 and 16 of this License; or
-
-    b) Requiring preservation of specified reasonable legal notices or
-    author attributions in that material or in the Appropriate Legal
-    Notices displayed by works containing it; or
-
-    c) Prohibiting misrepresentation of the origin of that material, or
-    requiring that modified versions of such material be marked in
-    reasonable ways as different from the original version; or
-
-    d) Limiting the use for publicity purposes of names of licensors or
-    authors of the material; or
-
-    e) Declining to grant rights under trademark law for use of some
-    trade names, trademarks, or service marks; or
-
-    f) Requiring indemnification of licensors and authors of that
-    material by anyone who conveys the material (or modified versions of
-    it) with contractual assumptions of liability to the recipient, for
-    any liability that these contractual assumptions directly impose on
-    those licensors and authors.
-
-  All other non-permissive additional terms are considered "further
-restrictions" within the meaning of section 10.  If the Program as you
-received it, or any part of it, contains a notice stating that it is
-governed by this License along with a term that is a further
-restriction, you may remove that term.  If a license document contains
-a further restriction but permits relicensing or conveying under this
-License, you may add to a covered work material governed by the terms
-of that license document, provided that the further restriction does
-not survive such relicensing or conveying.
-
-  If you add terms to a covered work in accord with this section, you
-must place, in the relevant source files, a statement of the
-additional terms that apply to those files, or a notice indicating
-where to find the applicable terms.
-
-  Additional terms, permissive or non-permissive, may be stated in the
-form of a separately written license, or stated as exceptions;
-the above requirements apply either way.
-
-  8. Termination.
-
-  You may not propagate or modify a covered work except as expressly
-provided under this License.  Any attempt otherwise to propagate or
-modify it is void, and will automatically terminate your rights under
-this License (including any patent licenses granted under the third
-paragraph of section 11).
-
-  However, if you cease all violation of this License, then your
-license from a particular copyright holder is reinstated (a)
-provisionally, unless and until the copyright holder explicitly and
-finally terminates your license, and (b) permanently, if the copyright
-holder fails to notify you of the violation by some reasonable means
-prior to 60 days after the cessation.
-
-  Moreover, your license from a particular copyright holder is
-reinstated permanently if the copyright holder notifies you of the
-violation by some reasonable means, this is the first time you have
-received notice of violation of this License (for any work) from that
-copyright holder, and you cure the violation prior to 30 days after
-your receipt of the notice.
-
-  Termination of your rights under this section does not terminate the
-licenses of parties who have received copies or rights from you under
-this License.  If your rights have been terminated and not permanently
-reinstated, you do not qualify to receive new licenses for the same
-material under section 10.
-
-  9. Acceptance Not Required for Having Copies.
-
-  You are not required to accept this License in order to receive or
-run a copy of the Program.  Ancillary propagation of a covered work
-occurring solely as a consequence of using peer-to-peer transmission
-to receive a copy likewise does not require acceptance.  However,
-nothing other than this License grants you permission to propagate or
-modify any covered work.  These actions infringe copyright if you do
-not accept this License.  Therefore, by modifying or propagating a
-covered work, you indicate your acceptance of this License to do so.
-
-  10. Automatic Licensing of Downstream Recipients.
-
-  Each time you convey a covered work, the recipient automatically
-receives a license from the original licensors, to run, modify and
-propagate that work, subject to this License.  You are not responsible
-for enforcing compliance by third parties with this License.
-
-  An "entity transaction" is a transaction transferring control of an
-organization, or substantially all assets of one, or subdividing an
-organization, or merging organizations.  If propagation of a covered
-work results from an entity transaction, each party to that
-transaction who receives a copy of the work also receives whatever
-licenses to the work the party's predecessor in interest had or could
-give under the previous paragraph, plus a right to possession of the
-Corresponding Source of the work from the predecessor in interest, if
-the predecessor has it or can get it with reasonable efforts.
-
-  You may not impose any further restrictions on the exercise of the
-rights granted or affirmed under this License.  For example, you may
-not impose a license fee, royalty, or other charge for exercise of
-rights granted under this License, and you may not initiate litigation
-(including a cross-claim or counterclaim in a lawsuit) alleging that
-any patent claim is infringed by making, using, selling, offering for
-sale, or importing the Program or any portion of it.
-
-  11. Patents.
-
-  A "contributor" is a copyright holder who authorizes use under this
-License of the Program or a work on which the Program is based.  The
-work thus licensed is called the contributor's "contributor version".
-
-  A contributor's "essential patent claims" are all patent claims
-owned or controlled by the contributor, whether already acquired or
-hereafter acquired, that would be infringed by some manner, permitted
-by this License, of making, using, or selling its contributor version,
-but do not include claims that would be infringed only as a
-consequence of further modification of the contributor version.  For
-purposes of this definition, "control" includes the right to grant
-patent sublicenses in a manner consistent with the requirements of
-this License.
-
-  Each contributor grants you a non-exclusive, worldwide, royalty-free
-patent license under the contributor's essential patent claims, to
-make, use, sell, offer for sale, import and otherwise run, modify and
-propagate the contents of its contributor version.
-
-  In the following three paragraphs, a "patent license" is any express
-agreement or commitment, however denominated, not to enforce a patent
-(such as an express permission to practice a patent or covenant not to
-sue for patent infringement).  To "grant" such a patent license to a
-party means to make such an agreement or commitment not to enforce a
-patent against the party.
-
-  If you convey a covered work, knowingly relying on a patent license,
-and the Corresponding Source of the work is not available for anyone
-to copy, free of charge and under the terms of this License, through a
-publicly available network server or other readily accessible means,
-then you must either (1) cause the Corresponding Source to be so
-available, or (2) arrange to deprive yourself of the benefit of the
-patent license for this particular work, or (3) arrange, in a manner
-consistent with the requirements of this License, to extend the patent
-license to downstream recipients.  "Knowingly relying" means you have
-actual knowledge that, but for the patent license, your conveying the
-covered work in a country, or your recipient's use of the covered work
-in a country, would infringe one or more identifiable patents in that
-country that you have reason to believe are valid.
-
-  If, pursuant to or in connection with a single transaction or
-arrangement, you convey, or propagate by procuring conveyance of, a
-covered work, and grant a patent license to some of the parties
-receiving the covered work authorizing them to use, propagate, modify
-or convey a specific copy of the covered work, then the patent license
-you grant is automatically extended to all recipients of the covered
-work and works based on it.
-
-  A patent license is "discriminatory" if it does not include within
-the scope of its coverage, prohibits the exercise of, or is
-conditioned on the non-exercise of one or more of the rights that are
-specifically granted under this License.  You may not convey a covered
-work if you are a party to an arrangement with a third party that is
-in the business of distributing software, under which you make payment
-to the third party based on the extent of your activity of conveying
-the work, and under which the third party grants, to any of the
-parties who would receive the covered work from you, a discriminatory
-patent license (a) in connection with copies of the covered work
-conveyed by you (or copies made from those copies), or (b) primarily
-for and in connection with specific products or compilations that
-contain the covered work, unless you entered into that arrangement,
-or that patent license was granted, prior to 28 March 2007.
-
-  Nothing in this License shall be construed as excluding or limiting
-any implied license or other defenses to infringement that may
-otherwise be available to you under applicable patent law.
-
-  12. No Surrender of Others' Freedom.
-
-  If conditions are imposed on you (whether by court order, agreement or
-otherwise) that contradict the conditions of this License, they do not
-excuse you from the conditions of this License.  If you cannot convey a
-covered work so as to satisfy simultaneously your obligations under this
-License and any other pertinent obligations, then as a consequence you may
-not convey it at all.  For example, if you agree to terms that obligate you
-to collect a royalty for further conveying from those to whom you convey
-the Program, the only way you could satisfy both those terms and this
-License would be to refrain entirely from conveying the Program.
-
-  13. Use with the GNU Affero General Public License.
-
-  Notwithstanding any other provision of this License, you have
-permission to link or combine any covered work with a work licensed
-under version 3 of the GNU Affero General Public License into a single
-combined work, and to convey the resulting work.  The terms of this
-License will continue to apply to the part which is the covered work,
-but the special requirements of the GNU Affero General Public License,
-section 13, concerning interaction through a network will apply to the
-combination as such.
-
-  14. Revised Versions of this License.
-
-  The Free Software Foundation may publish revised and/or new versions of
-the GNU General Public License from time to time.  Such new versions will
-be similar in spirit to the present version, but may differ in detail to
-address new problems or concerns.
-
-  Each version is given a distinguishing version number.  If the
-Program specifies that a certain numbered version of the GNU General
-Public License "or any later version" applies to it, you have the
-option of following the terms and conditions either of that numbered
-version or of any later version published by the Free Software
-Foundation.  If the Program does not specify a version number of the
-GNU General Public License, you may choose any version ever published
-by the Free Software Foundation.
-
-  If the Program specifies that a proxy can decide which future
-versions of the GNU General Public License can be used, that proxy's
-public statement of acceptance of a version permanently authorizes you
-to choose that version for the Program.
-
-  Later license versions may give you additional or different
-permissions.  However, no additional obligations are imposed on any
-author or copyright holder as a result of your choosing to follow a
-later version.
-
-  15. Disclaimer of Warranty.
-
-  THERE IS NO WARRANTY FOR THE PROGRAM, TO THE EXTENT PERMITTED BY
-APPLICABLE LAW.  EXCEPT WHEN OTHERWISE STATED IN WRITING THE COPYRIGHT
-HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM "AS IS" WITHOUT WARRANTY
-OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT NOT LIMITED TO,
-THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
-PURPOSE.  THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE OF THE PROGRAM
-IS WITH YOU.  SHOULD THE PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF
-ALL NECESSARY SERVICING, REPAIR OR CORRECTION.
-
-  16. Limitation of Liability.
-
-  IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING
-WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MODIFIES AND/OR CONVEYS
-THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU FOR DAMAGES, INCLUDING ANY
-GENERAL, SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING OUT OF THE
-USE OR INABILITY TO USE THE PROGRAM (INCLUDING BUT NOT LIMITED TO LOSS OF
-DATA OR DATA BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY YOU OR THIRD
-PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE WITH ANY OTHER PROGRAMS),
-EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE POSSIBILITY OF
-SUCH DAMAGES.
-
-  17. Interpretation of Sections 15 and 16.
-
-  If the disclaimer of warranty and limitation of liability provided
-above cannot be given local legal effect according to their terms,
-reviewing courts shall apply local law that most closely approximates
-an absolute waiver of all civil liability in connection with the
-Program, unless a warranty or assumption of liability accompanies a
-copy of the Program in return for a fee.
-
-                     END OF TERMS AND CONDITIONS
-
-            How to Apply These Terms to Your New Programs
-
-  If you develop a new program, and you want it to be of the greatest
-possible use to the public, the best way to achieve this is to make it
-free software which everyone can redistribute and change under these terms.
-
-  To do so, attach the following notices to the program.  It is safest
-to attach them to the start of each source file to most effectively
-state the exclusion of warranty; and each file should have at least
-the "copyright" line and a pointer to where the full notice is found.
-
-    {one line to give the program's name and a brief idea of what it does.}
-    Copyright (C) {year}  {name of author}
-
-    This program is free software: you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation, either version 3 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License
-    along with this program.  If not, see <http://www.gnu.org/licenses/>.
-
-Also add information on how to contact you by electronic and paper mail.
-
-  If the program does terminal interaction, make it output a short
-notice like this when it starts in an interactive mode:
-
-    {project}  Copyright (C) {year}  {fullname}
-    This program comes with ABSOLUTELY NO WARRANTY; for details type `show w'.
-    This is free software, and you are welcome to redistribute it
-    under certain conditions; type `show c' for details.
-
-The hypothetical commands `show w' and `show c' should show the appropriate
-parts of the General Public License.  Of course, your program's commands
-might be different; for a GUI interface, you would use an "about box".
-
-  You should also get your employer (if you work as a programmer) or school,
-if any, to sign a "copyright disclaimer" for the program, if necessary.
-For more information on this, and how to apply and follow the GNU GPL, see
-<http://www.gnu.org/licenses/>.
-
-  The GNU General Public License does not permit incorporating your program
-into proprietary programs.  If your program is a subroutine library, you
-may consider it more useful to permit linking proprietary applications with
-the library.  If this is what you want to do, use the GNU Lesser General
-Public License instead of this License.  But first, please read
-<http://www.gnu.org/philosophy/why-not-lgpl.html>.
\ No newline at end of file
+ccminer is available under the terms of the GNU Public License version 3.
+
+See LICENSE.TXT for details.
diff --git a/LICENSE.txt b/LICENSE.txt
new file mode 100644
index 0000000000..8e70fb3c86
--- /dev/null
+++ b/LICENSE.txt
@@ -0,0 +1,885 @@
+                    GNU GENERAL PUBLIC LICENSE
+                       Version 3, 29 June 2007
+
+ Copyright (C) 2007 Free Software Foundation, Inc. <http://fsf.org/>
+ Everyone is permitted to copy and distribute verbatim copies
+ of this license document, but changing it is not allowed.
+
+                            Preamble
+
+  The GNU General Public License is a free, copyleft license for
+software and other kinds of works.
+
+  The licenses for most software and other practical works are designed
+to take away your freedom to share and change the works.  By contrast,
+the GNU General Public License is intended to guarantee your freedom to
+share and change all versions of a program--to make sure it remains free
+software for all its users.  We, the Free Software Foundation, use the
+GNU General Public License for most of our software; it applies also to
+any other work released this way by its authors.  You can apply it to
+your programs, too.
+
+  When we speak of free software, we are referring to freedom, not
+price.  Our General Public Licenses are designed to make sure that you
+have the freedom to distribute copies of free software (and charge for
+them if you wish), that you receive source code or can get it if you
+want it, that you can change the software or use pieces of it in new
+free programs, and that you know you can do these things.
+
+  To protect your rights, we need to prevent others from denying you
+these rights or asking you to surrender the rights.  Therefore, you have
+certain responsibilities if you distribute copies of the software, or if
+you modify it: responsibilities to respect the freedom of others.
+
+  For example, if you distribute copies of such a program, whether
+gratis or for a fee, you must pass on to the recipients the same
+freedoms that you received.  You must make sure that they, too, receive
+or can get the source code.  And you must show them these terms so they
+know their rights.
+
+  Developers that use the GNU GPL protect your rights with two steps:
+(1) assert copyright on the software, and (2) offer you this License
+giving you legal permission to copy, distribute and/or modify it.
+
+  For the developers' and authors' protection, the GPL clearly explains
+that there is no warranty for this free software.  For both users' and
+authors' sake, the GPL requires that modified versions be marked as
+changed, so that their problems will not be attributed erroneously to
+authors of previous versions.
+
+  Some devices are designed to deny users access to install or run
+modified versions of the software inside them, although the manufacturer
+can do so.  This is fundamentally incompatible with the aim of
+protecting users' freedom to change the software.  The systematic
+pattern of such abuse occurs in the area of products for individuals to
+use, which is precisely where it is most unacceptable.  Therefore, we
+have designed this version of the GPL to prohibit the practice for those
+products.  If such problems arise substantially in other domains, we
+stand ready to extend this provision to those domains in future versions
+of the GPL, as needed to protect the freedom of users.
+
+  Finally, every program is threatened constantly by software patents.
+States should not allow patents to restrict development and use of
+software on general-purpose computers, but in those that do, we wish to
+avoid the special danger that patents applied to a free program could
+make it effectively proprietary.  To prevent this, the GPL assures that
+patents cannot be used to render the program non-free.
+
+  The precise terms and conditions for copying, distribution and
+modification follow.
+
+                       TERMS AND CONDITIONS
+
+  0. Definitions.
+
+  "This License" refers to version 3 of the GNU General Public License.
+
+  "Copyright" also means copyright-like laws that apply to other kinds of
+works, such as semiconductor masks.
+
+  "The Program" refers to any copyrightable work licensed under this
+License.  Each licensee is addressed as "you".  "Licensees" and
+"recipients" may be individuals or organizations.
+
+  To "modify" a work means to copy from or adapt all or part of the work
+in a fashion requiring copyright permission, other than the making of an
+exact copy.  The resulting work is called a "modified version" of the
+earlier work or a work "based on" the earlier work.
+
+  A "covered work" means either the unmodified Program or a work based
+on the Program.
+
+  To "propagate" a work means to do anything with it that, without
+permission, would make you directly or secondarily liable for
+infringement under applicable copyright law, except executing it on a
+computer or modifying a private copy.  Propagation includes copying,
+distribution (with or without modification), making available to the
+public, and in some countries other activities as well.
+
+  To "convey" a work means any kind of propagation that enables other
+parties to make or receive copies.  Mere interaction with a user through
+a computer network, with no transfer of a copy, is not conveying.
+
+  An interactive user interface displays "Appropriate Legal Notices"
+to the extent that it includes a convenient and prominently visible
+feature that (1) displays an appropriate copyright notice, and (2)
+tells the user that there is no warranty for the work (except to the
+extent that warranties are provided), that licensees may convey the
+work under this License, and how to view a copy of this License.  If
+the interface presents a list of user commands or options, such as a
+menu, a prominent item in the list meets this criterion.
+
+  1. Source Code.
+
+  The "source code" for a work means the preferred form of the work
+for making modifications to it.  "Object code" means any non-source
+form of a work.
+
+  A "Standard Interface" means an interface that either is an official
+standard defined by a recognized standards body, or, in the case of
+interfaces specified for a particular programming language, one that
+is widely used among developers working in that language.
+
+  The "System Libraries" of an executable work include anything, other
+than the work as a whole, that (a) is included in the normal form of
+packaging a Major Component, but which is not part of that Major
+Component, and (b) serves only to enable use of the work with that
+Major Component, or to implement a Standard Interface for which an
+implementation is available to the public in source code form.  A
+"Major Component", in this context, means a major essential component
+(kernel, window system, and so on) of the specific operating system
+(if any) on which the executable work runs, or a compiler used to
+produce the work, or an object code interpreter used to run it.
+
+  The "Corresponding Source" for a work in object code form means all
+the source code needed to generate, install, and (for an executable
+work) run the object code and to modify the work, including scripts to
+control those activities.  However, it does not include the work's
+System Libraries, or general-purpose tools or generally available free
+programs which are used unmodified in performing those activities but
+which are not part of the work.  For example, Corresponding Source
+includes interface definition files associated with source files for
+the work, and the source code for shared libraries and dynamically
+linked subprograms that the work is specifically designed to require,
+such as by intimate data communication or control flow between those
+subprograms and other parts of the work.
+
+  The Corresponding Source need not include anything that users
+can regenerate automatically from other parts of the Corresponding
+Source.
+
+  The Corresponding Source for a work in source code form is that
+same work.
+
+  2. Basic Permissions.
+
+  All rights granted under this License are granted for the term of
+copyright on the Program, and are irrevocable provided the stated
+conditions are met.  This License explicitly affirms your unlimited
+permission to run the unmodified Program.  The output from running a
+covered work is covered by this License only if the output, given its
+content, constitutes a covered work.  This License acknowledges your
+rights of fair use or other equivalent, as provided by copyright law.
+
+  You may make, run and propagate covered works that you do not
+convey, without conditions so long as your license otherwise remains
+in force.  You may convey covered works to others for the sole purpose
+of having them make modifications exclusively for you, or provide you
+with facilities for running those works, provided that you comply with
+the terms of this License in conveying all material for which you do
+not control copyright.  Those thus making or running the covered works
+for you must do so exclusively on your behalf, under your direction
+and control, on terms that prohibit them from making any copies of
+your copyrighted material outside their relationship with you.
+
+  Conveying under any other circumstances is permitted solely under
+the conditions stated below.  Sublicensing is not allowed; section 10
+makes it unnecessary.
+
+  3. Protecting Users' Legal Rights From Anti-Circumvention Law.
+
+  No covered work shall be deemed part of an effective technological
+measure under any applicable law fulfilling obligations under article
+11 of the WIPO copyright treaty adopted on 20 December 1996, or
+similar laws prohibiting or restricting circumvention of such
+measures.
+
+  When you convey a covered work, you waive any legal power to forbid
+circumvention of technological measures to the extent such circumvention
+is effected by exercising rights under this License with respect to
+the covered work, and you disclaim any intention to limit operation or
+modification of the work as a means of enforcing, against the work's
+users, your or third parties' legal rights to forbid circumvention of
+technological measures.
+
+  4. Conveying Verbatim Copies.
+
+  You may convey verbatim copies of the Program's source code as you
+receive it, in any medium, provided that you conspicuously and
+appropriately publish on each copy an appropriate copyright notice;
+keep intact all notices stating that this License and any
+non-permissive terms added in accord with section 7 apply to the code;
+keep intact all notices of the absence of any warranty; and give all
+recipients a copy of this License along with the Program.
+
+  You may charge any price or no price for each copy that you convey,
+and you may offer support or warranty protection for a fee.
+
+  5. Conveying Modified Source Versions.
+
+  You may convey a work based on the Program, or the modifications to
+produce it from the Program, in the form of source code under the
+terms of section 4, provided that you also meet all of these conditions:
+
+    a) The work must carry prominent notices stating that you modified
+    it, and giving a relevant date.
+
+    b) The work must carry prominent notices stating that it is
+    released under this License and any conditions added under section
+    7.  This requirement modifies the requirement in section 4 to
+    "keep intact all notices".
+
+    c) You must license the entire work, as a whole, under this
+    License to anyone who comes into possession of a copy.  This
+    License will therefore apply, along with any applicable section 7
+    additional terms, to the whole of the work, and all its parts,
+    regardless of how they are packaged.  This License gives no
+    permission to license the work in any other way, but it does not
+    invalidate such permission if you have separately received it.
+
+    d) If the work has interactive user interfaces, each must display
+    Appropriate Legal Notices; however, if the Program has interactive
+    interfaces that do not display Appropriate Legal Notices, your
+    work need not make them do so.
+
+  A compilation of a covered work with other separate and independent
+works, which are not by their nature extensions of the covered work,
+and which are not combined with it such as to form a larger program,
+in or on a volume of a storage or distribution medium, is called an
+"aggregate" if the compilation and its resulting copyright are not
+used to limit the access or legal rights of the compilation's users
+beyond what the individual works permit.  Inclusion of a covered work
+in an aggregate does not cause this License to apply to the other
+parts of the aggregate.
+
+  6. Conveying Non-Source Forms.
+
+  You may convey a covered work in object code form under the terms
+of sections 4 and 5, provided that you also convey the
+machine-readable Corresponding Source under the terms of this License,
+in one of these ways:
+
+    a) Convey the object code in, or embodied in, a physical product
+    (including a physical distribution medium), accompanied by the
+    Corresponding Source fixed on a durable physical medium
+    customarily used for software interchange.
+
+    b) Convey the object code in, or embodied in, a physical product
+    (including a physical distribution medium), accompanied by a
+    written offer, valid for at least three years and valid for as
+    long as you offer spare parts or customer support for that product
+    model, to give anyone who possesses the object code either (1) a
+    copy of the Corresponding Source for all the software in the
+    product that is covered by this License, on a durable physical
+    medium customarily used for software interchange, for a price no
+    more than your reasonable cost of physically performing this
+    conveying of source, or (2) access to copy the
+    Corresponding Source from a network server at no charge.
+
+    c) Convey individual copies of the object code with a copy of the
+    written offer to provide the Corresponding Source.  This
+    alternative is allowed only occasionally and noncommercially, and
+    only if you received the object code with such an offer, in accord
+    with subsection 6b.
+
+    d) Convey the object code by offering access from a designated
+    place (gratis or for a charge), and offer equivalent access to the
+    Corresponding Source in the same way through the same place at no
+    further charge.  You need not require recipients to copy the
+    Corresponding Source along with the object code.  If the place to
+    copy the object code is a network server, the Corresponding Source
+    may be on a different server (operated by you or a third party)
+    that supports equivalent copying facilities, provided you maintain
+    clear directions next to the object code saying where to find the
+    Corresponding Source.  Regardless of what server hosts the
+    Corresponding Source, you remain obligated to ensure that it is
+    available for as long as needed to satisfy these requirements.
+
+    e) Convey the object code using peer-to-peer transmission, provided
+    you inform other peers where the object code and Corresponding
+    Source of the work are being offered to the general public at no
+    charge under subsection 6d.
+
+  A separable portion of the object code, whose source code is excluded
+from the Corresponding Source as a System Library, need not be
+included in conveying the object code work.
+
+  A "User Product" is either (1) a "consumer product", which means any
+tangible personal property which is normally used for personal, family,
+or household purposes, or (2) anything designed or sold for incorporation
+into a dwelling.  In determining whether a product is a consumer product,
+doubtful cases shall be resolved in favor of coverage.  For a particular
+product received by a particular user, "normally used" refers to a
+typical or common use of that class of product, regardless of the status
+of the particular user or of the way in which the particular user
+actually uses, or expects or is expected to use, the product.  A product
+is a consumer product regardless of whether the product has substantial
+commercial, industrial or non-consumer uses, unless such uses represent
+the only significant mode of use of the product.
+
+  "Installation Information" for a User Product means any methods,
+procedures, authorization keys, or other information required to install
+and execute modified versions of a covered work in that User Product from
+a modified version of its Corresponding Source.  The information must
+suffice to ensure that the continued functioning of the modified object
+code is in no case prevented or interfered with solely because
+modification has been made.
+
+  If you convey an object code work under this section in, or with, or
+specifically for use in, a User Product, and the conveying occurs as
+part of a transaction in which the right of possession and use of the
+User Product is transferred to the recipient in perpetuity or for a
+fixed term (regardless of how the transaction is characterized), the
+Corresponding Source conveyed under this section must be accompanied
+by the Installation Information.  But this requirement does not apply
+if neither you nor any third party retains the ability to install
+modified object code on the User Product (for example, the work has
+been installed in ROM).
+
+  The requirement to provide Installation Information does not include a
+requirement to continue to provide support service, warranty, or updates
+for a work that has been modified or installed by the recipient, or for
+the User Product in which it has been modified or installed.  Access to a
+network may be denied when the modification itself materially and
+adversely affects the operation of the network or violates the rules and
+protocols for communication across the network.
+
+  Corresponding Source conveyed, and Installation Information provided,
+in accord with this section must be in a format that is publicly
+documented (and with an implementation available to the public in
+source code form), and must require no special password or key for
+unpacking, reading or copying.
+
+  7. Additional Terms.
+
+  "Additional permissions" are terms that supplement the terms of this
+License by making exceptions from one or more of its conditions.
+Additional permissions that are applicable to the entire Program shall
+be treated as though they were included in this License, to the extent
+that they are valid under applicable law.  If additional permissions
+apply only to part of the Program, that part may be used separately
+under those permissions, but the entire Program remains governed by
+this License without regard to the additional permissions.
+
+  When you convey a copy of a covered work, you may at your option
+remove any additional permissions from that copy, or from any part of
+it.  (Additional permissions may be written to require their own
+removal in certain cases when you modify the work.)  You may place
+additional permissions on material, added by you to a covered work,
+for which you have or can give appropriate copyright permission.
+
+  Notwithstanding any other provision of this License, for material you
+add to a covered work, you may (if authorized by the copyright holders of
+that material) supplement the terms of this License with terms:
+
+    a) Disclaiming warranty or limiting liability differently from the
+    terms of sections 15 and 16 of this License; or
+
+    b) Requiring preservation of specified reasonable legal notices or
+    author attributions in that material or in the Appropriate Legal
+    Notices displayed by works containing it; or
+
+    c) Prohibiting misrepresentation of the origin of that material, or
+    requiring that modified versions of such material be marked in
+    reasonable ways as different from the original version; or
+
+    d) Limiting the use for publicity purposes of names of licensors or
+    authors of the material; or
+
+    e) Declining to grant rights under trademark law for use of some
+    trade names, trademarks, or service marks; or
+
+    f) Requiring indemnification of licensors and authors of that
+    material by anyone who conveys the material (or modified versions of
+    it) with contractual assumptions of liability to the recipient, for
+    any liability that these contractual assumptions directly impose on
+    those licensors and authors.
+
+  All other non-permissive additional terms are considered "further
+restrictions" within the meaning of section 10.  If the Program as you
+received it, or any part of it, contains a notice stating that it is
+governed by this License along with a term that is a further
+restriction, you may remove that term.  If a license document contains
+a further restriction but permits relicensing or conveying under this
+License, you may add to a covered work material governed by the terms
+of that license document, provided that the further restriction does
+not survive such relicensing or conveying.
+
+  If you add terms to a covered work in accord with this section, you
+must place, in the relevant source files, a statement of the
+additional terms that apply to those files, or a notice indicating
+where to find the applicable terms.
+
+  Additional terms, permissive or non-permissive, may be stated in the
+form of a separately written license, or stated as exceptions;
+the above requirements apply either way.
+
+  8. Termination.
+
+  You may not propagate or modify a covered work except as expressly
+provided under this License.  Any attempt otherwise to propagate or
+modify it is void, and will automatically terminate your rights under
+this License (including any patent licenses granted under the third
+paragraph of section 11).
+
+  However, if you cease all violation of this License, then your
+license from a particular copyright holder is reinstated (a)
+provisionally, unless and until the copyright holder explicitly and
+finally terminates your license, and (b) permanently, if the copyright
+holder fails to notify you of the violation by some reasonable means
+prior to 60 days after the cessation.
+
+  Moreover, your license from a particular copyright holder is
+reinstated permanently if the copyright holder notifies you of the
+violation by some reasonable means, this is the first time you have
+received notice of violation of this License (for any work) from that
+copyright holder, and you cure the violation prior to 30 days after
+your receipt of the notice.
+
+  Termination of your rights under this section does not terminate the
+licenses of parties who have received copies or rights from you under
+this License.  If your rights have been terminated and not permanently
+reinstated, you do not qualify to receive new licenses for the same
+material under section 10.
+
+  9. Acceptance Not Required for Having Copies.
+
+  You are not required to accept this License in order to receive or
+run a copy of the Program.  Ancillary propagation of a covered work
+occurring solely as a consequence of using peer-to-peer transmission
+to receive a copy likewise does not require acceptance.  However,
+nothing other than this License grants you permission to propagate or
+modify any covered work.  These actions infringe copyright if you do
+not accept this License.  Therefore, by modifying or propagating a
+covered work, you indicate your acceptance of this License to do so.
+
+  10. Automatic Licensing of Downstream Recipients.
+
+  Each time you convey a covered work, the recipient automatically
+receives a license from the original licensors, to run, modify and
+propagate that work, subject to this License.  You are not responsible
+for enforcing compliance by third parties with this License.
+
+  An "entity transaction" is a transaction transferring control of an
+organization, or substantially all assets of one, or subdividing an
+organization, or merging organizations.  If propagation of a covered
+work results from an entity transaction, each party to that
+transaction who receives a copy of the work also receives whatever
+licenses to the work the party's predecessor in interest had or could
+give under the previous paragraph, plus a right to possession of the
+Corresponding Source of the work from the predecessor in interest, if
+the predecessor has it or can get it with reasonable efforts.
+
+  You may not impose any further restrictions on the exercise of the
+rights granted or affirmed under this License.  For example, you may
+not impose a license fee, royalty, or other charge for exercise of
+rights granted under this License, and you may not initiate litigation
+(including a cross-claim or counterclaim in a lawsuit) alleging that
+any patent claim is infringed by making, using, selling, offering for
+sale, or importing the Program or any portion of it.
+
+  11. Patents.
+
+  A "contributor" is a copyright holder who authorizes use under this
+License of the Program or a work on which the Program is based.  The
+work thus licensed is called the contributor's "contributor version".
+
+  A contributor's "essential patent claims" are all patent claims
+owned or controlled by the contributor, whether already acquired or
+hereafter acquired, that would be infringed by some manner, permitted
+by this License, of making, using, or selling its contributor version,
+but do not include claims that would be infringed only as a
+consequence of further modification of the contributor version.  For
+purposes of this definition, "control" includes the right to grant
+patent sublicenses in a manner consistent with the requirements of
+this License.
+
+  Each contributor grants you a non-exclusive, worldwide, royalty-free
+patent license under the contributor's essential patent claims, to
+make, use, sell, offer for sale, import and otherwise run, modify and
+propagate the contents of its contributor version.
+
+  In the following three paragraphs, a "patent license" is any express
+agreement or commitment, however denominated, not to enforce a patent
+(such as an express permission to practice a patent or covenant not to
+sue for patent infringement).  To "grant" such a patent license to a
+party means to make such an agreement or commitment not to enforce a
+patent against the party.
+
+  If you convey a covered work, knowingly relying on a patent license,
+and the Corresponding Source of the work is not available for anyone
+to copy, free of charge and under the terms of this License, through a
+publicly available network server or other readily accessible means,
+then you must either (1) cause the Corresponding Source to be so
+available, or (2) arrange to deprive yourself of the benefit of the
+patent license for this particular work, or (3) arrange, in a manner
+consistent with the requirements of this License, to extend the patent
+license to downstream recipients.  "Knowingly relying" means you have
+actual knowledge that, but for the patent license, your conveying the
+covered work in a country, or your recipient's use of the covered work
+in a country, would infringe one or more identifiable patents in that
+country that you have reason to believe are valid.
+
+  If, pursuant to or in connection with a single transaction or
+arrangement, you convey, or propagate by procuring conveyance of, a
+covered work, and grant a patent license to some of the parties
+receiving the covered work authorizing them to use, propagate, modify
+or convey a specific copy of the covered work, then the patent license
+you grant is automatically extended to all recipients of the covered
+work and works based on it.
+
+  A patent license is "discriminatory" if it does not include within
+the scope of its coverage, prohibits the exercise of, or is
+conditioned on the non-exercise of one or more of the rights that are
+specifically granted under this License.  You may not convey a covered
+work if you are a party to an arrangement with a third party that is
+in the business of distributing software, under which you make payment
+to the third party based on the extent of your activity of conveying
+the work, and under which the third party grants, to any of the
+parties who would receive the covered work from you, a discriminatory
+patent license (a) in connection with copies of the covered work
+conveyed by you (or copies made from those copies), or (b) primarily
+for and in connection with specific products or compilations that
+contain the covered work, unless you entered into that arrangement,
+or that patent license was granted, prior to 28 March 2007.
+
+  Nothing in this License shall be construed as excluding or limiting
+any implied license or other defenses to infringement that may
+otherwise be available to you under applicable patent law.
+
+  12. No Surrender of Others' Freedom.
+
+  If conditions are imposed on you (whether by court order, agreement or
+otherwise) that contradict the conditions of this License, they do not
+excuse you from the conditions of this License.  If you cannot convey a
+covered work so as to satisfy simultaneously your obligations under this
+License and any other pertinent obligations, then as a consequence you may
+not convey it at all.  For example, if you agree to terms that obligate you
+to collect a royalty for further conveying from those to whom you convey
+the Program, the only way you could satisfy both those terms and this
+License would be to refrain entirely from conveying the Program.
+
+  13. Use with the GNU Affero General Public License.
+
+  Notwithstanding any other provision of this License, you have
+permission to link or combine any covered work with a work licensed
+under version 3 of the GNU Affero General Public License into a single
+combined work, and to convey the resulting work.  The terms of this
+License will continue to apply to the part which is the covered work,
+but the special requirements of the GNU Affero General Public License,
+section 13, concerning interaction through a network will apply to the
+combination as such.
+
+  14. Revised Versions of this License.
+
+  The Free Software Foundation may publish revised and/or new versions of
+the GNU General Public License from time to time.  Such new versions will
+be similar in spirit to the present version, but may differ in detail to
+address new problems or concerns.
+
+  Each version is given a distinguishing version number.  If the
+Program specifies that a certain numbered version of the GNU General
+Public License "or any later version" applies to it, you have the
+option of following the terms and conditions either of that numbered
+version or of any later version published by the Free Software
+Foundation.  If the Program does not specify a version number of the
+GNU General Public License, you may choose any version ever published
+by the Free Software Foundation.
+
+  If the Program specifies that a proxy can decide which future
+versions of the GNU General Public License can be used, that proxy's
+public statement of acceptance of a version permanently authorizes you
+to choose that version for the Program.
+
+  Later license versions may give you additional or different
+permissions.  However, no additional obligations are imposed on any
+author or copyright holder as a result of your choosing to follow a
+later version.
+
+  15. Disclaimer of Warranty.
+
+  THERE IS NO WARRANTY FOR THE PROGRAM, TO THE EXTENT PERMITTED BY
+APPLICABLE LAW.  EXCEPT WHEN OTHERWISE STATED IN WRITING THE COPYRIGHT
+HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM "AS IS" WITHOUT WARRANTY
+OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT NOT LIMITED TO,
+THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+PURPOSE.  THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE OF THE PROGRAM
+IS WITH YOU.  SHOULD THE PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF
+ALL NECESSARY SERVICING, REPAIR OR CORRECTION.
+
+  16. Limitation of Liability.
+
+  IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING
+WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MODIFIES AND/OR CONVEYS
+THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU FOR DAMAGES, INCLUDING ANY
+GENERAL, SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING OUT OF THE
+USE OR INABILITY TO USE THE PROGRAM (INCLUDING BUT NOT LIMITED TO LOSS OF
+DATA OR DATA BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY YOU OR THIRD
+PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE WITH ANY OTHER PROGRAMS),
+EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE POSSIBILITY OF
+SUCH DAMAGES.
+
+  17. Interpretation of Sections 15 and 16.
+
+  If the disclaimer of warranty and limitation of liability provided
+above cannot be given local legal effect according to their terms,
+reviewing courts shall apply local law that most closely approximates
+an absolute waiver of all civil liability in connection with the
+Program, unless a warranty or assumption of liability accompanies a
+copy of the Program in return for a fee.
+
+                     END OF TERMS AND CONDITIONS
+
+            How to Apply These Terms to Your New Programs
+
+  If you develop a new program, and you want it to be of the greatest
+possible use to the public, the best way to achieve this is to make it
+free software which everyone can redistribute and change under these terms.
+
+  To do so, attach the following notices to the program.  It is safest
+to attach them to the start of each source file to most effectively
+state the exclusion of warranty; and each file should have at least
+the "copyright" line and a pointer to where the full notice is found.
+
+    <one line to give the program's name and a brief idea of what it does.>
+    Copyright (C) <year>  <name of author>
+
+    This program is free software: you can redistribute it and/or modify
+    it under the terms of the GNU General Public License as published by
+    the Free Software Foundation, either version 3 of the License, or
+    (at your option) any later version.
+
+    This program is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    GNU General Public License for more details.
+
+    You should have received a copy of the GNU General Public License
+    along with this program.  If not, see <http://www.gnu.org/licenses/>.
+
+Also add information on how to contact you by electronic and paper mail.
+
+  If the program does terminal interaction, make it output a short
+notice like this when it starts in an interactive mode:
+
+    <program>  Copyright (C) <year>  <name of author>
+    This program comes with ABSOLUTELY NO WARRANTY; for details type `show w'.
+    This is free software, and you are welcome to redistribute it
+    under certain conditions; type `show c' for details.
+
+The hypothetical commands `show w' and `show c' should show the appropriate
+parts of the General Public License.  Of course, your program's commands
+might be different; for a GUI interface, you would use an "about box".
+
+  You should also get your employer (if you work as a programmer) or school,
+if any, to sign a "copyright disclaimer" for the program, if necessary.
+For more information on this, and how to apply and follow the GNU GPL, see
+<http://www.gnu.org/licenses/>.
+
+  The GNU General Public License does not permit incorporating your program
+into proprietary programs.  If your program is a subroutine library, you
+may consider it more useful to permit linking proprietary applications with
+the library.  If this is what you want to do, use the GNU Lesser General
+Public License instead of this License.  But first, please read
+<http://www.gnu.org/philosophy/why-not-lgpl.html>.
+
+
+
+
+
+
+
+
+
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright [yyyy] [name of copyright owner]
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.
+
diff --git a/Makefile.am b/Makefile.am
new file mode 100644
index 0000000000..26f8d7e3e4
--- /dev/null
+++ b/Makefile.am
@@ -0,0 +1,45 @@
+
+if WANT_JANSSON
+JANSSON_INCLUDES= -I$(top_srcdir)/compat/jansson
+else
+JANSSON_INCLUDES=
+endif
+
+EXTRA_DIST		= autogen.sh README.txt LICENSE.txt \
+			  cudaminer.sln cudaminer.vcxproj cudaminer.vcxproj.filters \
+			  compat/gettimeofday.c compat/getopt/getopt_long.c cpuminer-config.h.in
+
+SUBDIRS		= compat
+
+bin_PROGRAMS	= ccminer
+
+ccminer_SOURCES		= elist.h miner.h compat.h \
+			  compat/inttypes.h compat/stdbool.h compat/unistd.h \
+			  compat/sys/time.h compat/getopt/getopt.h \
+			  cpu-miner.c util.c blake.c groestl.c hefty1.c keccak.c scrypt.c sha2.c \
+			  sph_blake.h sph_groestl.h sph_keccak.h sph_types.h \
+			  heavy.cu \
+			  cuda_blake512.cu cuda_blake512.h \
+			  cuda_combine.cu cuda_combine.h \
+			  cuda_groestl512.cu cuda_groestl512.h \
+			  cuda_hefty1.cu cuda_hefty1.h \
+			  cuda_keccak512.cu cuda_keccak512.h \
+			  cuda_sha256.cu cuda_sha256.h \
+			  cuda_fugue256.cu \
+			  fuguecoin.cpp fugue.c sph_fugue.h uint256.h
+			  
+
+ccminer_LDFLAGS		= $(PTHREAD_FLAGS) @CUDA_LDFLAGS@
+ccminer_LDADD		= @LIBCURL@ @JANSSON_LIBS@ @PTHREAD_LIBS@ @WS2_LIBS@ @CUDA_LIBS@ @OPENMP_CFLAGS@ @LIBS@
+ccminer_CPPFLAGS	= -msse2 @LIBCURL_CPPFLAGS@ @OPENMP_CFLAGS@ $(PTHREAD_FLAGS) -fno-strict-aliasing $(JANSSON_INCLUDES) -DSCRYPT_KECCAK512 -DSCRYPT_CHACHA -DSCRYPT_CHOOSE_COMPILETIME
+
+.cu.o:
+	$(NVCC) @CFLAGS@ -Xptxas "-abi=no -v" -arch=compute_20 --maxrregcount=63 --ptxas-options=-v $(JANSSON_INCLUDES) -o $@ -c $<
+
+## Thrust needs Compute 2.0 minimum
+#heavy.o: heavy.cu
+#	$(NVCC) @CFLAGS@ -Xptxas "-abi=no -v" -arch=compute_20 --maxrregcount=63 --ptxas-options=-v $(JANSSON_INCLUDES) -o $@ -c $<
+#
+#cuda_hefty1.o: cuda_hefty1.cu
+#	$(NVCC) @CFLAGS@ -Xptxas "-abi=no -v" -arch=compute_20 --maxrregcount=63 --ptxas-options=-v $(JANSSON_INCLUDES) -o $@ -c $<
+
diff --git a/Makefile.in b/Makefile.in
new file mode 100644
index 0000000000..963c70dfde
--- /dev/null
+++ b/Makefile.in
@@ -0,0 +1,1107 @@
+# Makefile.in generated by automake 1.13.3 from Makefile.am.
+# @configure_input@
+
+# Copyright (C) 1994-2013 Free Software Foundation, Inc.
+
+# This Makefile.in is free software; the Free Software Foundation
+# gives unlimited permission to copy and/or distribute it,
+# with or without modifications, as long as this notice is preserved.
+
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY, to the extent permitted by law; without
+# even the implied warranty of MERCHANTABILITY or FITNESS FOR A
+# PARTICULAR PURPOSE.
+
+@SET_MAKE@
+
+VPATH = @srcdir@
+am__is_gnu_make = test -n '$(MAKEFILE_LIST)' && test -n '$(MAKELEVEL)'
+am__make_running_with_option = \
+  case $${target_option-} in \
+      ?) ;; \
+      *) echo "am__make_running_with_option: internal error: invalid" \
+              "target option '$${target_option-}' specified" >&2; \
+         exit 1;; \
+  esac; \
+  has_opt=no; \
+  sane_makeflags=$$MAKEFLAGS; \
+  if $(am__is_gnu_make); then \
+    sane_makeflags=$$MFLAGS; \
+  else \
+    case $$MAKEFLAGS in \
+      *\\[\ \	]*) \
+        bs=\\; \
+        sane_makeflags=`printf '%s\n' "$$MAKEFLAGS" \
+          | sed "s/$$bs$$bs[$$bs $$bs	]*//g"`;; \
+    esac; \
+  fi; \
+  skip_next=no; \
+  strip_trailopt () \
+  { \
+    flg=`printf '%s\n' "$$flg" | sed "s/$$1.*$$//"`; \
+  }; \
+  for flg in $$sane_makeflags; do \
+    test $$skip_next = yes && { skip_next=no; continue; }; \
+    case $$flg in \
+      *=*|--*) continue;; \
+        -*I) strip_trailopt 'I'; skip_next=yes;; \
+      -*I?*) strip_trailopt 'I';; \
+        -*O) strip_trailopt 'O'; skip_next=yes;; \
+      -*O?*) strip_trailopt 'O';; \
+        -*l) strip_trailopt 'l'; skip_next=yes;; \
+      -*l?*) strip_trailopt 'l';; \
+      -[dEDm]) skip_next=yes;; \
+      -[JT]) skip_next=yes;; \
+    esac; \
+    case $$flg in \
+      *$$target_option*) has_opt=yes; break;; \
+    esac; \
+  done; \
+  test $$has_opt = yes
+am__make_dryrun = (target_option=n; $(am__make_running_with_option))
+am__make_keepgoing = (target_option=k; $(am__make_running_with_option))
+pkgdatadir = $(datadir)/@PACKAGE@
+pkgincludedir = $(includedir)/@PACKAGE@
+pkglibdir = $(libdir)/@PACKAGE@
+pkglibexecdir = $(libexecdir)/@PACKAGE@
+am__cd = CDPATH="$${ZSH_VERSION+.}$(PATH_SEPARATOR)" && cd
+install_sh_DATA = $(install_sh) -c -m 644
+install_sh_PROGRAM = $(install_sh) -c
+install_sh_SCRIPT = $(install_sh) -c
+INSTALL_HEADER = $(INSTALL_DATA)
+transform = $(program_transform_name)
+NORMAL_INSTALL = :
+PRE_INSTALL = :
+POST_INSTALL = :
+NORMAL_UNINSTALL = :
+PRE_UNINSTALL = :
+POST_UNINSTALL = :
+build_triplet = @build@
+host_triplet = @host@
+target_triplet = @target@
+bin_PROGRAMS = ccminer$(EXEEXT)
+subdir = .
+DIST_COMMON = INSTALL NEWS README AUTHORS ChangeLog \
+	$(srcdir)/Makefile.in $(srcdir)/Makefile.am \
+	$(top_srcdir)/configure $(am__configure_deps) \
+	$(srcdir)/cpuminer-config.h.in depcomp COPYING compile \
+	config.guess config.sub install-sh missing
+ACLOCAL_M4 = $(top_srcdir)/aclocal.m4
+am__aclocal_m4_deps = $(top_srcdir)/configure.ac
+am__configure_deps = $(am__aclocal_m4_deps) $(CONFIGURE_DEPENDENCIES) \
+	$(ACLOCAL_M4)
+am__CONFIG_DISTCLEAN_FILES = config.status config.cache config.log \
+ configure.lineno config.status.lineno
+mkinstalldirs = $(install_sh) -d
+CONFIG_HEADER = cpuminer-config.h
+CONFIG_CLEAN_FILES =
+CONFIG_CLEAN_VPATH_FILES =
+am__installdirs = "$(DESTDIR)$(bindir)"
+PROGRAMS = $(bin_PROGRAMS)
+am_ccminer_OBJECTS = ccminer-cpu-miner.$(OBJEXT) \
+	ccminer-util.$(OBJEXT) ccminer-blake.$(OBJEXT) \
+	ccminer-groestl.$(OBJEXT) ccminer-hefty1.$(OBJEXT) \
+	ccminer-keccak.$(OBJEXT) ccminer-scrypt.$(OBJEXT) \
+	ccminer-sha2.$(OBJEXT) heavy.$(OBJEXT) cuda_blake512.$(OBJEXT) \
+	cuda_combine.$(OBJEXT) cuda_groestl512.$(OBJEXT) \
+	cuda_hefty1.$(OBJEXT) cuda_keccak512.$(OBJEXT) \
+	cuda_sha256.$(OBJEXT) cuda_fugue256.$(OBJEXT) \
+	ccminer-fuguecoin.$(OBJEXT) ccminer-fugue.$(OBJEXT)
+ccminer_OBJECTS = $(am_ccminer_OBJECTS)
+ccminer_DEPENDENCIES =
+ccminer_LINK = $(CXXLD) $(AM_CXXFLAGS) $(CXXFLAGS) $(ccminer_LDFLAGS) \
+	$(LDFLAGS) -o $@
+AM_V_P = $(am__v_P_@AM_V@)
+am__v_P_ = $(am__v_P_@AM_DEFAULT_V@)
+am__v_P_0 = false
+am__v_P_1 = :
+AM_V_GEN = $(am__v_GEN_@AM_V@)
+am__v_GEN_ = $(am__v_GEN_@AM_DEFAULT_V@)
+am__v_GEN_0 = @echo "  GEN     " $@;
+am__v_GEN_1 = 
+AM_V_at = $(am__v_at_@AM_V@)
+am__v_at_ = $(am__v_at_@AM_DEFAULT_V@)
+am__v_at_0 = @
+am__v_at_1 = 
+DEFAULT_INCLUDES = -I.@am__isrc@
+depcomp = $(SHELL) $(top_srcdir)/depcomp
+am__depfiles_maybe = depfiles
+am__mv = mv -f
+AM_V_lt = $(am__v_lt_@AM_V@)
+am__v_lt_ = $(am__v_lt_@AM_DEFAULT_V@)
+am__v_lt_0 = --silent
+am__v_lt_1 = 
+COMPILE = $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) \
+	$(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS)
+AM_V_CC = $(am__v_CC_@AM_V@)
+am__v_CC_ = $(am__v_CC_@AM_DEFAULT_V@)
+am__v_CC_0 = @echo "  CC      " $@;
+am__v_CC_1 = 
+CCLD = $(CC)
+LINK = $(CCLD) $(AM_CFLAGS) $(CFLAGS) $(AM_LDFLAGS) $(LDFLAGS) -o $@
+AM_V_CCLD = $(am__v_CCLD_@AM_V@)
+am__v_CCLD_ = $(am__v_CCLD_@AM_DEFAULT_V@)
+am__v_CCLD_0 = @echo "  CCLD    " $@;
+am__v_CCLD_1 = 
+CXXCOMPILE = $(CXX) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) \
+	$(AM_CPPFLAGS) $(CPPFLAGS) $(AM_CXXFLAGS) $(CXXFLAGS)
+AM_V_CXX = $(am__v_CXX_@AM_V@)
+am__v_CXX_ = $(am__v_CXX_@AM_DEFAULT_V@)
+am__v_CXX_0 = @echo "  CXX     " $@;
+am__v_CXX_1 = 
+CXXLD = $(CXX)
+CXXLINK = $(CXXLD) $(AM_CXXFLAGS) $(CXXFLAGS) $(AM_LDFLAGS) $(LDFLAGS) \
+	-o $@
+AM_V_CXXLD = $(am__v_CXXLD_@AM_V@)
+am__v_CXXLD_ = $(am__v_CXXLD_@AM_DEFAULT_V@)
+am__v_CXXLD_0 = @echo "  CXXLD   " $@;
+am__v_CXXLD_1 = 
+SOURCES = $(ccminer_SOURCES)
+DIST_SOURCES = $(ccminer_SOURCES)
+RECURSIVE_TARGETS = all-recursive check-recursive cscopelist-recursive \
+	ctags-recursive dvi-recursive html-recursive info-recursive \
+	install-data-recursive install-dvi-recursive \
+	install-exec-recursive install-html-recursive \
+	install-info-recursive install-pdf-recursive \
+	install-ps-recursive install-recursive installcheck-recursive \
+	installdirs-recursive pdf-recursive ps-recursive \
+	tags-recursive uninstall-recursive
+am__can_run_installinfo = \
+  case $$AM_UPDATE_INFO_DIR in \
+    n|no|NO) false;; \
+    *) (install-info --version) >/dev/null 2>&1;; \
+  esac
+RECURSIVE_CLEAN_TARGETS = mostlyclean-recursive clean-recursive	\
+  distclean-recursive maintainer-clean-recursive
+am__recursive_targets = \
+  $(RECURSIVE_TARGETS) \
+  $(RECURSIVE_CLEAN_TARGETS) \
+  $(am__extra_recursive_targets)
+AM_RECURSIVE_TARGETS = $(am__recursive_targets:-recursive=) TAGS CTAGS \
+	cscope distdir dist dist-all distcheck
+am__tagged_files = $(HEADERS) $(SOURCES) $(TAGS_FILES) \
+	$(LISP)cpuminer-config.h.in
+# Read a list of newline-separated strings from the standard input,
+# and print each of them once, without duplicates.  Input order is
+# *not* preserved.
+am__uniquify_input = $(AWK) '\
+  BEGIN { nonempty = 0; } \
+  { items[$$0] = 1; nonempty = 1; } \
+  END { if (nonempty) { for (i in items) print i; }; } \
+'
+# Make sure the list of sources is unique.  This is necessary because,
+# e.g., the same source file might be shared among _SOURCES variables
+# for different programs/libraries.
+am__define_uniq_tagged_files = \
+  list='$(am__tagged_files)'; \
+  unique=`for i in $$list; do \
+    if test -f "$$i"; then echo $$i; else echo $(srcdir)/$$i; fi; \
+  done | $(am__uniquify_input)`
+ETAGS = etags
+CTAGS = ctags
+CSCOPE = cscope
+DIST_SUBDIRS = $(SUBDIRS)
+DISTFILES = $(DIST_COMMON) $(DIST_SOURCES) $(TEXINFOS) $(EXTRA_DIST)
+distdir = $(PACKAGE)-$(VERSION)
+top_distdir = $(distdir)
+am__remove_distdir = \
+  if test -d "$(distdir)"; then \
+    find "$(distdir)" -type d ! -perm -200 -exec chmod u+w {} ';' \
+      && rm -rf "$(distdir)" \
+      || { sleep 5 && rm -rf "$(distdir)"; }; \
+  else :; fi
+am__post_remove_distdir = $(am__remove_distdir)
+am__relativize = \
+  dir0=`pwd`; \
+  sed_first='s,^\([^/]*\)/.*$$,\1,'; \
+  sed_rest='s,^[^/]*/*,,'; \
+  sed_last='s,^.*/\([^/]*\)$$,\1,'; \
+  sed_butlast='s,/*[^/]*$$,,'; \
+  while test -n "$$dir1"; do \
+    first=`echo "$$dir1" | sed -e "$$sed_first"`; \
+    if test "$$first" != "."; then \
+      if test "$$first" = ".."; then \
+        dir2=`echo "$$dir0" | sed -e "$$sed_last"`/"$$dir2"; \
+        dir0=`echo "$$dir0" | sed -e "$$sed_butlast"`; \
+      else \
+        first2=`echo "$$dir2" | sed -e "$$sed_first"`; \
+        if test "$$first2" = "$$first"; then \
+          dir2=`echo "$$dir2" | sed -e "$$sed_rest"`; \
+        else \
+          dir2="../$$dir2"; \
+        fi; \
+        dir0="$$dir0"/"$$first"; \
+      fi; \
+    fi; \
+    dir1=`echo "$$dir1" | sed -e "$$sed_rest"`; \
+  done; \
+  reldir="$$dir2"
+DIST_ARCHIVES = $(distdir).tar.gz
+GZIP_ENV = --best
+DIST_TARGETS = dist-gzip
+distuninstallcheck_listfiles = find . -type f -print
+am__distuninstallcheck_listfiles = $(distuninstallcheck_listfiles) \
+  | sed 's|^\./|$(prefix)/|' | grep -v '$(infodir)/dir$$'
+distcleancheck_listfiles = find . -type f -print
+ACLOCAL = @ACLOCAL@
+ALLOCA = @ALLOCA@
+AMTAR = @AMTAR@
+AM_DEFAULT_VERBOSITY = @AM_DEFAULT_VERBOSITY@
+AUTOCONF = @AUTOCONF@
+AUTOHEADER = @AUTOHEADER@
+AUTOMAKE = @AUTOMAKE@
+AWK = @AWK@
+CC = @CC@
+CCAS = @CCAS@
+CCASDEPMODE = @CCASDEPMODE@
+CCASFLAGS = @CCASFLAGS@
+CCDEPMODE = @CCDEPMODE@
+CFLAGS = @CFLAGS@
+CPP = @CPP@
+CPPFLAGS = @CPPFLAGS@
+CUDA_CFLAGS = @CUDA_CFLAGS@
+CUDA_LDFLAGS = @CUDA_LDFLAGS@
+CUDA_LIBS = @CUDA_LIBS@
+CXX = @CXX@
+CXXDEPMODE = @CXXDEPMODE@
+CXXFLAGS = @CXXFLAGS@
+CYGPATH_W = @CYGPATH_W@
+DEFS = @DEFS@
+DEPDIR = @DEPDIR@
+ECHO_C = @ECHO_C@
+ECHO_N = @ECHO_N@
+ECHO_T = @ECHO_T@
+EGREP = @EGREP@
+EXEEXT = @EXEEXT@
+GREP = @GREP@
+INSTALL = @INSTALL@
+INSTALL_DATA = @INSTALL_DATA@
+INSTALL_PROGRAM = @INSTALL_PROGRAM@
+INSTALL_SCRIPT = @INSTALL_SCRIPT@
+INSTALL_STRIP_PROGRAM = @INSTALL_STRIP_PROGRAM@
+JANSSON_LIBS = @JANSSON_LIBS@
+LDFLAGS = @LDFLAGS@
+LIBCURL = @LIBCURL@
+LIBCURL_CPPFLAGS = @LIBCURL_CPPFLAGS@
+LIBOBJS = @LIBOBJS@
+LIBS = @LIBS@
+LTLIBOBJS = @LTLIBOBJS@
+MAINT = @MAINT@
+MAKEINFO = @MAKEINFO@
+MKDIR_P = @MKDIR_P@
+NVCC = @NVCC@
+OBJEXT = @OBJEXT@
+OPENMP_CFLAGS = @OPENMP_CFLAGS@
+PACKAGE = @PACKAGE@
+PACKAGE_BUGREPORT = @PACKAGE_BUGREPORT@
+PACKAGE_NAME = @PACKAGE_NAME@
+PACKAGE_STRING = @PACKAGE_STRING@
+PACKAGE_TARNAME = @PACKAGE_TARNAME@
+PACKAGE_URL = @PACKAGE_URL@
+PACKAGE_VERSION = @PACKAGE_VERSION@
+PATH_SEPARATOR = @PATH_SEPARATOR@
+PTHREAD_FLAGS = @PTHREAD_FLAGS@
+PTHREAD_LIBS = @PTHREAD_LIBS@
+RANLIB = @RANLIB@
+SET_MAKE = @SET_MAKE@
+SHELL = @SHELL@
+STRIP = @STRIP@
+VERSION = @VERSION@
+WS2_LIBS = @WS2_LIBS@
+_libcurl_config = @_libcurl_config@
+abs_builddir = @abs_builddir@
+abs_srcdir = @abs_srcdir@
+abs_top_builddir = @abs_top_builddir@
+abs_top_srcdir = @abs_top_srcdir@
+ac_ct_CC = @ac_ct_CC@
+ac_ct_CXX = @ac_ct_CXX@
+am__include = @am__include@
+am__leading_dot = @am__leading_dot@
+am__quote = @am__quote@
+am__tar = @am__tar@
+am__untar = @am__untar@
+bindir = @bindir@
+build = @build@
+build_alias = @build_alias@
+build_cpu = @build_cpu@
+build_os = @build_os@
+build_vendor = @build_vendor@
+builddir = @builddir@
+datadir = @datadir@
+datarootdir = @datarootdir@
+docdir = @docdir@
+dvidir = @dvidir@
+exec_prefix = @exec_prefix@
+host = @host@
+host_alias = @host_alias@
+host_cpu = @host_cpu@
+host_os = @host_os@
+host_vendor = @host_vendor@
+htmldir = @htmldir@
+includedir = @includedir@
+infodir = @infodir@
+install_sh = @install_sh@
+libdir = @libdir@
+libexecdir = @libexecdir@
+localedir = @localedir@
+localstatedir = @localstatedir@
+mandir = @mandir@
+mkdir_p = @mkdir_p@
+oldincludedir = @oldincludedir@
+pdfdir = @pdfdir@
+prefix = @prefix@
+program_transform_name = @program_transform_name@
+psdir = @psdir@
+sbindir = @sbindir@
+sharedstatedir = @sharedstatedir@
+srcdir = @srcdir@
+sysconfdir = @sysconfdir@
+target = @target@
+target_alias = @target_alias@
+target_cpu = @target_cpu@
+target_os = @target_os@
+target_vendor = @target_vendor@
+top_build_prefix = @top_build_prefix@
+top_builddir = @top_builddir@
+top_srcdir = @top_srcdir@
+@WANT_JANSSON_FALSE@JANSSON_INCLUDES = 
+@WANT_JANSSON_TRUE@JANSSON_INCLUDES = -I$(top_srcdir)/compat/jansson
+EXTRA_DIST = autogen.sh README.txt LICENSE.txt \
+			  cudaminer.sln cudaminer.vcxproj cudaminer.vcxproj.filters \
+			  compat/gettimeofday.c compat/getopt/getopt_long.c cpuminer-config.h.in
+
+SUBDIRS = compat
+ccminer_SOURCES = elist.h miner.h compat.h \
+			  compat/inttypes.h compat/stdbool.h compat/unistd.h \
+			  compat/sys/time.h compat/getopt/getopt.h \
+			  cpu-miner.c util.c blake.c groestl.c hefty1.c keccak.c scrypt.c sha2.c \
+			  sph_blake.h sph_groestl.h sph_keccak.h sph_types.h \
+			  heavy.cu \
+			  cuda_blake512.cu cuda_blake512.h \
+			  cuda_combine.cu cuda_combine.h \
+			  cuda_groestl512.cu cuda_groestl512.h \
+			  cuda_hefty1.cu cuda_hefty1.h \
+			  cuda_keccak512.cu cuda_keccak512.h \
+			  cuda_sha256.cu cuda_sha256.h \
+			  cuda_fugue256.cu \
+			  fuguecoin.cpp fugue.c sph_fugue.h uint256.h
+
+ccminer_LDFLAGS = $(PTHREAD_FLAGS) @CUDA_LDFLAGS@
+ccminer_LDADD = @LIBCURL@ @JANSSON_LIBS@ @PTHREAD_LIBS@ @WS2_LIBS@ @CUDA_LIBS@ @OPENMP_CFLAGS@ @LIBS@
+ccminer_CPPFLAGS = -msse2 @LIBCURL_CPPFLAGS@ @OPENMP_CFLAGS@ $(PTHREAD_FLAGS) -fno-strict-aliasing $(JANSSON_INCLUDES) -DSCRYPT_KECCAK512 -DSCRYPT_CHACHA -DSCRYPT_CHOOSE_COMPILETIME
+all: cpuminer-config.h
+	$(MAKE) $(AM_MAKEFLAGS) all-recursive
+
+.SUFFIXES:
+.SUFFIXES: .c .cpp .cu .o .obj
+am--refresh: Makefile
+	@:
+$(srcdir)/Makefile.in: @MAINTAINER_MODE_TRUE@ $(srcdir)/Makefile.am  $(am__configure_deps)
+	@for dep in $?; do \
+	  case '$(am__configure_deps)' in \
+	    *$$dep*) \
+	      echo ' cd $(srcdir) && $(AUTOMAKE) --gnu'; \
+	      $(am__cd) $(srcdir) && $(AUTOMAKE) --gnu \
+		&& exit 0; \
+	      exit 1;; \
+	  esac; \
+	done; \
+	echo ' cd $(top_srcdir) && $(AUTOMAKE) --gnu Makefile'; \
+	$(am__cd) $(top_srcdir) && \
+	  $(AUTOMAKE) --gnu Makefile
+.PRECIOUS: Makefile
+Makefile: $(srcdir)/Makefile.in $(top_builddir)/config.status
+	@case '$?' in \
+	  *config.status*) \
+	    echo ' $(SHELL) ./config.status'; \
+	    $(SHELL) ./config.status;; \
+	  *) \
+	    echo ' cd $(top_builddir) && $(SHELL) ./config.status $@ $(am__depfiles_maybe)'; \
+	    cd $(top_builddir) && $(SHELL) ./config.status $@ $(am__depfiles_maybe);; \
+	esac;
+
+$(top_builddir)/config.status: $(top_srcdir)/configure $(CONFIG_STATUS_DEPENDENCIES)
+	$(SHELL) ./config.status --recheck
+
+$(top_srcdir)/configure: @MAINTAINER_MODE_TRUE@ $(am__configure_deps)
+	$(am__cd) $(srcdir) && $(AUTOCONF)
+$(ACLOCAL_M4): @MAINTAINER_MODE_TRUE@ $(am__aclocal_m4_deps)
+	$(am__cd) $(srcdir) && $(ACLOCAL) $(ACLOCAL_AMFLAGS)
+$(am__aclocal_m4_deps):
+
+cpuminer-config.h: stamp-h1
+	@if test ! -f $@; then rm -f stamp-h1; else :; fi
+	@if test ! -f $@; then $(MAKE) $(AM_MAKEFLAGS) stamp-h1; else :; fi
+
+stamp-h1: $(srcdir)/cpuminer-config.h.in $(top_builddir)/config.status
+	@rm -f stamp-h1
+	cd $(top_builddir) && $(SHELL) ./config.status cpuminer-config.h
+$(srcdir)/cpuminer-config.h.in: @MAINTAINER_MODE_TRUE@ $(am__configure_deps) 
+	($(am__cd) $(top_srcdir) && $(AUTOHEADER))
+	rm -f stamp-h1
+	touch $@
+
+distclean-hdr:
+	-rm -f cpuminer-config.h stamp-h1
+install-binPROGRAMS: $(bin_PROGRAMS)
+	@$(NORMAL_INSTALL)
+	@list='$(bin_PROGRAMS)'; test -n "$(bindir)" || list=; \
+	if test -n "$$list"; then \
+	  echo " $(MKDIR_P) '$(DESTDIR)$(bindir)'"; \
+	  $(MKDIR_P) "$(DESTDIR)$(bindir)" || exit 1; \
+	fi; \
+	for p in $$list; do echo "$$p $$p"; done | \
+	sed 's/$(EXEEXT)$$//' | \
+	while read p p1; do if test -f $$p \
+	  ; then echo "$$p"; echo "$$p"; else :; fi; \
+	done | \
+	sed -e 'p;s,.*/,,;n;h' \
+	    -e 's|.*|.|' \
+	    -e 'p;x;s,.*/,,;s/$(EXEEXT)$$//;$(transform);s/$$/$(EXEEXT)/' | \
+	sed 'N;N;N;s,\n, ,g' | \
+	$(AWK) 'BEGIN { files["."] = ""; dirs["."] = 1 } \
+	  { d=$$3; if (dirs[d] != 1) { print "d", d; dirs[d] = 1 } \
+	    if ($$2 == $$4) files[d] = files[d] " " $$1; \
+	    else { print "f", $$3 "/" $$4, $$1; } } \
+	  END { for (d in files) print "f", d, files[d] }' | \
+	while read type dir files; do \
+	    if test "$$dir" = .; then dir=; else dir=/$$dir; fi; \
+	    test -z "$$files" || { \
+	      echo " $(INSTALL_PROGRAM_ENV) $(INSTALL_PROGRAM) $$files '$(DESTDIR)$(bindir)$$dir'"; \
+	      $(INSTALL_PROGRAM_ENV) $(INSTALL_PROGRAM) $$files "$(DESTDIR)$(bindir)$$dir" || exit $$?; \
+	    } \
+	; done
+
+uninstall-binPROGRAMS:
+	@$(NORMAL_UNINSTALL)
+	@list='$(bin_PROGRAMS)'; test -n "$(bindir)" || list=; \
+	files=`for p in $$list; do echo "$$p"; done | \
+	  sed -e 'h;s,^.*/,,;s/$(EXEEXT)$$//;$(transform)' \
+	      -e 's/$$/$(EXEEXT)/' \
+	`; \
+	test -n "$$list" || exit 0; \
+	echo " ( cd '$(DESTDIR)$(bindir)' && rm -f" $$files ")"; \
+	cd "$(DESTDIR)$(bindir)" && rm -f $$files
+
+clean-binPROGRAMS:
+	-test -z "$(bin_PROGRAMS)" || rm -f $(bin_PROGRAMS)
+
+ccminer$(EXEEXT): $(ccminer_OBJECTS) $(ccminer_DEPENDENCIES) $(EXTRA_ccminer_DEPENDENCIES) 
+	@rm -f ccminer$(EXEEXT)
+	$(AM_V_CXXLD)$(ccminer_LINK) $(ccminer_OBJECTS) $(ccminer_LDADD) $(LIBS)
+
+mostlyclean-compile:
+	-rm -f *.$(OBJEXT)
+
+distclean-compile:
+	-rm -f *.tab.c
+
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/ccminer-blake.Po@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/ccminer-cpu-miner.Po@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/ccminer-fugue.Po@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/ccminer-fuguecoin.Po@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/ccminer-groestl.Po@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/ccminer-hefty1.Po@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/ccminer-keccak.Po@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/ccminer-scrypt.Po@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/ccminer-sha2.Po@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/ccminer-util.Po@am__quote@
+
+.c.o:
+@am__fastdepCC_TRUE@	$(AM_V_CC)$(COMPILE) -MT $@ -MD -MP -MF $(DEPDIR)/$*.Tpo -c -o $@ $<
+@am__fastdepCC_TRUE@	$(AM_V_at)$(am__mv) $(DEPDIR)/$*.Tpo $(DEPDIR)/$*.Po
+@AMDEP_TRUE@@am__fastdepCC_FALSE@	$(AM_V_CC)source='$<' object='$@' libtool=no @AMDEPBACKSLASH@
+@AMDEP_TRUE@@am__fastdepCC_FALSE@	DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
+@am__fastdepCC_FALSE@	$(AM_V_CC@am__nodep@)$(COMPILE) -c $<
+
+.c.obj:
+@am__fastdepCC_TRUE@	$(AM_V_CC)$(COMPILE) -MT $@ -MD -MP -MF $(DEPDIR)/$*.Tpo -c -o $@ `$(CYGPATH_W) '$<'`
+@am__fastdepCC_TRUE@	$(AM_V_at)$(am__mv) $(DEPDIR)/$*.Tpo $(DEPDIR)/$*.Po
+@AMDEP_TRUE@@am__fastdepCC_FALSE@	$(AM_V_CC)source='$<' object='$@' libtool=no @AMDEPBACKSLASH@
+@AMDEP_TRUE@@am__fastdepCC_FALSE@	DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
+@am__fastdepCC_FALSE@	$(AM_V_CC@am__nodep@)$(COMPILE) -c `$(CYGPATH_W) '$<'`
+
+ccminer-cpu-miner.o: cpu-miner.c
+@am__fastdepCC_TRUE@	$(AM_V_CC)$(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(ccminer_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -MT ccminer-cpu-miner.o -MD -MP -MF $(DEPDIR)/ccminer-cpu-miner.Tpo -c -o ccminer-cpu-miner.o `test -f 'cpu-miner.c' || echo '$(srcdir)/'`cpu-miner.c
+@am__fastdepCC_TRUE@	$(AM_V_at)$(am__mv) $(DEPDIR)/ccminer-cpu-miner.Tpo $(DEPDIR)/ccminer-cpu-miner.Po
+@AMDEP_TRUE@@am__fastdepCC_FALSE@	$(AM_V_CC)source='cpu-miner.c' object='ccminer-cpu-miner.o' libtool=no @AMDEPBACKSLASH@
+@AMDEP_TRUE@@am__fastdepCC_FALSE@	DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
+@am__fastdepCC_FALSE@	$(AM_V_CC@am__nodep@)$(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(ccminer_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -c -o ccminer-cpu-miner.o `test -f 'cpu-miner.c' || echo '$(srcdir)/'`cpu-miner.c
+
+ccminer-cpu-miner.obj: cpu-miner.c
+@am__fastdepCC_TRUE@	$(AM_V_CC)$(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(ccminer_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -MT ccminer-cpu-miner.obj -MD -MP -MF $(DEPDIR)/ccminer-cpu-miner.Tpo -c -o ccminer-cpu-miner.obj `if test -f 'cpu-miner.c'; then $(CYGPATH_W) 'cpu-miner.c'; else $(CYGPATH_W) '$(srcdir)/cpu-miner.c'; fi`
+@am__fastdepCC_TRUE@	$(AM_V_at)$(am__mv) $(DEPDIR)/ccminer-cpu-miner.Tpo $(DEPDIR)/ccminer-cpu-miner.Po
+@AMDEP_TRUE@@am__fastdepCC_FALSE@	$(AM_V_CC)source='cpu-miner.c' object='ccminer-cpu-miner.obj' libtool=no @AMDEPBACKSLASH@
+@AMDEP_TRUE@@am__fastdepCC_FALSE@	DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
+@am__fastdepCC_FALSE@	$(AM_V_CC@am__nodep@)$(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(ccminer_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -c -o ccminer-cpu-miner.obj `if test -f 'cpu-miner.c'; then $(CYGPATH_W) 'cpu-miner.c'; else $(CYGPATH_W) '$(srcdir)/cpu-miner.c'; fi`
+
+ccminer-util.o: util.c
+@am__fastdepCC_TRUE@	$(AM_V_CC)$(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(ccminer_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -MT ccminer-util.o -MD -MP -MF $(DEPDIR)/ccminer-util.Tpo -c -o ccminer-util.o `test -f 'util.c' || echo '$(srcdir)/'`util.c
+@am__fastdepCC_TRUE@	$(AM_V_at)$(am__mv) $(DEPDIR)/ccminer-util.Tpo $(DEPDIR)/ccminer-util.Po
+@AMDEP_TRUE@@am__fastdepCC_FALSE@	$(AM_V_CC)source='util.c' object='ccminer-util.o' libtool=no @AMDEPBACKSLASH@
+@AMDEP_TRUE@@am__fastdepCC_FALSE@	DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
+@am__fastdepCC_FALSE@	$(AM_V_CC@am__nodep@)$(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(ccminer_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -c -o ccminer-util.o `test -f 'util.c' || echo '$(srcdir)/'`util.c
+
+ccminer-util.obj: util.c
+@am__fastdepCC_TRUE@	$(AM_V_CC)$(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(ccminer_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -MT ccminer-util.obj -MD -MP -MF $(DEPDIR)/ccminer-util.Tpo -c -o ccminer-util.obj `if test -f 'util.c'; then $(CYGPATH_W) 'util.c'; else $(CYGPATH_W) '$(srcdir)/util.c'; fi`
+@am__fastdepCC_TRUE@	$(AM_V_at)$(am__mv) $(DEPDIR)/ccminer-util.Tpo $(DEPDIR)/ccminer-util.Po
+@AMDEP_TRUE@@am__fastdepCC_FALSE@	$(AM_V_CC)source='util.c' object='ccminer-util.obj' libtool=no @AMDEPBACKSLASH@
+@AMDEP_TRUE@@am__fastdepCC_FALSE@	DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
+@am__fastdepCC_FALSE@	$(AM_V_CC@am__nodep@)$(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(ccminer_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -c -o ccminer-util.obj `if test -f 'util.c'; then $(CYGPATH_W) 'util.c'; else $(CYGPATH_W) '$(srcdir)/util.c'; fi`
+
+ccminer-blake.o: blake.c
+@am__fastdepCC_TRUE@	$(AM_V_CC)$(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(ccminer_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -MT ccminer-blake.o -MD -MP -MF $(DEPDIR)/ccminer-blake.Tpo -c -o ccminer-blake.o `test -f 'blake.c' || echo '$(srcdir)/'`blake.c
+@am__fastdepCC_TRUE@	$(AM_V_at)$(am__mv) $(DEPDIR)/ccminer-blake.Tpo $(DEPDIR)/ccminer-blake.Po
+@AMDEP_TRUE@@am__fastdepCC_FALSE@	$(AM_V_CC)source='blake.c' object='ccminer-blake.o' libtool=no @AMDEPBACKSLASH@
+@AMDEP_TRUE@@am__fastdepCC_FALSE@	DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
+@am__fastdepCC_FALSE@	$(AM_V_CC@am__nodep@)$(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(ccminer_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -c -o ccminer-blake.o `test -f 'blake.c' || echo '$(srcdir)/'`blake.c
+
+ccminer-blake.obj: blake.c
+@am__fastdepCC_TRUE@	$(AM_V_CC)$(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(ccminer_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -MT ccminer-blake.obj -MD -MP -MF $(DEPDIR)/ccminer-blake.Tpo -c -o ccminer-blake.obj `if test -f 'blake.c'; then $(CYGPATH_W) 'blake.c'; else $(CYGPATH_W) '$(srcdir)/blake.c'; fi`
+@am__fastdepCC_TRUE@	$(AM_V_at)$(am__mv) $(DEPDIR)/ccminer-blake.Tpo $(DEPDIR)/ccminer-blake.Po
+@AMDEP_TRUE@@am__fastdepCC_FALSE@	$(AM_V_CC)source='blake.c' object='ccminer-blake.obj' libtool=no @AMDEPBACKSLASH@
+@AMDEP_TRUE@@am__fastdepCC_FALSE@	DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
+@am__fastdepCC_FALSE@	$(AM_V_CC@am__nodep@)$(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(ccminer_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -c -o ccminer-blake.obj `if test -f 'blake.c'; then $(CYGPATH_W) 'blake.c'; else $(CYGPATH_W) '$(srcdir)/blake.c'; fi`
+
+ccminer-groestl.o: groestl.c
+@am__fastdepCC_TRUE@	$(AM_V_CC)$(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(ccminer_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -MT ccminer-groestl.o -MD -MP -MF $(DEPDIR)/ccminer-groestl.Tpo -c -o ccminer-groestl.o `test -f 'groestl.c' || echo '$(srcdir)/'`groestl.c
+@am__fastdepCC_TRUE@	$(AM_V_at)$(am__mv) $(DEPDIR)/ccminer-groestl.Tpo $(DEPDIR)/ccminer-groestl.Po
+@AMDEP_TRUE@@am__fastdepCC_FALSE@	$(AM_V_CC)source='groestl.c' object='ccminer-groestl.o' libtool=no @AMDEPBACKSLASH@
+@AMDEP_TRUE@@am__fastdepCC_FALSE@	DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
+@am__fastdepCC_FALSE@	$(AM_V_CC@am__nodep@)$(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(ccminer_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -c -o ccminer-groestl.o `test -f 'groestl.c' || echo '$(srcdir)/'`groestl.c
+
+ccminer-groestl.obj: groestl.c
+@am__fastdepCC_TRUE@	$(AM_V_CC)$(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(ccminer_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -MT ccminer-groestl.obj -MD -MP -MF $(DEPDIR)/ccminer-groestl.Tpo -c -o ccminer-groestl.obj `if test -f 'groestl.c'; then $(CYGPATH_W) 'groestl.c'; else $(CYGPATH_W) '$(srcdir)/groestl.c'; fi`
+@am__fastdepCC_TRUE@	$(AM_V_at)$(am__mv) $(DEPDIR)/ccminer-groestl.Tpo $(DEPDIR)/ccminer-groestl.Po
+@AMDEP_TRUE@@am__fastdepCC_FALSE@	$(AM_V_CC)source='groestl.c' object='ccminer-groestl.obj' libtool=no @AMDEPBACKSLASH@
+@AMDEP_TRUE@@am__fastdepCC_FALSE@	DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
+@am__fastdepCC_FALSE@	$(AM_V_CC@am__nodep@)$(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(ccminer_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -c -o ccminer-groestl.obj `if test -f 'groestl.c'; then $(CYGPATH_W) 'groestl.c'; else $(CYGPATH_W) '$(srcdir)/groestl.c'; fi`
+
+ccminer-hefty1.o: hefty1.c
+@am__fastdepCC_TRUE@	$(AM_V_CC)$(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(ccminer_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -MT ccminer-hefty1.o -MD -MP -MF $(DEPDIR)/ccminer-hefty1.Tpo -c -o ccminer-hefty1.o `test -f 'hefty1.c' || echo '$(srcdir)/'`hefty1.c
+@am__fastdepCC_TRUE@	$(AM_V_at)$(am__mv) $(DEPDIR)/ccminer-hefty1.Tpo $(DEPDIR)/ccminer-hefty1.Po
+@AMDEP_TRUE@@am__fastdepCC_FALSE@	$(AM_V_CC)source='hefty1.c' object='ccminer-hefty1.o' libtool=no @AMDEPBACKSLASH@
+@AMDEP_TRUE@@am__fastdepCC_FALSE@	DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
+@am__fastdepCC_FALSE@	$(AM_V_CC@am__nodep@)$(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(ccminer_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -c -o ccminer-hefty1.o `test -f 'hefty1.c' || echo '$(srcdir)/'`hefty1.c
+
+ccminer-hefty1.obj: hefty1.c
+@am__fastdepCC_TRUE@	$(AM_V_CC)$(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(ccminer_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -MT ccminer-hefty1.obj -MD -MP -MF $(DEPDIR)/ccminer-hefty1.Tpo -c -o ccminer-hefty1.obj `if test -f 'hefty1.c'; then $(CYGPATH_W) 'hefty1.c'; else $(CYGPATH_W) '$(srcdir)/hefty1.c'; fi`
+@am__fastdepCC_TRUE@	$(AM_V_at)$(am__mv) $(DEPDIR)/ccminer-hefty1.Tpo $(DEPDIR)/ccminer-hefty1.Po
+@AMDEP_TRUE@@am__fastdepCC_FALSE@	$(AM_V_CC)source='hefty1.c' object='ccminer-hefty1.obj' libtool=no @AMDEPBACKSLASH@
+@AMDEP_TRUE@@am__fastdepCC_FALSE@	DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
+@am__fastdepCC_FALSE@	$(AM_V_CC@am__nodep@)$(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(ccminer_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -c -o ccminer-hefty1.obj `if test -f 'hefty1.c'; then $(CYGPATH_W) 'hefty1.c'; else $(CYGPATH_W) '$(srcdir)/hefty1.c'; fi`
+
+ccminer-keccak.o: keccak.c
+@am__fastdepCC_TRUE@	$(AM_V_CC)$(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(ccminer_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -MT ccminer-keccak.o -MD -MP -MF $(DEPDIR)/ccminer-keccak.Tpo -c -o ccminer-keccak.o `test -f 'keccak.c' || echo '$(srcdir)/'`keccak.c
+@am__fastdepCC_TRUE@	$(AM_V_at)$(am__mv) $(DEPDIR)/ccminer-keccak.Tpo $(DEPDIR)/ccminer-keccak.Po
+@AMDEP_TRUE@@am__fastdepCC_FALSE@	$(AM_V_CC)source='keccak.c' object='ccminer-keccak.o' libtool=no @AMDEPBACKSLASH@
+@AMDEP_TRUE@@am__fastdepCC_FALSE@	DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
+@am__fastdepCC_FALSE@	$(AM_V_CC@am__nodep@)$(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(ccminer_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -c -o ccminer-keccak.o `test -f 'keccak.c' || echo '$(srcdir)/'`keccak.c
+
+ccminer-keccak.obj: keccak.c
+@am__fastdepCC_TRUE@	$(AM_V_CC)$(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(ccminer_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -MT ccminer-keccak.obj -MD -MP -MF $(DEPDIR)/ccminer-keccak.Tpo -c -o ccminer-keccak.obj `if test -f 'keccak.c'; then $(CYGPATH_W) 'keccak.c'; else $(CYGPATH_W) '$(srcdir)/keccak.c'; fi`
+@am__fastdepCC_TRUE@	$(AM_V_at)$(am__mv) $(DEPDIR)/ccminer-keccak.Tpo $(DEPDIR)/ccminer-keccak.Po
+@AMDEP_TRUE@@am__fastdepCC_FALSE@	$(AM_V_CC)source='keccak.c' object='ccminer-keccak.obj' libtool=no @AMDEPBACKSLASH@
+@AMDEP_TRUE@@am__fastdepCC_FALSE@	DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
+@am__fastdepCC_FALSE@	$(AM_V_CC@am__nodep@)$(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(ccminer_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -c -o ccminer-keccak.obj `if test -f 'keccak.c'; then $(CYGPATH_W) 'keccak.c'; else $(CYGPATH_W) '$(srcdir)/keccak.c'; fi`
+
+ccminer-scrypt.o: scrypt.c
+@am__fastdepCC_TRUE@	$(AM_V_CC)$(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(ccminer_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -MT ccminer-scrypt.o -MD -MP -MF $(DEPDIR)/ccminer-scrypt.Tpo -c -o ccminer-scrypt.o `test -f 'scrypt.c' || echo '$(srcdir)/'`scrypt.c
+@am__fastdepCC_TRUE@	$(AM_V_at)$(am__mv) $(DEPDIR)/ccminer-scrypt.Tpo $(DEPDIR)/ccminer-scrypt.Po
+@AMDEP_TRUE@@am__fastdepCC_FALSE@	$(AM_V_CC)source='scrypt.c' object='ccminer-scrypt.o' libtool=no @AMDEPBACKSLASH@
+@AMDEP_TRUE@@am__fastdepCC_FALSE@	DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
+@am__fastdepCC_FALSE@	$(AM_V_CC@am__nodep@)$(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(ccminer_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -c -o ccminer-scrypt.o `test -f 'scrypt.c' || echo '$(srcdir)/'`scrypt.c
+
+ccminer-scrypt.obj: scrypt.c
+@am__fastdepCC_TRUE@	$(AM_V_CC)$(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(ccminer_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -MT ccminer-scrypt.obj -MD -MP -MF $(DEPDIR)/ccminer-scrypt.Tpo -c -o ccminer-scrypt.obj `if test -f 'scrypt.c'; then $(CYGPATH_W) 'scrypt.c'; else $(CYGPATH_W) '$(srcdir)/scrypt.c'; fi`
+@am__fastdepCC_TRUE@	$(AM_V_at)$(am__mv) $(DEPDIR)/ccminer-scrypt.Tpo $(DEPDIR)/ccminer-scrypt.Po
+@AMDEP_TRUE@@am__fastdepCC_FALSE@	$(AM_V_CC)source='scrypt.c' object='ccminer-scrypt.obj' libtool=no @AMDEPBACKSLASH@
+@AMDEP_TRUE@@am__fastdepCC_FALSE@	DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
+@am__fastdepCC_FALSE@	$(AM_V_CC@am__nodep@)$(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(ccminer_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -c -o ccminer-scrypt.obj `if test -f 'scrypt.c'; then $(CYGPATH_W) 'scrypt.c'; else $(CYGPATH_W) '$(srcdir)/scrypt.c'; fi`
+
+ccminer-sha2.o: sha2.c
+@am__fastdepCC_TRUE@	$(AM_V_CC)$(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(ccminer_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -MT ccminer-sha2.o -MD -MP -MF $(DEPDIR)/ccminer-sha2.Tpo -c -o ccminer-sha2.o `test -f 'sha2.c' || echo '$(srcdir)/'`sha2.c
+@am__fastdepCC_TRUE@	$(AM_V_at)$(am__mv) $(DEPDIR)/ccminer-sha2.Tpo $(DEPDIR)/ccminer-sha2.Po
+@AMDEP_TRUE@@am__fastdepCC_FALSE@	$(AM_V_CC)source='sha2.c' object='ccminer-sha2.o' libtool=no @AMDEPBACKSLASH@
+@AMDEP_TRUE@@am__fastdepCC_FALSE@	DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
+@am__fastdepCC_FALSE@	$(AM_V_CC@am__nodep@)$(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(ccminer_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -c -o ccminer-sha2.o `test -f 'sha2.c' || echo '$(srcdir)/'`sha2.c
+
+ccminer-sha2.obj: sha2.c
+@am__fastdepCC_TRUE@	$(AM_V_CC)$(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(ccminer_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -MT ccminer-sha2.obj -MD -MP -MF $(DEPDIR)/ccminer-sha2.Tpo -c -o ccminer-sha2.obj `if test -f 'sha2.c'; then $(CYGPATH_W) 'sha2.c'; else $(CYGPATH_W) '$(srcdir)/sha2.c'; fi`
+@am__fastdepCC_TRUE@	$(AM_V_at)$(am__mv) $(DEPDIR)/ccminer-sha2.Tpo $(DEPDIR)/ccminer-sha2.Po
+@AMDEP_TRUE@@am__fastdepCC_FALSE@	$(AM_V_CC)source='sha2.c' object='ccminer-sha2.obj' libtool=no @AMDEPBACKSLASH@
+@AMDEP_TRUE@@am__fastdepCC_FALSE@	DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
+@am__fastdepCC_FALSE@	$(AM_V_CC@am__nodep@)$(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(ccminer_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -c -o ccminer-sha2.obj `if test -f 'sha2.c'; then $(CYGPATH_W) 'sha2.c'; else $(CYGPATH_W) '$(srcdir)/sha2.c'; fi`
+
+ccminer-fugue.o: fugue.c
+@am__fastdepCC_TRUE@	$(AM_V_CC)$(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(ccminer_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -MT ccminer-fugue.o -MD -MP -MF $(DEPDIR)/ccminer-fugue.Tpo -c -o ccminer-fugue.o `test -f 'fugue.c' || echo '$(srcdir)/'`fugue.c
+@am__fastdepCC_TRUE@	$(AM_V_at)$(am__mv) $(DEPDIR)/ccminer-fugue.Tpo $(DEPDIR)/ccminer-fugue.Po
+@AMDEP_TRUE@@am__fastdepCC_FALSE@	$(AM_V_CC)source='fugue.c' object='ccminer-fugue.o' libtool=no @AMDEPBACKSLASH@
+@AMDEP_TRUE@@am__fastdepCC_FALSE@	DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
+@am__fastdepCC_FALSE@	$(AM_V_CC@am__nodep@)$(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(ccminer_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -c -o ccminer-fugue.o `test -f 'fugue.c' || echo '$(srcdir)/'`fugue.c
+
+ccminer-fugue.obj: fugue.c
+@am__fastdepCC_TRUE@	$(AM_V_CC)$(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(ccminer_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -MT ccminer-fugue.obj -MD -MP -MF $(DEPDIR)/ccminer-fugue.Tpo -c -o ccminer-fugue.obj `if test -f 'fugue.c'; then $(CYGPATH_W) 'fugue.c'; else $(CYGPATH_W) '$(srcdir)/fugue.c'; fi`
+@am__fastdepCC_TRUE@	$(AM_V_at)$(am__mv) $(DEPDIR)/ccminer-fugue.Tpo $(DEPDIR)/ccminer-fugue.Po
+@AMDEP_TRUE@@am__fastdepCC_FALSE@	$(AM_V_CC)source='fugue.c' object='ccminer-fugue.obj' libtool=no @AMDEPBACKSLASH@
+@AMDEP_TRUE@@am__fastdepCC_FALSE@	DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
+@am__fastdepCC_FALSE@	$(AM_V_CC@am__nodep@)$(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(ccminer_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -c -o ccminer-fugue.obj `if test -f 'fugue.c'; then $(CYGPATH_W) 'fugue.c'; else $(CYGPATH_W) '$(srcdir)/fugue.c'; fi`
+
+.cpp.o:
+@am__fastdepCXX_TRUE@	$(AM_V_CXX)$(CXXCOMPILE) -MT $@ -MD -MP -MF $(DEPDIR)/$*.Tpo -c -o $@ $<
+@am__fastdepCXX_TRUE@	$(AM_V_at)$(am__mv) $(DEPDIR)/$*.Tpo $(DEPDIR)/$*.Po
+@AMDEP_TRUE@@am__fastdepCXX_FALSE@	$(AM_V_CXX)source='$<' object='$@' libtool=no @AMDEPBACKSLASH@
+@AMDEP_TRUE@@am__fastdepCXX_FALSE@	DEPDIR=$(DEPDIR) $(CXXDEPMODE) $(depcomp) @AMDEPBACKSLASH@
+@am__fastdepCXX_FALSE@	$(AM_V_CXX@am__nodep@)$(CXXCOMPILE) -c -o $@ $<
+
+.cpp.obj:
+@am__fastdepCXX_TRUE@	$(AM_V_CXX)$(CXXCOMPILE) -MT $@ -MD -MP -MF $(DEPDIR)/$*.Tpo -c -o $@ `$(CYGPATH_W) '$<'`
+@am__fastdepCXX_TRUE@	$(AM_V_at)$(am__mv) $(DEPDIR)/$*.Tpo $(DEPDIR)/$*.Po
+@AMDEP_TRUE@@am__fastdepCXX_FALSE@	$(AM_V_CXX)source='$<' object='$@' libtool=no @AMDEPBACKSLASH@
+@AMDEP_TRUE@@am__fastdepCXX_FALSE@	DEPDIR=$(DEPDIR) $(CXXDEPMODE) $(depcomp) @AMDEPBACKSLASH@
+@am__fastdepCXX_FALSE@	$(AM_V_CXX@am__nodep@)$(CXXCOMPILE) -c -o $@ `$(CYGPATH_W) '$<'`
+
+ccminer-fuguecoin.o: fuguecoin.cpp
+@am__fastdepCXX_TRUE@	$(AM_V_CXX)$(CXX) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(ccminer_CPPFLAGS) $(CPPFLAGS) $(AM_CXXFLAGS) $(CXXFLAGS) -MT ccminer-fuguecoin.o -MD -MP -MF $(DEPDIR)/ccminer-fuguecoin.Tpo -c -o ccminer-fuguecoin.o `test -f 'fuguecoin.cpp' || echo '$(srcdir)/'`fuguecoin.cpp
+@am__fastdepCXX_TRUE@	$(AM_V_at)$(am__mv) $(DEPDIR)/ccminer-fuguecoin.Tpo $(DEPDIR)/ccminer-fuguecoin.Po
+@AMDEP_TRUE@@am__fastdepCXX_FALSE@	$(AM_V_CXX)source='fuguecoin.cpp' object='ccminer-fuguecoin.o' libtool=no @AMDEPBACKSLASH@
+@AMDEP_TRUE@@am__fastdepCXX_FALSE@	DEPDIR=$(DEPDIR) $(CXXDEPMODE) $(depcomp) @AMDEPBACKSLASH@
+@am__fastdepCXX_FALSE@	$(AM_V_CXX@am__nodep@)$(CXX) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(ccminer_CPPFLAGS) $(CPPFLAGS) $(AM_CXXFLAGS) $(CXXFLAGS) -c -o ccminer-fuguecoin.o `test -f 'fuguecoin.cpp' || echo '$(srcdir)/'`fuguecoin.cpp
+
+ccminer-fuguecoin.obj: fuguecoin.cpp
+@am__fastdepCXX_TRUE@	$(AM_V_CXX)$(CXX) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(ccminer_CPPFLAGS) $(CPPFLAGS) $(AM_CXXFLAGS) $(CXXFLAGS) -MT ccminer-fuguecoin.obj -MD -MP -MF $(DEPDIR)/ccminer-fuguecoin.Tpo -c -o ccminer-fuguecoin.obj `if test -f 'fuguecoin.cpp'; then $(CYGPATH_W) 'fuguecoin.cpp'; else $(CYGPATH_W) '$(srcdir)/fuguecoin.cpp'; fi`
+@am__fastdepCXX_TRUE@	$(AM_V_at)$(am__mv) $(DEPDIR)/ccminer-fuguecoin.Tpo $(DEPDIR)/ccminer-fuguecoin.Po
+@AMDEP_TRUE@@am__fastdepCXX_FALSE@	$(AM_V_CXX)source='fuguecoin.cpp' object='ccminer-fuguecoin.obj' libtool=no @AMDEPBACKSLASH@
+@AMDEP_TRUE@@am__fastdepCXX_FALSE@	DEPDIR=$(DEPDIR) $(CXXDEPMODE) $(depcomp) @AMDEPBACKSLASH@
+@am__fastdepCXX_FALSE@	$(AM_V_CXX@am__nodep@)$(CXX) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(ccminer_CPPFLAGS) $(CPPFLAGS) $(AM_CXXFLAGS) $(CXXFLAGS) -c -o ccminer-fuguecoin.obj `if test -f 'fuguecoin.cpp'; then $(CYGPATH_W) 'fuguecoin.cpp'; else $(CYGPATH_W) '$(srcdir)/fuguecoin.cpp'; fi`
+
+# This directory's subdirectories are mostly independent; you can cd
+# into them and run 'make' without going through this Makefile.
+# To change the values of 'make' variables: instead of editing Makefiles,
+# (1) if the variable is set in 'config.status', edit 'config.status'
+#     (which will cause the Makefiles to be regenerated when you run 'make');
+# (2) otherwise, pass the desired values on the 'make' command line.
+$(am__recursive_targets):
+	@fail=; \
+	if $(am__make_keepgoing); then \
+	  failcom='fail=yes'; \
+	else \
+	  failcom='exit 1'; \
+	fi; \
+	dot_seen=no; \
+	target=`echo $@ | sed s/-recursive//`; \
+	case "$@" in \
+	  distclean-* | maintainer-clean-*) list='$(DIST_SUBDIRS)' ;; \
+	  *) list='$(SUBDIRS)' ;; \
+	esac; \
+	for subdir in $$list; do \
+	  echo "Making $$target in $$subdir"; \
+	  if test "$$subdir" = "."; then \
+	    dot_seen=yes; \
+	    local_target="$$target-am"; \
+	  else \
+	    local_target="$$target"; \
+	  fi; \
+	  ($(am__cd) $$subdir && $(MAKE) $(AM_MAKEFLAGS) $$local_target) \
+	  || eval $$failcom; \
+	done; \
+	if test "$$dot_seen" = "no"; then \
+	  $(MAKE) $(AM_MAKEFLAGS) "$$target-am" || exit 1; \
+	fi; test -z "$$fail"
+
+ID: $(am__tagged_files)
+	$(am__define_uniq_tagged_files); mkid -fID $$unique
+tags: tags-recursive
+TAGS: tags
+
+tags-am: $(TAGS_DEPENDENCIES) $(am__tagged_files)
+	set x; \
+	here=`pwd`; \
+	if ($(ETAGS) --etags-include --version) >/dev/null 2>&1; then \
+	  include_option=--etags-include; \
+	  empty_fix=.; \
+	else \
+	  include_option=--include; \
+	  empty_fix=; \
+	fi; \
+	list='$(SUBDIRS)'; for subdir in $$list; do \
+	  if test "$$subdir" = .; then :; else \
+	    test ! -f $$subdir/TAGS || \
+	      set "$$@" "$$include_option=$$here/$$subdir/TAGS"; \
+	  fi; \
+	done; \
+	$(am__define_uniq_tagged_files); \
+	shift; \
+	if test -z "$(ETAGS_ARGS)$$*$$unique"; then :; else \
+	  test -n "$$unique" || unique=$$empty_fix; \
+	  if test $$# -gt 0; then \
+	    $(ETAGS) $(ETAGSFLAGS) $(AM_ETAGSFLAGS) $(ETAGS_ARGS) \
+	      "$$@" $$unique; \
+	  else \
+	    $(ETAGS) $(ETAGSFLAGS) $(AM_ETAGSFLAGS) $(ETAGS_ARGS) \
+	      $$unique; \
+	  fi; \
+	fi
+ctags: ctags-recursive
+
+CTAGS: ctags
+ctags-am: $(TAGS_DEPENDENCIES) $(am__tagged_files)
+	$(am__define_uniq_tagged_files); \
+	test -z "$(CTAGS_ARGS)$$unique" \
+	  || $(CTAGS) $(CTAGSFLAGS) $(AM_CTAGSFLAGS) $(CTAGS_ARGS) \
+	     $$unique
+
+GTAGS:
+	here=`$(am__cd) $(top_builddir) && pwd` \
+	  && $(am__cd) $(top_srcdir) \
+	  && gtags -i $(GTAGS_ARGS) "$$here"
+cscope: cscope.files
+	test ! -s cscope.files \
+	  || $(CSCOPE) -b -q $(AM_CSCOPEFLAGS) $(CSCOPEFLAGS) -i cscope.files $(CSCOPE_ARGS)
+clean-cscope:
+	-rm -f cscope.files
+cscope.files: clean-cscope cscopelist
+cscopelist: cscopelist-recursive
+
+cscopelist-am: $(am__tagged_files)
+	list='$(am__tagged_files)'; \
+	case "$(srcdir)" in \
+	  [\\/]* | ?:[\\/]*) sdir="$(srcdir)" ;; \
+	  *) sdir=$(subdir)/$(srcdir) ;; \
+	esac; \
+	for i in $$list; do \
+	  if test -f "$$i"; then \
+	    echo "$(subdir)/$$i"; \
+	  else \
+	    echo "$$sdir/$$i"; \
+	  fi; \
+	done >> $(top_builddir)/cscope.files
+
+distclean-tags:
+	-rm -f TAGS ID GTAGS GRTAGS GSYMS GPATH tags
+	-rm -f cscope.out cscope.in.out cscope.po.out cscope.files
+
+distdir: $(DISTFILES)
+	$(am__remove_distdir)
+	test -d "$(distdir)" || mkdir "$(distdir)"
+	@srcdirstrip=`echo "$(srcdir)" | sed 's/[].[^$$\\*]/\\\\&/g'`; \
+	topsrcdirstrip=`echo "$(top_srcdir)" | sed 's/[].[^$$\\*]/\\\\&/g'`; \
+	list='$(DISTFILES)'; \
+	  dist_files=`for file in $$list; do echo $$file; done | \
+	  sed -e "s|^$$srcdirstrip/||;t" \
+	      -e "s|^$$topsrcdirstrip/|$(top_builddir)/|;t"`; \
+	case $$dist_files in \
+	  */*) $(MKDIR_P) `echo "$$dist_files" | \
+			   sed '/\//!d;s|^|$(distdir)/|;s,/[^/]*$$,,' | \
+			   sort -u` ;; \
+	esac; \
+	for file in $$dist_files; do \
+	  if test -f $$file || test -d $$file; then d=.; else d=$(srcdir); fi; \
+	  if test -d $$d/$$file; then \
+	    dir=`echo "/$$file" | sed -e 's,/[^/]*$$,,'`; \
+	    if test -d "$(distdir)/$$file"; then \
+	      find "$(distdir)/$$file" -type d ! -perm -700 -exec chmod u+rwx {} \;; \
+	    fi; \
+	    if test -d $(srcdir)/$$file && test $$d != $(srcdir); then \
+	      cp -fpR $(srcdir)/$$file "$(distdir)$$dir" || exit 1; \
+	      find "$(distdir)/$$file" -type d ! -perm -700 -exec chmod u+rwx {} \;; \
+	    fi; \
+	    cp -fpR $$d/$$file "$(distdir)$$dir" || exit 1; \
+	  else \
+	    test -f "$(distdir)/$$file" \
+	    || cp -p $$d/$$file "$(distdir)/$$file" \
+	    || exit 1; \
+	  fi; \
+	done
+	@list='$(DIST_SUBDIRS)'; for subdir in $$list; do \
+	  if test "$$subdir" = .; then :; else \
+	    $(am__make_dryrun) \
+	      || test -d "$(distdir)/$$subdir" \
+	      || $(MKDIR_P) "$(distdir)/$$subdir" \
+	      || exit 1; \
+	    dir1=$$subdir; dir2="$(distdir)/$$subdir"; \
+	    $(am__relativize); \
+	    new_distdir=$$reldir; \
+	    dir1=$$subdir; dir2="$(top_distdir)"; \
+	    $(am__relativize); \
+	    new_top_distdir=$$reldir; \
+	    echo " (cd $$subdir && $(MAKE) $(AM_MAKEFLAGS) top_distdir="$$new_top_distdir" distdir="$$new_distdir" \\"; \
+	    echo "     am__remove_distdir=: am__skip_length_check=: am__skip_mode_fix=: distdir)"; \
+	    ($(am__cd) $$subdir && \
+	      $(MAKE) $(AM_MAKEFLAGS) \
+	        top_distdir="$$new_top_distdir" \
+	        distdir="$$new_distdir" \
+		am__remove_distdir=: \
+		am__skip_length_check=: \
+		am__skip_mode_fix=: \
+	        distdir) \
+	      || exit 1; \
+	  fi; \
+	done
+	-test -n "$(am__skip_mode_fix)" \
+	|| find "$(distdir)" -type d ! -perm -755 \
+		-exec chmod u+rwx,go+rx {} \; -o \
+	  ! -type d ! -perm -444 -links 1 -exec chmod a+r {} \; -o \
+	  ! -type d ! -perm -400 -exec chmod a+r {} \; -o \
+	  ! -type d ! -perm -444 -exec $(install_sh) -c -m a+r {} {} \; \
+	|| chmod -R a+r "$(distdir)"
+dist-gzip: distdir
+	tardir=$(distdir) && $(am__tar) | GZIP=$(GZIP_ENV) gzip -c >$(distdir).tar.gz
+	$(am__post_remove_distdir)
+
+dist-bzip2: distdir
+	tardir=$(distdir) && $(am__tar) | BZIP2=$${BZIP2--9} bzip2 -c >$(distdir).tar.bz2
+	$(am__post_remove_distdir)
+
+dist-lzip: distdir
+	tardir=$(distdir) && $(am__tar) | lzip -c $${LZIP_OPT--9} >$(distdir).tar.lz
+	$(am__post_remove_distdir)
+
+dist-xz: distdir
+	tardir=$(distdir) && $(am__tar) | XZ_OPT=$${XZ_OPT--e} xz -c >$(distdir).tar.xz
+	$(am__post_remove_distdir)
+
+dist-tarZ: distdir
+	tardir=$(distdir) && $(am__tar) | compress -c >$(distdir).tar.Z
+	$(am__post_remove_distdir)
+
+dist-shar: distdir
+	shar $(distdir) | GZIP=$(GZIP_ENV) gzip -c >$(distdir).shar.gz
+	$(am__post_remove_distdir)
+
+dist-zip: distdir
+	-rm -f $(distdir).zip
+	zip -rq $(distdir).zip $(distdir)
+	$(am__post_remove_distdir)
+
+dist dist-all:
+	$(MAKE) $(AM_MAKEFLAGS) $(DIST_TARGETS) am__post_remove_distdir='@:'
+	$(am__post_remove_distdir)
+
+# This target untars the dist file and tries a VPATH configuration.  Then
+# it guarantees that the distribution is self-contained by making another
+# tarfile.
+distcheck: dist
+	case '$(DIST_ARCHIVES)' in \
+	*.tar.gz*) \
+	  GZIP=$(GZIP_ENV) gzip -dc $(distdir).tar.gz | $(am__untar) ;;\
+	*.tar.bz2*) \
+	  bzip2 -dc $(distdir).tar.bz2 | $(am__untar) ;;\
+	*.tar.lz*) \
+	  lzip -dc $(distdir).tar.lz | $(am__untar) ;;\
+	*.tar.xz*) \
+	  xz -dc $(distdir).tar.xz | $(am__untar) ;;\
+	*.tar.Z*) \
+	  uncompress -c $(distdir).tar.Z | $(am__untar) ;;\
+	*.shar.gz*) \
+	  GZIP=$(GZIP_ENV) gzip -dc $(distdir).shar.gz | unshar ;;\
+	*.zip*) \
+	  unzip $(distdir).zip ;;\
+	esac
+	chmod -R a-w $(distdir)
+	chmod u+w $(distdir)
+	mkdir $(distdir)/_build $(distdir)/_inst
+	chmod a-w $(distdir)
+	test -d $(distdir)/_build || exit 0; \
+	dc_install_base=`$(am__cd) $(distdir)/_inst && pwd | sed -e 's,^[^:\\/]:[\\/],/,'` \
+	  && dc_destdir="$${TMPDIR-/tmp}/am-dc-$$$$/" \
+	  && am__cwd=`pwd` \
+	  && $(am__cd) $(distdir)/_build \
+	  && ../configure --srcdir=.. --prefix="$$dc_install_base" \
+	    $(AM_DISTCHECK_CONFIGURE_FLAGS) \
+	    $(DISTCHECK_CONFIGURE_FLAGS) \
+	  && $(MAKE) $(AM_MAKEFLAGS) \
+	  && $(MAKE) $(AM_MAKEFLAGS) dvi \
+	  && $(MAKE) $(AM_MAKEFLAGS) check \
+	  && $(MAKE) $(AM_MAKEFLAGS) install \
+	  && $(MAKE) $(AM_MAKEFLAGS) installcheck \
+	  && $(MAKE) $(AM_MAKEFLAGS) uninstall \
+	  && $(MAKE) $(AM_MAKEFLAGS) distuninstallcheck_dir="$$dc_install_base" \
+	        distuninstallcheck \
+	  && chmod -R a-w "$$dc_install_base" \
+	  && ({ \
+	       (cd ../.. && umask 077 && mkdir "$$dc_destdir") \
+	       && $(MAKE) $(AM_MAKEFLAGS) DESTDIR="$$dc_destdir" install \
+	       && $(MAKE) $(AM_MAKEFLAGS) DESTDIR="$$dc_destdir" uninstall \
+	       && $(MAKE) $(AM_MAKEFLAGS) DESTDIR="$$dc_destdir" \
+	            distuninstallcheck_dir="$$dc_destdir" distuninstallcheck; \
+	      } || { rm -rf "$$dc_destdir"; exit 1; }) \
+	  && rm -rf "$$dc_destdir" \
+	  && $(MAKE) $(AM_MAKEFLAGS) dist \
+	  && rm -rf $(DIST_ARCHIVES) \
+	  && $(MAKE) $(AM_MAKEFLAGS) distcleancheck \
+	  && cd "$$am__cwd" \
+	  || exit 1
+	$(am__post_remove_distdir)
+	@(echo "$(distdir) archives ready for distribution: "; \
+	  list='$(DIST_ARCHIVES)'; for i in $$list; do echo $$i; done) | \
+	  sed -e 1h -e 1s/./=/g -e 1p -e 1x -e '$$p' -e '$$x'
+distuninstallcheck:
+	@test -n '$(distuninstallcheck_dir)' || { \
+	  echo 'ERROR: trying to run $@ with an empty' \
+	       '$$(distuninstallcheck_dir)' >&2; \
+	  exit 1; \
+	}; \
+	$(am__cd) '$(distuninstallcheck_dir)' || { \
+	  echo 'ERROR: cannot chdir into $(distuninstallcheck_dir)' >&2; \
+	  exit 1; \
+	}; \
+	test `$(am__distuninstallcheck_listfiles) | wc -l` -eq 0 \
+	   || { echo "ERROR: files left after uninstall:" ; \
+	        if test -n "$(DESTDIR)"; then \
+	          echo "  (check DESTDIR support)"; \
+	        fi ; \
+	        $(distuninstallcheck_listfiles) ; \
+	        exit 1; } >&2
+distcleancheck: distclean
+	@if test '$(srcdir)' = . ; then \
+	  echo "ERROR: distcleancheck can only run from a VPATH build" ; \
+	  exit 1 ; \
+	fi
+	@test `$(distcleancheck_listfiles) | wc -l` -eq 0 \
+	  || { echo "ERROR: files left in build directory after distclean:" ; \
+	       $(distcleancheck_listfiles) ; \
+	       exit 1; } >&2
+check-am: all-am
+check: check-recursive
+all-am: Makefile $(PROGRAMS) cpuminer-config.h
+installdirs: installdirs-recursive
+installdirs-am:
+	for dir in "$(DESTDIR)$(bindir)"; do \
+	  test -z "$$dir" || $(MKDIR_P) "$$dir"; \
+	done
+install: install-recursive
+install-exec: install-exec-recursive
+install-data: install-data-recursive
+uninstall: uninstall-recursive
+
+install-am: all-am
+	@$(MAKE) $(AM_MAKEFLAGS) install-exec-am install-data-am
+
+installcheck: installcheck-recursive
+install-strip:
+	if test -z '$(STRIP)'; then \
+	  $(MAKE) $(AM_MAKEFLAGS) INSTALL_PROGRAM="$(INSTALL_STRIP_PROGRAM)" \
+	    install_sh_PROGRAM="$(INSTALL_STRIP_PROGRAM)" INSTALL_STRIP_FLAG=-s \
+	      install; \
+	else \
+	  $(MAKE) $(AM_MAKEFLAGS) INSTALL_PROGRAM="$(INSTALL_STRIP_PROGRAM)" \
+	    install_sh_PROGRAM="$(INSTALL_STRIP_PROGRAM)" INSTALL_STRIP_FLAG=-s \
+	    "INSTALL_PROGRAM_ENV=STRIPPROG='$(STRIP)'" install; \
+	fi
+mostlyclean-generic:
+
+clean-generic:
+
+distclean-generic:
+	-test -z "$(CONFIG_CLEAN_FILES)" || rm -f $(CONFIG_CLEAN_FILES)
+	-test . = "$(srcdir)" || test -z "$(CONFIG_CLEAN_VPATH_FILES)" || rm -f $(CONFIG_CLEAN_VPATH_FILES)
+
+maintainer-clean-generic:
+	@echo "This command is intended for maintainers to use"
+	@echo "it deletes files that may require special tools to rebuild."
+clean: clean-recursive
+
+clean-am: clean-binPROGRAMS clean-generic mostlyclean-am
+
+distclean: distclean-recursive
+	-rm -f $(am__CONFIG_DISTCLEAN_FILES)
+	-rm -rf ./$(DEPDIR)
+	-rm -f Makefile
+distclean-am: clean-am distclean-compile distclean-generic \
+	distclean-hdr distclean-tags
+
+dvi: dvi-recursive
+
+dvi-am:
+
+html: html-recursive
+
+html-am:
+
+info: info-recursive
+
+info-am:
+
+install-data-am:
+
+install-dvi: install-dvi-recursive
+
+install-dvi-am:
+
+install-exec-am: install-binPROGRAMS
+
+install-html: install-html-recursive
+
+install-html-am:
+
+install-info: install-info-recursive
+
+install-info-am:
+
+install-man:
+
+install-pdf: install-pdf-recursive
+
+install-pdf-am:
+
+install-ps: install-ps-recursive
+
+install-ps-am:
+
+installcheck-am:
+
+maintainer-clean: maintainer-clean-recursive
+	-rm -f $(am__CONFIG_DISTCLEAN_FILES)
+	-rm -rf $(top_srcdir)/autom4te.cache
+	-rm -rf ./$(DEPDIR)
+	-rm -f Makefile
+maintainer-clean-am: distclean-am maintainer-clean-generic
+
+mostlyclean: mostlyclean-recursive
+
+mostlyclean-am: mostlyclean-compile mostlyclean-generic
+
+pdf: pdf-recursive
+
+pdf-am:
+
+ps: ps-recursive
+
+ps-am:
+
+uninstall-am: uninstall-binPROGRAMS
+
+.MAKE: $(am__recursive_targets) all install-am install-strip
+
+.PHONY: $(am__recursive_targets) CTAGS GTAGS TAGS all all-am \
+	am--refresh check check-am clean clean-binPROGRAMS \
+	clean-cscope clean-generic cscope cscopelist-am ctags ctags-am \
+	dist dist-all dist-bzip2 dist-gzip dist-lzip dist-shar \
+	dist-tarZ dist-xz dist-zip distcheck distclean \
+	distclean-compile distclean-generic distclean-hdr \
+	distclean-tags distcleancheck distdir distuninstallcheck dvi \
+	dvi-am html html-am info info-am install install-am \
+	install-binPROGRAMS install-data install-data-am install-dvi \
+	install-dvi-am install-exec install-exec-am install-html \
+	install-html-am install-info install-info-am install-man \
+	install-pdf install-pdf-am install-ps install-ps-am \
+	install-strip installcheck installcheck-am installdirs \
+	installdirs-am maintainer-clean maintainer-clean-generic \
+	mostlyclean mostlyclean-compile mostlyclean-generic pdf pdf-am \
+	ps ps-am tags tags-am uninstall uninstall-am \
+	uninstall-binPROGRAMS
+
+
+.cu.o:
+	$(NVCC) @CFLAGS@ -Xptxas "-abi=no -v" -arch=compute_20 --maxrregcount=63 --ptxas-options=-v $(JANSSON_INCLUDES) -o $@ -c $<
+
+#heavy.o: heavy.cu
+#	$(NVCC) @CFLAGS@ -Xptxas "-abi=no -v" -arch=compute_20 --maxrregcount=63 --ptxas-options=-v $(JANSSON_INCLUDES) -o $@ -c $<
+#
+#cuda_hefty1.o: cuda_hefty1.cu
+#	$(NVCC) @CFLAGS@ -Xptxas "-abi=no -v" -arch=compute_20 --maxrregcount=63 --ptxas-options=-v $(JANSSON_INCLUDES) -o $@ -c $<
+
+# Tell versions [3.59,3.63) of GNU make to not export all variables.
+# Otherwise a system limit (for SysV at least) may be exceeded.
+.NOEXPORT:
diff --git a/NEWS b/NEWS
new file mode 100644
index 0000000000..d3f5a12faa
--- /dev/null
+++ b/NEWS
@@ -0,0 +1 @@
+
diff --git a/README b/README
new file mode 100644
index 0000000000..f5c26fca65
--- /dev/null
+++ b/README
@@ -0,0 +1,3 @@
+
+A CUDA based miner for Heavycoin and Fuguecoin. For more information, take a look at README.txt
+
diff --git a/README.txt b/README.txt
new file mode 100644
index 0000000000..12981970d1
--- /dev/null
+++ b/README.txt
@@ -0,0 +1,117 @@
+
+HeavyCUDA release Mar 18th 2014 - Initial Release
+-------------------------------------------------------------
+
+***************************************************************
+If you find this tool useful and like to support its continued 
+          development, then consider a donation.
+
+   LTC donation address: LKS1WDKGED647msBQfLBHV3Ls8sveGncnm
+   BTC donation address: 16hJF5mceSojnTD3ZTUDqdRhDyPJzoRakM
+   YAC donation address: Y87sptDEcpLkLeAuex6qZioDbvy1qXZEj4
+   VTC donation address: VrjeFzMgvteCGarLw85KivBzmsiH9fqp4a
+   MAX donation address: mHrhQP9EFArechWxTFJ97s9D3jvcCvEEnt
+  DOGE donation address: DT9ghsGmez6ojVdEZgvaZbT2Z3TruXG6yP
+ PANDA donation address: PvgtxJ2ZKaudRogCXfUMLXVaWUMcKQgRed
+   MRC donation address: 1Lxc4JPDpQRJB8BN4YwhmSQ3Rcu8gjj2Kd
+   HVC donation address: HNN3PyyTMkDo4RkEjkWSGMwqia1yD8mwJN
+***************************************************************
+
+>>> Introduction <<<
+
+This is a CUDA accelerated mining application for use with
+HeavyCoin and FugueCoin. We did not take effort on usability,
+so please set your parameters carefuly.
+
+THIS PROGRAMM IS PROVIDED "AS-IS", USE IT AT YOUR OWN RISK!
+
+If you're interessted and read the source-code, please excuse
+that the most of our comments are in german.
+
+>>> Command Line Interface <<<
+
+This code is based on the pooler cpuminer 2.3.2 release and inherits
+its command line interface and options.
+
+  -a, --algo=ALGO       specify the algorithm to use
+                          heavy       use to mine Heavycoin
+                          fugue256    use to mine Fuguecoin
+
+  -o, --url=URL         URL of mining server (default: " DEF_RPC_URL ")
+  -O, --userpass=U:P    username:password pair for mining server
+  -u, --user=USERNAME   username for mining server
+  -p, --pass=PASSWORD   password for mining server
+  -v, --vote		Heavycoin block vote (default: 512)
+      --cert=FILE       certificate for mining server using SSL
+  -x, --proxy=[PROTOCOL://]HOST[:PORT]  connect through a proxy
+  -t, --threads=N       number of miner threads (default: number of nVidia GPUs in your system)
+  -r, --retries=N       number of times to retry if a network call fails
+                          (default: retry indefinitely)
+  -R, --retry-pause=N   time to pause between retries, in seconds (default: 15)
+  -T, --timeout=N       network timeout, in seconds (default: 270)
+  -s, --scantime=N      upper bound on time spent scanning current work when
+                        long polling is unavailable, in seconds (default: 5)
+      --no-longpoll     disable X-Long-Polling support
+      --no-stratum      disable X-Stratum support
+  -q, --quiet           disable per-thread hashmeter output
+  -D, --debug           enable debug output
+  -P, --protocol-dump   verbose dump of protocol-level activities
+  -B, --background      run the miner in the background
+      --benchmark       run in offline benchmark mode
+  -c, --config=FILE     load a JSON-format configuration file
+  -V, --version         display version information and exit
+  -h, --help            display this help text and exit
+
+>>> Examples <<<
+
+Example for Heavycoin Mining on heavycoinpool.com with a single gpu in your system
+
+cudaminer.exe -t 1 -a heavy -o stratum+tcp://stratum01.heavycoinpool.com:5333 -u <<username.worker>> -p <<workerpassword>> -v 512
+
+
+
+Example for Heavycoin Mining on hvc.1gh.com with a dual gpu in your system
+
+cudaminer.exe -t 2 -a heavy -o stratum+tcp://hvcpool.1gh.com:5333 -u <<WALLET>> -p x -v 512
+
+
+
+Example for Fuguecoin solo-mining with 4 gpu's in your system and a Fuguecoin-wallet running on localhost
+
+cudaminer.exe -q -s 1 -t 4 -a fugue256 -o http://localhost:9089 -u <<myusername>> -p <<mypassword>>
+
+For solo-mining you typically use -o 127.0.0.1:xxxx where xxxx represents
+the RPC portnumber specified in your wallet's .conf file and you have to
+pass the same username and password with -O as specified in the wallet's
+.conf file. The wallet must also be started with the -server option and
+the server flag in the wallet's .conf file set to 1
+
+
+>>> Additional Notes <<<
+
+This code should be running on nVidia GPUs ranging from compute capability
+2.0 up to compute capability 3.5. Just don't expect any hashing miracles
+from your old clunkers.
+
+>>> RELEASE HISTORY <<<
+
+  March, 18 2014 initial release.
+
+
+>>> AUTHORS <<<
+
+Notable contributors to this application are:
+
+Christian Buchner, Christian H. (Germany): CUDA implementation 
+
+and also many thanks to anyone else who contributed to the original
+cpuminer application (Jeff Garzik, pooler), it's original HVC-fork
+and the HVC-fork available at hvc.1gh.com
+
+Source code is included to satisfy GNU GPL V2 requirements.
+
+
+With kind regards,
+
+   Christian Buchner ( Christian.Buchner@gmail.com )
+   Christian H. ( Chris84 )
diff --git a/aclocal.m4 b/aclocal.m4
new file mode 100644
index 0000000000..1b2c558401
--- /dev/null
+++ b/aclocal.m4
@@ -0,0 +1,1379 @@
+# generated automatically by aclocal 1.13.3 -*- Autoconf -*-
+
+# Copyright (C) 1996-2013 Free Software Foundation, Inc.
+
+# This file is free software; the Free Software Foundation
+# gives unlimited permission to copy and/or distribute it,
+# with or without modifications, as long as this notice is preserved.
+
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY, to the extent permitted by law; without
+# even the implied warranty of MERCHANTABILITY or FITNESS FOR A
+# PARTICULAR PURPOSE.
+
+m4_ifndef([AC_CONFIG_MACRO_DIRS], [m4_defun([_AM_CONFIG_MACRO_DIRS], [])m4_defun([AC_CONFIG_MACRO_DIRS], [_AM_CONFIG_MACRO_DIRS($@)])])
+m4_ifndef([AC_AUTOCONF_VERSION],
+  [m4_copy([m4_PACKAGE_VERSION], [AC_AUTOCONF_VERSION])])dnl
+m4_if(m4_defn([AC_AUTOCONF_VERSION]), [2.69],,
+[m4_warning([this file was generated for autoconf 2.69.
+You have another version of autoconf.  It may work, but is not guaranteed to.
+If you have problems, you may need to regenerate the build system entirely.
+To do so, use the procedure documented by the package, typically 'autoreconf'.])])
+
+# LIBCURL_CHECK_CONFIG ([DEFAULT-ACTION], [MINIMUM-VERSION],
+#                       [ACTION-IF-YES], [ACTION-IF-NO])
+# ----------------------------------------------------------
+#      David Shaw <dshaw@jabberwocky.com>   May-09-2006
+#
+# Checks for libcurl.  DEFAULT-ACTION is the string yes or no to
+# specify whether to default to --with-libcurl or --without-libcurl.
+# If not supplied, DEFAULT-ACTION is yes.  MINIMUM-VERSION is the
+# minimum version of libcurl to accept.  Pass the version as a regular
+# version number like 7.10.1. If not supplied, any version is
+# accepted.  ACTION-IF-YES is a list of shell commands to run if
+# libcurl was successfully found and passed the various tests.
+# ACTION-IF-NO is a list of shell commands that are run otherwise.
+# Note that using --without-libcurl does run ACTION-IF-NO.
+#
+# This macro #defines HAVE_LIBCURL if a working libcurl setup is
+# found, and sets @LIBCURL@ and @LIBCURL_CPPFLAGS@ to the necessary
+# values.  Other useful defines are LIBCURL_FEATURE_xxx where xxx are
+# the various features supported by libcurl, and LIBCURL_PROTOCOL_yyy
+# where yyy are the various protocols supported by libcurl.  Both xxx
+# and yyy are capitalized.  See the list of AH_TEMPLATEs at the top of
+# the macro for the complete list of possible defines.  Shell
+# variables $libcurl_feature_xxx and $libcurl_protocol_yyy are also
+# defined to 'yes' for those features and protocols that were found.
+# Note that xxx and yyy keep the same capitalization as in the
+# curl-config list (e.g. it's "HTTP" and not "http").
+#
+# Users may override the detected values by doing something like:
+# LIBCURL="-lcurl" LIBCURL_CPPFLAGS="-I/usr/myinclude" ./configure
+#
+# For the sake of sanity, this macro assumes that any libcurl that is
+# found is after version 7.7.2, the first version that included the
+# curl-config script.  Note that it is very important for people
+# packaging binary versions of libcurl to include this script!
+# Without curl-config, we can only guess what protocols are available,
+# or use curl_version_info to figure it out at runtime.
+
+AC_DEFUN([LIBCURL_CHECK_CONFIG],
+[
+  AH_TEMPLATE([LIBCURL_FEATURE_SSL],[Defined if libcurl supports SSL])
+  AH_TEMPLATE([LIBCURL_FEATURE_KRB4],[Defined if libcurl supports KRB4])
+  AH_TEMPLATE([LIBCURL_FEATURE_IPV6],[Defined if libcurl supports IPv6])
+  AH_TEMPLATE([LIBCURL_FEATURE_LIBZ],[Defined if libcurl supports libz])
+  AH_TEMPLATE([LIBCURL_FEATURE_ASYNCHDNS],[Defined if libcurl supports AsynchDNS])
+  AH_TEMPLATE([LIBCURL_FEATURE_IDN],[Defined if libcurl supports IDN])
+  AH_TEMPLATE([LIBCURL_FEATURE_SSPI],[Defined if libcurl supports SSPI])
+  AH_TEMPLATE([LIBCURL_FEATURE_NTLM],[Defined if libcurl supports NTLM])
+
+  AH_TEMPLATE([LIBCURL_PROTOCOL_HTTP],[Defined if libcurl supports HTTP])
+  AH_TEMPLATE([LIBCURL_PROTOCOL_HTTPS],[Defined if libcurl supports HTTPS])
+  AH_TEMPLATE([LIBCURL_PROTOCOL_FTP],[Defined if libcurl supports FTP])
+  AH_TEMPLATE([LIBCURL_PROTOCOL_FTPS],[Defined if libcurl supports FTPS])
+  AH_TEMPLATE([LIBCURL_PROTOCOL_FILE],[Defined if libcurl supports FILE])
+  AH_TEMPLATE([LIBCURL_PROTOCOL_TELNET],[Defined if libcurl supports TELNET])
+  AH_TEMPLATE([LIBCURL_PROTOCOL_LDAP],[Defined if libcurl supports LDAP])
+  AH_TEMPLATE([LIBCURL_PROTOCOL_DICT],[Defined if libcurl supports DICT])
+  AH_TEMPLATE([LIBCURL_PROTOCOL_TFTP],[Defined if libcurl supports TFTP])
+  AH_TEMPLATE([LIBCURL_PROTOCOL_RTSP],[Defined if libcurl supports RTSP])
+  AH_TEMPLATE([LIBCURL_PROTOCOL_POP3],[Defined if libcurl supports POP3])
+  AH_TEMPLATE([LIBCURL_PROTOCOL_IMAP],[Defined if libcurl supports IMAP])
+  AH_TEMPLATE([LIBCURL_PROTOCOL_SMTP],[Defined if libcurl supports SMTP])
+
+  AC_ARG_WITH(libcurl,
+     AC_HELP_STRING([--with-libcurl=PREFIX],[look for the curl library in PREFIX/lib and headers in PREFIX/include]),
+     [_libcurl_with=$withval],[_libcurl_with=ifelse([$1],,[yes],[$1])])
+
+  if test "$_libcurl_with" != "no" ; then
+
+     AC_PROG_AWK
+
+     _libcurl_version_parse="eval $AWK '{split(\$NF,A,\".\"); X=256*256*A[[1]]+256*A[[2]]+A[[3]]; print X;}'"
+
+     _libcurl_try_link=yes
+
+     if test -d "$_libcurl_with" ; then
+        LIBCURL_CPPFLAGS="-I$withval/include"
+        _libcurl_ldflags="-L$withval/lib"
+        AC_PATH_PROG([_libcurl_config],[curl-config],[],
+                     ["$withval/bin"])
+     else
+        AC_PATH_PROG([_libcurl_config],[curl-config],[],[$PATH])
+     fi
+
+     if test x$_libcurl_config != "x" ; then
+        AC_CACHE_CHECK([for the version of libcurl],
+           [libcurl_cv_lib_curl_version],
+           [libcurl_cv_lib_curl_version=`$_libcurl_config --version | $AWK '{print $[]2}'`])
+
+        _libcurl_version=`echo $libcurl_cv_lib_curl_version | $_libcurl_version_parse`
+        _libcurl_wanted=`echo ifelse([$2],,[0],[$2]) | $_libcurl_version_parse`
+
+        if test $_libcurl_wanted -gt 0 ; then
+           AC_CACHE_CHECK([for libcurl >= version $2],
+              [libcurl_cv_lib_version_ok],
+              [
+              if test $_libcurl_version -ge $_libcurl_wanted ; then
+                 libcurl_cv_lib_version_ok=yes
+              else
+                 libcurl_cv_lib_version_ok=no
+              fi
+              ])
+        fi
+
+        if test $_libcurl_wanted -eq 0 || test x$libcurl_cv_lib_version_ok = xyes ; then
+           if test x"$LIBCURL_CPPFLAGS" = "x" ; then
+              LIBCURL_CPPFLAGS=`$_libcurl_config --cflags`
+           fi
+           if test x"$LIBCURL" = "x" ; then
+              LIBCURL=`$_libcurl_config --libs`
+
+              # This is so silly, but Apple actually has a bug in their
+              # curl-config script.  Fixed in Tiger, but there are still
+              # lots of Panther installs around.
+              case "${host}" in
+                 powerpc-apple-darwin7*)
+                    LIBCURL=`echo $LIBCURL | sed -e 's|-arch i386||g'`
+                 ;;
+              esac
+           fi
+
+           # All curl-config scripts support --feature
+           _libcurl_features=`$_libcurl_config --feature`
+
+           # Is it modern enough to have --protocols? (7.12.4)
+           if test $_libcurl_version -ge 461828 ; then
+              _libcurl_protocols=`$_libcurl_config --protocols`
+           fi
+        else
+           _libcurl_try_link=no
+        fi
+
+        unset _libcurl_wanted
+     fi
+
+     if test $_libcurl_try_link = yes ; then
+
+        # we didn't find curl-config, so let's see if the user-supplied
+        # link line (or failing that, "-lcurl") is enough.
+        LIBCURL=${LIBCURL-"$_libcurl_ldflags -lcurl"}
+
+        AC_CACHE_CHECK([whether libcurl is usable],
+           [libcurl_cv_lib_curl_usable],
+           [
+           _libcurl_save_cppflags=$CPPFLAGS
+           CPPFLAGS="$LIBCURL_CPPFLAGS $CPPFLAGS"
+           _libcurl_save_libs=$LIBS
+           LIBS="$LIBCURL $LIBS"
+
+           AC_LINK_IFELSE([AC_LANG_PROGRAM([[#include <curl/curl.h>]],[[
+/* Try and use a few common options to force a failure if we are
+   missing symbols or can't link. */
+int x;
+curl_easy_setopt(NULL,CURLOPT_URL,NULL);
+x=CURL_ERROR_SIZE;
+x=CURLOPT_WRITEFUNCTION;
+x=CURLOPT_FILE;
+x=CURLOPT_ERRORBUFFER;
+x=CURLOPT_STDERR;
+x=CURLOPT_VERBOSE;
+if (x) ;
+]])],libcurl_cv_lib_curl_usable=yes,libcurl_cv_lib_curl_usable=no)
+
+           CPPFLAGS=$_libcurl_save_cppflags
+           LIBS=$_libcurl_save_libs
+           unset _libcurl_save_cppflags
+           unset _libcurl_save_libs
+           ])
+
+        if test $libcurl_cv_lib_curl_usable = yes ; then
+
+           # Does curl_free() exist in this version of libcurl?
+           # If not, fake it with free()
+
+           _libcurl_save_cppflags=$CPPFLAGS
+           CPPFLAGS="$CPPFLAGS $LIBCURL_CPPFLAGS"
+           _libcurl_save_libs=$LIBS
+           LIBS="$LIBS $LIBCURL"
+
+           AC_CHECK_FUNC(curl_free,,
+              AC_DEFINE(curl_free,free,
+                [Define curl_free() as free() if our version of curl lacks curl_free.]))
+
+           CPPFLAGS=$_libcurl_save_cppflags
+           LIBS=$_libcurl_save_libs
+           unset _libcurl_save_cppflags
+           unset _libcurl_save_libs
+
+           AC_DEFINE(HAVE_LIBCURL,1,
+             [Define to 1 if you have a functional curl library.])
+           AC_SUBST(LIBCURL_CPPFLAGS)
+           AC_SUBST(LIBCURL)
+
+           for _libcurl_feature in $_libcurl_features ; do
+              AC_DEFINE_UNQUOTED(AS_TR_CPP(libcurl_feature_$_libcurl_feature),[1])
+              eval AS_TR_SH(libcurl_feature_$_libcurl_feature)=yes
+           done
+
+           if test "x$_libcurl_protocols" = "x" ; then
+
+              # We don't have --protocols, so just assume that all
+              # protocols are available
+              _libcurl_protocols="HTTP FTP FILE TELNET LDAP DICT TFTP"
+
+              if test x$libcurl_feature_SSL = xyes ; then
+                 _libcurl_protocols="$_libcurl_protocols HTTPS"
+
+                 # FTPS wasn't standards-compliant until version
+                 # 7.11.0 (0x070b00 == 461568)
+                 if test $_libcurl_version -ge 461568; then
+                    _libcurl_protocols="$_libcurl_protocols FTPS"
+                 fi
+              fi
+
+              # RTSP, IMAP, POP3 and SMTP were added in
+              # 7.20.0 (0x071400 == 463872)
+              if test $_libcurl_version -ge 463872; then
+                 _libcurl_protocols="$_libcurl_protocols RTSP IMAP POP3 SMTP"
+              fi
+           fi
+
+           for _libcurl_protocol in $_libcurl_protocols ; do
+              AC_DEFINE_UNQUOTED(AS_TR_CPP(libcurl_protocol_$_libcurl_protocol),[1])
+              eval AS_TR_SH(libcurl_protocol_$_libcurl_protocol)=yes
+           done
+        else
+           unset LIBCURL
+           unset LIBCURL_CPPFLAGS
+        fi
+     fi
+
+     unset _libcurl_try_link
+     unset _libcurl_version_parse
+     unset _libcurl_config
+     unset _libcurl_feature
+     unset _libcurl_features
+     unset _libcurl_protocol
+     unset _libcurl_protocols
+     unset _libcurl_version
+     unset _libcurl_ldflags
+  fi
+
+  if test x$_libcurl_with = xno || test x$libcurl_cv_lib_curl_usable != xyes ; then
+     # This is the IF-NO path
+     ifelse([$4],,:,[$4])
+  else
+     # This is the IF-YES path
+     ifelse([$3],,:,[$3])
+  fi
+
+  unset _libcurl_with
+])dnl
+
+# Copyright (C) 2002-2013 Free Software Foundation, Inc.
+#
+# This file is free software; the Free Software Foundation
+# gives unlimited permission to copy and/or distribute it,
+# with or without modifications, as long as this notice is preserved.
+
+# AM_AUTOMAKE_VERSION(VERSION)
+# ----------------------------
+# Automake X.Y traces this macro to ensure aclocal.m4 has been
+# generated from the m4 files accompanying Automake X.Y.
+# (This private macro should not be called outside this file.)
+AC_DEFUN([AM_AUTOMAKE_VERSION],
+[am__api_version='1.13'
+dnl Some users find AM_AUTOMAKE_VERSION and mistake it for a way to
+dnl require some minimum version.  Point them to the right macro.
+m4_if([$1], [1.13.3], [],
+      [AC_FATAL([Do not call $0, use AM_INIT_AUTOMAKE([$1]).])])dnl
+])
+
+# _AM_AUTOCONF_VERSION(VERSION)
+# -----------------------------
+# aclocal traces this macro to find the Autoconf version.
+# This is a private macro too.  Using m4_define simplifies
+# the logic in aclocal, which can simply ignore this definition.
+m4_define([_AM_AUTOCONF_VERSION], [])
+
+# AM_SET_CURRENT_AUTOMAKE_VERSION
+# -------------------------------
+# Call AM_AUTOMAKE_VERSION and AM_AUTOMAKE_VERSION so they can be traced.
+# This function is AC_REQUIREd by AM_INIT_AUTOMAKE.
+AC_DEFUN([AM_SET_CURRENT_AUTOMAKE_VERSION],
+[AM_AUTOMAKE_VERSION([1.13.3])dnl
+m4_ifndef([AC_AUTOCONF_VERSION],
+  [m4_copy([m4_PACKAGE_VERSION], [AC_AUTOCONF_VERSION])])dnl
+_AM_AUTOCONF_VERSION(m4_defn([AC_AUTOCONF_VERSION]))])
+
+# Figure out how to run the assembler.                      -*- Autoconf -*-
+
+# Copyright (C) 2001-2013 Free Software Foundation, Inc.
+#
+# This file is free software; the Free Software Foundation
+# gives unlimited permission to copy and/or distribute it,
+# with or without modifications, as long as this notice is preserved.
+
+# AM_PROG_AS
+# ----------
+AC_DEFUN([AM_PROG_AS],
+[# By default we simply use the C compiler to build assembly code.
+AC_REQUIRE([AC_PROG_CC])
+test "${CCAS+set}" = set || CCAS=$CC
+test "${CCASFLAGS+set}" = set || CCASFLAGS=$CFLAGS
+AC_ARG_VAR([CCAS],      [assembler compiler command (defaults to CC)])
+AC_ARG_VAR([CCASFLAGS], [assembler compiler flags (defaults to CFLAGS)])
+_AM_IF_OPTION([no-dependencies],, [_AM_DEPENDENCIES([CCAS])])dnl
+])
+
+# AM_AUX_DIR_EXPAND                                         -*- Autoconf -*-
+
+# Copyright (C) 2001-2013 Free Software Foundation, Inc.
+#
+# This file is free software; the Free Software Foundation
+# gives unlimited permission to copy and/or distribute it,
+# with or without modifications, as long as this notice is preserved.
+
+# For projects using AC_CONFIG_AUX_DIR([foo]), Autoconf sets
+# $ac_aux_dir to '$srcdir/foo'.  In other projects, it is set to
+# '$srcdir', '$srcdir/..', or '$srcdir/../..'.
+#
+# Of course, Automake must honor this variable whenever it calls a
+# tool from the auxiliary directory.  The problem is that $srcdir (and
+# therefore $ac_aux_dir as well) can be either absolute or relative,
+# depending on how configure is run.  This is pretty annoying, since
+# it makes $ac_aux_dir quite unusable in subdirectories: in the top
+# source directory, any form will work fine, but in subdirectories a
+# relative path needs to be adjusted first.
+#
+# $ac_aux_dir/missing
+#    fails when called from a subdirectory if $ac_aux_dir is relative
+# $top_srcdir/$ac_aux_dir/missing
+#    fails if $ac_aux_dir is absolute,
+#    fails when called from a subdirectory in a VPATH build with
+#          a relative $ac_aux_dir
+#
+# The reason of the latter failure is that $top_srcdir and $ac_aux_dir
+# are both prefixed by $srcdir.  In an in-source build this is usually
+# harmless because $srcdir is '.', but things will broke when you
+# start a VPATH build or use an absolute $srcdir.
+#
+# So we could use something similar to $top_srcdir/$ac_aux_dir/missing,
+# iff we strip the leading $srcdir from $ac_aux_dir.  That would be:
+#   am_aux_dir='\$(top_srcdir)/'`expr "$ac_aux_dir" : "$srcdir//*\(.*\)"`
+# and then we would define $MISSING as
+#   MISSING="\${SHELL} $am_aux_dir/missing"
+# This will work as long as MISSING is not called from configure, because
+# unfortunately $(top_srcdir) has no meaning in configure.
+# However there are other variables, like CC, which are often used in
+# configure, and could therefore not use this "fixed" $ac_aux_dir.
+#
+# Another solution, used here, is to always expand $ac_aux_dir to an
+# absolute PATH.  The drawback is that using absolute paths prevent a
+# configured tree to be moved without reconfiguration.
+
+AC_DEFUN([AM_AUX_DIR_EXPAND],
+[dnl Rely on autoconf to set up CDPATH properly.
+AC_PREREQ([2.50])dnl
+# expand $ac_aux_dir to an absolute path
+am_aux_dir=`cd $ac_aux_dir && pwd`
+])
+
+# AM_CONDITIONAL                                            -*- Autoconf -*-
+
+# Copyright (C) 1997-2013 Free Software Foundation, Inc.
+#
+# This file is free software; the Free Software Foundation
+# gives unlimited permission to copy and/or distribute it,
+# with or without modifications, as long as this notice is preserved.
+
+# AM_CONDITIONAL(NAME, SHELL-CONDITION)
+# -------------------------------------
+# Define a conditional.
+AC_DEFUN([AM_CONDITIONAL],
+[AC_PREREQ([2.52])dnl
+ m4_if([$1], [TRUE],  [AC_FATAL([$0: invalid condition: $1])],
+       [$1], [FALSE], [AC_FATAL([$0: invalid condition: $1])])dnl
+AC_SUBST([$1_TRUE])dnl
+AC_SUBST([$1_FALSE])dnl
+_AM_SUBST_NOTMAKE([$1_TRUE])dnl
+_AM_SUBST_NOTMAKE([$1_FALSE])dnl
+m4_define([_AM_COND_VALUE_$1], [$2])dnl
+if $2; then
+  $1_TRUE=
+  $1_FALSE='#'
+else
+  $1_TRUE='#'
+  $1_FALSE=
+fi
+AC_CONFIG_COMMANDS_PRE(
+[if test -z "${$1_TRUE}" && test -z "${$1_FALSE}"; then
+  AC_MSG_ERROR([[conditional "$1" was never defined.
+Usually this means the macro was only invoked conditionally.]])
+fi])])
+
+# Copyright (C) 1999-2013 Free Software Foundation, Inc.
+#
+# This file is free software; the Free Software Foundation
+# gives unlimited permission to copy and/or distribute it,
+# with or without modifications, as long as this notice is preserved.
+
+
+# There are a few dirty hacks below to avoid letting 'AC_PROG_CC' be
+# written in clear, in which case automake, when reading aclocal.m4,
+# will think it sees a *use*, and therefore will trigger all it's
+# C support machinery.  Also note that it means that autoscan, seeing
+# CC etc. in the Makefile, will ask for an AC_PROG_CC use...
+
+
+# _AM_DEPENDENCIES(NAME)
+# ----------------------
+# See how the compiler implements dependency checking.
+# NAME is "CC", "CXX", "OBJC", "OBJCXX", "UPC", or "GJC".
+# We try a few techniques and use that to set a single cache variable.
+#
+# We don't AC_REQUIRE the corresponding AC_PROG_CC since the latter was
+# modified to invoke _AM_DEPENDENCIES(CC); we would have a circular
+# dependency, and given that the user is not expected to run this macro,
+# just rely on AC_PROG_CC.
+AC_DEFUN([_AM_DEPENDENCIES],
+[AC_REQUIRE([AM_SET_DEPDIR])dnl
+AC_REQUIRE([AM_OUTPUT_DEPENDENCY_COMMANDS])dnl
+AC_REQUIRE([AM_MAKE_INCLUDE])dnl
+AC_REQUIRE([AM_DEP_TRACK])dnl
+
+m4_if([$1], [CC],   [depcc="$CC"   am_compiler_list=],
+      [$1], [CXX],  [depcc="$CXX"  am_compiler_list=],
+      [$1], [OBJC], [depcc="$OBJC" am_compiler_list='gcc3 gcc'],
+      [$1], [OBJCXX], [depcc="$OBJCXX" am_compiler_list='gcc3 gcc'],
+      [$1], [UPC],  [depcc="$UPC"  am_compiler_list=],
+      [$1], [GCJ],  [depcc="$GCJ"  am_compiler_list='gcc3 gcc'],
+                    [depcc="$$1"   am_compiler_list=])
+
+AC_CACHE_CHECK([dependency style of $depcc],
+               [am_cv_$1_dependencies_compiler_type],
+[if test -z "$AMDEP_TRUE" && test -f "$am_depcomp"; then
+  # We make a subdir and do the tests there.  Otherwise we can end up
+  # making bogus files that we don't know about and never remove.  For
+  # instance it was reported that on HP-UX the gcc test will end up
+  # making a dummy file named 'D' -- because '-MD' means "put the output
+  # in D".
+  rm -rf conftest.dir
+  mkdir conftest.dir
+  # Copy depcomp to subdir because otherwise we won't find it if we're
+  # using a relative directory.
+  cp "$am_depcomp" conftest.dir
+  cd conftest.dir
+  # We will build objects and dependencies in a subdirectory because
+  # it helps to detect inapplicable dependency modes.  For instance
+  # both Tru64's cc and ICC support -MD to output dependencies as a
+  # side effect of compilation, but ICC will put the dependencies in
+  # the current directory while Tru64 will put them in the object
+  # directory.
+  mkdir sub
+
+  am_cv_$1_dependencies_compiler_type=none
+  if test "$am_compiler_list" = ""; then
+     am_compiler_list=`sed -n ['s/^#*\([a-zA-Z0-9]*\))$/\1/p'] < ./depcomp`
+  fi
+  am__universal=false
+  m4_case([$1], [CC],
+    [case " $depcc " in #(
+     *\ -arch\ *\ -arch\ *) am__universal=true ;;
+     esac],
+    [CXX],
+    [case " $depcc " in #(
+     *\ -arch\ *\ -arch\ *) am__universal=true ;;
+     esac])
+
+  for depmode in $am_compiler_list; do
+    # Setup a source with many dependencies, because some compilers
+    # like to wrap large dependency lists on column 80 (with \), and
+    # we should not choose a depcomp mode which is confused by this.
+    #
+    # We need to recreate these files for each test, as the compiler may
+    # overwrite some of them when testing with obscure command lines.
+    # This happens at least with the AIX C compiler.
+    : > sub/conftest.c
+    for i in 1 2 3 4 5 6; do
+      echo '#include "conftst'$i'.h"' >> sub/conftest.c
+      # Using ": > sub/conftst$i.h" creates only sub/conftst1.h with
+      # Solaris 10 /bin/sh.
+      echo '/* dummy */' > sub/conftst$i.h
+    done
+    echo "${am__include} ${am__quote}sub/conftest.Po${am__quote}" > confmf
+
+    # We check with '-c' and '-o' for the sake of the "dashmstdout"
+    # mode.  It turns out that the SunPro C++ compiler does not properly
+    # handle '-M -o', and we need to detect this.  Also, some Intel
+    # versions had trouble with output in subdirs.
+    am__obj=sub/conftest.${OBJEXT-o}
+    am__minus_obj="-o $am__obj"
+    case $depmode in
+    gcc)
+      # This depmode causes a compiler race in universal mode.
+      test "$am__universal" = false || continue
+      ;;
+    nosideeffect)
+      # After this tag, mechanisms are not by side-effect, so they'll
+      # only be used when explicitly requested.
+      if test "x$enable_dependency_tracking" = xyes; then
+	continue
+      else
+	break
+      fi
+      ;;
+    msvc7 | msvc7msys | msvisualcpp | msvcmsys)
+      # This compiler won't grok '-c -o', but also, the minuso test has
+      # not run yet.  These depmodes are late enough in the game, and
+      # so weak that their functioning should not be impacted.
+      am__obj=conftest.${OBJEXT-o}
+      am__minus_obj=
+      ;;
+    none) break ;;
+    esac
+    if depmode=$depmode \
+       source=sub/conftest.c object=$am__obj \
+       depfile=sub/conftest.Po tmpdepfile=sub/conftest.TPo \
+       $SHELL ./depcomp $depcc -c $am__minus_obj sub/conftest.c \
+         >/dev/null 2>conftest.err &&
+       grep sub/conftst1.h sub/conftest.Po > /dev/null 2>&1 &&
+       grep sub/conftst6.h sub/conftest.Po > /dev/null 2>&1 &&
+       grep $am__obj sub/conftest.Po > /dev/null 2>&1 &&
+       ${MAKE-make} -s -f confmf > /dev/null 2>&1; then
+      # icc doesn't choke on unknown options, it will just issue warnings
+      # or remarks (even with -Werror).  So we grep stderr for any message
+      # that says an option was ignored or not supported.
+      # When given -MP, icc 7.0 and 7.1 complain thusly:
+      #   icc: Command line warning: ignoring option '-M'; no argument required
+      # The diagnosis changed in icc 8.0:
+      #   icc: Command line remark: option '-MP' not supported
+      if (grep 'ignoring option' conftest.err ||
+          grep 'not supported' conftest.err) >/dev/null 2>&1; then :; else
+        am_cv_$1_dependencies_compiler_type=$depmode
+        break
+      fi
+    fi
+  done
+
+  cd ..
+  rm -rf conftest.dir
+else
+  am_cv_$1_dependencies_compiler_type=none
+fi
+])
+AC_SUBST([$1DEPMODE], [depmode=$am_cv_$1_dependencies_compiler_type])
+AM_CONDITIONAL([am__fastdep$1], [
+  test "x$enable_dependency_tracking" != xno \
+  && test "$am_cv_$1_dependencies_compiler_type" = gcc3])
+])
+
+
+# AM_SET_DEPDIR
+# -------------
+# Choose a directory name for dependency files.
+# This macro is AC_REQUIREd in _AM_DEPENDENCIES.
+AC_DEFUN([AM_SET_DEPDIR],
+[AC_REQUIRE([AM_SET_LEADING_DOT])dnl
+AC_SUBST([DEPDIR], ["${am__leading_dot}deps"])dnl
+])
+
+
+# AM_DEP_TRACK
+# ------------
+AC_DEFUN([AM_DEP_TRACK],
+[AC_ARG_ENABLE([dependency-tracking], [dnl
+AS_HELP_STRING(
+  [--enable-dependency-tracking],
+  [do not reject slow dependency extractors])
+AS_HELP_STRING(
+  [--disable-dependency-tracking],
+  [speeds up one-time build])])
+if test "x$enable_dependency_tracking" != xno; then
+  am_depcomp="$ac_aux_dir/depcomp"
+  AMDEPBACKSLASH='\'
+  am__nodep='_no'
+fi
+AM_CONDITIONAL([AMDEP], [test "x$enable_dependency_tracking" != xno])
+AC_SUBST([AMDEPBACKSLASH])dnl
+_AM_SUBST_NOTMAKE([AMDEPBACKSLASH])dnl
+AC_SUBST([am__nodep])dnl
+_AM_SUBST_NOTMAKE([am__nodep])dnl
+])
+
+# Generate code to set up dependency tracking.              -*- Autoconf -*-
+
+# Copyright (C) 1999-2013 Free Software Foundation, Inc.
+#
+# This file is free software; the Free Software Foundation
+# gives unlimited permission to copy and/or distribute it,
+# with or without modifications, as long as this notice is preserved.
+
+
+# _AM_OUTPUT_DEPENDENCY_COMMANDS
+# ------------------------------
+AC_DEFUN([_AM_OUTPUT_DEPENDENCY_COMMANDS],
+[{
+  # Older Autoconf quotes --file arguments for eval, but not when files
+  # are listed without --file.  Let's play safe and only enable the eval
+  # if we detect the quoting.
+  case $CONFIG_FILES in
+  *\'*) eval set x "$CONFIG_FILES" ;;
+  *)   set x $CONFIG_FILES ;;
+  esac
+  shift
+  for mf
+  do
+    # Strip MF so we end up with the name of the file.
+    mf=`echo "$mf" | sed -e 's/:.*$//'`
+    # Check whether this is an Automake generated Makefile or not.
+    # We used to match only the files named 'Makefile.in', but
+    # some people rename them; so instead we look at the file content.
+    # Grep'ing the first line is not enough: some people post-process
+    # each Makefile.in and add a new line on top of each file to say so.
+    # Grep'ing the whole file is not good either: AIX grep has a line
+    # limit of 2048, but all sed's we know have understand at least 4000.
+    if sed -n 's,^#.*generated by automake.*,X,p' "$mf" | grep X >/dev/null 2>&1; then
+      dirpart=`AS_DIRNAME("$mf")`
+    else
+      continue
+    fi
+    # Extract the definition of DEPDIR, am__include, and am__quote
+    # from the Makefile without running 'make'.
+    DEPDIR=`sed -n 's/^DEPDIR = //p' < "$mf"`
+    test -z "$DEPDIR" && continue
+    am__include=`sed -n 's/^am__include = //p' < "$mf"`
+    test -z "$am__include" && continue
+    am__quote=`sed -n 's/^am__quote = //p' < "$mf"`
+    # Find all dependency output files, they are included files with
+    # $(DEPDIR) in their names.  We invoke sed twice because it is the
+    # simplest approach to changing $(DEPDIR) to its actual value in the
+    # expansion.
+    for file in `sed -n "
+      s/^$am__include $am__quote\(.*(DEPDIR).*\)$am__quote"'$/\1/p' <"$mf" | \
+	 sed -e 's/\$(DEPDIR)/'"$DEPDIR"'/g'`; do
+      # Make sure the directory exists.
+      test -f "$dirpart/$file" && continue
+      fdir=`AS_DIRNAME(["$file"])`
+      AS_MKDIR_P([$dirpart/$fdir])
+      # echo "creating $dirpart/$file"
+      echo '# dummy' > "$dirpart/$file"
+    done
+  done
+}
+])# _AM_OUTPUT_DEPENDENCY_COMMANDS
+
+
+# AM_OUTPUT_DEPENDENCY_COMMANDS
+# -----------------------------
+# This macro should only be invoked once -- use via AC_REQUIRE.
+#
+# This code is only required when automatic dependency tracking
+# is enabled.  FIXME.  This creates each '.P' file that we will
+# need in order to bootstrap the dependency handling code.
+AC_DEFUN([AM_OUTPUT_DEPENDENCY_COMMANDS],
+[AC_CONFIG_COMMANDS([depfiles],
+     [test x"$AMDEP_TRUE" != x"" || _AM_OUTPUT_DEPENDENCY_COMMANDS],
+     [AMDEP_TRUE="$AMDEP_TRUE" ac_aux_dir="$ac_aux_dir"])
+])
+
+# Do all the work for Automake.                             -*- Autoconf -*-
+
+# Copyright (C) 1996-2013 Free Software Foundation, Inc.
+#
+# This file is free software; the Free Software Foundation
+# gives unlimited permission to copy and/or distribute it,
+# with or without modifications, as long as this notice is preserved.
+
+# This macro actually does too much.  Some checks are only needed if
+# your package does certain things.  But this isn't really a big deal.
+
+# AM_INIT_AUTOMAKE(PACKAGE, VERSION, [NO-DEFINE])
+# AM_INIT_AUTOMAKE([OPTIONS])
+# -----------------------------------------------
+# The call with PACKAGE and VERSION arguments is the old style
+# call (pre autoconf-2.50), which is being phased out.  PACKAGE
+# and VERSION should now be passed to AC_INIT and removed from
+# the call to AM_INIT_AUTOMAKE.
+# We support both call styles for the transition.  After
+# the next Automake release, Autoconf can make the AC_INIT
+# arguments mandatory, and then we can depend on a new Autoconf
+# release and drop the old call support.
+AC_DEFUN([AM_INIT_AUTOMAKE],
+[AC_PREREQ([2.65])dnl
+dnl Autoconf wants to disallow AM_ names.  We explicitly allow
+dnl the ones we care about.
+m4_pattern_allow([^AM_[A-Z]+FLAGS$])dnl
+AC_REQUIRE([AM_SET_CURRENT_AUTOMAKE_VERSION])dnl
+AC_REQUIRE([AC_PROG_INSTALL])dnl
+if test "`cd $srcdir && pwd`" != "`pwd`"; then
+  # Use -I$(srcdir) only when $(srcdir) != ., so that make's output
+  # is not polluted with repeated "-I."
+  AC_SUBST([am__isrc], [' -I$(srcdir)'])_AM_SUBST_NOTMAKE([am__isrc])dnl
+  # test to see if srcdir already configured
+  if test -f $srcdir/config.status; then
+    AC_MSG_ERROR([source directory already configured; run "make distclean" there first])
+  fi
+fi
+
+# test whether we have cygpath
+if test -z "$CYGPATH_W"; then
+  if (cygpath --version) >/dev/null 2>/dev/null; then
+    CYGPATH_W='cygpath -w'
+  else
+    CYGPATH_W=echo
+  fi
+fi
+AC_SUBST([CYGPATH_W])
+
+# Define the identity of the package.
+dnl Distinguish between old-style and new-style calls.
+m4_ifval([$2],
+[AC_DIAGNOSE([obsolete],
+             [$0: two- and three-arguments forms are deprecated.])
+m4_ifval([$3], [_AM_SET_OPTION([no-define])])dnl
+ AC_SUBST([PACKAGE], [$1])dnl
+ AC_SUBST([VERSION], [$2])],
+[_AM_SET_OPTIONS([$1])dnl
+dnl Diagnose old-style AC_INIT with new-style AM_AUTOMAKE_INIT.
+m4_if(
+  m4_ifdef([AC_PACKAGE_NAME], [ok]):m4_ifdef([AC_PACKAGE_VERSION], [ok]),
+  [ok:ok],,
+  [m4_fatal([AC_INIT should be called with package and version arguments])])dnl
+ AC_SUBST([PACKAGE], ['AC_PACKAGE_TARNAME'])dnl
+ AC_SUBST([VERSION], ['AC_PACKAGE_VERSION'])])dnl
+
+_AM_IF_OPTION([no-define],,
+[AC_DEFINE_UNQUOTED([PACKAGE], ["$PACKAGE"], [Name of package])
+ AC_DEFINE_UNQUOTED([VERSION], ["$VERSION"], [Version number of package])])dnl
+
+# Some tools Automake needs.
+AC_REQUIRE([AM_SANITY_CHECK])dnl
+AC_REQUIRE([AC_ARG_PROGRAM])dnl
+AM_MISSING_PROG([ACLOCAL], [aclocal-${am__api_version}])
+AM_MISSING_PROG([AUTOCONF], [autoconf])
+AM_MISSING_PROG([AUTOMAKE], [automake-${am__api_version}])
+AM_MISSING_PROG([AUTOHEADER], [autoheader])
+AM_MISSING_PROG([MAKEINFO], [makeinfo])
+AC_REQUIRE([AM_PROG_INSTALL_SH])dnl
+AC_REQUIRE([AM_PROG_INSTALL_STRIP])dnl
+AC_REQUIRE([AC_PROG_MKDIR_P])dnl
+# For better backward compatibility.  To be removed once Automake 1.9.x
+# dies out for good.  For more background, see:
+# <http://lists.gnu.org/archive/html/automake/2012-07/msg00001.html>
+# <http://lists.gnu.org/archive/html/automake/2012-07/msg00014.html>
+AC_SUBST([mkdir_p], ['$(MKDIR_P)'])
+# We need awk for the "check" target.  The system "awk" is bad on
+# some platforms.
+AC_REQUIRE([AC_PROG_AWK])dnl
+AC_REQUIRE([AC_PROG_MAKE_SET])dnl
+AC_REQUIRE([AM_SET_LEADING_DOT])dnl
+_AM_IF_OPTION([tar-ustar], [_AM_PROG_TAR([ustar])],
+	      [_AM_IF_OPTION([tar-pax], [_AM_PROG_TAR([pax])],
+			     [_AM_PROG_TAR([v7])])])
+_AM_IF_OPTION([no-dependencies],,
+[AC_PROVIDE_IFELSE([AC_PROG_CC],
+		  [_AM_DEPENDENCIES([CC])],
+		  [m4_define([AC_PROG_CC],
+			     m4_defn([AC_PROG_CC])[_AM_DEPENDENCIES([CC])])])dnl
+AC_PROVIDE_IFELSE([AC_PROG_CXX],
+		  [_AM_DEPENDENCIES([CXX])],
+		  [m4_define([AC_PROG_CXX],
+			     m4_defn([AC_PROG_CXX])[_AM_DEPENDENCIES([CXX])])])dnl
+AC_PROVIDE_IFELSE([AC_PROG_OBJC],
+		  [_AM_DEPENDENCIES([OBJC])],
+		  [m4_define([AC_PROG_OBJC],
+			     m4_defn([AC_PROG_OBJC])[_AM_DEPENDENCIES([OBJC])])])dnl
+AC_PROVIDE_IFELSE([AC_PROG_OBJCXX],
+		  [_AM_DEPENDENCIES([OBJCXX])],
+		  [m4_define([AC_PROG_OBJCXX],
+			     m4_defn([AC_PROG_OBJCXX])[_AM_DEPENDENCIES([OBJCXX])])])dnl
+])
+AC_REQUIRE([AM_SILENT_RULES])dnl
+dnl The testsuite driver may need to know about EXEEXT, so add the
+dnl 'am__EXEEXT' conditional if _AM_COMPILER_EXEEXT was seen.  This
+dnl macro is hooked onto _AC_COMPILER_EXEEXT early, see below.
+AC_CONFIG_COMMANDS_PRE(dnl
+[m4_provide_if([_AM_COMPILER_EXEEXT],
+  [AM_CONDITIONAL([am__EXEEXT], [test -n "$EXEEXT"])])])dnl
+])
+
+dnl Hook into '_AC_COMPILER_EXEEXT' early to learn its expansion.  Do not
+dnl add the conditional right here, as _AC_COMPILER_EXEEXT may be further
+dnl mangled by Autoconf and run in a shell conditional statement.
+m4_define([_AC_COMPILER_EXEEXT],
+m4_defn([_AC_COMPILER_EXEEXT])[m4_provide([_AM_COMPILER_EXEEXT])])
+
+
+# When config.status generates a header, we must update the stamp-h file.
+# This file resides in the same directory as the config header
+# that is generated.  The stamp files are numbered to have different names.
+
+# Autoconf calls _AC_AM_CONFIG_HEADER_HOOK (when defined) in the
+# loop where config.status creates the headers, so we can generate
+# our stamp files there.
+AC_DEFUN([_AC_AM_CONFIG_HEADER_HOOK],
+[# Compute $1's index in $config_headers.
+_am_arg=$1
+_am_stamp_count=1
+for _am_header in $config_headers :; do
+  case $_am_header in
+    $_am_arg | $_am_arg:* )
+      break ;;
+    * )
+      _am_stamp_count=`expr $_am_stamp_count + 1` ;;
+  esac
+done
+echo "timestamp for $_am_arg" >`AS_DIRNAME(["$_am_arg"])`/stamp-h[]$_am_stamp_count])
+
+# Copyright (C) 2001-2013 Free Software Foundation, Inc.
+#
+# This file is free software; the Free Software Foundation
+# gives unlimited permission to copy and/or distribute it,
+# with or without modifications, as long as this notice is preserved.
+
+# AM_PROG_INSTALL_SH
+# ------------------
+# Define $install_sh.
+AC_DEFUN([AM_PROG_INSTALL_SH],
+[AC_REQUIRE([AM_AUX_DIR_EXPAND])dnl
+if test x"${install_sh}" != xset; then
+  case $am_aux_dir in
+  *\ * | *\	*)
+    install_sh="\${SHELL} '$am_aux_dir/install-sh'" ;;
+  *)
+    install_sh="\${SHELL} $am_aux_dir/install-sh"
+  esac
+fi
+AC_SUBST([install_sh])])
+
+# Copyright (C) 2003-2013 Free Software Foundation, Inc.
+#
+# This file is free software; the Free Software Foundation
+# gives unlimited permission to copy and/or distribute it,
+# with or without modifications, as long as this notice is preserved.
+
+# Check whether the underlying file-system supports filenames
+# with a leading dot.  For instance MS-DOS doesn't.
+AC_DEFUN([AM_SET_LEADING_DOT],
+[rm -rf .tst 2>/dev/null
+mkdir .tst 2>/dev/null
+if test -d .tst; then
+  am__leading_dot=.
+else
+  am__leading_dot=_
+fi
+rmdir .tst 2>/dev/null
+AC_SUBST([am__leading_dot])])
+
+# Add --enable-maintainer-mode option to configure.         -*- Autoconf -*-
+# From Jim Meyering
+
+# Copyright (C) 1996-2013 Free Software Foundation, Inc.
+#
+# This file is free software; the Free Software Foundation
+# gives unlimited permission to copy and/or distribute it,
+# with or without modifications, as long as this notice is preserved.
+
+# AM_MAINTAINER_MODE([DEFAULT-MODE])
+# ----------------------------------
+# Control maintainer-specific portions of Makefiles.
+# Default is to disable them, unless 'enable' is passed literally.
+# For symmetry, 'disable' may be passed as well.  Anyway, the user
+# can override the default with the --enable/--disable switch.
+AC_DEFUN([AM_MAINTAINER_MODE],
+[m4_case(m4_default([$1], [disable]),
+       [enable], [m4_define([am_maintainer_other], [disable])],
+       [disable], [m4_define([am_maintainer_other], [enable])],
+       [m4_define([am_maintainer_other], [enable])
+        m4_warn([syntax], [unexpected argument to AM@&t@_MAINTAINER_MODE: $1])])
+AC_MSG_CHECKING([whether to enable maintainer-specific portions of Makefiles])
+  dnl maintainer-mode's default is 'disable' unless 'enable' is passed
+  AC_ARG_ENABLE([maintainer-mode],
+    [AS_HELP_STRING([--]am_maintainer_other[-maintainer-mode],
+      am_maintainer_other[ make rules and dependencies not useful
+      (and sometimes confusing) to the casual installer])],
+    [USE_MAINTAINER_MODE=$enableval],
+    [USE_MAINTAINER_MODE=]m4_if(am_maintainer_other, [enable], [no], [yes]))
+  AC_MSG_RESULT([$USE_MAINTAINER_MODE])
+  AM_CONDITIONAL([MAINTAINER_MODE], [test $USE_MAINTAINER_MODE = yes])
+  MAINT=$MAINTAINER_MODE_TRUE
+  AC_SUBST([MAINT])dnl
+]
+)
+
+# Check to see how 'make' treats includes.	            -*- Autoconf -*-
+
+# Copyright (C) 2001-2013 Free Software Foundation, Inc.
+#
+# This file is free software; the Free Software Foundation
+# gives unlimited permission to copy and/or distribute it,
+# with or without modifications, as long as this notice is preserved.
+
+# AM_MAKE_INCLUDE()
+# -----------------
+# Check to see how make treats includes.
+AC_DEFUN([AM_MAKE_INCLUDE],
+[am_make=${MAKE-make}
+cat > confinc << 'END'
+am__doit:
+	@echo this is the am__doit target
+.PHONY: am__doit
+END
+# If we don't find an include directive, just comment out the code.
+AC_MSG_CHECKING([for style of include used by $am_make])
+am__include="#"
+am__quote=
+_am_result=none
+# First try GNU make style include.
+echo "include confinc" > confmf
+# Ignore all kinds of additional output from 'make'.
+case `$am_make -s -f confmf 2> /dev/null` in #(
+*the\ am__doit\ target*)
+  am__include=include
+  am__quote=
+  _am_result=GNU
+  ;;
+esac
+# Now try BSD make style include.
+if test "$am__include" = "#"; then
+   echo '.include "confinc"' > confmf
+   case `$am_make -s -f confmf 2> /dev/null` in #(
+   *the\ am__doit\ target*)
+     am__include=.include
+     am__quote="\""
+     _am_result=BSD
+     ;;
+   esac
+fi
+AC_SUBST([am__include])
+AC_SUBST([am__quote])
+AC_MSG_RESULT([$_am_result])
+rm -f confinc confmf
+])
+
+# Copyright (C) 1999-2013 Free Software Foundation, Inc.
+#
+# This file is free software; the Free Software Foundation
+# gives unlimited permission to copy and/or distribute it,
+# with or without modifications, as long as this notice is preserved.
+
+# AM_PROG_CC_C_O
+# --------------
+# Like AC_PROG_CC_C_O, but changed for automake.
+AC_DEFUN([AM_PROG_CC_C_O],
+[AC_REQUIRE([AC_PROG_CC_C_O])dnl
+AC_REQUIRE([AM_AUX_DIR_EXPAND])dnl
+AC_REQUIRE_AUX_FILE([compile])dnl
+# FIXME: we rely on the cache variable name because
+# there is no other way.
+set dummy $CC
+am_cc=`echo $[2] | sed ['s/[^a-zA-Z0-9_]/_/g;s/^[0-9]/_/']`
+eval am_t=\$ac_cv_prog_cc_${am_cc}_c_o
+if test "$am_t" != yes; then
+   # Losing compiler, so override with the script.
+   # FIXME: It is wrong to rewrite CC.
+   # But if we don't then we get into trouble of one sort or another.
+   # A longer-term fix would be to have automake use am__CC in this case,
+   # and then we could set am__CC="\$(top_srcdir)/compile \$(CC)"
+   CC="$am_aux_dir/compile $CC"
+fi
+dnl Make sure AC_PROG_CC is never called again, or it will override our
+dnl setting of CC.
+m4_define([AC_PROG_CC],
+          [m4_fatal([AC_PROG_CC cannot be called after AM_PROG_CC_C_O])])
+])
+
+# Fake the existence of programs that GNU maintainers use.  -*- Autoconf -*-
+
+# Copyright (C) 1997-2013 Free Software Foundation, Inc.
+#
+# This file is free software; the Free Software Foundation
+# gives unlimited permission to copy and/or distribute it,
+# with or without modifications, as long as this notice is preserved.
+
+# AM_MISSING_PROG(NAME, PROGRAM)
+# ------------------------------
+AC_DEFUN([AM_MISSING_PROG],
+[AC_REQUIRE([AM_MISSING_HAS_RUN])
+$1=${$1-"${am_missing_run}$2"}
+AC_SUBST($1)])
+
+# AM_MISSING_HAS_RUN
+# ------------------
+# Define MISSING if not defined so far and test if it is modern enough.
+# If it is, set am_missing_run to use it, otherwise, to nothing.
+AC_DEFUN([AM_MISSING_HAS_RUN],
+[AC_REQUIRE([AM_AUX_DIR_EXPAND])dnl
+AC_REQUIRE_AUX_FILE([missing])dnl
+if test x"${MISSING+set}" != xset; then
+  case $am_aux_dir in
+  *\ * | *\	*)
+    MISSING="\${SHELL} \"$am_aux_dir/missing\"" ;;
+  *)
+    MISSING="\${SHELL} $am_aux_dir/missing" ;;
+  esac
+fi
+# Use eval to expand $SHELL
+if eval "$MISSING --is-lightweight"; then
+  am_missing_run="$MISSING "
+else
+  am_missing_run=
+  AC_MSG_WARN(['missing' script is too old or missing])
+fi
+])
+
+# Helper functions for option handling.                     -*- Autoconf -*-
+
+# Copyright (C) 2001-2013 Free Software Foundation, Inc.
+#
+# This file is free software; the Free Software Foundation
+# gives unlimited permission to copy and/or distribute it,
+# with or without modifications, as long as this notice is preserved.
+
+# _AM_MANGLE_OPTION(NAME)
+# -----------------------
+AC_DEFUN([_AM_MANGLE_OPTION],
+[[_AM_OPTION_]m4_bpatsubst($1, [[^a-zA-Z0-9_]], [_])])
+
+# _AM_SET_OPTION(NAME)
+# --------------------
+# Set option NAME.  Presently that only means defining a flag for this option.
+AC_DEFUN([_AM_SET_OPTION],
+[m4_define(_AM_MANGLE_OPTION([$1]), [1])])
+
+# _AM_SET_OPTIONS(OPTIONS)
+# ------------------------
+# OPTIONS is a space-separated list of Automake options.
+AC_DEFUN([_AM_SET_OPTIONS],
+[m4_foreach_w([_AM_Option], [$1], [_AM_SET_OPTION(_AM_Option)])])
+
+# _AM_IF_OPTION(OPTION, IF-SET, [IF-NOT-SET])
+# -------------------------------------------
+# Execute IF-SET if OPTION is set, IF-NOT-SET otherwise.
+AC_DEFUN([_AM_IF_OPTION],
+[m4_ifset(_AM_MANGLE_OPTION([$1]), [$2], [$3])])
+
+# Check to make sure that the build environment is sane.    -*- Autoconf -*-
+
+# Copyright (C) 1996-2013 Free Software Foundation, Inc.
+#
+# This file is free software; the Free Software Foundation
+# gives unlimited permission to copy and/or distribute it,
+# with or without modifications, as long as this notice is preserved.
+
+# AM_SANITY_CHECK
+# ---------------
+AC_DEFUN([AM_SANITY_CHECK],
+[AC_MSG_CHECKING([whether build environment is sane])
+# Reject unsafe characters in $srcdir or the absolute working directory
+# name.  Accept space and tab only in the latter.
+am_lf='
+'
+case `pwd` in
+  *[[\\\"\#\$\&\'\`$am_lf]]*)
+    AC_MSG_ERROR([unsafe absolute working directory name]);;
+esac
+case $srcdir in
+  *[[\\\"\#\$\&\'\`$am_lf\ \	]]*)
+    AC_MSG_ERROR([unsafe srcdir value: '$srcdir']);;
+esac
+
+# Do 'set' in a subshell so we don't clobber the current shell's
+# arguments.  Must try -L first in case configure is actually a
+# symlink; some systems play weird games with the mod time of symlinks
+# (eg FreeBSD returns the mod time of the symlink's containing
+# directory).
+if (
+   am_has_slept=no
+   for am_try in 1 2; do
+     echo "timestamp, slept: $am_has_slept" > conftest.file
+     set X `ls -Lt "$srcdir/configure" conftest.file 2> /dev/null`
+     if test "$[*]" = "X"; then
+	# -L didn't work.
+	set X `ls -t "$srcdir/configure" conftest.file`
+     fi
+     if test "$[*]" != "X $srcdir/configure conftest.file" \
+	&& test "$[*]" != "X conftest.file $srcdir/configure"; then
+
+	# If neither matched, then we have a broken ls.  This can happen
+	# if, for instance, CONFIG_SHELL is bash and it inherits a
+	# broken ls alias from the environment.  This has actually
+	# happened.  Such a system could not be considered "sane".
+	AC_MSG_ERROR([ls -t appears to fail.  Make sure there is not a broken
+  alias in your environment])
+     fi
+     if test "$[2]" = conftest.file || test $am_try -eq 2; then
+       break
+     fi
+     # Just in case.
+     sleep 1
+     am_has_slept=yes
+   done
+   test "$[2]" = conftest.file
+   )
+then
+   # Ok.
+   :
+else
+   AC_MSG_ERROR([newly created file is older than distributed files!
+Check your system clock])
+fi
+AC_MSG_RESULT([yes])
+# If we didn't sleep, we still need to ensure time stamps of config.status and
+# generated files are strictly newer.
+am_sleep_pid=
+if grep 'slept: no' conftest.file >/dev/null 2>&1; then
+  ( sleep 1 ) &
+  am_sleep_pid=$!
+fi
+AC_CONFIG_COMMANDS_PRE(
+  [AC_MSG_CHECKING([that generated files are newer than configure])
+   if test -n "$am_sleep_pid"; then
+     # Hide warnings about reused PIDs.
+     wait $am_sleep_pid 2>/dev/null
+   fi
+   AC_MSG_RESULT([done])])
+rm -f conftest.file
+])
+
+# Copyright (C) 2009-2013 Free Software Foundation, Inc.
+#
+# This file is free software; the Free Software Foundation
+# gives unlimited permission to copy and/or distribute it,
+# with or without modifications, as long as this notice is preserved.
+
+# AM_SILENT_RULES([DEFAULT])
+# --------------------------
+# Enable less verbose build rules; with the default set to DEFAULT
+# ("yes" being less verbose, "no" or empty being verbose).
+AC_DEFUN([AM_SILENT_RULES],
+[AC_ARG_ENABLE([silent-rules], [dnl
+AS_HELP_STRING(
+  [--enable-silent-rules],
+  [less verbose build output (undo: "make V=1")])
+AS_HELP_STRING(
+  [--disable-silent-rules],
+  [verbose build output (undo: "make V=0")])dnl
+])
+case $enable_silent_rules in @%:@ (((
+  yes) AM_DEFAULT_VERBOSITY=0;;
+   no) AM_DEFAULT_VERBOSITY=1;;
+    *) AM_DEFAULT_VERBOSITY=m4_if([$1], [yes], [0], [1]);;
+esac
+dnl
+dnl A few 'make' implementations (e.g., NonStop OS and NextStep)
+dnl do not support nested variable expansions.
+dnl See automake bug#9928 and bug#10237.
+am_make=${MAKE-make}
+AC_CACHE_CHECK([whether $am_make supports nested variables],
+   [am_cv_make_support_nested_variables],
+   [if AS_ECHO([['TRUE=$(BAR$(V))
+BAR0=false
+BAR1=true
+V=1
+am__doit:
+	@$(TRUE)
+.PHONY: am__doit']]) | $am_make -f - >/dev/null 2>&1; then
+  am_cv_make_support_nested_variables=yes
+else
+  am_cv_make_support_nested_variables=no
+fi])
+if test $am_cv_make_support_nested_variables = yes; then
+  dnl Using '$V' instead of '$(V)' breaks IRIX make.
+  AM_V='$(V)'
+  AM_DEFAULT_V='$(AM_DEFAULT_VERBOSITY)'
+else
+  AM_V=$AM_DEFAULT_VERBOSITY
+  AM_DEFAULT_V=$AM_DEFAULT_VERBOSITY
+fi
+AC_SUBST([AM_V])dnl
+AM_SUBST_NOTMAKE([AM_V])dnl
+AC_SUBST([AM_DEFAULT_V])dnl
+AM_SUBST_NOTMAKE([AM_DEFAULT_V])dnl
+AC_SUBST([AM_DEFAULT_VERBOSITY])dnl
+AM_BACKSLASH='\'
+AC_SUBST([AM_BACKSLASH])dnl
+_AM_SUBST_NOTMAKE([AM_BACKSLASH])dnl
+])
+
+# Copyright (C) 2001-2013 Free Software Foundation, Inc.
+#
+# This file is free software; the Free Software Foundation
+# gives unlimited permission to copy and/or distribute it,
+# with or without modifications, as long as this notice is preserved.
+
+# AM_PROG_INSTALL_STRIP
+# ---------------------
+# One issue with vendor 'install' (even GNU) is that you can't
+# specify the program used to strip binaries.  This is especially
+# annoying in cross-compiling environments, where the build's strip
+# is unlikely to handle the host's binaries.
+# Fortunately install-sh will honor a STRIPPROG variable, so we
+# always use install-sh in "make install-strip", and initialize
+# STRIPPROG with the value of the STRIP variable (set by the user).
+AC_DEFUN([AM_PROG_INSTALL_STRIP],
+[AC_REQUIRE([AM_PROG_INSTALL_SH])dnl
+# Installed binaries are usually stripped using 'strip' when the user
+# run "make install-strip".  However 'strip' might not be the right
+# tool to use in cross-compilation environments, therefore Automake
+# will honor the 'STRIP' environment variable to overrule this program.
+dnl Don't test for $cross_compiling = yes, because it might be 'maybe'.
+if test "$cross_compiling" != no; then
+  AC_CHECK_TOOL([STRIP], [strip], :)
+fi
+INSTALL_STRIP_PROGRAM="\$(install_sh) -c -s"
+AC_SUBST([INSTALL_STRIP_PROGRAM])])
+
+# Copyright (C) 2006-2013 Free Software Foundation, Inc.
+#
+# This file is free software; the Free Software Foundation
+# gives unlimited permission to copy and/or distribute it,
+# with or without modifications, as long as this notice is preserved.
+
+# _AM_SUBST_NOTMAKE(VARIABLE)
+# ---------------------------
+# Prevent Automake from outputting VARIABLE = @VARIABLE@ in Makefile.in.
+# This macro is traced by Automake.
+AC_DEFUN([_AM_SUBST_NOTMAKE])
+
+# AM_SUBST_NOTMAKE(VARIABLE)
+# --------------------------
+# Public sister of _AM_SUBST_NOTMAKE.
+AC_DEFUN([AM_SUBST_NOTMAKE], [_AM_SUBST_NOTMAKE($@)])
+
+# Check how to create a tarball.                            -*- Autoconf -*-
+
+# Copyright (C) 2004-2013 Free Software Foundation, Inc.
+#
+# This file is free software; the Free Software Foundation
+# gives unlimited permission to copy and/or distribute it,
+# with or without modifications, as long as this notice is preserved.
+
+# _AM_PROG_TAR(FORMAT)
+# --------------------
+# Check how to create a tarball in format FORMAT.
+# FORMAT should be one of 'v7', 'ustar', or 'pax'.
+#
+# Substitute a variable $(am__tar) that is a command
+# writing to stdout a FORMAT-tarball containing the directory
+# $tardir.
+#     tardir=directory && $(am__tar) > result.tar
+#
+# Substitute a variable $(am__untar) that extract such
+# a tarball read from stdin.
+#     $(am__untar) < result.tar
+#
+AC_DEFUN([_AM_PROG_TAR],
+[# Always define AMTAR for backward compatibility.  Yes, it's still used
+# in the wild :-(  We should find a proper way to deprecate it ...
+AC_SUBST([AMTAR], ['$${TAR-tar}'])
+
+# We'll loop over all known methods to create a tar archive until one works.
+_am_tools='gnutar m4_if([$1], [ustar], [plaintar]) pax cpio none'
+
+m4_if([$1], [v7],
+  [am__tar='$${TAR-tar} chof - "$$tardir"' am__untar='$${TAR-tar} xf -'],
+
+  [m4_case([$1],
+    [ustar],
+     [# The POSIX 1988 'ustar' format is defined with fixed-size fields.
+      # There is notably a 21 bits limit for the UID and the GID.  In fact,
+      # the 'pax' utility can hang on bigger UID/GID (see automake bug#8343
+      # and bug#13588).
+      am_max_uid=2097151 # 2^21 - 1
+      am_max_gid=$am_max_uid
+      # The $UID and $GID variables are not portable, so we need to resort
+      # to the POSIX-mandated id(1) utility.  Errors in the 'id' calls
+      # below are definitely unexpected, so allow the users to see them
+      # (that is, avoid stderr redirection).
+      am_uid=`id -u || echo unknown`
+      am_gid=`id -g || echo unknown`
+      AC_MSG_CHECKING([whether UID '$am_uid' is supported by ustar format])
+      if test $am_uid -le $am_max_uid; then
+         AC_MSG_RESULT([yes])
+      else
+         AC_MSG_RESULT([no])
+         _am_tools=none
+      fi
+      AC_MSG_CHECKING([whether GID '$am_gid' is supported by ustar format])
+      if test $am_gid -le $am_max_gid; then
+         AC_MSG_RESULT([yes])
+      else
+        AC_MSG_RESULT([no])
+        _am_tools=none
+      fi],
+
+  [pax],
+    [],
+
+  [m4_fatal([Unknown tar format])])
+
+  AC_MSG_CHECKING([how to create a $1 tar archive])
+
+  # Go ahead even if we have the value already cached.  We do so because we
+  # need to set the values for the 'am__tar' and 'am__untar' variables.
+  _am_tools=${am_cv_prog_tar_$1-$_am_tools}
+
+  for _am_tool in $_am_tools; do
+    case $_am_tool in
+    gnutar)
+      for _am_tar in tar gnutar gtar; do
+        AM_RUN_LOG([$_am_tar --version]) && break
+      done
+      am__tar="$_am_tar --format=m4_if([$1], [pax], [posix], [$1]) -chf - "'"$$tardir"'
+      am__tar_="$_am_tar --format=m4_if([$1], [pax], [posix], [$1]) -chf - "'"$tardir"'
+      am__untar="$_am_tar -xf -"
+      ;;
+    plaintar)
+      # Must skip GNU tar: if it does not support --format= it doesn't create
+      # ustar tarball either.
+      (tar --version) >/dev/null 2>&1 && continue
+      am__tar='tar chf - "$$tardir"'
+      am__tar_='tar chf - "$tardir"'
+      am__untar='tar xf -'
+      ;;
+    pax)
+      am__tar='pax -L -x $1 -w "$$tardir"'
+      am__tar_='pax -L -x $1 -w "$tardir"'
+      am__untar='pax -r'
+      ;;
+    cpio)
+      am__tar='find "$$tardir" -print | cpio -o -H $1 -L'
+      am__tar_='find "$tardir" -print | cpio -o -H $1 -L'
+      am__untar='cpio -i -H $1 -d'
+      ;;
+    none)
+      am__tar=false
+      am__tar_=false
+      am__untar=false
+      ;;
+    esac
+
+    # If the value was cached, stop now.  We just wanted to have am__tar
+    # and am__untar set.
+    test -n "${am_cv_prog_tar_$1}" && break
+
+    # tar/untar a dummy directory, and stop if the command works.
+    rm -rf conftest.dir
+    mkdir conftest.dir
+    echo GrepMe > conftest.dir/file
+    AM_RUN_LOG([tardir=conftest.dir && eval $am__tar_ >conftest.tar])
+    rm -rf conftest.dir
+    if test -s conftest.tar; then
+      AM_RUN_LOG([$am__untar <conftest.tar])
+      AM_RUN_LOG([cat conftest.dir/file])
+      grep GrepMe conftest.dir/file >/dev/null 2>&1 && break
+    fi
+  done
+  rm -rf conftest.dir
+
+  AC_CACHE_VAL([am_cv_prog_tar_$1], [am_cv_prog_tar_$1=$_am_tool])
+  AC_MSG_RESULT([$am_cv_prog_tar_$1])])
+
+AC_SUBST([am__tar])
+AC_SUBST([am__untar])
+]) # _AM_PROG_TAR
+
diff --git a/autogen.sh b/autogen.sh
new file mode 100755
index 0000000000..8261a2c136
--- /dev/null
+++ b/autogen.sh
@@ -0,0 +1 @@
+aclocal && autoheader && automake --add-missing --gnu --copy && autoconf
diff --git a/blake.c b/blake.c
new file mode 100644
index 0000000000..a9043e9780
--- /dev/null
+++ b/blake.c
@@ -0,0 +1,1120 @@
+/* $Id: blake.c 252 2011-06-07 17:55:14Z tp $ */
+/*
+ * BLAKE implementation.
+ *
+ * ==========================(LICENSE BEGIN)============================
+ *
+ * Copyright (c) 2007-2010  Projet RNRT SAPHIR
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining
+ * a copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sublicense, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be
+ * included in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ * IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY
+ * CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * ===========================(LICENSE END)=============================
+ *
+ * @author   Thomas Pornin <thomas.pornin@cryptolog.com>
+ */
+
+#include <stddef.h>
+#include <string.h>
+#include <limits.h>
+
+#include "sph_blake.h"
+
+#ifdef __cplusplus
+extern "C"{
+#endif
+
+#if SPH_SMALL_FOOTPRINT && !defined SPH_SMALL_FOOTPRINT_BLAKE
+#define SPH_SMALL_FOOTPRINT_BLAKE   1
+#endif
+
+#if SPH_SMALL_FOOTPRINT_BLAKE
+#define SPH_COMPACT_BLAKE_32   1
+#endif
+
+#if SPH_64 && (SPH_SMALL_FOOTPRINT_BLAKE || !SPH_64_TRUE)
+#define SPH_COMPACT_BLAKE_64   1
+#endif
+
+#ifdef _MSC_VER
+#pragma warning (disable: 4146)
+#endif
+
+static const sph_u32 IV224[8] = {
+	SPH_C32(0xC1059ED8), SPH_C32(0x367CD507),
+	SPH_C32(0x3070DD17), SPH_C32(0xF70E5939),
+	SPH_C32(0xFFC00B31), SPH_C32(0x68581511),
+	SPH_C32(0x64F98FA7), SPH_C32(0xBEFA4FA4)
+};
+
+static const sph_u32 IV256[8] = {
+	SPH_C32(0x6A09E667), SPH_C32(0xBB67AE85),
+	SPH_C32(0x3C6EF372), SPH_C32(0xA54FF53A),
+	SPH_C32(0x510E527F), SPH_C32(0x9B05688C),
+	SPH_C32(0x1F83D9AB), SPH_C32(0x5BE0CD19)
+};
+
+#if SPH_64
+
+static const sph_u64 IV384[8] = {
+	SPH_C64(0xCBBB9D5DC1059ED8), SPH_C64(0x629A292A367CD507),
+	SPH_C64(0x9159015A3070DD17), SPH_C64(0x152FECD8F70E5939),
+	SPH_C64(0x67332667FFC00B31), SPH_C64(0x8EB44A8768581511),
+	SPH_C64(0xDB0C2E0D64F98FA7), SPH_C64(0x47B5481DBEFA4FA4)
+};
+
+static const sph_u64 IV512[8] = {
+	SPH_C64(0x6A09E667F3BCC908), SPH_C64(0xBB67AE8584CAA73B),
+	SPH_C64(0x3C6EF372FE94F82B), SPH_C64(0xA54FF53A5F1D36F1),
+	SPH_C64(0x510E527FADE682D1), SPH_C64(0x9B05688C2B3E6C1F),
+	SPH_C64(0x1F83D9ABFB41BD6B), SPH_C64(0x5BE0CD19137E2179)
+};
+
+#endif
+
+#if SPH_COMPACT_BLAKE_32 || SPH_COMPACT_BLAKE_64
+
+static const unsigned sigma[16][16] = {
+	{  0,  1,  2,  3,  4,  5,  6,  7,  8,  9, 10, 11, 12, 13, 14, 15 },
+	{ 14, 10,  4,  8,  9, 15, 13,  6,  1, 12,  0,  2, 11,  7,  5,  3 },
+	{ 11,  8, 12,  0,  5,  2, 15, 13, 10, 14,  3,  6,  7,  1,  9,  4 },
+	{  7,  9,  3,  1, 13, 12, 11, 14,  2,  6,  5, 10,  4,  0, 15,  8 },
+	{  9,  0,  5,  7,  2,  4, 10, 15, 14,  1, 11, 12,  6,  8,  3, 13 },
+	{  2, 12,  6, 10,  0, 11,  8,  3,  4, 13,  7,  5, 15, 14,  1,  9 },
+	{ 12,  5,  1, 15, 14, 13,  4, 10,  0,  7,  6,  3,  9,  2,  8, 11 },
+	{ 13, 11,  7, 14, 12,  1,  3,  9,  5,  0, 15,  4,  8,  6,  2, 10 },
+	{  6, 15, 14,  9, 11,  3,  0,  8, 12,  2, 13,  7,  1,  4, 10,  5 },
+	{ 10,  2,  8,  4,  7,  6,  1,  5, 15, 11,  9, 14,  3, 12, 13,  0 },
+	{  0,  1,  2,  3,  4,  5,  6,  7,  8,  9, 10, 11, 12, 13, 14, 15 },
+	{ 14, 10,  4,  8,  9, 15, 13,  6,  1, 12,  0,  2, 11,  7,  5,  3 },
+	{ 11,  8, 12,  0,  5,  2, 15, 13, 10, 14,  3,  6,  7,  1,  9,  4 },
+	{  7,  9,  3,  1, 13, 12, 11, 14,  2,  6,  5, 10,  4,  0, 15,  8 },
+	{  9,  0,  5,  7,  2,  4, 10, 15, 14,  1, 11, 12,  6,  8,  3, 13 },
+	{  2, 12,  6, 10,  0, 11,  8,  3,  4, 13,  7,  5, 15, 14,  1,  9 }
+};
+
+/*
+  0  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15
+ 14 10  4  8  9 15 13  6  1 12  0  2 11  7  5  3
+ 11  8 12  0  5  2 15 13 10 14  3  6  7  1  9  4
+  7  9  3  1 13 12 11 14  2  6  5 10  4  0 15  8
+  9  0  5  7  2  4 10 15 14  1 11 12  6  8  3 13
+  2 12  6 10  0 11  8  3  4 13  7  5 15 14  1  9
+ 12  5  1 15 14 13  4 10  0  7  6  3  9  2  8 11
+ 13 11  7 14 12  1  3  9  5  0 15  4  8  6  2 10
+  6 15 14  9 11  3  0  8 12  2 13  7  1  4 10  5
+ 10  2  8  4  7  6  1  5 15 11  9 14  3 12 13  0
+*/
+#endif
+
+#define Z00   0
+#define Z01   1
+#define Z02   2
+#define Z03   3
+#define Z04   4
+#define Z05   5
+#define Z06   6
+#define Z07   7
+#define Z08   8
+#define Z09   9
+#define Z0A   A
+#define Z0B   B
+#define Z0C   C
+#define Z0D   D
+#define Z0E   E
+#define Z0F   F
+
+#define Z10   E
+#define Z11   A
+#define Z12   4
+#define Z13   8
+#define Z14   9
+#define Z15   F
+#define Z16   D
+#define Z17   6
+#define Z18   1
+#define Z19   C
+#define Z1A   0
+#define Z1B   2
+#define Z1C   B
+#define Z1D   7
+#define Z1E   5
+#define Z1F   3
+
+#define Z20   B
+#define Z21   8
+#define Z22   C
+#define Z23   0
+#define Z24   5
+#define Z25   2
+#define Z26   F
+#define Z27   D
+#define Z28   A
+#define Z29   E
+#define Z2A   3
+#define Z2B   6
+#define Z2C   7
+#define Z2D   1
+#define Z2E   9
+#define Z2F   4
+
+#define Z30   7
+#define Z31   9
+#define Z32   3
+#define Z33   1
+#define Z34   D
+#define Z35   C
+#define Z36   B
+#define Z37   E
+#define Z38   2
+#define Z39   6
+#define Z3A   5
+#define Z3B   A
+#define Z3C   4
+#define Z3D   0
+#define Z3E   F
+#define Z3F   8
+
+#define Z40   9
+#define Z41   0
+#define Z42   5
+#define Z43   7
+#define Z44   2
+#define Z45   4
+#define Z46   A
+#define Z47   F
+#define Z48   E
+#define Z49   1
+#define Z4A   B
+#define Z4B   C
+#define Z4C   6
+#define Z4D   8
+#define Z4E   3
+#define Z4F   D
+
+#define Z50   2
+#define Z51   C
+#define Z52   6
+#define Z53   A
+#define Z54   0
+#define Z55   B
+#define Z56   8
+#define Z57   3
+#define Z58   4
+#define Z59   D
+#define Z5A   7
+#define Z5B   5
+#define Z5C   F
+#define Z5D   E
+#define Z5E   1
+#define Z5F   9
+
+#define Z60   C
+#define Z61   5
+#define Z62   1
+#define Z63   F
+#define Z64   E
+#define Z65   D
+#define Z66   4
+#define Z67   A
+#define Z68   0
+#define Z69   7
+#define Z6A   6
+#define Z6B   3
+#define Z6C   9
+#define Z6D   2
+#define Z6E   8
+#define Z6F   B
+
+#define Z70   D
+#define Z71   B
+#define Z72   7
+#define Z73   E
+#define Z74   C
+#define Z75   1
+#define Z76   3
+#define Z77   9
+#define Z78   5
+#define Z79   0
+#define Z7A   F
+#define Z7B   4
+#define Z7C   8
+#define Z7D   6
+#define Z7E   2
+#define Z7F   A
+
+#define Z80   6
+#define Z81   F
+#define Z82   E
+#define Z83   9
+#define Z84   B
+#define Z85   3
+#define Z86   0
+#define Z87   8
+#define Z88   C
+#define Z89   2
+#define Z8A   D
+#define Z8B   7
+#define Z8C   1
+#define Z8D   4
+#define Z8E   A
+#define Z8F   5
+
+#define Z90   A
+#define Z91   2
+#define Z92   8
+#define Z93   4
+#define Z94   7
+#define Z95   6
+#define Z96   1
+#define Z97   5
+#define Z98   F
+#define Z99   B
+#define Z9A   9
+#define Z9B   E
+#define Z9C   3
+#define Z9D   C
+#define Z9E   D
+#define Z9F   0
+
+#define Mx(r, i)    Mx_(Z ## r ## i)
+#define Mx_(n)      Mx__(n)
+#define Mx__(n)     M ## n
+
+#define CSx(r, i)   CSx_(Z ## r ## i)
+#define CSx_(n)     CSx__(n)
+#define CSx__(n)    CS ## n
+
+#define CS0   SPH_C32(0x243F6A88)
+#define CS1   SPH_C32(0x85A308D3)
+#define CS2   SPH_C32(0x13198A2E)
+#define CS3   SPH_C32(0x03707344)
+#define CS4   SPH_C32(0xA4093822)
+#define CS5   SPH_C32(0x299F31D0)
+#define CS6   SPH_C32(0x082EFA98)
+#define CS7   SPH_C32(0xEC4E6C89)
+#define CS8   SPH_C32(0x452821E6)
+#define CS9   SPH_C32(0x38D01377)
+#define CSA   SPH_C32(0xBE5466CF)
+#define CSB   SPH_C32(0x34E90C6C)
+#define CSC   SPH_C32(0xC0AC29B7)
+#define CSD   SPH_C32(0xC97C50DD)
+#define CSE   SPH_C32(0x3F84D5B5)
+#define CSF   SPH_C32(0xB5470917)
+
+#if SPH_COMPACT_BLAKE_32
+
+static const sph_u32 CS[16] = {
+	SPH_C32(0x243F6A88), SPH_C32(0x85A308D3),
+	SPH_C32(0x13198A2E), SPH_C32(0x03707344),
+	SPH_C32(0xA4093822), SPH_C32(0x299F31D0),
+	SPH_C32(0x082EFA98), SPH_C32(0xEC4E6C89),
+	SPH_C32(0x452821E6), SPH_C32(0x38D01377),
+	SPH_C32(0xBE5466CF), SPH_C32(0x34E90C6C),
+	SPH_C32(0xC0AC29B7), SPH_C32(0xC97C50DD),
+	SPH_C32(0x3F84D5B5), SPH_C32(0xB5470917)
+};
+
+#endif
+
+#if SPH_64
+
+#define CBx(r, i)   CBx_(Z ## r ## i)
+#define CBx_(n)     CBx__(n)
+#define CBx__(n)    CB ## n
+
+#define CB0   SPH_C64(0x243F6A8885A308D3)
+#define CB1   SPH_C64(0x13198A2E03707344)
+#define CB2   SPH_C64(0xA4093822299F31D0)
+#define CB3   SPH_C64(0x082EFA98EC4E6C89)
+#define CB4   SPH_C64(0x452821E638D01377)
+#define CB5   SPH_C64(0xBE5466CF34E90C6C)
+#define CB6   SPH_C64(0xC0AC29B7C97C50DD)
+#define CB7   SPH_C64(0x3F84D5B5B5470917)
+#define CB8   SPH_C64(0x9216D5D98979FB1B)
+#define CB9   SPH_C64(0xD1310BA698DFB5AC)
+#define CBA   SPH_C64(0x2FFD72DBD01ADFB7)
+#define CBB   SPH_C64(0xB8E1AFED6A267E96)
+#define CBC   SPH_C64(0xBA7C9045F12C7F99)
+#define CBD   SPH_C64(0x24A19947B3916CF7)
+#define CBE   SPH_C64(0x0801F2E2858EFC16)
+#define CBF   SPH_C64(0x636920D871574E69)
+
+#if SPH_COMPACT_BLAKE_64
+
+static const sph_u64 CB[16] = {
+	SPH_C64(0x243F6A8885A308D3), SPH_C64(0x13198A2E03707344),
+	SPH_C64(0xA4093822299F31D0), SPH_C64(0x082EFA98EC4E6C89),
+	SPH_C64(0x452821E638D01377), SPH_C64(0xBE5466CF34E90C6C),
+	SPH_C64(0xC0AC29B7C97C50DD), SPH_C64(0x3F84D5B5B5470917),
+	SPH_C64(0x9216D5D98979FB1B), SPH_C64(0xD1310BA698DFB5AC),
+	SPH_C64(0x2FFD72DBD01ADFB7), SPH_C64(0xB8E1AFED6A267E96),
+	SPH_C64(0xBA7C9045F12C7F99), SPH_C64(0x24A19947B3916CF7),
+	SPH_C64(0x0801F2E2858EFC16), SPH_C64(0x636920D871574E69)
+};
+
+#endif
+
+#endif
+
+#define GS(m0, m1, c0, c1, a, b, c, d)   do { \
+		a = SPH_T32(a + b + (m0 ^ c1)); \
+		d = SPH_ROTR32(d ^ a, 16); \
+		c = SPH_T32(c + d); \
+		b = SPH_ROTR32(b ^ c, 12); \
+		a = SPH_T32(a + b + (m1 ^ c0)); \
+		d = SPH_ROTR32(d ^ a, 8); \
+		c = SPH_T32(c + d); \
+		b = SPH_ROTR32(b ^ c, 7); \
+	} while (0)
+
+#if SPH_COMPACT_BLAKE_32
+
+#define ROUND_S(r)   do { \
+		GS(M[sigma[r][0x0]], M[sigma[r][0x1]], \
+			CS[sigma[r][0x0]], CS[sigma[r][0x1]], V0, V4, V8, VC); \
+		GS(M[sigma[r][0x2]], M[sigma[r][0x3]], \
+			CS[sigma[r][0x2]], CS[sigma[r][0x3]], V1, V5, V9, VD); \
+		GS(M[sigma[r][0x4]], M[sigma[r][0x5]], \
+			CS[sigma[r][0x4]], CS[sigma[r][0x5]], V2, V6, VA, VE); \
+		GS(M[sigma[r][0x6]], M[sigma[r][0x7]], \
+			CS[sigma[r][0x6]], CS[sigma[r][0x7]], V3, V7, VB, VF); \
+		GS(M[sigma[r][0x8]], M[sigma[r][0x9]], \
+			CS[sigma[r][0x8]], CS[sigma[r][0x9]], V0, V5, VA, VF); \
+		GS(M[sigma[r][0xA]], M[sigma[r][0xB]], \
+			CS[sigma[r][0xA]], CS[sigma[r][0xB]], V1, V6, VB, VC); \
+		GS(M[sigma[r][0xC]], M[sigma[r][0xD]], \
+			CS[sigma[r][0xC]], CS[sigma[r][0xD]], V2, V7, V8, VD); \
+		GS(M[sigma[r][0xE]], M[sigma[r][0xF]], \
+			CS[sigma[r][0xE]], CS[sigma[r][0xF]], V3, V4, V9, VE); \
+	} while (0)
+
+#else
+
+#define ROUND_S(r)   do { \
+		GS(Mx(r, 0), Mx(r, 1), CSx(r, 0), CSx(r, 1), V0, V4, V8, VC); \
+		GS(Mx(r, 2), Mx(r, 3), CSx(r, 2), CSx(r, 3), V1, V5, V9, VD); \
+		GS(Mx(r, 4), Mx(r, 5), CSx(r, 4), CSx(r, 5), V2, V6, VA, VE); \
+		GS(Mx(r, 6), Mx(r, 7), CSx(r, 6), CSx(r, 7), V3, V7, VB, VF); \
+		GS(Mx(r, 8), Mx(r, 9), CSx(r, 8), CSx(r, 9), V0, V5, VA, VF); \
+		GS(Mx(r, A), Mx(r, B), CSx(r, A), CSx(r, B), V1, V6, VB, VC); \
+		GS(Mx(r, C), Mx(r, D), CSx(r, C), CSx(r, D), V2, V7, V8, VD); \
+		GS(Mx(r, E), Mx(r, F), CSx(r, E), CSx(r, F), V3, V4, V9, VE); \
+	} while (0)
+
+#endif
+
+#if SPH_64
+
+#define GB(m0, m1, c0, c1, a, b, c, d)   do { \
+		a = SPH_T64(a + b + (m0 ^ c1)); \
+		d = SPH_ROTR64(d ^ a, 32); \
+		c = SPH_T64(c + d); \
+		b = SPH_ROTR64(b ^ c, 25); \
+		a = SPH_T64(a + b + (m1 ^ c0)); \
+		d = SPH_ROTR64(d ^ a, 16); \
+		c = SPH_T64(c + d); \
+		b = SPH_ROTR64(b ^ c, 11); \
+	} while (0)
+
+#if SPH_COMPACT_BLAKE_64
+
+#define ROUND_B(r)   do { \
+		GB(M[sigma[r][0x0]], M[sigma[r][0x1]], \
+			CB[sigma[r][0x0]], CB[sigma[r][0x1]], V0, V4, V8, VC); \
+		GB(M[sigma[r][0x2]], M[sigma[r][0x3]], \
+			CB[sigma[r][0x2]], CB[sigma[r][0x3]], V1, V5, V9, VD); \
+		GB(M[sigma[r][0x4]], M[sigma[r][0x5]], \
+			CB[sigma[r][0x4]], CB[sigma[r][0x5]], V2, V6, VA, VE); \
+		GB(M[sigma[r][0x6]], M[sigma[r][0x7]], \
+			CB[sigma[r][0x6]], CB[sigma[r][0x7]], V3, V7, VB, VF); \
+		GB(M[sigma[r][0x8]], M[sigma[r][0x9]], \
+			CB[sigma[r][0x8]], CB[sigma[r][0x9]], V0, V5, VA, VF); \
+		GB(M[sigma[r][0xA]], M[sigma[r][0xB]], \
+			CB[sigma[r][0xA]], CB[sigma[r][0xB]], V1, V6, VB, VC); \
+		GB(M[sigma[r][0xC]], M[sigma[r][0xD]], \
+			CB[sigma[r][0xC]], CB[sigma[r][0xD]], V2, V7, V8, VD); \
+		GB(M[sigma[r][0xE]], M[sigma[r][0xF]], \
+			CB[sigma[r][0xE]], CB[sigma[r][0xF]], V3, V4, V9, VE); \
+	} while (0)
+
+#else
+
+#define ROUND_B(r)   do { \
+		GB(Mx(r, 0), Mx(r, 1), CBx(r, 0), CBx(r, 1), V0, V4, V8, VC); \
+		GB(Mx(r, 2), Mx(r, 3), CBx(r, 2), CBx(r, 3), V1, V5, V9, VD); \
+		GB(Mx(r, 4), Mx(r, 5), CBx(r, 4), CBx(r, 5), V2, V6, VA, VE); \
+		GB(Mx(r, 6), Mx(r, 7), CBx(r, 6), CBx(r, 7), V3, V7, VB, VF); \
+		GB(Mx(r, 8), Mx(r, 9), CBx(r, 8), CBx(r, 9), V0, V5, VA, VF); \
+		GB(Mx(r, A), Mx(r, B), CBx(r, A), CBx(r, B), V1, V6, VB, VC); \
+		GB(Mx(r, C), Mx(r, D), CBx(r, C), CBx(r, D), V2, V7, V8, VD); \
+		GB(Mx(r, E), Mx(r, F), CBx(r, E), CBx(r, F), V3, V4, V9, VE); \
+	} while (0)
+
+#endif
+
+#endif
+
+#define DECL_STATE32 \
+	sph_u32 H0, H1, H2, H3, H4, H5, H6, H7; \
+	sph_u32 S0, S1, S2, S3, T0, T1;
+
+#define READ_STATE32(state)   do { \
+		H0 = (state)->H[0]; \
+		H1 = (state)->H[1]; \
+		H2 = (state)->H[2]; \
+		H3 = (state)->H[3]; \
+		H4 = (state)->H[4]; \
+		H5 = (state)->H[5]; \
+		H6 = (state)->H[6]; \
+		H7 = (state)->H[7]; \
+		S0 = (state)->S[0]; \
+		S1 = (state)->S[1]; \
+		S2 = (state)->S[2]; \
+		S3 = (state)->S[3]; \
+		T0 = (state)->T0; \
+		T1 = (state)->T1; \
+	} while (0)
+
+#define WRITE_STATE32(state)   do { \
+		(state)->H[0] = H0; \
+		(state)->H[1] = H1; \
+		(state)->H[2] = H2; \
+		(state)->H[3] = H3; \
+		(state)->H[4] = H4; \
+		(state)->H[5] = H5; \
+		(state)->H[6] = H6; \
+		(state)->H[7] = H7; \
+		(state)->S[0] = S0; \
+		(state)->S[1] = S1; \
+		(state)->S[2] = S2; \
+		(state)->S[3] = S3; \
+		(state)->T0 = T0; \
+		(state)->T1 = T1; \
+	} while (0)
+
+#if SPH_COMPACT_BLAKE_32
+
+#define COMPRESS32   do { \
+		sph_u32 M[16]; \
+		sph_u32 V0, V1, V2, V3, V4, V5, V6, V7; \
+		sph_u32 V8, V9, VA, VB, VC, VD, VE, VF; \
+		unsigned r; \
+		V0 = H0; \
+		V1 = H1; \
+		V2 = H2; \
+		V3 = H3; \
+		V4 = H4; \
+		V5 = H5; \
+		V6 = H6; \
+		V7 = H7; \
+		V8 = S0 ^ CS0; \
+		V9 = S1 ^ CS1; \
+		VA = S2 ^ CS2; \
+		VB = S3 ^ CS3; \
+		VC = T0 ^ CS4; \
+		VD = T0 ^ CS5; \
+		VE = T1 ^ CS6; \
+		VF = T1 ^ CS7; \
+		M[0x0] = sph_dec32be_aligned(buf +  0); \
+		M[0x1] = sph_dec32be_aligned(buf +  4); \
+		M[0x2] = sph_dec32be_aligned(buf +  8); \
+		M[0x3] = sph_dec32be_aligned(buf + 12); \
+		M[0x4] = sph_dec32be_aligned(buf + 16); \
+		M[0x5] = sph_dec32be_aligned(buf + 20); \
+		M[0x6] = sph_dec32be_aligned(buf + 24); \
+		M[0x7] = sph_dec32be_aligned(buf + 28); \
+		M[0x8] = sph_dec32be_aligned(buf + 32); \
+		M[0x9] = sph_dec32be_aligned(buf + 36); \
+		M[0xA] = sph_dec32be_aligned(buf + 40); \
+		M[0xB] = sph_dec32be_aligned(buf + 44); \
+		M[0xC] = sph_dec32be_aligned(buf + 48); \
+		M[0xD] = sph_dec32be_aligned(buf + 52); \
+		M[0xE] = sph_dec32be_aligned(buf + 56); \
+		M[0xF] = sph_dec32be_aligned(buf + 60); \
+		for (r = 0; r < 14; r ++) \
+			ROUND_S(r); \
+		H0 ^= S0 ^ V0 ^ V8; \
+		H1 ^= S1 ^ V1 ^ V9; \
+		H2 ^= S2 ^ V2 ^ VA; \
+		H3 ^= S3 ^ V3 ^ VB; \
+		H4 ^= S0 ^ V4 ^ VC; \
+		H5 ^= S1 ^ V5 ^ VD; \
+		H6 ^= S2 ^ V6 ^ VE; \
+		H7 ^= S3 ^ V7 ^ VF; \
+	} while (0)
+
+#else
+
+#define COMPRESS32   do { \
+		sph_u32 M0, M1, M2, M3, M4, M5, M6, M7; \
+		sph_u32 M8, M9, MA, MB, MC, MD, ME, MF; \
+		sph_u32 V0, V1, V2, V3, V4, V5, V6, V7; \
+		sph_u32 V8, V9, VA, VB, VC, VD, VE, VF; \
+		V0 = H0; \
+		V1 = H1; \
+		V2 = H2; \
+		V3 = H3; \
+		V4 = H4; \
+		V5 = H5; \
+		V6 = H6; \
+		V7 = H7; \
+		V8 = S0 ^ CS0; \
+		V9 = S1 ^ CS1; \
+		VA = S2 ^ CS2; \
+		VB = S3 ^ CS3; \
+		VC = T0 ^ CS4; \
+		VD = T0 ^ CS5; \
+		VE = T1 ^ CS6; \
+		VF = T1 ^ CS7; \
+		M0 = sph_dec32be_aligned(buf +  0); \
+		M1 = sph_dec32be_aligned(buf +  4); \
+		M2 = sph_dec32be_aligned(buf +  8); \
+		M3 = sph_dec32be_aligned(buf + 12); \
+		M4 = sph_dec32be_aligned(buf + 16); \
+		M5 = sph_dec32be_aligned(buf + 20); \
+		M6 = sph_dec32be_aligned(buf + 24); \
+		M7 = sph_dec32be_aligned(buf + 28); \
+		M8 = sph_dec32be_aligned(buf + 32); \
+		M9 = sph_dec32be_aligned(buf + 36); \
+		MA = sph_dec32be_aligned(buf + 40); \
+		MB = sph_dec32be_aligned(buf + 44); \
+		MC = sph_dec32be_aligned(buf + 48); \
+		MD = sph_dec32be_aligned(buf + 52); \
+		ME = sph_dec32be_aligned(buf + 56); \
+		MF = sph_dec32be_aligned(buf + 60); \
+		ROUND_S(0); \
+		ROUND_S(1); \
+		ROUND_S(2); \
+		ROUND_S(3); \
+		ROUND_S(4); \
+		ROUND_S(5); \
+		ROUND_S(6); \
+		ROUND_S(7); \
+		ROUND_S(8); \
+		ROUND_S(9); \
+		ROUND_S(0); \
+		ROUND_S(1); \
+		ROUND_S(2); \
+		ROUND_S(3); \
+		H0 ^= S0 ^ V0 ^ V8; \
+		H1 ^= S1 ^ V1 ^ V9; \
+		H2 ^= S2 ^ V2 ^ VA; \
+		H3 ^= S3 ^ V3 ^ VB; \
+		H4 ^= S0 ^ V4 ^ VC; \
+		H5 ^= S1 ^ V5 ^ VD; \
+		H6 ^= S2 ^ V6 ^ VE; \
+		H7 ^= S3 ^ V7 ^ VF; \
+	} while (0)
+
+#endif
+
+#if SPH_64
+
+#define DECL_STATE64 \
+	sph_u64 H0, H1, H2, H3, H4, H5, H6, H7; \
+	sph_u64 S0, S1, S2, S3, T0, T1;
+
+#define READ_STATE64(state)   do { \
+		H0 = (state)->H[0]; \
+		H1 = (state)->H[1]; \
+		H2 = (state)->H[2]; \
+		H3 = (state)->H[3]; \
+		H4 = (state)->H[4]; \
+		H5 = (state)->H[5]; \
+		H6 = (state)->H[6]; \
+		H7 = (state)->H[7]; \
+		S0 = (state)->S[0]; \
+		S1 = (state)->S[1]; \
+		S2 = (state)->S[2]; \
+		S3 = (state)->S[3]; \
+		T0 = (state)->T0; \
+		T1 = (state)->T1; \
+	} while (0)
+
+#define WRITE_STATE64(state)   do { \
+		(state)->H[0] = H0; \
+		(state)->H[1] = H1; \
+		(state)->H[2] = H2; \
+		(state)->H[3] = H3; \
+		(state)->H[4] = H4; \
+		(state)->H[5] = H5; \
+		(state)->H[6] = H6; \
+		(state)->H[7] = H7; \
+		(state)->S[0] = S0; \
+		(state)->S[1] = S1; \
+		(state)->S[2] = S2; \
+		(state)->S[3] = S3; \
+		(state)->T0 = T0; \
+		(state)->T1 = T1; \
+	} while (0)
+
+#if SPH_COMPACT_BLAKE_64
+
+#define COMPRESS64   do { \
+		sph_u64 M[16]; \
+		sph_u64 V0, V1, V2, V3, V4, V5, V6, V7; \
+		sph_u64 V8, V9, VA, VB, VC, VD, VE, VF; \
+		unsigned r; \
+		V0 = H0; \
+		V1 = H1; \
+		V2 = H2; \
+		V3 = H3; \
+		V4 = H4; \
+		V5 = H5; \
+		V6 = H6; \
+		V7 = H7; \
+		V8 = S0 ^ CB0; \
+		V9 = S1 ^ CB1; \
+		VA = S2 ^ CB2; \
+		VB = S3 ^ CB3; \
+		VC = T0 ^ CB4; \
+		VD = T0 ^ CB5; \
+		VE = T1 ^ CB6; \
+		VF = T1 ^ CB7; \
+		M[0x0] = sph_dec64be_aligned(buf +   0); \
+		M[0x1] = sph_dec64be_aligned(buf +   8); \
+		M[0x2] = sph_dec64be_aligned(buf +  16); \
+		M[0x3] = sph_dec64be_aligned(buf +  24); \
+		M[0x4] = sph_dec64be_aligned(buf +  32); \
+		M[0x5] = sph_dec64be_aligned(buf +  40); \
+		M[0x6] = sph_dec64be_aligned(buf +  48); \
+		M[0x7] = sph_dec64be_aligned(buf +  56); \
+		M[0x8] = sph_dec64be_aligned(buf +  64); \
+		M[0x9] = sph_dec64be_aligned(buf +  72); \
+		M[0xA] = sph_dec64be_aligned(buf +  80); \
+		M[0xB] = sph_dec64be_aligned(buf +  88); \
+		M[0xC] = sph_dec64be_aligned(buf +  96); \
+		M[0xD] = sph_dec64be_aligned(buf + 104); \
+		M[0xE] = sph_dec64be_aligned(buf + 112); \
+		M[0xF] = sph_dec64be_aligned(buf + 120); \
+		for (r = 0; r < 16; r ++) \
+			ROUND_B(r); \
+		H0 ^= S0 ^ V0 ^ V8; \
+		H1 ^= S1 ^ V1 ^ V9; \
+		H2 ^= S2 ^ V2 ^ VA; \
+		H3 ^= S3 ^ V3 ^ VB; \
+		H4 ^= S0 ^ V4 ^ VC; \
+		H5 ^= S1 ^ V5 ^ VD; \
+		H6 ^= S2 ^ V6 ^ VE; \
+		H7 ^= S3 ^ V7 ^ VF; \
+	} while (0)
+
+#else
+
+#define COMPRESS64   do { \
+		sph_u64 M0, M1, M2, M3, M4, M5, M6, M7; \
+		sph_u64 M8, M9, MA, MB, MC, MD, ME, MF; \
+		sph_u64 V0, V1, V2, V3, V4, V5, V6, V7; \
+		sph_u64 V8, V9, VA, VB, VC, VD, VE, VF; \
+		V0 = H0; \
+		V1 = H1; \
+		V2 = H2; \
+		V3 = H3; \
+		V4 = H4; \
+		V5 = H5; \
+		V6 = H6; \
+		V7 = H7; \
+		V8 = S0 ^ CB0; \
+		V9 = S1 ^ CB1; \
+		VA = S2 ^ CB2; \
+		VB = S3 ^ CB3; \
+		VC = T0 ^ CB4; \
+		VD = T0 ^ CB5; \
+		VE = T1 ^ CB6; \
+		VF = T1 ^ CB7; \
+		M0 = sph_dec64be_aligned(buf +   0); \
+		M1 = sph_dec64be_aligned(buf +   8); \
+		M2 = sph_dec64be_aligned(buf +  16); \
+		M3 = sph_dec64be_aligned(buf +  24); \
+		M4 = sph_dec64be_aligned(buf +  32); \
+		M5 = sph_dec64be_aligned(buf +  40); \
+		M6 = sph_dec64be_aligned(buf +  48); \
+		M7 = sph_dec64be_aligned(buf +  56); \
+		M8 = sph_dec64be_aligned(buf +  64); \
+		M9 = sph_dec64be_aligned(buf +  72); \
+		MA = sph_dec64be_aligned(buf +  80); \
+		MB = sph_dec64be_aligned(buf +  88); \
+		MC = sph_dec64be_aligned(buf +  96); \
+		MD = sph_dec64be_aligned(buf + 104); \
+		ME = sph_dec64be_aligned(buf + 112); \
+		MF = sph_dec64be_aligned(buf + 120); \
+		ROUND_B(0); \
+		ROUND_B(1); \
+		ROUND_B(2); \
+		ROUND_B(3); \
+		ROUND_B(4); \
+		ROUND_B(5); \
+		ROUND_B(6); \
+		ROUND_B(7); \
+		ROUND_B(8); \
+		ROUND_B(9); \
+		ROUND_B(0); \
+		ROUND_B(1); \
+		ROUND_B(2); \
+		ROUND_B(3); \
+		ROUND_B(4); \
+		ROUND_B(5); \
+		H0 ^= S0 ^ V0 ^ V8; \
+		H1 ^= S1 ^ V1 ^ V9; \
+		H2 ^= S2 ^ V2 ^ VA; \
+		H3 ^= S3 ^ V3 ^ VB; \
+		H4 ^= S0 ^ V4 ^ VC; \
+		H5 ^= S1 ^ V5 ^ VD; \
+		H6 ^= S2 ^ V6 ^ VE; \
+		H7 ^= S3 ^ V7 ^ VF; \
+	} while (0)
+
+#endif
+
+#endif
+
+static const sph_u32 salt_zero_small[4] = { 0, 0, 0, 0 };
+
+static void
+blake32_init(sph_blake_small_context *sc,
+	const sph_u32 *iv, const sph_u32 *salt)
+{
+	memcpy(sc->H, iv, 8 * sizeof(sph_u32));
+	memcpy(sc->S, salt, 4 * sizeof(sph_u32));
+	sc->T0 = sc->T1 = 0;
+	sc->ptr = 0;
+}
+
+static void
+blake32(sph_blake_small_context *sc, const void *data, size_t len)
+{
+	unsigned char *buf;
+	size_t ptr;
+	DECL_STATE32
+
+	buf = sc->buf;
+	ptr = sc->ptr;
+	if (len < (sizeof sc->buf) - ptr) {
+		memcpy(buf + ptr, data, len);
+		ptr += len;
+		sc->ptr = ptr;
+		return;
+	}
+
+	READ_STATE32(sc);
+	while (len > 0) {
+		size_t clen;
+
+		clen = (sizeof sc->buf) - ptr;
+		if (clen > len)
+			clen = len;
+		memcpy(buf + ptr, data, clen);
+		ptr += clen;
+		data = (const unsigned char *)data + clen;
+		len -= clen;
+		if (ptr == sizeof sc->buf) {
+			if ((T0 = SPH_T32(T0 + 512)) < 512)
+				T1 = SPH_T32(T1 + 1);
+			COMPRESS32;
+			ptr = 0;
+		}
+	}
+	WRITE_STATE32(sc);
+	sc->ptr = ptr;
+}
+
+static void
+blake32_close(sph_blake_small_context *sc,
+	unsigned ub, unsigned n, void *dst, size_t out_size_w32)
+{
+	union {
+		unsigned char buf[64];
+		sph_u32 dummy;
+	} u;
+	size_t ptr, k;
+	unsigned bit_len;
+	unsigned z;
+	sph_u32 th, tl;
+	unsigned char *out;
+
+	ptr = sc->ptr;
+	bit_len = ((unsigned)ptr << 3) + n;
+	z = 0x80 >> n;
+	u.buf[ptr] = ((ub & -z) | z) & 0xFF;
+	tl = sc->T0 + bit_len;
+	th = sc->T1;
+	if (ptr == 0 && n == 0) {
+		sc->T0 = SPH_C32(0xFFFFFE00);
+		sc->T1 = SPH_C32(0xFFFFFFFF);
+	} else if (sc->T0 == 0) {
+		sc->T0 = SPH_C32(0xFFFFFE00) + bit_len;
+		sc->T1 = SPH_T32(sc->T1 - 1);
+	} else {
+		sc->T0 -= 512 - bit_len;
+	}
+	if (bit_len <= 446) {
+		memset(u.buf + ptr + 1, 0, 55 - ptr);
+		if (out_size_w32 == 8)
+			u.buf[55] |= 1;
+		sph_enc32be_aligned(u.buf + 56, th);
+		sph_enc32be_aligned(u.buf + 60, tl);
+		blake32(sc, u.buf + ptr, 64 - ptr);
+	} else {
+		memset(u.buf + ptr + 1, 0, 63 - ptr);
+		blake32(sc, u.buf + ptr, 64 - ptr);
+		sc->T0 = SPH_C32(0xFFFFFE00);
+		sc->T1 = SPH_C32(0xFFFFFFFF);
+		memset(u.buf, 0, 56);
+		if (out_size_w32 == 8)
+			u.buf[55] = 1;
+		sph_enc32be_aligned(u.buf + 56, th);
+		sph_enc32be_aligned(u.buf + 60, tl);
+		blake32(sc, u.buf, 64);
+	}
+	out = dst;
+	for (k = 0; k < out_size_w32; k ++)
+		sph_enc32be(out + (k << 2), sc->H[k]);
+}
+
+#if SPH_64
+
+static const sph_u64 salt_zero_big[4] = { 0, 0, 0, 0 };
+
+static void
+blake64_init(sph_blake_big_context *sc,
+	const sph_u64 *iv, const sph_u64 *salt)
+{
+	memcpy(sc->H, iv, 8 * sizeof(sph_u64));
+	memcpy(sc->S, salt, 4 * sizeof(sph_u64));
+	sc->T0 = sc->T1 = 0;
+	sc->ptr = 0;
+}
+
+static void
+blake64(sph_blake_big_context *sc, const void *data, size_t len)
+{
+	unsigned char *buf;
+	size_t ptr;
+	DECL_STATE64
+
+	buf = sc->buf;
+	ptr = sc->ptr;
+	if (len < (sizeof sc->buf) - ptr) {
+		memcpy(buf + ptr, data, len);
+		ptr += len;
+		sc->ptr = ptr;
+		return;
+	}
+
+	READ_STATE64(sc);
+	while (len > 0) {
+		size_t clen;
+
+		clen = (sizeof sc->buf) - ptr;
+		if (clen > len)
+			clen = len;
+		memcpy(buf + ptr, data, clen);
+		ptr += clen;
+		data = (const unsigned char *)data + clen;
+		len -= clen;
+		if (ptr == sizeof sc->buf) {
+			if ((T0 = SPH_T64(T0 + 1024)) < 1024)
+				T1 = SPH_T64(T1 + 1);
+			COMPRESS64;
+			ptr = 0;
+		}
+	}
+	WRITE_STATE64(sc);
+	sc->ptr = ptr;
+}
+
+static void
+blake64_close(sph_blake_big_context *sc,
+	unsigned ub, unsigned n, void *dst, size_t out_size_w64)
+{
+	union {
+		unsigned char buf[128];
+		sph_u64 dummy;
+	} u;
+	size_t ptr, k;
+	unsigned bit_len;
+	unsigned z;
+	sph_u64 th, tl;
+	unsigned char *out;
+
+	ptr = sc->ptr;
+	bit_len = ((unsigned)ptr << 3) + n;
+	z = 0x80 >> n;
+	u.buf[ptr] = ((ub & -z) | z) & 0xFF;
+	tl = sc->T0 + bit_len;
+	th = sc->T1;
+	if (ptr == 0 && n == 0) {
+		sc->T0 = SPH_C64(0xFFFFFFFFFFFFFC00);
+		sc->T1 = SPH_C64(0xFFFFFFFFFFFFFFFF);
+	} else if (sc->T0 == 0) {
+		sc->T0 = SPH_C64(0xFFFFFFFFFFFFFC00) + bit_len;
+		sc->T1 = SPH_T64(sc->T1 - 1);
+	} else {
+		sc->T0 -= 1024 - bit_len;
+	}
+	if (bit_len <= 894) {
+		memset(u.buf + ptr + 1, 0, 111 - ptr);
+		if (out_size_w64 == 8)
+			u.buf[111] |= 1;
+		sph_enc64be_aligned(u.buf + 112, th);
+		sph_enc64be_aligned(u.buf + 120, tl);
+		blake64(sc, u.buf + ptr, 128 - ptr);
+	} else {
+		memset(u.buf + ptr + 1, 0, 127 - ptr);
+		blake64(sc, u.buf + ptr, 128 - ptr);
+		sc->T0 = SPH_C64(0xFFFFFFFFFFFFFC00);
+		sc->T1 = SPH_C64(0xFFFFFFFFFFFFFFFF);
+		memset(u.buf, 0, 112);
+		if (out_size_w64 == 8)
+			u.buf[111] = 1;
+		sph_enc64be_aligned(u.buf + 112, th);
+		sph_enc64be_aligned(u.buf + 120, tl);
+		blake64(sc, u.buf, 128);
+	}
+	out = dst;
+	for (k = 0; k < out_size_w64; k ++)
+		sph_enc64be(out + (k << 3), sc->H[k]);
+}
+
+#endif
+
+/* see sph_blake.h */
+void
+sph_blake224_init(void *cc)
+{
+	blake32_init(cc, IV224, salt_zero_small);
+}
+
+/* see sph_blake.h */
+void
+sph_blake224(void *cc, const void *data, size_t len)
+{
+	blake32(cc, data, len);
+}
+
+/* see sph_blake.h */
+void
+sph_blake224_close(void *cc, void *dst)
+{
+	sph_blake224_addbits_and_close(cc, 0, 0, dst);
+}
+
+/* see sph_blake.h */
+void
+sph_blake224_addbits_and_close(void *cc, unsigned ub, unsigned n, void *dst)
+{
+	blake32_close(cc, ub, n, dst, 7);
+	sph_blake224_init(cc);
+}
+
+/* see sph_blake.h */
+void
+sph_blake256_init(void *cc)
+{
+	blake32_init(cc, IV256, salt_zero_small);
+}
+
+/* see sph_blake.h */
+void
+sph_blake256(void *cc, const void *data, size_t len)
+{
+	blake32(cc, data, len);
+}
+
+/* see sph_blake.h */
+void
+sph_blake256_close(void *cc, void *dst)
+{
+	sph_blake256_addbits_and_close(cc, 0, 0, dst);
+}
+
+/* see sph_blake.h */
+void
+sph_blake256_addbits_and_close(void *cc, unsigned ub, unsigned n, void *dst)
+{
+	blake32_close(cc, ub, n, dst, 8);
+	sph_blake256_init(cc);
+}
+
+#if SPH_64
+
+/* see sph_blake.h */
+void
+sph_blake384_init(void *cc)
+{
+	blake64_init(cc, IV384, salt_zero_big);
+}
+
+/* see sph_blake.h */
+void
+sph_blake384(void *cc, const void *data, size_t len)
+{
+	blake64(cc, data, len);
+}
+
+/* see sph_blake.h */
+void
+sph_blake384_close(void *cc, void *dst)
+{
+	sph_blake384_addbits_and_close(cc, 0, 0, dst);
+}
+
+/* see sph_blake.h */
+void
+sph_blake384_addbits_and_close(void *cc, unsigned ub, unsigned n, void *dst)
+{
+	blake64_close(cc, ub, n, dst, 6);
+	sph_blake384_init(cc);
+}
+
+/* see sph_blake.h */
+void
+sph_blake512_init(void *cc)
+{
+	blake64_init(cc, IV512, salt_zero_big);
+}
+
+/* see sph_blake.h */
+void
+sph_blake512(void *cc, const void *data, size_t len)
+{
+	blake64(cc, data, len);
+}
+
+/* see sph_blake.h */
+void
+sph_blake512_close(void *cc, void *dst)
+{
+	sph_blake512_addbits_and_close(cc, 0, 0, dst);
+}
+
+/* see sph_blake.h */
+void
+sph_blake512_addbits_and_close(void *cc, unsigned ub, unsigned n, void *dst)
+{
+	blake64_close(cc, ub, n, dst, 8);
+	sph_blake512_init(cc);
+}
+
+#endif
+
+#ifdef __cplusplus
+}
+#endif
diff --git a/ccminer.sln b/ccminer.sln
new file mode 100644
index 0000000000..a28278f2db
--- /dev/null
+++ b/ccminer.sln
@@ -0,0 +1,26 @@
+
+Microsoft Visual Studio Solution File, Format Version 11.00
+# Visual Studio 2010
+Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "ccminer", "ccminer.vcxproj", "{36DC07F9-A4A6-4877-A146-1B960083CF6F}"
+EndProject
+Global
+	GlobalSection(SolutionConfigurationPlatforms) = preSolution
+		Debug|Win32 = Debug|Win32
+		Debug|x64 = Debug|x64
+		Release|Win32 = Release|Win32
+		Release|x64 = Release|x64
+	EndGlobalSection
+	GlobalSection(ProjectConfigurationPlatforms) = postSolution
+		{36DC07F9-A4A6-4877-A146-1B960083CF6F}.Debug|Win32.ActiveCfg = Debug|Win32
+		{36DC07F9-A4A6-4877-A146-1B960083CF6F}.Debug|Win32.Build.0 = Debug|Win32
+		{36DC07F9-A4A6-4877-A146-1B960083CF6F}.Debug|x64.ActiveCfg = Debug|x64
+		{36DC07F9-A4A6-4877-A146-1B960083CF6F}.Debug|x64.Build.0 = Debug|x64
+		{36DC07F9-A4A6-4877-A146-1B960083CF6F}.Release|Win32.ActiveCfg = Release|Win32
+		{36DC07F9-A4A6-4877-A146-1B960083CF6F}.Release|Win32.Build.0 = Release|Win32
+		{36DC07F9-A4A6-4877-A146-1B960083CF6F}.Release|x64.ActiveCfg = Release|x64
+		{36DC07F9-A4A6-4877-A146-1B960083CF6F}.Release|x64.Build.0 = Release|x64
+	EndGlobalSection
+	GlobalSection(SolutionProperties) = preSolution
+		HideSolutionNode = FALSE
+	EndGlobalSection
+EndGlobal
diff --git a/ccminer.vcxproj b/ccminer.vcxproj
new file mode 100644
index 0000000000..b721b6c964
--- /dev/null
+++ b/ccminer.vcxproj
@@ -0,0 +1,291 @@
+<?xml version="1.0" encoding="utf-8"?>
+<Project DefaultTargets="Build" ToolsVersion="4.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
+  <ItemGroup Label="ProjectConfigurations">
+    <ProjectConfiguration Include="Debug|Win32">
+      <Configuration>Debug</Configuration>
+      <Platform>Win32</Platform>
+    </ProjectConfiguration>
+    <ProjectConfiguration Include="Debug|x64">
+      <Configuration>Debug</Configuration>
+      <Platform>x64</Platform>
+    </ProjectConfiguration>
+    <ProjectConfiguration Include="Release|Win32">
+      <Configuration>Release</Configuration>
+      <Platform>Win32</Platform>
+    </ProjectConfiguration>
+    <ProjectConfiguration Include="Release|x64">
+      <Configuration>Release</Configuration>
+      <Platform>x64</Platform>
+    </ProjectConfiguration>
+  </ItemGroup>
+  <PropertyGroup Label="Globals">
+    <ProjectGuid>{36DC07F9-A4A6-4877-A146-1B960083CF6F}</ProjectGuid>
+    <RootNamespace>ccminer</RootNamespace>
+  </PropertyGroup>
+  <Import Project="$(VCTargetsPath)\Microsoft.Cpp.Default.props" />
+  <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'" Label="Configuration">
+    <ConfigurationType>Application</ConfigurationType>
+    <UseDebugLibraries>true</UseDebugLibraries>
+    <CharacterSet>MultiByte</CharacterSet>
+  </PropertyGroup>
+  <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Debug|x64'" Label="Configuration">
+    <ConfigurationType>Application</ConfigurationType>
+    <UseDebugLibraries>true</UseDebugLibraries>
+    <CharacterSet>MultiByte</CharacterSet>
+  </PropertyGroup>
+  <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Release|Win32'" Label="Configuration">
+    <ConfigurationType>Application</ConfigurationType>
+    <UseDebugLibraries>false</UseDebugLibraries>
+    <WholeProgramOptimization>true</WholeProgramOptimization>
+    <CharacterSet>MultiByte</CharacterSet>
+  </PropertyGroup>
+  <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Release|x64'" Label="Configuration">
+    <ConfigurationType>Application</ConfigurationType>
+    <UseDebugLibraries>false</UseDebugLibraries>
+    <WholeProgramOptimization>true</WholeProgramOptimization>
+    <CharacterSet>MultiByte</CharacterSet>
+  </PropertyGroup>
+  <Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
+  <ImportGroup Label="ExtensionSettings">
+    <Import Project="$(VCTargetsPath)\BuildCustomizations\CUDA 5.5.props" />
+  </ImportGroup>
+  <ImportGroup Label="PropertySheets" Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">
+    <Import Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" />
+  </ImportGroup>
+  <ImportGroup Condition="'$(Configuration)|$(Platform)'=='Debug|x64'" Label="PropertySheets">
+    <Import Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" />
+  </ImportGroup>
+  <ImportGroup Label="PropertySheets" Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">
+    <Import Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" />
+  </ImportGroup>
+  <ImportGroup Condition="'$(Configuration)|$(Platform)'=='Release|x64'" Label="PropertySheets">
+    <Import Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" />
+  </ImportGroup>
+  <PropertyGroup Label="UserMacros" />
+  <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">
+    <LinkIncremental>true</LinkIncremental>
+  </PropertyGroup>
+  <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">
+    <LinkIncremental>true</LinkIncremental>
+  </PropertyGroup>
+  <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">
+    <LinkIncremental>true</LinkIncremental>
+  </PropertyGroup>
+  <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Release|x64'">
+    <LinkIncremental>true</LinkIncremental>
+  </PropertyGroup>
+  <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">
+    <ClCompile>
+      <WarningLevel>Level3</WarningLevel>
+      <Optimization>Disabled</Optimization>
+      <PreprocessorDefinitions>WIN32;_DEBUG;_CONSOLE;_CRT_SECURE_NO_WARNINGS;CURL_STATICLIB;SCRYPT_KECCAK512;SCRYPT_CHACHA;SCRYPT_CHOOSE_COMPILETIME;%(PreprocessorDefinitions)</PreprocessorDefinitions>
+      <AdditionalIncludeDirectories>.;compat;compat\jansson;compat\getopt;..\pthreads\Pre-built.2\include;..\curl-7.29.0\include;..\OpenSSL-Win32\include;%(AdditionalIncludeDirectories);$(CudaToolkitIncludeDir)</AdditionalIncludeDirectories>
+    </ClCompile>
+    <Link>
+      <GenerateDebugInformation>true</GenerateDebugInformation>
+      <SubSystem>Console</SubSystem>
+      <AdditionalDependencies>cudart.lib;cuda.lib;kernel32.lib;user32.lib;gdi32.lib;winspool.lib;comdlg32.lib;advapi32.lib;shell32.lib;ole32.lib;oleaut32.lib;uuid.lib;odbc32.lib;odbccp32.lib;pthreadVC2.lib;libcurl.lib;ws2_32.lib;Wldap32.lib;libeay32MTd.lib;ssleay32MTd.lib;%(AdditionalDependencies)</AdditionalDependencies>
+      <AdditionalLibraryDirectories>..\pthreads\Pre-built.2\lib\x86;..\curl-7.29.0\build\lib\Debug;..\OpenSSL-Win32\lib\VC\static;%(AdditionalLibraryDirectories);$(CudaToolkitLibDir)</AdditionalLibraryDirectories>
+    </Link>
+    <PostBuildEvent>
+      <Command>echo copy "$(CudaToolkitBinDir)\cudart*.dll" "$(OutDir)"
+copy "$(CudaToolkitBinDir)\cudart*.dll" "$(OutDir)"</Command>
+    </PostBuildEvent>
+    <CudaCompile>
+      <CInterleavedPTX>true</CInterleavedPTX>
+    </CudaCompile>
+    <CudaCompile>
+      <MaxRegCount>128</MaxRegCount>
+    </CudaCompile>
+    <CudaCompile>
+      <PtxAsOptionV>true</PtxAsOptionV>
+      <Keep>true</Keep>
+      <CodeGeneration>compute_35,sm_35</CodeGeneration>
+      <Include>
+      </Include>
+      <AdditionalOptions>-Xptxas "-abi=no -v" %(AdditionalOptions)</AdditionalOptions>
+    </CudaCompile>
+  </ItemDefinitionGroup>
+  <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">
+    <ClCompile>
+      <WarningLevel>Level3</WarningLevel>
+      <Optimization>Disabled</Optimization>
+      <PreprocessorDefinitions>WIN32;_DEBUG;_CONSOLE;_CRT_SECURE_NO_WARNINGS;CURL_STATICLIB;SCRYPT_KECCAK512;SCRYPT_CHACHA;SCRYPT_CHOOSE_COMPILETIME;%(PreprocessorDefinitions)</PreprocessorDefinitions>
+      <AdditionalIncludeDirectories>.;compat;compat\jansson;compat\getopt;..\pthreads\Pre-built.2\include;..\curl-7.29.0_x64\include;..\OpenSSL-Win64\include;%(AdditionalIncludeDirectories);$(CudaToolkitIncludeDir)</AdditionalIncludeDirectories>
+    </ClCompile>
+    <Link>
+      <GenerateDebugInformation>true</GenerateDebugInformation>
+      <SubSystem>Console</SubSystem>
+      <AdditionalDependencies>cudart.lib;cuda.lib;kernel32.lib;user32.lib;gdi32.lib;winspool.lib;comdlg32.lib;advapi32.lib;shell32.lib;ole32.lib;oleaut32.lib;uuid.lib;odbc32.lib;odbccp32.lib;pthreadVC2.lib;libcurl.lib;ws2_32.lib;Wldap32.lib;libeay32MTd.lib;ssleay32MTd.lib;%(AdditionalDependencies)</AdditionalDependencies>
+      <AdditionalLibraryDirectories>..\pthreads\Pre-built.2\lib\x64;..\curl-7.29.0_x64\build\lib\Debug;..\OpenSSL-Win64\lib\VC\static;%(AdditionalLibraryDirectories);$(CudaToolkitLibDir)</AdditionalLibraryDirectories>
+    </Link>
+    <PostBuildEvent>
+      <Command>echo copy "$(CudaToolkitBinDir)\cudart*.dll" "$(OutDir)"
+copy "$(CudaToolkitBinDir)\cudart*.dll" "$(OutDir)"</Command>
+    </PostBuildEvent>
+    <CudaCompile>
+      <CInterleavedPTX>true</CInterleavedPTX>
+    </CudaCompile>
+    <CudaCompile>
+      <MaxRegCount>128</MaxRegCount>
+    </CudaCompile>
+    <CudaCompile>
+      <PtxAsOptionV>true</PtxAsOptionV>
+      <Keep>true</Keep>
+      <CodeGeneration>compute_35,sm_35</CodeGeneration>
+      <Include>
+      </Include>
+      <AdditionalOptions>-Xptxas "-abi=no -v" %(AdditionalOptions)</AdditionalOptions>
+    </CudaCompile>
+  </ItemDefinitionGroup>
+  <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">
+    <ClCompile>
+      <WarningLevel>Level3</WarningLevel>
+      <Optimization>MaxSpeed</Optimization>
+      <FunctionLevelLinking>true</FunctionLevelLinking>
+      <IntrinsicFunctions>true</IntrinsicFunctions>
+      <PreprocessorDefinitions>WIN32;NDEBUG;_CONSOLE;_CRT_SECURE_NO_WARNINGS;CURL_STATICLIB;SCRYPT_KECCAK512;SCRYPT_CHACHA;SCRYPT_CHOOSE_COMPILETIME;%(PreprocessorDefinitions)</PreprocessorDefinitions>
+      <AdditionalIncludeDirectories>.;compat;compat\jansson;compat\getopt;..\pthreads\Pre-built.2\include;..\curl-7.29.0\include;..\OpenSSL-Win32\include;%(AdditionalIncludeDirectories);$(CudaToolkitIncludeDir)</AdditionalIncludeDirectories>
+    </ClCompile>
+    <Link>
+      <GenerateDebugInformation>true</GenerateDebugInformation>
+      <EnableCOMDATFolding>true</EnableCOMDATFolding>
+      <OptimizeReferences>true</OptimizeReferences>
+      <SubSystem>Console</SubSystem>
+      <AdditionalDependencies>cudart.lib;cuda.lib;kernel32.lib;user32.lib;gdi32.lib;winspool.lib;comdlg32.lib;advapi32.lib;shell32.lib;ole32.lib;oleaut32.lib;uuid.lib;odbc32.lib;odbccp32.lib;pthreadVC2.lib;libcurl.lib;ws2_32.lib;Wldap32.lib;libeay32MT.lib;ssleay32MT.lib;%(AdditionalDependencies)</AdditionalDependencies>
+      <AdditionalLibraryDirectories>..\pthreads\Pre-built.2\lib\x86;..\curl-7.29.0\build\lib\Release;..\OpenSSL-Win32\lib\VC\static;%(AdditionalLibraryDirectories);$(CudaToolkitLibDir)</AdditionalLibraryDirectories>
+    </Link>
+    <PostBuildEvent>
+      <Command>echo copy "$(CudaToolkitBinDir)\cudart*.dll" "$(OutDir)"
+copy "$(CudaToolkitBinDir)\cudart*.dll" "$(OutDir)"</Command>
+    </PostBuildEvent>
+    <CudaCompile>
+      <CInterleavedPTX>true</CInterleavedPTX>
+    </CudaCompile>
+    <CudaCompile>
+      <MaxRegCount>128</MaxRegCount>
+    </CudaCompile>
+    <CudaCompile>
+      <PtxAsOptionV>true</PtxAsOptionV>
+      <Keep>true</Keep>
+      <CodeGeneration>compute_35,sm_35</CodeGeneration>
+      <Include>
+      </Include>
+      <AdditionalOptions>-Xptxas "-abi=no -v" %(AdditionalOptions)</AdditionalOptions>
+    </CudaCompile>
+  </ItemDefinitionGroup>
+  <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Release|x64'">
+    <ClCompile>
+      <WarningLevel>Level3</WarningLevel>
+      <Optimization>MaxSpeed</Optimization>
+      <FunctionLevelLinking>true</FunctionLevelLinking>
+      <IntrinsicFunctions>true</IntrinsicFunctions>
+      <PreprocessorDefinitions>WIN32;NDEBUG;_CONSOLE;_CRT_SECURE_NO_WARNINGS;CURL_STATICLIB;SCRYPT_KECCAK512;SCRYPT_CHACHA;SCRYPT_CHOOSE_COMPILETIME;%(PreprocessorDefinitions)</PreprocessorDefinitions>
+      <AdditionalIncludeDirectories>.;compat;compat\jansson;compat\getopt;..\pthreads\Pre-built.2\include;..\curl-7.29.0_x64\include;..\OpenSSL-Win64\include;%(AdditionalIncludeDirectories);$(CudaToolkitIncludeDir)</AdditionalIncludeDirectories>
+    </ClCompile>
+    <Link>
+      <GenerateDebugInformation>true</GenerateDebugInformation>
+      <EnableCOMDATFolding>true</EnableCOMDATFolding>
+      <OptimizeReferences>true</OptimizeReferences>
+      <SubSystem>Console</SubSystem>
+      <AdditionalDependencies>cudart.lib;cuda.lib;kernel32.lib;user32.lib;gdi32.lib;winspool.lib;comdlg32.lib;advapi32.lib;shell32.lib;ole32.lib;oleaut32.lib;uuid.lib;odbc32.lib;odbccp32.lib;pthreadVC2.lib;libcurl.lib;ws2_32.lib;Wldap32.lib;libeay32MT.lib;ssleay32MT.lib;%(AdditionalDependencies)</AdditionalDependencies>
+      <AdditionalLibraryDirectories>..\pthreads\Pre-built.2\lib\x64;..\curl-7.29.0_x64\build\lib\Release;..\OpenSSL-Win64\lib\VC\static;%(AdditionalLibraryDirectories);$(CudaToolkitLibDir)</AdditionalLibraryDirectories>
+    </Link>
+    <PostBuildEvent>
+      <Command>echo copy "$(CudaToolkitBinDir)\cudart*.dll" "$(OutDir)"
+copy "$(CudaToolkitBinDir)\cudart*.dll" "$(OutDir)"</Command>
+    </PostBuildEvent>
+    <CudaCompile>
+      <CInterleavedPTX>true</CInterleavedPTX>
+    </CudaCompile>
+    <CudaCompile>
+      <MaxRegCount>128</MaxRegCount>
+    </CudaCompile>
+    <CudaCompile>
+      <PtxAsOptionV>true</PtxAsOptionV>
+      <Keep>true</Keep>
+      <CodeGeneration>compute_35,sm_35</CodeGeneration>
+      <Include>
+      </Include>
+      <AdditionalOptions>-Xptxas "-abi=no -v" %(AdditionalOptions)</AdditionalOptions>
+    </CudaCompile>
+  </ItemDefinitionGroup>
+  <ItemGroup>
+    <ClCompile Include="blake.c" />
+    <ClCompile Include="compat\getopt\getopt_long.c" />
+    <ClCompile Include="compat\gettimeofday.c" />
+    <ClCompile Include="compat\jansson\dump.c" />
+    <ClCompile Include="compat\jansson\hashtable.c" />
+    <ClCompile Include="compat\jansson\load.c" />
+    <ClCompile Include="compat\jansson\strbuffer.c" />
+    <ClCompile Include="compat\jansson\utf.c" />
+    <ClCompile Include="compat\jansson\value.c" />
+    <ClCompile Include="cpu-miner.c">
+      <AdditionalOptions Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">/TP %(AdditionalOptions)</AdditionalOptions>
+      <AdditionalOptions Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">/TP %(AdditionalOptions)</AdditionalOptions>
+      <AdditionalOptions Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">/TP %(AdditionalOptions)</AdditionalOptions>
+      <AdditionalOptions Condition="'$(Configuration)|$(Platform)'=='Release|x64'">/TP %(AdditionalOptions)</AdditionalOptions>
+    </ClCompile>
+    <ClCompile Include="fugue.c" />
+    <ClCompile Include="fuguecoin.cpp" />
+    <ClCompile Include="groestl.c" />
+    <ClCompile Include="hefty1.c" />
+    <ClCompile Include="keccak.c" />
+    <ClCompile Include="scrypt.c" />
+    <ClCompile Include="sha2.c">
+      <AdditionalOptions Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">/TP %(AdditionalOptions)</AdditionalOptions>
+      <AdditionalOptions Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">/TP %(AdditionalOptions)</AdditionalOptions>
+      <AdditionalOptions Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">/TP %(AdditionalOptions)</AdditionalOptions>
+      <AdditionalOptions Condition="'$(Configuration)|$(Platform)'=='Release|x64'">/TP %(AdditionalOptions)</AdditionalOptions>
+    </ClCompile>
+    <ClCompile Include="util.c">
+      <AdditionalOptions Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">/TP %(AdditionalOptions)</AdditionalOptions>
+      <AdditionalOptions Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">/TP %(AdditionalOptions)</AdditionalOptions>
+      <AdditionalOptions Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">/TP %(AdditionalOptions)</AdditionalOptions>
+      <AdditionalOptions Condition="'$(Configuration)|$(Platform)'=='Release|x64'">/TP %(AdditionalOptions)</AdditionalOptions>
+    </ClCompile>
+  </ItemGroup>
+  <ItemGroup>
+    <ClInclude Include="compat.h" />
+    <ClInclude Include="compat\getopt\getopt.h" />
+    <ClInclude Include="compat\inttypes.h" />
+    <ClInclude Include="compat\stdbool.h" />
+    <ClInclude Include="compat\sys\time.h" />
+    <ClInclude Include="compat\unistd.h" />
+    <ClInclude Include="cpuminer-config.h" />
+    <ClInclude Include="cuda_blake512.h" />
+    <ClInclude Include="cuda_combine.h" />
+    <ClInclude Include="cuda_groestl512.h" />
+    <ClInclude Include="cuda_hefty1.h" />
+    <ClInclude Include="cuda_keccak512.h" />
+    <ClInclude Include="cuda_sha256.h" />
+    <ClInclude Include="elist.h" />
+    <ClInclude Include="hefty1.h" />
+    <ClInclude Include="miner.h" />
+    <ClInclude Include="sph_blake.h" />
+    <ClInclude Include="sph_fugue.h" />
+    <ClInclude Include="sph_groestl.h" />
+    <ClInclude Include="sph_keccak.h" />
+    <ClInclude Include="sph_types.h" />
+    <ClInclude Include="uint256.h" />
+  </ItemGroup>
+  <ItemGroup>
+    <CudaCompile Include="cuda_blake512.cu" />
+    <CudaCompile Include="cuda_combine.cu" />
+    <CudaCompile Include="cuda_fugue256.cu" />
+    <CudaCompile Include="cuda_groestl512.cu" />
+    <CudaCompile Include="cuda_hefty1.cu">
+      <CodeGeneration Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">compute_10,sm_10</CodeGeneration>
+      <CodeGeneration Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">compute_10,sm_10</CodeGeneration>
+      <CodeGeneration Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">compute_10,sm_10</CodeGeneration>
+      <CodeGeneration Condition="'$(Configuration)|$(Platform)'=='Release|x64'">compute_10,sm_10</CodeGeneration>
+    </CudaCompile>
+    <CudaCompile Include="cuda_keccak512.cu" />
+    <CudaCompile Include="cuda_sha256.cu" />
+    <CudaCompile Include="heavy.cu" />
+  </ItemGroup>
+  <Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
+  <ImportGroup Label="ExtensionTargets">
+    <Import Project="$(VCTargetsPath)\BuildCustomizations\CUDA 5.5.targets" />
+  </ImportGroup>
+</Project>
\ No newline at end of file
diff --git a/ccminer.vcxproj.filters b/ccminer.vcxproj.filters
new file mode 100644
index 0000000000..f3b22cef6b
--- /dev/null
+++ b/ccminer.vcxproj.filters
@@ -0,0 +1,188 @@
+<?xml version="1.0" encoding="utf-8"?>
+<Project ToolsVersion="4.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
+  <ItemGroup>
+    <Filter Include="Source Files">
+      <UniqueIdentifier>{2450a9c7-a97a-49e1-ba19-c8dbc5a4e3e7}</UniqueIdentifier>
+    </Filter>
+    <Filter Include="Source Files\gettimeofday">
+      <UniqueIdentifier>{c53ce808-c5c5-4c6c-99a2-3947090c62f1}</UniqueIdentifier>
+    </Filter>
+    <Filter Include="Source Files\getopt">
+      <UniqueIdentifier>{5a45c1bf-81d2-4bc6-97b5-714e34f51a82}</UniqueIdentifier>
+    </Filter>
+    <Filter Include="Header Files">
+      <UniqueIdentifier>{431cec61-9376-4de9-aae9-04c4250652e7}</UniqueIdentifier>
+    </Filter>
+    <Filter Include="Header Files\compat">
+      <UniqueIdentifier>{cc8bb259-5332-4a45-ba81-f4840a55b604}</UniqueIdentifier>
+    </Filter>
+    <Filter Include="Header Files\compat\getopt">
+      <UniqueIdentifier>{89362bd8-4690-4f0c-a4f7-6b2fa67a1f34}</UniqueIdentifier>
+    </Filter>
+    <Filter Include="Header Files\compat\sys">
+      <UniqueIdentifier>{6c3cd392-b6b8-424c-87d2-10e33dbd4b41}</UniqueIdentifier>
+    </Filter>
+    <Filter Include="Source Files\CUDA">
+      <UniqueIdentifier>{5a31b6f4-4943-4b22-b69a-230f3cc96269}</UniqueIdentifier>
+    </Filter>
+    <Filter Include="Header Files\CUDA">
+      <UniqueIdentifier>{a0f072d0-a831-4c23-8d64-7a026521df9c}</UniqueIdentifier>
+    </Filter>
+    <Filter Include="Source Files\code">
+      <UniqueIdentifier>{fe39ded0-754b-415f-a284-038a15a0aa55}</UniqueIdentifier>
+    </Filter>
+    <Filter Include="Source Files\CUDA\jansson">
+      <UniqueIdentifier>{17b56151-79ec-4a32-bac3-9d94ae7f68fe}</UniqueIdentifier>
+    </Filter>
+  </ItemGroup>
+  <ItemGroup>
+    <ClCompile Include="compat\jansson\dump.c">
+      <Filter>Source Files\CUDA\jansson</Filter>
+    </ClCompile>
+    <ClCompile Include="compat\jansson\hashtable.c">
+      <Filter>Source Files\CUDA\jansson</Filter>
+    </ClCompile>
+    <ClCompile Include="compat\jansson\load.c">
+      <Filter>Source Files\CUDA\jansson</Filter>
+    </ClCompile>
+    <ClCompile Include="compat\jansson\strbuffer.c">
+      <Filter>Source Files\CUDA\jansson</Filter>
+    </ClCompile>
+    <ClCompile Include="compat\jansson\utf.c">
+      <Filter>Source Files\CUDA\jansson</Filter>
+    </ClCompile>
+    <ClCompile Include="compat\jansson\value.c">
+      <Filter>Source Files\CUDA\jansson</Filter>
+    </ClCompile>
+    <ClCompile Include="compat\getopt\getopt_long.c">
+      <Filter>Source Files\getopt</Filter>
+    </ClCompile>
+    <ClCompile Include="compat\gettimeofday.c">
+      <Filter>Source Files\gettimeofday</Filter>
+    </ClCompile>
+    <ClCompile Include="util.c">
+      <Filter>Source Files</Filter>
+    </ClCompile>
+    <ClCompile Include="sha2.c">
+      <Filter>Source Files</Filter>
+    </ClCompile>
+    <ClCompile Include="cpu-miner.c">
+      <Filter>Source Files</Filter>
+    </ClCompile>
+    <ClCompile Include="blake.c">
+      <Filter>Source Files</Filter>
+    </ClCompile>
+    <ClCompile Include="groestl.c">
+      <Filter>Source Files</Filter>
+    </ClCompile>
+    <ClCompile Include="hefty1.c">
+      <Filter>Source Files</Filter>
+    </ClCompile>
+    <ClCompile Include="keccak.c">
+      <Filter>Source Files</Filter>
+    </ClCompile>
+    <ClCompile Include="scrypt.c">
+      <Filter>Source Files</Filter>
+    </ClCompile>
+    <ClCompile Include="fugue.c">
+      <Filter>Source Files</Filter>
+    </ClCompile>
+    <ClCompile Include="fuguecoin.cpp">
+      <Filter>Source Files</Filter>
+    </ClCompile>
+  </ItemGroup>
+  <ItemGroup>
+    <ClInclude Include="compat.h">
+      <Filter>Header Files</Filter>
+    </ClInclude>
+    <ClInclude Include="elist.h">
+      <Filter>Header Files</Filter>
+    </ClInclude>
+    <ClInclude Include="miner.h">
+      <Filter>Header Files</Filter>
+    </ClInclude>
+    <ClInclude Include="compat\sys\time.h">
+      <Filter>Header Files\compat\sys</Filter>
+    </ClInclude>
+    <ClInclude Include="compat\unistd.h">
+      <Filter>Header Files\compat</Filter>
+    </ClInclude>
+    <ClInclude Include="compat\inttypes.h">
+      <Filter>Header Files\compat</Filter>
+    </ClInclude>
+    <ClInclude Include="compat\getopt\getopt.h">
+      <Filter>Header Files\compat\getopt</Filter>
+    </ClInclude>
+    <ClInclude Include="compat\stdbool.h">
+      <Filter>Header Files\compat</Filter>
+    </ClInclude>
+    <ClInclude Include="cpuminer-config.h">
+      <Filter>Header Files</Filter>
+    </ClInclude>
+    <ClInclude Include="sph_blake.h">
+      <Filter>Header Files</Filter>
+    </ClInclude>
+    <ClInclude Include="sph_types.h">
+      <Filter>Header Files</Filter>
+    </ClInclude>
+    <ClInclude Include="sph_groestl.h">
+      <Filter>Header Files</Filter>
+    </ClInclude>
+    <ClInclude Include="sph_keccak.h">
+      <Filter>Header Files</Filter>
+    </ClInclude>
+    <ClInclude Include="hefty1.h">
+      <Filter>Header Files</Filter>
+    </ClInclude>
+    <ClInclude Include="cuda_sha256.h">
+      <Filter>Header Files\CUDA</Filter>
+    </ClInclude>
+    <ClInclude Include="cuda_hefty1.h">
+      <Filter>Header Files\CUDA</Filter>
+    </ClInclude>
+    <ClInclude Include="cuda_keccak512.h">
+      <Filter>Header Files\CUDA</Filter>
+    </ClInclude>
+    <ClInclude Include="cuda_combine.h">
+      <Filter>Header Files\CUDA</Filter>
+    </ClInclude>
+    <ClInclude Include="cuda_blake512.h">
+      <Filter>Header Files\CUDA</Filter>
+    </ClInclude>
+    <ClInclude Include="cuda_groestl512.h">
+      <Filter>Header Files\CUDA</Filter>
+    </ClInclude>
+    <ClInclude Include="sph_fugue.h">
+      <Filter>Header Files</Filter>
+    </ClInclude>
+    <ClInclude Include="uint256.h">
+      <Filter>Header Files</Filter>
+    </ClInclude>
+  </ItemGroup>
+  <ItemGroup>
+    <CudaCompile Include="cuda_sha256.cu">
+      <Filter>Source Files\CUDA</Filter>
+    </CudaCompile>
+    <CudaCompile Include="cuda_blake512.cu">
+      <Filter>Source Files\CUDA</Filter>
+    </CudaCompile>
+    <CudaCompile Include="cuda_groestl512.cu">
+      <Filter>Source Files\CUDA</Filter>
+    </CudaCompile>
+    <CudaCompile Include="cuda_hefty1.cu">
+      <Filter>Source Files\CUDA</Filter>
+    </CudaCompile>
+    <CudaCompile Include="cuda_keccak512.cu">
+      <Filter>Source Files\CUDA</Filter>
+    </CudaCompile>
+    <CudaCompile Include="cuda_combine.cu">
+      <Filter>Source Files\CUDA</Filter>
+    </CudaCompile>
+    <CudaCompile Include="heavy.cu">
+      <Filter>Source Files\CUDA</Filter>
+    </CudaCompile>
+    <CudaCompile Include="cuda_fugue256.cu">
+      <Filter>Source Files\CUDA</Filter>
+    </CudaCompile>
+  </ItemGroup>
+</Project>
\ No newline at end of file
diff --git a/compat.h b/compat.h
new file mode 100644
index 0000000000..ac7b8b9f11
--- /dev/null
+++ b/compat.h
@@ -0,0 +1,24 @@
+#ifndef __COMPAT_H__
+#define __COMPAT_H__
+
+#ifdef WIN32
+
+#include <windows.h>
+
+static __inline void sleep(int secs)
+{
+	Sleep(secs * 1000);
+}
+
+enum {
+	PRIO_PROCESS		= 0,
+};
+
+static __inline int setpriority(int which, int who, int prio)
+{
+	return -!SetThreadPriority(GetCurrentThread(), THREAD_PRIORITY_IDLE /*THREAD_PRIORITY_TIME_CRITICAL*/);
+}
+
+#endif /* WIN32 */
+
+#endif /* __COMPAT_H__ */
diff --git a/compat/Makefile.am b/compat/Makefile.am
new file mode 100644
index 0000000000..9401c8e510
--- /dev/null
+++ b/compat/Makefile.am
@@ -0,0 +1,7 @@
+
+if WANT_JANSSON
+SUBDIRS	= jansson
+else
+SUBDIRS	=
+endif
+
diff --git a/compat/Makefile.in b/compat/Makefile.in
new file mode 100644
index 0000000000..d1d76d9c3b
--- /dev/null
+++ b/compat/Makefile.in
@@ -0,0 +1,601 @@
+# Makefile.in generated by automake 1.13.3 from Makefile.am.
+# @configure_input@
+
+# Copyright (C) 1994-2013 Free Software Foundation, Inc.
+
+# This Makefile.in is free software; the Free Software Foundation
+# gives unlimited permission to copy and/or distribute it,
+# with or without modifications, as long as this notice is preserved.
+
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY, to the extent permitted by law; without
+# even the implied warranty of MERCHANTABILITY or FITNESS FOR A
+# PARTICULAR PURPOSE.
+
+@SET_MAKE@
+VPATH = @srcdir@
+am__is_gnu_make = test -n '$(MAKEFILE_LIST)' && test -n '$(MAKELEVEL)'
+am__make_running_with_option = \
+  case $${target_option-} in \
+      ?) ;; \
+      *) echo "am__make_running_with_option: internal error: invalid" \
+              "target option '$${target_option-}' specified" >&2; \
+         exit 1;; \
+  esac; \
+  has_opt=no; \
+  sane_makeflags=$$MAKEFLAGS; \
+  if $(am__is_gnu_make); then \
+    sane_makeflags=$$MFLAGS; \
+  else \
+    case $$MAKEFLAGS in \
+      *\\[\ \	]*) \
+        bs=\\; \
+        sane_makeflags=`printf '%s\n' "$$MAKEFLAGS" \
+          | sed "s/$$bs$$bs[$$bs $$bs	]*//g"`;; \
+    esac; \
+  fi; \
+  skip_next=no; \
+  strip_trailopt () \
+  { \
+    flg=`printf '%s\n' "$$flg" | sed "s/$$1.*$$//"`; \
+  }; \
+  for flg in $$sane_makeflags; do \
+    test $$skip_next = yes && { skip_next=no; continue; }; \
+    case $$flg in \
+      *=*|--*) continue;; \
+        -*I) strip_trailopt 'I'; skip_next=yes;; \
+      -*I?*) strip_trailopt 'I';; \
+        -*O) strip_trailopt 'O'; skip_next=yes;; \
+      -*O?*) strip_trailopt 'O';; \
+        -*l) strip_trailopt 'l'; skip_next=yes;; \
+      -*l?*) strip_trailopt 'l';; \
+      -[dEDm]) skip_next=yes;; \
+      -[JT]) skip_next=yes;; \
+    esac; \
+    case $$flg in \
+      *$$target_option*) has_opt=yes; break;; \
+    esac; \
+  done; \
+  test $$has_opt = yes
+am__make_dryrun = (target_option=n; $(am__make_running_with_option))
+am__make_keepgoing = (target_option=k; $(am__make_running_with_option))
+pkgdatadir = $(datadir)/@PACKAGE@
+pkgincludedir = $(includedir)/@PACKAGE@
+pkglibdir = $(libdir)/@PACKAGE@
+pkglibexecdir = $(libexecdir)/@PACKAGE@
+am__cd = CDPATH="$${ZSH_VERSION+.}$(PATH_SEPARATOR)" && cd
+install_sh_DATA = $(install_sh) -c -m 644
+install_sh_PROGRAM = $(install_sh) -c
+install_sh_SCRIPT = $(install_sh) -c
+INSTALL_HEADER = $(INSTALL_DATA)
+transform = $(program_transform_name)
+NORMAL_INSTALL = :
+PRE_INSTALL = :
+POST_INSTALL = :
+NORMAL_UNINSTALL = :
+PRE_UNINSTALL = :
+POST_UNINSTALL = :
+build_triplet = @build@
+host_triplet = @host@
+target_triplet = @target@
+subdir = compat
+DIST_COMMON = $(srcdir)/Makefile.in $(srcdir)/Makefile.am
+ACLOCAL_M4 = $(top_srcdir)/aclocal.m4
+am__aclocal_m4_deps = $(top_srcdir)/configure.ac
+am__configure_deps = $(am__aclocal_m4_deps) $(CONFIGURE_DEPENDENCIES) \
+	$(ACLOCAL_M4)
+mkinstalldirs = $(install_sh) -d
+CONFIG_HEADER = $(top_builddir)/cpuminer-config.h
+CONFIG_CLEAN_FILES =
+CONFIG_CLEAN_VPATH_FILES =
+AM_V_P = $(am__v_P_@AM_V@)
+am__v_P_ = $(am__v_P_@AM_DEFAULT_V@)
+am__v_P_0 = false
+am__v_P_1 = :
+AM_V_GEN = $(am__v_GEN_@AM_V@)
+am__v_GEN_ = $(am__v_GEN_@AM_DEFAULT_V@)
+am__v_GEN_0 = @echo "  GEN     " $@;
+am__v_GEN_1 = 
+AM_V_at = $(am__v_at_@AM_V@)
+am__v_at_ = $(am__v_at_@AM_DEFAULT_V@)
+am__v_at_0 = @
+am__v_at_1 = 
+SOURCES =
+DIST_SOURCES =
+RECURSIVE_TARGETS = all-recursive check-recursive cscopelist-recursive \
+	ctags-recursive dvi-recursive html-recursive info-recursive \
+	install-data-recursive install-dvi-recursive \
+	install-exec-recursive install-html-recursive \
+	install-info-recursive install-pdf-recursive \
+	install-ps-recursive install-recursive installcheck-recursive \
+	installdirs-recursive pdf-recursive ps-recursive \
+	tags-recursive uninstall-recursive
+am__can_run_installinfo = \
+  case $$AM_UPDATE_INFO_DIR in \
+    n|no|NO) false;; \
+    *) (install-info --version) >/dev/null 2>&1;; \
+  esac
+RECURSIVE_CLEAN_TARGETS = mostlyclean-recursive clean-recursive	\
+  distclean-recursive maintainer-clean-recursive
+am__recursive_targets = \
+  $(RECURSIVE_TARGETS) \
+  $(RECURSIVE_CLEAN_TARGETS) \
+  $(am__extra_recursive_targets)
+AM_RECURSIVE_TARGETS = $(am__recursive_targets:-recursive=) TAGS CTAGS \
+	distdir
+am__tagged_files = $(HEADERS) $(SOURCES) $(TAGS_FILES) $(LISP)
+# Read a list of newline-separated strings from the standard input,
+# and print each of them once, without duplicates.  Input order is
+# *not* preserved.
+am__uniquify_input = $(AWK) '\
+  BEGIN { nonempty = 0; } \
+  { items[$$0] = 1; nonempty = 1; } \
+  END { if (nonempty) { for (i in items) print i; }; } \
+'
+# Make sure the list of sources is unique.  This is necessary because,
+# e.g., the same source file might be shared among _SOURCES variables
+# for different programs/libraries.
+am__define_uniq_tagged_files = \
+  list='$(am__tagged_files)'; \
+  unique=`for i in $$list; do \
+    if test -f "$$i"; then echo $$i; else echo $(srcdir)/$$i; fi; \
+  done | $(am__uniquify_input)`
+ETAGS = etags
+CTAGS = ctags
+DIST_SUBDIRS = jansson
+DISTFILES = $(DIST_COMMON) $(DIST_SOURCES) $(TEXINFOS) $(EXTRA_DIST)
+am__relativize = \
+  dir0=`pwd`; \
+  sed_first='s,^\([^/]*\)/.*$$,\1,'; \
+  sed_rest='s,^[^/]*/*,,'; \
+  sed_last='s,^.*/\([^/]*\)$$,\1,'; \
+  sed_butlast='s,/*[^/]*$$,,'; \
+  while test -n "$$dir1"; do \
+    first=`echo "$$dir1" | sed -e "$$sed_first"`; \
+    if test "$$first" != "."; then \
+      if test "$$first" = ".."; then \
+        dir2=`echo "$$dir0" | sed -e "$$sed_last"`/"$$dir2"; \
+        dir0=`echo "$$dir0" | sed -e "$$sed_butlast"`; \
+      else \
+        first2=`echo "$$dir2" | sed -e "$$sed_first"`; \
+        if test "$$first2" = "$$first"; then \
+          dir2=`echo "$$dir2" | sed -e "$$sed_rest"`; \
+        else \
+          dir2="../$$dir2"; \
+        fi; \
+        dir0="$$dir0"/"$$first"; \
+      fi; \
+    fi; \
+    dir1=`echo "$$dir1" | sed -e "$$sed_rest"`; \
+  done; \
+  reldir="$$dir2"
+ACLOCAL = @ACLOCAL@
+ALLOCA = @ALLOCA@
+AMTAR = @AMTAR@
+AM_DEFAULT_VERBOSITY = @AM_DEFAULT_VERBOSITY@
+AUTOCONF = @AUTOCONF@
+AUTOHEADER = @AUTOHEADER@
+AUTOMAKE = @AUTOMAKE@
+AWK = @AWK@
+CC = @CC@
+CCAS = @CCAS@
+CCASDEPMODE = @CCASDEPMODE@
+CCASFLAGS = @CCASFLAGS@
+CCDEPMODE = @CCDEPMODE@
+CFLAGS = @CFLAGS@
+CPP = @CPP@
+CPPFLAGS = @CPPFLAGS@
+CUDA_CFLAGS = @CUDA_CFLAGS@
+CUDA_LDFLAGS = @CUDA_LDFLAGS@
+CUDA_LIBS = @CUDA_LIBS@
+CXX = @CXX@
+CXXDEPMODE = @CXXDEPMODE@
+CXXFLAGS = @CXXFLAGS@
+CYGPATH_W = @CYGPATH_W@
+DEFS = @DEFS@
+DEPDIR = @DEPDIR@
+ECHO_C = @ECHO_C@
+ECHO_N = @ECHO_N@
+ECHO_T = @ECHO_T@
+EGREP = @EGREP@
+EXEEXT = @EXEEXT@
+GREP = @GREP@
+INSTALL = @INSTALL@
+INSTALL_DATA = @INSTALL_DATA@
+INSTALL_PROGRAM = @INSTALL_PROGRAM@
+INSTALL_SCRIPT = @INSTALL_SCRIPT@
+INSTALL_STRIP_PROGRAM = @INSTALL_STRIP_PROGRAM@
+JANSSON_LIBS = @JANSSON_LIBS@
+LDFLAGS = @LDFLAGS@
+LIBCURL = @LIBCURL@
+LIBCURL_CPPFLAGS = @LIBCURL_CPPFLAGS@
+LIBOBJS = @LIBOBJS@
+LIBS = @LIBS@
+LTLIBOBJS = @LTLIBOBJS@
+MAINT = @MAINT@
+MAKEINFO = @MAKEINFO@
+MKDIR_P = @MKDIR_P@
+NVCC = @NVCC@
+OBJEXT = @OBJEXT@
+OPENMP_CFLAGS = @OPENMP_CFLAGS@
+PACKAGE = @PACKAGE@
+PACKAGE_BUGREPORT = @PACKAGE_BUGREPORT@
+PACKAGE_NAME = @PACKAGE_NAME@
+PACKAGE_STRING = @PACKAGE_STRING@
+PACKAGE_TARNAME = @PACKAGE_TARNAME@
+PACKAGE_URL = @PACKAGE_URL@
+PACKAGE_VERSION = @PACKAGE_VERSION@
+PATH_SEPARATOR = @PATH_SEPARATOR@
+PTHREAD_FLAGS = @PTHREAD_FLAGS@
+PTHREAD_LIBS = @PTHREAD_LIBS@
+RANLIB = @RANLIB@
+SET_MAKE = @SET_MAKE@
+SHELL = @SHELL@
+STRIP = @STRIP@
+VERSION = @VERSION@
+WS2_LIBS = @WS2_LIBS@
+_libcurl_config = @_libcurl_config@
+abs_builddir = @abs_builddir@
+abs_srcdir = @abs_srcdir@
+abs_top_builddir = @abs_top_builddir@
+abs_top_srcdir = @abs_top_srcdir@
+ac_ct_CC = @ac_ct_CC@
+ac_ct_CXX = @ac_ct_CXX@
+am__include = @am__include@
+am__leading_dot = @am__leading_dot@
+am__quote = @am__quote@
+am__tar = @am__tar@
+am__untar = @am__untar@
+bindir = @bindir@
+build = @build@
+build_alias = @build_alias@
+build_cpu = @build_cpu@
+build_os = @build_os@
+build_vendor = @build_vendor@
+builddir = @builddir@
+datadir = @datadir@
+datarootdir = @datarootdir@
+docdir = @docdir@
+dvidir = @dvidir@
+exec_prefix = @exec_prefix@
+host = @host@
+host_alias = @host_alias@
+host_cpu = @host_cpu@
+host_os = @host_os@
+host_vendor = @host_vendor@
+htmldir = @htmldir@
+includedir = @includedir@
+infodir = @infodir@
+install_sh = @install_sh@
+libdir = @libdir@
+libexecdir = @libexecdir@
+localedir = @localedir@
+localstatedir = @localstatedir@
+mandir = @mandir@
+mkdir_p = @mkdir_p@
+oldincludedir = @oldincludedir@
+pdfdir = @pdfdir@
+prefix = @prefix@
+program_transform_name = @program_transform_name@
+psdir = @psdir@
+sbindir = @sbindir@
+sharedstatedir = @sharedstatedir@
+srcdir = @srcdir@
+sysconfdir = @sysconfdir@
+target = @target@
+target_alias = @target_alias@
+target_cpu = @target_cpu@
+target_os = @target_os@
+target_vendor = @target_vendor@
+top_build_prefix = @top_build_prefix@
+top_builddir = @top_builddir@
+top_srcdir = @top_srcdir@
+@WANT_JANSSON_FALSE@SUBDIRS = 
+@WANT_JANSSON_TRUE@SUBDIRS = jansson
+all: all-recursive
+
+.SUFFIXES:
+$(srcdir)/Makefile.in: @MAINTAINER_MODE_TRUE@ $(srcdir)/Makefile.am  $(am__configure_deps)
+	@for dep in $?; do \
+	  case '$(am__configure_deps)' in \
+	    *$$dep*) \
+	      ( cd $(top_builddir) && $(MAKE) $(AM_MAKEFLAGS) am--refresh ) \
+	        && { if test -f $@; then exit 0; else break; fi; }; \
+	      exit 1;; \
+	  esac; \
+	done; \
+	echo ' cd $(top_srcdir) && $(AUTOMAKE) --gnu compat/Makefile'; \
+	$(am__cd) $(top_srcdir) && \
+	  $(AUTOMAKE) --gnu compat/Makefile
+.PRECIOUS: Makefile
+Makefile: $(srcdir)/Makefile.in $(top_builddir)/config.status
+	@case '$?' in \
+	  *config.status*) \
+	    cd $(top_builddir) && $(MAKE) $(AM_MAKEFLAGS) am--refresh;; \
+	  *) \
+	    echo ' cd $(top_builddir) && $(SHELL) ./config.status $(subdir)/$@ $(am__depfiles_maybe)'; \
+	    cd $(top_builddir) && $(SHELL) ./config.status $(subdir)/$@ $(am__depfiles_maybe);; \
+	esac;
+
+$(top_builddir)/config.status: $(top_srcdir)/configure $(CONFIG_STATUS_DEPENDENCIES)
+	cd $(top_builddir) && $(MAKE) $(AM_MAKEFLAGS) am--refresh
+
+$(top_srcdir)/configure: @MAINTAINER_MODE_TRUE@ $(am__configure_deps)
+	cd $(top_builddir) && $(MAKE) $(AM_MAKEFLAGS) am--refresh
+$(ACLOCAL_M4): @MAINTAINER_MODE_TRUE@ $(am__aclocal_m4_deps)
+	cd $(top_builddir) && $(MAKE) $(AM_MAKEFLAGS) am--refresh
+$(am__aclocal_m4_deps):
+
+# This directory's subdirectories are mostly independent; you can cd
+# into them and run 'make' without going through this Makefile.
+# To change the values of 'make' variables: instead of editing Makefiles,
+# (1) if the variable is set in 'config.status', edit 'config.status'
+#     (which will cause the Makefiles to be regenerated when you run 'make');
+# (2) otherwise, pass the desired values on the 'make' command line.
+$(am__recursive_targets):
+	@fail=; \
+	if $(am__make_keepgoing); then \
+	  failcom='fail=yes'; \
+	else \
+	  failcom='exit 1'; \
+	fi; \
+	dot_seen=no; \
+	target=`echo $@ | sed s/-recursive//`; \
+	case "$@" in \
+	  distclean-* | maintainer-clean-*) list='$(DIST_SUBDIRS)' ;; \
+	  *) list='$(SUBDIRS)' ;; \
+	esac; \
+	for subdir in $$list; do \
+	  echo "Making $$target in $$subdir"; \
+	  if test "$$subdir" = "."; then \
+	    dot_seen=yes; \
+	    local_target="$$target-am"; \
+	  else \
+	    local_target="$$target"; \
+	  fi; \
+	  ($(am__cd) $$subdir && $(MAKE) $(AM_MAKEFLAGS) $$local_target) \
+	  || eval $$failcom; \
+	done; \
+	if test "$$dot_seen" = "no"; then \
+	  $(MAKE) $(AM_MAKEFLAGS) "$$target-am" || exit 1; \
+	fi; test -z "$$fail"
+
+ID: $(am__tagged_files)
+	$(am__define_uniq_tagged_files); mkid -fID $$unique
+tags: tags-recursive
+TAGS: tags
+
+tags-am: $(TAGS_DEPENDENCIES) $(am__tagged_files)
+	set x; \
+	here=`pwd`; \
+	if ($(ETAGS) --etags-include --version) >/dev/null 2>&1; then \
+	  include_option=--etags-include; \
+	  empty_fix=.; \
+	else \
+	  include_option=--include; \
+	  empty_fix=; \
+	fi; \
+	list='$(SUBDIRS)'; for subdir in $$list; do \
+	  if test "$$subdir" = .; then :; else \
+	    test ! -f $$subdir/TAGS || \
+	      set "$$@" "$$include_option=$$here/$$subdir/TAGS"; \
+	  fi; \
+	done; \
+	$(am__define_uniq_tagged_files); \
+	shift; \
+	if test -z "$(ETAGS_ARGS)$$*$$unique"; then :; else \
+	  test -n "$$unique" || unique=$$empty_fix; \
+	  if test $$# -gt 0; then \
+	    $(ETAGS) $(ETAGSFLAGS) $(AM_ETAGSFLAGS) $(ETAGS_ARGS) \
+	      "$$@" $$unique; \
+	  else \
+	    $(ETAGS) $(ETAGSFLAGS) $(AM_ETAGSFLAGS) $(ETAGS_ARGS) \
+	      $$unique; \
+	  fi; \
+	fi
+ctags: ctags-recursive
+
+CTAGS: ctags
+ctags-am: $(TAGS_DEPENDENCIES) $(am__tagged_files)
+	$(am__define_uniq_tagged_files); \
+	test -z "$(CTAGS_ARGS)$$unique" \
+	  || $(CTAGS) $(CTAGSFLAGS) $(AM_CTAGSFLAGS) $(CTAGS_ARGS) \
+	     $$unique
+
+GTAGS:
+	here=`$(am__cd) $(top_builddir) && pwd` \
+	  && $(am__cd) $(top_srcdir) \
+	  && gtags -i $(GTAGS_ARGS) "$$here"
+cscopelist: cscopelist-recursive
+
+cscopelist-am: $(am__tagged_files)
+	list='$(am__tagged_files)'; \
+	case "$(srcdir)" in \
+	  [\\/]* | ?:[\\/]*) sdir="$(srcdir)" ;; \
+	  *) sdir=$(subdir)/$(srcdir) ;; \
+	esac; \
+	for i in $$list; do \
+	  if test -f "$$i"; then \
+	    echo "$(subdir)/$$i"; \
+	  else \
+	    echo "$$sdir/$$i"; \
+	  fi; \
+	done >> $(top_builddir)/cscope.files
+
+distclean-tags:
+	-rm -f TAGS ID GTAGS GRTAGS GSYMS GPATH tags
+
+distdir: $(DISTFILES)
+	@srcdirstrip=`echo "$(srcdir)" | sed 's/[].[^$$\\*]/\\\\&/g'`; \
+	topsrcdirstrip=`echo "$(top_srcdir)" | sed 's/[].[^$$\\*]/\\\\&/g'`; \
+	list='$(DISTFILES)'; \
+	  dist_files=`for file in $$list; do echo $$file; done | \
+	  sed -e "s|^$$srcdirstrip/||;t" \
+	      -e "s|^$$topsrcdirstrip/|$(top_builddir)/|;t"`; \
+	case $$dist_files in \
+	  */*) $(MKDIR_P) `echo "$$dist_files" | \
+			   sed '/\//!d;s|^|$(distdir)/|;s,/[^/]*$$,,' | \
+			   sort -u` ;; \
+	esac; \
+	for file in $$dist_files; do \
+	  if test -f $$file || test -d $$file; then d=.; else d=$(srcdir); fi; \
+	  if test -d $$d/$$file; then \
+	    dir=`echo "/$$file" | sed -e 's,/[^/]*$$,,'`; \
+	    if test -d "$(distdir)/$$file"; then \
+	      find "$(distdir)/$$file" -type d ! -perm -700 -exec chmod u+rwx {} \;; \
+	    fi; \
+	    if test -d $(srcdir)/$$file && test $$d != $(srcdir); then \
+	      cp -fpR $(srcdir)/$$file "$(distdir)$$dir" || exit 1; \
+	      find "$(distdir)/$$file" -type d ! -perm -700 -exec chmod u+rwx {} \;; \
+	    fi; \
+	    cp -fpR $$d/$$file "$(distdir)$$dir" || exit 1; \
+	  else \
+	    test -f "$(distdir)/$$file" \
+	    || cp -p $$d/$$file "$(distdir)/$$file" \
+	    || exit 1; \
+	  fi; \
+	done
+	@list='$(DIST_SUBDIRS)'; for subdir in $$list; do \
+	  if test "$$subdir" = .; then :; else \
+	    $(am__make_dryrun) \
+	      || test -d "$(distdir)/$$subdir" \
+	      || $(MKDIR_P) "$(distdir)/$$subdir" \
+	      || exit 1; \
+	    dir1=$$subdir; dir2="$(distdir)/$$subdir"; \
+	    $(am__relativize); \
+	    new_distdir=$$reldir; \
+	    dir1=$$subdir; dir2="$(top_distdir)"; \
+	    $(am__relativize); \
+	    new_top_distdir=$$reldir; \
+	    echo " (cd $$subdir && $(MAKE) $(AM_MAKEFLAGS) top_distdir="$$new_top_distdir" distdir="$$new_distdir" \\"; \
+	    echo "     am__remove_distdir=: am__skip_length_check=: am__skip_mode_fix=: distdir)"; \
+	    ($(am__cd) $$subdir && \
+	      $(MAKE) $(AM_MAKEFLAGS) \
+	        top_distdir="$$new_top_distdir" \
+	        distdir="$$new_distdir" \
+		am__remove_distdir=: \
+		am__skip_length_check=: \
+		am__skip_mode_fix=: \
+	        distdir) \
+	      || exit 1; \
+	  fi; \
+	done
+check-am: all-am
+check: check-recursive
+all-am: Makefile
+installdirs: installdirs-recursive
+installdirs-am:
+install: install-recursive
+install-exec: install-exec-recursive
+install-data: install-data-recursive
+uninstall: uninstall-recursive
+
+install-am: all-am
+	@$(MAKE) $(AM_MAKEFLAGS) install-exec-am install-data-am
+
+installcheck: installcheck-recursive
+install-strip:
+	if test -z '$(STRIP)'; then \
+	  $(MAKE) $(AM_MAKEFLAGS) INSTALL_PROGRAM="$(INSTALL_STRIP_PROGRAM)" \
+	    install_sh_PROGRAM="$(INSTALL_STRIP_PROGRAM)" INSTALL_STRIP_FLAG=-s \
+	      install; \
+	else \
+	  $(MAKE) $(AM_MAKEFLAGS) INSTALL_PROGRAM="$(INSTALL_STRIP_PROGRAM)" \
+	    install_sh_PROGRAM="$(INSTALL_STRIP_PROGRAM)" INSTALL_STRIP_FLAG=-s \
+	    "INSTALL_PROGRAM_ENV=STRIPPROG='$(STRIP)'" install; \
+	fi
+mostlyclean-generic:
+
+clean-generic:
+
+distclean-generic:
+	-test -z "$(CONFIG_CLEAN_FILES)" || rm -f $(CONFIG_CLEAN_FILES)
+	-test . = "$(srcdir)" || test -z "$(CONFIG_CLEAN_VPATH_FILES)" || rm -f $(CONFIG_CLEAN_VPATH_FILES)
+
+maintainer-clean-generic:
+	@echo "This command is intended for maintainers to use"
+	@echo "it deletes files that may require special tools to rebuild."
+clean: clean-recursive
+
+clean-am: clean-generic mostlyclean-am
+
+distclean: distclean-recursive
+	-rm -f Makefile
+distclean-am: clean-am distclean-generic distclean-tags
+
+dvi: dvi-recursive
+
+dvi-am:
+
+html: html-recursive
+
+html-am:
+
+info: info-recursive
+
+info-am:
+
+install-data-am:
+
+install-dvi: install-dvi-recursive
+
+install-dvi-am:
+
+install-exec-am:
+
+install-html: install-html-recursive
+
+install-html-am:
+
+install-info: install-info-recursive
+
+install-info-am:
+
+install-man:
+
+install-pdf: install-pdf-recursive
+
+install-pdf-am:
+
+install-ps: install-ps-recursive
+
+install-ps-am:
+
+installcheck-am:
+
+maintainer-clean: maintainer-clean-recursive
+	-rm -f Makefile
+maintainer-clean-am: distclean-am maintainer-clean-generic
+
+mostlyclean: mostlyclean-recursive
+
+mostlyclean-am: mostlyclean-generic
+
+pdf: pdf-recursive
+
+pdf-am:
+
+ps: ps-recursive
+
+ps-am:
+
+uninstall-am:
+
+.MAKE: $(am__recursive_targets) install-am install-strip
+
+.PHONY: $(am__recursive_targets) CTAGS GTAGS TAGS all all-am check \
+	check-am clean clean-generic cscopelist-am ctags ctags-am \
+	distclean distclean-generic distclean-tags distdir dvi dvi-am \
+	html html-am info info-am install install-am install-data \
+	install-data-am install-dvi install-dvi-am install-exec \
+	install-exec-am install-html install-html-am install-info \
+	install-info-am install-man install-pdf install-pdf-am \
+	install-ps install-ps-am install-strip installcheck \
+	installcheck-am installdirs installdirs-am maintainer-clean \
+	maintainer-clean-generic mostlyclean mostlyclean-generic pdf \
+	pdf-am ps ps-am tags tags-am uninstall uninstall-am
+
+
+# Tell versions [3.59,3.63) of GNU make to not export all variables.
+# Otherwise a system limit (for SysV at least) may be exceeded.
+.NOEXPORT:
diff --git a/compat/getopt/getopt.h b/compat/getopt/getopt.h
new file mode 100644
index 0000000000..068cc24618
--- /dev/null
+++ b/compat/getopt/getopt.h
@@ -0,0 +1,93 @@
+/*	$Id: getopt.h,v 1.1 2009/10/16 19:50:28 rodney Exp rodney $ */
+/*	$OpenBSD: getopt.h,v 1.1 2002/12/03 20:24:29 millert Exp $	*/
+/*	$NetBSD: getopt.h,v 1.4 2000/07/07 10:43:54 ad Exp $	*/
+
+/*-
+ * Copyright (c) 2000 The NetBSD Foundation, Inc.
+ * All rights reserved.
+ *
+ * This code is derived from software contributed to The NetBSD Foundation
+ * by Dieter Baron and Thomas Klausner.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ * 3. All advertising materials mentioning features or use of this software
+ *    must display the following acknowledgement:
+ *        This product includes software developed by the NetBSD
+ *        Foundation, Inc. and its contributors.
+ * 4. Neither the name of The NetBSD Foundation nor the names of its
+ *    contributors may be used to endorse or promote products derived
+ *    from this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
+ * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
+ * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+ * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
+ * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+ * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+ * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+ * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+ * POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#ifndef _GETOPT_H_
+#define _GETOPT_H_
+
+#if 0
+#include <sys/cdefs.h>
+#endif
+
+/*
+ * GNU-like getopt_long() and 4.4BSD getsubopt()/optreset extensions
+ */
+#define no_argument        0
+#define required_argument  1
+#define optional_argument  2
+
+struct option {
+	/* name of long option */
+	const char *name;
+	/*
+	 * one of no_argument, required_argument, and optional_argument:
+	 * whether option takes an argument
+	 */
+	int has_arg;
+	/* if not NULL, set *flag to val when option found */
+	int *flag;
+	/* if flag not NULL, value to set *flag to; else return value */
+	int val;
+};
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+int	 getopt_long(int, char * const *, const char *,
+	    const struct option *, int *);
+int	 getopt_long_only(int, char * const *, const char *,
+	    const struct option *, int *);
+#ifndef _GETOPT_DEFINED
+#define _GETOPT_DEFINED
+int	 getopt(int, char * const *, const char *);
+int	 getsubopt(char **, char * const *, char **);
+
+extern   char *optarg;                  /* getopt(3) external variables */
+extern   int opterr;
+extern   int optind;
+extern   int optopt;
+extern   int optreset;
+extern   char *suboptarg;               /* getsubopt(3) external variable */
+#endif /* _GETOPT_DEFINED */
+ 
+#ifdef __cplusplus
+}
+#endif
+#endif /* !_GETOPT_H_ */
diff --git a/compat/getopt/getopt_long.c b/compat/getopt/getopt_long.c
new file mode 100644
index 0000000000..90fc0cfaac
--- /dev/null
+++ b/compat/getopt/getopt_long.c
@@ -0,0 +1,554 @@
+/*	$Id: getopt_long.c,v 1.1 2009/10/16 19:50:28 rodney Exp rodney $	*/
+/*	$OpenBSD: getopt_long.c,v 1.23 2007/10/31 12:34:57 chl Exp $	*/
+/*	$NetBSD: getopt_long.c,v 1.15 2002/01/31 22:43:40 tv Exp $	*/
+
+/*
+ * Copyright (c) 2002 Todd C. Miller <Todd.Miller@courtesan.com>
+ *
+ * Permission to use, copy, modify, and distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ *
+ * Sponsored in part by the Defense Advanced Research Projects
+ * Agency (DARPA) and Air Force Research Laboratory, Air Force
+ * Materiel Command, USAF, under agreement number F39502-99-1-0512.
+ */
+
+#ifndef lint
+static const char rcsid[]="$Id: getopt_long.c,v 1.1 2009/10/16 19:50:28 rodney Exp rodney $";
+#endif /* lint */
+/*-
+ * Copyright (c) 2000 The NetBSD Foundation, Inc.
+ * All rights reserved.
+ *
+ * This code is derived from software contributed to The NetBSD Foundation
+ * by Dieter Baron and Thomas Klausner.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
+ * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
+ * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+ * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
+ * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+ * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+ * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+ * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+ * POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#if 0
+#include <err.h>
+#endif
+#include <errno.h>
+#include <getopt.h>
+#include <stdlib.h>
+#include <string.h>
+
+#ifdef _WIN32
+
+/* Windows needs warnx().  We change the definition though:
+ *  1. (another) global is defined, opterrmsg, which holds the error message
+ *  2. errors are always printed out on stderr w/o the program name
+ * Note that opterrmsg always gets set no matter what opterr is set to.  The
+ * error message will not be printed if opterr is 0 as usual.
+ */
+
+#include <stdio.h>
+#include <stdarg.h>
+#include <stdarg.h>
+
+char opterrmsg[128]; /* last error message is stored here */
+
+static void warnx(const char *fmt, ...)
+{
+	va_list ap;
+	va_start(ap, fmt);
+	if (fmt != NULL)
+		_vsnprintf(opterrmsg, 128, fmt, ap);
+	else
+		opterrmsg[0]='\0';
+	va_end(ap);
+	fprintf(stderr, opterrmsg);
+	fprintf(stderr, "\n");
+}
+
+#endif /*_WIN32*/
+
+#define	REPLACE_GETOPT		/* use this getopt as the system getopt(3) */
+
+#ifdef REPLACE_GETOPT
+int	opterr = 1;		/* if error message should be printed */
+int	optind = 1;		/* index into parent argv vector */
+int	optopt = '?';		/* character checked for validity */
+int	optreset;		/* reset getopt */
+char    *optarg;		/* argument associated with option */
+#endif
+
+#define PRINT_ERROR	((opterr) && (*options != ':'))
+
+#define FLAG_PERMUTE	0x01	/* permute non-options to the end of argv */
+#define FLAG_ALLARGS	0x02	/* treat non-options as args to option "-1" */
+#define FLAG_LONGONLY	0x04	/* operate as getopt_long_only */
+
+/* return values */
+#define	BADCH		(int)'?'
+#define	BADARG		((*options == ':') ? (int)':' : (int)'?')
+#define	INORDER 	(int)1
+
+#define	EMSG		""
+
+static int getopt_internal(int, char * const *, const char *,
+			   const struct option *, int *, int);
+static int parse_long_options(char * const *, const char *,
+			      const struct option *, int *, int);
+static int gcd(int, int);
+static void permute_args(int, int, int, char * const *);
+
+static char *place = EMSG; /* option letter processing */
+
+/* XXX: set optreset to 1 rather than these two */
+static int nonopt_start = -1; /* first non option argument (for permute) */
+static int nonopt_end = -1;   /* first option after non options (for permute) */
+
+/* Error messages */
+static const char recargchar[] = "option requires an argument -- %c";
+static const char recargstring[] = "option requires an argument -- %s";
+static const char ambig[] = "ambiguous option -- %.*s";
+static const char noarg[] = "option doesn't take an argument -- %.*s";
+static const char illoptchar[] = "unknown option -- %c";
+static const char illoptstring[] = "unknown option -- %s";
+
+/*
+ * Compute the greatest common divisor of a and b.
+ */
+static int
+gcd(int a, int b)
+{
+	int c;
+
+	c = a % b;
+	while (c != 0) {
+		a = b;
+		b = c;
+		c = a % b;
+	}
+
+	return (b);
+}
+
+/*
+ * Exchange the block from nonopt_start to nonopt_end with the block
+ * from nonopt_end to opt_end (keeping the same order of arguments
+ * in each block).
+ */
+static void
+permute_args(int panonopt_start, int panonopt_end, int opt_end,
+	char * const *nargv)
+{
+	int cstart, cyclelen, i, j, ncycle, nnonopts, nopts, pos;
+	char *swap;
+
+	/*
+	 * compute lengths of blocks and number and size of cycles
+	 */
+	nnonopts = panonopt_end - panonopt_start;
+	nopts = opt_end - panonopt_end;
+	ncycle = gcd(nnonopts, nopts);
+	cyclelen = (opt_end - panonopt_start) / ncycle;
+
+	for (i = 0; i < ncycle; i++) {
+		cstart = panonopt_end+i;
+		pos = cstart;
+		for (j = 0; j < cyclelen; j++) {
+			if (pos >= panonopt_end)
+				pos -= nnonopts;
+			else
+				pos += nopts;
+			swap = nargv[pos];
+			/* LINTED const cast */
+			((char **) nargv)[pos] = nargv[cstart];
+			/* LINTED const cast */
+			((char **)nargv)[cstart] = swap;
+		}
+	}
+}
+
+/*
+ * parse_long_options --
+ *	Parse long options in argc/argv argument vector.
+ * Returns -1 if short_too is set and the option does not match long_options.
+ */
+static int
+parse_long_options(char * const *nargv, const char *options,
+	const struct option *long_options, int *idx, int short_too)
+{
+	char *current_argv, *has_equal;
+	size_t current_argv_len;
+	int i, match;
+
+	current_argv = place;
+	match = -1;
+
+	optind++;
+
+	if ((has_equal = strchr(current_argv, '=')) != NULL) {
+		/* argument found (--option=arg) */
+		current_argv_len = has_equal - current_argv;
+		has_equal++;
+	} else
+		current_argv_len = strlen(current_argv);
+
+	for (i = 0; long_options[i].name; i++) {
+		/* find matching long option */
+		if (strncmp(current_argv, long_options[i].name,
+		    current_argv_len))
+			continue;
+
+		if (strlen(long_options[i].name) == current_argv_len) {
+			/* exact match */
+			match = i;
+			break;
+		}
+		/*
+		 * If this is a known short option, don't allow
+		 * a partial match of a single character.
+		 */
+		if (short_too && current_argv_len == 1)
+			continue;
+
+		if (match == -1)	/* partial match */
+			match = i;
+		else {
+			/* ambiguous abbreviation */
+			if (PRINT_ERROR)
+				warnx(ambig, (int)current_argv_len,
+				     current_argv);
+			optopt = 0;
+			return (BADCH);
+		}
+	}
+	if (match != -1) {		/* option found */
+		if (long_options[match].has_arg == no_argument
+		    && has_equal) {
+			if (PRINT_ERROR)
+				warnx(noarg, (int)current_argv_len,
+				     current_argv);
+			/*
+			 * XXX: GNU sets optopt to val regardless of flag
+			 */
+			if (long_options[match].flag == NULL)
+				optopt = long_options[match].val;
+			else
+				optopt = 0;
+			return (BADARG);
+		}
+		if (long_options[match].has_arg == required_argument ||
+		    long_options[match].has_arg == optional_argument) {
+			if (has_equal)
+				optarg = has_equal;
+			else if (long_options[match].has_arg ==
+			    required_argument) {
+				/*
+				 * optional argument doesn't use next nargv
+				 */
+				optarg = nargv[optind++];
+			}
+		}
+		if ((long_options[match].has_arg == required_argument)
+		    && (optarg == NULL)) {
+			/*
+			 * Missing argument; leading ':' indicates no error
+			 * should be generated.
+			 */
+			if (PRINT_ERROR)
+				warnx(recargstring,
+				    current_argv);
+			/*
+			 * XXX: GNU sets optopt to val regardless of flag
+			 */
+			if (long_options[match].flag == NULL)
+				optopt = long_options[match].val;
+			else
+				optopt = 0;
+			--optind;
+			return (BADARG);
+		}
+	} else {			/* unknown option */
+		if (short_too) {
+			--optind;
+			return (-1);
+		}
+		if (PRINT_ERROR)
+			warnx(illoptstring, current_argv);
+		optopt = 0;
+		return (BADCH);
+	}
+	if (idx)
+		*idx = match;
+	if (long_options[match].flag) {
+		*long_options[match].flag = long_options[match].val;
+		return (0);
+	} else
+		return (long_options[match].val);
+}
+
+/*
+ * getopt_internal --
+ *	Parse argc/argv argument vector.  Called by user level routines.
+ */
+static int
+getopt_internal(int nargc, char * const *nargv, const char *options,
+	const struct option *long_options, int *idx, int flags)
+{
+	char *oli;				/* option letter list index */
+	int optchar, short_too;
+	static int posixly_correct = -1;
+
+	if (options == NULL)
+		return (-1);
+
+	/*
+	 * Disable GNU extensions if POSIXLY_CORRECT is set or options
+	 * string begins with a '+'.
+	 */
+	if (posixly_correct == -1)
+		posixly_correct = (getenv("POSIXLY_CORRECT") != NULL);
+	if (posixly_correct || *options == '+')
+		flags &= ~FLAG_PERMUTE;
+	else if (*options == '-')
+		flags |= FLAG_ALLARGS;
+	if (*options == '+' || *options == '-')
+		options++;
+
+	/*
+	 * XXX Some GNU programs (like cvs) set optind to 0 instead of
+	 * XXX using optreset.  Work around this braindamage.
+	 */
+	if (optind == 0)
+		optind = optreset = 1;
+
+	optarg = NULL;
+	if (optreset)
+		nonopt_start = nonopt_end = -1;
+start:
+	if (optreset || !*place) {		/* update scanning pointer */
+		optreset = 0;
+		if (optind >= nargc) {          /* end of argument vector */
+			place = EMSG;
+			if (nonopt_end != -1) {
+				/* do permutation, if we have to */
+				permute_args(nonopt_start, nonopt_end,
+				    optind, nargv);
+				optind -= nonopt_end - nonopt_start;
+			}
+			else if (nonopt_start != -1) {
+				/*
+				 * If we skipped non-options, set optind
+				 * to the first of them.
+				 */
+				optind = nonopt_start;
+			}
+			nonopt_start = nonopt_end = -1;
+			return (-1);
+		}
+		if (*(place = nargv[optind]) != '-' ||
+		    (place[1] == '\0' && strchr(options, '-') == NULL)) {
+			place = EMSG;		/* found non-option */
+			if (flags & FLAG_ALLARGS) {
+				/*
+				 * GNU extension:
+				 * return non-option as argument to option 1
+				 */
+				optarg = nargv[optind++];
+				return (INORDER);
+			}
+			if (!(flags & FLAG_PERMUTE)) {
+				/*
+				 * If no permutation wanted, stop parsing
+				 * at first non-option.
+				 */
+				return (-1);
+			}
+			/* do permutation */
+			if (nonopt_start == -1)
+				nonopt_start = optind;
+			else if (nonopt_end != -1) {
+				permute_args(nonopt_start, nonopt_end,
+				    optind, nargv);
+				nonopt_start = optind -
+				    (nonopt_end - nonopt_start);
+				nonopt_end = -1;
+			}
+			optind++;
+			/* process next argument */
+			goto start;
+		}
+		if (nonopt_start != -1 && nonopt_end == -1)
+			nonopt_end = optind;
+
+		/*
+		 * If we have "-" do nothing, if "--" we are done.
+		 */
+		if (place[1] != '\0' && *++place == '-' && place[1] == '\0') {
+			optind++;
+			place = EMSG;
+			/*
+			 * We found an option (--), so if we skipped
+			 * non-options, we have to permute.
+			 */
+			if (nonopt_end != -1) {
+				permute_args(nonopt_start, nonopt_end,
+				    optind, nargv);
+				optind -= nonopt_end - nonopt_start;
+			}
+			nonopt_start = nonopt_end = -1;
+			return (-1);
+		}
+	}
+
+	/*
+	 * Check long options if:
+	 *  1) we were passed some
+	 *  2) the arg is not just "-"
+	 *  3) either the arg starts with -- we are getopt_long_only()
+	 */
+	if (long_options != NULL && place != nargv[optind] &&
+	    (*place == '-' || (flags & FLAG_LONGONLY))) {
+		short_too = 0;
+		if (*place == '-')
+			place++;		/* --foo long option */
+		else if (*place != ':' && strchr(options, *place) != NULL)
+			short_too = 1;		/* could be short option too */
+
+		optchar = parse_long_options(nargv, options, long_options,
+		    idx, short_too);
+		if (optchar != -1) {
+			place = EMSG;
+			return (optchar);
+		}
+	}
+
+	if ((optchar = (int)*place++) == (int)':' ||
+	    (optchar == (int)'-' && *place != '\0') ||
+	    (oli = strchr(options, optchar)) == NULL) {
+		/*
+		 * If the user specified "-" and  '-' isn't listed in
+		 * options, return -1 (non-option) as per POSIX.
+		 * Otherwise, it is an unknown option character (or ':').
+		 */
+		if (optchar == (int)'-' && *place == '\0')
+			return (-1);
+		if (!*place)
+			++optind;
+		if (PRINT_ERROR)
+			warnx(illoptchar, optchar);
+		optopt = optchar;
+		return (BADCH);
+	}
+	if (long_options != NULL && optchar == 'W' && oli[1] == ';') {
+		/* -W long-option */
+		if (*place)			/* no space */
+			/* NOTHING */;
+		else if (++optind >= nargc) {	/* no arg */
+			place = EMSG;
+			if (PRINT_ERROR)
+				warnx(recargchar, optchar);
+			optopt = optchar;
+			return (BADARG);
+		} else				/* white space */
+			place = nargv[optind];
+		optchar = parse_long_options(nargv, options, long_options,
+		    idx, 0);
+		place = EMSG;
+		return (optchar);
+	}
+	if (*++oli != ':') {			/* doesn't take argument */
+		if (!*place)
+			++optind;
+	} else {				/* takes (optional) argument */
+		optarg = NULL;
+		if (*place)			/* no white space */
+			optarg = place;
+		else if (oli[1] != ':') {	/* arg not optional */
+			if (++optind >= nargc) {	/* no arg */
+				place = EMSG;
+				if (PRINT_ERROR)
+					warnx(recargchar, optchar);
+				optopt = optchar;
+				return (BADARG);
+			} else
+				optarg = nargv[optind];
+		}
+		place = EMSG;
+		++optind;
+	}
+	/* dump back option letter */
+	return (optchar);
+}
+
+#ifdef REPLACE_GETOPT
+/*
+ * getopt --
+ *	Parse argc/argv argument vector.
+ *
+ * [eventually this will replace the BSD getopt]
+ */
+int
+getopt(int nargc, char * const *nargv, const char *options)
+{
+
+	/*
+	 * We don't pass FLAG_PERMUTE to getopt_internal() since
+	 * the BSD getopt(3) (unlike GNU) has never done this.
+	 *
+	 * Furthermore, since many privileged programs call getopt()
+	 * before dropping privileges it makes sense to keep things
+	 * as simple (and bug-free) as possible.
+	 */
+	return (getopt_internal(nargc, nargv, options, NULL, NULL, 0));
+}
+#endif /* REPLACE_GETOPT */
+
+/*
+ * getopt_long --
+ *	Parse argc/argv argument vector.
+ */
+int
+getopt_long(int nargc, char * const *nargv, const char *options,
+    const struct option *long_options, int *idx)
+{
+
+	return (getopt_internal(nargc, nargv, options, long_options, idx,
+	    FLAG_PERMUTE));
+}
+
+/*
+ * getopt_long_only --
+ *	Parse argc/argv argument vector.
+ */
+int
+getopt_long_only(int nargc, char * const *nargv, const char *options,
+    const struct option *long_options, int *idx)
+{
+
+	return (getopt_internal(nargc, nargv, options, long_options, idx,
+	    FLAG_PERMUTE|FLAG_LONGONLY));
+}
diff --git a/compat/gettimeofday.c b/compat/gettimeofday.c
new file mode 100644
index 0000000000..da17893d0f
--- /dev/null
+++ b/compat/gettimeofday.c
@@ -0,0 +1,83 @@
+#include < time.h >
+#include <windows.h> //I've ommited this line.
+#if defined(_MSC_VER) || defined(_MSC_EXTENSIONS)
+  #define DELTA_EPOCH_IN_MICROSECS  11644473600000000Ui64
+#else
+  #define DELTA_EPOCH_IN_MICROSECS  11644473600000000ULL
+#endif
+ 
+struct timezone 
+{
+  int  tz_minuteswest; /* minutes W of Greenwich */
+  int  tz_dsttime;     /* type of dst correction */
+};
+ 
+int gettimeofday(struct timeval *tv, struct timezone *tz)
+{
+  FILETIME ft;
+  unsigned __int64 tmpres = 0;
+  static int tzflag;
+ 
+  if (NULL != tv)
+  {
+    GetSystemTimeAsFileTime(&ft);
+ 
+    tmpres |= ft.dwHighDateTime;
+    tmpres <<= 32;
+    tmpres |= ft.dwLowDateTime;
+ 
+    /*converting file time to unix epoch*/
+    tmpres /= 10;  /*convert into microseconds*/
+    tmpres -= DELTA_EPOCH_IN_MICROSECS; 
+    tv->tv_sec = (long)(tmpres / 1000000UL);
+    tv->tv_usec = (long)(tmpres % 1000000UL);
+  }
+ 
+  if (NULL != tz)
+  {
+    if (!tzflag)
+    {
+      _tzset();
+      tzflag++;
+    }
+    tz->tz_minuteswest = _timezone / 60;
+    tz->tz_dsttime = _daylight;
+  }
+ 
+  return 0;
+}
+
+void usleep(__int64 waitTime) 
+{ 
+    if (waitTime > 0)
+    {
+        if (waitTime > 100)
+        {
+            // use a waitable timer for larger intervals > 0.1ms
+
+            HANDLE timer; 
+            LARGE_INTEGER ft; 
+
+            ft.QuadPart = -(10*waitTime); // Convert to 100 nanosecond interval, negative value indicates relative time
+
+            timer = CreateWaitableTimer(NULL, TRUE, NULL); 
+            SetWaitableTimer(timer, &ft, 0, NULL, NULL, 0); 
+            WaitForSingleObject(timer, INFINITE); 
+            CloseHandle(timer); 
+        }
+        else
+        {
+            // use a polling loop for short intervals <= 100ms
+
+            LARGE_INTEGER perfCnt, start, now;
+            __int64 elapsed;
+ 
+            QueryPerformanceFrequency(&perfCnt);
+            QueryPerformanceCounter(&start);
+            do {
+                QueryPerformanceCounter((LARGE_INTEGER*) &now);
+                elapsed = (__int64)((now.QuadPart - start.QuadPart) / (float)perfCnt.QuadPart * 1000 * 1000);
+            } while ( elapsed < waitTime );
+        }
+    }
+}
diff --git a/compat/inttypes.h b/compat/inttypes.h
new file mode 100644
index 0000000000..f07d50f391
--- /dev/null
+++ b/compat/inttypes.h
@@ -0,0 +1,2 @@
+#pragma once
+#include <stdint.h>
diff --git a/compat/jansson/Makefile.am b/compat/jansson/Makefile.am
new file mode 100644
index 0000000000..ff38e51653
--- /dev/null
+++ b/compat/jansson/Makefile.am
@@ -0,0 +1,18 @@
+
+noinst_LIBRARIES	= libjansson.a
+
+libjansson_a_SOURCES	= \
+			  config.h		\
+			  dump.c		\
+			  hashtable.c		\
+			  hashtable.h		\
+			  jansson.h		\
+			  jansson_private.h	\
+			  load.c		\
+			  strbuffer.c		\
+			  strbuffer.h		\
+			  utf.c			\
+			  utf.h			\
+			  util.h		\
+			  value.c
+
diff --git a/compat/jansson/Makefile.in b/compat/jansson/Makefile.in
new file mode 100644
index 0000000000..f5e30acea8
--- /dev/null
+++ b/compat/jansson/Makefile.in
@@ -0,0 +1,571 @@
+# Makefile.in generated by automake 1.13.3 from Makefile.am.
+# @configure_input@
+
+# Copyright (C) 1994-2013 Free Software Foundation, Inc.
+
+# This Makefile.in is free software; the Free Software Foundation
+# gives unlimited permission to copy and/or distribute it,
+# with or without modifications, as long as this notice is preserved.
+
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY, to the extent permitted by law; without
+# even the implied warranty of MERCHANTABILITY or FITNESS FOR A
+# PARTICULAR PURPOSE.
+
+@SET_MAKE@
+
+VPATH = @srcdir@
+am__is_gnu_make = test -n '$(MAKEFILE_LIST)' && test -n '$(MAKELEVEL)'
+am__make_running_with_option = \
+  case $${target_option-} in \
+      ?) ;; \
+      *) echo "am__make_running_with_option: internal error: invalid" \
+              "target option '$${target_option-}' specified" >&2; \
+         exit 1;; \
+  esac; \
+  has_opt=no; \
+  sane_makeflags=$$MAKEFLAGS; \
+  if $(am__is_gnu_make); then \
+    sane_makeflags=$$MFLAGS; \
+  else \
+    case $$MAKEFLAGS in \
+      *\\[\ \	]*) \
+        bs=\\; \
+        sane_makeflags=`printf '%s\n' "$$MAKEFLAGS" \
+          | sed "s/$$bs$$bs[$$bs $$bs	]*//g"`;; \
+    esac; \
+  fi; \
+  skip_next=no; \
+  strip_trailopt () \
+  { \
+    flg=`printf '%s\n' "$$flg" | sed "s/$$1.*$$//"`; \
+  }; \
+  for flg in $$sane_makeflags; do \
+    test $$skip_next = yes && { skip_next=no; continue; }; \
+    case $$flg in \
+      *=*|--*) continue;; \
+        -*I) strip_trailopt 'I'; skip_next=yes;; \
+      -*I?*) strip_trailopt 'I';; \
+        -*O) strip_trailopt 'O'; skip_next=yes;; \
+      -*O?*) strip_trailopt 'O';; \
+        -*l) strip_trailopt 'l'; skip_next=yes;; \
+      -*l?*) strip_trailopt 'l';; \
+      -[dEDm]) skip_next=yes;; \
+      -[JT]) skip_next=yes;; \
+    esac; \
+    case $$flg in \
+      *$$target_option*) has_opt=yes; break;; \
+    esac; \
+  done; \
+  test $$has_opt = yes
+am__make_dryrun = (target_option=n; $(am__make_running_with_option))
+am__make_keepgoing = (target_option=k; $(am__make_running_with_option))
+pkgdatadir = $(datadir)/@PACKAGE@
+pkgincludedir = $(includedir)/@PACKAGE@
+pkglibdir = $(libdir)/@PACKAGE@
+pkglibexecdir = $(libexecdir)/@PACKAGE@
+am__cd = CDPATH="$${ZSH_VERSION+.}$(PATH_SEPARATOR)" && cd
+install_sh_DATA = $(install_sh) -c -m 644
+install_sh_PROGRAM = $(install_sh) -c
+install_sh_SCRIPT = $(install_sh) -c
+INSTALL_HEADER = $(INSTALL_DATA)
+transform = $(program_transform_name)
+NORMAL_INSTALL = :
+PRE_INSTALL = :
+POST_INSTALL = :
+NORMAL_UNINSTALL = :
+PRE_UNINSTALL = :
+POST_UNINSTALL = :
+build_triplet = @build@
+host_triplet = @host@
+target_triplet = @target@
+subdir = compat/jansson
+DIST_COMMON = $(srcdir)/Makefile.in $(srcdir)/Makefile.am \
+	$(top_srcdir)/depcomp
+ACLOCAL_M4 = $(top_srcdir)/aclocal.m4
+am__aclocal_m4_deps = $(top_srcdir)/configure.ac
+am__configure_deps = $(am__aclocal_m4_deps) $(CONFIGURE_DEPENDENCIES) \
+	$(ACLOCAL_M4)
+mkinstalldirs = $(install_sh) -d
+CONFIG_HEADER = $(top_builddir)/cpuminer-config.h
+CONFIG_CLEAN_FILES =
+CONFIG_CLEAN_VPATH_FILES =
+LIBRARIES = $(noinst_LIBRARIES)
+AR = ar
+ARFLAGS = cru
+AM_V_AR = $(am__v_AR_@AM_V@)
+am__v_AR_ = $(am__v_AR_@AM_DEFAULT_V@)
+am__v_AR_0 = @echo "  AR      " $@;
+am__v_AR_1 = 
+libjansson_a_AR = $(AR) $(ARFLAGS)
+libjansson_a_LIBADD =
+am_libjansson_a_OBJECTS = dump.$(OBJEXT) hashtable.$(OBJEXT) \
+	load.$(OBJEXT) strbuffer.$(OBJEXT) utf.$(OBJEXT) \
+	value.$(OBJEXT)
+libjansson_a_OBJECTS = $(am_libjansson_a_OBJECTS)
+AM_V_P = $(am__v_P_@AM_V@)
+am__v_P_ = $(am__v_P_@AM_DEFAULT_V@)
+am__v_P_0 = false
+am__v_P_1 = :
+AM_V_GEN = $(am__v_GEN_@AM_V@)
+am__v_GEN_ = $(am__v_GEN_@AM_DEFAULT_V@)
+am__v_GEN_0 = @echo "  GEN     " $@;
+am__v_GEN_1 = 
+AM_V_at = $(am__v_at_@AM_V@)
+am__v_at_ = $(am__v_at_@AM_DEFAULT_V@)
+am__v_at_0 = @
+am__v_at_1 = 
+DEFAULT_INCLUDES = -I.@am__isrc@ -I$(top_builddir)
+depcomp = $(SHELL) $(top_srcdir)/depcomp
+am__depfiles_maybe = depfiles
+am__mv = mv -f
+COMPILE = $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) \
+	$(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS)
+AM_V_CC = $(am__v_CC_@AM_V@)
+am__v_CC_ = $(am__v_CC_@AM_DEFAULT_V@)
+am__v_CC_0 = @echo "  CC      " $@;
+am__v_CC_1 = 
+CCLD = $(CC)
+LINK = $(CCLD) $(AM_CFLAGS) $(CFLAGS) $(AM_LDFLAGS) $(LDFLAGS) -o $@
+AM_V_CCLD = $(am__v_CCLD_@AM_V@)
+am__v_CCLD_ = $(am__v_CCLD_@AM_DEFAULT_V@)
+am__v_CCLD_0 = @echo "  CCLD    " $@;
+am__v_CCLD_1 = 
+SOURCES = $(libjansson_a_SOURCES)
+DIST_SOURCES = $(libjansson_a_SOURCES)
+am__can_run_installinfo = \
+  case $$AM_UPDATE_INFO_DIR in \
+    n|no|NO) false;; \
+    *) (install-info --version) >/dev/null 2>&1;; \
+  esac
+am__tagged_files = $(HEADERS) $(SOURCES) $(TAGS_FILES) $(LISP)
+# Read a list of newline-separated strings from the standard input,
+# and print each of them once, without duplicates.  Input order is
+# *not* preserved.
+am__uniquify_input = $(AWK) '\
+  BEGIN { nonempty = 0; } \
+  { items[$$0] = 1; nonempty = 1; } \
+  END { if (nonempty) { for (i in items) print i; }; } \
+'
+# Make sure the list of sources is unique.  This is necessary because,
+# e.g., the same source file might be shared among _SOURCES variables
+# for different programs/libraries.
+am__define_uniq_tagged_files = \
+  list='$(am__tagged_files)'; \
+  unique=`for i in $$list; do \
+    if test -f "$$i"; then echo $$i; else echo $(srcdir)/$$i; fi; \
+  done | $(am__uniquify_input)`
+ETAGS = etags
+CTAGS = ctags
+DISTFILES = $(DIST_COMMON) $(DIST_SOURCES) $(TEXINFOS) $(EXTRA_DIST)
+ACLOCAL = @ACLOCAL@
+ALLOCA = @ALLOCA@
+AMTAR = @AMTAR@
+AM_DEFAULT_VERBOSITY = @AM_DEFAULT_VERBOSITY@
+AUTOCONF = @AUTOCONF@
+AUTOHEADER = @AUTOHEADER@
+AUTOMAKE = @AUTOMAKE@
+AWK = @AWK@
+CC = @CC@
+CCAS = @CCAS@
+CCASDEPMODE = @CCASDEPMODE@
+CCASFLAGS = @CCASFLAGS@
+CCDEPMODE = @CCDEPMODE@
+CFLAGS = @CFLAGS@
+CPP = @CPP@
+CPPFLAGS = @CPPFLAGS@
+CUDA_CFLAGS = @CUDA_CFLAGS@
+CUDA_LDFLAGS = @CUDA_LDFLAGS@
+CUDA_LIBS = @CUDA_LIBS@
+CXX = @CXX@
+CXXDEPMODE = @CXXDEPMODE@
+CXXFLAGS = @CXXFLAGS@
+CYGPATH_W = @CYGPATH_W@
+DEFS = @DEFS@
+DEPDIR = @DEPDIR@
+ECHO_C = @ECHO_C@
+ECHO_N = @ECHO_N@
+ECHO_T = @ECHO_T@
+EGREP = @EGREP@
+EXEEXT = @EXEEXT@
+GREP = @GREP@
+INSTALL = @INSTALL@
+INSTALL_DATA = @INSTALL_DATA@
+INSTALL_PROGRAM = @INSTALL_PROGRAM@
+INSTALL_SCRIPT = @INSTALL_SCRIPT@
+INSTALL_STRIP_PROGRAM = @INSTALL_STRIP_PROGRAM@
+JANSSON_LIBS = @JANSSON_LIBS@
+LDFLAGS = @LDFLAGS@
+LIBCURL = @LIBCURL@
+LIBCURL_CPPFLAGS = @LIBCURL_CPPFLAGS@
+LIBOBJS = @LIBOBJS@
+LIBS = @LIBS@
+LTLIBOBJS = @LTLIBOBJS@
+MAINT = @MAINT@
+MAKEINFO = @MAKEINFO@
+MKDIR_P = @MKDIR_P@
+NVCC = @NVCC@
+OBJEXT = @OBJEXT@
+OPENMP_CFLAGS = @OPENMP_CFLAGS@
+PACKAGE = @PACKAGE@
+PACKAGE_BUGREPORT = @PACKAGE_BUGREPORT@
+PACKAGE_NAME = @PACKAGE_NAME@
+PACKAGE_STRING = @PACKAGE_STRING@
+PACKAGE_TARNAME = @PACKAGE_TARNAME@
+PACKAGE_URL = @PACKAGE_URL@
+PACKAGE_VERSION = @PACKAGE_VERSION@
+PATH_SEPARATOR = @PATH_SEPARATOR@
+PTHREAD_FLAGS = @PTHREAD_FLAGS@
+PTHREAD_LIBS = @PTHREAD_LIBS@
+RANLIB = @RANLIB@
+SET_MAKE = @SET_MAKE@
+SHELL = @SHELL@
+STRIP = @STRIP@
+VERSION = @VERSION@
+WS2_LIBS = @WS2_LIBS@
+_libcurl_config = @_libcurl_config@
+abs_builddir = @abs_builddir@
+abs_srcdir = @abs_srcdir@
+abs_top_builddir = @abs_top_builddir@
+abs_top_srcdir = @abs_top_srcdir@
+ac_ct_CC = @ac_ct_CC@
+ac_ct_CXX = @ac_ct_CXX@
+am__include = @am__include@
+am__leading_dot = @am__leading_dot@
+am__quote = @am__quote@
+am__tar = @am__tar@
+am__untar = @am__untar@
+bindir = @bindir@
+build = @build@
+build_alias = @build_alias@
+build_cpu = @build_cpu@
+build_os = @build_os@
+build_vendor = @build_vendor@
+builddir = @builddir@
+datadir = @datadir@
+datarootdir = @datarootdir@
+docdir = @docdir@
+dvidir = @dvidir@
+exec_prefix = @exec_prefix@
+host = @host@
+host_alias = @host_alias@
+host_cpu = @host_cpu@
+host_os = @host_os@
+host_vendor = @host_vendor@
+htmldir = @htmldir@
+includedir = @includedir@
+infodir = @infodir@
+install_sh = @install_sh@
+libdir = @libdir@
+libexecdir = @libexecdir@
+localedir = @localedir@
+localstatedir = @localstatedir@
+mandir = @mandir@
+mkdir_p = @mkdir_p@
+oldincludedir = @oldincludedir@
+pdfdir = @pdfdir@
+prefix = @prefix@
+program_transform_name = @program_transform_name@
+psdir = @psdir@
+sbindir = @sbindir@
+sharedstatedir = @sharedstatedir@
+srcdir = @srcdir@
+sysconfdir = @sysconfdir@
+target = @target@
+target_alias = @target_alias@
+target_cpu = @target_cpu@
+target_os = @target_os@
+target_vendor = @target_vendor@
+top_build_prefix = @top_build_prefix@
+top_builddir = @top_builddir@
+top_srcdir = @top_srcdir@
+noinst_LIBRARIES = libjansson.a
+libjansson_a_SOURCES = \
+			  config.h		\
+			  dump.c		\
+			  hashtable.c		\
+			  hashtable.h		\
+			  jansson.h		\
+			  jansson_private.h	\
+			  load.c		\
+			  strbuffer.c		\
+			  strbuffer.h		\
+			  utf.c			\
+			  utf.h			\
+			  util.h		\
+			  value.c
+
+all: all-am
+
+.SUFFIXES:
+.SUFFIXES: .c .o .obj
+$(srcdir)/Makefile.in: @MAINTAINER_MODE_TRUE@ $(srcdir)/Makefile.am  $(am__configure_deps)
+	@for dep in $?; do \
+	  case '$(am__configure_deps)' in \
+	    *$$dep*) \
+	      ( cd $(top_builddir) && $(MAKE) $(AM_MAKEFLAGS) am--refresh ) \
+	        && { if test -f $@; then exit 0; else break; fi; }; \
+	      exit 1;; \
+	  esac; \
+	done; \
+	echo ' cd $(top_srcdir) && $(AUTOMAKE) --gnu compat/jansson/Makefile'; \
+	$(am__cd) $(top_srcdir) && \
+	  $(AUTOMAKE) --gnu compat/jansson/Makefile
+.PRECIOUS: Makefile
+Makefile: $(srcdir)/Makefile.in $(top_builddir)/config.status
+	@case '$?' in \
+	  *config.status*) \
+	    cd $(top_builddir) && $(MAKE) $(AM_MAKEFLAGS) am--refresh;; \
+	  *) \
+	    echo ' cd $(top_builddir) && $(SHELL) ./config.status $(subdir)/$@ $(am__depfiles_maybe)'; \
+	    cd $(top_builddir) && $(SHELL) ./config.status $(subdir)/$@ $(am__depfiles_maybe);; \
+	esac;
+
+$(top_builddir)/config.status: $(top_srcdir)/configure $(CONFIG_STATUS_DEPENDENCIES)
+	cd $(top_builddir) && $(MAKE) $(AM_MAKEFLAGS) am--refresh
+
+$(top_srcdir)/configure: @MAINTAINER_MODE_TRUE@ $(am__configure_deps)
+	cd $(top_builddir) && $(MAKE) $(AM_MAKEFLAGS) am--refresh
+$(ACLOCAL_M4): @MAINTAINER_MODE_TRUE@ $(am__aclocal_m4_deps)
+	cd $(top_builddir) && $(MAKE) $(AM_MAKEFLAGS) am--refresh
+$(am__aclocal_m4_deps):
+
+clean-noinstLIBRARIES:
+	-test -z "$(noinst_LIBRARIES)" || rm -f $(noinst_LIBRARIES)
+
+libjansson.a: $(libjansson_a_OBJECTS) $(libjansson_a_DEPENDENCIES) $(EXTRA_libjansson_a_DEPENDENCIES) 
+	$(AM_V_at)-rm -f libjansson.a
+	$(AM_V_AR)$(libjansson_a_AR) libjansson.a $(libjansson_a_OBJECTS) $(libjansson_a_LIBADD)
+	$(AM_V_at)$(RANLIB) libjansson.a
+
+mostlyclean-compile:
+	-rm -f *.$(OBJEXT)
+
+distclean-compile:
+	-rm -f *.tab.c
+
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/dump.Po@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/hashtable.Po@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/load.Po@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/strbuffer.Po@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/utf.Po@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/value.Po@am__quote@
+
+.c.o:
+@am__fastdepCC_TRUE@	$(AM_V_CC)$(COMPILE) -MT $@ -MD -MP -MF $(DEPDIR)/$*.Tpo -c -o $@ $<
+@am__fastdepCC_TRUE@	$(AM_V_at)$(am__mv) $(DEPDIR)/$*.Tpo $(DEPDIR)/$*.Po
+@AMDEP_TRUE@@am__fastdepCC_FALSE@	$(AM_V_CC)source='$<' object='$@' libtool=no @AMDEPBACKSLASH@
+@AMDEP_TRUE@@am__fastdepCC_FALSE@	DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
+@am__fastdepCC_FALSE@	$(AM_V_CC@am__nodep@)$(COMPILE) -c $<
+
+.c.obj:
+@am__fastdepCC_TRUE@	$(AM_V_CC)$(COMPILE) -MT $@ -MD -MP -MF $(DEPDIR)/$*.Tpo -c -o $@ `$(CYGPATH_W) '$<'`
+@am__fastdepCC_TRUE@	$(AM_V_at)$(am__mv) $(DEPDIR)/$*.Tpo $(DEPDIR)/$*.Po
+@AMDEP_TRUE@@am__fastdepCC_FALSE@	$(AM_V_CC)source='$<' object='$@' libtool=no @AMDEPBACKSLASH@
+@AMDEP_TRUE@@am__fastdepCC_FALSE@	DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
+@am__fastdepCC_FALSE@	$(AM_V_CC@am__nodep@)$(COMPILE) -c `$(CYGPATH_W) '$<'`
+
+ID: $(am__tagged_files)
+	$(am__define_uniq_tagged_files); mkid -fID $$unique
+tags: tags-am
+TAGS: tags
+
+tags-am: $(TAGS_DEPENDENCIES) $(am__tagged_files)
+	set x; \
+	here=`pwd`; \
+	$(am__define_uniq_tagged_files); \
+	shift; \
+	if test -z "$(ETAGS_ARGS)$$*$$unique"; then :; else \
+	  test -n "$$unique" || unique=$$empty_fix; \
+	  if test $$# -gt 0; then \
+	    $(ETAGS) $(ETAGSFLAGS) $(AM_ETAGSFLAGS) $(ETAGS_ARGS) \
+	      "$$@" $$unique; \
+	  else \
+	    $(ETAGS) $(ETAGSFLAGS) $(AM_ETAGSFLAGS) $(ETAGS_ARGS) \
+	      $$unique; \
+	  fi; \
+	fi
+ctags: ctags-am
+
+CTAGS: ctags
+ctags-am: $(TAGS_DEPENDENCIES) $(am__tagged_files)
+	$(am__define_uniq_tagged_files); \
+	test -z "$(CTAGS_ARGS)$$unique" \
+	  || $(CTAGS) $(CTAGSFLAGS) $(AM_CTAGSFLAGS) $(CTAGS_ARGS) \
+	     $$unique
+
+GTAGS:
+	here=`$(am__cd) $(top_builddir) && pwd` \
+	  && $(am__cd) $(top_srcdir) \
+	  && gtags -i $(GTAGS_ARGS) "$$here"
+cscopelist: cscopelist-am
+
+cscopelist-am: $(am__tagged_files)
+	list='$(am__tagged_files)'; \
+	case "$(srcdir)" in \
+	  [\\/]* | ?:[\\/]*) sdir="$(srcdir)" ;; \
+	  *) sdir=$(subdir)/$(srcdir) ;; \
+	esac; \
+	for i in $$list; do \
+	  if test -f "$$i"; then \
+	    echo "$(subdir)/$$i"; \
+	  else \
+	    echo "$$sdir/$$i"; \
+	  fi; \
+	done >> $(top_builddir)/cscope.files
+
+distclean-tags:
+	-rm -f TAGS ID GTAGS GRTAGS GSYMS GPATH tags
+
+distdir: $(DISTFILES)
+	@srcdirstrip=`echo "$(srcdir)" | sed 's/[].[^$$\\*]/\\\\&/g'`; \
+	topsrcdirstrip=`echo "$(top_srcdir)" | sed 's/[].[^$$\\*]/\\\\&/g'`; \
+	list='$(DISTFILES)'; \
+	  dist_files=`for file in $$list; do echo $$file; done | \
+	  sed -e "s|^$$srcdirstrip/||;t" \
+	      -e "s|^$$topsrcdirstrip/|$(top_builddir)/|;t"`; \
+	case $$dist_files in \
+	  */*) $(MKDIR_P) `echo "$$dist_files" | \
+			   sed '/\//!d;s|^|$(distdir)/|;s,/[^/]*$$,,' | \
+			   sort -u` ;; \
+	esac; \
+	for file in $$dist_files; do \
+	  if test -f $$file || test -d $$file; then d=.; else d=$(srcdir); fi; \
+	  if test -d $$d/$$file; then \
+	    dir=`echo "/$$file" | sed -e 's,/[^/]*$$,,'`; \
+	    if test -d "$(distdir)/$$file"; then \
+	      find "$(distdir)/$$file" -type d ! -perm -700 -exec chmod u+rwx {} \;; \
+	    fi; \
+	    if test -d $(srcdir)/$$file && test $$d != $(srcdir); then \
+	      cp -fpR $(srcdir)/$$file "$(distdir)$$dir" || exit 1; \
+	      find "$(distdir)/$$file" -type d ! -perm -700 -exec chmod u+rwx {} \;; \
+	    fi; \
+	    cp -fpR $$d/$$file "$(distdir)$$dir" || exit 1; \
+	  else \
+	    test -f "$(distdir)/$$file" \
+	    || cp -p $$d/$$file "$(distdir)/$$file" \
+	    || exit 1; \
+	  fi; \
+	done
+check-am: all-am
+check: check-am
+all-am: Makefile $(LIBRARIES)
+installdirs:
+install: install-am
+install-exec: install-exec-am
+install-data: install-data-am
+uninstall: uninstall-am
+
+install-am: all-am
+	@$(MAKE) $(AM_MAKEFLAGS) install-exec-am install-data-am
+
+installcheck: installcheck-am
+install-strip:
+	if test -z '$(STRIP)'; then \
+	  $(MAKE) $(AM_MAKEFLAGS) INSTALL_PROGRAM="$(INSTALL_STRIP_PROGRAM)" \
+	    install_sh_PROGRAM="$(INSTALL_STRIP_PROGRAM)" INSTALL_STRIP_FLAG=-s \
+	      install; \
+	else \
+	  $(MAKE) $(AM_MAKEFLAGS) INSTALL_PROGRAM="$(INSTALL_STRIP_PROGRAM)" \
+	    install_sh_PROGRAM="$(INSTALL_STRIP_PROGRAM)" INSTALL_STRIP_FLAG=-s \
+	    "INSTALL_PROGRAM_ENV=STRIPPROG='$(STRIP)'" install; \
+	fi
+mostlyclean-generic:
+
+clean-generic:
+
+distclean-generic:
+	-test -z "$(CONFIG_CLEAN_FILES)" || rm -f $(CONFIG_CLEAN_FILES)
+	-test . = "$(srcdir)" || test -z "$(CONFIG_CLEAN_VPATH_FILES)" || rm -f $(CONFIG_CLEAN_VPATH_FILES)
+
+maintainer-clean-generic:
+	@echo "This command is intended for maintainers to use"
+	@echo "it deletes files that may require special tools to rebuild."
+clean: clean-am
+
+clean-am: clean-generic clean-noinstLIBRARIES mostlyclean-am
+
+distclean: distclean-am
+	-rm -rf ./$(DEPDIR)
+	-rm -f Makefile
+distclean-am: clean-am distclean-compile distclean-generic \
+	distclean-tags
+
+dvi: dvi-am
+
+dvi-am:
+
+html: html-am
+
+html-am:
+
+info: info-am
+
+info-am:
+
+install-data-am:
+
+install-dvi: install-dvi-am
+
+install-dvi-am:
+
+install-exec-am:
+
+install-html: install-html-am
+
+install-html-am:
+
+install-info: install-info-am
+
+install-info-am:
+
+install-man:
+
+install-pdf: install-pdf-am
+
+install-pdf-am:
+
+install-ps: install-ps-am
+
+install-ps-am:
+
+installcheck-am:
+
+maintainer-clean: maintainer-clean-am
+	-rm -rf ./$(DEPDIR)
+	-rm -f Makefile
+maintainer-clean-am: distclean-am maintainer-clean-generic
+
+mostlyclean: mostlyclean-am
+
+mostlyclean-am: mostlyclean-compile mostlyclean-generic
+
+pdf: pdf-am
+
+pdf-am:
+
+ps: ps-am
+
+ps-am:
+
+uninstall-am:
+
+.MAKE: install-am install-strip
+
+.PHONY: CTAGS GTAGS TAGS all all-am check check-am clean clean-generic \
+	clean-noinstLIBRARIES cscopelist-am ctags ctags-am distclean \
+	distclean-compile distclean-generic distclean-tags distdir dvi \
+	dvi-am html html-am info info-am install install-am \
+	install-data install-data-am install-dvi install-dvi-am \
+	install-exec install-exec-am install-html install-html-am \
+	install-info install-info-am install-man install-pdf \
+	install-pdf-am install-ps install-ps-am install-strip \
+	installcheck installcheck-am installdirs maintainer-clean \
+	maintainer-clean-generic mostlyclean mostlyclean-compile \
+	mostlyclean-generic pdf pdf-am ps ps-am tags tags-am uninstall \
+	uninstall-am
+
+
+# Tell versions [3.59,3.63) of GNU make to not export all variables.
+# Otherwise a system limit (for SysV at least) may be exceeded.
+.NOEXPORT:
diff --git a/compat/jansson/config.h b/compat/jansson/config.h
new file mode 100644
index 0000000000..f11075a3c5
--- /dev/null
+++ b/compat/jansson/config.h
@@ -0,0 +1,73 @@
+/* config.h.  Generated from config.h.in by configure.  */
+/* config.h.in.  Generated from configure.ac by autoheader.  */
+
+/* Define to 1 if you have the <dlfcn.h> header file. */
+#define HAVE_DLFCN_H 1
+
+/* Define to 1 if you have the <inttypes.h> header file. */
+#define HAVE_INTTYPES_H 1
+
+/* Define to 1 if you have the <memory.h> header file. */
+#define HAVE_MEMORY_H 1
+
+/* Define to 1 if you have the <stdint.h> header file. */
+#define HAVE_STDINT_H 1
+
+/* Define to 1 if you have the <stdlib.h> header file. */
+#define HAVE_STDLIB_H 1
+
+/* Define to 1 if you have the <strings.h> header file. */
+#define HAVE_STRINGS_H 1
+
+/* Define to 1 if you have the <string.h> header file. */
+#define HAVE_STRING_H 1
+
+/* Define to 1 if you have the <sys/stat.h> header file. */
+#define HAVE_SYS_STAT_H 1
+
+/* Define to 1 if you have the <sys/types.h> header file. */
+#define HAVE_SYS_TYPES_H 1
+
+/* Define to 1 if you have the <unistd.h> header file. */
+#define HAVE_UNISTD_H 1
+
+/* Define to the sub-directory in which libtool stores uninstalled libraries.
+   */
+#define LT_OBJDIR ".libs/"
+
+/* Name of package */
+#define PACKAGE "jansson"
+
+/* Define to the address where bug reports for this package should be sent. */
+#define PACKAGE_BUGREPORT "petri@digip.org"
+
+/* Define to the full name of this package. */
+#define PACKAGE_NAME "jansson"
+
+/* Define to the full name and version of this package. */
+#define PACKAGE_STRING "jansson 1.3"
+
+/* Define to the one symbol short name of this package. */
+#define PACKAGE_TARNAME "jansson"
+
+/* Define to the home page for this package. */
+#define PACKAGE_URL ""
+
+/* Define to the version of this package. */
+#define PACKAGE_VERSION "1.3"
+
+/* Define to 1 if you have the ANSI C header files. */
+#define STDC_HEADERS 1
+
+/* Version number of package */
+#define VERSION "1.3"
+
+/* Define to `__inline__' or `__inline' if that's what the C compiler
+   calls it, or to nothing if 'inline' is not supported under any name.  */
+#ifndef __cplusplus
+/* #undef inline */
+#endif
+
+/* Define to the type of a signed integer type of width exactly 32 bits if
+   such a type exists and the standard includes do not define it. */
+/* #undef int32_t */
diff --git a/compat/jansson/dump.c b/compat/jansson/dump.c
new file mode 100644
index 0000000000..a8c9cc646a
--- /dev/null
+++ b/compat/jansson/dump.c
@@ -0,0 +1,460 @@
+/*
+ * Copyright (c) 2009, 2010 Petri Lehtinen <petri@digip.org>
+ *
+ * Jansson is free software; you can redistribute it and/or modify
+ * it under the terms of the MIT license. See LICENSE for details.
+ */
+
+#define _GNU_SOURCE
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <assert.h>
+
+#include <jansson.h>
+#include "jansson_private.h"
+#include "strbuffer.h"
+#include "utf.h"
+
+#define MAX_INTEGER_STR_LENGTH  100
+#define MAX_REAL_STR_LENGTH     100
+
+typedef int (*dump_func)(const char *buffer, int size, void *data);
+
+struct string
+{
+    char *buffer;
+    int length;
+    int size;
+};
+
+static int dump_to_strbuffer(const char *buffer, int size, void *data)
+{
+    return strbuffer_append_bytes((strbuffer_t *)data, buffer, size);
+}
+
+static int dump_to_file(const char *buffer, int size, void *data)
+{
+    FILE *dest = (FILE *)data;
+    if(fwrite(buffer, size, 1, dest) != 1)
+        return -1;
+    return 0;
+}
+
+/* 256 spaces (the maximum indentation size) */
+static char whitespace[] = "                                                                                                                                                                                                                                                                ";
+
+static int dump_indent(unsigned long flags, int depth, int space, dump_func dump, void *data)
+{
+    if(JSON_INDENT(flags) > 0)
+    {
+        int i, ws_count = JSON_INDENT(flags);
+
+        if(dump("\n", 1, data))
+            return -1;
+
+        for(i = 0; i < depth; i++)
+        {
+            if(dump(whitespace, ws_count, data))
+                return -1;
+        }
+    }
+    else if(space && !(flags & JSON_COMPACT))
+    {
+        return dump(" ", 1, data);
+    }
+    return 0;
+}
+
+static int dump_string(const char *str, int ascii, dump_func dump, void *data)
+{
+    const char *pos, *end;
+    int32_t codepoint;
+
+    if(dump("\"", 1, data))
+        return -1;
+
+    end = pos = str;
+    while(1)
+    {
+        const char *text;
+        char seq[13];
+        int length;
+
+        while(*end)
+        {
+            end = utf8_iterate(pos, &codepoint);
+            if(!end)
+                return -1;
+
+            /* mandatory escape or control char */
+            if(codepoint == '\\' || codepoint == '"' || codepoint < 0x20)
+                break;
+
+            /* non-ASCII */
+            if(ascii && codepoint > 0x7F)
+                break;
+
+            pos = end;
+        }
+
+        if(pos != str) {
+            if(dump(str, pos - str, data))
+                return -1;
+        }
+
+        if(end == pos)
+            break;
+
+        /* handle \, ", and control codes */
+        length = 2;
+        switch(codepoint)
+        {
+            case '\\': text = "\\\\"; break;
+            case '\"': text = "\\\""; break;
+            case '\b': text = "\\b"; break;
+            case '\f': text = "\\f"; break;
+            case '\n': text = "\\n"; break;
+            case '\r': text = "\\r"; break;
+            case '\t': text = "\\t"; break;
+            default:
+            {
+                /* codepoint is in BMP */
+                if(codepoint < 0x10000)
+                {
+                    sprintf(seq, "\\u%04x", codepoint);
+                    length = 6;
+                }
+
+                /* not in BMP -> construct a UTF-16 surrogate pair */
+                else
+                {
+                    int32_t first, last;
+
+                    codepoint -= 0x10000;
+                    first = 0xD800 | ((codepoint & 0xffc00) >> 10);
+                    last = 0xDC00 | (codepoint & 0x003ff);
+
+                    sprintf(seq, "\\u%04x\\u%04x", first, last);
+                    length = 12;
+                }
+
+                text = seq;
+                break;
+            }
+        }
+
+        if(dump(text, length, data))
+            return -1;
+
+        str = pos = end;
+    }
+
+    return dump("\"", 1, data);
+}
+
+static int object_key_compare_keys(const void *key1, const void *key2)
+{
+    return strcmp((*(const object_key_t **)key1)->key,
+                  (*(const object_key_t **)key2)->key);
+}
+
+static int object_key_compare_serials(const void *key1, const void *key2)
+{
+    return (*(const object_key_t **)key1)->serial -
+           (*(const object_key_t **)key2)->serial;
+}
+
+static int do_dump(const json_t *json, unsigned long flags, int depth,
+                   dump_func dump, void *data)
+{
+    int ascii = flags & JSON_ENSURE_ASCII ? 1 : 0;
+
+    switch(json_typeof(json)) {
+        case JSON_NULL:
+            return dump("null", 4, data);
+
+        case JSON_TRUE:
+            return dump("true", 4, data);
+
+        case JSON_FALSE:
+            return dump("false", 5, data);
+
+        case JSON_INTEGER:
+        {
+            char buffer[MAX_INTEGER_STR_LENGTH];
+            int size;
+
+            size = snprintf(buffer, MAX_INTEGER_STR_LENGTH, "%d", json_integer_value(json));
+            if(size >= MAX_INTEGER_STR_LENGTH)
+                return -1;
+
+            return dump(buffer, size, data);
+        }
+
+        case JSON_REAL:
+        {
+            char buffer[MAX_REAL_STR_LENGTH];
+            int size;
+
+            size = snprintf(buffer, MAX_REAL_STR_LENGTH, "%.17g",
+                            json_real_value(json));
+            if(size >= MAX_REAL_STR_LENGTH)
+                return -1;
+
+            /* Make sure there's a dot or 'e' in the output. Otherwise
+               a real is converted to an integer when decoding */
+            if(strchr(buffer, '.') == NULL &&
+               strchr(buffer, 'e') == NULL)
+            {
+                if(size + 2 >= MAX_REAL_STR_LENGTH) {
+                    /* No space to append ".0" */
+                    return -1;
+                }
+                buffer[size] = '.';
+                buffer[size + 1] = '0';
+                size += 2;
+            }
+
+            return dump(buffer, size, data);
+        }
+
+        case JSON_STRING:
+            return dump_string(json_string_value(json), ascii, dump, data);
+
+        case JSON_ARRAY:
+        {
+            int i;
+            int n;
+            json_array_t *array;
+
+            /* detect circular references */
+            array = json_to_array(json);
+            if(array->visited)
+                goto array_error;
+            array->visited = 1;
+
+            n = json_array_size(json);
+
+            if(dump("[", 1, data))
+                goto array_error;
+            if(n == 0) {
+                array->visited = 0;
+                return dump("]", 1, data);
+            }
+            if(dump_indent(flags, depth + 1, 0, dump, data))
+                goto array_error;
+
+            for(i = 0; i < n; ++i) {
+                if(do_dump(json_array_get(json, i), flags, depth + 1,
+                           dump, data))
+                    goto array_error;
+
+                if(i < n - 1)
+                {
+                    if(dump(",", 1, data) ||
+                       dump_indent(flags, depth + 1, 1, dump, data))
+                        goto array_error;
+                }
+                else
+                {
+                    if(dump_indent(flags, depth, 0, dump, data))
+                        goto array_error;
+                }
+            }
+
+            array->visited = 0;
+            return dump("]", 1, data);
+
+        array_error:
+            array->visited = 0;
+            return -1;
+        }
+
+        case JSON_OBJECT:
+        {
+            json_object_t *object;
+            void *iter;
+            const char *separator;
+            int separator_length;
+
+            if(flags & JSON_COMPACT) {
+                separator = ":";
+                separator_length = 1;
+            }
+            else {
+                separator = ": ";
+                separator_length = 2;
+            }
+
+            /* detect circular references */
+            object = json_to_object(json);
+            if(object->visited)
+                goto object_error;
+            object->visited = 1;
+
+            iter = json_object_iter((json_t *)json);
+
+            if(dump("{", 1, data))
+                goto object_error;
+            if(!iter) {
+                object->visited = 0;
+                return dump("}", 1, data);
+            }
+            if(dump_indent(flags, depth + 1, 0, dump, data))
+                goto object_error;
+
+            if(flags & JSON_SORT_KEYS || flags & JSON_PRESERVE_ORDER)
+            {
+                const object_key_t **keys;
+                unsigned int size;
+                unsigned int i;
+                int (*cmp_func)(const void *, const void *);
+
+                size = json_object_size(json);
+                keys = malloc(size * sizeof(object_key_t *));
+                if(!keys)
+                    goto object_error;
+
+                i = 0;
+                while(iter)
+                {
+                    keys[i] = jsonp_object_iter_fullkey(iter);
+                    iter = json_object_iter_next((json_t *)json, iter);
+                    i++;
+                }
+                assert(i == size);
+
+                if(flags & JSON_SORT_KEYS)
+                    cmp_func = object_key_compare_keys;
+                else
+                    cmp_func = object_key_compare_serials;
+
+                qsort((void*)keys, size, sizeof(object_key_t *), cmp_func);
+
+                for(i = 0; i < size; i++)
+                {
+                    const char *key;
+                    json_t *value;
+
+                    key = keys[i]->key;
+                    value = json_object_get(json, key);
+                    assert(value);
+
+                    dump_string(key, ascii, dump, data);
+                    if(dump(separator, separator_length, data) ||
+                       do_dump(value, flags, depth + 1, dump, data))
+                    {
+                        free((void*)keys);
+                        goto object_error;
+                    }
+
+                    if(i < size - 1)
+                    {
+                        if(dump(",", 1, data) ||
+                           dump_indent(flags, depth + 1, 1, dump, data))
+                        {
+                            free((void*)keys);
+                            goto object_error;
+                        }
+                    }
+                    else
+                    {
+                        if(dump_indent(flags, depth, 0, dump, data))
+                        {
+                            free((void*)keys);
+                            goto object_error;
+                        }
+                    }
+                }
+
+                free((void*)keys);
+            }
+            else
+            {
+                /* Don't sort keys */
+
+                while(iter)
+                {
+                    void *next = json_object_iter_next((json_t *)json, iter);
+
+                    dump_string(json_object_iter_key(iter), ascii, dump, data);
+                    if(dump(separator, separator_length, data) ||
+                       do_dump(json_object_iter_value(iter), flags, depth + 1,
+                               dump, data))
+                        goto object_error;
+
+                    if(next)
+                    {
+                        if(dump(",", 1, data) ||
+                           dump_indent(flags, depth + 1, 1, dump, data))
+                            goto object_error;
+                    }
+                    else
+                    {
+                        if(dump_indent(flags, depth, 0, dump, data))
+                            goto object_error;
+                    }
+
+                    iter = next;
+                }
+            }
+
+            object->visited = 0;
+            return dump("}", 1, data);
+
+        object_error:
+            object->visited = 0;
+            return -1;
+        }
+
+        default:
+            /* not reached */
+            return -1;
+    }
+}
+
+
+char *json_dumps(const json_t *json, unsigned long flags)
+{
+    strbuffer_t strbuff;
+    char *result;
+
+    if(!json_is_array(json) && !json_is_object(json))
+        return NULL;
+
+    if(strbuffer_init(&strbuff))
+        return NULL;
+
+    if(do_dump(json, flags, 0, dump_to_strbuffer, (void *)&strbuff)) {
+        strbuffer_close(&strbuff);
+        return NULL;
+    }
+
+    result = strdup(strbuffer_value(&strbuff));
+    strbuffer_close(&strbuff);
+
+    return result;
+}
+
+int json_dumpf(const json_t *json, FILE *output, unsigned long flags)
+{
+    if(!json_is_array(json) && !json_is_object(json))
+        return -1;
+
+    return do_dump(json, flags, 0, dump_to_file, (void *)output);
+}
+
+int json_dump_file(const json_t *json, const char *path, unsigned long flags)
+{
+    int result;
+
+    FILE *output = fopen(path, "w");
+    if(!output)
+        return -1;
+
+    result = json_dumpf(json, output, flags);
+
+    fclose(output);
+    return result;
+}
diff --git a/compat/jansson/hashtable.c b/compat/jansson/hashtable.c
new file mode 100644
index 0000000000..791f9acbe2
--- /dev/null
+++ b/compat/jansson/hashtable.c
@@ -0,0 +1,379 @@
+/*
+ * Copyright (c) 2009, 2010 Petri Lehtinen <petri@digip.org>
+ *
+ * This library is free software; you can redistribute it and/or modify
+ * it under the terms of the MIT license. See LICENSE for details.
+ */
+
+#include <config.h>
+
+#include <stdlib.h>
+#include "hashtable.h"
+
+#ifdef WIN32
+#define inline __inline
+#endif
+
+typedef struct hashtable_list list_t;
+typedef struct hashtable_pair pair_t;
+typedef struct hashtable_bucket bucket_t;
+
+#define container_of(ptr_, type_, member_)                      \
+    ((type_ *)((char *)ptr_ - (size_t)&((type_ *)0)->member_))
+
+#define list_to_pair(list_)  container_of(list_, pair_t, list)
+
+static inline void list_init(list_t *list)
+{
+    list->next = list;
+    list->prev = list;
+}
+
+static inline void list_insert(list_t *list, list_t *node)
+{
+    node->next = list;
+    node->prev = list->prev;
+    list->prev->next = node;
+    list->prev = node;
+}
+
+static inline void list_remove(list_t *list)
+{
+    list->prev->next = list->next;
+    list->next->prev = list->prev;
+}
+
+static inline int bucket_is_empty(hashtable_t *hashtable, bucket_t *bucket)
+{
+    return bucket->first == &hashtable->list && bucket->first == bucket->last;
+}
+
+static void insert_to_bucket(hashtable_t *hashtable, bucket_t *bucket,
+                             list_t *list)
+{
+    if(bucket_is_empty(hashtable, bucket))
+    {
+        list_insert(&hashtable->list, list);
+        bucket->first = bucket->last = list;
+    }
+    else
+    {
+        list_insert(bucket->first, list);
+        bucket->first = list;
+    }
+}
+
+static unsigned int primes[] = {
+    5, 13, 23, 53, 97, 193, 389, 769, 1543, 3079, 6151, 12289, 24593,
+    49157, 98317, 196613, 393241, 786433, 1572869, 3145739, 6291469,
+    12582917, 25165843, 50331653, 100663319, 201326611, 402653189,
+    805306457, 1610612741
+};
+static const unsigned int num_primes = sizeof(primes) / sizeof(unsigned int);
+
+static inline unsigned int num_buckets(hashtable_t *hashtable)
+{
+    return primes[hashtable->num_buckets];
+}
+
+
+static pair_t *hashtable_find_pair(hashtable_t *hashtable, bucket_t *bucket,
+                                   const void *key, unsigned int hash)
+{
+    list_t *list;
+    pair_t *pair;
+
+    if(bucket_is_empty(hashtable, bucket))
+        return NULL;
+
+    list = bucket->first;
+    while(1)
+    {
+        pair = list_to_pair(list);
+        if(pair->hash == hash && hashtable->cmp_keys(pair->key, key))
+            return pair;
+
+        if(list == bucket->last)
+            break;
+
+        list = list->next;
+    }
+
+    return NULL;
+}
+
+/* returns 0 on success, -1 if key was not found */
+static int hashtable_do_del(hashtable_t *hashtable,
+                            const void *key, unsigned int hash)
+{
+    pair_t *pair;
+    bucket_t *bucket;
+    unsigned int index;
+
+    index = hash % num_buckets(hashtable);
+    bucket = &hashtable->buckets[index];
+
+    pair = hashtable_find_pair(hashtable, bucket, key, hash);
+    if(!pair)
+        return -1;
+
+    if(&pair->list == bucket->first && &pair->list == bucket->last)
+        bucket->first = bucket->last = &hashtable->list;
+
+    else if(&pair->list == bucket->first)
+        bucket->first = pair->list.next;
+
+    else if(&pair->list == bucket->last)
+        bucket->last = pair->list.prev;
+
+    list_remove(&pair->list);
+
+    if(hashtable->free_key)
+        hashtable->free_key(pair->key);
+    if(hashtable->free_value)
+        hashtable->free_value(pair->value);
+
+    free(pair);
+    hashtable->size--;
+
+    return 0;
+}
+
+static void hashtable_do_clear(hashtable_t *hashtable)
+{
+    list_t *list, *next;
+    pair_t *pair;
+
+    for(list = hashtable->list.next; list != &hashtable->list; list = next)
+    {
+        next = list->next;
+        pair = list_to_pair(list);
+        if(hashtable->free_key)
+            hashtable->free_key(pair->key);
+        if(hashtable->free_value)
+            hashtable->free_value(pair->value);
+        free(pair);
+    }
+}
+
+static int hashtable_do_rehash(hashtable_t *hashtable)
+{
+    list_t *list, *next;
+    pair_t *pair;
+    unsigned int i, index, new_size;
+
+    free(hashtable->buckets);
+
+    hashtable->num_buckets++;
+    new_size = num_buckets(hashtable);
+
+    hashtable->buckets = malloc(new_size * sizeof(bucket_t));
+    if(!hashtable->buckets)
+        return -1;
+
+    for(i = 0; i < num_buckets(hashtable); i++)
+    {
+        hashtable->buckets[i].first = hashtable->buckets[i].last =
+            &hashtable->list;
+    }
+
+    list = hashtable->list.next;
+    list_init(&hashtable->list);
+
+    for(; list != &hashtable->list; list = next) {
+        next = list->next;
+        pair = list_to_pair(list);
+        index = pair->hash % new_size;
+        insert_to_bucket(hashtable, &hashtable->buckets[index], &pair->list);
+    }
+
+    return 0;
+}
+
+
+hashtable_t *hashtable_create(key_hash_fn hash_key, key_cmp_fn cmp_keys,
+                              free_fn free_key, free_fn free_value)
+{
+    hashtable_t *hashtable = malloc(sizeof(hashtable_t));
+    if(!hashtable)
+        return NULL;
+
+    if(hashtable_init(hashtable, hash_key, cmp_keys, free_key, free_value))
+    {
+        free(hashtable);
+        return NULL;
+    }
+
+    return hashtable;
+}
+
+void hashtable_destroy(hashtable_t *hashtable)
+{
+    hashtable_close(hashtable);
+    free(hashtable);
+}
+
+int hashtable_init(hashtable_t *hashtable,
+                   key_hash_fn hash_key, key_cmp_fn cmp_keys,
+                   free_fn free_key, free_fn free_value)
+{
+    unsigned int i;
+
+    hashtable->size = 0;
+    hashtable->num_buckets = 0;  /* index to primes[] */
+    hashtable->buckets = malloc(num_buckets(hashtable) * sizeof(bucket_t));
+    if(!hashtable->buckets)
+        return -1;
+
+    list_init(&hashtable->list);
+
+    hashtable->hash_key = hash_key;
+    hashtable->cmp_keys = cmp_keys;
+    hashtable->free_key = free_key;
+    hashtable->free_value = free_value;
+
+    for(i = 0; i < num_buckets(hashtable); i++)
+    {
+        hashtable->buckets[i].first = hashtable->buckets[i].last =
+            &hashtable->list;
+    }
+
+    return 0;
+}
+
+void hashtable_close(hashtable_t *hashtable)
+{
+    hashtable_do_clear(hashtable);
+    free(hashtable->buckets);
+}
+
+int hashtable_set(hashtable_t *hashtable, void *key, void *value)
+{
+    pair_t *pair;
+    bucket_t *bucket;
+    unsigned int hash, index;
+
+    /* rehash if the load ratio exceeds 1 */
+    if(hashtable->size >= num_buckets(hashtable))
+        if(hashtable_do_rehash(hashtable))
+            return -1;
+
+    hash = hashtable->hash_key(key);
+    index = hash % num_buckets(hashtable);
+    bucket = &hashtable->buckets[index];
+    pair = hashtable_find_pair(hashtable, bucket, key, hash);
+
+    if(pair)
+    {
+        if(hashtable->free_key)
+            hashtable->free_key(key);
+        if(hashtable->free_value)
+            hashtable->free_value(pair->value);
+        pair->value = value;
+    }
+    else
+    {
+        pair = malloc(sizeof(pair_t));
+        if(!pair)
+            return -1;
+
+        pair->key = key;
+        pair->value = value;
+        pair->hash = hash;
+        list_init(&pair->list);
+
+        insert_to_bucket(hashtable, bucket, &pair->list);
+
+        hashtable->size++;
+    }
+    return 0;
+}
+
+void *hashtable_get(hashtable_t *hashtable, const void *key)
+{
+    pair_t *pair;
+    unsigned int hash;
+    bucket_t *bucket;
+
+    hash = hashtable->hash_key(key);
+    bucket = &hashtable->buckets[hash % num_buckets(hashtable)];
+
+    pair = hashtable_find_pair(hashtable, bucket, key, hash);
+    if(!pair)
+        return NULL;
+
+    return pair->value;
+}
+
+int hashtable_del(hashtable_t *hashtable, const void *key)
+{
+    unsigned int hash = hashtable->hash_key(key);
+    return hashtable_do_del(hashtable, key, hash);
+}
+
+void hashtable_clear(hashtable_t *hashtable)
+{
+    unsigned int i;
+
+    hashtable_do_clear(hashtable);
+
+    for(i = 0; i < num_buckets(hashtable); i++)
+    {
+        hashtable->buckets[i].first = hashtable->buckets[i].last =
+            &hashtable->list;
+    }
+
+    list_init(&hashtable->list);
+    hashtable->size = 0;
+}
+
+void *hashtable_iter(hashtable_t *hashtable)
+{
+    return hashtable_iter_next(hashtable, &hashtable->list);
+}
+
+void *hashtable_iter_at(hashtable_t *hashtable, const void *key)
+{
+    pair_t *pair;
+    unsigned int hash;
+    bucket_t *bucket;
+
+    hash = hashtable->hash_key(key);
+    bucket = &hashtable->buckets[hash % num_buckets(hashtable)];
+
+    pair = hashtable_find_pair(hashtable, bucket, key, hash);
+    if(!pair)
+        return NULL;
+
+    return &pair->list;
+}
+
+void *hashtable_iter_next(hashtable_t *hashtable, void *iter)
+{
+    list_t *list = (list_t *)iter;
+    if(list->next == &hashtable->list)
+        return NULL;
+    return list->next;
+}
+
+void *hashtable_iter_key(void *iter)
+{
+    pair_t *pair = list_to_pair((list_t *)iter);
+    return pair->key;
+}
+
+void *hashtable_iter_value(void *iter)
+{
+    pair_t *pair = list_to_pair((list_t *)iter);
+    return pair->value;
+}
+
+void hashtable_iter_set(hashtable_t *hashtable, void *iter, void *value)
+{
+    pair_t *pair = list_to_pair((list_t *)iter);
+
+    if(hashtable->free_value)
+        hashtable->free_value(pair->value);
+
+    pair->value = value;
+}
diff --git a/compat/jansson/hashtable.h b/compat/jansson/hashtable.h
new file mode 100644
index 0000000000..52f8549281
--- /dev/null
+++ b/compat/jansson/hashtable.h
@@ -0,0 +1,207 @@
+/*
+ * Copyright (c) 2009, 2010 Petri Lehtinen <petri@digip.org>
+ *
+ * This library is free software; you can redistribute it and/or modify
+ * it under the terms of the MIT license. See LICENSE for details.
+ */
+
+#ifndef HASHTABLE_H
+#define HASHTABLE_H
+
+typedef unsigned int (*key_hash_fn)(const void *key);
+typedef int (*key_cmp_fn)(const void *key1, const void *key2);
+typedef void (*free_fn)(void *key);
+
+struct hashtable_list {
+    struct hashtable_list *prev;
+    struct hashtable_list *next;
+};
+
+struct hashtable_pair {
+    void *key;
+    void *value;
+    unsigned int hash;
+    struct hashtable_list list;
+};
+
+struct hashtable_bucket {
+    struct hashtable_list *first;
+    struct hashtable_list *last;
+};
+
+typedef struct hashtable {
+    unsigned int size;
+    struct hashtable_bucket *buckets;
+    unsigned int num_buckets;  /* index to primes[] */
+    struct hashtable_list list;
+
+    key_hash_fn hash_key;
+    key_cmp_fn cmp_keys;  /* returns non-zero for equal keys */
+    free_fn free_key;
+    free_fn free_value;
+} hashtable_t;
+
+/**
+ * hashtable_create - Create a hashtable object
+ *
+ * @hash_key: The key hashing function
+ * @cmp_keys: The key compare function. Returns non-zero for equal and
+ *     zero for unequal unequal keys
+ * @free_key: If non-NULL, called for a key that is no longer referenced.
+ * @free_value: If non-NULL, called for a value that is no longer referenced.
+ *
+ * Returns a new hashtable object that should be freed with
+ * hashtable_destroy when it's no longer used, or NULL on failure (out
+ * of memory).
+ */
+hashtable_t *hashtable_create(key_hash_fn hash_key, key_cmp_fn cmp_keys,
+                              free_fn free_key, free_fn free_value);
+
+/**
+ * hashtable_destroy - Destroy a hashtable object
+ *
+ * @hashtable: The hashtable
+ *
+ * Destroys a hashtable created with hashtable_create().
+ */
+void hashtable_destroy(hashtable_t *hashtable);
+
+/**
+ * hashtable_init - Initialize a hashtable object
+ *
+ * @hashtable: The (statically allocated) hashtable object
+ * @hash_key: The key hashing function
+ * @cmp_keys: The key compare function. Returns non-zero for equal and
+ *     zero for unequal unequal keys
+ * @free_key: If non-NULL, called for a key that is no longer referenced.
+ * @free_value: If non-NULL, called for a value that is no longer referenced.
+ *
+ * Initializes a statically allocated hashtable object. The object
+ * should be cleared with hashtable_close when it's no longer used.
+ *
+ * Returns 0 on success, -1 on error (out of memory).
+ */
+int hashtable_init(hashtable_t *hashtable,
+                   key_hash_fn hash_key, key_cmp_fn cmp_keys,
+                   free_fn free_key, free_fn free_value);
+
+/**
+ * hashtable_close - Release all resources used by a hashtable object
+ *
+ * @hashtable: The hashtable
+ *
+ * Destroys a statically allocated hashtable object.
+ */
+void hashtable_close(hashtable_t *hashtable);
+
+/**
+ * hashtable_set - Add/modify value in hashtable
+ *
+ * @hashtable: The hashtable object
+ * @key: The key
+ * @value: The value
+ *
+ * If a value with the given key already exists, its value is replaced
+ * with the new value.
+ *
+ * Key and value are "stealed" in the sense that hashtable frees them
+ * automatically when they are no longer used. The freeing is
+ * accomplished by calling free_key and free_value functions that were
+ * supplied to hashtable_new. In case one or both of the free
+ * functions is NULL, the corresponding item is not "stealed".
+ *
+ * Returns 0 on success, -1 on failure (out of memory).
+ */
+int hashtable_set(hashtable_t *hashtable, void *key, void *value);
+
+/**
+ * hashtable_get - Get a value associated with a key
+ *
+ * @hashtable: The hashtable object
+ * @key: The key
+ *
+ * Returns value if it is found, or NULL otherwise.
+ */
+void *hashtable_get(hashtable_t *hashtable, const void *key);
+
+/**
+ * hashtable_del - Remove a value from the hashtable
+ *
+ * @hashtable: The hashtable object
+ * @key: The key
+ *
+ * Returns 0 on success, or -1 if the key was not found.
+ */
+int hashtable_del(hashtable_t *hashtable, const void *key);
+
+/**
+ * hashtable_clear - Clear hashtable
+ *
+ * @hashtable: The hashtable object
+ *
+ * Removes all items from the hashtable.
+ */
+void hashtable_clear(hashtable_t *hashtable);
+
+/**
+ * hashtable_iter - Iterate over hashtable
+ *
+ * @hashtable: The hashtable object
+ *
+ * Returns an opaque iterator to the first element in the hashtable.
+ * The iterator should be passed to hashtable_iter_* functions.
+ * The hashtable items are not iterated over in any particular order.
+ *
+ * There's no need to free the iterator in any way. The iterator is
+ * valid as long as the item that is referenced by the iterator is not
+ * deleted. Other values may be added or deleted. In particular,
+ * hashtable_iter_next() may be called on an iterator, and after that
+ * the key/value pair pointed by the old iterator may be deleted.
+ */
+void *hashtable_iter(hashtable_t *hashtable);
+
+/**
+ * hashtable_iter_at - Return an iterator at a specific key
+ *
+ * @hashtable: The hashtable object
+ * @key: The key that the iterator should point to
+ *
+ * Like hashtable_iter() but returns an iterator pointing to a
+ * specific key.
+ */
+void *hashtable_iter_at(hashtable_t *hashtable, const void *key);
+
+/**
+ * hashtable_iter_next - Advance an iterator
+ *
+ * @hashtable: The hashtable object
+ * @iter: The iterator
+ *
+ * Returns a new iterator pointing to the next element in the
+ * hashtable or NULL if the whole hastable has been iterated over.
+ */
+void *hashtable_iter_next(hashtable_t *hashtable, void *iter);
+
+/**
+ * hashtable_iter_key - Retrieve the key pointed by an iterator
+ *
+ * @iter: The iterator
+ */
+void *hashtable_iter_key(void *iter);
+
+/**
+ * hashtable_iter_value - Retrieve the value pointed by an iterator
+ *
+ * @iter: The iterator
+ */
+void *hashtable_iter_value(void *iter);
+
+/**
+ * hashtable_iter_set - Set the value pointed by an iterator
+ *
+ * @iter: The iterator
+ * @value: The value to set
+ */
+void hashtable_iter_set(hashtable_t *hashtable, void *iter, void *value);
+
+#endif
diff --git a/compat/jansson/jansson.h b/compat/jansson/jansson.h
new file mode 100644
index 0000000000..781896ce9f
--- /dev/null
+++ b/compat/jansson/jansson.h
@@ -0,0 +1,200 @@
+/*
+ * Copyright (c) 2009, 2010 Petri Lehtinen <petri@digip.org>
+ *
+ * Jansson is free software; you can redistribute it and/or modify
+ * it under the terms of the MIT license. See LICENSE for details.
+ */
+
+#ifndef JANSSON_H
+#define JANSSON_H
+
+#include <stdio.h>
+
+#ifdef WIN32
+#define snprintf(...) _snprintf(__VA_ARGS__)
+#define strdup(x) _strdup(x)
+#endif
+
+#ifdef WIN32
+#define inline __inline
+#endif
+
+#ifndef __cplusplus
+#define JSON_INLINE inline
+#else
+#define JSON_INLINE inline
+extern "C" {
+#endif
+
+/* types */
+
+typedef enum {
+    JSON_OBJECT,
+    JSON_ARRAY,
+    JSON_STRING,
+    JSON_INTEGER,
+    JSON_REAL,
+    JSON_TRUE,
+    JSON_FALSE,
+    JSON_NULL
+} json_type;
+
+typedef struct {
+    json_type type;
+    unsigned long refcount;
+} json_t;
+
+#define json_typeof(json)      ((json)->type)
+#define json_is_object(json)   (json && json_typeof(json) == JSON_OBJECT)
+#define json_is_array(json)    (json && json_typeof(json) == JSON_ARRAY)
+#define json_is_string(json)   (json && json_typeof(json) == JSON_STRING)
+#define json_is_integer(json)  (json && json_typeof(json) == JSON_INTEGER)
+#define json_is_real(json)     (json && json_typeof(json) == JSON_REAL)
+#define json_is_number(json)   (json_is_integer(json) || json_is_real(json))
+#define json_is_true(json)     (json && json_typeof(json) == JSON_TRUE)
+#define json_is_false(json)    (json && json_typeof(json) == JSON_FALSE)
+#define json_is_boolean(json)  (json_is_true(json) || json_is_false(json))
+#define json_is_null(json)     (json && json_typeof(json) == JSON_NULL)
+
+/* construction, destruction, reference counting */
+
+json_t *json_object(void);
+json_t *json_array(void);
+json_t *json_string(const char *value);
+json_t *json_string_nocheck(const char *value);
+json_t *json_integer(int value);
+json_t *json_real(double value);
+json_t *json_true(void);
+json_t *json_false(void);
+json_t *json_null(void);
+
+static JSON_INLINE
+json_t *json_incref(json_t *json)
+{
+    if(json && json->refcount != (unsigned int)-1)
+        ++json->refcount;
+    return json;
+}
+
+/* do not call json_delete directly */
+void json_delete(json_t *json);
+
+static JSON_INLINE
+void json_decref(json_t *json)
+{
+    if(json && json->refcount != (unsigned int)-1 && --json->refcount == 0)
+        json_delete(json);
+}
+
+
+/* getters, setters, manipulation */
+
+unsigned int json_object_size(const json_t *object);
+json_t *json_object_get(const json_t *object, const char *key);
+int json_object_set_new(json_t *object, const char *key, json_t *value);
+int json_object_set_new_nocheck(json_t *object, const char *key, json_t *value);
+int json_object_del(json_t *object, const char *key);
+int json_object_clear(json_t *object);
+int json_object_update(json_t *object, json_t *other);
+void *json_object_iter(json_t *object);
+void *json_object_iter_at(json_t *object, const char *key);
+void *json_object_iter_next(json_t *object, void *iter);
+const char *json_object_iter_key(void *iter);
+json_t *json_object_iter_value(void *iter);
+int json_object_iter_set_new(json_t *object, void *iter, json_t *value);
+
+static JSON_INLINE
+int json_object_set(json_t *object, const char *key, json_t *value)
+{
+    return json_object_set_new(object, key, json_incref(value));
+}
+
+static JSON_INLINE
+int json_object_set_nocheck(json_t *object, const char *key, json_t *value)
+{
+    return json_object_set_new_nocheck(object, key, json_incref(value));
+}
+
+static inline
+int json_object_iter_set(json_t *object, void *iter, json_t *value)
+{
+    return json_object_iter_set_new(object, iter, json_incref(value));
+}
+
+unsigned int json_array_size(const json_t *array);
+json_t *json_array_get(const json_t *array, unsigned int index);
+int json_array_set_new(json_t *array, unsigned int index, json_t *value);
+int json_array_append_new(json_t *array, json_t *value);
+int json_array_insert_new(json_t *array, unsigned int index, json_t *value);
+int json_array_remove(json_t *array, unsigned int index);
+int json_array_clear(json_t *array);
+int json_array_extend(json_t *array, json_t *other);
+
+static JSON_INLINE
+int json_array_set(json_t *array, unsigned int index, json_t *value)
+{
+    return json_array_set_new(array, index, json_incref(value));
+}
+
+static JSON_INLINE
+int json_array_append(json_t *array, json_t *value)
+{
+    return json_array_append_new(array, json_incref(value));
+}
+
+static JSON_INLINE
+int json_array_insert(json_t *array, unsigned int index, json_t *value)
+{
+    return json_array_insert_new(array, index, json_incref(value));
+}
+
+const char *json_string_value(const json_t *string);
+int json_integer_value(const json_t *integer);
+double json_real_value(const json_t *real);
+double json_number_value(const json_t *json);
+
+int json_string_set(json_t *string, const char *value);
+int json_string_set_nocheck(json_t *string, const char *value);
+int json_integer_set(json_t *integer, int value);
+int json_real_set(json_t *real, double value);
+
+
+/* equality */
+
+int json_equal(json_t *value1, json_t *value2);
+
+
+/* copying */
+
+json_t *json_copy(json_t *value);
+json_t *json_deep_copy(json_t *value);
+
+
+/* loading, printing */
+
+#define JSON_ERROR_TEXT_LENGTH  160
+
+typedef struct {
+    char text[JSON_ERROR_TEXT_LENGTH];
+    int line;
+} json_error_t;
+
+json_t *json_loads(const char *input, json_error_t *error);
+json_t *json_loadf(FILE *input, json_error_t *error);
+json_t *json_load_file(const char *path, json_error_t *error);
+
+#define JSON_INDENT(n)      (n & 0xFF)
+#define JSON_COMPACT        0x100
+#define JSON_ENSURE_ASCII   0x200
+#define JSON_SORT_KEYS      0x400
+#define JSON_PRESERVE_ORDER 0x800
+
+char *json_dumps(const json_t *json, unsigned long flags);
+int json_dumpf(const json_t *json, FILE *output, unsigned long flags);
+int json_dump_file(const json_t *json, const char *path, unsigned long flags);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif
diff --git a/compat/jansson/jansson_private.h b/compat/jansson/jansson_private.h
new file mode 100644
index 0000000000..3a3ed75480
--- /dev/null
+++ b/compat/jansson/jansson_private.h
@@ -0,0 +1,60 @@
+/*
+ * Copyright (c) 2009, 2010 Petri Lehtinen <petri@digip.org>
+ *
+ * Jansson is free software; you can redistribute it and/or modify
+ * it under the terms of the MIT license. See LICENSE for details.
+ */
+
+#ifndef JANSSON_PRIVATE_H
+#define JANSSON_PRIVATE_H
+
+#include "jansson.h"
+#include "hashtable.h"
+
+#define container_of(ptr_, type_, member_)  \
+    ((type_ *)((char *)ptr_ - (size_t)&((type_ *)0)->member_))
+
+typedef struct {
+    json_t json;
+    hashtable_t hashtable;
+    unsigned long serial;
+    int visited;
+} json_object_t;
+
+typedef struct {
+    json_t json;
+    unsigned int size;
+    unsigned int entries;
+    json_t **table;
+    int visited;
+} json_array_t;
+
+typedef struct {
+    json_t json;
+    char *value;
+} json_string_t;
+
+typedef struct {
+    json_t json;
+    double value;
+} json_real_t;
+
+typedef struct {
+    json_t json;
+    int value;
+} json_integer_t;
+
+#define json_to_object(json_)  container_of(json_, json_object_t, json)
+#define json_to_array(json_)   container_of(json_, json_array_t, json)
+#define json_to_string(json_)  container_of(json_, json_string_t, json)
+#define json_to_real(json_)   container_of(json_, json_real_t, json)
+#define json_to_integer(json_) container_of(json_, json_integer_t, json)
+
+typedef struct {
+    unsigned long serial;
+    char key[];
+} object_key_t;
+
+const object_key_t *jsonp_object_iter_fullkey(void *iter);
+
+#endif
diff --git a/compat/jansson/load.c b/compat/jansson/load.c
new file mode 100644
index 0000000000..ee56fbe733
--- /dev/null
+++ b/compat/jansson/load.c
@@ -0,0 +1,879 @@
+/*
+ * Copyright (c) 2009, 2010 Petri Lehtinen <petri@digip.org>
+ *
+ * Jansson is free software; you can redistribute it and/or modify
+ * it under the terms of the MIT license. See LICENSE for details.
+ */
+
+#define _GNU_SOURCE
+#include <ctype.h>
+#include <errno.h>
+#include <limits.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <stdarg.h>
+#include <assert.h>
+
+#include <jansson.h>
+#include "jansson_private.h"
+#include "strbuffer.h"
+#include "utf.h"
+
+#define TOKEN_INVALID         -1
+#define TOKEN_EOF              0
+#define TOKEN_STRING         256
+#define TOKEN_INTEGER        257
+#define TOKEN_REAL           258
+#define TOKEN_TRUE           259
+#define TOKEN_FALSE          260
+#define TOKEN_NULL           261
+
+/* read one byte from stream, return EOF on end of file */
+typedef int (*get_func)(void *data);
+
+/* return non-zero if end of file has been reached */
+typedef int (*eof_func)(void *data);
+
+typedef struct {
+    get_func get;
+    eof_func eof;
+    void *data;
+    int stream_pos;
+    char buffer[5];
+    int buffer_pos;
+} stream_t;
+
+
+typedef struct {
+    stream_t stream;
+    strbuffer_t saved_text;
+    int token;
+    int line, column;
+    union {
+        char *string;
+        int integer;
+        double real;
+    } value;
+} lex_t;
+
+
+/*** error reporting ***/
+
+static void error_init(json_error_t *error)
+{
+    if(error)
+    {
+        error->text[0] = '\0';
+        error->line = -1;
+    }
+}
+
+static void error_set(json_error_t *error, const lex_t *lex,
+                      const char *msg, ...)
+{
+    va_list ap;
+    char text[JSON_ERROR_TEXT_LENGTH];
+
+    if(!error || error->text[0] != '\0') {
+        /* error already set */
+        return;
+    }
+
+    va_start(ap, msg);
+    vsnprintf(text, JSON_ERROR_TEXT_LENGTH, msg, ap);
+    va_end(ap);
+
+    if(lex)
+    {
+        const char *saved_text = strbuffer_value(&lex->saved_text);
+        error->line = lex->line;
+        if(saved_text && saved_text[0])
+        {
+            if(lex->saved_text.length <= 20) {
+                snprintf(error->text, JSON_ERROR_TEXT_LENGTH,
+                         "%s near '%s'", text, saved_text);
+            }
+            else
+                snprintf(error->text, JSON_ERROR_TEXT_LENGTH, "%s", text);
+        }
+        else
+        {
+            snprintf(error->text, JSON_ERROR_TEXT_LENGTH,
+                     "%s near end of file", text);
+        }
+    }
+    else
+    {
+        error->line = -1;
+        snprintf(error->text, JSON_ERROR_TEXT_LENGTH, "%s", text);
+    }
+}
+
+
+/*** lexical analyzer ***/
+
+static void
+stream_init(stream_t *stream, get_func get, eof_func eof, void *data)
+{
+    stream->get = get;
+    stream->eof = eof;
+    stream->data = data;
+    stream->stream_pos = 0;
+    stream->buffer[0] = '\0';
+    stream->buffer_pos = 0;
+}
+
+static char stream_get(stream_t *stream, json_error_t *error)
+{
+    char c;
+
+    if(!stream->buffer[stream->buffer_pos])
+    {
+        stream->buffer[0] = stream->get(stream->data);
+        stream->buffer_pos = 0;
+
+        c = stream->buffer[0];
+
+        if((unsigned char)c >= 0x80 && c != (char)EOF)
+        {
+            /* multi-byte UTF-8 sequence */
+            int i, count;
+
+            count = utf8_check_first(c);
+            if(!count)
+                goto out;
+
+            assert(count >= 2);
+
+            for(i = 1; i < count; i++)
+                stream->buffer[i] = stream->get(stream->data);
+
+            if(!utf8_check_full(stream->buffer, count, NULL))
+                goto out;
+
+            stream->stream_pos += count;
+            stream->buffer[count] = '\0';
+        }
+        else {
+            stream->buffer[1] = '\0';
+            stream->stream_pos++;
+        }
+    }
+
+    return stream->buffer[stream->buffer_pos++];
+
+out:
+    error_set(error, NULL, "unable to decode byte 0x%x at position %d",
+              (unsigned char)c, stream->stream_pos);
+
+    stream->buffer[0] = EOF;
+    stream->buffer[1] = '\0';
+    stream->buffer_pos = 1;
+
+    return EOF;
+}
+
+static void stream_unget(stream_t *stream, char c)
+{
+    assert(stream->buffer_pos > 0);
+    stream->buffer_pos--;
+    assert(stream->buffer[stream->buffer_pos] == c);
+}
+
+
+static int lex_get(lex_t *lex, json_error_t *error)
+{
+    return stream_get(&lex->stream, error);
+}
+
+static int lex_eof(lex_t *lex)
+{
+    return lex->stream.eof(lex->stream.data);
+}
+
+static void lex_save(lex_t *lex, char c)
+{
+    strbuffer_append_byte(&lex->saved_text, c);
+}
+
+static int lex_get_save(lex_t *lex, json_error_t *error)
+{
+    char c = stream_get(&lex->stream, error);
+    lex_save(lex, c);
+    return c;
+}
+
+static void lex_unget_unsave(lex_t *lex, char c)
+{
+    char d;
+    stream_unget(&lex->stream, c);
+    d = strbuffer_pop(&lex->saved_text);
+    assert(c == d);
+}
+
+static void lex_save_cached(lex_t *lex)
+{
+    while(lex->stream.buffer[lex->stream.buffer_pos] != '\0')
+    {
+        lex_save(lex, lex->stream.buffer[lex->stream.buffer_pos]);
+        lex->stream.buffer_pos++;
+    }
+}
+
+/* assumes that str points to 'u' plus at least 4 valid hex digits */
+static int32_t decode_unicode_escape(const char *str)
+{
+    int i;
+    int32_t value = 0;
+
+    assert(str[0] == 'u');
+
+    for(i = 1; i <= 4; i++) {
+        char c = str[i];
+        value <<= 4;
+        if(isdigit(c))
+            value += c - '0';
+        else if(islower(c))
+            value += c - 'a' + 10;
+        else if(isupper(c))
+            value += c - 'A' + 10;
+        else
+            assert(0);
+    }
+
+    return value;
+}
+
+static void lex_scan_string(lex_t *lex, json_error_t *error)
+{
+    char c;
+    const char *p;
+    char *t;
+    int i;
+
+    lex->value.string = NULL;
+    lex->token = TOKEN_INVALID;
+
+    c = lex_get_save(lex, error);
+
+    while(c != '"') {
+        if(c == (char)EOF) {
+            lex_unget_unsave(lex, c);
+            if(lex_eof(lex))
+                error_set(error, lex, "premature end of input");
+            goto out;
+        }
+
+        else if((unsigned char)c <= 0x1F) {
+            /* control character */
+            lex_unget_unsave(lex, c);
+            if(c == '\n')
+                error_set(error, lex, "unexpected newline", c);
+            else
+                error_set(error, lex, "control character 0x%x", c);
+            goto out;
+        }
+
+        else if(c == '\\') {
+            c = lex_get_save(lex, error);
+            if(c == 'u') {
+                c = lex_get_save(lex, error);
+                for(i = 0; i < 4; i++) {
+                    if(!isxdigit(c)) {
+                        lex_unget_unsave(lex, c);
+                        error_set(error, lex, "invalid escape");
+                        goto out;
+                    }
+                    c = lex_get_save(lex, error);
+                }
+            }
+            else if(c == '"' || c == '\\' || c == '/' || c == 'b' ||
+                    c == 'f' || c == 'n' || c == 'r' || c == 't')
+                c = lex_get_save(lex, error);
+            else {
+                lex_unget_unsave(lex, c);
+                error_set(error, lex, "invalid escape");
+                goto out;
+            }
+        }
+        else
+            c = lex_get_save(lex, error);
+    }
+
+    /* the actual value is at most of the same length as the source
+       string, because:
+         - shortcut escapes (e.g. "\t") (length 2) are converted to 1 byte
+         - a single \uXXXX escape (length 6) is converted to at most 3 bytes
+         - two \uXXXX escapes (length 12) forming an UTF-16 surrogate pair
+           are converted to 4 bytes
+    */
+    lex->value.string = malloc(lex->saved_text.length + 1);
+    if(!lex->value.string) {
+        /* this is not very nice, since TOKEN_INVALID is returned */
+        goto out;
+    }
+
+    /* the target */
+    t = lex->value.string;
+
+    /* + 1 to skip the " */
+    p = strbuffer_value(&lex->saved_text) + 1;
+
+    while(*p != '"') {
+        if(*p == '\\') {
+            p++;
+            if(*p == 'u') {
+                char buffer[4];
+                int length;
+                int32_t value;
+
+                value = decode_unicode_escape(p);
+                p += 5;
+
+                if(0xD800 <= value && value <= 0xDBFF) {
+                    /* surrogate pair */
+                    if(*p == '\\' && *(p + 1) == 'u') {
+                        int32_t value2 = decode_unicode_escape(++p);
+                        p += 5;
+
+                        if(0xDC00 <= value2 && value2 <= 0xDFFF) {
+                            /* valid second surrogate */
+                            value =
+                                ((value - 0xD800) << 10) +
+                                (value2 - 0xDC00) +
+                                0x10000;
+                        }
+                        else {
+                            /* invalid second surrogate */
+                            error_set(error, lex,
+                                      "invalid Unicode '\\u%04X\\u%04X'",
+                                      value, value2);
+                            goto out;
+                        }
+                    }
+                    else {
+                        /* no second surrogate */
+                        error_set(error, lex, "invalid Unicode '\\u%04X'",
+                                  value);
+                        goto out;
+                    }
+                }
+                else if(0xDC00 <= value && value <= 0xDFFF) {
+                    error_set(error, lex, "invalid Unicode '\\u%04X'", value);
+                    goto out;
+                }
+                else if(value == 0)
+                {
+                    error_set(error, lex, "\\u0000 is not allowed");
+                    goto out;
+                }
+
+                if(utf8_encode(value, buffer, &length))
+                    assert(0);
+
+                memcpy(t, buffer, length);
+                t += length;
+            }
+            else {
+                switch(*p) {
+                    case '"': case '\\': case '/':
+                        *t = *p; break;
+                    case 'b': *t = '\b'; break;
+                    case 'f': *t = '\f'; break;
+                    case 'n': *t = '\n'; break;
+                    case 'r': *t = '\r'; break;
+                    case 't': *t = '\t'; break;
+                    default: assert(0);
+                }
+                t++;
+                p++;
+            }
+        }
+        else
+            *(t++) = *(p++);
+    }
+    *t = '\0';
+    lex->token = TOKEN_STRING;
+    return;
+
+out:
+    free(lex->value.string);
+}
+
+static int lex_scan_number(lex_t *lex, char c, json_error_t *error)
+{
+    const char *saved_text;
+    char *end;
+    double value;
+
+    lex->token = TOKEN_INVALID;
+
+    if(c == '-')
+        c = lex_get_save(lex, error);
+
+    if(c == '0') {
+        c = lex_get_save(lex, error);
+        if(isdigit(c)) {
+            lex_unget_unsave(lex, c);
+            goto out;
+        }
+    }
+    else if(isdigit(c)) {
+        c = lex_get_save(lex, error);
+        while(isdigit(c))
+            c = lex_get_save(lex, error);
+    }
+    else {
+      lex_unget_unsave(lex, c);
+      goto out;
+    }
+
+    if(c != '.' && c != 'E' && c != 'e') {
+        long value;
+
+        lex_unget_unsave(lex, c);
+
+        saved_text = strbuffer_value(&lex->saved_text);
+        value = strtol(saved_text, &end, 10);
+        assert(end == saved_text + lex->saved_text.length);
+
+        if((value == LONG_MAX && errno == ERANGE) || value > INT_MAX) {
+            error_set(error, lex, "too big integer");
+            goto out;
+        }
+        else if((value == LONG_MIN && errno == ERANGE) || value < INT_MIN) {
+            error_set(error, lex, "too big negative integer");
+            goto out;
+        }
+
+        lex->token = TOKEN_INTEGER;
+        lex->value.integer = (int)value;
+        return 0;
+    }
+
+    if(c == '.') {
+        c = lex_get(lex, error);
+        if(!isdigit(c))
+            goto out;
+        lex_save(lex, c);
+
+        c = lex_get_save(lex, error);
+        while(isdigit(c))
+            c = lex_get_save(lex, error);
+    }
+
+    if(c == 'E' || c == 'e') {
+        c = lex_get_save(lex, error);
+        if(c == '+' || c == '-')
+            c = lex_get_save(lex, error);
+
+        if(!isdigit(c)) {
+            lex_unget_unsave(lex, c);
+            goto out;
+        }
+
+        c = lex_get_save(lex, error);
+        while(isdigit(c))
+            c = lex_get_save(lex, error);
+    }
+
+    lex_unget_unsave(lex, c);
+
+    saved_text = strbuffer_value(&lex->saved_text);
+    value = strtod(saved_text, &end);
+    assert(end == saved_text + lex->saved_text.length);
+
+    if(errno == ERANGE && value != 0) {
+        error_set(error, lex, "real number overflow");
+        goto out;
+    }
+
+    lex->token = TOKEN_REAL;
+    lex->value.real = value;
+    return 0;
+
+out:
+    return -1;
+}
+
+static int lex_scan(lex_t *lex, json_error_t *error)
+{
+    char c;
+
+    strbuffer_clear(&lex->saved_text);
+
+    if(lex->token == TOKEN_STRING) {
+        free(lex->value.string);
+        lex->value.string = NULL;
+    }
+
+    c = lex_get(lex, error);
+    while(c == ' ' || c == '\t' || c == '\n' || c == '\r')
+    {
+        if(c == '\n')
+            lex->line++;
+
+        c = lex_get(lex, error);
+    }
+
+    if(c == (char)EOF) {
+        if(lex_eof(lex))
+            lex->token = TOKEN_EOF;
+        else
+            lex->token = TOKEN_INVALID;
+        goto out;
+    }
+
+    lex_save(lex, c);
+
+    if(c == '{' || c == '}' || c == '[' || c == ']' || c == ':' || c == ',')
+        lex->token = c;
+
+    else if(c == '"')
+        lex_scan_string(lex, error);
+
+    else if(isdigit(c) || c == '-') {
+        if(lex_scan_number(lex, c, error))
+            goto out;
+    }
+
+    else if(isupper(c) || islower(c)) {
+        /* eat up the whole identifier for clearer error messages */
+        const char *saved_text;
+
+        c = lex_get_save(lex, error);
+        while(isupper(c) || islower(c))
+            c = lex_get_save(lex, error);
+        lex_unget_unsave(lex, c);
+
+        saved_text = strbuffer_value(&lex->saved_text);
+
+        if(strcmp(saved_text, "true") == 0)
+            lex->token = TOKEN_TRUE;
+        else if(strcmp(saved_text, "false") == 0)
+            lex->token = TOKEN_FALSE;
+        else if(strcmp(saved_text, "null") == 0)
+            lex->token = TOKEN_NULL;
+        else
+            lex->token = TOKEN_INVALID;
+    }
+
+    else {
+        /* save the rest of the input UTF-8 sequence to get an error
+           message of valid UTF-8 */
+        lex_save_cached(lex);
+        lex->token = TOKEN_INVALID;
+    }
+
+out:
+    return lex->token;
+}
+
+static char *lex_steal_string(lex_t *lex)
+{
+    char *result = NULL;
+    if(lex->token == TOKEN_STRING)
+    {
+        result = lex->value.string;
+        lex->value.string = NULL;
+    }
+    return result;
+}
+
+static int lex_init(lex_t *lex, get_func get, eof_func eof, void *data)
+{
+    stream_init(&lex->stream, get, eof, data);
+    if(strbuffer_init(&lex->saved_text))
+        return -1;
+
+    lex->token = TOKEN_INVALID;
+    lex->line = 1;
+
+    return 0;
+}
+
+static void lex_close(lex_t *lex)
+{
+    if(lex->token == TOKEN_STRING)
+        free(lex->value.string);
+    strbuffer_close(&lex->saved_text);
+}
+
+
+/*** parser ***/
+
+static json_t *parse_value(lex_t *lex, json_error_t *error);
+
+static json_t *parse_object(lex_t *lex, json_error_t *error)
+{
+    json_t *object = json_object();
+    if(!object)
+        return NULL;
+
+    lex_scan(lex, error);
+    if(lex->token == '}')
+        return object;
+
+    while(1) {
+        char *key;
+        json_t *value;
+
+        if(lex->token != TOKEN_STRING) {
+            error_set(error, lex, "string or '}' expected");
+            goto error;
+        }
+
+        key = lex_steal_string(lex);
+        if(!key)
+            return NULL;
+
+        lex_scan(lex, error);
+        if(lex->token != ':') {
+            free(key);
+            error_set(error, lex, "':' expected");
+            goto error;
+        }
+
+        lex_scan(lex, error);
+        value = parse_value(lex, error);
+        if(!value) {
+            free(key);
+            goto error;
+        }
+
+        if(json_object_set_nocheck(object, key, value)) {
+            free(key);
+            json_decref(value);
+            goto error;
+        }
+
+        json_decref(value);
+        free(key);
+
+        lex_scan(lex, error);
+        if(lex->token != ',')
+            break;
+
+        lex_scan(lex, error);
+    }
+
+    if(lex->token != '}') {
+        error_set(error, lex, "'}' expected");
+        goto error;
+    }
+
+    return object;
+
+error:
+    json_decref(object);
+    return NULL;
+}
+
+static json_t *parse_array(lex_t *lex, json_error_t *error)
+{
+    json_t *array = json_array();
+    if(!array)
+        return NULL;
+
+    lex_scan(lex, error);
+    if(lex->token == ']')
+        return array;
+
+    while(lex->token) {
+        json_t *elem = parse_value(lex, error);
+        if(!elem)
+            goto error;
+
+        if(json_array_append(array, elem)) {
+            json_decref(elem);
+            goto error;
+        }
+        json_decref(elem);
+
+        lex_scan(lex, error);
+        if(lex->token != ',')
+            break;
+
+        lex_scan(lex, error);
+    }
+
+    if(lex->token != ']') {
+        error_set(error, lex, "']' expected");
+        goto error;
+    }
+
+    return array;
+
+error:
+    json_decref(array);
+    return NULL;
+}
+
+static json_t *parse_value(lex_t *lex, json_error_t *error)
+{
+    json_t *json;
+
+    switch(lex->token) {
+        case TOKEN_STRING: {
+            json = json_string_nocheck(lex->value.string);
+            break;
+        }
+
+        case TOKEN_INTEGER: {
+            json = json_integer(lex->value.integer);
+            break;
+        }
+
+        case TOKEN_REAL: {
+            json = json_real(lex->value.real);
+            break;
+        }
+
+        case TOKEN_TRUE:
+            json = json_true();
+            break;
+
+        case TOKEN_FALSE:
+            json = json_false();
+            break;
+
+        case TOKEN_NULL:
+            json = json_null();
+            break;
+
+        case '{':
+            json = parse_object(lex, error);
+            break;
+
+        case '[':
+            json = parse_array(lex, error);
+            break;
+
+        case TOKEN_INVALID:
+            error_set(error, lex, "invalid token");
+            return NULL;
+
+        default:
+            error_set(error, lex, "unexpected token");
+            return NULL;
+    }
+
+    if(!json)
+        return NULL;
+
+    return json;
+}
+
+static json_t *parse_json(lex_t *lex, json_error_t *error)
+{
+    error_init(error);
+
+    lex_scan(lex, error);
+    if(lex->token != '[' && lex->token != '{') {
+        error_set(error, lex, "'[' or '{' expected");
+        return NULL;
+    }
+
+    return parse_value(lex, error);
+}
+
+typedef struct
+{
+    const char *data;
+    int pos;
+} string_data_t;
+
+static int string_get(void *data)
+{
+    char c;
+    string_data_t *stream = (string_data_t *)data;
+    c = stream->data[stream->pos];
+    if(c == '\0')
+        return EOF;
+    else
+    {
+        stream->pos++;
+        return c;
+    }
+}
+
+static int string_eof(void *data)
+{
+    string_data_t *stream = (string_data_t *)data;
+    return (stream->data[stream->pos] == '\0');
+}
+
+json_t *json_loads(const char *string, json_error_t *error)
+{
+    lex_t lex;
+    json_t *result;
+
+    string_data_t stream_data = {
+        string,
+        0
+    };
+
+    if(lex_init(&lex, string_get, string_eof, (void *)&stream_data))
+        return NULL;
+
+    result = parse_json(&lex, error);
+    if(!result)
+        goto out;
+
+    lex_scan(&lex, error);
+    if(lex.token != TOKEN_EOF) {
+        error_set(error, &lex, "end of file expected");
+        json_decref(result);
+        result = NULL;
+    }
+
+out:
+    lex_close(&lex);
+    return result;
+}
+
+json_t *json_loadf(FILE *input, json_error_t *error)
+{
+    lex_t lex;
+    json_t *result;
+
+    if(lex_init(&lex, (get_func)fgetc, (eof_func)feof, input))
+        return NULL;
+
+    result = parse_json(&lex, error);
+    if(!result)
+        goto out;
+
+    lex_scan(&lex, error);
+    if(lex.token != TOKEN_EOF) {
+        error_set(error, &lex, "end of file expected");
+        json_decref(result);
+        result = NULL;
+    }
+
+out:
+    lex_close(&lex);
+    return result;
+}
+
+json_t *json_load_file(const char *path, json_error_t *error)
+{
+    json_t *result;
+    FILE *fp;
+
+    error_init(error);
+
+    fp = fopen(path, "r");
+    if(!fp)
+    {
+        error_set(error, NULL, "unable to open %s: %s",
+                  path, strerror(errno));
+        return NULL;
+    }
+
+    result = json_loadf(fp, error);
+
+    fclose(fp);
+    return result;
+}
diff --git a/compat/jansson/strbuffer.c b/compat/jansson/strbuffer.c
new file mode 100644
index 0000000000..0019645ed1
--- /dev/null
+++ b/compat/jansson/strbuffer.c
@@ -0,0 +1,95 @@
+/*
+ * Copyright (c) 2009, 2010 Petri Lehtinen <petri@digip.org>
+ *
+ * Jansson is free software; you can redistribute it and/or modify
+ * it under the terms of the MIT license. See LICENSE for details.
+ */
+
+#define _GNU_SOURCE
+#include <stdlib.h>
+#include <string.h>
+#include "strbuffer.h"
+#include "util.h"
+
+#define STRBUFFER_MIN_SIZE  16
+#define STRBUFFER_FACTOR    2
+
+int strbuffer_init(strbuffer_t *strbuff)
+{
+    strbuff->size = STRBUFFER_MIN_SIZE;
+    strbuff->length = 0;
+
+    strbuff->value = malloc(strbuff->size);
+    if(!strbuff->value)
+        return -1;
+
+    /* initialize to empty */
+    strbuff->value[0] = '\0';
+    return 0;
+}
+
+void strbuffer_close(strbuffer_t *strbuff)
+{
+    free(strbuff->value);
+    strbuff->size = 0;
+    strbuff->length = 0;
+    strbuff->value = NULL;
+}
+
+void strbuffer_clear(strbuffer_t *strbuff)
+{
+    strbuff->length = 0;
+    strbuff->value[0] = '\0';
+}
+
+const char *strbuffer_value(const strbuffer_t *strbuff)
+{
+    return strbuff->value;
+}
+
+char *strbuffer_steal_value(strbuffer_t *strbuff)
+{
+    char *result = strbuff->value;
+    strbuffer_init(strbuff);
+    return result;
+}
+
+int strbuffer_append(strbuffer_t *strbuff, const char *string)
+{
+    return strbuffer_append_bytes(strbuff, string, strlen(string));
+}
+
+int strbuffer_append_byte(strbuffer_t *strbuff, char byte)
+{
+    return strbuffer_append_bytes(strbuff, &byte, 1);
+}
+
+int strbuffer_append_bytes(strbuffer_t *strbuff, const char *data, int size)
+{
+    if(strbuff->length + size >= strbuff->size)
+    {
+        strbuff->size = max(strbuff->size * STRBUFFER_FACTOR,
+                            strbuff->length + size + 1);
+
+        strbuff->value = realloc(strbuff->value, strbuff->size);
+        if(!strbuff->value)
+            return -1;
+    }
+
+    memcpy(strbuff->value + strbuff->length, data, size);
+    strbuff->length += size;
+    strbuff->value[strbuff->length] = '\0';
+
+    return 0;
+}
+
+char strbuffer_pop(strbuffer_t *strbuff)
+{
+    if(strbuff->length > 0) {
+        char c = strbuff->value[--strbuff->length];
+        strbuff->value[strbuff->length] = '\0';
+        return c;
+    }
+    else
+        return '\0';
+}
diff --git a/compat/jansson/strbuffer.h b/compat/jansson/strbuffer.h
new file mode 100644
index 0000000000..816594a826
--- /dev/null
+++ b/compat/jansson/strbuffer.h
@@ -0,0 +1,31 @@
+/*
+ * Copyright (c) 2009, 2010 Petri Lehtinen <petri@digip.org>
+ *
+ * Jansson is free software; you can redistribute it and/or modify
+ * it under the terms of the MIT license. See LICENSE for details.
+ */
+
+#ifndef STRBUFFER_H
+#define STRBUFFER_H
+
+typedef struct {
+    char *value;
+    int length;   /* bytes used */
+    int size;     /* bytes allocated */
+} strbuffer_t;
+
+int strbuffer_init(strbuffer_t *strbuff);
+void strbuffer_close(strbuffer_t *strbuff);
+
+void strbuffer_clear(strbuffer_t *strbuff);
+
+const char *strbuffer_value(const strbuffer_t *strbuff);
+char *strbuffer_steal_value(strbuffer_t *strbuff);
+
+int strbuffer_append(strbuffer_t *strbuff, const char *string);
+int strbuffer_append_byte(strbuffer_t *strbuff, char byte);
+int strbuffer_append_bytes(strbuffer_t *strbuff, const char *data, int size);
+
+char strbuffer_pop(strbuffer_t *strbuff);
+
+#endif
diff --git a/compat/jansson/utf.c b/compat/jansson/utf.c
new file mode 100644
index 0000000000..2b644501cb
--- /dev/null
+++ b/compat/jansson/utf.c
@@ -0,0 +1,190 @@
+/*
+ * Copyright (c) 2009, 2010 Petri Lehtinen <petri@digip.org>
+ *
+ * Jansson is free software; you can redistribute it and/or modify
+ * it under the terms of the MIT license. See LICENSE for details.
+ */
+
+#include <string.h>
+#include "utf.h"
+
+int utf8_encode(int32_t codepoint, char *buffer, int *size)
+{
+    if(codepoint < 0)
+        return -1;
+    else if(codepoint < 0x80)
+    {
+        buffer[0] = (char)codepoint;
+        *size = 1;
+    }
+    else if(codepoint < 0x800)
+    {
+        buffer[0] = 0xC0 + ((codepoint & 0x7C0) >> 6);
+        buffer[1] = 0x80 + ((codepoint & 0x03F));
+        *size = 2;
+    }
+    else if(codepoint < 0x10000)
+    {
+        buffer[0] = 0xE0 + ((codepoint & 0xF000) >> 12);
+        buffer[1] = 0x80 + ((codepoint & 0x0FC0) >> 6);
+        buffer[2] = 0x80 + ((codepoint & 0x003F));
+        *size = 3;
+    }
+    else if(codepoint <= 0x10FFFF)
+    {
+        buffer[0] = 0xF0 + ((codepoint & 0x1C0000) >> 18);
+        buffer[1] = 0x80 + ((codepoint & 0x03F000) >> 12);
+        buffer[2] = 0x80 + ((codepoint & 0x000FC0) >> 6);
+        buffer[3] = 0x80 + ((codepoint & 0x00003F));
+        *size = 4;
+    }
+    else
+        return -1;
+
+    return 0;
+}
+
+int utf8_check_first(char byte)
+{
+    unsigned char u = (unsigned char)byte;
+
+    if(u < 0x80)
+        return 1;
+
+    if(0x80 <= u && u <= 0xBF) {
+        /* second, third or fourth byte of a multi-byte
+           sequence, i.e. a "continuation byte" */
+        return 0;
+    }
+    else if(u == 0xC0 || u == 0xC1) {
+        /* overlong encoding of an ASCII byte */
+        return 0;
+    }
+    else if(0xC2 <= u && u <= 0xDF) {
+        /* 2-byte sequence */
+        return 2;
+    }
+
+    else if(0xE0 <= u && u <= 0xEF) {
+        /* 3-byte sequence */
+        return 3;
+    }
+    else if(0xF0 <= u && u <= 0xF4) {
+        /* 4-byte sequence */
+        return 4;
+    }
+    else { /* u >= 0xF5 */
+        /* Restricted (start of 4-, 5- or 6-byte sequence) or invalid
+           UTF-8 */
+        return 0;
+    }
+}
+
+int utf8_check_full(const char *buffer, int size, int32_t *codepoint)
+{
+    int i;
+    int32_t value = 0;
+    unsigned char u = (unsigned char)buffer[0];
+
+    if(size == 2)
+    {
+        value = u & 0x1F;
+    }
+    else if(size == 3)
+    {
+        value = u & 0xF;
+    }
+    else if(size == 4)
+    {
+        value = u & 0x7;
+    }
+    else
+        return 0;
+
+    for(i = 1; i < size; i++)
+    {
+        u = (unsigned char)buffer[i];
+
+        if(u < 0x80 || u > 0xBF) {
+            /* not a continuation byte */
+            return 0;
+        }
+
+        value = (value << 6) + (u & 0x3F);
+    }
+
+    if(value > 0x10FFFF) {
+        /* not in Unicode range */
+        return 0;
+    }
+
+    else if(0xD800 <= value && value <= 0xDFFF) {
+        /* invalid code point (UTF-16 surrogate halves) */
+        return 0;
+    }
+
+    else if((size == 2 && value < 0x80) ||
+            (size == 3 && value < 0x800) ||
+            (size == 4 && value < 0x10000)) {
+        /* overlong encoding */
+        return 0;
+    }
+
+    if(codepoint)
+        *codepoint = value;
+
+    return 1;
+}
+
+const char *utf8_iterate(const char *buffer, int32_t *codepoint)
+{
+    int count;
+    int32_t value;
+
+    if(!*buffer)
+        return buffer;
+
+    count = utf8_check_first(buffer[0]);
+    if(count <= 0)
+        return NULL;
+
+    if(count == 1)
+        value = (unsigned char)buffer[0];
+    else
+    {
+        if(!utf8_check_full(buffer, count, &value))
+            return NULL;
+    }
+
+    if(codepoint)
+        *codepoint = value;
+
+    return buffer + count;
+}
+
+int utf8_check_string(const char *string, int length)
+{
+    int i;
+
+    if(length == -1)
+        length = strlen(string);
+
+    for(i = 0; i < length; i++)
+    {
+        int count = utf8_check_first(string[i]);
+        if(count == 0)
+            return 0;
+        else if(count > 1)
+        {
+            if(i + count > length)
+                return 0;
+
+            if(!utf8_check_full(&string[i], count, NULL))
+                return 0;
+
+            i += count - 1;
+        }
+    }
+
+    return 1;
+}
diff --git a/compat/jansson/utf.h b/compat/jansson/utf.h
new file mode 100644
index 0000000000..8e95296404
--- /dev/null
+++ b/compat/jansson/utf.h
@@ -0,0 +1,28 @@
+/*
+ * Copyright (c) 2009, 2010 Petri Lehtinen <petri@digip.org>
+ *
+ * Jansson is free software; you can redistribute it and/or modify
+ * it under the terms of the MIT license. See LICENSE for details.
+ */
+
+#ifndef UTF_H
+#define UTF_H
+
+#include <config.h>
+
+#ifdef HAVE_INTTYPES_H
+/* inttypes.h includes stdint.h in a standard environment, so there's
+no need to include stdint.h separately. If inttypes.h doesn't define
+int32_t, it's defined in config.h. */
+#include <inttypes.h>
+#endif
+
+int utf8_encode(int codepoint, char *buffer, int *size);
+
+int utf8_check_first(char byte);
+int utf8_check_full(const char *buffer, int size, int32_t *codepoint);
+const char *utf8_iterate(const char *buffer, int32_t *codepoint);
+
+int utf8_check_string(const char *string, int length);
+
+#endif
diff --git a/compat/jansson/util.h b/compat/jansson/util.h
new file mode 100644
index 0000000000..33e5d62680
--- /dev/null
+++ b/compat/jansson/util.h
@@ -0,0 +1,15 @@
+/*
+ * Copyright (c) 2009, 2010 Petri Lehtinen <petri@digip.org>
+ *
+ * Jansson is free software; you can redistribute it and/or modify
+ * it under the terms of the MIT license. See LICENSE for details.
+ */
+
+#ifndef UTIL_H
+#define UTIL_H
+
+#ifndef max
+#define max(a, b)  ((a) > (b) ? (a) : (b))
+#endif
+
+#endif
diff --git a/compat/jansson/value.c b/compat/jansson/value.c
new file mode 100644
index 0000000000..591b89e1fe
--- /dev/null
+++ b/compat/jansson/value.c
@@ -0,0 +1,976 @@
+/*
+ * Copyright (c) 2009, 2010 Petri Lehtinen <petri@digip.org>
+ *
+ * Jansson is free software; you can redistribute it and/or modify
+ * it under the terms of the MIT license. See LICENSE for details.
+ */
+
+#define _GNU_SOURCE
+
+#include <config.h>
+
+#include <stdlib.h>
+#include <string.h>
+
+#include <jansson.h>
+#include "hashtable.h"
+#include "jansson_private.h"
+#include "utf.h"
+#include "util.h"
+
+
+static inline void json_init(json_t *json, json_type type)
+{
+    json->type = type;
+    json->refcount = 1;
+}
+
+
+/*** object ***/
+
+/* This macro just returns a pointer that's a few bytes backwards from
+   string. This makes it possible to pass a pointer to object_key_t
+   when only the string inside it is used, without actually creating
+   an object_key_t instance. */
+#define string_to_key(string)  container_of(string, object_key_t, key)
+
+static unsigned int hash_key(const void *ptr)
+{
+    const char *str = ((const object_key_t *)ptr)->key;
+
+    unsigned int hash = 5381;
+    unsigned int c;
+
+    while((c = (unsigned int)*str))
+    {
+        hash = ((hash << 5) + hash) + c;
+        str++;
+    }
+
+    return hash;
+}
+
+static int key_equal(const void *ptr1, const void *ptr2)
+{
+    return strcmp(((const object_key_t *)ptr1)->key,
+                  ((const object_key_t *)ptr2)->key) == 0;
+}
+
+static void value_decref(void *value)
+{
+    json_decref((json_t *)value);
+}
+
+json_t *json_object(void)
+{
+    json_object_t *object = malloc(sizeof(json_object_t));
+    if(!object)
+        return NULL;
+    json_init(&object->json, JSON_OBJECT);
+
+    if(hashtable_init(&object->hashtable, hash_key, key_equal,
+                      free, value_decref))
+    {
+        free(object);
+        return NULL;
+    }
+
+    object->serial = 0;
+    object->visited = 0;
+
+    return &object->json;
+}
+
+static void json_delete_object(json_object_t *object)
+{
+    hashtable_close(&object->hashtable);
+    free(object);
+}
+
+unsigned int json_object_size(const json_t *json)
+{
+    json_object_t *object;
+
+    if(!json_is_object(json))
+        return -1;
+
+    object = json_to_object(json);
+    return object->hashtable.size;
+}
+
+json_t *json_object_get(const json_t *json, const char *key)
+{
+    json_object_t *object;
+
+    if(!json_is_object(json))
+        return NULL;
+
+    object = json_to_object(json);
+    return hashtable_get(&object->hashtable, string_to_key(key));
+}
+
+int json_object_set_new_nocheck(json_t *json, const char *key, json_t *value)
+{
+    json_object_t *object;
+    object_key_t *k;
+
+    if(!key || !value)
+        return -1;
+
+    if(!json_is_object(json) || json == value)
+    {
+        json_decref(value);
+        return -1;
+    }
+    object = json_to_object(json);
+
+    k = malloc(sizeof(object_key_t) + strlen(key) + 1);
+    if(!k)
+        return -1;
+
+    k->serial = object->serial++;
+    strcpy(k->key, key);
+
+    if(hashtable_set(&object->hashtable, k, value))
+    {
+        json_decref(value);
+        return -1;
+    }
+
+    return 0;
+}
+
+int json_object_set_new(json_t *json, const char *key, json_t *value)
+{
+    if(!key || !utf8_check_string(key, -1))
+    {
+        json_decref(value);
+        return -1;
+    }
+
+    return json_object_set_new_nocheck(json, key, value);
+}
+
+int json_object_del(json_t *json, const char *key)
+{
+    json_object_t *object;
+
+    if(!json_is_object(json))
+        return -1;
+
+    object = json_to_object(json);
+    return hashtable_del(&object->hashtable, string_to_key(key));
+}
+
+int json_object_clear(json_t *json)
+{
+    json_object_t *object;
+
+    if(!json_is_object(json))
+        return -1;
+
+    object = json_to_object(json);
+    hashtable_clear(&object->hashtable);
+
+    return 0;
+}
+
+int json_object_update(json_t *object, json_t *other)
+{
+    void *iter;
+
+    if(!json_is_object(object) || !json_is_object(other))
+        return -1;
+
+    iter = json_object_iter(other);
+    while(iter) {
+        const char *key;
+        json_t *value;
+
+        key = json_object_iter_key(iter);
+        value = json_object_iter_value(iter);
+
+        if(json_object_set_nocheck(object, key, value))
+            return -1;
+
+        iter = json_object_iter_next(other, iter);
+    }
+
+    return 0;
+}
+
+void *json_object_iter(json_t *json)
+{
+    json_object_t *object;
+
+    if(!json_is_object(json))
+        return NULL;
+
+    object = json_to_object(json);
+    return hashtable_iter(&object->hashtable);
+}
+
+void *json_object_iter_at(json_t *json, const char *key)
+{
+    json_object_t *object;
+
+    if(!key || !json_is_object(json))
+        return NULL;
+
+    object = json_to_object(json);
+    return hashtable_iter_at(&object->hashtable, string_to_key(key));
+}
+
+void *json_object_iter_next(json_t *json, void *iter)
+{
+    json_object_t *object;
+
+    if(!json_is_object(json) || iter == NULL)
+        return NULL;
+
+    object = json_to_object(json);
+    return hashtable_iter_next(&object->hashtable, iter);
+}
+
+const object_key_t *jsonp_object_iter_fullkey(void *iter)
+{
+    if(!iter)
+        return NULL;
+
+    return hashtable_iter_key(iter);
+}
+
+const char *json_object_iter_key(void *iter)
+{
+    if(!iter)
+        return NULL;
+
+    return jsonp_object_iter_fullkey(iter)->key;
+}
+
+json_t *json_object_iter_value(void *iter)
+{
+    if(!iter)
+        return NULL;
+
+    return (json_t *)hashtable_iter_value(iter);
+}
+
+int json_object_iter_set_new(json_t *json, void *iter, json_t *value)
+{
+    json_object_t *object;
+
+    if(!json_is_object(json) || !iter || !value)
+        return -1;
+
+    object = json_to_object(json);
+    hashtable_iter_set(&object->hashtable, iter, value);
+
+    return 0;
+}
+
+static int json_object_equal(json_t *object1, json_t *object2)
+{
+    void *iter;
+
+    if(json_object_size(object1) != json_object_size(object2))
+        return 0;
+
+    iter = json_object_iter(object1);
+    while(iter)
+    {
+        const char *key;
+        json_t *value1, *value2;
+
+        key = json_object_iter_key(iter);
+        value1 = json_object_iter_value(iter);
+        value2 = json_object_get(object2, key);
+
+        if(!json_equal(value1, value2))
+            return 0;
+
+        iter = json_object_iter_next(object1, iter);
+    }
+
+    return 1;
+}
+
+static json_t *json_object_copy(json_t *object)
+{
+    json_t *result;
+    void *iter;
+
+    result = json_object();
+    if(!result)
+        return NULL;
+
+    iter = json_object_iter(object);
+    while(iter)
+    {
+        const char *key;
+        json_t *value;
+
+        key = json_object_iter_key(iter);
+        value = json_object_iter_value(iter);
+        json_object_set_nocheck(result, key, value);
+
+        iter = json_object_iter_next(object, iter);
+    }
+
+    return result;
+}
+
+static json_t *json_object_deep_copy(json_t *object)
+{
+    json_t *result;
+    void *iter;
+
+    result = json_object();
+    if(!result)
+        return NULL;
+
+    iter = json_object_iter(object);
+    while(iter)
+    {
+        const char *key;
+        json_t *value;
+
+        key = json_object_iter_key(iter);
+        value = json_object_iter_value(iter);
+        json_object_set_new_nocheck(result, key, json_deep_copy(value));
+
+        iter = json_object_iter_next(object, iter);
+    }
+
+    return result;
+}
+
+
+/*** array ***/
+
+json_t *json_array(void)
+{
+    json_array_t *array = malloc(sizeof(json_array_t));
+    if(!array)
+        return NULL;
+    json_init(&array->json, JSON_ARRAY);
+
+    array->entries = 0;
+    array->size = 8;
+
+    array->table = malloc(array->size * sizeof(json_t *));
+    if(!array->table) {
+        free(array);
+        return NULL;
+    }
+
+    array->visited = 0;
+
+    return &array->json;
+}
+
+static void json_delete_array(json_array_t *array)
+{
+    unsigned int i;
+
+    for(i = 0; i < array->entries; i++)
+        json_decref(array->table[i]);
+
+    free(array->table);
+    free(array);
+}
+
+unsigned int json_array_size(const json_t *json)
+{
+    if(!json_is_array(json))
+        return 0;
+
+    return json_to_array(json)->entries;
+}
+
+json_t *json_array_get(const json_t *json, unsigned int index)
+{
+    json_array_t *array;
+    if(!json_is_array(json))
+        return NULL;
+    array = json_to_array(json);
+
+    if(index >= array->entries)
+        return NULL;
+
+    return array->table[index];
+}
+
+int json_array_set_new(json_t *json, unsigned int index, json_t *value)
+{
+    json_array_t *array;
+
+    if(!value)
+        return -1;
+
+    if(!json_is_array(json) || json == value)
+    {
+        json_decref(value);
+        return -1;
+    }
+    array = json_to_array(json);
+
+    if(index >= array->entries)
+    {
+        json_decref(value);
+        return -1;
+    }
+
+    json_decref(array->table[index]);
+    array->table[index] = value;
+
+    return 0;
+}
+
+static void array_move(json_array_t *array, unsigned int dest,
+                       unsigned int src, unsigned int count)
+{
+    memmove(&array->table[dest], &array->table[src], count * sizeof(json_t *));
+}
+
+static void array_copy(json_t **dest, unsigned int dpos,
+                       json_t **src, unsigned int spos,
+                       unsigned int count)
+{
+    memcpy(&dest[dpos], &src[spos], count * sizeof(json_t *));
+}
+
+static json_t **json_array_grow(json_array_t *array,
+                                unsigned int amount,
+                                int copy)
+{
+    unsigned int new_size;
+    json_t **old_table, **new_table;
+
+    if(array->entries + amount <= array->size)
+        return array->table;
+
+    old_table = array->table;
+
+    new_size = max(array->size + amount, array->size * 2);
+    new_table = malloc(new_size * sizeof(json_t *));
+    if(!new_table)
+        return NULL;
+
+    array->size = new_size;
+    array->table = new_table;
+
+    if(copy) {
+        array_copy(array->table, 0, old_table, 0, array->entries);
+        free(old_table);
+        return array->table;
+    }
+
+    return old_table;
+}
+
+int json_array_append_new(json_t *json, json_t *value)
+{
+    json_array_t *array;
+
+    if(!value)
+        return -1;
+
+    if(!json_is_array(json) || json == value)
+    {
+        json_decref(value);
+        return -1;
+    }
+    array = json_to_array(json);
+
+    if(!json_array_grow(array, 1, 1)) {
+        json_decref(value);
+        return -1;
+    }
+
+    array->table[array->entries] = value;
+    array->entries++;
+
+    return 0;
+}
+
+int json_array_insert_new(json_t *json, unsigned int index, json_t *value)
+{
+    json_array_t *array;
+    json_t **old_table;
+
+    if(!value)
+        return -1;
+
+    if(!json_is_array(json) || json == value) {
+        json_decref(value);
+        return -1;
+    }
+    array = json_to_array(json);
+
+    if(index > array->entries) {
+        json_decref(value);
+        return -1;
+    }
+
+    old_table = json_array_grow(array, 1, 0);
+    if(!old_table) {
+        json_decref(value);
+        return -1;
+    }
+
+    if(old_table != array->table) {
+        array_copy(array->table, 0, old_table, 0, index);
+        array_copy(array->table, index + 1, old_table, index,
+                   array->entries - index);
+        free(old_table);
+    }
+    else
+        array_move(array, index + 1, index, array->entries - index);
+
+    array->table[index] = value;
+    array->entries++;
+
+    return 0;
+}
+
+int json_array_remove(json_t *json, unsigned int index)
+{
+    json_array_t *array;
+
+    if(!json_is_array(json))
+        return -1;
+    array = json_to_array(json);
+
+    if(index >= array->entries)
+        return -1;
+
+    json_decref(array->table[index]);
+
+    array_move(array, index, index + 1, array->entries - index);
+    array->entries--;
+
+    return 0;
+}
+
+int json_array_clear(json_t *json)
+{
+    json_array_t *array;
+    unsigned int i;
+
+    if(!json_is_array(json))
+        return -1;
+    array = json_to_array(json);
+
+    for(i = 0; i < array->entries; i++)
+        json_decref(array->table[i]);
+
+    array->entries = 0;
+    return 0;
+}
+
+int json_array_extend(json_t *json, json_t *other_json)
+{
+    json_array_t *array, *other;
+    unsigned int i;
+
+    if(!json_is_array(json) || !json_is_array(other_json))
+        return -1;
+    array = json_to_array(json);
+    other = json_to_array(other_json);
+
+    if(!json_array_grow(array, other->entries, 1))
+        return -1;
+
+    for(i = 0; i < other->entries; i++)
+        json_incref(other->table[i]);
+
+    array_copy(array->table, array->entries, other->table, 0, other->entries);
+
+    array->entries += other->entries;
+    return 0;
+}
+
+static int json_array_equal(json_t *array1, json_t *array2)
+{
+    unsigned int i, size;
+
+    size = json_array_size(array1);
+    if(size != json_array_size(array2))
+        return 0;
+
+    for(i = 0; i < size; i++)
+    {
+        json_t *value1, *value2;
+
+        value1 = json_array_get(array1, i);
+        value2 = json_array_get(array2, i);
+
+        if(!json_equal(value1, value2))
+            return 0;
+    }
+
+    return 1;
+}
+
+static json_t *json_array_copy(json_t *array)
+{
+    json_t *result;
+    unsigned int i;
+
+    result = json_array();
+    if(!result)
+        return NULL;
+
+    for(i = 0; i < json_array_size(array); i++)
+        json_array_append(result, json_array_get(array, i));
+
+    return result;
+}
+
+static json_t *json_array_deep_copy(json_t *array)
+{
+    json_t *result;
+    unsigned int i;
+
+    result = json_array();
+    if(!result)
+        return NULL;
+
+    for(i = 0; i < json_array_size(array); i++)
+        json_array_append_new(result, json_deep_copy(json_array_get(array, i)));
+
+    return result;
+}
+
+/*** string ***/
+
+json_t *json_string_nocheck(const char *value)
+{
+    json_string_t *string;
+
+    if(!value)
+        return NULL;
+
+    string = malloc(sizeof(json_string_t));
+    if(!string)
+        return NULL;
+    json_init(&string->json, JSON_STRING);
+
+    string->value = strdup(value);
+    if(!string->value) {
+        free(string);
+        return NULL;
+    }
+
+    return &string->json;
+}
+
+json_t *json_string(const char *value)
+{
+    if(!value || !utf8_check_string(value, -1))
+        return NULL;
+
+    return json_string_nocheck(value);
+}
+
+const char *json_string_value(const json_t *json)
+{
+    if(!json_is_string(json))
+        return NULL;
+
+    return json_to_string(json)->value;
+}
+
+int json_string_set_nocheck(json_t *json, const char *value)
+{
+    char *dup;
+    json_string_t *string;
+
+    dup = strdup(value);
+    if(!dup)
+        return -1;
+
+    string = json_to_string(json);
+    free(string->value);
+    string->value = dup;
+
+    return 0;
+}
+
+int json_string_set(json_t *json, const char *value)
+{
+    if(!value || !utf8_check_string(value, -1))
+        return -1;
+
+    return json_string_set_nocheck(json, value);
+}
+
+static void json_delete_string(json_string_t *string)
+{
+    free(string->value);
+    free(string);
+}
+
+static int json_string_equal(json_t *string1, json_t *string2)
+{
+    return strcmp(json_string_value(string1), json_string_value(string2)) == 0;
+}
+
+static json_t *json_string_copy(json_t *string)
+{
+    return json_string_nocheck(json_string_value(string));
+}
+
+
+/*** integer ***/
+
+json_t *json_integer(int value)
+{
+    json_integer_t *integer = malloc(sizeof(json_integer_t));
+    if(!integer)
+        return NULL;
+    json_init(&integer->json, JSON_INTEGER);
+
+    integer->value = value;
+    return &integer->json;
+}
+
+int json_integer_value(const json_t *json)
+{
+    if(!json_is_integer(json))
+        return 0;
+
+    return json_to_integer(json)->value;
+}
+
+int json_integer_set(json_t *json, int value)
+{
+    if(!json_is_integer(json))
+        return -1;
+
+    json_to_integer(json)->value = value;
+
+    return 0;
+}
+
+static void json_delete_integer(json_integer_t *integer)
+{
+    free(integer);
+}
+
+static int json_integer_equal(json_t *integer1, json_t *integer2)
+{
+    return json_integer_value(integer1) == json_integer_value(integer2);
+}
+
+static json_t *json_integer_copy(json_t *integer)
+{
+    return json_integer(json_integer_value(integer));
+}
+
+
+/*** real ***/
+
+json_t *json_real(double value)
+{
+    json_real_t *real = malloc(sizeof(json_real_t));
+    if(!real)
+        return NULL;
+    json_init(&real->json, JSON_REAL);
+
+    real->value = value;
+    return &real->json;
+}
+
+double json_real_value(const json_t *json)
+{
+    if(!json_is_real(json))
+        return 0;
+
+    return json_to_real(json)->value;
+}
+
+int json_real_set(json_t *json, double value)
+{
+    if(!json_is_real(json))
+        return 0;
+
+    json_to_real(json)->value = value;
+
+    return 0;
+}
+
+static void json_delete_real(json_real_t *real)
+{
+    free(real);
+}
+
+static int json_real_equal(json_t *real1, json_t *real2)
+{
+    return json_real_value(real1) == json_real_value(real2);
+}
+
+static json_t *json_real_copy(json_t *real)
+{
+    return json_real(json_real_value(real));
+}
+
+
+/*** number ***/
+
+double json_number_value(const json_t *json)
+{
+    if(json_is_integer(json))
+        return json_integer_value(json);
+    else if(json_is_real(json))
+        return json_real_value(json);
+    else
+        return 0.0;
+}
+
+
+/*** simple values ***/
+
+json_t *json_true(void)
+{
+    static json_t the_true = {
+        JSON_TRUE,
+        (unsigned int)-1
+    };
+    return &the_true;
+}
+
+
+json_t *json_false(void)
+{
+    static json_t the_false = {
+        JSON_FALSE,
+        (unsigned int)-1
+    };
+    return &the_false;
+}
+
+
+json_t *json_null(void)
+{
+    static json_t the_null = {
+        JSON_NULL,
+        (unsigned int)-1
+    };
+    return &the_null;
+}
+
+
+/*** deletion ***/
+
+void json_delete(json_t *json)
+{
+    if(json_is_object(json))
+        json_delete_object(json_to_object(json));
+
+    else if(json_is_array(json))
+        json_delete_array(json_to_array(json));
+
+    else if(json_is_string(json))
+        json_delete_string(json_to_string(json));
+
+    else if(json_is_integer(json))
+        json_delete_integer(json_to_integer(json));
+
+    else if(json_is_real(json))
+        json_delete_real(json_to_real(json));
+
+    /* json_delete is not called for true, false or null */
+}
+
+
+/*** equality ***/
+
+int json_equal(json_t *json1, json_t *json2)
+{
+    if(!json1 || !json2)
+        return 0;
+
+    if(json_typeof(json1) != json_typeof(json2))
+        return 0;
+
+    /* this covers true, false and null as they are singletons */
+    if(json1 == json2)
+        return 1;
+
+    if(json_is_object(json1))
+        return json_object_equal(json1, json2);
+
+    if(json_is_array(json1))
+        return json_array_equal(json1, json2);
+
+    if(json_is_string(json1))
+        return json_string_equal(json1, json2);
+
+    if(json_is_integer(json1))
+        return json_integer_equal(json1, json2);
+
+    if(json_is_real(json1))
+        return json_real_equal(json1, json2);
+
+    return 0;
+}
+
+
+/*** copying ***/
+
+json_t *json_copy(json_t *json)
+{
+    if(!json)
+        return NULL;
+
+    if(json_is_object(json))
+        return json_object_copy(json);
+
+    if(json_is_array(json))
+        return json_array_copy(json);
+
+    if(json_is_string(json))
+        return json_string_copy(json);
+
+    if(json_is_integer(json))
+        return json_integer_copy(json);
+
+    if(json_is_real(json))
+        return json_real_copy(json);
+
+    if(json_is_true(json) || json_is_false(json) || json_is_null(json))
+        return json;
+
+    return NULL;
+}
+
+json_t *json_deep_copy(json_t *json)
+{
+    if(!json)
+        return NULL;
+
+    if(json_is_object(json))
+        return json_object_deep_copy(json);
+
+    if(json_is_array(json))
+        return json_array_deep_copy(json);
+
+    /* for the rest of the types, deep copying doesn't differ from
+       shallow copying */
+
+    if(json_is_string(json))
+        return json_string_copy(json);
+
+    if(json_is_integer(json))
+        return json_integer_copy(json);
+
+    if(json_is_real(json))
+        return json_real_copy(json);
+
+    if(json_is_true(json) || json_is_false(json) || json_is_null(json))
+        return json;
+
+    return NULL;
+}
diff --git a/compat/stdbool.h b/compat/stdbool.h
new file mode 100644
index 0000000000..31d04564de
--- /dev/null
+++ b/compat/stdbool.h
@@ -0,0 +1,6 @@
+#pragma once
+
+#define false   0
+#define true    1
+
+#define bool int
diff --git a/compat/sys/time.h b/compat/sys/time.h
new file mode 100644
index 0000000000..0326e1dfd7
--- /dev/null
+++ b/compat/sys/time.h
@@ -0,0 +1,11 @@
+#pragma once
+#ifdef __cplusplus
+extern "C"
+{
+#endif
+int gettimeofday(struct timeval *tv, struct timezone *tz);
+void usleep(__int64 usec);
+#ifdef __cplusplus
+}
+#endif
+typedef __int64 useconds_t;
diff --git a/compat/thrust/CHANGELOG b/compat/thrust/CHANGELOG
new file mode 100644
index 0000000000..110c6689c2
--- /dev/null
+++ b/compat/thrust/CHANGELOG
@@ -0,0 +1,662 @@
+#######################################
+#           Thrust v1.7.0             #
+#######################################
+
+Summary
+    Thrust 1.7.0 introduces a new interface for controlling algorithm execution as
+    well as several new algorithms and performance improvements. With this new
+    interface, users may directly control how algorithms execute as well as details
+    such as the allocation of temporary storage. Key/value versions of thrust::merge
+    and the set operation algorithms have been added, as well stencil versions of
+    partitioning algorithms. thrust::tabulate has been introduced to tabulate the
+    values of functions taking integers. For 32b types, new CUDA merge and set
+    operations provide 2-15x faster performance while a new CUDA comparison sort
+    provides 1.3-4x faster performance. Finally, a new TBB reduce_by_key implementation
+    provides 80% faster performance.
+
+Breaking API Changes
+    Dispatch
+      Custom user backend systems' tag types must now inherit from the corresponding system's execution_policy template (e.g. thrust::cuda::execution_policy) instead
+      of the tag struct (e.g. thrust::cuda::tag). Otherwise, algorithm specializations will silently go unfound during dispatch.
+      See examples/minimal_custom_backend.cu and examples/cuda/fallback_allocator.cu for usage examples.
+
+      thrust::advance and thrust::distance are no longer dispatched based on iterator system type and thus may no longer be customized.
+
+    Iterators
+      iterator_facade and iterator_adaptor's Pointer template parameters have been eliminated.
+      iterator_adaptor has been moved into the thrust namespace (previously thrust::experimental::iterator_adaptor).
+      iterator_facade has been moved into the thrust namespace (previously thrust::experimental::iterator_facade).
+      iterator_core_access has been moved into the thrust namespace (previously thrust::experimental::iterator_core_access).
+      All iterators' nested pointer typedef (the type of the result of operator->) is now void instead of a pointer type to indicate that such expressions are currently impossible.
+      Floating point counting_iterators' nested difference_type typedef is now a signed integral type instead of a floating point type.
+
+    Other
+      normal_distribution has been moved into the thrust::random namespace (previously thrust::random::experimental::normal_distribution).
+      Placeholder expressions may no longer include the comma operator.
+
+New Features
+    Execution Policies
+      Users may directly control the dispatch of algorithm invocations with optional execution policy arguments.
+      For example, instead of wrapping raw pointers allocated by cudaMalloc with thrust::device_ptr, the thrust::device execution_policy may be passed as an argument to an algorithm invocation to enable CUDA execution.
+      The following execution policies are supported in this version:
+
+        thrust::host
+        thrust::device
+        thrust::cpp::par
+        thrust::cuda::par
+        thrust::omp::par
+        thrust::tbb::par
+
+    Algorithms
+	free
+	get_temporary_buffer
+	malloc
+        merge_by_key
+        partition with stencil
+        partition_copy with stencil
+	return_temporary_buffer
+        set_difference_by_key
+        set_intersection_by_key
+        set_symmetric_difference_by_key
+        set_union_by_key
+        stable_partition with stencil
+        stable_partition_copy with stencil
+	tabulate
+
+New Examples
+    uninitialized_vector demonstrates how to use a custom allocator to avoid the automatic initialization of elements in thrust::device_vector.
+
+Other Enhancements
+    Authors of custom backend systems may manipulate arbitrary state during algorithm dispatch by incorporating it into their execution_policy parameter.
+    Users may control the allocation of temporary storage during algorithm execution by passing standard allocators as parameters via execution policies such as thrust::device.
+    THRUST_DEVICE_SYSTEM_CPP has been added as a compile-time target for the device backend. 
+    CUDA merge performance is 2-15x faster.
+    CUDA comparison sort performance is 1.3-4x faster.
+    CUDA set operation performance is 1.5-15x faster.
+    TBB reduce_by_key performance is 80% faster.
+    Several algorithms have been parallelized with TBB.
+    Support for user allocators in vectors has been improved.
+    The sparse_vector example is now implemented with merge_by_key instead of sort_by_key.
+    Warnings have been eliminated in various contexts.
+    Warnings about __host__ or __device__-only functions called from __host__ __device__ functions have been eliminated in various contexts.
+    Documentation about algorithm requirements have been improved.
+    Simplified the minimal_custom_backend example.
+    Simplified the cuda/custom_temporary_allocation example.
+    Simplified the cuda/fallback_allocator example.
+
+Bug Fixes
+    #248 fix broken counting_iterator<float> behavior with OpenMP
+    #231, #209 fix set operation failures with CUDA
+    #187 fix incorrect occupancy calculation with CUDA
+    #153 fix broken multigpu behavior with CUDA
+    #142 eliminate warning produced by thrust::random::taus88 and MSVC 2010
+    #208 correctly initialize elements in temporary storage when necessary
+    #16 fix compilation error when sorting bool with CUDA
+    #10 fix ambiguous overloads of reinterpret_tag
+
+Known Issues
+    g++ versions 4.3 and lower may fail to dispatch thrust::get_temporary_buffer correctly causing infinite recursion in examples such as cuda/custom_temporary_allocation.
+
+Acknowledgments
+    Thanks to Sean Baxter, Bryan Catanzaro, and Manjunath Kudlur for contributing a faster merge implementation for CUDA.
+    Thanks to Sean Baxter for contributing a faster set operation implementation for CUDA.
+    Thanks to Cliff Woolley for contributing a correct occupancy calculation algorithm.
+
+#######################################
+#           Thrust v1.6.0             #
+#######################################
+
+Summary
+    Thrust v1.6.0 provides an interface for customization and extension and a new
+    backend system based on the Threading Building Blocks library. With this
+    new interface, programmers may customize the behavior of specific algorithms
+    as well as control the allocation of temporary storage or invent entirely new
+    backends. These enhancements also allow multiple different backend systems
+    such as CUDA and OpenMP to coexist within a single program. Support for TBB
+    allows Thrust programs to integrate more naturally into applications which
+    may already employ the TBB task scheduler.
+
+Breaking API Changes
+    The header <thrust/experimental/cuda/pinned_allocator.h> has been moved to <thrust/system/cuda/experimental/pinned_allocator.h>
+    thrust::experimental::cuda::pinned_allocator has been moved to thrust::cuda::experimental::pinned_allocator
+    The macro THRUST_DEVICE_BACKEND has been renamed THRUST_DEVICE_SYSTEM
+    The macro THRUST_DEVICE_BACKEND_CUDA has been renamed THRUST_DEVICE_SYSTEM_CUDA
+    The macro THRUST_DEVICE_BACKEND_OMP has been renamed THRUST_DEVICE_SYSTEM_OMP
+    thrust::host_space_tag has been renamed thrust::host_system_tag
+    thrust::device_space_tag has been renamed thrust::device_system_tag
+    thrust::any_space_tag has been renamed thrust::any_system_tag
+    thrust::iterator_space has been renamed thrust::iterator_system
+    
+
+New Features
+    Backend Systems
+        Threading Building Blocks (TBB) is now supported
+    Functions
+        for_each_n
+        raw_reference_cast
+    Types
+        pointer
+        reference
+
+New Examples
+    cuda/custom_temporary_allocation
+    cuda/fallback_allocator
+    device_ptr
+    expand
+    minimal_custom_backend
+    raw_reference_cast
+    set_operations
+
+Other Enhancements
+    thrust::for_each now returns the end of the input range similar to most other algorithms
+    thrust::pair and thrust::tuple have swap functionality
+    all CUDA algorithms now support large data types
+    iterators may be dereferenced in user __device__ or __global__ functions
+    the safe use of different backend systems is now possible within a single binary
+
+Bug Fixes
+    #469 min_element and max_element algorithms no longer require a const comparison operator
+
+Known Issues
+    cudafe++.exe may crash when parsing TBB headers on Windows. 
+
+#######################################
+#           Thrust v1.5.3             #
+#######################################
+
+Summary
+    Small bug fixes
+
+Bug Fixes
+    Avoid warnings about potential race due to __shared__ non-POD variable
+
+#######################################
+#           Thrust v1.5.2             #
+#######################################
+
+Summary
+    Small bug fixes
+
+Bug Fixes
+    Fixed warning about C-style initialization of structures
+
+#######################################
+#           Thrust v1.5.1             #
+#######################################
+
+Summary
+    Small bug fixes
+
+Bug Fixes
+    Sorting data referenced by permutation_iterators on CUDA produces invalid results
+
+#######################################
+#           Thrust v1.5.0             #
+#######################################
+
+Summary
+    Thrust v1.5.0 provides introduces new programmer productivity and performance
+    enhancements. New functionality for creating anonymous "lambda" functions has
+    been added. A faster host sort provides 2-10x faster performance for sorting
+    arithmetic types on (single-threaded) CPUs. A new OpenMP sort provides
+    2.5x-3.0x speedup over the host sort using a quad-core CPU. When sorting
+    arithmetic types with the OpenMP backend the combined performance improvement
+    is 5.9x for 32-bit integers and ranges from 3.0x (64-bit types) to 14.2x
+    (8-bit types). A new CUDA reduce_by_key implementation provides 2-3x faster
+    performance.
+
+Breaking API Changes
+    device_ptr<void> no longer unsafely converts to device_ptr<T> without an
+    explicit cast. Use the expression
+    device_pointer_cast(static_cast<int*>(void_ptr.get()))
+    to convert, for example, device_ptr<void> to device_ptr<int>.
+
+New Features
+    Functions
+        stencil-less transform_if
+
+    Types
+        lambda placeholders
+
+New Examples
+    lambda
+
+Other Enhancements
+    host sort is 2-10x faster for arithmetic types
+    OMP sort provides speedup over host sort
+    reduce_by_key is 2-3x faster
+    reduce_by_key no longer requires O(N) temporary storage
+    CUDA scan algorithms are 10-40% faster
+    host_vector and device_vector are now documented
+    out-of-memory exceptions now provide detailed information from CUDART
+    improved histogram example
+    device_reference now has a specialized swap
+    reduce_by_key and scan algorithms are compatible with discard_iterator
+
+Removed Functionality
+
+Bug Fixes
+     #44 allow host_vector to compile when value_type uses __align__
+    #198 allow adjacent_difference to permit safe in-situ operation
+    #303 make thrust thread-safe
+    #313 avoid race conditions in device_vector::insert
+    #314 avoid unintended adl invocation when dispatching copy
+    #365 fix merge and set operation failures
+
+Known Issues
+    None
+
+Acknowledgments
+    Thanks to Manjunath Kudlur for contributing his Carbon library, from which the lambda functionality is derived.
+    Thanks to Jean-Francois Bastien for suggesting a fix for issue 303.
+
+#######################################
+#           Thrust v1.4.0             #
+#######################################
+
+Summary
+    Thrust v1.4.0 provides support for CUDA 4.0 in addition to many feature
+    and performance improvements.  New set theoretic algorithms operating on
+    sorted sequences have been added.  Additionally, a new fancy iterator
+    allows discarding redundant or otherwise unnecessary output from
+    algorithms, conserving memory storage and bandwidth.
+
+Breaking API Changes
+    Eliminations
+        thrust/is_sorted.h
+        thrust/utility.h
+        thrust/set_intersection.h
+        thrust/experimental/cuda/ogl_interop_allocator.h and the functionality therein
+        thrust::deprecated::copy_when
+        thrust::deprecated::absolute_value
+
+New Features
+    Functions
+        copy_n
+        merge
+        set_difference
+        set_symmetric_difference
+        set_union
+
+    Types
+        discard_iterator
+
+    Device support
+        Compute Capability 2.1 GPUs
+
+New Examples
+    run_length_decoding
+
+Other Enhancements
+    Compilation warnings are substantially reduced in various contexts.
+    The compilation time of thrust::sort, thrust::stable_sort, thrust::sort_by_key,
+    and thrust::stable_sort_by_key are substantially reduced.
+    A fast sort implementation is used when sorting primitive types with thrust::greater.
+    The performance of thrust::set_intersection is improved.
+    The performance of thrust::fill is improved on SM 1.x devices.
+    A code example is now provided in each algorithm's documentation.
+    thrust::reverse now operates in-place
+
+Removed Functionality
+    thrust::deprecated::copy_when
+    thrust::deprecated::absolute_value
+    thrust::experimental::cuda::ogl_interop_allocator
+    thrust::gather and thrust::scatter from host to device and vice versa are no longer supported.
+    Operations which modify the elements of a thrust::device_vector are no longer
+    available from source code compiled without nvcc when the device backend is CUDA.
+    Instead, use the idiom from the cpp_interop example.
+
+Bug Fixes
+    #212 set_intersection works correctly for large input sizes.
+    #275 counting_iterator and constant_iterator work correctly with OpenMP as the
+    backend when compiling with optimization
+    #256 min and max correctly return their first argument as a tie-breaker
+    #248 NDEBUG is interpreted correctly
+
+Known Issues
+    nvcc may generate code containing warnings when compiling some Thrust algorithms.
+    When compiling with -arch=sm_1x, some Thrust algorithms may cause nvcc to issue
+    benign pointer advisories.
+    When compiling with -arch=sm_1x and -G, some Thrust algorithms may fail to execute correctly.
+    thrust::inclusive_scan, thrust::exclusive_scan, thrust::inclusive_scan_by_key,
+    and thrust::exclusive_scan_by_key are currently incompatible with thrust::discard_iterator.
+
+Acknowledgments
+    Thanks to David Tarjan for improving the performance of set_intersection.
+    Thanks to Duane Merrill for continued help with sort.
+    Thanks to Nathan Whitehead for help with CUDA Toolkit integration.
+
+#######################################
+#           Thrust v1.3.0             #
+#######################################
+
+Summary
+    Thrust v1.3.0 provides support for CUDA 3.2 in addition to many feature
+    and performance enhancements.
+    
+    Performance of the sort and sort_by_key algorithms is improved by as much 
+    as 3x in certain situations.  The performance of stream compaction algorithms,
+    such as copy_if, is improved by as much as 2x.  Reduction performance is 
+    also improved, particularly for small input sizes.
+    
+    CUDA errors are now converted to runtime exceptions using the system_error
+    interface.  Combined with a debug mode, also new in v1.3, runtime errors
+    can be located with greater precision.
+
+    Lastly, a few header files have been consolidated or renamed for clarity.
+    See the deprecations section below for additional details.
+
+
+Breaking API Changes
+    Promotions
+        thrust::experimental::inclusive_segmented_scan has been renamed thrust::inclusive_scan_by_key and exposes a different interface
+        thrust::experimental::exclusive_segmented_scan has been renamed thrust::exclusive_scan_by_key and exposes a different interface
+        thrust::experimental::partition_copy has been renamed thrust::partition_copy and exposes a different interface
+        thrust::next::gather has been renamed thrust::gather
+        thrust::next::gather_if has been renamed thrust::gather_if
+        thrust::unique_copy_by_key has been renamed thrust::unique_by_key_copy
+    Deprecations
+        thrust::copy_when has been renamed thrust::deprecated::copy_when
+        thrust::absolute_value has been renamed thrust::deprecated::absolute_value
+        The header thrust/set_intersection.h is now deprecated; use thrust/set_operations.h instead
+        The header thrust/utility.h is now deprecated; use thrust/swap.h instead
+        The header thrust/swap_ranges.h is now deprecated; use thrust/swap.h instead
+    Eliminations
+        thrust::deprecated::gather
+        thrust::deprecated::gather_if
+        thrust/experimental/arch.h and the functions therein
+        thrust/sorting/merge_sort.h
+        thrust/sorting/radix_sort.h
+
+New Features
+    Functions
+        exclusive_scan_by_key
+        find
+        find_if
+        find_if_not
+        inclusive_scan_by_key
+        is_partitioned
+        is_sorted_until
+        mismatch
+        partition_point
+        reverse
+        reverse_copy
+        stable_partition_copy
+
+    Types
+        system_error and related types
+        experimental::cuda::ogl_interop_allocator
+        bit_and, bit_or, and bit_xor
+
+    Device support
+        gf104-based GPUs
+
+New Examples
+    opengl_interop.cu
+    repeated_range.cu
+    simple_moving_average.cu
+    sparse_vector.cu
+    strided_range.cu
+
+Other Enhancements
+    Performance of thrust::sort and thrust::sort_by_key is substantially improved for primitive key types
+    Performance of thrust::copy_if is substantially improved
+    Performance of thrust::reduce and related reductions is improved
+    THRUST_DEBUG mode added
+    Callers of Thrust functions may detect error conditions by catching thrust::system_error, which derives from std::runtime_error
+    The number of compiler warnings generated by Thrust has been substantially reduced
+    Comparison sort now works correctly for input sizes > 32M
+    min & max usage no longer collides with <windows.h> definitions
+    Compiling against the OpenMP backend no longer requires nvcc
+    Performance of device_vector initialized in .cpp files is substantially improved in common cases
+    Performance of thrust::sort_by_key on the host is substantially improved
+
+Removed Functionality
+    nvcc 2.3 is no longer supported
+
+Bug Fixes
+    Debug device code now compiles correctly
+    thrust::uninitialized_copy and thrust::unintialized_fill now dispatch constructors on the device rather than the host
+
+Known Issues
+    #212 set_intersection is known to fail for large input sizes
+    partition_point is known to fail for 64b types with nvcc 3.2
+
+Acknowledgments
+    Thanks to Duane Merrill for contributing a fast CUDA radix sort implementation
+    Thanks to Erich Elsen for contributing an implementation of find_if
+    Thanks to Andrew Corrigan for contributing changes which allow the OpenMP backend to compile in the absence of nvcc
+    Thanks to Andrew Corrigan, Cliff Wooley, David Coeurjolly, Janick Martinez Esturo, John Bowers, Maxim Naumov, Michael Garland, and Ryuta Suzuki for bug reports
+    Thanks to Cliff Woolley for help with testing
+
+#######################################
+#           Thrust v1.2.1             #
+#######################################
+
+Summary
+    Small fixes for compatibility with CUDA 3.1
+
+Known Issues
+    inclusive_scan & exclusive_scan may fail with very large types
+    the Microsoft compiler may fail to compile code using both sort and binary search algorithms
+    uninitialized_fill & uninitialized_copy dispatch constructors on the host rather than the device
+    # 109 some algorithms may exhibit poor performance with the OpenMP backend with large numbers (>= 6) of CPU threads
+    default_random_engine::discard is not accelerated with nvcc 2.3
+    nvcc 3.1 may fail to compile code using types derived from thrust::subtract_with_carry_engine, such as thrust::ranlux24 & thrust::ranlux48.
+
+#######################################
+#           Thrust v1.2.0             #
+#######################################
+
+Summary
+    Thrust v1.2 introduces support for compilation to multicore CPUs
+    and the Ocelot virtual machine, and several new facilities for
+    pseudo-random number generation.  New algorithms such as set
+    intersection and segmented reduction have also been added.  Lastly,
+    improvements to the robustness of the CUDA backend ensure
+    correctness across a broad set of (uncommon) use cases.
+
+Breaking API Changes
+    thrust::gather's interface was incorrect and has been removed.
+    The old interface is deprecated but will be preserved for Thrust
+    version 1.2 at thrust::deprecated::gather &
+    thrust::deprecated::gather_if. The new interface is provided at
+    thrust::next::gather & thrust::next::gather_if.  The new interface
+    will be promoted to thrust:: in Thrust version 1.3. For more details,
+    please refer to this thread:
+    http://groups.google.com/group/thrust-users/browse_thread/thread/f5f0583cb97b51fd
+
+    The thrust::sorting namespace has been deprecated in favor of the
+    top-level sorting functions, such as thrust::sort() and
+    thrust::sort_by_key().
+
+New Features
+    Functions
+        reduce_by_key
+        set_intersection
+        tie
+        unique_copy
+        unique_by_key
+        unique_copy_by_key
+
+    Types
+        Random Number Generation
+            discard_block_engine
+            default_random_engine
+            linear_congruential_engine
+            linear_feedback_shift_engine
+            minstd_rand
+            minstd_rand0
+            normal_distribution (experimental)
+            ranlux24
+            ranlux48
+            ranlux24_base
+            ranlux48_base
+            subtract_with_carry_engine
+            taus88
+            uniform_int_distribution
+            uniform_real_distribution
+            xor_combine_engine
+        Functionals
+            project1st
+            project2nd
+
+    Fancy Iterators
+        permutation_iterator
+        reverse_iterator
+
+    Device support
+        Add support for multicore CPUs via OpenMP
+        Add support for Fermi-class GPUs
+        Add support for Ocelot virtual machine
+
+New Examples
+    cpp_integration
+    histogram
+    mode
+    monte_carlo
+    monte_carlo_disjoint_sequences
+    padded_grid_reduction
+    permutation_iterator
+    row_sum
+    run_length_encoding
+    segmented_scan
+    stream_compaction
+    summary_statistics
+    transform_iterator
+    word_count
+
+Other Enhancements
+    vector functions operator!=, rbegin, crbegin, rend, crend, data, & shrink_to_fit
+    integer sorting performance is improved when max is large but (max - min) is small and when min is negative
+    performance of inclusive_scan() and exclusive_scan() is improved by 20-25% for primitive types
+    support for nvcc 3.0
+
+Removed Functionality
+    removed support for equal between host & device sequences
+    removed support for gather() and scatter() between host & device sequences
+
+Bug Fixes
+    # 8 cause a compiler error if the required compiler is not found rather than a mysterious error at link time
+    # 42 device_ptr & device_reference are classes rather than structs, eliminating warnings on certain platforms
+    # 46 gather & scatter handle any space iterators correctly
+    # 51 thrust::experimental::arch functions gracefully handle unrecognized GPUs
+    # 52 avoid collisions with common user macros such as BLOCK_SIZE
+    # 62 provide better documentation for device_reference
+    # 68 allow built-in CUDA vector types to work with device_vector in pure C++ mode
+    # 102 eliminated a race condition in device_vector::erase
+    various compilation warnings eliminated
+
+Known Issues
+   inclusive_scan & exclusive_scan may fail with very large types
+   the Microsoft compiler may fail to compile code using both sort and binary search algorithms
+   uninitialized_fill & uninitialized_copy dispatch constructors on the host rather than the device
+   # 109 some algorithms may exhibit poor performance with the OpenMP backend with large numbers (>= 6) of CPU threads
+   default_random_engine::discard is not accelerated with nvcc 2.3
+
+Acknowledgments
+   Thanks to Gregory Diamos for contributing a CUDA implementation of set_intersection
+   Thanks to Ryuta Suzuki & Gregory Diamos for rigorously testing Thrust's unit tests and examples against Ocelot
+   Thanks to Tom Bradley for contributing an implementation of normal_distribution
+   Thanks to Joseph Rhoads for contributing the example summary_statistics
+
+#######################################
+#           Thrust v1.1.1             #
+#######################################
+
+Summary
+    Small fixes for compatibility with CUDA 2.3a and Mac OSX Snow Leopard.
+
+#######################################
+#           Thrust v1.1.0             #
+#######################################
+
+Summary
+    Thrust v1.1 introduces fancy iterators, binary search functions, and
+    several specialized reduction functions.  Experimental support for
+    segmented scan has also been added.
+
+Breaking API Changes
+    counting_iterator has been moved into the thrust namespace (previously thrust::experimental)
+
+New Features
+    Functions
+        copy_if
+        lower_bound
+        upper_bound
+        vectorized lower_bound
+        vectorized upper_bound
+        equal_range
+        binary_search
+        vectorized binary_search
+        all_of
+        any_of
+        none_of
+        minmax_element
+        advance
+        inclusive_segmented_scan (experimental)
+        exclusive_segmented_scan (experimental)
+
+    Types
+        pair
+        tuple
+        device_malloc_allocator
+
+    Fancy Iterators
+        constant_iterator
+        counting_iterator
+        transform_iterator
+        zip_iterator
+
+New Examples
+    computing the maximum absolute difference between vectors
+    computing the bounding box of a two-dimensional point set
+    sorting multiple arrays together (lexicographical sorting)
+    constructing a summed area table
+    using zip_iterator to mimic an array of structs
+    using constant_iterator to increment array values
+
+Other Enhancements
+    added pinned memory allocator (experimental)
+    added more methods to host_vector & device_vector (issue #4)
+    added variant of remove_if with a stencil argument (issue #29)
+    scan and reduce use cudaFuncGetAttributes to determine grid size
+    exceptions are reported when temporary device arrays cannot be allocated 
+
+Bug Fixes
+     #5 make vector work for larger data types
+     #9 stable_partition_copy doesn't respect OutputIterator concept semantics
+    #10 scans should return OutputIterator
+    #16 make algorithms work for larger data types
+    #27 dispatch radix_sort even when comp=less<T> is explicitly provided
+
+Known Issues
+    Using functors with Thrust entry points may not compile on Mac OSX with gcc-4.0.1
+    uninitialized_copy & uninitialized_fill dispatch constructors on the host rather than the device.
+    inclusive_scan, inclusive_scan_by_key, exclusive_scan, and exclusive_scan_by_key may fail when used with large types with the CUDA 3.1 driver
+
+
+#######################################
+#           Thrust v1.0.0             #
+#######################################
+
+Breaking API changes
+    Rename top level namespace komrade to thrust.
+    Move partition_copy() & stable_partition_copy() into thrust::experimental namespace until we can easily provide the standard interface.
+    Rename range() to sequence() to avoid collision with Boost.Range.
+    Rename copy_if() to copy_when() due to semantic differences with C++0x copy_if().
+
+New Features
+    Add C++0x style cbegin() & cend() methods to host_vector & device_vector.
+    Add transform_if function.
+    Add stencil versions of replace_if() & replace_copy_if().
+    Allow counting_iterator to work with for_each().
+    Allow types with constructors in comparison sort & reduce.
+
+Other Enhancements
+    merge_sort and stable_merge_sort are now 2 to 5x faster when executed on the parallel device.
+
+Bug fixes
+    Workaround an issue where an incremented iterator causes nvcc to crash. (Komrade issue #6)
+    Fix an issue where const_iterators could not be passed to transform. (Komrade issue #7)
+
diff --git a/compat/thrust/adjacent_difference.h b/compat/thrust/adjacent_difference.h
new file mode 100644
index 0000000000..772b5f993f
--- /dev/null
+++ b/compat/thrust/adjacent_difference.h
@@ -0,0 +1,244 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file adjacent_difference.h
+ *  \brief Compute difference between consecutive elements of a range
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/execution_policy.h>
+
+namespace thrust
+{
+
+
+/*! \addtogroup transformations Transformations
+ *  \{
+ */
+
+
+/*! \p adjacent_difference calculates the differences of adjacent elements in the
+ *  range <tt>[first, last)</tt>. That is, <tt>\*first</tt> is assigned to
+ *  <tt>\*result</tt>, and, for each iterator \p i in the range
+ *  <tt>[first + 1, last)</tt>, the difference of <tt>\*i</tt> and <tt>*(i - 1)</tt>
+ *  is assigned to <tt>\*(result + (i - first))</tt>.
+ *
+ *  This version of \p adjacent_difference uses <tt>operator-</tt> to calculate
+ *  differences.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the input range.
+ *  \param last  The end of the input range.
+ *  \param result The beginning of the output range.
+ *  \return The iterator <tt>result + (last - first)</tt>
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \c x and \c y are objects of \p InputIterator's \c value_type, then \c x - \c is defined,
+ *          and \p InputIterator's \c value_type is convertible to a type in \p OutputIterator's set of \c value_types,
+ *          and the return type of <tt>x - y</tt> is convertible to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *
+ *  \remark Note that \p result is permitted to be the same iterator as \p first. This is
+ *          useful for computing differences "in place".
+ *
+ *  The following code snippet demonstrates how to use \p adjacent_difference to compute
+ *  the difference between adjacent elements of a range using the \p thrust::device execution policy:
+ *
+ *  \code
+ *  #include <thrust/adjacent_difference.h>
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  int h_data[8] = {1, 2, 1, 2, 1, 2, 1, 2};
+ *  thrust::device_vector<int> d_data(h_data, h_data + 8);
+ *  thrust::device_vector<int> d_result(8);
+ *
+ *  thrust::adjacent_difference(thrust::device, d_data.begin(), d_data.end(), d_result.begin());
+ *
+ *  // d_result is now [1, 1, -1, 1, -1, 1, -1, 1]
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/adjacent_difference.html
+ *  \see inclusive_scan
+ */
+template<typename DerivedPolicy, typename InputIterator, typename OutputIterator>
+OutputIterator adjacent_difference(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                   InputIterator first, InputIterator last, 
+                                   OutputIterator result);
+
+/*! \p adjacent_difference calculates the differences of adjacent elements in the
+ *  range <tt>[first, last)</tt>. That is, <tt>*first</tt> is assigned to
+ *  <tt>\*result</tt>, and, for each iterator \p i in the range
+ *  <tt>[first + 1, last)</tt>, <tt>binary_op(\*i, \*(i - 1))</tt> is assigned to
+ *  <tt>\*(result + (i - first))</tt>.
+ *  
+ *  This version of \p adjacent_difference uses the binary function \p binary_op to
+ *  calculate differences.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the input range.
+ *  \param last  The end of the input range.
+ *  \param result The beginning of the output range.
+ *  \param binary_op The binary function used to compute differences.
+ *  \return The iterator <tt>result + (last - first)</tt>
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator's \c value_type is convertible to \p BinaryFunction's \c first_argument_type and \c second_argument_type,
+ *          and \p InputIterator's \c value_type is convertible to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam BinaryFunction's \c result_type is convertible to a type in \p OutputIterator's set of \c value_types.
+ *
+ *  \remark Note that \p result is permitted to be the same iterator as \p first. This is
+ *          useful for computing differences "in place".
+ *
+ *  The following code snippet demonstrates how to use \p adjacent_difference to compute
+ *  the sum between adjacent elements of a range using the \p thrust::device execution policy:
+ *
+ *  \code
+ *  #include <thrust/adjacent_difference.h>
+ *  #include <thrust/functional.h>
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  int h_data[8] = {1, 2, 1, 2, 1, 2, 1, 2};
+ *  thrust::device_vector<int> d_data(h_data, h_data + 8);
+ *  thrust::device_vector<int> d_result(8);
+ *
+ *  thrust::adjacent_difference(thrust::device, d_data.begin(), d_data.end(), d_result.begin(), thrust::plus<int>());
+ *
+ *  // d_data is now [1, 3, 3, 3, 3, 3, 3, 3]
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/adjacent_difference.html
+ *  \see inclusive_scan
+ */
+template<typename DerivedPolicy, typename InputIterator, typename OutputIterator, typename BinaryFunction>
+OutputIterator adjacent_difference(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                   InputIterator first, InputIterator last,
+                                   OutputIterator result,
+                                   BinaryFunction binary_op);
+
+/*! \p adjacent_difference calculates the differences of adjacent elements in the
+ *  range <tt>[first, last)</tt>. That is, <tt>\*first</tt> is assigned to
+ *  <tt>\*result</tt>, and, for each iterator \p i in the range
+ *  <tt>[first + 1, last)</tt>, the difference of <tt>\*i</tt> and <tt>*(i - 1)</tt>
+ *  is assigned to <tt>\*(result + (i - first))</tt>.
+ *
+ *  This version of \p adjacent_difference uses <tt>operator-</tt> to calculate
+ *  differences.
+ *
+ *  \param first The beginning of the input range.
+ *  \param last  The end of the input range.
+ *  \param result The beginning of the output range.
+ *  \return The iterator <tt>result + (last - first)</tt>
+ *
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \c x and \c y are objects of \p InputIterator's \c value_type, then \c x - \c is defined,
+ *          and \p InputIterator's \c value_type is convertible to a type in \p OutputIterator's set of \c value_types,
+ *          and the return type of <tt>x - y</tt> is convertible to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *
+ *  \remark Note that \p result is permitted to be the same iterator as \p first. This is
+ *          useful for computing differences "in place".
+ *
+ *  The following code snippet demonstrates how to use \p adjacent_difference to compute
+ *  the difference between adjacent elements of a range.
+ *
+ *  \code
+ *  #include <thrust/adjacent_difference.h>
+ *  #include <thrust/device_vector.h>
+ *  ...
+ *  int h_data[8] = {1, 2, 1, 2, 1, 2, 1, 2};
+ *  thrust::device_vector<int> d_data(h_data, h_data + 8);
+ *  thrust::device_vector<int> d_result(8);
+ *
+ *  thrust::adjacent_difference(d_data.begin(), d_data.end(), d_result.begin());
+ *
+ *  // d_result is now [1, 1, -1, 1, -1, 1, -1, 1]
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/adjacent_difference.html
+ *  \see inclusive_scan
+ */
+template <typename InputIterator, typename OutputIterator>
+OutputIterator adjacent_difference(InputIterator first, InputIterator last, 
+                                   OutputIterator result);
+
+/*! \p adjacent_difference calculates the differences of adjacent elements in the
+ *  range <tt>[first, last)</tt>. That is, <tt>*first</tt> is assigned to
+ *  <tt>\*result</tt>, and, for each iterator \p i in the range
+ *  <tt>[first + 1, last)</tt>, <tt>binary_op(\*i, \*(i - 1))</tt> is assigned to
+ *  <tt>\*(result + (i - first))</tt>.
+ *  
+ *  This version of \p adjacent_difference uses the binary function \p binary_op to
+ *  calculate differences.
+ *
+ *  \param first The beginning of the input range.
+ *  \param last  The end of the input range.
+ *  \param result The beginning of the output range.
+ *  \param binary_op The binary function used to compute differences.
+ *  \return The iterator <tt>result + (last - first)</tt>
+ *
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator's \c value_type is convertible to \p BinaryFunction's \c first_argument_type and \c second_argument_type,
+ *          and \p InputIterator's \c value_type is convertible to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam BinaryFunction's \c result_type is convertible to a type in \p OutputIterator's set of \c value_types.
+ *
+ *  \remark Note that \p result is permitted to be the same iterator as \p first. This is
+ *          useful for computing differences "in place".
+ *
+ *  The following code snippet demonstrates how to use \p adjacent_difference to compute
+ *  the sum between adjacent elements of a range.
+ *
+ *  \code
+ *  #include <thrust/adjacent_difference.h>
+ *  #include <thrust/functional.h>
+ *  #include <thrust/device_vector.h>
+ *  ...
+ *  int h_data[8] = {1, 2, 1, 2, 1, 2, 1, 2};
+ *  thrust::device_vector<int> d_data(h_data, h_data + 8);
+ *  thrust::device_vector<int> d_result(8);
+ *
+ *  thrust::adjacent_difference(d_data.begin(), d_data.end(), d_result.begin(), thrust::plus<int>());
+ *
+ *  // d_data is now [1, 3, 3, 3, 3, 3, 3, 3]
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/adjacent_difference.html
+ *  \see inclusive_scan
+ */
+template <typename InputIterator, typename OutputIterator, typename BinaryFunction>
+OutputIterator adjacent_difference(InputIterator first, InputIterator last,
+                                   OutputIterator result,
+                                   BinaryFunction binary_op);
+
+/*! \}
+ */
+
+} // end namespace thrust
+
+#include <thrust/detail/adjacent_difference.inl>
+
diff --git a/compat/thrust/advance.h b/compat/thrust/advance.h
new file mode 100644
index 0000000000..e7f60b0d54
--- /dev/null
+++ b/compat/thrust/advance.h
@@ -0,0 +1,73 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file advance.h
+ *  \brief Advance an iterator by a given distance.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+namespace thrust
+{
+
+
+/*! \addtogroup iterators
+ *  \{
+ */
+
+/*! \p advance(i, n) increments the iterator \p i by the distance \p n. 
+ *  If <tt>n > 0</tt> it is equivalent to executing <tt>++i</tt> \p n
+ *  times, and if <tt>n < 0</tt> it is equivalent to executing <tt>--i</tt>
+ *  \p n times. If <tt>n == 0</tt>, the call has no effect.
+ *
+ *  \param i The iterator to be advanced.
+ *  \param n The distance by which to advance the iterator.
+ *
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>.
+ *  \tparam Distance is an integral type that is convertible to \p InputIterator's distance type. 
+ *
+ *  \pre \p n shall be negative only for bidirectional and random access iterators.
+ *
+ *  The following code snippet demonstrates how to use \p advance to increment
+ *  an iterator a given number of times.
+ *
+ *  \code
+ *  #include <thrust/advance.h>
+ *  #include <thrust/device_vector.h>
+ *  ...
+ *  thrust::device_vector<int> vec(13);
+ *  thrust::device_vector<int>::iterator iter = vec.begin();
+ *
+ *  thrust::advance(iter, 7);
+ *
+ *  // iter - vec.begin() == 7
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/advance.html
+ */
+template <typename InputIterator, typename Distance>
+void advance(InputIterator& i, Distance n);
+
+/*! \} // end iterators
+ */
+
+} // end thrust
+
+#include <thrust/detail/advance.inl>
+
diff --git a/compat/thrust/binary_search.h b/compat/thrust/binary_search.h
new file mode 100644
index 0000000000..d2ac5a621e
--- /dev/null
+++ b/compat/thrust/binary_search.h
@@ -0,0 +1,1888 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file binary_search.h
+ *  \brief Search for values in sorted ranges.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/execution_policy.h>
+#include <thrust/pair.h>
+
+namespace thrust
+{
+
+    
+/*! \addtogroup algorithms
+ */
+
+
+/*! \addtogroup searching
+ *  \ingroup algorithms
+ *  \{
+ */
+
+
+/*! \addtogroup binary_search Binary Search
+ *  \ingroup searching
+ *  \{
+ */
+
+
+//////////////////////   
+// Scalar Functions //
+//////////////////////
+
+
+/*! \p lower_bound is a version of binary search: it attempts to find
+ * the element value in an ordered range <tt>[first, last)</tt>. 
+ * Specifically, it returns the first position where value could be
+ * inserted without violating the ordering. This version of 
+ * \p lower_bound uses <tt>operator<</tt> for comparison and returns
+ * the furthermost iterator \c i in <tt>[first, last)</tt> such that,
+ * for every iterator \c j in <tt>[first, i)</tt>, <tt>*j < value</tt>. 
+ *
+ * The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the ordered sequence.
+ *  \param last The end of the ordered sequence.
+ *  \param value The value to be searched.
+ *  \return The furthermost iterator \c i, such that <tt>*i < value</tt>.
+ * 
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator">Forward Iterator</a>.
+ *  \tparam LessThanComparable is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable.html">LessThanComparable</a>. 
+ *
+ *  The following code snippet demonstrates how to use \p lower_bound
+ *  to search for values in a ordered range using the \p thrust::device execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/binary_search.h>
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  thrust::device_vector<int> input(5);
+ *
+ *  input[0] = 0;
+ *  input[1] = 2;
+ *  input[2] = 5;
+ *  input[3] = 7;
+ *  input[4] = 8;
+ *
+ *  thrust::lower_bound(thrust::device, input.begin(), input.end(), 0); // returns input.begin()
+ *  thrust::lower_bound(thrust::device, input.begin(), input.end(), 1); // returns input.begin() + 1
+ *  thrust::lower_bound(thrust::device, input.begin(), input.end(), 2); // returns input.begin() + 1
+ *  thrust::lower_bound(thrust::device, input.begin(), input.end(), 3); // returns input.begin() + 2
+ *  thrust::lower_bound(thrust::device, input.begin(), input.end(), 8); // returns input.begin() + 4
+ *  thrust::lower_bound(thrust::device, input.begin(), input.end(), 9); // returns input.end()
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/lower_bound.html
+ *  \see \p upper_bound
+ *  \see \p equal_range
+ *  \see \p binary_search
+ */
+template<typename DerivedPolicy, typename ForwardIterator, typename LessThanComparable>
+ForwardIterator lower_bound(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                            ForwardIterator first,
+                            ForwardIterator last,
+                            const LessThanComparable &value);
+
+
+/*! \p lower_bound is a version of binary search: it attempts to find
+ * the element value in an ordered range <tt>[first, last)</tt>. 
+ * Specifically, it returns the first position where value could be
+ * inserted without violating the ordering. This version of 
+ * \p lower_bound uses <tt>operator<</tt> for comparison and returns
+ * the furthermost iterator \c i in <tt>[first, last)</tt> such that,
+ * for every iterator \c j in <tt>[first, i)</tt>, <tt>*j < value</tt>. 
+ *
+ *  \param first The beginning of the ordered sequence.
+ *  \param last The end of the ordered sequence.
+ *  \param value The value to be searched.
+ *  \return The furthermost iterator \c i, such that <tt>*i < value</tt>.
+ * 
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator">Forward Iterator</a>.
+ *  \tparam LessThanComparable is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable.html">LessThanComparable</a>. 
+ *
+ *  The following code snippet demonstrates how to use \p lower_bound
+ *  to search for values in a ordered range.
+ *
+ *  \code
+ *  #include <thrust/binary_search.h>
+ *  #include <thrust/device_vector.h>
+ *  ...
+ *  thrust::device_vector<int> input(5);
+ *
+ *  input[0] = 0;
+ *  input[1] = 2;
+ *  input[2] = 5;
+ *  input[3] = 7;
+ *  input[4] = 8;
+ *
+ *  thrust::lower_bound(input.begin(), input.end(), 0); // returns input.begin()
+ *  thrust::lower_bound(input.begin(), input.end(), 1); // returns input.begin() + 1
+ *  thrust::lower_bound(input.begin(), input.end(), 2); // returns input.begin() + 1
+ *  thrust::lower_bound(input.begin(), input.end(), 3); // returns input.begin() + 2
+ *  thrust::lower_bound(input.begin(), input.end(), 8); // returns input.begin() + 4
+ *  thrust::lower_bound(input.begin(), input.end(), 9); // returns input.end()
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/lower_bound.html
+ *  \see \p upper_bound
+ *  \see \p equal_range
+ *  \see \p binary_search
+ */
+template <class ForwardIterator, class LessThanComparable>
+ForwardIterator lower_bound(ForwardIterator first, 
+                            ForwardIterator last,
+                            const LessThanComparable& value);
+
+
+/*! \p lower_bound is a version of binary search: it attempts to find
+ * the element value in an ordered range <tt>[first, last)</tt>. 
+ * Specifically, it returns the first position where value could be
+ * inserted without violating the ordering. This version of 
+ * \p lower_bound uses function object \c comp for comparison 
+ * and returns the furthermost iterator \c i in <tt>[first, last)</tt>
+ * such that, for every iterator \c j in <tt>[first, i)</tt>, 
+ * <tt>comp(*j, value)</tt> is \c true. 
+ *
+ * The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the ordered sequence.
+ *  \param last The end of the ordered sequence.
+ *  \param value The value to be searched.
+ *  \param comp The comparison operator.
+ *  \return The furthermost iterator \c i, such that <tt>comp(*i, value)</tt> is \c true.
+ * 
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator">Forward Iterator</a>.
+ *  \tparam T is comparable to \p ForwardIterator's \c value_type.
+ *  \tparam StrictWeakOrdering is a model of <a href="http://www.sgi.com/tech/stl/StrictWeakOrdering.html">Strict Weak Ordering</a>.
+ *
+ *  The following code snippet demonstrates how to use \p lower_bound
+ *  to search for values in a ordered range using the \p thrust::device execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/binary_search.h>
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/functional.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  thrust::device_vector<int> input(5);
+ *
+ *  input[0] = 0;
+ *  input[1] = 2;
+ *  input[2] = 5;
+ *  input[3] = 7;
+ *  input[4] = 8;
+ *
+ *  thrust::lower_bound(input.begin(), input.end(), 0, thrust::less<int>()); // returns input.begin()
+ *  thrust::lower_bound(input.begin(), input.end(), 1, thrust::less<int>()); // returns input.begin() + 1
+ *  thrust::lower_bound(input.begin(), input.end(), 2, thrust::less<int>()); // returns input.begin() + 1
+ *  thrust::lower_bound(input.begin(), input.end(), 3, thrust::less<int>()); // returns input.begin() + 2
+ *  thrust::lower_bound(input.begin(), input.end(), 8, thrust::less<int>()); // returns input.begin() + 4
+ *  thrust::lower_bound(input.begin(), input.end(), 9, thrust::less<int>()); // returns input.end()
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/lower_bound.html
+ *  \see \p upper_bound
+ *  \see \p equal_range
+ *  \see \p binary_search
+ */
+template<typename DerivedPolicy, typename ForwardIterator, typename T, typename StrictWeakOrdering>
+ForwardIterator lower_bound(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                            ForwardIterator first,
+                            ForwardIterator last,
+                            const T &value,
+                            StrictWeakOrdering comp);
+
+
+/*! \p lower_bound is a version of binary search: it attempts to find
+ * the element value in an ordered range <tt>[first, last)</tt>. 
+ * Specifically, it returns the first position where value could be
+ * inserted without violating the ordering. This version of 
+ * \p lower_bound uses function object \c comp for comparison 
+ * and returns the furthermost iterator \c i in <tt>[first, last)</tt>
+ * such that, for every iterator \c j in <tt>[first, i)</tt>, 
+ * <tt>comp(*j, value)</tt> is \c true. 
+ *
+ *  \param first The beginning of the ordered sequence.
+ *  \param last The end of the ordered sequence.
+ *  \param value The value to be searched.
+ *  \param comp The comparison operator.
+ *  \return The furthermost iterator \c i, such that <tt>comp(*i, value)</tt> is \c true.
+ * 
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator">Forward Iterator</a>.
+ *  \tparam T is comparable to \p ForwardIterator's \c value_type.
+ *  \tparam StrictWeakOrdering is a model of <a href="http://www.sgi.com/tech/stl/StrictWeakOrdering.html">Strict Weak Ordering</a>.
+ *
+ *  The following code snippet demonstrates how to use \p lower_bound
+ *  to search for values in a ordered range.
+ *
+ *  \code
+ *  #include <thrust/binary_search.h>
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/functional.h>
+ *  ...
+ *  thrust::device_vector<int> input(5);
+ *
+ *  input[0] = 0;
+ *  input[1] = 2;
+ *  input[2] = 5;
+ *  input[3] = 7;
+ *  input[4] = 8;
+ *
+ *  thrust::lower_bound(input.begin(), input.end(), 0, thrust::less<int>()); // returns input.begin()
+ *  thrust::lower_bound(input.begin(), input.end(), 1, thrust::less<int>()); // returns input.begin() + 1
+ *  thrust::lower_bound(input.begin(), input.end(), 2, thrust::less<int>()); // returns input.begin() + 1
+ *  thrust::lower_bound(input.begin(), input.end(), 3, thrust::less<int>()); // returns input.begin() + 2
+ *  thrust::lower_bound(input.begin(), input.end(), 8, thrust::less<int>()); // returns input.begin() + 4
+ *  thrust::lower_bound(input.begin(), input.end(), 9, thrust::less<int>()); // returns input.end()
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/lower_bound.html
+ *  \see \p upper_bound
+ *  \see \p equal_range
+ *  \see \p binary_search
+ */
+template <class ForwardIterator, class T, class StrictWeakOrdering>
+ForwardIterator lower_bound(ForwardIterator first,
+                            ForwardIterator last,
+                            const T& value, 
+                            StrictWeakOrdering comp);
+
+
+/*! \p upper_bound is a version of binary search: it attempts to find
+ * the element value in an ordered range <tt>[first, last)</tt>. 
+ * Specifically, it returns the last position where value could be
+ * inserted without violating the ordering. This version of 
+ * \p upper_bound uses <tt>operator<</tt> for comparison and returns
+ * the furthermost iterator \c i in <tt>[first, last)</tt> such that,
+ * for every iterator \c j in <tt>[first, i)</tt>, <tt>value < *j</tt>
+ * is \c false.
+ *
+ * The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the ordered sequence.
+ *  \param last The end of the ordered sequence.
+ *  \param value The value to be searched.
+ *  \return The furthermost iterator \c i, such that <tt>value < *i</tt> is \c false.
+ * 
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator">Forward Iterator</a>.
+ *  \tparam LessThanComparable is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable.html">LessThanComparable</a>. 
+ *
+ *  The following code snippet demonstrates how to use \p upper_bound
+ *  to search for values in a ordered range using the \p thrust::device execution policy for parallelism:
+ *
+ *  \code
+ *  #include <thrust/binary_search.h>
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  thrust::device_vector<int> input(5);
+ *
+ *  input[0] = 0;
+ *  input[1] = 2;
+ *  input[2] = 5;
+ *  input[3] = 7;
+ *  input[4] = 8;
+ *
+ *  thrust::upper_bound(thrust::device, input.begin(), input.end(), 0); // returns input.begin() + 1
+ *  thrust::upper_bound(thrust::device, input.begin(), input.end(), 1); // returns input.begin() + 1
+ *  thrust::upper_bound(thrust::device, input.begin(), input.end(), 2); // returns input.begin() + 2
+ *  thrust::upper_bound(thrust::device, input.begin(), input.end(), 3); // returns input.begin() + 2
+ *  thrust::upper_bound(thrust::device, input.begin(), input.end(), 8); // returns input.end()
+ *  thrust::upper_bound(thrust::device, input.begin(), input.end(), 9); // returns input.end()
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/upper_bound.html
+ *  \see \p lower_bound
+ *  \see \p equal_range
+ *  \see \p binary_search
+ */
+template<typename DerivedPolicy, typename ForwardIterator, typename LessThanComparable>
+ForwardIterator upper_bound(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                            ForwardIterator first,
+                            ForwardIterator last,
+                            const LessThanComparable &value);
+
+
+/*! \p upper_bound is a version of binary search: it attempts to find
+ * the element value in an ordered range <tt>[first, last)</tt>. 
+ * Specifically, it returns the last position where value could be
+ * inserted without violating the ordering. This version of 
+ * \p upper_bound uses <tt>operator<</tt> for comparison and returns
+ * the furthermost iterator \c i in <tt>[first, last)</tt> such that,
+ * for every iterator \c j in <tt>[first, i)</tt>, <tt>value < *j</tt>
+ * is \c false.
+ *
+ *  \param first The beginning of the ordered sequence.
+ *  \param last The end of the ordered sequence.
+ *  \param value The value to be searched.
+ *  \return The furthermost iterator \c i, such that <tt>value < *i</tt> is \c false.
+ * 
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator">Forward Iterator</a>.
+ *  \tparam LessThanComparable is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable.html">LessThanComparable</a>. 
+ *
+ *  The following code snippet demonstrates how to use \p upper_bound
+ *  to search for values in a ordered range.
+ *
+ *  \code
+ *  #include <thrust/binary_search.h>
+ *  #include <thrust/device_vector.h>
+ *  ...
+ *  thrust::device_vector<int> input(5);
+ *
+ *  input[0] = 0;
+ *  input[1] = 2;
+ *  input[2] = 5;
+ *  input[3] = 7;
+ *  input[4] = 8;
+ *
+ *  thrust::upper_bound(input.begin(), input.end(), 0); // returns input.begin() + 1
+ *  thrust::upper_bound(input.begin(), input.end(), 1); // returns input.begin() + 1
+ *  thrust::upper_bound(input.begin(), input.end(), 2); // returns input.begin() + 2
+ *  thrust::upper_bound(input.begin(), input.end(), 3); // returns input.begin() + 2
+ *  thrust::upper_bound(input.begin(), input.end(), 8); // returns input.end()
+ *  thrust::upper_bound(input.begin(), input.end(), 9); // returns input.end()
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/upper_bound.html
+ *  \see \p lower_bound
+ *  \see \p equal_range
+ *  \see \p binary_search
+ */
+template <class ForwardIterator, class LessThanComparable>
+ForwardIterator upper_bound(ForwardIterator first, 
+                            ForwardIterator last,
+                            const LessThanComparable& value);
+
+
+/*! \p upper_bound is a version of binary search: it attempts to find
+ * the element value in an ordered range <tt>[first, last)</tt>. 
+ * Specifically, it returns the last position where value could be
+ * inserted without violating the ordering. This version of 
+ * \p upper_bound uses function object \c comp for comparison and returns
+ * the furthermost iterator \c i in <tt>[first, last)</tt> such that,
+ * for every iterator \c j in <tt>[first, i)</tt>, <tt>comp(value, *j)</tt>
+ * is \c false.
+ *
+ * The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the ordered sequence.
+ *  \param last The end of the ordered sequence.
+ *  \param value The value to be searched.
+ *  \param comp The comparison operator.
+ *  \return The furthermost iterator \c i, such that <tt>comp(value, *i)</tt> is \c false.
+ * 
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator">Forward Iterator</a>.
+ *  \tparam T is comparable to \p ForwardIterator's \c value_type.
+ *  \tparam StrictWeakOrdering is a model of <a href="http://www.sgi.com/tech/stl/StrictWeakOrdering.html">Strict Weak Ordering</a>.
+ *
+ *  The following code snippet demonstrates how to use \p upper_bound
+ *  to search for values in a ordered range using the \p thrust::device execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/binary_search.h>
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/functional.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  thrust::device_vector<int> input(5);
+ *
+ *  input[0] = 0;
+ *  input[1] = 2;
+ *  input[2] = 5;
+ *  input[3] = 7;
+ *  input[4] = 8;
+ *
+ *  thrust::upper_bound(thrust::device, input.begin(), input.end(), 0, thrust::less<int>()); // returns input.begin() + 1
+ *  thrust::upper_bound(thrust::device, input.begin(), input.end(), 1, thrust::less<int>()); // returns input.begin() + 1
+ *  thrust::upper_bound(thrust::device, input.begin(), input.end(), 2, thrust::less<int>()); // returns input.begin() + 2
+ *  thrust::upper_bound(thrust::device, input.begin(), input.end(), 3, thrust::less<int>()); // returns input.begin() + 2
+ *  thrust::upper_bound(thrust::device, input.begin(), input.end(), 8, thrust::less<int>()); // returns input.end()
+ *  thrust::upper_bound(thrust::device, input.begin(), input.end(), 9, thrust::less<int>()); // returns input.end()
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/upper_bound.html
+ *  \see \p lower_bound
+ *  \see \p equal_range
+ *  \see \p binary_search
+ */
+template<typename DerivedPolicy, typename ForwardIterator, typename T, typename StrictWeakOrdering>
+ForwardIterator upper_bound(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                            ForwardIterator first,
+                            ForwardIterator last,
+                            const T &value,
+                            StrictWeakOrdering comp);
+
+/*! \p upper_bound is a version of binary search: it attempts to find
+ * the element value in an ordered range <tt>[first, last)</tt>. 
+ * Specifically, it returns the last position where value could be
+ * inserted without violating the ordering. This version of 
+ * \p upper_bound uses function object \c comp for comparison and returns
+ * the furthermost iterator \c i in <tt>[first, last)</tt> such that,
+ * for every iterator \c j in <tt>[first, i)</tt>, <tt>comp(value, *j)</tt>
+ * is \c false.
+ *
+ *  \param first The beginning of the ordered sequence.
+ *  \param last The end of the ordered sequence.
+ *  \param value The value to be searched.
+ *  \param comp The comparison operator.
+ *  \return The furthermost iterator \c i, such that <tt>comp(value, *i)</tt> is \c false.
+ * 
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator">Forward Iterator</a>.
+ *  \tparam T is comparable to \p ForwardIterator's \c value_type.
+ *  \tparam StrictWeakOrdering is a model of <a href="http://www.sgi.com/tech/stl/StrictWeakOrdering.html">Strict Weak Ordering</a>.
+ *
+ *  The following code snippet demonstrates how to use \p upper_bound
+ *  to search for values in a ordered range.
+ *
+ *  \code
+ *  #include <thrust/binary_search.h>
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/functional.h>
+ *  ...
+ *  thrust::device_vector<int> input(5);
+ *
+ *  input[0] = 0;
+ *  input[1] = 2;
+ *  input[2] = 5;
+ *  input[3] = 7;
+ *  input[4] = 8;
+ *
+ *  thrust::upper_bound(input.begin(), input.end(), 0, thrust::less<int>()); // returns input.begin() + 1
+ *  thrust::upper_bound(input.begin(), input.end(), 1, thrust::less<int>()); // returns input.begin() + 1
+ *  thrust::upper_bound(input.begin(), input.end(), 2, thrust::less<int>()); // returns input.begin() + 2
+ *  thrust::upper_bound(input.begin(), input.end(), 3, thrust::less<int>()); // returns input.begin() + 2
+ *  thrust::upper_bound(input.begin(), input.end(), 8, thrust::less<int>()); // returns input.end()
+ *  thrust::upper_bound(input.begin(), input.end(), 9, thrust::less<int>()); // returns input.end()
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/upper_bound.html
+ *  \see \p lower_bound
+ *  \see \p equal_range
+ *  \see \p binary_search
+ */
+template <class ForwardIterator, class T, class StrictWeakOrdering>
+ForwardIterator upper_bound(ForwardIterator first,
+                            ForwardIterator last,
+                            const T& value, 
+                            StrictWeakOrdering comp);
+
+
+/*! \p binary_search is a version of binary search: it attempts to find
+ * the element value in an ordered range <tt>[first, last)</tt>. 
+ * It returns \c true if an element that is equivalent to \c value 
+ * is present in <tt>[first, last)</tt> and \c false if no such element
+ * exists.  Specifically, this version returns \c true if and only if 
+ * there exists an iterator \c i in <tt>[first, last)</tt> such that 
+ * <tt>*i < value</tt> and <tt>value < *i</tt> are both \c false.
+ *
+ * The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the ordered sequence.
+ *  \param last The end of the ordered sequence.
+ *  \param value The value to be searched.
+ *  \return \c true if an equivalent element exists in <tt>[first, last)</tt>, otherwise \c false.
+ * 
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator">Forward Iterator</a>.
+ *  \tparam LessThanComparable is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable.html">LessThanComparable</a>. 
+ *
+ *  The following code snippet demonstrates how to use \p binary_search
+ *  to search for values in a ordered range using the \p thrust::device execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/binary_search.h>
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  thrust::device_vector<int> input(5);
+ *
+ *  input[0] = 0;
+ *  input[1] = 2;
+ *  input[2] = 5;
+ *  input[3] = 7;
+ *  input[4] = 8;
+ *
+ *  thrust::binary_search(thrust::device, input.begin(), input.end(), 0); // returns true
+ *  thrust::binary_search(thrust::device, input.begin(), input.end(), 1); // returns false
+ *  thrust::binary_search(thrust::device, input.begin(), input.end(), 2); // returns true
+ *  thrust::binary_search(thrust::device, input.begin(), input.end(), 3); // returns false
+ *  thrust::binary_search(thrust::device, input.begin(), input.end(), 8); // returns true
+ *  thrust::binary_search(thrust::device, input.begin(), input.end(), 9); // returns false
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/binary_search.html
+ *  \see \p lower_bound
+ *  \see \p upper_bound
+ *  \see \p equal_range
+ */
+template <typename DerivedPolicy, typename ForwardIterator, typename LessThanComparable>
+bool binary_search(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                   ForwardIterator first, 
+                   ForwardIterator last,
+                   const LessThanComparable& value);
+
+
+/*! \p binary_search is a version of binary search: it attempts to find
+ * the element value in an ordered range <tt>[first, last)</tt>. 
+ * It returns \c true if an element that is equivalent to \c value 
+ * is present in <tt>[first, last)</tt> and \c false if no such element
+ * exists.  Specifically, this version returns \c true if and only if 
+ * there exists an iterator \c i in <tt>[first, last)</tt> such that 
+ * <tt>*i < value</tt> and <tt>value < *i</tt> are both \c false.
+ *
+ *  \param first The beginning of the ordered sequence.
+ *  \param last The end of the ordered sequence.
+ *  \param value The value to be searched.
+ *  \return \c true if an equivalent element exists in <tt>[first, last)</tt>, otherwise \c false.
+ * 
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator">Forward Iterator</a>.
+ *  \tparam LessThanComparable is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable.html">LessThanComparable</a>. 
+ *
+ *  The following code snippet demonstrates how to use \p binary_search
+ *  to search for values in a ordered range.
+ *
+ *  \code
+ *  #include <thrust/binary_search.h>
+ *  #include <thrust/device_vector.h>
+ *  ...
+ *  thrust::device_vector<int> input(5);
+ *
+ *  input[0] = 0;
+ *  input[1] = 2;
+ *  input[2] = 5;
+ *  input[3] = 7;
+ *  input[4] = 8;
+ *
+ *  thrust::binary_search(input.begin(), input.end(), 0); // returns true
+ *  thrust::binary_search(input.begin(), input.end(), 1); // returns false
+ *  thrust::binary_search(input.begin(), input.end(), 2); // returns true
+ *  thrust::binary_search(input.begin(), input.end(), 3); // returns false
+ *  thrust::binary_search(input.begin(), input.end(), 8); // returns true
+ *  thrust::binary_search(input.begin(), input.end(), 9); // returns false
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/binary_search.html
+ *  \see \p lower_bound
+ *  \see \p upper_bound
+ *  \see \p equal_range
+ */
+template <class ForwardIterator, class LessThanComparable>
+bool binary_search(ForwardIterator first, 
+                   ForwardIterator last,
+                   const LessThanComparable& value);
+
+
+/*! \p binary_search is a version of binary search: it attempts to find
+ * the element value in an ordered range <tt>[first, last)</tt>. 
+ * It returns \c true if an element that is equivalent to \c value 
+ * is present in <tt>[first, last)</tt> and \c false if no such element
+ * exists.  Specifically, this version returns \c true if and only if 
+ * there exists an iterator \c i in <tt>[first, last)</tt> such that 
+ * <tt>comp(*i, value)</tt> and <tt>comp(value, *i)</tt> are both \c false.
+ *
+ * The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the ordered sequence.
+ *  \param last The end of the ordered sequence.
+ *  \param value The value to be searched.
+ *  \param comp The comparison operator.
+ *  \return \c true if an equivalent element exists in <tt>[first, last)</tt>, otherwise \c false.
+ * 
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator">Forward Iterator</a>.
+ *  \tparam T is comparable to \p ForwardIterator's \c value_type.
+ *  \tparam StrictWeakOrdering is a model of <a href="http://www.sgi.com/tech/stl/StrictWeakOrdering.html">Strict Weak Ordering</a>.
+ *
+ *  The following code snippet demonstrates how to use \p binary_search
+ *  to search for values in a ordered range using the \p thrust::device execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/binary_search.h>
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/functional.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  thrust::device_vector<int> input(5);
+ *
+ *  input[0] = 0;
+ *  input[1] = 2;
+ *  input[2] = 5;
+ *  input[3] = 7;
+ *  input[4] = 8;
+ *
+ *  thrust::binary_search(thrust::device, input.begin(), input.end(), 0, thrust::less<int>()); // returns true
+ *  thrust::binary_search(thrust::device, input.begin(), input.end(), 1, thrust::less<int>()); // returns false
+ *  thrust::binary_search(thrust::device, input.begin(), input.end(), 2, thrust::less<int>()); // returns true
+ *  thrust::binary_search(thrust::device, input.begin(), input.end(), 3, thrust::less<int>()); // returns false
+ *  thrust::binary_search(thrust::device, input.begin(), input.end(), 8, thrust::less<int>()); // returns true
+ *  thrust::binary_search(thrust::device, input.begin(), input.end(), 9, thrust::less<int>()); // returns false
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/binary_search.html
+ *  \see \p lower_bound
+ *  \see \p upper_bound
+ *  \see \p equal_range
+ */
+template <typename DerivedPolicy, typename ForwardIterator, typename T, typename StrictWeakOrdering>
+bool binary_search(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                   ForwardIterator first,
+                   ForwardIterator last,
+                   const T& value, 
+                   StrictWeakOrdering comp);
+
+
+/*! \p binary_search is a version of binary search: it attempts to find
+ * the element value in an ordered range <tt>[first, last)</tt>. 
+ * It returns \c true if an element that is equivalent to \c value 
+ * is present in <tt>[first, last)</tt> and \c false if no such element
+ * exists.  Specifically, this version returns \c true if and only if 
+ * there exists an iterator \c i in <tt>[first, last)</tt> such that 
+ * <tt>comp(*i, value)</tt> and <tt>comp(value, *i)</tt> are both \c false.
+ *
+ *  \param first The beginning of the ordered sequence.
+ *  \param last The end of the ordered sequence.
+ *  \param value The value to be searched.
+ *  \param comp The comparison operator.
+ *  \return \c true if an equivalent element exists in <tt>[first, last)</tt>, otherwise \c false.
+ * 
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator">Forward Iterator</a>.
+ *  \tparam T is comparable to \p ForwardIterator's \c value_type.
+ *  \tparam StrictWeakOrdering is a model of <a href="http://www.sgi.com/tech/stl/StrictWeakOrdering.html">Strict Weak Ordering</a>.
+ *
+ *  The following code snippet demonstrates how to use \p binary_search
+ *  to search for values in a ordered range.
+ *
+ *  \code
+ *  #include <thrust/binary_search.h>
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/functional.h>
+ *  ...
+ *  thrust::device_vector<int> input(5);
+ *
+ *  input[0] = 0;
+ *  input[1] = 2;
+ *  input[2] = 5;
+ *  input[3] = 7;
+ *  input[4] = 8;
+ *
+ *  thrust::binary_search(input.begin(), input.end(), 0, thrust::less<int>()); // returns true
+ *  thrust::binary_search(input.begin(), input.end(), 1, thrust::less<int>()); // returns false
+ *  thrust::binary_search(input.begin(), input.end(), 2, thrust::less<int>()); // returns true
+ *  thrust::binary_search(input.begin(), input.end(), 3, thrust::less<int>()); // returns false
+ *  thrust::binary_search(input.begin(), input.end(), 8, thrust::less<int>()); // returns true
+ *  thrust::binary_search(input.begin(), input.end(), 9, thrust::less<int>()); // returns false
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/binary_search.html
+ *  \see \p lower_bound
+ *  \see \p upper_bound
+ *  \see \p equal_range
+ */
+template <class ForwardIterator, class T, class StrictWeakOrdering>
+bool binary_search(ForwardIterator first,
+                   ForwardIterator last,
+                   const T& value, 
+                   StrictWeakOrdering comp);
+
+
+/*! \p equal_range is a version of binary search: it attempts to find
+ * the element value in an ordered range <tt>[first, last)</tt>. The 
+ * value returned by \p equal_range is essentially a combination of
+ * the values returned by \p lower_bound and \p upper_bound: it returns
+ * a \p pair of iterators \c i and \c j such that \c i is the first
+ * position where value could be inserted without violating the 
+ * ordering and \c j is the last position where value could be inserted
+ * without violating the ordering. It follows that every element in the
+ * range <tt>[i, j)</tt> is equivalent to value, and that 
+ * <tt>[i, j)</tt> is the largest subrange of <tt>[first, last)</tt> that
+ * has this property. 
+ *
+ * This version of \p equal_range returns a \p pair of iterators 
+ * <tt>[i, j)</tt>, where \c i is the furthermost iterator in 
+ * <tt>[first, last)</tt> such that, for every iterator \c k in 
+ * <tt>[first, i)</tt>, <tt>*k < value</tt>.  \c j is the furthermost
+ * iterator in <tt>[first, last)</tt> such that, for every iterator 
+ * \c k in <tt>[first, j)</tt>, <tt>value < *k</tt> is \c false. 
+ * For every iterator \c k in <tt>[i, j)</tt>, neither 
+ * <tt>value < *k</tt> nor <tt>*k < value</tt> is \c true.
+ *
+ * The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the ordered sequence.
+ *  \param last The end of the ordered sequence.
+ *  \param value The value to be searched.
+ *  \return A \p pair of iterators <tt>[i, j)</tt> that define the range of equivalent elements.
+ * 
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator">Forward Iterator</a>.
+ *  \tparam LessThanComparable is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable.html">LessThanComparable</a>. 
+ *
+ *  The following code snippet demonstrates how to use \p equal_range
+ *  to search for values in a ordered range using the \p thrust::device execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/binary_search.h>
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  thrust::device_vector<int> input(5);
+ *
+ *  input[0] = 0;
+ *  input[1] = 2;
+ *  input[2] = 5;
+ *  input[3] = 7;
+ *  input[4] = 8;
+ *
+ *  thrust::equal_range(thrust::device, input.begin(), input.end(), 0); // returns [input.begin(), input.begin() + 1)
+ *  thrust::equal_range(thrust::device, input.begin(), input.end(), 1); // returns [input.begin() + 1, input.begin() + 1)
+ *  thrust::equal_range(thrust::device, input.begin(), input.end(), 2); // returns [input.begin() + 1, input.begin() + 2)
+ *  thrust::equal_range(thrust::device, input.begin(), input.end(), 3); // returns [input.begin() + 2, input.begin() + 2)
+ *  thrust::equal_range(thrust::device, input.begin(), input.end(), 8); // returns [input.begin() + 4, input.end)
+ *  thrust::equal_range(thrust::device, input.begin(), input.end(), 9); // returns [input.end(), input.end)
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/equal_range.html
+ *  \see \p lower_bound
+ *  \see \p upper_bound
+ *  \see \p binary_search
+ */
+template <typename DerivedPolicy, typename ForwardIterator, typename LessThanComparable>
+thrust::pair<ForwardIterator, ForwardIterator>
+equal_range(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+            ForwardIterator first,
+            ForwardIterator last,
+            const LessThanComparable& value);
+
+
+/*! \p equal_range is a version of binary search: it attempts to find
+ * the element value in an ordered range <tt>[first, last)</tt>. The 
+ * value returned by \p equal_range is essentially a combination of
+ * the values returned by \p lower_bound and \p upper_bound: it returns
+ * a \p pair of iterators \c i and \c j such that \c i is the first
+ * position where value could be inserted without violating the 
+ * ordering and \c j is the last position where value could be inserted
+ * without violating the ordering. It follows that every element in the
+ * range <tt>[i, j)</tt> is equivalent to value, and that 
+ * <tt>[i, j)</tt> is the largest subrange of <tt>[first, last)</tt> that
+ * has this property. 
+ *
+ * This version of \p equal_range returns a \p pair of iterators 
+ * <tt>[i, j)</tt>, where \c i is the furthermost iterator in 
+ * <tt>[first, last)</tt> such that, for every iterator \c k in 
+ * <tt>[first, i)</tt>, <tt>*k < value</tt>.  \c j is the furthermost
+ * iterator in <tt>[first, last)</tt> such that, for every iterator 
+ * \c k in <tt>[first, j)</tt>, <tt>value < *k</tt> is \c false. 
+ * For every iterator \c k in <tt>[i, j)</tt>, neither 
+ * <tt>value < *k</tt> nor <tt>*k < value</tt> is \c true.
+ *
+ *  \param first The beginning of the ordered sequence.
+ *  \param last The end of the ordered sequence.
+ *  \param value The value to be searched.
+ *  \return A \p pair of iterators <tt>[i, j)</tt> that define the range of equivalent elements.
+ * 
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator">Forward Iterator</a>.
+ *  \tparam LessThanComparable is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable.html">LessThanComparable</a>. 
+ *
+ *  The following code snippet demonstrates how to use \p equal_range
+ *  to search for values in a ordered range.
+ *
+ *  \code
+ *  #include <thrust/binary_search.h>
+ *  #include <thrust/device_vector.h>
+ *  ...
+ *  thrust::device_vector<int> input(5);
+ *
+ *  input[0] = 0;
+ *  input[1] = 2;
+ *  input[2] = 5;
+ *  input[3] = 7;
+ *  input[4] = 8;
+ *
+ *  thrust::equal_range(input.begin(), input.end(), 0); // returns [input.begin(), input.begin() + 1)
+ *  thrust::equal_range(input.begin(), input.end(), 1); // returns [input.begin() + 1, input.begin() + 1)
+ *  thrust::equal_range(input.begin(), input.end(), 2); // returns [input.begin() + 1, input.begin() + 2)
+ *  thrust::equal_range(input.begin(), input.end(), 3); // returns [input.begin() + 2, input.begin() + 2)
+ *  thrust::equal_range(input.begin(), input.end(), 8); // returns [input.begin() + 4, input.end)
+ *  thrust::equal_range(input.begin(), input.end(), 9); // returns [input.end(), input.end)
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/equal_range.html
+ *  \see \p lower_bound
+ *  \see \p upper_bound
+ *  \see \p binary_search
+ */
+template <class ForwardIterator, class LessThanComparable>
+thrust::pair<ForwardIterator, ForwardIterator>
+equal_range(ForwardIterator first,
+            ForwardIterator last,
+            const LessThanComparable& value);
+
+
+/*! \p equal_range is a version of binary search: it attempts to find
+ * the element value in an ordered range <tt>[first, last)</tt>. The 
+ * value returned by \p equal_range is essentially a combination of
+ * the values returned by \p lower_bound and \p upper_bound: it returns
+ * a \p pair of iterators \c i and \c j such that \c i is the first
+ * position where value could be inserted without violating the 
+ * ordering and \c j is the last position where value could be inserted
+ * without violating the ordering. It follows that every element in the
+ * range <tt>[i, j)</tt> is equivalent to value, and that 
+ * <tt>[i, j)</tt> is the largest subrange of <tt>[first, last)</tt> that
+ * has this property. 
+ *
+ * This version of \p equal_range returns a \p pair of iterators 
+ * <tt>[i, j)</tt>. \c i is the furthermost iterator in 
+ * <tt>[first, last)</tt> such that, for every iterator \c k in 
+ * <tt>[first, i)</tt>, <tt>comp(*k, value)</tt> is \c true.
+ * \c j is the furthermost iterator in <tt>[first, last)</tt> such
+ * that, for every iterator \c k in <tt>[first, last)</tt>, 
+ * <tt>comp(value, *k)</tt> is \c false. For every iterator \c k 
+ * in <tt>[i, j)</tt>, neither <tt>comp(value, *k)</tt> nor 
+ * <tt>comp(*k, value)</tt> is \c true.
+ *
+ * The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the ordered sequence.
+ *  \param last The end of the ordered sequence.
+ *  \param value The value to be searched.
+ *  \param comp The comparison operator.
+ *  \return A \p pair of iterators <tt>[i, j)</tt> that define the range of equivalent elements.
+ * 
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator">Forward Iterator</a>.
+ *  \tparam T is comparable to \p ForwardIterator's \c value_type.
+ *  \tparam StrictWeakOrdering is a model of <a href="http://www.sgi.com/tech/stl/StrictWeakOrdering.html">Strict Weak Ordering</a>.
+ *
+ *  The following code snippet demonstrates how to use \p equal_range
+ *  to search for values in a ordered range using the \p thrust::device execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/binary_search.h>
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/functional.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  thrust::device_vector<int> input(5);
+ *
+ *  input[0] = 0;
+ *  input[1] = 2;
+ *  input[2] = 5;
+ *  input[3] = 7;
+ *  input[4] = 8;
+ *
+ *  thrust::equal_range(thrust::device, input.begin(), input.end(), 0, thrust::less<int>()); // returns [input.begin(), input.begin() + 1)
+ *  thrust::equal_range(thrust::device, input.begin(), input.end(), 1, thrust::less<int>()); // returns [input.begin() + 1, input.begin() + 1)
+ *  thrust::equal_range(thrust::device, input.begin(), input.end(), 2, thrust::less<int>()); // returns [input.begin() + 1, input.begin() + 2)
+ *  thrust::equal_range(thrust::device, input.begin(), input.end(), 3, thrust::less<int>()); // returns [input.begin() + 2, input.begin() + 2)
+ *  thrust::equal_range(thrust::device, input.begin(), input.end(), 8, thrust::less<int>()); // returns [input.begin() + 4, input.end)
+ *  thrust::equal_range(thrust::device, input.begin(), input.end(), 9, thrust::less<int>()); // returns [input.end(), input.end)
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/equal_range.html
+ *  \see \p lower_bound
+ *  \see \p upper_bound
+ *  \see \p binary_search
+ */
+template <typename DerivedPolicy, typename ForwardIterator, typename T, typename StrictWeakOrdering>
+thrust::pair<ForwardIterator, ForwardIterator>
+equal_range(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+            ForwardIterator first,
+            ForwardIterator last,
+            const T& value,
+            StrictWeakOrdering comp);
+
+
+/*! \p equal_range is a version of binary search: it attempts to find
+ * the element value in an ordered range <tt>[first, last)</tt>. The 
+ * value returned by \p equal_range is essentially a combination of
+ * the values returned by \p lower_bound and \p upper_bound: it returns
+ * a \p pair of iterators \c i and \c j such that \c i is the first
+ * position where value could be inserted without violating the 
+ * ordering and \c j is the last position where value could be inserted
+ * without violating the ordering. It follows that every element in the
+ * range <tt>[i, j)</tt> is equivalent to value, and that 
+ * <tt>[i, j)</tt> is the largest subrange of <tt>[first, last)</tt> that
+ * has this property. 
+ *
+ * This version of \p equal_range returns a \p pair of iterators 
+ * <tt>[i, j)</tt>. \c i is the furthermost iterator in 
+ * <tt>[first, last)</tt> such that, for every iterator \c k in 
+ * <tt>[first, i)</tt>, <tt>comp(*k, value)</tt> is \c true.
+ * \c j is the furthermost iterator in <tt>[first, last)</tt> such
+ * that, for every iterator \c k in <tt>[first, last)</tt>, 
+ * <tt>comp(value, *k)</tt> is \c false. For every iterator \c k 
+ * in <tt>[i, j)</tt>, neither <tt>comp(value, *k)</tt> nor 
+ * <tt>comp(*k, value)</tt> is \c true.
+ *
+ *  \param first The beginning of the ordered sequence.
+ *  \param last The end of the ordered sequence.
+ *  \param value The value to be searched.
+ *  \param comp The comparison operator.
+ *  \return A \p pair of iterators <tt>[i, j)</tt> that define the range of equivalent elements.
+ * 
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator">Forward Iterator</a>.
+ *  \tparam T is comparable to \p ForwardIterator's \c value_type.
+ *  \tparam StrictWeakOrdering is a model of <a href="http://www.sgi.com/tech/stl/StrictWeakOrdering.html">Strict Weak Ordering</a>.
+ *
+ *  The following code snippet demonstrates how to use \p equal_range
+ *  to search for values in a ordered range.
+ *
+ *  \code
+ *  #include <thrust/binary_search.h>
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/functional.h>
+ *  ...
+ *  thrust::device_vector<int> input(5);
+ *
+ *  input[0] = 0;
+ *  input[1] = 2;
+ *  input[2] = 5;
+ *  input[3] = 7;
+ *  input[4] = 8;
+ *
+ *  thrust::equal_range(input.begin(), input.end(), 0, thrust::less<int>()); // returns [input.begin(), input.begin() + 1)
+ *  thrust::equal_range(input.begin(), input.end(), 1, thrust::less<int>()); // returns [input.begin() + 1, input.begin() + 1)
+ *  thrust::equal_range(input.begin(), input.end(), 2, thrust::less<int>()); // returns [input.begin() + 1, input.begin() + 2)
+ *  thrust::equal_range(input.begin(), input.end(), 3, thrust::less<int>()); // returns [input.begin() + 2, input.begin() + 2)
+ *  thrust::equal_range(input.begin(), input.end(), 8, thrust::less<int>()); // returns [input.begin() + 4, input.end)
+ *  thrust::equal_range(input.begin(), input.end(), 9, thrust::less<int>()); // returns [input.end(), input.end)
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/equal_range.html
+ *  \see \p lower_bound
+ *  \see \p upper_bound
+ *  \see \p binary_search
+ */
+template <class ForwardIterator, class T, class StrictWeakOrdering>
+thrust::pair<ForwardIterator, ForwardIterator>
+equal_range(ForwardIterator first,
+            ForwardIterator last,
+            const T& value,
+            StrictWeakOrdering comp);
+
+
+/*! \addtogroup vectorized_binary_search Vectorized Searches
+ *  \ingroup binary_search
+ *  \{
+ */
+
+
+//////////////////////
+// Vector Functions //
+//////////////////////
+
+
+/*! \p lower_bound is a vectorized version of binary search: for each 
+ * iterator \c v in <tt>[values_first, values_last)</tt> it attempts to
+ * find the value <tt>*v</tt> in an ordered range <tt>[first, last)</tt>.
+ * Specifically, it returns the index of first position where value could
+ * be inserted without violating the ordering.
+ *
+ * The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the ordered sequence.
+ *  \param last The end of the ordered sequence.
+ *  \param values_first The beginning of the search values sequence.
+ *  \param values_last The end of the search values sequence.
+ *  \param result The beginning of the output sequence.
+ * 
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator">Forward Iterator</a>.
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>.
+ *                        and \c InputIterator's \c value_type is <a href="http://www.sgi.com/tech/stl/LessThanComparable.html">LessThanComparable</a>.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *                        and \c ForwardIterator's difference_type is convertible to \c OutputIterator's \c value_type.
+ *
+ *  \pre The ranges <tt>[first,last)</tt> and <tt>[result, result + (last - first))</tt> shall not overlap.
+ *
+ *  The following code snippet demonstrates how to use \p lower_bound
+ *  to search for multiple values in a ordered range using the \p thrust::device execution policy for
+ *  parallelization:
+ *
+ *  \code
+ *  #include <thrust/binary_search.h>
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  thrust::device_vector<int> input(5);
+ *
+ *  input[0] = 0;
+ *  input[1] = 2;
+ *  input[2] = 5;
+ *  input[3] = 7;
+ *  input[4] = 8;
+ *
+ *  thrust::device_vector<int> values(6);
+ *  values[0] = 0; 
+ *  values[1] = 1;
+ *  values[2] = 2;
+ *  values[3] = 3;
+ *  values[4] = 8;
+ *  values[5] = 9;
+ *
+ *  thrust::device_vector<unsigned int> output(6);
+ *
+ *  thrust::lower_bound(thrust::device,
+ *                      input.begin(), input.end(),
+ *                      values.begin(), values.end(),
+ *                      output.begin());
+ *
+ *  // output is now [0, 1, 1, 2, 4, 5]
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/lower_bound.html
+ *  \see \p upper_bound
+ *  \see \p equal_range
+ *  \see \p binary_search
+ */
+template <typename DerivedPolicy, typename ForwardIterator, typename InputIterator, typename OutputIterator>
+OutputIterator lower_bound(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                           ForwardIterator first, 
+                           ForwardIterator last,
+                           InputIterator values_first, 
+                           InputIterator values_last,
+                           OutputIterator result);
+
+
+/*! \p lower_bound is a vectorized version of binary search: for each 
+ * iterator \c v in <tt>[values_first, values_last)</tt> it attempts to
+ * find the value <tt>*v</tt> in an ordered range <tt>[first, last)</tt>.
+ * Specifically, it returns the index of first position where value could
+ * be inserted without violating the ordering.
+ *
+ *  \param first The beginning of the ordered sequence.
+ *  \param last The end of the ordered sequence.
+ *  \param values_first The beginning of the search values sequence.
+ *  \param values_last The end of the search values sequence.
+ *  \param result The beginning of the output sequence.
+ * 
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator">Forward Iterator</a>.
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>.
+ *                        and \c InputIterator's \c value_type is <a href="http://www.sgi.com/tech/stl/LessThanComparable.html">LessThanComparable</a>.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *                        and \c ForwardIterator's difference_type is convertible to \c OutputIterator's \c value_type.
+ *
+ *  \pre The ranges <tt>[first,last)</tt> and <tt>[result, result + (last - first))</tt> shall not overlap.
+ *
+ *  The following code snippet demonstrates how to use \p lower_bound
+ *  to search for multiple values in a ordered range.
+ *
+ *  \code
+ *  #include <thrust/binary_search.h>
+ *  #include <thrust/device_vector.h>
+ *  ...
+ *  thrust::device_vector<int> input(5);
+ *
+ *  input[0] = 0;
+ *  input[1] = 2;
+ *  input[2] = 5;
+ *  input[3] = 7;
+ *  input[4] = 8;
+ *
+ *  thrust::device_vector<int> values(6);
+ *  values[0] = 0; 
+ *  values[1] = 1;
+ *  values[2] = 2;
+ *  values[3] = 3;
+ *  values[4] = 8;
+ *  values[5] = 9;
+ *
+ *  thrust::device_vector<unsigned int> output(6);
+ *
+ *  thrust::lower_bound(input.begin(), input.end(),
+ *                      values.begin(), values.end(),
+ *                      output.begin());
+ *
+ *  // output is now [0, 1, 1, 2, 4, 5]
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/lower_bound.html
+ *  \see \p upper_bound
+ *  \see \p equal_range
+ *  \see \p binary_search
+ */
+template <class ForwardIterator, class InputIterator, class OutputIterator>
+OutputIterator lower_bound(ForwardIterator first, 
+                           ForwardIterator last,
+                           InputIterator values_first, 
+                           InputIterator values_last,
+                           OutputIterator result);
+
+
+/*! \p lower_bound is a vectorized version of binary search: for each 
+ * iterator \c v in <tt>[values_first, values_last)</tt> it attempts to
+ * find the value <tt>*v</tt> in an ordered range <tt>[first, last)</tt>.
+ * Specifically, it returns the index of first position where value could
+ * be inserted without violating the ordering.  This version of 
+ * \p lower_bound uses function object \c comp for comparison.
+ *
+ * The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the ordered sequence.
+ *  \param last The end of the ordered sequence.
+ *  \param values_first The beginning of the search values sequence.
+ *  \param values_last The end of the search values sequence.
+ *  \param result The beginning of the output sequence.
+ *  \param comp The comparison operator.
+ * 
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator">Forward Iterator</a>.
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>.
+ *                        and \c InputIterator's \c value_type is comparable to \p ForwardIterator's \c value_type.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *                        and \c ForwardIterator's difference_type is convertible to \c OutputIterator's \c value_type.
+ *  \tparam StrictWeakOrdering is a model of <a href="http://www.sgi.com/tech/stl/StrictWeakOrdering.html">Strict Weak Ordering</a>.
+ *
+ *  \pre The ranges <tt>[first,last)</tt> and <tt>[result, result + (last - first))</tt> shall not overlap.
+ *
+ *  The following code snippet demonstrates how to use \p lower_bound
+ *  to search for multiple values in a ordered range.
+ *
+ *  \code
+ *  #include <thrust/binary_search.h>
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/functional.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  thrust::device_vector<int> input(5);
+ *
+ *  input[0] = 0;
+ *  input[1] = 2;
+ *  input[2] = 5;
+ *  input[3] = 7;
+ *  input[4] = 8;
+ *
+ *  thrust::device_vector<int> values(6);
+ *  values[0] = 0; 
+ *  values[1] = 1;
+ *  values[2] = 2;
+ *  values[3] = 3;
+ *  values[4] = 8;
+ *  values[5] = 9;
+ *
+ *  thrust::device_vector<unsigned int> output(6);
+ *
+ *  thrust::lower_bound(input.begin(), input.end(),
+ *                      values.begin(), values.end(), 
+ *                      output.begin(),
+ *                      thrust::less<int>());
+ *
+ *  // output is now [0, 1, 1, 2, 4, 5]
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/lower_bound.html
+ *  \see \p upper_bound
+ *  \see \p equal_range
+ *  \see \p binary_search
+ */
+template <typename DerivedPolicy, typename ForwardIterator, typename InputIterator, typename OutputIterator, typename StrictWeakOrdering>
+OutputIterator lower_bound(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                           ForwardIterator first, 
+                           ForwardIterator last,
+                           InputIterator values_first, 
+                           InputIterator values_last,
+                           OutputIterator result,
+                           StrictWeakOrdering comp);
+
+
+/*! \p lower_bound is a vectorized version of binary search: for each 
+ * iterator \c v in <tt>[values_first, values_last)</tt> it attempts to
+ * find the value <tt>*v</tt> in an ordered range <tt>[first, last)</tt>.
+ * Specifically, it returns the index of first position where value could
+ * be inserted without violating the ordering.  This version of 
+ * \p lower_bound uses function object \c comp for comparison.
+ *
+ *  \param first The beginning of the ordered sequence.
+ *  \param last The end of the ordered sequence.
+ *  \param values_first The beginning of the search values sequence.
+ *  \param values_last The end of the search values sequence.
+ *  \param result The beginning of the output sequence.
+ *  \param comp The comparison operator.
+ * 
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator">Forward Iterator</a>.
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>.
+ *                        and \c InputIterator's \c value_type is comparable to \p ForwardIterator's \c value_type.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *                        and \c ForwardIterator's difference_type is convertible to \c OutputIterator's \c value_type.
+ *  \tparam StrictWeakOrdering is a model of <a href="http://www.sgi.com/tech/stl/StrictWeakOrdering.html">Strict Weak Ordering</a>.
+ *
+ *  \pre The ranges <tt>[first,last)</tt> and <tt>[result, result + (last - first))</tt> shall not overlap.
+ *
+ *  The following code snippet demonstrates how to use \p lower_bound
+ *  to search for multiple values in a ordered range.
+ *
+ *  \code
+ *  #include <thrust/binary_search.h>
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/functional.h>
+ *  ...
+ *  thrust::device_vector<int> input(5);
+ *
+ *  input[0] = 0;
+ *  input[1] = 2;
+ *  input[2] = 5;
+ *  input[3] = 7;
+ *  input[4] = 8;
+ *
+ *  thrust::device_vector<int> values(6);
+ *  values[0] = 0; 
+ *  values[1] = 1;
+ *  values[2] = 2;
+ *  values[3] = 3;
+ *  values[4] = 8;
+ *  values[5] = 9;
+ *
+ *  thrust::device_vector<unsigned int> output(6);
+ *
+ *  thrust::lower_bound(input.begin(), input.end(),
+ *                      values.begin(), values.end(), 
+ *                      output.begin(),
+ *                      thrust::less<int>());
+ *
+ *  // output is now [0, 1, 1, 2, 4, 5]
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/lower_bound.html
+ *  \see \p upper_bound
+ *  \see \p equal_range
+ *  \see \p binary_search
+ */
+template <class ForwardIterator, class InputIterator, class OutputIterator, class StrictWeakOrdering>
+OutputIterator lower_bound(ForwardIterator first, 
+                           ForwardIterator last,
+                           InputIterator values_first, 
+                           InputIterator values_last,
+                           OutputIterator result,
+                           StrictWeakOrdering comp);
+
+
+/*! \p upper_bound is a vectorized version of binary search: for each 
+ * iterator \c v in <tt>[values_first, values_last)</tt> it attempts to
+ * find the value <tt>*v</tt> in an ordered range <tt>[first, last)</tt>.
+ * Specifically, it returns the index of last position where value could
+ * be inserted without violating the ordering.
+ *
+ * The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the ordered sequence.
+ *  \param last The end of the ordered sequence.
+ *  \param values_first The beginning of the search values sequence.
+ *  \param values_last The end of the search values sequence.
+ *  \param result The beginning of the output sequence.
+ * 
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator">Forward Iterator</a>.
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>.
+ *                        and \c InputIterator's \c value_type is <a href="http://www.sgi.com/tech/stl/LessThanComparable.html">LessThanComparable</a>.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *                        and \c ForwardIterator's difference_type is convertible to \c OutputIterator's \c value_type.
+ *
+ *  \pre The ranges <tt>[first,last)</tt> and <tt>[result, result + (last - first))</tt> shall not overlap.
+ *
+ *  The following code snippet demonstrates how to use \p lower_bound
+ *  to search for multiple values in a ordered range using the \p thrust::device execution policy for
+ *  parallelization:
+ *
+ *  \code
+ *  #include <thrust/binary_search.h>
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  thrust::device_vector<int> input(5);
+ *
+ *  input[0] = 0;
+ *  input[1] = 2;
+ *  input[2] = 5;
+ *  input[3] = 7;
+ *  input[4] = 8;
+ *
+ *  thrust::device_vector<int> values(6);
+ *  values[0] = 0; 
+ *  values[1] = 1;
+ *  values[2] = 2;
+ *  values[3] = 3;
+ *  values[4] = 8;
+ *  values[5] = 9;
+ *
+ *  thrust::device_vector<unsigned int> output(6);
+ *
+ *  thrust::upper_bound(thrust::device,
+ *                      input.begin(), input.end(),
+ *                      values.begin(), values.end(),
+ *                      output.begin());
+ *
+ *  // output is now [1, 1, 2, 2, 5, 5]
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/upper_bound.html
+ *  \see \p upper_bound
+ *  \see \p equal_range
+ *  \see \p binary_search
+ */
+template <typename DerivedPolicy, typename ForwardIterator, typename InputIterator, typename OutputIterator>
+OutputIterator upper_bound(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                           ForwardIterator first, 
+                           ForwardIterator last,
+                           InputIterator values_first, 
+                           InputIterator values_last,
+                           OutputIterator result);
+
+
+/*! \p upper_bound is a vectorized version of binary search: for each 
+ * iterator \c v in <tt>[values_first, values_last)</tt> it attempts to
+ * find the value <tt>*v</tt> in an ordered range <tt>[first, last)</tt>.
+ * Specifically, it returns the index of last position where value could
+ * be inserted without violating the ordering.
+ *
+ *  \param first The beginning of the ordered sequence.
+ *  \param last The end of the ordered sequence.
+ *  \param values_first The beginning of the search values sequence.
+ *  \param values_last The end of the search values sequence.
+ *  \param result The beginning of the output sequence.
+ * 
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator">Forward Iterator</a>.
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>.
+ *                        and \c InputIterator's \c value_type is <a href="http://www.sgi.com/tech/stl/LessThanComparable.html">LessThanComparable</a>.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *                        and \c ForwardIterator's difference_type is convertible to \c OutputIterator's \c value_type.
+ *
+ *  \pre The ranges <tt>[first,last)</tt> and <tt>[result, result + (last - first))</tt> shall not overlap.
+ *
+ *  The following code snippet demonstrates how to use \p lower_bound
+ *  to search for multiple values in a ordered range.
+ *
+ *  \code
+ *  #include <thrust/binary_search.h>
+ *  #include <thrust/device_vector.h>
+ *  ...
+ *  thrust::device_vector<int> input(5);
+ *
+ *  input[0] = 0;
+ *  input[1] = 2;
+ *  input[2] = 5;
+ *  input[3] = 7;
+ *  input[4] = 8;
+ *
+ *  thrust::device_vector<int> values(6);
+ *  values[0] = 0; 
+ *  values[1] = 1;
+ *  values[2] = 2;
+ *  values[3] = 3;
+ *  values[4] = 8;
+ *  values[5] = 9;
+ *
+ *  thrust::device_vector<unsigned int> output(6);
+ *
+ *  thrust::upper_bound(input.begin(), input.end(),
+ *                      values.begin(), values.end(),
+ *                      output.begin());
+ *
+ *  // output is now [1, 1, 2, 2, 5, 5]
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/upper_bound.html
+ *  \see \p upper_bound
+ *  \see \p equal_range
+ *  \see \p binary_search
+ */
+template <class ForwardIterator, class InputIterator, class OutputIterator>
+OutputIterator upper_bound(ForwardIterator first, 
+                           ForwardIterator last,
+                           InputIterator values_first, 
+                           InputIterator values_last,
+                           OutputIterator result);
+
+
+/*! \p upper_bound is a vectorized version of binary search: for each 
+ * iterator \c v in <tt>[values_first, values_last)</tt> it attempts to
+ * find the value <tt>*v</tt> in an ordered range <tt>[first, last)</tt>.
+ * Specifically, it returns the index of first position where value could
+ * be inserted without violating the ordering.  This version of 
+ * \p upper_bound uses function object \c comp for comparison.
+ *
+ * The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the ordered sequence.
+ *  \param last The end of the ordered sequence.
+ *  \param values_first The beginning of the search values sequence.
+ *  \param values_last The end of the search values sequence.
+ *  \param result The beginning of the output sequence.
+ *  \param comp The comparison operator.
+ * 
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator">Forward Iterator</a>.
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>.
+ *                        and \c InputIterator's \c value_type is comparable to \p ForwardIterator's \c value_type.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *                        and \c ForwardIterator's difference_type is convertible to \c OutputIterator's \c value_type.
+ *  \tparam StrictWeakOrdering is a model of <a href="http://www.sgi.com/tech/stl/StrictWeakOrdering.html">Strict Weak Ordering</a>.
+ *
+ *  \pre The ranges <tt>[first,last)</tt> and <tt>[result, result + (last - first))</tt> shall not overlap.
+ *
+ *  The following code snippet demonstrates how to use \p lower_bound
+ *  to search for multiple values in a ordered range using the \p thrust::device execution policy for
+ *  parallelization:
+ *
+ *  \code
+ *  #include <thrust/binary_search.h>
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/functional.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  thrust::device_vector<int> input(5);
+ *
+ *  input[0] = 0;
+ *  input[1] = 2;
+ *  input[2] = 5;
+ *  input[3] = 7;
+ *  input[4] = 8;
+ *
+ *  thrust::device_vector<int> values(6);
+ *  values[0] = 0; 
+ *  values[1] = 1;
+ *  values[2] = 2;
+ *  values[3] = 3;
+ *  values[4] = 8;
+ *  values[5] = 9;
+ *
+ *  thrust::device_vector<unsigned int> output(6);
+ *
+ *  thrust::upper_bound(thrust::device,
+ *                      input.begin(), input.end(),
+ *                      values.begin(), values.end(), 
+ *                      output.begin(),
+ *                      thrust::less<int>());
+ *
+ *  // output is now [1, 1, 2, 2, 5, 5]
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/upper_bound.html
+ *  \see \p lower_bound
+ *  \see \p equal_range
+ *  \see \p binary_search
+ */
+template <typename DerivedPolicy, typename ForwardIterator, typename InputIterator, typename OutputIterator, typename StrictWeakOrdering>
+OutputIterator upper_bound(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                           ForwardIterator first, 
+                           ForwardIterator last,
+                           InputIterator values_first, 
+                           InputIterator values_last,
+                           OutputIterator result,
+                           StrictWeakOrdering comp);
+
+
+/*! \p upper_bound is a vectorized version of binary search: for each 
+ * iterator \c v in <tt>[values_first, values_last)</tt> it attempts to
+ * find the value <tt>*v</tt> in an ordered range <tt>[first, last)</tt>.
+ * Specifically, it returns the index of first position where value could
+ * be inserted without violating the ordering.  This version of 
+ * \p upper_bound uses function object \c comp for comparison.
+ *
+ *  \param first The beginning of the ordered sequence.
+ *  \param last The end of the ordered sequence.
+ *  \param values_first The beginning of the search values sequence.
+ *  \param values_last The end of the search values sequence.
+ *  \param result The beginning of the output sequence.
+ *  \param comp The comparison operator.
+ * 
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator">Forward Iterator</a>.
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>.
+ *                        and \c InputIterator's \c value_type is comparable to \p ForwardIterator's \c value_type.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *                        and \c ForwardIterator's difference_type is convertible to \c OutputIterator's \c value_type.
+ *  \tparam StrictWeakOrdering is a model of <a href="http://www.sgi.com/tech/stl/StrictWeakOrdering.html">Strict Weak Ordering</a>.
+ *
+ *  \pre The ranges <tt>[first,last)</tt> and <tt>[result, result + (last - first))</tt> shall not overlap.
+ *
+ *  The following code snippet demonstrates how to use \p lower_bound
+ *  to search for multiple values in a ordered range.
+ *
+ *  \code
+ *  #include <thrust/binary_search.h>
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/functional.h>
+ *  ...
+ *  thrust::device_vector<int> input(5);
+ *
+ *  input[0] = 0;
+ *  input[1] = 2;
+ *  input[2] = 5;
+ *  input[3] = 7;
+ *  input[4] = 8;
+ *
+ *  thrust::device_vector<int> values(6);
+ *  values[0] = 0; 
+ *  values[1] = 1;
+ *  values[2] = 2;
+ *  values[3] = 3;
+ *  values[4] = 8;
+ *  values[5] = 9;
+ *
+ *  thrust::device_vector<unsigned int> output(6);
+ *
+ *  thrust::upper_bound(input.begin(), input.end(),
+ *                      values.begin(), values.end(), 
+ *                      output.begin(),
+ *                      thrust::less<int>());
+ *
+ *  // output is now [1, 1, 2, 2, 5, 5]
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/upper_bound.html
+ *  \see \p lower_bound
+ *  \see \p equal_range
+ *  \see \p binary_search
+ */
+template <class ForwardIterator, class InputIterator, class OutputIterator, class StrictWeakOrdering>
+OutputIterator upper_bound(ForwardIterator first, 
+                           ForwardIterator last,
+                           InputIterator values_first, 
+                           InputIterator values_last,
+                           OutputIterator result,
+                           StrictWeakOrdering comp);
+
+
+/*! \p binary_search is a vectorized version of binary search: for each 
+ * iterator \c v in <tt>[values_first, values_last)</tt> it attempts to
+ * find the value <tt>*v</tt> in an ordered range <tt>[first, last)</tt>.
+ * It returns \c true if an element that is equivalent to \c value 
+ * is present in <tt>[first, last)</tt> and \c false if no such element
+ * exists.
+ *
+ * The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the ordered sequence.
+ *  \param last The end of the ordered sequence.
+ *  \param values_first The beginning of the search values sequence.
+ *  \param values_last The end of the search values sequence.
+ *  \param result The beginning of the output sequence.
+ * 
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator">Forward Iterator</a>.
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>.
+ *                        and \c InputIterator's \c value_type is <a href="http://www.sgi.com/tech/stl/LessThanComparable.html">LessThanComparable</a>.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *                        and bool is convertible to \c OutputIterator's \c value_type.
+ *
+ *  \pre The ranges <tt>[first,last)</tt> and <tt>[result, result + (last - first))</tt> shall not overlap.
+ *
+ *  The following code snippet demonstrates how to use \p binary_search
+ *  to search for multiple values in a ordered range using the \p thrust::device execution policy for
+ *  parallelization:
+ *
+ *  \code
+ *  #include <thrust/binary_search.h>
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  thrust::device_vector<int> input(5);
+ *
+ *  input[0] = 0;
+ *  input[1] = 2;
+ *  input[2] = 5;
+ *  input[3] = 7;
+ *  input[4] = 8;
+ *
+ *  thrust::device_vector<int> values(6);
+ *  values[0] = 0; 
+ *  values[1] = 1;
+ *  values[2] = 2;
+ *  values[3] = 3;
+ *  values[4] = 8;
+ *  values[5] = 9;
+ *
+ *  thrust::device_vector<bool> output(6);
+ *
+ *  thrust::binary_search(thrust::device,
+ *                        input.begin(), input.end(),
+ *                        values.begin(), values.end(),
+ *                        output.begin());
+ *
+ *  // output is now [true, false, true, false, true, false]
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/binary_search.html
+ *  \see \p lower_bound
+ *  \see \p upper_bound
+ *  \see \p equal_range
+ */
+template <typename DerivedPolicy, typename ForwardIterator, typename InputIterator, typename OutputIterator>
+OutputIterator binary_search(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                             ForwardIterator first, 
+                             ForwardIterator last,
+                             InputIterator values_first, 
+                             InputIterator values_last,
+                             OutputIterator result);
+
+
+/*! \p binary_search is a vectorized version of binary search: for each 
+ * iterator \c v in <tt>[values_first, values_last)</tt> it attempts to
+ * find the value <tt>*v</tt> in an ordered range <tt>[first, last)</tt>.
+ * It returns \c true if an element that is equivalent to \c value 
+ * is present in <tt>[first, last)</tt> and \c false if no such element
+ * exists.
+ *
+ *  \param first The beginning of the ordered sequence.
+ *  \param last The end of the ordered sequence.
+ *  \param values_first The beginning of the search values sequence.
+ *  \param values_last The end of the search values sequence.
+ *  \param result The beginning of the output sequence.
+ * 
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator">Forward Iterator</a>.
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>.
+ *                        and \c InputIterator's \c value_type is <a href="http://www.sgi.com/tech/stl/LessThanComparable.html">LessThanComparable</a>.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *                        and bool is convertible to \c OutputIterator's \c value_type.
+ *
+ *  \pre The ranges <tt>[first,last)</tt> and <tt>[result, result + (last - first))</tt> shall not overlap.
+ *
+ *  The following code snippet demonstrates how to use \p binary_search
+ *  to search for multiple values in a ordered range.
+ *
+ *  \code
+ *  #include <thrust/binary_search.h>
+ *  #include <thrust/device_vector.h>
+ *  ...
+ *  thrust::device_vector<int> input(5);
+ *
+ *  input[0] = 0;
+ *  input[1] = 2;
+ *  input[2] = 5;
+ *  input[3] = 7;
+ *  input[4] = 8;
+ *
+ *  thrust::device_vector<int> values(6);
+ *  values[0] = 0; 
+ *  values[1] = 1;
+ *  values[2] = 2;
+ *  values[3] = 3;
+ *  values[4] = 8;
+ *  values[5] = 9;
+ *
+ *  thrust::device_vector<bool> output(6);
+ *
+ *  thrust::binary_search(input.begin(), input.end(),
+ *                        values.begin(), values.end(),
+ *                        output.begin());
+ *
+ *  // output is now [true, false, true, false, true, false]
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/binary_search.html
+ *  \see \p lower_bound
+ *  \see \p upper_bound
+ *  \see \p equal_range
+ */
+template <class ForwardIterator, class InputIterator, class OutputIterator>
+OutputIterator binary_search(ForwardIterator first, 
+                             ForwardIterator last,
+                             InputIterator values_first, 
+                             InputIterator values_last,
+                             OutputIterator result);
+
+
+/*! \p binary_search is a vectorized version of binary search: for each 
+ * iterator \c v in <tt>[values_first, values_last)</tt> it attempts to
+ * find the value <tt>*v</tt> in an ordered range <tt>[first, last)</tt>.
+ * It returns \c true if an element that is equivalent to \c value 
+ * is present in <tt>[first, last)</tt> and \c false if no such element
+ * exists.  This version of \p binary_search uses function object 
+ * \c comp for comparison.
+ *
+ * The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the ordered sequence.
+ *  \param last The end of the ordered sequence.
+ *  \param values_first The beginning of the search values sequence.
+ *  \param values_last The end of the search values sequence.
+ *  \param result The beginning of the output sequence.
+ *  \param comp The comparison operator.
+ * 
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator">Forward Iterator</a>.
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>.
+ *                        and \c InputIterator's \c value_type is <a href="http://www.sgi.com/tech/stl/LessThanComparable.html">LessThanComparable</a>.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *                        and bool is convertible to \c OutputIterator's \c value_type.
+ *  \tparam StrictWeakOrdering is a model of <a href="http://www.sgi.com/tech/stl/StrictWeakOrdering.html">Strict Weak Ordering</a>.
+ *
+ *  \pre The ranges <tt>[first,last)</tt> and <tt>[result, result + (last - first))</tt> shall not overlap.
+ *
+ *  The following code snippet demonstrates how to use \p binary_search
+ *  to search for multiple values in a ordered range using the \p thrust::device execution policy for
+ *  parallelization:
+ *
+ *  \code
+ *  #include <thrust/binary_search.h>
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/functional.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  thrust::device_vector<int> input(5);
+ *
+ *  input[0] = 0;
+ *  input[1] = 2;
+ *  input[2] = 5;
+ *  input[3] = 7;
+ *  input[4] = 8;
+ *
+ *  thrust::device_vector<int> values(6);
+ *  values[0] = 0; 
+ *  values[1] = 1;
+ *  values[2] = 2;
+ *  values[3] = 3;
+ *  values[4] = 8;
+ *  values[5] = 9;
+ *
+ *  thrust::device_vector<bool> output(6);
+ *
+ *  thrust::binary_search(thrust::device,
+ *                        input.begin(), input.end(),
+ *                        values.begin(), values.end(),
+ *                        output.begin(),
+ *                        thrust::less<T>());
+ *
+ *  // output is now [true, false, true, false, true, false]
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/binary_search.html
+ *  \see \p lower_bound
+ *  \see \p upper_bound
+ *  \see \p equal_range
+ */
+template <typename DerivedPolicy, typename ForwardIterator, typename InputIterator, typename OutputIterator, typename StrictWeakOrdering>
+OutputIterator binary_search(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                             ForwardIterator first, 
+                             ForwardIterator last,
+                             InputIterator values_first, 
+                             InputIterator values_last,
+                             OutputIterator result,
+                             StrictWeakOrdering comp);
+
+
+/*! \p binary_search is a vectorized version of binary search: for each 
+ * iterator \c v in <tt>[values_first, values_last)</tt> it attempts to
+ * find the value <tt>*v</tt> in an ordered range <tt>[first, last)</tt>.
+ * It returns \c true if an element that is equivalent to \c value 
+ * is present in <tt>[first, last)</tt> and \c false if no such element
+ * exists.  This version of \p binary_search uses function object 
+ * \c comp for comparison.
+ *
+ *  \param first The beginning of the ordered sequence.
+ *  \param last The end of the ordered sequence.
+ *  \param values_first The beginning of the search values sequence.
+ *  \param values_last The end of the search values sequence.
+ *  \param result The beginning of the output sequence.
+ *  \param comp The comparison operator.
+ * 
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator">Forward Iterator</a>.
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>.
+ *                        and \c InputIterator's \c value_type is <a href="http://www.sgi.com/tech/stl/LessThanComparable.html">LessThanComparable</a>.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *                        and bool is convertible to \c OutputIterator's \c value_type.
+ *  \tparam StrictWeakOrdering is a model of <a href="http://www.sgi.com/tech/stl/StrictWeakOrdering.html">Strict Weak Ordering</a>.
+ *
+ *  \pre The ranges <tt>[first,last)</tt> and <tt>[result, result + (last - first))</tt> shall not overlap.
+ *
+ *  The following code snippet demonstrates how to use \p binary_search
+ *  to search for multiple values in a ordered range.
+ *
+ *  \code
+ *  #include <thrust/binary_search.h>
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/functional.h>
+ *  ...
+ *  thrust::device_vector<int> input(5);
+ *
+ *  input[0] = 0;
+ *  input[1] = 2;
+ *  input[2] = 5;
+ *  input[3] = 7;
+ *  input[4] = 8;
+ *
+ *  thrust::device_vector<int> values(6);
+ *  values[0] = 0; 
+ *  values[1] = 1;
+ *  values[2] = 2;
+ *  values[3] = 3;
+ *  values[4] = 8;
+ *  values[5] = 9;
+ *
+ *  thrust::device_vector<bool> output(6);
+ *
+ *  thrust::binary_search(input.begin(), input.end(),
+ *                        values.begin(), values.end(),
+ *                        output.begin(),
+ *                        thrust::less<T>());
+ *
+ *  // output is now [true, false, true, false, true, false]
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/binary_search.html
+ *  \see \p lower_bound
+ *  \see \p upper_bound
+ *  \see \p equal_range
+ */
+template <class ForwardIterator, class InputIterator, class OutputIterator, class StrictWeakOrdering>
+OutputIterator binary_search(ForwardIterator first, 
+                             ForwardIterator last,
+                             InputIterator values_first, 
+                             InputIterator values_last,
+                             OutputIterator result,
+                             StrictWeakOrdering comp);
+
+
+/*! \} // end vectorized_binary_search
+ */
+
+
+/*! \} // end binary_search
+ */
+
+
+/*! \} // end searching
+ */
+
+
+} // end namespace thrust
+
+#include <thrust/detail/binary_search.inl>
+
diff --git a/compat/thrust/copy.h b/compat/thrust/copy.h
new file mode 100644
index 0000000000..eaa9719459
--- /dev/null
+++ b/compat/thrust/copy.h
@@ -0,0 +1,505 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file copy.h
+ *  \brief Copies elements from one range to another
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/execution_policy.h>
+
+namespace thrust
+{
+
+/*! \addtogroup algorithms
+ */
+
+/*! \addtogroup copying
+ *  \ingroup algorithms
+ *  \{
+ */
+
+
+/*! \p copy copies elements from the range [\p first, \p last) to the range
+ *  [\p result, \p result + (\p last - \p first)). That is, it performs
+ *  the assignments *\p result = *\p first, *(\p result + \c 1) = *(\p first + \c 1),
+ *  and so on. Generally, for every integer \c n from \c 0 to \p last - \p first, \p copy
+ *  performs the assignment *(\p result + \c n) = *(\p first + \c n). Unlike
+ *  \c std::copy, \p copy offers no guarantee on order of operation.  As a result,
+ *  calling \p copy with overlapping source and destination ranges has undefined
+ *  behavior.
+ *
+ *  The return value is \p result + (\p last - \p first).
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the sequence to copy.
+ *  \param last The end of the sequence to copy.
+ *  \param result The destination sequence.
+ *  \return The end of the destination sequence.
+ *  \see http://www.sgi.com/tech/stl/copy.html
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator must be a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a> and \c InputIterator's \c value_type must be convertible to \c OutputIterator's \c value_type.
+ *  \tparam OutputIterator must be a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *
+ *  \pre \p result may be equal to \p first, but \p result shall not be in the range <tt>[first, last)</tt> otherwise.
+ *
+ *  The following code snippet demonstrates how to use \p copy
+ *  to copy from one range to another using the \p thrust::device parallelization policy:
+ *
+ *  \code
+ *  #include <thrust/copy.h>
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *
+ *  thrust::device_vector<int> vec0(100);
+ *  thrust::device_vector<int> vec1(100);
+ *  ...
+ *
+ *  thrust::copy(thrust::device, vec0.begin(), vec0.end(), vec1.begin());
+ *
+ *  // vec1 is now a copy of vec0
+ *  \endcode
+ */
+template<typename DerivedPolicy, typename InputIterator, typename OutputIterator>
+  OutputIterator copy(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                      InputIterator first,
+                      InputIterator last,
+                      OutputIterator result);
+
+
+/*! \p copy_n copies elements from the range <tt>[first, first + n)</tt> to the range
+ *  <tt>[result, result + n)</tt>. That is, it performs the assignments <tt>*result = *first, *(result + 1) = *(first + 1)</tt>,
+ *  and so on. Generally, for every integer \c i from \c 0 to \c n, \p copy
+ *  performs the assignment *(\p result + \c i) = *(\p first + \c i). Unlike
+ *  \c std::copy_n, \p copy_n offers no guarantee on order of operation. As a result,
+ *  calling \p copy_n with overlapping source and destination ranges has undefined
+ *  behavior.
+ *
+ *  The return value is \p result + \p n.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the range to copy.
+ *  \param n The number of elements to copy.
+ *  \param result The beginning destination range.
+ *  \return The end of the destination range.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator must be a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a> and \c InputIterator's \c value_type must be convertible to \c OutputIterator's \c value_type.
+ *  \tparam Size is an integral type.
+ *  \tparam OutputIterator must be a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *
+ *  \pre \p result may be equal to \p first, but \p result shall not be in the range <tt>[first, first + n)</tt> otherwise.
+ *
+ *  The following code snippet demonstrates how to use \p copy
+ *  to copy from one range to another using the \p thrust::device parallelization policy:
+ *
+ *  \code
+ *  #include <thrust/copy.h>
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  size_t n = 100;
+ *  thrust::device_vector<int> vec0(n);
+ *  thrust::device_vector<int> vec1(n);
+ *  ...
+ *  thrust::copy_n(thrust::device, vec0.begin(), n, vec1.begin());
+ *
+ *  // vec1 is now a copy of vec0
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/copy_n.html
+ *  \see thrust::copy
+ */
+template<typename DerivedPolicy, typename InputIterator, typename Size, typename OutputIterator>
+  OutputIterator copy_n(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                        InputIterator first,
+                        Size n,
+                        OutputIterator result);
+
+
+	
+/*! \p copy copies elements from the range [\p first, \p last) to the range
+ *  [\p result, \p result + (\p last - \p first)). That is, it performs
+ *  the assignments *\p result = *\p first, *(\p result + \c 1) = *(\p first + \c 1),
+ *  and so on. Generally, for every integer \c n from \c 0 to \p last - \p first, \p copy
+ *  performs the assignment *(\p result + \c n) = *(\p first + \c n). Unlike
+ *  \c std::copy, \p copy offers no guarantee on order of operation.  As a result,
+ *  calling \p copy with overlapping source and destination ranges has undefined
+ *  behavior.
+ *
+ *  The return value is \p result + (\p last - \p first).
+ *
+ *  \param first The beginning of the sequence to copy.
+ *  \param last The end of the sequence to copy.
+ *  \param result The destination sequence.
+ *  \return The end of the destination sequence.
+ *  \see http://www.sgi.com/tech/stl/copy.html
+ *
+ *  \tparam InputIterator must be a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a> and \c InputIterator's \c value_type must be convertible to \c OutputIterator's \c value_type.
+ *  \tparam OutputIterator must be a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *
+ *  \pre \p result may be equal to \p first, but \p result shall not be in the range <tt>[first, last)</tt> otherwise.
+ *
+ *  The following code snippet demonstrates how to use \p copy
+ *  to copy from one range to another.
+ *
+ *  \code
+ *  #include <thrust/copy.h>
+ *  #include <thrust/device_vector.h>
+ *  ...
+ *
+ *  thrust::device_vector<int> vec0(100);
+ *  thrust::device_vector<int> vec1(100);
+ *  ...
+ *
+ *  thrust::copy(vec0.begin(), vec0.end(),
+ *               vec1.begin());
+ *
+ *  // vec1 is now a copy of vec0
+ *  \endcode
+ */
+template<typename InputIterator, typename OutputIterator>
+  OutputIterator copy(InputIterator first,
+                      InputIterator last,
+                      OutputIterator result);
+
+/*! \p copy_n copies elements from the range <tt>[first, first + n)</tt> to the range
+ *  <tt>[result, result + n)</tt>. That is, it performs the assignments <tt>*result = *first, *(result + 1) = *(first + 1)</tt>,
+ *  and so on. Generally, for every integer \c i from \c 0 to \c n, \p copy
+ *  performs the assignment *(\p result + \c i) = *(\p first + \c i). Unlike
+ *  \c std::copy_n, \p copy_n offers no guarantee on order of operation. As a result,
+ *  calling \p copy_n with overlapping source and destination ranges has undefined
+ *  behavior.
+ *
+ *  The return value is \p result + \p n.
+ *
+ *  \param first The beginning of the range to copy.
+ *  \param n The number of elements to copy.
+ *  \param result The beginning destination range.
+ *  \return The end of the destination range.
+ *
+ *  \tparam InputIterator must be a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a> and \c InputIterator's \c value_type must be convertible to \c OutputIterator's \c value_type.
+ *  \tparam Size is an integral type.
+ *  \tparam OutputIterator must be a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *
+ *  \pre \p result may be equal to \p first, but \p result shall not be in the range <tt>[first, first + n)</tt> otherwise.
+ *
+ *  The following code snippet demonstrates how to use \p copy
+ *  to copy from one range to another.
+ *
+ *  \code
+ *  #include <thrust/copy.h>
+ *  #include <thrust/device_vector.h>
+ *  ...
+ *  size_t n = 100;
+ *  thrust::device_vector<int> vec0(n);
+ *  thrust::device_vector<int> vec1(n);
+ *  ...
+ *  thrust::copy_n(vec0.begin(), n, vec1.begin());
+ *
+ *  // vec1 is now a copy of vec0
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/copy_n.html
+ *  \see thrust::copy
+ */
+template<typename InputIterator, typename Size, typename OutputIterator>
+  OutputIterator copy_n(InputIterator first,
+                        Size n,
+                        OutputIterator result);
+
+/*! \} // end copying
+ */
+
+/*! \addtogroup stream_compaction
+ *  \{
+ */
+
+
+/*! This version of \p copy_if copies elements from the range <tt>[first,last)</tt>
+ *  to a range beginning at \ presult, except that any element which causes \p pred
+ *  to be \p pred to be \c false is not copied.
+ *
+ *  More precisely, for every integer \c n such that <tt>0 <= n < last-first</tt>,
+ *  \p copy_if performs the assignment <tt>*result = *(first+n)</tt> and \p result
+ *  is advanced one position if <tt>pred(*(first+n))</tt>. Otherwise, no assignment
+ *  occurs and \p result is not advanced.
+ *
+ *  The algorithm's execution is parallelized as determined by \p system.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the sequence from which to copy.
+ *  \param last The end of the sequence from which to copy.
+ *  \param result The beginning of the sequence into which to copy.
+ *  \param pred The predicate to test on every value of the range <tt>[first, last)</tt>.
+ *  \return <tt>result + n</tt>, where \c n is equal to the number of times \p pred
+ *          evaluated to \c true in the range <tt>[first, last)</tt>.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *                        and \p InputIterator's \c value_type is convertible to \p Predicate's \c argument_type.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam Predicate is a model of <a href="http://www.sgi.com/tech/stl/Predicate.html">Predicate</a>.
+ *
+ *  \pre The ranges <tt>[first, last)</tt> and <tt>[result, result + (last - first))</tt> shall not overlap.
+ *
+ *  The following code snippet demonstrates how to use \p copy_if to perform stream compaction
+ *  to copy even numbers to an output range using the \p thrust::host parallelization policy:
+ *
+ *  \code
+ *  #include <thrust/copy.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  struct is_even
+ *  {
+ *    __host__ __device__
+ *    bool operator()(const int x)
+ *    {
+ *      return (x % 2) == 0;
+ *    }
+ *  };
+ *  ...
+ *  const int N = 6;
+ *  int V[N] = {-2, 0, -1, 0, 1, 2};
+ *  int result[4];
+ *
+ *  thrust::copy_if(thrust::host, V, V + N, result, is_even());
+ *
+ *  // V remains {-2, 0, -1, 0, 1, 2}
+ *  // result is now {-2, 0, 0, 2}
+ *  \endcode
+ *
+ *  \see \c remove_copy_if
+ */
+template<typename DerivedPolicy, typename InputIterator, typename OutputIterator, typename Predicate>
+  OutputIterator copy_if(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                         InputIterator first,
+                         InputIterator last,
+                         OutputIterator result,
+                         Predicate pred);
+
+
+
+/*! This version of \p copy_if copies elements from the range <tt>[first,last)</tt>
+ *  to a range beginning at \ presult, except that any element which causes \p pred
+ *  to be \p pred to be \c false is not copied.
+ *
+ *  More precisely, for every integer \c n such that <tt>0 <= n < last-first</tt>,
+ *  \p copy_if performs the assignment <tt>*result = *(first+n)</tt> and \p result
+ *  is advanced one position if <tt>pred(*(first+n))</tt>. Otherwise, no assignment
+ *  occurs and \p result is not advanced.
+ *
+ *  \param first The beginning of the sequence from which to copy.
+ *  \param last The end of the sequence from which to copy.
+ *  \param result The beginning of the sequence into which to copy.
+ *  \param pred The predicate to test on every value of the range <tt>[first, last)</tt>.
+ *  \return <tt>result + n</tt>, where \c n is equal to the number of times \p pred
+ *          evaluated to \c true in the range <tt>[first, last)</tt>.
+ *
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *                        and \p InputIterator's \c value_type is convertible to \p Predicate's \c argument_type.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam Predicate is a model of <a href="http://www.sgi.com/tech/stl/Predicate.html">Predicate</a>.
+ *
+ *  \pre The ranges <tt>[first, last)</tt> and <tt>[result, result + (last - first))</tt> shall not overlap.
+ *
+ *  The following code snippet demonstrates how to use \p copy_if to perform stream compaction
+ *  to copy even numbers to an output range.
+ *
+ *  \code
+ *  #include <thrust/copy.h>
+ *  ...
+ *  struct is_even
+ *  {
+ *    __host__ __device__
+ *    bool operator()(const int x)
+ *    {
+ *      return (x % 2) == 0;
+ *    }
+ *  };
+ *  ...
+ *  const int N = 6;
+ *  int V[N] = {-2, 0, -1, 0, 1, 2};
+ *  int result[4];
+ *
+ *  thrust::copy_if(V, V + N, result, is_even());
+ *
+ *  // V remains {-2, 0, -1, 0, 1, 2}
+ *  // result is now {-2, 0, 0, 2}
+ *  \endcode
+ *
+ *  \see \c remove_copy_if
+ */
+template<typename InputIterator,
+         typename OutputIterator,
+         typename Predicate>
+  OutputIterator copy_if(InputIterator first,
+                         InputIterator last,
+                         OutputIterator result,
+                         Predicate pred);
+
+
+/*! This version of \p copy_if copies elements from the range <tt>[first,last)</tt>
+ *  to a range beginning at \p result, except that any element whose corresponding stencil
+ *  element causes \p pred to be \c false is not copied.
+ *
+ *  More precisely, for every integer \c n such that <tt>0 <= n < last-first</tt>,
+ *  \p copy_if performs the assignment <tt>*result = *(first+n)</tt> and \p result
+ *  is advanced one position if <tt>pred(*(stencil+n))</tt>. Otherwise, no assignment
+ *  occurs and \p result is not advanced.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the sequence from which to copy.
+ *  \param last The end of the sequence from which to copy.
+ *  \param stencil The beginning of the stencil sequence.
+ *  \param result The beginning of the sequence into which to copy.
+ *  \param pred The predicate to test on every value of the range <tt>[stencil, stencil + (last-first))</tt>.
+ *  \return <tt>result + n</tt>, where \c n is equal to the number of times \p pred
+ *          evaluated to \c true in the range <tt>[stencil, stencil + (last-first))</tt>.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *                         and \p InputIterator2's \c value_type is convertible to \p Predicate's \c argument_type.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator">Output Iterator</a>.
+ *  \tparam Predicate is a model of <a href="http://www.sgi.com/tech/stl/Predicate.html">Predicate</a>.
+ *
+ *  \pre The ranges <tt>[first, last)</tt> and <tt>[result, result + (last - first))</tt> shall not overlap.
+ *  \pre The ranges <tt>[stencil, stencil + (last - first))</tt> and <tt>[result, result + (last - first))</tt> shall not overlap.
+ *
+ *  The following code snippet demonstrates how to use \p copy_if to perform stream compaction
+ *  to copy numbers to an output range when corresponding stencil elements are even using the \p thrust::host execution policy:
+ *
+ *  \code
+ *  #include <thrust/copy.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  struct is_even
+ *  {
+ *    __host__ __device__
+ *    bool operator()(const int x)
+ *    {
+ *      return (x % 2) == 0;
+ *    }
+ *  };
+ *  ...
+ *  int N = 6;
+ *  int data[N]    = { 0, 1,  2, 3, 4, 5};
+ *  int stencil[N] = {-2, 0, -1, 0, 1, 2};
+ *  int result[4];
+ *
+ *  thrust::copy_if(thrust::host, data, data + N, stencil, result, is_even());
+ *
+ *  // data remains    = { 0, 1,  2, 3, 4, 5};
+ *  // stencil remains = {-2, 0, -1, 0, 1, 2};
+ *  // result is now     { 0, 1,  3, 5}
+ *  \endcode
+ *
+ *  \see \c remove_copy_if
+ */
+template<typename DerivedPolicy, typename InputIterator1, typename InputIterator2, typename OutputIterator, typename Predicate>
+  OutputIterator copy_if(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                         InputIterator1 first,
+                         InputIterator1 last,
+                         InputIterator2 stencil,
+                         OutputIterator result,
+                         Predicate pred);
+
+
+/*! This version of \p copy_if copies elements from the range <tt>[first,last)</tt>
+ *  to a range beginning at \p result, except that any element whose corresponding stencil
+ *  element causes \p pred to be \c false is not copied.
+ *
+ *  More precisely, for every integer \c n such that <tt>0 <= n < last-first</tt>,
+ *  \p copy_if performs the assignment <tt>*result = *(first+n)</tt> and \p result
+ *  is advanced one position if <tt>pred(*(stencil+n))</tt>. Otherwise, no assignment
+ *  occurs and \p result is not advanced.
+ *
+ *  \param first The beginning of the sequence from which to copy.
+ *  \param last The end of the sequence from which to copy.
+ *  \param stencil The beginning of the stencil sequence.
+ *  \param result The beginning of the sequence into which to copy.
+ *  \param pred The predicate to test on every value of the range <tt>[stencil, stencil + (last-first))</tt>.
+ *  \return <tt>result + n</tt>, where \c n is equal to the number of times \p pred
+ *          evaluated to \c true in the range <tt>[stencil, stencil + (last-first))</tt>.
+ *
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *                         and \p InputIterator2's \c value_type is convertible to \p Predicate's \c argument_type.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator">Output Iterator</a>.
+ *  \tparam Predicate is a model of <a href="http://www.sgi.com/tech/stl/Predicate.html">Predicate</a>.
+ *
+ *  \pre The ranges <tt>[first, last)</tt> and <tt>[result, result + (last - first))</tt> shall not overlap.
+ *  \pre The ranges <tt>[stencil, stencil + (last - first))</tt> and <tt>[result, result + (last - first))</tt> shall not overlap.
+ *
+ *  The following code snippet demonstrates how to use \p copy_if to perform stream compaction
+ *  to copy numbers to an output range when corresponding stencil elements are even:
+ *
+ *  \code
+ *  #include <thrust/copy.h>
+ *  ...
+ *  struct is_even
+ *  {
+ *    __host__ __device__
+ *    bool operator()(const int x)
+ *    {
+ *      return (x % 2) == 0;
+ *    }
+ *  };
+ *  ...
+ *  int N = 6;
+ *  int data[N]    = { 0, 1,  2, 3, 4, 5};
+ *  int stencil[N] = {-2, 0, -1, 0, 1, 2};
+ *  int result[4];
+ *
+ *  thrust::copy_if(data, data + N, stencil, result, is_even());
+ *
+ *  // data remains    = { 0, 1,  2, 3, 4, 5};
+ *  // stencil remains = {-2, 0, -1, 0, 1, 2};
+ *  // result is now     { 0, 1,  3, 5}
+ *  \endcode
+ *
+ *  \see \c remove_copy_if
+ */
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename Predicate>
+  OutputIterator copy_if(InputIterator1 first,
+                         InputIterator1 last,
+                         InputIterator2 stencil,
+                         OutputIterator result,
+                         Predicate pred);
+
+/*! \} // end stream_compaction
+ */
+	
+} // end namespace thrust
+
+#include <thrust/detail/copy.h>
+#include <thrust/detail/copy_if.h>
+
diff --git a/compat/thrust/count.h b/compat/thrust/count.h
new file mode 100644
index 0000000000..cddd1dd68b
--- /dev/null
+++ b/compat/thrust/count.h
@@ -0,0 +1,231 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file count.h
+ *  \brief Counting elements in a range
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/execution_policy.h>
+#include <thrust/iterator/iterator_traits.h>
+
+namespace thrust
+{
+
+
+/*! \addtogroup algorithms
+ */
+
+/*! \addtogroup reductions
+ *  \ingroup algorithms
+ *  \{
+ */
+
+/*! \addtogroup counting
+ *  \ingroup reductions
+ *  \{
+ */
+
+
+/*! \p count finds the number of elements in <tt>[first,last)</tt> that are equal
+ *  to \p value. More precisely, \p count returns the number of iterators \c i in
+ *  <tt>[first, last)</tt> such that <tt>*i == value</tt>.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the sequence.
+ *  \param last The end of the sequence.
+ *  \param value The value to be counted.
+ *  \return The number of elements equal to \p value.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator must be a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a> and \c InputIterator's \c value_type must be a model of must be a model of <a href="http://www.sgi.com/tech/stl/EqualityComparable.html">Equality Comparable</a>.
+ *  \tparam EqualityComparable must be a model of <a href="http://www.sgi.com/tech/stl/EqualityComparable.html">Equality Comparable</a> and can be compared for equality with \c InputIterator's \c value_type
+ *
+ *  The following code snippet demonstrates how to use \p count to 
+ *  count the number of instances in a range of a value of interest using the \p thrust::device execution policy:
+ *
+ *  \code
+ *  #include <thrust/count.h>
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  // put 3 1s in a device_vector
+ *  thrust::device_vector<int> vec(5,0);
+ *  vec[1] = 1;
+ *  vec[3] = 1;
+ *  vec[4] = 1;
+ *  
+ *  // count the 1s
+ *  int result = thrust::count(thrust::device, vec.begin(), vec.end(), 1);
+ *  // result == 3
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/count.html
+ */
+template<typename DerivedPolicy, typename InputIterator, typename EqualityComparable>
+  typename thrust::iterator_traits<InputIterator>::difference_type
+    count(const thrust::detail::execution_policy_base<DerivedPolicy> &exec, InputIterator first, InputIterator last, const EqualityComparable& value);
+
+
+
+/*! \p count finds the number of elements in <tt>[first,last)</tt> that are equal
+ *  to \p value. More precisely, \p count returns the number of iterators \c i in
+ *  <tt>[first, last)</tt> such that <tt>*i == value</tt>.
+ *
+ *  \param first The beginning of the sequence.
+ *  \param last The end of the sequence.
+ *  \param value The value to be counted.
+ *  \return The number of elements equal to \p value.
+ *
+ *  \tparam InputIterator must be a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a> and \c InputIterator's \c value_type must be a model of must be a model of <a href="http://www.sgi.com/tech/stl/EqualityComparable.html">Equality Comparable</a>.
+ *  \tparam EqualityComparable must be a model of <a href="http://www.sgi.com/tech/stl/EqualityComparable.html">Equality Comparable</a> and can be compared for equality with \c InputIterator's \c value_type
+ *
+ *  The following code snippet demonstrates how to use \p count to 
+ *  count the number of instances in a range of a value of interest.
+ *  \code
+ *  #include <thrust/count.h>
+ *  #include <thrust/device_vector.h>
+ *  ...
+ *  // put 3 1s in a device_vector
+ *  thrust::device_vector<int> vec(5,0);
+ *  vec[1] = 1;
+ *  vec[3] = 1;
+ *  vec[4] = 1;
+ *  
+ *  // count the 1s
+ *  int result = thrust::count(vec.begin(), vec.end(), 1);
+ *  // result == 3
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/count.html
+ */
+template <typename InputIterator, typename EqualityComparable>
+  typename thrust::iterator_traits<InputIterator>::difference_type
+    count(InputIterator first, InputIterator last, const EqualityComparable& value);
+
+
+/*! \p count_if finds the number of elements in <tt>[first,last)</tt> for which 
+ *  a predicate is \c true. More precisely, \p count_if returns the number of iterators
+ *  \c i in <tt>[first, last)</tt> such that <tt>pred(*i) == true</tt>.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the sequence.
+ *  \param last The end of the sequence.
+ *  \param pred The predicate.
+ *  \return The number of elements where \p pred is \c true.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator must be a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a> and \c InputIterator's \c value_type must be convertible to \c Predicate's \c argument_type.
+ *  \tparam Predicate must be a model of <a href="http://www.sgi.com/tech/stl/Predicate.html">Predicate</a>.
+ *
+ *  The following code snippet demonstrates how to use \p count to
+ *  count the number of odd numbers in a range using the \p thrust::device execution policy:
+ *
+ *  \code
+ *  #include <thrust/count.h>
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  struct is_odd
+ *  {
+ *    __host__ __device__
+ *    bool operator()(int &x)
+ *    {
+ *      return x & 1;
+ *    }
+ *  };
+ *  ...
+ *  // fill a device_vector with even & odd numbers
+ *  thrust::device_vector<int> vec(5);
+ *  vec[0] = 0;
+ *  vec[1] = 1;
+ *  vec[2] = 2;
+ *  vec[3] = 3;
+ *  vec[4] = 4;
+ *
+ *  // count the odd elements in vec
+ *  int result = thrust::count_if(thrust::device, vec.begin(), vec.end(), is_odd());
+ *  // result == 2
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/count.html
+ */
+template<typename DerivedPolicy, typename InputIterator, typename Predicate>
+  typename thrust::iterator_traits<InputIterator>::difference_type
+    count_if(const thrust::detail::execution_policy_base<DerivedPolicy> &exec, InputIterator first, InputIterator last, Predicate pred);
+
+
+/*! \p count_if finds the number of elements in <tt>[first,last)</tt> for which 
+ *  a predicate is \c true. More precisely, \p count_if returns the number of iterators
+ *  \c i in <tt>[first, last)</tt> such that <tt>pred(*i) == true</tt>.
+ *
+ *  \param first The beginning of the sequence.
+ *  \param last The end of the sequence.
+ *  \param pred The predicate.
+ *  \return The number of elements where \p pred is \c true.
+ *
+ *  \tparam InputIterator must be a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a> and \c InputIterator's \c value_type must be convertible to \c Predicate's \c argument_type.
+ *  \tparam Predicate must be a model of <a href="http://www.sgi.com/tech/stl/Predicate.html">Predicate</a>.
+ *
+ *  The following code snippet demonstrates how to use \p count to
+ *  count the number of odd numbers in a range.
+ *  \code
+ *  #include <thrust/count.h>
+ *  #include <thrust/device_vector.h>
+ *  ...
+ *  struct is_odd
+ *  {
+ *    __host__ __device__
+ *    bool operator()(int &x)
+ *    {
+ *      return x & 1;
+ *    }
+ *  };
+ *  ...
+ *  // fill a device_vector with even & odd numbers
+ *  thrust::device_vector<int> vec(5);
+ *  vec[0] = 0;
+ *  vec[1] = 1;
+ *  vec[2] = 2;
+ *  vec[3] = 3;
+ *  vec[4] = 4;
+ *
+ *  // count the odd elements in vec
+ *  int result = thrust::count_if(vec.begin(), vec.end(), is_odd());
+ *  // result == 2
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/count.html
+ */
+template <typename InputIterator, typename Predicate>
+  typename thrust::iterator_traits<InputIterator>::difference_type
+    count_if(InputIterator first, InputIterator last, Predicate pred);
+
+/*! \} // end counting
+ *  \} // end reductions
+ */
+
+} // end thrust
+
+#include <thrust/detail/count.inl>
+
diff --git a/compat/thrust/detail/adjacent_difference.inl b/compat/thrust/detail/adjacent_difference.inl
new file mode 100644
index 0000000000..6590f9d15d
--- /dev/null
+++ b/compat/thrust/detail/adjacent_difference.inl
@@ -0,0 +1,88 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file adjacent_difference.inl
+ *  \brief Inline file for adjacent_difference.h
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/select_system.h>
+#include <thrust/system/detail/generic/adjacent_difference.h>
+#include <thrust/system/detail/adl/adjacent_difference.h>
+
+namespace thrust
+{
+
+
+template <typename DerivedPolicy, typename InputIterator, typename OutputIterator>
+OutputIterator adjacent_difference(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                   InputIterator first, InputIterator last, 
+                                   OutputIterator result)
+{
+  using thrust::system::detail::generic::adjacent_difference;
+
+  return adjacent_difference(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, result);
+} // end adjacent_difference()
+
+
+template <typename DerivedPolicy, typename InputIterator, typename OutputIterator, typename BinaryFunction>
+OutputIterator adjacent_difference(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                   InputIterator first, InputIterator last, 
+                                   OutputIterator result,
+                                   BinaryFunction binary_op)
+{
+  using thrust::system::detail::generic::adjacent_difference;
+
+  return adjacent_difference(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, result, binary_op);
+} // end adjacent_difference()
+
+
+template <typename InputIterator, typename OutputIterator>
+OutputIterator adjacent_difference(InputIterator first, InputIterator last, 
+                                   OutputIterator result)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator>::type  System1;
+  typedef typename thrust::iterator_system<OutputIterator>::type System2;
+
+  System1 system1;
+  System2 system2;
+
+  return thrust::adjacent_difference(select_system(system1, system2), first, last, result);
+} // end adjacent_difference()
+
+
+template <typename InputIterator, typename OutputIterator, typename BinaryFunction>
+OutputIterator adjacent_difference(InputIterator first, InputIterator last,
+                                   OutputIterator result,
+                                   BinaryFunction binary_op)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator>::type  System1;
+  typedef typename thrust::iterator_system<OutputIterator>::type System2;
+
+  System1 system1;
+  System2 system2;
+
+  return thrust::adjacent_difference(select_system(system1, system2), first, last, result, binary_op);
+} // end adjacent_difference()
+
+
+} // end namespace thrust
+
diff --git a/compat/thrust/detail/advance.inl b/compat/thrust/detail/advance.inl
new file mode 100644
index 0000000000..2907be7534
--- /dev/null
+++ b/compat/thrust/detail/advance.inl
@@ -0,0 +1,38 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file advance.inl
+ *  \brief Inline file for advance.h
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/advance.h>
+#include <thrust/system/detail/generic/advance.h>
+
+namespace thrust
+{
+
+
+template <typename InputIterator, typename Distance>
+void advance(InputIterator& i, Distance n)
+{
+  thrust::system::detail::generic::advance(i, n);
+} // end advance()
+
+
+} // end namespace thrust
+
diff --git a/compat/thrust/detail/allocator/allocator_traits.h b/compat/thrust/detail/allocator/allocator_traits.h
new file mode 100644
index 0000000000..6ee99b453f
--- /dev/null
+++ b/compat/thrust/detail/allocator/allocator_traits.h
@@ -0,0 +1,240 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/type_traits/pointer_traits.h>
+#include <thrust/detail/type_traits/has_nested_type.h>
+#include <thrust/detail/type_traits.h>
+
+namespace thrust
+{
+namespace detail
+{
+namespace allocator_traits_detail
+{
+
+__THRUST_DEFINE_HAS_NESTED_TYPE(has_pointer, pointer)
+__THRUST_DEFINE_HAS_NESTED_TYPE(has_const_pointer, const_pointer)
+__THRUST_DEFINE_HAS_NESTED_TYPE(has_reference, reference)
+__THRUST_DEFINE_HAS_NESTED_TYPE(has_const_reference, const_reference)
+__THRUST_DEFINE_HAS_NESTED_TYPE(has_void_pointer, void_pointer)
+__THRUST_DEFINE_HAS_NESTED_TYPE(has_const_void_pointer, const_void_pointer)
+__THRUST_DEFINE_HAS_NESTED_TYPE(has_difference_type, difference_type)
+__THRUST_DEFINE_HAS_NESTED_TYPE(has_size_type, size_type)
+__THRUST_DEFINE_HAS_NESTED_TYPE(has_propagate_on_container_copy_assignment, propagate_on_container_copy_assignment)
+__THRUST_DEFINE_HAS_NESTED_TYPE(has_propagate_on_container_move_assignment, propagate_on_container_move_assignment)
+__THRUST_DEFINE_HAS_NESTED_TYPE(has_propagate_on_container_swap, propagate_on_container_swap)
+__THRUST_DEFINE_HAS_NESTED_TYPE(has_system_type, system_type)
+
+template<typename T>
+  struct nested_pointer
+{
+  typedef typename T::pointer type;
+};
+
+template<typename T>
+  struct nested_const_pointer
+{
+  typedef typename T::const_pointer type;
+};
+
+template<typename T>
+  struct nested_reference
+{
+  typedef typename T::reference type;
+};
+
+template<typename T>
+  struct nested_const_reference
+{
+  typedef typename T::const_reference type;
+};
+
+template<typename T>
+  struct nested_void_pointer
+{
+  typedef typename T::void_pointer type;
+};
+
+template<typename T>
+  struct nested_const_void_pointer
+{
+  typedef typename T::const_void_pointer type;
+};
+
+template<typename T>
+  struct nested_difference_type
+{
+  typedef typename T::difference_type type;
+};
+
+template<typename T>
+  struct nested_size_type
+{
+  typedef typename T::size_type type;
+};
+
+template<typename T>
+  struct nested_propagate_on_container_copy_assignment
+{
+  typedef typename T::propagate_on_container_copy_assignment type;
+};
+
+template<typename T>
+  struct nested_propagate_on_container_move_assignment
+{
+  typedef typename T::propagate_on_container_move_assignment type;
+};
+
+template<typename T>
+  struct nested_propagate_on_container_swap
+{
+  typedef typename T::propagate_on_container_swap type;
+};
+
+template<typename T>
+  struct nested_system_type
+{
+  typedef typename T::system_type type;
+};
+
+} // end allocator_traits_detail
+
+
+template<typename Alloc>
+  struct allocator_traits
+{
+  typedef Alloc allocator_type;
+
+  typedef typename allocator_type::value_type value_type;
+
+  typedef typename eval_if<
+    allocator_traits_detail::has_pointer<allocator_type>::value,
+    allocator_traits_detail::nested_pointer<allocator_type>,
+    identity_<value_type*>
+  >::type pointer;
+
+  private:
+    template<typename T>
+      struct rebind_pointer
+    {
+      typedef typename pointer_traits<pointer>::template rebind<T>::other type;
+    };
+
+  public:
+
+  typedef typename eval_if<
+    allocator_traits_detail::has_const_pointer<allocator_type>::value,
+    allocator_traits_detail::nested_const_pointer<allocator_type>,
+    rebind_pointer<const value_type>
+  >::type const_pointer;
+
+  typedef typename eval_if<
+    allocator_traits_detail::has_void_pointer<allocator_type>::value,
+    allocator_traits_detail::nested_void_pointer<allocator_type>,
+    rebind_pointer<void>
+  >::type void_pointer;
+
+  typedef typename eval_if<
+    allocator_traits_detail::has_const_void_pointer<allocator_type>::value,
+    allocator_traits_detail::nested_const_void_pointer<allocator_type>,
+    rebind_pointer<const void>
+  >::type const_void_pointer;
+
+  typedef typename eval_if<
+    allocator_traits_detail::has_difference_type<allocator_type>::value,
+    allocator_traits_detail::nested_difference_type<allocator_type>,
+    pointer_difference<pointer>
+  >::type difference_type;
+
+  typedef typename eval_if<
+    allocator_traits_detail::has_size_type<allocator_type>::value,
+    allocator_traits_detail::nested_size_type<allocator_type>,
+    make_unsigned<difference_type>
+  >::type size_type;
+
+  typedef typename eval_if<
+    allocator_traits_detail::has_propagate_on_container_copy_assignment<allocator_type>::value,
+    allocator_traits_detail::nested_propagate_on_container_copy_assignment<allocator_type>,
+    identity_<false_type>
+  >::type propagate_on_container_copy_assignment;
+
+  typedef typename eval_if<
+    allocator_traits_detail::has_propagate_on_container_move_assignment<allocator_type>::value,
+    allocator_traits_detail::nested_propagate_on_container_move_assignment<allocator_type>,
+    identity_<false_type>
+  >::type propagate_on_container_move_assignment;
+
+  typedef typename eval_if<
+    allocator_traits_detail::has_propagate_on_container_swap<allocator_type>::value,
+    allocator_traits_detail::nested_propagate_on_container_swap<allocator_type>,
+    identity_<false_type>
+  >::type propagate_on_container_swap;
+
+  typedef typename eval_if<
+    allocator_traits_detail::has_system_type<allocator_type>::value,
+    allocator_traits_detail::nested_system_type<allocator_type>,
+    thrust::iterator_system<pointer>
+  >::type system_type;
+
+  // XXX rebind and rebind_traits are alias templates
+  //     and so are omitted while c++11 is unavailable
+
+  inline static pointer allocate(allocator_type &a, size_type n);
+
+  inline static pointer allocate(allocator_type &a, size_type n, const_void_pointer hint);
+
+  inline static void deallocate(allocator_type &a, pointer p, size_type n);
+
+  // XXX should probably change T* to pointer below and then relax later
+
+  template<typename T>
+  inline __host__ __device__ static void construct(allocator_type &a, T *p);
+  
+  template<typename T, typename Arg1>
+  inline __host__ __device__ static void construct(allocator_type &a, T *p, const Arg1 &arg1);
+
+  template<typename T>
+  inline __host__ __device__ static void destroy(allocator_type &a, T *p);
+
+  inline static size_type max_size(const allocator_type &a);
+}; // end allocator_traits
+
+
+// XXX consider moving this non-standard functionality inside allocator_traits
+template<typename Alloc>
+  struct allocator_system
+{
+  // the type of the allocator's system
+  typedef typename eval_if<
+    allocator_traits_detail::has_system_type<Alloc>::value,
+    allocator_traits_detail::nested_system_type<Alloc>,
+    thrust::iterator_system<
+      typename allocator_traits<Alloc>::pointer
+    >
+  >::type type;
+
+  inline static type &get(Alloc &a);
+};
+
+
+} // end detail
+} // end thrust
+
+#include <thrust/detail/allocator/allocator_traits.inl>
+
diff --git a/compat/thrust/detail/allocator/allocator_traits.inl b/compat/thrust/detail/allocator/allocator_traits.inl
new file mode 100644
index 0000000000..83193355d5
--- /dev/null
+++ b/compat/thrust/detail/allocator/allocator_traits.inl
@@ -0,0 +1,287 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/allocator/allocator_traits.h>
+#include <thrust/detail/type_traits/has_member_function.h>
+#include <thrust/detail/type_traits/is_call_possible.h>
+#include <new>
+#include <limits>
+
+namespace thrust
+{
+namespace detail
+{
+namespace allocator_traits_detail
+{
+
+__THRUST_DEFINE_IS_CALL_POSSIBLE(has_member_allocate_with_hint_impl, allocate)
+
+template<typename Alloc>
+  class has_member_allocate_with_hint
+{
+  typedef typename allocator_traits<Alloc>::pointer            pointer;
+  typedef typename allocator_traits<Alloc>::size_type          size_type;
+  typedef typename allocator_traits<Alloc>::const_void_pointer const_void_pointer;
+
+  public:
+    typedef typename has_member_allocate_with_hint_impl<Alloc, pointer(size_type,const_void_pointer)>::type type;
+    static const bool value = type::value;
+};
+
+template<typename Alloc>
+  typename enable_if<
+    has_member_allocate_with_hint<Alloc>::value,
+    typename allocator_traits<Alloc>::pointer
+  >::type
+    allocate(Alloc &a, typename allocator_traits<Alloc>::size_type n, typename allocator_traits<Alloc>::const_void_pointer hint)
+{
+  return a.allocate(n,hint);
+}
+
+template<typename Alloc>
+  typename disable_if<
+    has_member_allocate_with_hint<Alloc>::value,
+    typename allocator_traits<Alloc>::pointer
+  >::type
+    allocate(Alloc &a, typename allocator_traits<Alloc>::size_type n, typename allocator_traits<Alloc>::const_void_pointer)
+{
+  return a.allocate(n);
+}
+
+
+__THRUST_DEFINE_IS_CALL_POSSIBLE(has_member_construct1_impl, construct)
+
+template<typename Alloc, typename T>
+  struct has_member_construct1
+    : has_member_construct1_impl<Alloc, void(T*)>
+{};
+
+template<typename Alloc, typename T>
+  inline __host__ __device__
+    typename enable_if<
+      has_member_construct1<Alloc,T>::value
+    >::type
+      construct(Alloc &a, T *p)
+{
+  a.construct(p);
+}
+
+template<typename Alloc, typename T>
+  inline __host__ __device__
+    typename disable_if<
+      has_member_construct1<Alloc,T>::value
+    >::type
+      construct(Alloc &a, T *p)
+{
+  ::new(static_cast<void*>(p)) T();
+}
+
+
+__THRUST_DEFINE_IS_CALL_POSSIBLE(has_member_construct2_impl, construct)
+
+template<typename Alloc, typename T, typename Arg1>
+  struct has_member_construct2
+    : has_member_construct2_impl<Alloc, void(T*,const Arg1 &)>
+{};
+
+template<typename Alloc, typename T, typename Arg1>
+  inline __host__ __device__
+    typename enable_if<
+      has_member_construct2<Alloc,T,Arg1>::value
+    >::type
+      construct(Alloc &a, T *p, const Arg1 &arg1)
+{
+  a.construct(p,arg1);
+}
+
+template<typename Alloc, typename T, typename Arg1>
+  inline __host__ __device__
+    typename disable_if<
+      has_member_construct2<Alloc,T,Arg1>::value
+    >::type
+      construct(Alloc &, T *p, const Arg1 &arg1)
+{
+  ::new(static_cast<void*>(p)) T(arg1);
+}
+
+
+__THRUST_DEFINE_IS_CALL_POSSIBLE(has_member_destroy_impl, destroy)
+
+template<typename Alloc, typename T>
+  struct has_member_destroy
+    : has_member_destroy_impl<Alloc, void(T*)>
+{};
+
+template<typename Alloc, typename T>
+  inline __host__ __device__
+    typename enable_if<
+      has_member_destroy<Alloc,T>::value
+    >::type
+      destroy(Alloc &a, T *p)
+{
+  a.destroy(p);
+}
+
+template<typename Alloc, typename T>
+  inline __host__ __device__
+    typename disable_if<
+      has_member_destroy<Alloc,T>::value
+    >::type
+      destroy(Alloc &, T *p)
+{
+  p->~T();
+}
+
+
+__THRUST_DEFINE_IS_CALL_POSSIBLE(has_member_max_size_impl, max_size)
+
+template<typename Alloc>
+  class has_member_max_size
+{
+  typedef typename allocator_traits<Alloc>::size_type size_type;
+
+  public:
+    typedef typename has_member_max_size_impl<Alloc, size_type(void)>::type type;
+    static const bool value = type::value;
+};
+
+template<typename Alloc>
+  typename enable_if<
+    has_member_max_size<Alloc>::value,
+    typename allocator_traits<Alloc>::size_type
+  >::type
+    max_size(const Alloc &a)
+{
+  return a.max_size();
+}
+
+template<typename Alloc>
+  typename disable_if<
+    has_member_max_size<Alloc>::value,
+    typename allocator_traits<Alloc>::size_type
+  >::type
+    max_size(const Alloc &a)
+{
+  typedef typename allocator_traits<Alloc>::size_type size_type;
+  return std::numeric_limits<size_type>::max();
+}
+
+__THRUST_DEFINE_HAS_MEMBER_FUNCTION(has_member_system_impl, system)
+
+template<typename Alloc>
+  class has_member_system
+{
+  typedef typename allocator_system<Alloc>::type system_type;
+
+  public:
+    typedef typename has_member_system_impl<Alloc, system_type&(void)>::type type;
+    static const bool value = type::value;
+};
+
+template<typename Alloc>
+  typename enable_if<
+    has_member_system<Alloc>::value,
+    typename allocator_system<Alloc>::type &
+  >::type
+    system(Alloc &a)
+{
+  return a.system();
+}
+
+template<typename Alloc>
+  typename disable_if<
+    has_member_system<Alloc>::value,
+    typename allocator_system<Alloc>::type &
+  >::type
+    system(Alloc &a)
+{
+  // assumes the system is default-constructible
+  static typename allocator_system<Alloc>::type state;
+  return state;
+}
+
+
+} // end allocator_traits_detail
+
+
+template<typename Alloc>
+  typename allocator_traits<Alloc>::pointer
+    allocator_traits<Alloc>
+      ::allocate(Alloc &a, typename allocator_traits<Alloc>::size_type n)
+{
+  return a.allocate(n);
+}
+
+template<typename Alloc>
+  typename allocator_traits<Alloc>::pointer
+    allocator_traits<Alloc>
+      ::allocate(Alloc &a, typename allocator_traits<Alloc>::size_type n, typename allocator_traits<Alloc>::const_void_pointer hint)
+{
+  return allocator_traits_detail::allocate(a, n, hint);
+}
+
+template<typename Alloc>
+  void allocator_traits<Alloc>
+    ::deallocate(Alloc &a, typename allocator_traits<Alloc>::pointer p, typename allocator_traits<Alloc>::size_type n)
+{
+  return a.deallocate(p,n);
+}
+
+template<typename Alloc>
+  template<typename T>
+    void allocator_traits<Alloc>
+      ::construct(allocator_type &a, T *p)
+{
+  return allocator_traits_detail::construct(a,p);
+}
+
+template<typename Alloc>
+  template<typename T, typename Arg1>
+    void allocator_traits<Alloc>
+      ::construct(allocator_type &a, T *p, const Arg1 &arg1)
+{
+  return allocator_traits_detail::construct(a,p,arg1);
+}
+
+template<typename Alloc>
+  template<typename T>
+    void allocator_traits<Alloc>
+      ::destroy(allocator_type &a, T *p)
+{
+  return allocator_traits_detail::destroy(a,p);
+}
+
+template<typename Alloc>
+  typename allocator_traits<Alloc>::size_type
+    allocator_traits<Alloc>
+      ::max_size(const allocator_type &a)
+{
+  return allocator_traits_detail::max_size(a);
+}
+
+template<typename Alloc>
+  typename allocator_system<Alloc>::type &
+    allocator_system<Alloc>
+      ::get(Alloc &a)
+{
+  return allocator_traits_detail::system(a);
+}
+
+
+} // end detail
+} // end thrust
+
diff --git a/compat/thrust/detail/allocator/copy_construct_range.h b/compat/thrust/detail/allocator/copy_construct_range.h
new file mode 100644
index 0000000000..5d99e1fa14
--- /dev/null
+++ b/compat/thrust/detail/allocator/copy_construct_range.h
@@ -0,0 +1,45 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/execution_policy.h>
+
+namespace thrust
+{
+namespace detail
+{
+
+template<typename System, typename Allocator, typename InputIterator, typename Pointer>
+  Pointer copy_construct_range(thrust::execution_policy<System> &from_system,
+                               Allocator &a,
+                               InputIterator first,
+                               InputIterator last,
+                               Pointer result);
+
+template<typename System, typename Allocator, typename InputIterator, typename Size, typename Pointer>
+  Pointer copy_construct_range_n(thrust::execution_policy<System> &from_system,
+                                 Allocator &a,
+                                 InputIterator first,
+                                 Size n,
+                                 Pointer result);
+
+} // end detail
+} // end thrust
+
+#include <thrust/detail/allocator/copy_construct_range.inl>
+
diff --git a/compat/thrust/detail/allocator/copy_construct_range.inl b/compat/thrust/detail/allocator/copy_construct_range.inl
new file mode 100644
index 0000000000..7c5478b65b
--- /dev/null
+++ b/compat/thrust/detail/allocator/copy_construct_range.inl
@@ -0,0 +1,298 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/allocator/allocator_traits.h>
+#include <thrust/detail/type_traits/pointer_traits.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/detail/copy.h>
+#include <thrust/tuple.h>
+#include <thrust/advance.h>
+#include <thrust/distance.h>
+#include <thrust/iterator/zip_iterator.h>
+#include <thrust/for_each.h>
+#include <memory>
+
+namespace thrust
+{
+namespace detail
+{
+namespace allocator_traits_detail
+{
+
+
+template<typename Allocator, typename InputType, typename OutputType>
+  struct copy_construct_with_allocator
+{
+  Allocator &a;
+
+  copy_construct_with_allocator(Allocator &a)
+    : a(a)
+  {}
+
+  template<typename Tuple>
+  inline __host__ __device__
+  void operator()(Tuple t)
+  {
+    const InputType &in = thrust::get<0>(t);
+    OutputType &out = thrust::get<1>(t);
+
+    allocator_traits<Allocator>::construct(a, &out, in);
+  }
+};
+
+
+template<typename Allocator, typename T>
+  struct needs_copy_construct_via_allocator
+    : has_member_construct2<
+        Allocator,
+        T,
+        T
+      >
+{};
+
+
+// we know that std::allocator::construct's only effect is to call T's
+// copy constructor, so we needn't use it for copy construction
+template<typename U, typename T>
+  struct needs_copy_construct_via_allocator<std::allocator<U>, T>
+    : thrust::detail::false_type
+{};
+
+
+// XXX it's regrettable that this implementation is copied almost
+//     exactly from system::detail::generic::uninitialized_copy
+//     perhaps generic::uninitialized_copy could call this routine
+//     with a default allocator
+template<typename Allocator, typename FromSystem, typename ToSystem, typename InputIterator, typename Pointer>
+  typename enable_if_convertible<
+    FromSystem,
+    ToSystem,
+    Pointer
+  >::type
+    uninitialized_copy_with_allocator(Allocator &a,
+                                      thrust::execution_policy<FromSystem> &from_system,
+                                      thrust::execution_policy<ToSystem> &to_system,
+                                      InputIterator first,
+                                      InputIterator last,
+                                      Pointer result)
+{
+  // zip up the iterators
+  typedef thrust::tuple<InputIterator,Pointer> IteratorTuple;
+  typedef thrust::zip_iterator<IteratorTuple>  ZipIterator;
+
+  ZipIterator begin = thrust::make_zip_iterator(thrust::make_tuple(first,result));
+  ZipIterator end = begin;
+
+  // get a zip_iterator pointing to the end
+  const typename thrust::iterator_difference<InputIterator>::type n = thrust::distance(first,last);
+  thrust::advance(end,n);
+
+  // create a functor
+  typedef typename iterator_traits<InputIterator>::value_type InputType;
+  typedef typename iterator_traits<Pointer>::value_type       OutputType;
+
+  // do the for_each
+  // note we use to_system to dispatch the for_each
+  thrust::for_each(to_system, begin, end, copy_construct_with_allocator<Allocator,InputType,OutputType>(a));
+
+  // return the end of the output range
+  return thrust::get<1>(end.get_iterator_tuple());
+}
+
+
+// XXX it's regrettable that this implementation is copied almost
+//     exactly from system::detail::generic::uninitialized_copy_n
+//     perhaps generic::uninitialized_copy_n could call this routine
+//     with a default allocator
+template<typename Allocator, typename FromSystem, typename ToSystem, typename InputIterator, typename Size, typename Pointer>
+  typename enable_if_convertible<
+    FromSystem,
+    ToSystem,
+    Pointer
+  >::type
+    uninitialized_copy_with_allocator_n(Allocator &a,
+                                        thrust::execution_policy<FromSystem> &from_system,
+                                        thrust::execution_policy<ToSystem> &to_system,
+                                        InputIterator first,
+                                        Size n,
+                                        Pointer result)
+{
+  // zip up the iterators
+  typedef thrust::tuple<InputIterator,Pointer> IteratorTuple;
+  typedef thrust::zip_iterator<IteratorTuple>  ZipIterator;
+
+  ZipIterator begin = thrust::make_zip_iterator(thrust::make_tuple(first,result));
+
+  // create a functor
+  typedef typename iterator_traits<InputIterator>::value_type InputType;
+  typedef typename iterator_traits<Pointer>::value_type       OutputType;
+
+  // do the for_each_n
+  // note we use to_system to dispatch the for_each_n
+  ZipIterator end = thrust::for_each_n(to_system, begin, n, copy_construct_with_allocator<Allocator,InputType,OutputType>(a));
+
+  // return the end of the output range
+  return thrust::get<1>(end.get_iterator_tuple());
+}
+
+
+template<typename Allocator, typename FromSystem, typename ToSystem, typename InputIterator, typename Pointer>
+  typename disable_if_convertible<
+    FromSystem,
+    ToSystem,
+    Pointer
+  >::type
+    uninitialized_copy_with_allocator(Allocator &,
+                                      thrust::execution_policy<FromSystem> &from_system,
+                                      thrust::execution_policy<ToSystem> &to_system,
+                                      InputIterator first,
+                                      InputIterator last,
+                                      Pointer result)
+{
+  // the systems aren't trivially interoperable
+  // just call two_system_copy and hope for the best
+  return thrust::detail::two_system_copy(from_system, to_system, first, last, result);
+} // end uninitialized_copy_with_allocator()
+
+
+template<typename Allocator, typename FromSystem, typename ToSystem, typename InputIterator, typename Size, typename Pointer>
+  typename disable_if_convertible<
+    FromSystem,
+    ToSystem,
+    Pointer
+  >::type
+    uninitialized_copy_with_allocator_n(Allocator &,
+                                        thrust::execution_policy<FromSystem> &from_system,
+                                        thrust::execution_policy<ToSystem> &to_system,
+                                        InputIterator first,
+                                        Size n,
+                                        Pointer result)
+{
+  // the systems aren't trivially interoperable
+  // just call two_system_copy_n and hope for the best
+  return thrust::detail::two_system_copy_n(from_system, to_system, first, n, result);
+} // end uninitialized_copy_with_allocator_n()
+
+
+template<typename FromSystem, typename Allocator, typename InputIterator, typename Pointer>
+  typename disable_if<
+    needs_copy_construct_via_allocator<
+      Allocator,
+      typename pointer_element<Pointer>::type
+    >::value,
+    Pointer
+  >::type
+    copy_construct_range(thrust::execution_policy<FromSystem> &from_system,
+                         Allocator &a,
+                         InputIterator first,
+                         InputIterator last,
+                         Pointer result)
+{
+  typename allocator_system<Allocator>::type &to_system = allocator_system<Allocator>::get(a);
+
+  // just call two_system_copy
+  return thrust::detail::two_system_copy(from_system, to_system, first, last, result);
+}
+
+
+template<typename FromSystem, typename Allocator, typename InputIterator, typename Size, typename Pointer>
+  typename disable_if<
+    needs_copy_construct_via_allocator<
+      Allocator,
+      typename pointer_element<Pointer>::type
+    >::value,
+    Pointer
+  >::type
+    copy_construct_range_n(thrust::execution_policy<FromSystem> &from_system,
+                           Allocator &a,
+                           InputIterator first,
+                           Size n,
+                           Pointer result)
+{
+  typename allocator_system<Allocator>::type &to_system = allocator_system<Allocator>::get(a);
+
+  // just call two_system_copy_n
+  return thrust::detail::two_system_copy_n(from_system, to_system, first, n, result);
+}
+
+
+template<typename FromSystem, typename Allocator, typename InputIterator, typename Pointer>
+  typename enable_if<
+    needs_copy_construct_via_allocator<
+      Allocator,
+      typename pointer_element<Pointer>::type
+    >::value,
+    Pointer
+  >::type
+    copy_construct_range(thrust::execution_policy<FromSystem> &from_system,
+                         Allocator &a,
+                         InputIterator first,
+                         InputIterator last,
+                         Pointer result)
+{
+  typename allocator_system<Allocator>::type &to_system = allocator_system<Allocator>::get(a);
+  return uninitialized_copy_with_allocator(a, from_system, to_system, first, last, result);
+}
+
+
+template<typename FromSystem, typename Allocator, typename InputIterator, typename Size, typename Pointer>
+  typename enable_if<
+    needs_copy_construct_via_allocator<
+      Allocator,
+      typename pointer_element<Pointer>::type
+    >::value,
+    Pointer
+  >::type
+    copy_construct_range_n(thrust::execution_policy<FromSystem> &from_system,
+                           Allocator &a,
+                           InputIterator first,
+                           Size n,
+                           Pointer result)
+{
+  typename allocator_system<Allocator>::type &to_system = allocator_system<Allocator>::get(a);
+  return uninitialized_copy_with_allocator_n(a, from_system, to_system, first, n, result);
+}
+
+
+} // end allocator_traits_detail
+
+
+template<typename System, typename Allocator, typename InputIterator, typename Pointer>
+  Pointer copy_construct_range(thrust::execution_policy<System> &from_system,
+                               Allocator &a,
+                               InputIterator first,
+                               InputIterator last,
+                               Pointer result)
+{
+  return allocator_traits_detail::copy_construct_range(from_system, a, first, last, result);
+}
+
+
+template<typename System, typename Allocator, typename InputIterator, typename Size, typename Pointer>
+  Pointer copy_construct_range_n(thrust::execution_policy<System> &from_system,
+                                 Allocator &a,
+                                 InputIterator first,
+                                 Size n,
+                                 Pointer result)
+{
+  return allocator_traits_detail::copy_construct_range_n(from_system, a, first, n, result);
+}
+
+
+} // end detail
+} // end thrust
+
diff --git a/compat/thrust/detail/allocator/default_construct_range.h b/compat/thrust/detail/allocator/default_construct_range.h
new file mode 100644
index 0000000000..d83cb31f35
--- /dev/null
+++ b/compat/thrust/detail/allocator/default_construct_range.h
@@ -0,0 +1,36 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+namespace thrust
+{
+namespace detail
+{
+
+
+template<typename Allocator, typename Pointer, typename Size>
+inline void default_construct_range(Allocator &a, Pointer p, Size n);
+
+
+} // end detail
+} // end thrust
+
+#include <thrust/detail/allocator/default_construct_range.inl>
+
+
diff --git a/compat/thrust/detail/allocator/default_construct_range.inl b/compat/thrust/detail/allocator/default_construct_range.inl
new file mode 100644
index 0000000000..45fe9c69a1
--- /dev/null
+++ b/compat/thrust/detail/allocator/default_construct_range.inl
@@ -0,0 +1,105 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/allocator/allocator_traits.h>
+#include <thrust/detail/type_traits.h>
+#include <thrust/detail/type_traits/pointer_traits.h>
+#include <thrust/for_each.h>
+#include <thrust/uninitialized_fill.h>
+
+namespace thrust
+{
+namespace detail
+{
+namespace allocator_traits_detail
+{
+
+
+template<typename Allocator>
+  struct construct1_via_allocator
+{
+  Allocator &a;
+
+  construct1_via_allocator(Allocator &a)
+    : a(a)
+  {}
+
+  template<typename T>
+  inline __host__ __device__
+  void operator()(T &x)
+  {
+    allocator_traits<Allocator>::construct(a, &x);
+  }
+};
+
+
+template<typename Allocator, typename T>
+  struct needs_default_construct_via_allocator
+    : has_member_construct1<
+        Allocator,
+        T
+      >
+{};
+
+
+// we know that std::allocator::construct's only effect is to call T's 
+// default constructor, so we needn't use it for default construction
+template<typename U, typename T>
+  struct needs_default_construct_via_allocator<std::allocator<U>, T>
+    : thrust::detail::false_type
+{};
+
+
+template<typename Allocator, typename Pointer, typename Size>
+  typename enable_if<
+    needs_default_construct_via_allocator<
+      Allocator,
+      typename pointer_element<Pointer>::type
+    >::value
+  >::type
+    default_construct_range(Allocator &a, Pointer p, Size n)
+{
+  thrust::for_each_n(allocator_system<Allocator>::get(a), p, n, construct1_via_allocator<Allocator>(a));
+}
+
+
+template<typename Allocator, typename Pointer, typename Size>
+  typename disable_if<
+    needs_default_construct_via_allocator<
+      Allocator,
+      typename pointer_element<Pointer>::type
+    >::value
+  >::type
+    default_construct_range(Allocator &a, Pointer p, Size n)
+{
+  thrust::uninitialized_fill_n(allocator_system<Allocator>::get(a), p, n, typename pointer_element<Pointer>::type());
+}
+
+
+} // end allocator_traits_detail
+
+
+template<typename Allocator, typename Pointer, typename Size>
+  void default_construct_range(Allocator &a, Pointer p, Size n)
+{
+  return allocator_traits_detail::default_construct_range(a,p,n);
+}
+
+
+} // end detail
+} // end thrust
+
diff --git a/compat/thrust/detail/allocator/destroy_range.h b/compat/thrust/detail/allocator/destroy_range.h
new file mode 100644
index 0000000000..d690a60a79
--- /dev/null
+++ b/compat/thrust/detail/allocator/destroy_range.h
@@ -0,0 +1,33 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+namespace thrust
+{
+namespace detail
+{
+
+template<typename Allocator, typename Pointer, typename Size>
+  inline void destroy_range(Allocator &a, Pointer p, Size n);
+
+} // end detail
+} // end thrust
+
+#include <thrust/detail/allocator/destroy_range.inl>
+
diff --git a/compat/thrust/detail/allocator/destroy_range.inl b/compat/thrust/detail/allocator/destroy_range.inl
new file mode 100644
index 0000000000..ace222356f
--- /dev/null
+++ b/compat/thrust/detail/allocator/destroy_range.inl
@@ -0,0 +1,158 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/detail/allocator/destroy_range.h>
+#include <thrust/detail/allocator/allocator_traits.h>
+#include <thrust/detail/type_traits/pointer_traits.h>
+#include <thrust/for_each.h>
+#include <memory>
+
+namespace thrust
+{
+namespace detail
+{
+namespace allocator_traits_detail
+{
+
+
+// destroy_range has three cases:
+// if Allocator has an effectful member function destroy:
+//   1. destroy via the allocator
+// else
+//   2. if T has a non-trivial destructor, destroy the range without using the allocator
+//   3. if T has a trivial destructor, do a no-op
+
+template<typename Allocator, typename T>
+  struct has_effectful_member_destroy
+    : has_member_destroy<Allocator,T>
+{};
+
+// std::allocator::destroy's only effect is to invoke its argument's destructor
+template<typename U, typename T>
+  struct has_effectful_member_destroy<std::allocator<U>, T>
+    : thrust::detail::false_type
+{};
+
+// case 1: Allocator has an effectful 1-argument member function "destroy"
+template<typename Allocator, typename Pointer>
+  struct enable_if_destroy_range_case1
+    : thrust::detail::enable_if<
+        has_effectful_member_destroy<
+          Allocator,
+          typename pointer_element<Pointer>::type
+        >::value
+      >
+{};
+
+// case 2: Allocator has no member function "destroy", but T has a non-trivial destructor
+template<typename Allocator, typename Pointer>
+  struct enable_if_destroy_range_case2
+    : thrust::detail::enable_if<
+        !has_effectful_member_destroy<
+          Allocator,
+          typename pointer_element<Pointer>::type
+        >::value &&
+        !has_trivial_destructor<
+          typename pointer_element<Pointer>::type
+        >::value
+      >
+{};
+
+// case 3: Allocator has no member function "destroy", and T has a trivial destructor
+template<typename Allocator, typename Pointer>
+  struct enable_if_destroy_range_case3
+    : thrust::detail::enable_if<
+        !has_effectful_member_destroy<
+          Allocator,
+          typename pointer_element<Pointer>::type
+        >::value &&
+        has_trivial_destructor<
+          typename pointer_element<Pointer>::type
+        >::value
+      >
+{};
+
+
+
+template<typename Allocator>
+  struct destroy_via_allocator
+{
+  Allocator &a;
+
+  destroy_via_allocator(Allocator &a)
+    : a(a)
+  {}
+
+  template<typename T>
+  inline __host__ __device__
+  void operator()(T &x)
+  {
+    allocator_traits<Allocator>::destroy(a, &x);
+  }
+};
+
+
+// destroy_range case 1: destroy via allocator
+template<typename Allocator, typename Pointer, typename Size>
+  typename enable_if_destroy_range_case1<Allocator,Pointer>::type
+    destroy_range(Allocator &a, Pointer p, Size n)
+{
+  thrust::for_each_n(allocator_system<Allocator>::get(a), p, n, destroy_via_allocator<Allocator>(a));
+}
+
+
+// we must prepare for His coming
+struct gozer
+{
+  template<typename T>
+  inline __host__ __device__
+  void operator()(T &x)
+  {
+    x.~T();
+  }
+};
+
+// destroy_range case 2: destroy without the allocator
+template<typename Allocator, typename Pointer, typename Size>
+  typename enable_if_destroy_range_case2<Allocator,Pointer>::type
+    destroy_range(Allocator &a, Pointer p, Size n)
+{
+  thrust::for_each_n(allocator_system<Allocator>::get(a), p, n, gozer());
+}
+
+
+// destroy_range case 3: no-op
+template<typename Allocator, typename Pointer, typename Size>
+  typename enable_if_destroy_range_case3<Allocator,Pointer>::type
+    destroy_range(Allocator &, Pointer, Size)
+{
+  // no op
+}
+
+
+} // end allocator_traits_detail
+
+
+template<typename Allocator, typename Pointer, typename Size>
+  void destroy_range(Allocator &a, Pointer p, Size n)
+{
+  return allocator_traits_detail::destroy_range(a,p,n);
+}
+
+
+} // end detail
+} // end thrust
+
diff --git a/compat/thrust/detail/allocator/fill_construct_range.h b/compat/thrust/detail/allocator/fill_construct_range.h
new file mode 100644
index 0000000000..66fec416c3
--- /dev/null
+++ b/compat/thrust/detail/allocator/fill_construct_range.h
@@ -0,0 +1,35 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+namespace thrust
+{
+namespace detail
+{
+
+
+template<typename Allocator, typename Pointer, typename Size, typename T>
+inline void fill_construct_range(Allocator &a, Pointer p, Size n, const T &value);
+
+
+} // end detail
+} // end thrust
+
+#include <thrust/detail/allocator/fill_construct_range.inl>
+
diff --git a/compat/thrust/detail/allocator/fill_construct_range.inl b/compat/thrust/detail/allocator/fill_construct_range.inl
new file mode 100644
index 0000000000..e2c9c09c80
--- /dev/null
+++ b/compat/thrust/detail/allocator/fill_construct_range.inl
@@ -0,0 +1,109 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/type_traits.h>
+#include <thrust/detail/allocator/allocator_traits.h>
+#include <thrust/detail/type_traits/pointer_traits.h>
+#include <thrust/for_each.h>
+#include <thrust/uninitialized_fill.h>
+#include <memory>
+
+namespace thrust
+{
+namespace detail
+{
+namespace allocator_traits_detail
+{
+
+// fill_construct_range has 2 cases:
+// if Allocator has an effectful member function construct:
+//   1. construct via the allocator
+// else
+//   2. construct via uninitialized_fill
+
+template<typename Allocator, typename T, typename Arg1>
+  struct has_effectful_member_construct2
+    : has_member_construct2<Allocator,T,Arg1>
+{};
+
+// std::allocator::construct's only effect is to invoke placement new
+template<typename U, typename T, typename Arg1>
+  struct has_effectful_member_construct2<std::allocator<U>,T,Arg1>
+    : thrust::detail::false_type
+{};
+
+
+template<typename Allocator, typename Arg1>
+  struct construct2_via_allocator
+{
+  Allocator &a;
+  Arg1 arg;
+
+  construct2_via_allocator(Allocator &a, const Arg1 &arg)
+    : a(a), arg(arg)
+  {}
+
+  template<typename T>
+  inline __host__ __device__
+  void operator()(T &x)
+  {
+    allocator_traits<Allocator>::construct(a, &x, arg);
+  }
+};
+
+
+template<typename Allocator, typename Pointer, typename Size, typename T>
+  typename enable_if<
+    has_effectful_member_construct2<
+      Allocator,
+      typename pointer_element<Pointer>::type,
+      T
+    >::value
+  >::type
+    fill_construct_range(Allocator &a, Pointer p, Size n, const T &value)
+{
+  thrust::for_each_n(allocator_system<Allocator>::get(a), p, n, construct2_via_allocator<Allocator,T>(a, value));
+}
+
+
+template<typename Allocator, typename Pointer, typename Size, typename T>
+  typename disable_if<
+    has_effectful_member_construct2<
+      Allocator,
+      typename pointer_element<Pointer>::type,
+      T
+    >::value
+  >::type
+    fill_construct_range(Allocator &a, Pointer p, Size n, const T &value)
+{
+  thrust::uninitialized_fill_n(allocator_system<Allocator>::get(a), p, n, value);
+}
+
+
+} // end allocator_traits_detail
+
+
+template<typename Alloc, typename Pointer, typename Size, typename T>
+  void fill_construct_range(Alloc &a, Pointer p, Size n, const T &value)
+{
+  return allocator_traits_detail::fill_construct_range(a,p,n,value);
+}
+
+
+} // end detail
+} // end thrust
+
diff --git a/compat/thrust/detail/allocator/malloc_allocator.h b/compat/thrust/detail/allocator/malloc_allocator.h
new file mode 100644
index 0000000000..cf4567e419
--- /dev/null
+++ b/compat/thrust/detail/allocator/malloc_allocator.h
@@ -0,0 +1,52 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/type_traits/pointer_traits.h>
+#include <thrust/detail/allocator/tagged_allocator.h>
+
+namespace thrust
+{
+namespace detail
+{
+
+template<typename T, typename System, typename Pointer>
+  class malloc_allocator
+    : public thrust::detail::tagged_allocator<
+               T, System, Pointer
+             >
+{
+  private:
+    typedef thrust::detail::tagged_allocator<
+      T, System, Pointer
+    > super_t;
+
+  public:
+    typedef typename super_t::pointer   pointer;
+    typedef typename super_t::size_type size_type;
+
+    pointer allocate(size_type cnt);
+
+    void deallocate(pointer p, size_type n);
+};
+
+} // end detail
+} // end thrust
+
+#include <thrust/detail/allocator/malloc_allocator.inl>
+
diff --git a/compat/thrust/detail/allocator/malloc_allocator.inl b/compat/thrust/detail/allocator/malloc_allocator.inl
new file mode 100644
index 0000000000..dd70202a04
--- /dev/null
+++ b/compat/thrust/detail/allocator/malloc_allocator.inl
@@ -0,0 +1,64 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/allocator/malloc_allocator.h>
+#include <thrust/system/detail/generic/select_system.h>
+#include <thrust/system/detail/bad_alloc.h>
+#include <thrust/detail/raw_pointer_cast.h>
+#include <thrust/detail/malloc_and_free.h>
+
+namespace thrust
+{
+namespace detail
+{
+
+
+template<typename T, typename System, typename Pointer>
+  typename malloc_allocator<T,System,Pointer>::pointer
+    malloc_allocator<T,System,Pointer>
+      ::allocate(typename malloc_allocator<T,System,Pointer>::size_type cnt)
+{
+  using thrust::system::detail::generic::select_system;
+
+  // XXX should use a hypothetical thrust::static_pointer_cast here
+  System system;
+
+  pointer result = thrust::malloc<T>(select_system(system), cnt);
+
+  if(result.get() == 0)
+  {
+    throw thrust::system::detail::bad_alloc("malloc_allocator::allocate: malloc failed");
+  } // end if
+
+  return result;
+} // end malloc_allocator::allocate()
+
+
+template<typename T, typename System, typename Pointer>
+  void malloc_allocator<T,System,Pointer>
+    ::deallocate(typename malloc_allocator<T,System,Pointer>::pointer p, typename malloc_allocator<T,System,Pointer>::size_type n)
+{
+  using thrust::system::detail::generic::select_system;
+
+  System system;
+  thrust::free(select_system(system), p);
+} // end malloc_allocator
+
+
+} // end detail
+} // end thrust
+
diff --git a/compat/thrust/detail/allocator/no_throw_allocator.h b/compat/thrust/detail/allocator/no_throw_allocator.h
new file mode 100644
index 0000000000..ce397dbdb5
--- /dev/null
+++ b/compat/thrust/detail/allocator/no_throw_allocator.h
@@ -0,0 +1,62 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+namespace thrust
+{
+namespace detail
+{
+
+template<typename BaseAllocator>
+  struct no_throw_allocator : BaseAllocator
+{
+  private:
+    typedef BaseAllocator super_t;
+  
+  public:
+    inline no_throw_allocator(const BaseAllocator &other = BaseAllocator())
+      : super_t(other)
+    {}
+
+    template<typename U>
+      struct rebind
+    {
+      typedef no_throw_allocator<typename super_t::template rebind<U>::other> other;
+    }; // end rebind
+
+    void deallocate(typename super_t::pointer p, typename super_t::size_type n)
+    {
+      try
+      {
+        super_t::deallocate(p, n);
+      } // end try
+      catch(...)
+      {
+        // catch anything
+      } // end catch
+    } // end deallocate()
+
+    inline bool operator==(no_throw_allocator const &other) { return super_t::operator==(other); }
+    inline bool operator!=(no_throw_allocator const &other) { return super_t::operator!=(other); }
+}; // end no_throw_allocator
+
+} // end detail
+} // end thrust
+
+
diff --git a/compat/thrust/detail/allocator/tagged_allocator.h b/compat/thrust/detail/allocator/tagged_allocator.h
new file mode 100644
index 0000000000..3cb87a32bf
--- /dev/null
+++ b/compat/thrust/detail/allocator/tagged_allocator.h
@@ -0,0 +1,101 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/type_traits/pointer_traits.h>
+#include <thrust/iterator/iterator_traits.h>
+
+namespace thrust
+{
+namespace detail
+{
+
+template<typename T, typename Tag, typename Pointer> class tagged_allocator;
+
+template<typename Tag, typename Pointer>
+  class tagged_allocator<void, Tag, Pointer>
+{
+  public:
+    typedef void                                                                                 value_type;
+    typedef typename thrust::detail::pointer_traits<Pointer>::template rebind<void>::other       pointer;
+    typedef typename thrust::detail::pointer_traits<Pointer>::template rebind<const void>::other const_pointer;
+    typedef std::size_t                                                                          size_type;
+    typedef typename thrust::detail::pointer_traits<Pointer>::difference_type                    difference_type;
+    typedef Tag                                                                                  system_type;
+
+    template<typename U>
+      struct rebind
+    {
+      typedef tagged_allocator<U,Tag,Pointer> other;
+    }; // end rebind
+};
+
+template<typename T, typename Tag, typename Pointer>
+  class tagged_allocator
+{
+  public:
+    typedef T                                                                                 value_type;
+    typedef typename thrust::detail::pointer_traits<Pointer>::template rebind<T>::other       pointer;
+    typedef typename thrust::detail::pointer_traits<Pointer>::template rebind<const T>::other const_pointer;
+    typedef typename thrust::iterator_reference<pointer>::type                                reference;
+    typedef typename thrust::iterator_reference<const_pointer>::type                          const_reference;
+    typedef std::size_t                                                                       size_type;
+    typedef typename thrust::detail::pointer_traits<pointer>::difference_type                 difference_type;
+    typedef Tag                                                                               system_type;
+
+    template<typename U>
+      struct rebind
+    {
+      typedef tagged_allocator<U,Tag,Pointer> other;
+    }; // end rebind
+
+    __host__ __device__
+    inline tagged_allocator();
+
+    __host__ __device__
+    inline tagged_allocator(const tagged_allocator &);
+
+    template<typename U, typename OtherPointer>
+    __host__ __device__
+    inline tagged_allocator(const tagged_allocator<U, Tag, OtherPointer> &);
+
+    __host__ __device__
+    inline ~tagged_allocator();
+
+    __host__ __device__
+    pointer address(reference x) const;
+
+    __host__ __device__
+    const_pointer address(const_reference x) const;
+
+    size_type max_size() const;
+};
+
+template<typename T1, typename Pointer1, typename T2, typename Pointer2, typename Tag>
+__host__ __device__
+bool operator==(const tagged_allocator<T1,Pointer1,Tag> &, const tagged_allocator<T2,Pointer2,Tag> &);
+
+template<typename T1, typename Pointer1, typename T2, typename Pointer2, typename Tag>
+__host__ __device__
+bool operator!=(const tagged_allocator<T1,Pointer1,Tag> &, const tagged_allocator<T2,Pointer2,Tag> &);
+
+} // end detail
+} // end thrust
+
+#include <thrust/detail/allocator/tagged_allocator.inl>
+
diff --git a/compat/thrust/detail/allocator/tagged_allocator.inl b/compat/thrust/detail/allocator/tagged_allocator.inl
new file mode 100644
index 0000000000..cb362a840d
--- /dev/null
+++ b/compat/thrust/detail/allocator/tagged_allocator.inl
@@ -0,0 +1,97 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/allocator/tagged_allocator.h>
+#include <limits>
+
+namespace thrust
+{
+namespace detail
+{
+
+
+template<typename T, typename Tag, typename Pointer>
+  tagged_allocator<T,Tag,Pointer>
+    ::tagged_allocator()
+{}
+
+
+template<typename T, typename Tag, typename Pointer>
+  tagged_allocator<T,Tag,Pointer>
+    ::tagged_allocator(const tagged_allocator<T,Tag,Pointer> &)
+{}
+
+
+template<typename T, typename Tag, typename Pointer>
+  template<typename U, typename OtherPointer>
+    tagged_allocator<T,Tag,Pointer>
+      ::tagged_allocator(const tagged_allocator<U,Tag,OtherPointer> &)
+{}
+
+
+template<typename T, typename Tag, typename Pointer>
+  tagged_allocator<T,Tag,Pointer>
+    ::~tagged_allocator()
+{}
+
+
+template<typename T, typename Tag, typename Pointer>
+  typename tagged_allocator<T,Tag,Pointer>::pointer
+    tagged_allocator<T,Tag,Pointer>
+      ::address(reference x) const
+{
+  return &x;
+}
+
+
+template<typename T, typename Tag, typename Pointer>
+  typename tagged_allocator<T,Tag,Pointer>::const_pointer
+    tagged_allocator<T,Tag,Pointer>
+      ::address(const_reference x) const
+{
+  return &x;
+}
+
+
+template<typename T, typename Tag, typename Pointer>
+  typename tagged_allocator<T,Tag,Pointer>::size_type
+    tagged_allocator<T,Tag,Pointer>
+      ::max_size() const
+{
+  return (std::numeric_limits<size_type>::max)() / sizeof(T);
+}
+
+
+template<typename T1, typename Pointer1, typename T2, typename Pointer2, typename Tag>
+__host__ __device__
+bool operator==(const tagged_allocator<T1,Pointer1,Tag> &, const tagged_allocator<T2,Pointer2,Tag> &)
+{
+  return true;
+}
+
+
+template<typename T1, typename Pointer1, typename T2, typename Pointer2, typename Tag>
+__host__ __device__
+bool operator!=(const tagged_allocator<T1,Pointer1,Tag> &, const tagged_allocator<T2,Pointer2,Tag> &)
+{
+  return false;
+}
+    
+
+} // end detail
+} // end thrust
+
diff --git a/compat/thrust/detail/allocator/temporary_allocator.h b/compat/thrust/detail/allocator/temporary_allocator.h
new file mode 100644
index 0000000000..f0496f9fe1
--- /dev/null
+++ b/compat/thrust/detail/allocator/temporary_allocator.h
@@ -0,0 +1,75 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/allocator/tagged_allocator.h>
+#include <thrust/detail/allocator/allocator_traits.h>
+#include <thrust/pair.h>
+#include <thrust/memory.h>
+#include <thrust/detail/execution_policy.h>
+
+namespace thrust
+{
+namespace detail
+{
+
+
+// XXX the pointer parameter given to tagged_allocator should be related to
+//     the type of the expression get_temporary_buffer(system, n).first
+//     without decltype, compromise on pointer<T,System>
+template<typename T, typename System>
+  class temporary_allocator
+    : public thrust::detail::tagged_allocator<
+               T, System, thrust::pointer<T,System>
+             >
+{
+  private:
+    typedef thrust::detail::tagged_allocator<
+      T, System, thrust::pointer<T,System>
+    > super_t;
+
+    System &m_system;
+
+  public:
+    typedef typename super_t::pointer   pointer;
+    typedef typename super_t::size_type size_type;
+
+    inline explicit temporary_allocator(thrust::execution_policy<System> &system) :
+      super_t(),
+      m_system(thrust::detail::derived_cast(system))
+    {}
+
+    pointer allocate(size_type cnt);
+
+    void deallocate(pointer p, size_type n);
+
+    inline System &system()
+    {
+      return m_system;
+    } // end system()
+
+  private:
+    typedef thrust::pair<pointer, size_type> pointer_and_size;
+}; // end temporary_allocator
+
+
+} // end detail
+} // end thrust
+
+#include <thrust/detail/allocator/temporary_allocator.inl>
+
diff --git a/compat/thrust/detail/allocator/temporary_allocator.inl b/compat/thrust/detail/allocator/temporary_allocator.inl
new file mode 100644
index 0000000000..63221d57e9
--- /dev/null
+++ b/compat/thrust/detail/allocator/temporary_allocator.inl
@@ -0,0 +1,59 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/allocator/temporary_allocator.h>
+#include <thrust/detail/temporary_buffer.h>
+#include <thrust/system/detail/bad_alloc.h>
+
+namespace thrust
+{
+namespace detail
+{
+
+
+template<typename T, typename System>
+  typename temporary_allocator<T,System>::pointer
+    temporary_allocator<T,System>
+      ::allocate(typename temporary_allocator<T,System>::size_type cnt)
+{
+  pointer_and_size result = thrust::get_temporary_buffer<T>(system(), cnt);
+
+  // handle failure
+  if(result.second < cnt)
+  {
+    // deallocate and throw
+    // note that we pass cnt to deallocate, not a value derived from result.second
+    deallocate(result.first, cnt);
+
+    throw thrust::system::detail::bad_alloc("temporary_buffer::allocate: get_temporary_buffer failed");
+  } // end if
+
+  return result.first;
+} // end temporary_allocator::allocate()
+
+
+template<typename T, typename System>
+  void temporary_allocator<T,System>
+    ::deallocate(typename temporary_allocator<T,System>::pointer p, typename temporary_allocator<T,System>::size_type n)
+{
+  return thrust::return_temporary_buffer(system(), p);
+} // end temporary_allocator
+
+
+} // end detail
+} // end thrust
+
diff --git a/compat/thrust/detail/binary_search.inl b/compat/thrust/detail/binary_search.inl
new file mode 100644
index 0000000000..0fd799a311
--- /dev/null
+++ b/compat/thrust/detail/binary_search.inl
@@ -0,0 +1,458 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file binary_search.inl
+ *  \brief Inline file for binary_search.h.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/binary_search.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/system/detail/generic/select_system.h>
+#include <thrust/system/detail/generic/binary_search.h>
+#include <thrust/system/detail/adl/binary_search.h>
+
+namespace thrust
+{
+
+
+template <typename DerivedPolicy, typename ForwardIterator, typename LessThanComparable>
+ForwardIterator lower_bound(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                            ForwardIterator first,
+                            ForwardIterator last,
+                            const LessThanComparable &value)
+{
+    using thrust::system::detail::generic::lower_bound;
+    return lower_bound(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, value);
+}
+
+
+template<typename DerivedPolicy, typename ForwardIterator, typename T, typename StrictWeakOrdering>
+ForwardIterator lower_bound(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                            ForwardIterator first,
+                            ForwardIterator last,
+                            const T &value,
+                            StrictWeakOrdering comp)
+{
+    using thrust::system::detail::generic::lower_bound;
+    return lower_bound(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, value, comp);
+}
+
+
+template<typename DerivedPolicy, typename ForwardIterator, typename LessThanComparable>
+ForwardIterator upper_bound(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                            ForwardIterator first,
+                            ForwardIterator last,
+                            const LessThanComparable &value)
+{
+    using thrust::system::detail::generic::upper_bound;
+    return upper_bound(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, value);
+}
+
+
+template<typename DerivedPolicy, typename ForwardIterator, typename T, typename StrictWeakOrdering>
+ForwardIterator upper_bound(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                            ForwardIterator first,
+                            ForwardIterator last,
+                            const T &value,
+                            StrictWeakOrdering comp)
+{
+    using thrust::system::detail::generic::upper_bound;
+    return upper_bound(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, value, comp);
+}
+
+
+template <typename DerivedPolicy, typename ForwardIterator, typename LessThanComparable>
+bool binary_search(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                   ForwardIterator first, 
+                   ForwardIterator last,
+                   const LessThanComparable& value)
+{
+    using thrust::system::detail::generic::binary_search;
+    return binary_search(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, value);
+}
+
+
+template <typename DerivedPolicy, typename ForwardIterator, typename T, typename StrictWeakOrdering>
+bool binary_search(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                   ForwardIterator first,
+                   ForwardIterator last,
+                   const T& value, 
+                   StrictWeakOrdering comp)
+{
+    using thrust::system::detail::generic::binary_search;
+    return binary_search(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, value, comp);
+}
+
+
+template <typename DerivedPolicy, typename ForwardIterator, typename T, typename StrictWeakOrdering>
+thrust::pair<ForwardIterator, ForwardIterator>
+equal_range(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+            ForwardIterator first,
+            ForwardIterator last,
+            const T& value,
+            StrictWeakOrdering comp)
+{
+    using thrust::system::detail::generic::equal_range;
+    return equal_range(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, value, comp);
+}
+
+
+template <typename DerivedPolicy, typename ForwardIterator, typename LessThanComparable>
+thrust::pair<ForwardIterator, ForwardIterator>
+equal_range(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+            ForwardIterator first,
+            ForwardIterator last,
+            const LessThanComparable& value)
+{
+    using thrust::system::detail::generic::equal_range;
+    return equal_range(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, value);
+}
+
+
+template <typename DerivedPolicy, typename ForwardIterator, typename InputIterator, typename OutputIterator>
+OutputIterator lower_bound(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                           ForwardIterator first, 
+                           ForwardIterator last,
+                           InputIterator values_first, 
+                           InputIterator values_last,
+                           OutputIterator output)
+{
+    using thrust::system::detail::generic::lower_bound;
+    return lower_bound(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, values_first, values_last, output);
+}
+
+
+template <typename DerivedPolicy, typename ForwardIterator, typename InputIterator, typename OutputIterator, typename StrictWeakOrdering>
+OutputIterator lower_bound(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                           ForwardIterator first, 
+                           ForwardIterator last,
+                           InputIterator values_first, 
+                           InputIterator values_last,
+                           OutputIterator output,
+                           StrictWeakOrdering comp)
+{
+    using thrust::system::detail::generic::lower_bound;
+    return lower_bound(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, values_first, values_last, output, comp);
+}
+
+
+template <typename DerivedPolicy, typename ForwardIterator, typename InputIterator, typename OutputIterator>
+OutputIterator upper_bound(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                           ForwardIterator first, 
+                           ForwardIterator last,
+                           InputIterator values_first, 
+                           InputIterator values_last,
+                           OutputIterator output)
+{
+    using thrust::system::detail::generic::upper_bound;
+    return upper_bound(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, values_first, values_last, output);
+}
+
+
+template <typename DerivedPolicy, typename ForwardIterator, typename InputIterator, typename OutputIterator, typename StrictWeakOrdering>
+OutputIterator upper_bound(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                           ForwardIterator first, 
+                           ForwardIterator last,
+                           InputIterator values_first, 
+                           InputIterator values_last,
+                           OutputIterator output,
+                           StrictWeakOrdering comp)
+{
+    using thrust::system::detail::generic::upper_bound;
+    return upper_bound(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, values_first, values_last, output, comp);
+}
+
+
+template <typename DerivedPolicy, typename ForwardIterator, typename InputIterator, typename OutputIterator>
+OutputIterator binary_search(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                             ForwardIterator first, 
+                             ForwardIterator last,
+                             InputIterator values_first, 
+                             InputIterator values_last,
+                             OutputIterator output)
+{
+    using thrust::system::detail::generic::binary_search;
+    return binary_search(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, values_first, values_last, output);
+}
+
+
+template <typename DerivedPolicy, typename ForwardIterator, typename InputIterator, typename OutputIterator, typename StrictWeakOrdering>
+OutputIterator binary_search(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                             ForwardIterator first, 
+                             ForwardIterator last,
+                             InputIterator values_first, 
+                             InputIterator values_last,
+                             OutputIterator output,
+                             StrictWeakOrdering comp)
+{
+    using thrust::system::detail::generic::binary_search;
+    return binary_search(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, values_first, values_last, output, comp);
+}
+
+
+//////////////////////
+// Scalar Functions //
+//////////////////////
+
+template <typename ForwardIterator, typename LessThanComparable>
+ForwardIterator lower_bound(ForwardIterator first, 
+                            ForwardIterator last,
+                            const LessThanComparable& value)
+{
+    using thrust::system::detail::generic::select_system;
+
+    typedef typename thrust::iterator_system<ForwardIterator>::type System; 
+
+    System system;
+
+    return thrust::lower_bound(select_system(system), first, last, value);
+}
+
+template <typename ForwardIterator, typename T, typename StrictWeakOrdering>
+ForwardIterator lower_bound(ForwardIterator first,
+                            ForwardIterator last,
+                            const T& value, 
+                            StrictWeakOrdering comp)
+{
+    using thrust::system::detail::generic::select_system;
+
+    typedef typename thrust::iterator_system<ForwardIterator>::type System; 
+
+    System system;
+
+    return thrust::lower_bound(select_system(system), first, last, value, comp);
+}
+
+template <typename ForwardIterator, typename LessThanComparable>
+ForwardIterator upper_bound(ForwardIterator first, 
+                            ForwardIterator last,
+                            const LessThanComparable& value)
+{
+    using thrust::system::detail::generic::select_system;
+
+    typedef typename thrust::iterator_system<ForwardIterator>::type System;
+
+    System system;
+
+    return thrust::upper_bound(select_system(system), first, last, value);
+}
+
+template <typename ForwardIterator, typename T, typename StrictWeakOrdering>
+ForwardIterator upper_bound(ForwardIterator first,
+                            ForwardIterator last,
+                            const T& value, 
+                            StrictWeakOrdering comp)
+{
+    using thrust::system::detail::generic::select_system;
+
+    typedef typename thrust::iterator_system<ForwardIterator>::type System;
+
+    System system;
+
+    return thrust::upper_bound(select_system(system), first, last, value, comp);
+}
+
+template <typename ForwardIterator, typename LessThanComparable>
+bool binary_search(ForwardIterator first, 
+                   ForwardIterator last,
+                   const LessThanComparable& value)
+{
+    using thrust::system::detail::generic::select_system;
+
+    typedef typename thrust::iterator_system<ForwardIterator>::type System;
+
+    System system;
+
+    return thrust::binary_search(select_system(system), first, last, value);
+}
+
+template <typename ForwardIterator, typename T, typename StrictWeakOrdering>
+bool binary_search(ForwardIterator first,
+                   ForwardIterator last,
+                   const T& value, 
+                   StrictWeakOrdering comp)
+{
+    using thrust::system::detail::generic::select_system;
+
+    typedef typename thrust::iterator_system<ForwardIterator>::type System;
+
+    System system;
+
+    return thrust::binary_search(select_system(system), first, last, value, comp);
+}
+
+template <typename ForwardIterator, typename LessThanComparable>
+thrust::pair<ForwardIterator, ForwardIterator>
+equal_range(ForwardIterator first,
+            ForwardIterator last,
+            const LessThanComparable& value)
+{
+    using thrust::system::detail::generic::select_system;
+
+    typedef typename thrust::iterator_system<ForwardIterator>::type System;
+
+    System system;
+
+    return thrust::equal_range(select_system(system), first, last, value);
+}
+
+template <typename ForwardIterator, typename T, typename StrictWeakOrdering>
+thrust::pair<ForwardIterator, ForwardIterator>
+equal_range(ForwardIterator first,
+            ForwardIterator last,
+            const T& value,
+            StrictWeakOrdering comp)
+{
+    using thrust::system::detail::generic::select_system;
+
+    typedef typename thrust::iterator_system<ForwardIterator>::type System;
+
+    System system;
+
+    return thrust::equal_range(select_system(system), first, last, value, comp);
+}
+
+//////////////////////
+// Vector Functions //
+//////////////////////
+
+template <typename ForwardIterator, typename InputIterator, typename OutputIterator>
+OutputIterator lower_bound(ForwardIterator first, 
+                           ForwardIterator last,
+                           InputIterator values_first, 
+                           InputIterator values_last,
+                           OutputIterator output)
+{
+    using thrust::system::detail::generic::select_system;
+
+    typedef typename thrust::iterator_system<ForwardIterator>::type System1;
+    typedef typename thrust::iterator_system<InputIterator>::type   System2;
+    typedef typename thrust::iterator_system<OutputIterator>::type  System3;
+
+    System1 system1;
+    System2 system2;
+    System3 system3;
+
+    return thrust::lower_bound(select_system(system1,system2,system3), first, last, values_first, values_last, output);
+}
+
+template <typename ForwardIterator, typename InputIterator, typename OutputIterator, typename StrictWeakOrdering>
+OutputIterator lower_bound(ForwardIterator first, 
+                           ForwardIterator last,
+                           InputIterator values_first, 
+                           InputIterator values_last,
+                           OutputIterator output,
+                           StrictWeakOrdering comp)
+{
+    using thrust::system::detail::generic::select_system;
+
+    typedef typename thrust::iterator_system<ForwardIterator>::type System1;
+    typedef typename thrust::iterator_system<InputIterator>::type   System2;
+    typedef typename thrust::iterator_system<OutputIterator>::type  System3;
+
+    System1 system1;
+    System2 system2;
+    System3 system3;
+
+    return thrust::lower_bound(select_system(system1,system2,system3), first, last, values_first, values_last, output, comp);
+}
+    
+template <typename ForwardIterator, typename InputIterator, typename OutputIterator>
+OutputIterator upper_bound(ForwardIterator first, 
+                           ForwardIterator last,
+                           InputIterator values_first, 
+                           InputIterator values_last,
+                           OutputIterator output)
+{
+    using thrust::system::detail::generic::select_system;
+
+    typedef typename thrust::iterator_system<ForwardIterator>::type System1;
+    typedef typename thrust::iterator_system<InputIterator>::type   System2;
+    typedef typename thrust::iterator_system<OutputIterator>::type  System3;
+
+    System1 system1;
+    System2 system2;
+    System3 system3;
+
+    return thrust::upper_bound(select_system(system1,system2,system3), first, last, values_first, values_last, output);
+}
+
+template <typename ForwardIterator, typename InputIterator, typename OutputIterator, typename StrictWeakOrdering>
+OutputIterator upper_bound(ForwardIterator first, 
+                           ForwardIterator last,
+                           InputIterator values_first, 
+                           InputIterator values_last,
+                           OutputIterator output,
+                           StrictWeakOrdering comp)
+{
+    using thrust::system::detail::generic::select_system;
+
+    typedef typename thrust::iterator_system<ForwardIterator>::type System1;
+    typedef typename thrust::iterator_system<InputIterator>::type   System2;
+    typedef typename thrust::iterator_system<OutputIterator>::type  System3;
+
+    System1 system1;
+    System2 system2;
+    System3 system3;
+
+    return thrust::upper_bound(select_system(system1,system2,system3), first, last, values_first, values_last, output, comp);
+}
+
+template <typename ForwardIterator, typename InputIterator, typename OutputIterator>
+OutputIterator binary_search(ForwardIterator first, 
+                             ForwardIterator last,
+                             InputIterator values_first, 
+                             InputIterator values_last,
+                             OutputIterator output)
+{
+    using thrust::system::detail::generic::select_system;
+
+    typedef typename thrust::iterator_system<ForwardIterator>::type System1;
+    typedef typename thrust::iterator_system<InputIterator>::type   System2;
+    typedef typename thrust::iterator_system<OutputIterator>::type  System3;
+
+    System1 system1;
+    System2 system2;
+    System3 system3;
+
+    return thrust::binary_search(select_system(system1,system2,system3), first, last, values_first, values_last, output);
+}
+
+template <typename ForwardIterator, typename InputIterator, typename OutputIterator, typename StrictWeakOrdering>
+OutputIterator binary_search(ForwardIterator first, 
+                             ForwardIterator last,
+                             InputIterator values_first, 
+                             InputIterator values_last,
+                             OutputIterator output,
+                             StrictWeakOrdering comp)
+{
+    using thrust::system::detail::generic::select_system;
+
+    typedef typename thrust::iterator_system<ForwardIterator>::type System1;
+    typedef typename thrust::iterator_system<InputIterator>::type   System2;
+    typedef typename thrust::iterator_system<OutputIterator>::type  System3;
+
+    System1 system1;
+    System2 system2;
+    System3 system3;
+
+    return thrust::binary_search(select_system(system1,system2,system3), first, last, values_first, values_last, output, comp);
+}
+
+} // end namespace thrust
+
diff --git a/compat/thrust/detail/config.h b/compat/thrust/detail/config.h
new file mode 100644
index 0000000000..d6b6691089
--- /dev/null
+++ b/compat/thrust/detail/config.h
@@ -0,0 +1,23 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+/*! \file config.h
+ *  \brief Defines platform configuration.
+ */
+
+#pragma once
+
+#include <thrust/detail/config/config.h>
+
diff --git a/compat/thrust/detail/config/compiler.h b/compat/thrust/detail/config/compiler.h
new file mode 100644
index 0000000000..90ce911155
--- /dev/null
+++ b/compat/thrust/detail/config/compiler.h
@@ -0,0 +1,103 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+/*! \file compiler.h
+ *  \brief Compiler-specific configuration
+ */
+
+#pragma once
+
+#ifdef __CUDACC__
+
+#include <cuda.h>
+
+// Thrust supports CUDA >= 3.0
+#if CUDA_VERSION < 3000
+#error "CUDA v3.0 or newer is required"
+#endif // CUDA_VERSION
+
+#endif // __CUDACC__
+
+// enumerate host compilers we know about
+#define THRUST_HOST_COMPILER_UNKNOWN 0
+#define THRUST_HOST_COMPILER_MSVC    1
+#define THRUST_HOST_COMPILER_GCC     2
+
+// enumerate host compilers we know about
+#define THRUST_DEVICE_COMPILER_UNKNOWN 0
+#define THRUST_DEVICE_COMPILER_MSVC    1
+#define THRUST_DEVICE_COMPILER_GCC     2
+#define THRUST_DEVICE_COMPILER_NVCC    3
+
+// figure out which host compiler we're using
+// XXX we should move the definition of THRUST_DEPRECATED out of this logic
+#if   defined(_MSC_VER)
+#define THRUST_HOST_COMPILER THRUST_HOST_COMPILER_MSVC
+#define THRUST_DEPRECATED __declspec(deprecated)
+#elif defined(__GNUC__)
+#define THRUST_HOST_COMPILER THRUST_HOST_COMPILER_GCC
+#define THRUST_DEPRECATED __attribute__ ((deprecated)) 
+#define THRUST_GCC_VERSION (__GNUC__ * 10000 + __GNUC_MINOR__ * 100 + __GNUC_PATCHLEVEL__)
+#else
+#define THRUST_HOST_COMPILER THRUST_HOST_COMPILER_UNKNOWN
+#define THRUST_DEPRECATED
+#endif // THRUST_HOST_COMPILER
+
+// figure out which device compiler we're using
+#if defined(__CUDACC__)
+#define THRUST_DEVICE_COMPILER THRUST_DEVICE_COMPILER_NVCC
+#elif THRUST_HOST_COMPILER == THRUST_HOST_COMPILER_MSVC
+#define THRUST_DEVICE_COMPILER THRUST_DEVICE_COMPILER_MSVC
+#elif THRUST_HOST_COMPILER == THRUST_HOST_COMPILER_GCC
+#define THRUST_DEVICE_COMPILER THRUST_DEVICE_COMPILER_GCC
+#else
+#define THRUST_DEVICE_COMPILER THRUST_DEVICE_COMPILER_UNKNOWN
+#endif
+
+// is the device compiler capable of compiling omp?
+#ifdef _OPENMP
+#define THRUST_DEVICE_COMPILER_IS_OMP_CAPABLE THRUST_TRUE
+#else
+#define THRUST_DEVICE_COMPILER_IS_OMP_CAPABLE THRUST_FALSE
+#endif // _OPENMP
+
+// disable specific MSVC warnings
+#if (THRUST_HOST_COMPILER == THRUST_HOST_COMPILER_MSVC) && !defined(__CUDA_ARCH__)
+#define __THRUST_DISABLE_MSVC_WARNING_BEGIN(x) \
+__pragma(warning(push)) \
+__pragma(warning(disable : x))
+#define __THRUST_DISABLE_MSVC_WARNING_END(x) \
+__pragma(warning(pop))
+#else
+#define __THRUST_DISABLE_MSVC_WARNING_BEGIN(x)
+#define __THRUST_DISABLE_MSVC_WARNING_END(x)
+#endif
+#define __THRUST_DISABLE_MSVC_POSSIBLE_LOSS_OF_DATA_WARNING(x) \
+__THRUST_DISABLE_MSVC_WARNING_BEGIN(4244 4267) \
+x;\
+__THRUST_DISABLE_MSVC_WARNING_END(4244 4267)
+#define __THRUST_DISABLE_MSVC_POSSIBLE_LOSS_OF_DATA_WARNING_BEGIN \
+__THRUST_DISABLE_MSVC_WARNING_BEGIN(4244 4267)
+#define __THRUST_DISABLE_MSVC_POSSIBLE_LOSS_OF_DATA_WARNING_END \
+__THRUST_DISABLE_MSVC_WARNING_END(4244 4267)
+#define __THRUST_DISABLE_MSVC_FORCING_VALUE_TO_BOOL(x) \
+__THRUST_DISABLE_MSVC_WARNING_BEGIN(4800) \
+x;\
+__THRUST_DISABLE_MSVC_WARNING_END(4800)
+#define __THRUST_DISABLE_MSVC_FORCING_VALUE_TO_BOOL_BEGIN \
+__THRUST_DISABLE_MSVC_WARNING_BEGIN(4800)
+#define __THRUST_DISABLE_MSVC_FORCING_VALUE_TO_BOOL_END \
+__THRUST_DISABLE_MSVC_WARNING_END(4800)
diff --git a/compat/thrust/detail/config/compiler_fence.h b/compat/thrust/detail/config/compiler_fence.h
new file mode 100644
index 0000000000..f5cbf98204
--- /dev/null
+++ b/compat/thrust/detail/config/compiler_fence.h
@@ -0,0 +1,52 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// msvc case
+#if THRUST_HOST_COMPILER == THRUST_HOST_COMPILER_MSVC
+
+#ifndef _DEBUG
+
+#include <intrin.h>
+#pragma intrinsic(_ReadWriteBarrier)
+#define __thrust_compiler_fence() _ReadWriteBarrier()
+#else
+
+#define __thrust_compiler_fence() do {} while (0)
+
+#endif // _DEBUG
+
+// gcc case
+#elif THRUST_HOST_COMPILER == THRUST_HOST_COMPILER_GCC
+
+#if THRUST_GCC_VERSION >= 40200 // atomic built-ins were introduced ~4.2
+#define __thrust_compiler_fence() __sync_synchronize()
+#else
+// allow the code to compile without any guarantees
+#define __thrust_compiler_fence() do {} while (0)
+#endif // THRUST_GCC_VERSION
+
+// unknown case
+#elif THRUST_HOST_COMPILER == THRUST_HOST_COMPILER_UNKNOWN
+
+// allow the code to compile without any guarantees
+#define __thrust_compiler_fence() do {} while (0)
+
+#endif
+
diff --git a/compat/thrust/detail/config/config.h b/compat/thrust/detail/config/config.h
new file mode 100644
index 0000000000..f3498acd14
--- /dev/null
+++ b/compat/thrust/detail/config/config.h
@@ -0,0 +1,36 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+/*! \file config.h
+ *  \brief Defines platform configuration.
+ */
+
+#pragma once
+
+// XXX the order of these #includes matters
+
+#include <thrust/detail/config/simple_defines.h>
+#include <thrust/detail/config/compiler.h>
+// host_system.h & device_system.h must be #included as early as possible
+// because other config headers depend on it
+#include <thrust/detail/config/host_system.h>
+#include <thrust/detail/config/device_system.h>
+#include <thrust/detail/config/host_device.h>
+#include <thrust/detail/config/debug.h>
+#include <thrust/detail/config/compiler_fence.h>
+#include <thrust/detail/config/forceinline.h>
+#include <thrust/detail/config/hd_warning_disable.h>
+
diff --git a/compat/thrust/detail/config/debug.h b/compat/thrust/detail/config/debug.h
new file mode 100644
index 0000000000..56c1bad207
--- /dev/null
+++ b/compat/thrust/detail/config/debug.h
@@ -0,0 +1,32 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#ifndef THRUST_DEBUG
+#  ifndef NDEBUG
+#    if (DEBUG || _DEBUG)
+#      define THRUST_DEBUG 1
+#    endif // (DEBUG || _DEBUG)
+#  endif // NDEBUG
+#endif // THRUST_DEBUG
+
+#if THRUST_DEBUG
+#  ifndef __THRUST_SYNCHRONOUS
+#    define __THRUST_SYNCHRONOUS 1
+#  endif // __THRUST_SYNCHRONOUS
+#endif // THRUST_DEBUG
+
diff --git a/compat/thrust/detail/config/device_system.h b/compat/thrust/detail/config/device_system.h
new file mode 100644
index 0000000000..a104906753
--- /dev/null
+++ b/compat/thrust/detail/config/device_system.h
@@ -0,0 +1,61 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+// reserve 0 for undefined
+#define THRUST_DEVICE_SYSTEM_CUDA    1
+#define THRUST_DEVICE_SYSTEM_OMP     2
+#define THRUST_DEVICE_SYSTEM_TBB     3
+#define THRUST_DEVICE_SYSTEM_CPP     4
+
+#ifndef THRUST_DEVICE_SYSTEM
+#define THRUST_DEVICE_SYSTEM THRUST_DEVICE_SYSTEM_CUDA
+#endif // THRUST_DEVICE_SYSTEM
+
+// XXX make the use of THRUST_DEVICE_BACKEND an error in Thrust 1.7
+// XXX eliminate the following in Thrust 1.7
+
+#define THRUST_DEVICE_BACKEND_CUDA THRUST_DEVICE_SYSTEM_CUDA
+#define THRUST_DEVICE_BACKEND_OMP  THRUST_DEVICE_SYSTEM_OMP
+#define THRUST_DEVICE_BACKEND_TBB  THRUST_DEVICE_SYSTEM_TBB
+
+#ifdef THRUST_DEVICE_BACKEND
+#  if THRUST_HOST_COMPILER == THRUST_HOST_COMPILER_MSVC
+#    pragma message("----------------------------------------------------------------------------------")
+#    pragma message("| WARNING: THRUST_DEVICE_BACKEND is deprecated; use THRUST_DEVICE_SYSTEM instead |")
+#    pragma message("----------------------------------------------------------------------------------")
+#  else
+#    warning ----------------------------------------------------------------------------------
+#    warning | WARNING: THRUST_DEVICE_BACKEND is deprecated; use THRUST_DEVICE_SYSTEM instead |
+#    warning ----------------------------------------------------------------------------------
+#  endif // THRUST_HOST_COMPILER
+#  undef THRUST_DEVICE_SYSTEM
+#  define THRUST_DEVICE_SYSTEM THRUST_DEVICE_BACKEND
+#endif // THRUST_DEVICE_BACKEND
+
+#if THRUST_DEVICE_SYSTEM == THRUST_DEVICE_SYSTEM_CUDA
+#define __THRUST_DEVICE_SYSTEM_NAMESPACE cuda
+#elif THRUST_DEVICE_SYSTEM == THRUST_DEVICE_SYSTEM_OMP
+#define __THRUST_DEVICE_SYSTEM_NAMESPACE omp
+#elif THRUST_DEVICE_SYSTEM == THRUST_DEVICE_SYSTEM_TBB
+#define __THRUST_DEVICE_SYSTEM_NAMESPACE tbb
+#elif THRUST_DEVICE_SYSTEM == THRUST_DEVICE_SYSTEM_CPP
+#define __THRUST_DEVICE_SYSTEM_NAMESPACE cpp
+#endif
+
+#define __THRUST_DEVICE_SYSTEM_ROOT thrust/system/__THRUST_DEVICE_SYSTEM_NAMESPACE
+
diff --git a/compat/thrust/detail/config/forceinline.h b/compat/thrust/detail/config/forceinline.h
new file mode 100644
index 0000000000..620769b999
--- /dev/null
+++ b/compat/thrust/detail/config/forceinline.h
@@ -0,0 +1,36 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+/*! \file forceinline.h
+ *  \brief Defines __thrust_forceinline__
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+#if defined(__CUDACC__)
+
+#define __thrust_forceinline__ __forceinline__
+
+#else
+
+// TODO add 
+
+#define __thrust_forceinline__
+
+#endif
+
diff --git a/compat/thrust/detail/config/hd_warning_disable.h b/compat/thrust/detail/config/hd_warning_disable.h
new file mode 100644
index 0000000000..b993ef2828
--- /dev/null
+++ b/compat/thrust/detail/config/hd_warning_disable.h
@@ -0,0 +1,35 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+/*! \file hd_warning_disable.h
+ *  \brief Defines __thrust_hd_warning_disable__
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+#if defined(__CUDACC__)
+
+#define __thrust_hd_warning_disable__ \
+#pragma hd_warning_disable
+#else
+
+#define __thrust_hd_warning_disable__
+
+#endif
+
+
diff --git a/compat/thrust/detail/config/host_device.h b/compat/thrust/detail/config/host_device.h
new file mode 100644
index 0000000000..5d0975d106
--- /dev/null
+++ b/compat/thrust/detail/config/host_device.h
@@ -0,0 +1,44 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+/*! \file host_device.h
+ *  \brief Defines __host__ and __device__ and other CUDA-isms
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+#if THRUST_DEVICE_SYSTEM == THRUST_DEVICE_SYSTEM_CUDA
+
+#include <host_defines.h>
+
+#else
+
+// since __host__ & __device__ might have already be defined, only
+// #define them if not defined already
+// XXX this will break if the client does #include <host_defines.h> later
+
+#ifndef __host__
+#define __host__
+#endif // __host__
+
+#ifndef __device__
+#define __device__
+#endif // __device__
+
+#endif
+
diff --git a/compat/thrust/detail/config/host_system.h b/compat/thrust/detail/config/host_system.h
new file mode 100644
index 0000000000..fb8edabc55
--- /dev/null
+++ b/compat/thrust/detail/config/host_system.h
@@ -0,0 +1,58 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+// reserve 0 for undefined
+#define THRUST_HOST_SYSTEM_CPP    1
+#define THRUST_HOST_SYSTEM_OMP    2
+#define THRUST_HOST_SYSTEM_TBB    3
+
+#ifndef THRUST_HOST_SYSTEM
+#define THRUST_HOST_SYSTEM THRUST_HOST_SYSTEM_CPP
+#endif // THRUST_HOST_SYSTEM
+
+// XXX make the use of THRUST_HOST_BACKEND an error in Thrust 1.7
+// XXX eliminate the following in Thrust 1.7
+
+#define THRUST_HOST_BACKEND_CPP THRUST_HOST_SYSTEM_CPP
+#define THRUST_HOST_BACKEND_OMP THRUST_HOST_SYSTEM_OMP
+#define THRUST_HOST_BACKEND_TBB THRUST_HOST_SYSTEM_TBB
+
+#ifdef THRUST_HOST_BACKEND
+#  if THRUST_HOST_COMPILER == THRUST_HOST_COMPILER_MSVC
+#    pragma message("------------------------------------------------------------------------------")
+#    pragma message("| WARNING: THRUST_HOST_BACKEND is deprecated; use THRUST_HOST_SYSTEM instead |")
+#    pragma message("------------------------------------------------------------------------------")
+#  else
+#    warning ------------------------------------------------------------------------------
+#    warning | WARNING: THRUST_HOST_BACKEND is deprecated; use THRUST_HOST_SYSTEM instead |
+#    warning ------------------------------------------------------------------------------
+#  endif // THRUST_HOST_COMPILER
+#  undef THRUST_HOST_SYSTEM
+#  define THRUST_HOST_SYSTEM THRUST_HOST_BACKEND
+#endif // THRUST_HOST_BACKEND
+
+#if THRUST_HOST_SYSTEM == THRUST_HOST_SYSTEM_CPP
+#define __THRUST_HOST_SYSTEM_NAMESPACE cpp
+#elif THRUST_HOST_SYSTEM == THRUST_HOST_SYSTEM_OMP
+#define __THRUST_HOST_SYSTEM_NAMESPACE omp
+#elif THRUST_HOST_SYSTEM == THRUST_HOST_SYSTEM_TBB
+#define __THRUST_HOST_SYSTEM_NAMESPACE tbb
+#endif
+
+#define __THRUST_HOST_SYSTEM_ROOT thrust/system/__THRUST_HOST_SYSTEM_NAMESPACE
+
diff --git a/compat/thrust/detail/config/simple_defines.h b/compat/thrust/detail/config/simple_defines.h
new file mode 100644
index 0000000000..f9510ee9ce
--- /dev/null
+++ b/compat/thrust/detail/config/simple_defines.h
@@ -0,0 +1,28 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+/*! \file simple_defines.h
+ *  \brief Primitive macros without dependencies.
+ */
+
+#pragma once
+
+#define THRUST_UNKNOWN 0
+#define THRUST_FALSE   0
+#define THRUST_TRUE    1
+
+#define THRUST_PREVENT_MACRO_SUBSTITUTION
+
diff --git a/compat/thrust/detail/contiguous_storage.h b/compat/thrust/detail/contiguous_storage.h
new file mode 100644
index 0000000000..fe72bce2d2
--- /dev/null
+++ b/compat/thrust/detail/contiguous_storage.h
@@ -0,0 +1,129 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/iterator/detail/normal_iterator.h>
+#include <thrust/detail/execution_policy.h>
+#include <thrust/detail/allocator/allocator_traits.h>
+
+namespace thrust
+{
+
+namespace detail
+{
+
+// XXX parameter T is redundant with parameter Alloc
+template<typename T, typename Alloc>
+  class contiguous_storage
+{
+  private:
+    typedef thrust::detail::allocator_traits<Alloc> alloc_traits;
+
+  public:
+    typedef Alloc                                      allocator_type;
+    typedef T                                          value_type;
+    typedef typename alloc_traits::pointer             pointer;
+    typedef typename alloc_traits::const_pointer       const_pointer;
+    typedef typename alloc_traits::size_type           size_type;
+    typedef typename alloc_traits::difference_type     difference_type;
+
+    // XXX we should bring reference & const_reference into allocator_traits
+    //     at the moment, it's unclear how -- we have nothing analogous to
+    //     rebind_pointer for references
+    //     we either need to add reference_traits or extend the existing
+    //     pointer_traits to support wrapped references
+    typedef typename Alloc::reference                  reference;
+    typedef typename Alloc::const_reference            const_reference;
+
+    typedef thrust::detail::normal_iterator<pointer>       iterator;
+    typedef thrust::detail::normal_iterator<const_pointer> const_iterator;
+
+    explicit contiguous_storage(const allocator_type &alloc = allocator_type());
+
+    explicit contiguous_storage(size_type n, const allocator_type &alloc = allocator_type());
+
+    ~contiguous_storage(void);
+
+    size_type size(void) const;
+
+    size_type max_size(void) const;
+
+    iterator begin(void);
+    
+    const_iterator begin(void) const;
+
+    iterator end(void);
+
+    const_iterator end(void) const;
+
+    reference operator[](size_type n);
+
+    const_reference operator[](size_type n) const;
+
+    allocator_type get_allocator(void) const;
+
+    // note that allocate does *not* automatically call deallocate
+    void allocate(size_type n);
+
+    void deallocate(void);
+
+    void swap(contiguous_storage &x);
+
+    void default_construct_n(iterator first, size_type n);
+
+    void uninitialized_fill_n(iterator first, size_type n, const value_type &value);
+
+    template<typename InputIterator>
+    iterator uninitialized_copy(InputIterator first, InputIterator last, iterator result);
+
+    template<typename System, typename InputIterator>
+    iterator uninitialized_copy(thrust::execution_policy<System> &from_system,
+                                InputIterator first,
+                                InputIterator last,
+                                iterator result);
+
+    template<typename InputIterator, typename Size>
+    iterator uninitialized_copy_n(InputIterator first, Size n, iterator result);
+
+    template<typename System, typename InputIterator, typename Size>
+    iterator uninitialized_copy_n(thrust::execution_policy<System> &from_system,
+                                  InputIterator first,
+                                  Size n,
+                                  iterator result);
+
+    void destroy(iterator first, iterator last);
+
+  private:
+    // XXX we could inherit from this to take advantage of empty base class optimization
+    allocator_type m_allocator;
+
+    iterator m_begin;
+    
+    size_type m_size;
+
+    // disallow assignment
+    contiguous_storage &operator=(const contiguous_storage &x);
+}; // end contiguous_storage
+
+} // end detail
+
+template<typename T, typename Alloc> void swap(detail::contiguous_storage<T,Alloc> &lhs, detail::contiguous_storage<T,Alloc> &rhs);
+
+} // end thrust
+
+#include <thrust/detail/contiguous_storage.inl>
+
diff --git a/compat/thrust/detail/contiguous_storage.inl b/compat/thrust/detail/contiguous_storage.inl
new file mode 100644
index 0000000000..7e26c26b1a
--- /dev/null
+++ b/compat/thrust/detail/contiguous_storage.inl
@@ -0,0 +1,245 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/contiguous_storage.h>
+#include <thrust/detail/swap.h>
+#include <thrust/detail/allocator/allocator_traits.h>
+#include <thrust/detail/allocator/copy_construct_range.h>
+#include <thrust/detail/allocator/default_construct_range.h>
+#include <thrust/detail/allocator/destroy_range.h>
+#include <thrust/detail/allocator/fill_construct_range.h>
+#include <utility> // for use of std::swap in the WAR below
+
+namespace thrust
+{
+
+namespace detail
+{
+
+template<typename T, typename Alloc>
+  contiguous_storage<T,Alloc>
+    ::contiguous_storage(const Alloc &alloc)
+      :m_allocator(alloc),
+       m_begin(pointer(static_cast<T*>(0))),
+       m_size(0)
+{
+  ;
+} // end contiguous_storage::contiguous_storage()
+
+template<typename T, typename Alloc>
+  contiguous_storage<T,Alloc>
+    ::contiguous_storage(size_type n, const Alloc &alloc)
+      :m_allocator(alloc),
+       m_begin(pointer(static_cast<T*>(0))),
+       m_size(0)
+{
+  allocate(n);
+} // end contiguous_storage::contiguous_storage()
+
+template<typename T, typename Alloc>
+  contiguous_storage<T,Alloc>
+    ::~contiguous_storage(void)
+{
+  deallocate();
+} // end contiguous_storage::~contiguous_storage()
+
+template<typename T, typename Alloc>
+  typename contiguous_storage<T,Alloc>::size_type
+    contiguous_storage<T,Alloc>
+      ::size(void) const
+{
+  return m_size;
+} // end contiguous_storage::size()
+
+template<typename T, typename Alloc>
+  typename contiguous_storage<T,Alloc>::size_type
+    contiguous_storage<T,Alloc>
+      ::max_size(void) const
+{
+  return alloc_traits::max_size(m_allocator);
+} // end contiguous_storage::max_size()
+
+template<typename T, typename Alloc>
+  typename contiguous_storage<T,Alloc>::iterator
+    contiguous_storage<T,Alloc>
+      ::begin(void)
+{
+  return m_begin;
+} // end contiguous_storage::begin()
+
+template<typename T, typename Alloc>
+  typename contiguous_storage<T,Alloc>::const_iterator
+    contiguous_storage<T,Alloc>
+      ::begin(void) const
+{
+  return m_begin;
+} // end contiguous_storage::begin()
+
+template<typename T, typename Alloc>
+  typename contiguous_storage<T,Alloc>::iterator
+    contiguous_storage<T,Alloc>
+      ::end(void)
+{
+  return m_begin + size();
+} // end contiguous_storage::end()
+
+template<typename T, typename Alloc>
+  typename contiguous_storage<T,Alloc>::const_iterator
+    contiguous_storage<T,Alloc>
+      ::end(void) const
+{
+  return m_begin + size();
+} // end contiguous_storage::end()
+
+template<typename T, typename Alloc>
+  typename contiguous_storage<T,Alloc>::reference
+    contiguous_storage<T,Alloc>
+      ::operator[](size_type n)
+{
+  return m_begin[n];
+} // end contiguous_storage::operator[]()
+
+template<typename T, typename Alloc>
+  typename contiguous_storage<T,Alloc>::const_reference
+    contiguous_storage<T,Alloc>
+      ::operator[](size_type n) const
+{
+  return m_begin[n];
+} // end contiguous_storage::operator[]()
+
+template<typename T, typename Alloc>
+  typename contiguous_storage<T,Alloc>::allocator_type
+    contiguous_storage<T,Alloc>
+      ::get_allocator(void) const
+{
+  return m_allocator;
+} // end contiguous_storage::get_allocator()
+
+template<typename T, typename Alloc>
+  void contiguous_storage<T,Alloc>
+    ::allocate(size_type n)
+{
+  if(n > 0)
+  {
+    m_begin = iterator(m_allocator.allocate(n));
+    m_size = n;
+  } // end if
+  else
+  {
+    m_begin = iterator(pointer(static_cast<T*>(0)));
+    m_size = 0;
+  } // end else
+} // end contiguous_storage::allocate()
+
+template<typename T, typename Alloc>
+  void contiguous_storage<T,Alloc>
+    ::deallocate(void)
+{
+  if(size() > 0)
+  {
+    m_allocator.deallocate(m_begin.base(), size());
+    m_begin = iterator(pointer(static_cast<T*>(0)));
+    m_size = 0;
+  } // end if
+} // end contiguous_storage::deallocate()
+
+template<typename T, typename Alloc>
+  void contiguous_storage<T,Alloc>
+    ::swap(contiguous_storage &x)
+{
+  thrust::swap(m_begin, x.m_begin);
+  thrust::swap(m_size, x.m_size);
+
+  // XXX WAR nvcc 4.0's "calling a __host__ function from a __host__ __device__ function is not allowed" warning
+  //thrust::swap(m_allocator, x.m_allocator);
+  std::swap(m_allocator, x.m_allocator);
+} // end contiguous_storage::swap()
+
+template<typename T, typename Alloc>
+  void contiguous_storage<T,Alloc>
+    ::default_construct_n(iterator first, size_type n)
+{
+  default_construct_range(m_allocator, first.base(), n);
+} // end contiguous_storage::default_construct_n()
+
+template<typename T, typename Alloc>
+  void contiguous_storage<T,Alloc>
+    ::uninitialized_fill_n(iterator first, size_type n, const value_type &x)
+{
+  fill_construct_range(m_allocator, first.base(), n, x);
+} // end contiguous_storage::uninitialized_fill()
+
+template<typename T, typename Alloc>
+  template<typename System, typename InputIterator>
+    typename contiguous_storage<T,Alloc>::iterator
+      contiguous_storage<T,Alloc>
+        ::uninitialized_copy(thrust::execution_policy<System> &from_system, InputIterator first, InputIterator last, iterator result)
+{
+  return iterator(copy_construct_range(from_system, m_allocator, first, last, result.base()));
+} // end contiguous_storage::uninitialized_copy()
+
+template<typename T, typename Alloc>
+  template<typename InputIterator>
+    typename contiguous_storage<T,Alloc>::iterator
+      contiguous_storage<T,Alloc>
+        ::uninitialized_copy(InputIterator first, InputIterator last, iterator result)
+{
+  // XXX assumes InputIterator's associated System is default-constructible
+  typename thrust::iterator_system<InputIterator>::type from_system;
+
+  return iterator(copy_construct_range(from_system, m_allocator, first, last, result.base()));
+} // end contiguous_storage::uninitialized_copy()
+
+template<typename T, typename Alloc>
+  template<typename System, typename InputIterator, typename Size>
+    typename contiguous_storage<T,Alloc>::iterator
+      contiguous_storage<T,Alloc>
+        ::uninitialized_copy_n(thrust::execution_policy<System> &from_system, InputIterator first, Size n, iterator result)
+{
+  return iterator(copy_construct_range_n(from_system, m_allocator, first, n, result.base()));
+} // end contiguous_storage::uninitialized_copy_n()
+
+template<typename T, typename Alloc>
+  template<typename InputIterator, typename Size>
+    typename contiguous_storage<T,Alloc>::iterator
+      contiguous_storage<T,Alloc>
+        ::uninitialized_copy_n(InputIterator first, Size n, iterator result)
+{
+  // XXX assumes InputIterator's associated System is default-constructible
+  typename thrust::iterator_system<InputIterator>::type from_system;
+
+  return iterator(copy_construct_range_n(from_system, m_allocator, first, n, result.base()));
+} // end contiguous_storage::uninitialized_copy_n()
+
+template<typename T, typename Alloc>
+  void contiguous_storage<T,Alloc>
+    ::destroy(iterator first, iterator last)
+{
+  destroy_range(m_allocator, first.base(), last - first);
+} // end contiguous_storage::destroy()
+
+} // end detail
+
+template<typename T, typename Alloc>
+  void swap(detail::contiguous_storage<T,Alloc> &lhs, detail::contiguous_storage<T,Alloc> &rhs)
+{
+  lhs.swap(rhs);
+} // end swap()
+
+} // end thrust
+
diff --git a/compat/thrust/detail/copy.h b/compat/thrust/detail/copy.h
new file mode 100644
index 0000000000..8ed3abd219
--- /dev/null
+++ b/compat/thrust/detail/copy.h
@@ -0,0 +1,87 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/execution_policy.h>
+
+namespace thrust
+{
+
+template<typename System,
+         typename InputIterator,
+         typename OutputIterator>
+  OutputIterator copy(const thrust::detail::execution_policy_base<System> &system,
+                      InputIterator first,
+                      InputIterator last,
+                      OutputIterator result);
+
+template<typename System,
+         typename InputIterator,
+         typename Size,
+         typename OutputIterator>
+  OutputIterator copy_n(const thrust::detail::execution_policy_base<System> &system,
+                        InputIterator first,
+                        Size n,
+                        OutputIterator result);
+
+template<typename InputIterator,
+         typename OutputIterator>
+  OutputIterator copy(InputIterator first,
+                      InputIterator last,
+                      OutputIterator result);
+
+template<typename InputIterator,
+         typename Size,
+         typename OutputIterator>
+  OutputIterator copy_n(InputIterator first,
+                        Size n,
+                        OutputIterator result);
+
+
+namespace detail
+{
+
+
+template<typename FromSystem,
+         typename ToSystem,
+         typename InputIterator,
+         typename OutputIterator>
+  OutputIterator two_system_copy(thrust::execution_policy<FromSystem> &from_system,
+                                 thrust::execution_policy<ToSystem>   &two_system,
+                                 InputIterator first,
+                                 InputIterator last,
+                                 OutputIterator result);
+
+
+template<typename FromSystem,
+         typename ToSystem,
+         typename InputIterator,
+         typename Size,
+         typename OutputIterator>
+  OutputIterator two_system_copy_n(thrust::execution_policy<FromSystem> &from_system,
+                                   thrust::execution_policy<ToSystem>   &two_system,
+                                   InputIterator first,
+                                   Size n,
+                                   OutputIterator result);
+
+
+} // end detail
+} // end thrust
+
+#include <thrust/detail/copy.inl>
+
diff --git a/compat/thrust/detail/copy.inl b/compat/thrust/detail/copy.inl
new file mode 100644
index 0000000000..9ac48074a0
--- /dev/null
+++ b/compat/thrust/detail/copy.inl
@@ -0,0 +1,124 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/copy.h>
+#include <thrust/system/detail/generic/select_system.h>
+#include <thrust/system/detail/generic/copy.h>
+#include <thrust/system/detail/adl/copy.h>
+
+namespace thrust
+{
+
+
+template<typename DerivedPolicy, typename InputIterator, typename OutputIterator>
+  OutputIterator copy(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                      InputIterator first,
+                      InputIterator last,
+                      OutputIterator result)
+{
+  using thrust::system::detail::generic::copy;
+  return copy(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, result);
+} // end copy()
+
+
+template<typename DerivedPolicy, typename InputIterator, typename Size, typename OutputIterator>
+  OutputIterator copy_n(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                        InputIterator first,
+                        Size n,
+                        OutputIterator result)
+{
+  using thrust::system::detail::generic::copy_n;
+  return copy_n(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, n, result);
+} // end copy_n()
+
+
+namespace detail
+{
+
+
+template<typename System1,
+         typename System2,
+         typename InputIterator,
+         typename OutputIterator>
+  OutputIterator two_system_copy(thrust::execution_policy<System1> &system1,
+                                 thrust::execution_policy<System2> &system2,
+                                 InputIterator first,
+                                 InputIterator last,
+                                 OutputIterator result)
+{
+  using thrust::system::detail::generic::select_system;
+
+  return thrust::copy(select_system(thrust::detail::derived_cast(thrust::detail::strip_const(system1)), thrust::detail::derived_cast(thrust::detail::strip_const(system2))), first, last, result);
+} // end two_system_copy()
+
+
+template<typename System1,
+         typename System2,
+         typename InputIterator,
+         typename Size,
+         typename OutputIterator>
+  OutputIterator two_system_copy_n(thrust::execution_policy<System1> &system1,
+                                   thrust::execution_policy<System2> &system2,
+                                   InputIterator first,
+                                   Size n,
+                                   OutputIterator result)
+{
+  using thrust::system::detail::generic::select_system;
+
+  return thrust::copy_n(select_system(thrust::detail::derived_cast(thrust::detail::strip_const(system1)), thrust::detail::derived_cast(thrust::detail::strip_const(system2))), first, n, result);
+} // end two_system_copy_n()
+
+
+} // end detail
+
+
+template<typename InputIterator,
+         typename OutputIterator>
+  OutputIterator copy(InputIterator first,
+                      InputIterator last,
+                      OutputIterator result)
+{
+  typedef typename thrust::iterator_system<InputIterator>::type  System1;
+  typedef typename thrust::iterator_system<OutputIterator>::type System2;
+
+  System1 system1;
+  System2 system2;
+
+  return thrust::detail::two_system_copy(system1, system2, first, last, result);
+} // end copy()
+
+
+template<typename InputIterator,
+         typename Size,
+         typename OutputIterator>
+  OutputIterator copy_n(InputIterator first,
+                        Size n,
+                        OutputIterator result)
+{
+  typedef typename thrust::iterator_system<InputIterator>::type  System1;
+  typedef typename thrust::iterator_system<OutputIterator>::type System2;
+
+  System1 system1;
+  System2 system2;
+
+  return thrust::detail::two_system_copy_n(system1, system2, first, n, result);
+} // end copy_n()
+
+
+} // end namespace thrust
+
diff --git a/compat/thrust/detail/copy_if.h b/compat/thrust/detail/copy_if.h
new file mode 100644
index 0000000000..54e1ef4027
--- /dev/null
+++ b/compat/thrust/detail/copy_if.h
@@ -0,0 +1,68 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/execution_policy.h>
+
+namespace thrust
+{
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator,
+         typename Predicate>
+  OutputIterator copy_if(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                         InputIterator first,
+                         InputIterator last,
+                         OutputIterator result,
+                         Predicate pred);
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename Predicate>
+  OutputIterator copy_if(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                         InputIterator1 first,
+                         InputIterator1 last,
+                         InputIterator2 stencil,
+                         OutputIterator result,
+                         Predicate pred);
+
+template<typename InputIterator,
+         typename OutputIterator,
+         typename Predicate>
+  OutputIterator copy_if(InputIterator first,
+                         InputIterator last,
+                         OutputIterator result,
+                         Predicate pred);
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename Predicate>
+  OutputIterator copy_if(InputIterator1 first,
+                         InputIterator1 last,
+                         InputIterator2 stencil,
+                         OutputIterator result,
+                         Predicate pred);
+
+} // end thrust
+
+#include <thrust/detail/copy_if.inl>
+
diff --git a/compat/thrust/detail/copy_if.inl b/compat/thrust/detail/copy_if.inl
new file mode 100644
index 0000000000..e443bb7d0b
--- /dev/null
+++ b/compat/thrust/detail/copy_if.inl
@@ -0,0 +1,105 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/copy_if.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/system/detail/generic/copy_if.h>
+#include <thrust/system/detail/generic/select_system.h>
+#include <thrust/system/detail/adl/copy_if.h>
+
+namespace thrust
+{
+
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator,
+         typename Predicate>
+  OutputIterator copy_if(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                         InputIterator first,
+                         InputIterator last,
+                         OutputIterator result,
+                         Predicate pred)
+{
+  using thrust::system::detail::generic::copy_if;
+  return copy_if(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, result, pred);
+} // end copy_if()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename Predicate>
+  OutputIterator copy_if(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                         InputIterator1 first,
+                         InputIterator1 last,
+                         InputIterator2 stencil,
+                         OutputIterator result,
+                         Predicate pred)
+{
+  using thrust::system::detail::generic::copy_if;
+  return copy_if(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, stencil, result, pred);
+} // end copy_if()
+
+
+template<typename InputIterator,
+         typename OutputIterator,
+         typename Predicate>
+  OutputIterator copy_if(InputIterator first,
+                         InputIterator last,
+                         OutputIterator result,
+                         Predicate pred)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator>::type  System1;
+  typedef typename thrust::iterator_system<OutputIterator>::type System2;
+
+  System1 system1;
+  System2 system2;
+
+  return thrust::copy_if(select_system(system1,system2), first, last, result, pred);
+} // end copy_if()
+
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename Predicate>
+  OutputIterator copy_if(InputIterator1 first,
+                         InputIterator1 last,
+                         InputIterator2 stencil,
+                         OutputIterator result,
+                         Predicate pred)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator1>::type System1;
+  typedef typename thrust::iterator_system<InputIterator2>::type System2;
+  typedef typename thrust::iterator_system<OutputIterator>::type System3;
+
+  System1 system1;
+  System2 system2;
+  System3 system3;
+
+  return thrust::copy_if(select_system(system1,system2,system3), first, last, stencil, result, pred);
+} // end copy_if()
+
+
+} // end thrust
+
diff --git a/compat/thrust/detail/count.inl b/compat/thrust/detail/count.inl
new file mode 100644
index 0000000000..d2856ae1ce
--- /dev/null
+++ b/compat/thrust/detail/count.inl
@@ -0,0 +1,80 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file count.inl
+ *  \brief Inline file for count.h.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/count.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/system/detail/generic/select_system.h>
+#include <thrust/system/detail/generic/count.h>
+#include <thrust/system/detail/adl/count.h>
+
+namespace thrust
+{
+
+
+template<typename DerivedPolicy, typename InputIterator, typename EqualityComparable>
+  typename thrust::iterator_traits<InputIterator>::difference_type
+    count(const thrust::detail::execution_policy_base<DerivedPolicy> &exec, InputIterator first, InputIterator last, const EqualityComparable& value)
+{
+  using thrust::system::detail::generic::count;
+  return count(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, value);
+} // end count()
+
+
+template<typename DerivedPolicy, typename InputIterator, typename Predicate>
+  typename thrust::iterator_traits<InputIterator>::difference_type
+    count_if(const thrust::detail::execution_policy_base<DerivedPolicy> &exec, InputIterator first, InputIterator last, Predicate pred)
+{
+  using thrust::system::detail::generic::count_if;
+  return count_if(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, pred);
+} // end count_if()
+
+
+template <typename InputIterator, typename EqualityComparable>
+typename thrust::iterator_traits<InputIterator>::difference_type
+count(InputIterator first, InputIterator last, const EqualityComparable& value)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator>::type System;
+
+  System system;
+
+  return thrust::count(select_system(system), first, last, value);
+} // end count()
+
+
+template <typename InputIterator, typename Predicate>
+typename thrust::iterator_traits<InputIterator>::difference_type
+count_if(InputIterator first, InputIterator last, Predicate pred)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator>::type System;
+
+  System system;
+
+  return thrust::count_if(select_system(system), first, last, pred);
+} // end count_if()
+
+
+} // end namespace thrust
+
diff --git a/compat/thrust/detail/cstdint.h b/compat/thrust/detail/cstdint.h
new file mode 100644
index 0000000000..25d30fd5b2
--- /dev/null
+++ b/compat/thrust/detail/cstdint.h
@@ -0,0 +1,79 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#if (THRUST_HOST_COMPILER == THRUST_HOST_COMPILER_GCC)
+#include <stdint.h>
+#endif
+
+namespace thrust
+{
+namespace detail
+{
+
+#if (THRUST_HOST_COMPILER == THRUST_HOST_COMPILER_MSVC)
+
+#if (_MSC_VER < 1300)
+   typedef signed   char     int8_t;
+   typedef signed   short    int16_t;
+   typedef signed   int      int32_t;
+   typedef unsigned char     uint8_t;
+   typedef unsigned short    uint16_t;
+   typedef unsigned int      uint32_t;
+#else
+   typedef signed   __int8   int8_t;
+   typedef signed   __int16  int16_t;
+   typedef signed   __int32  int32_t;
+   typedef unsigned __int8   uint8_t;
+   typedef unsigned __int16  uint16_t;
+   typedef unsigned __int32  uint32_t;
+#endif
+typedef signed   __int64     int64_t;
+typedef unsigned __int64     uint64_t;
+
+#else
+
+typedef ::int8_t   int8_t;
+typedef ::int16_t  int16_t;
+typedef ::int32_t  int32_t;
+typedef ::int64_t  int64_t;
+typedef ::uint8_t  uint8_t;
+typedef ::uint16_t uint16_t;
+typedef ::uint32_t uint32_t;
+typedef ::uint64_t uint64_t;
+
+#endif
+
+
+// an oracle to tell us how to define intptr_t
+template<int word_size = sizeof(void*)> struct divine_intptr_t;
+template<int word_size = sizeof(void*)> struct divine_uintptr_t;
+
+// 32b platforms
+template<>  struct divine_intptr_t<4>  {  typedef thrust::detail::int32_t  type; };
+template<>  struct divine_uintptr_t<4> {  typedef thrust::detail::uint32_t type; };
+
+// 64b platforms
+template<>  struct divine_intptr_t<8>  { typedef thrust::detail::int64_t  type; };
+template<>  struct divine_uintptr_t<8> { typedef thrust::detail::uint64_t type; };
+
+typedef divine_intptr_t<>::type   intptr_t;
+typedef divine_uintptr_t<>::type  uintptr_t;
+
+} // end detail
+} // end thrust
+
diff --git a/compat/thrust/detail/device_delete.inl b/compat/thrust/detail/device_delete.inl
new file mode 100644
index 0000000000..dd70d76891
--- /dev/null
+++ b/compat/thrust/detail/device_delete.inl
@@ -0,0 +1,47 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file device_delete.inl
+ *  \brief Inline file for device_delete.h.
+ */
+
+#include <thrust/device_delete.h>
+#include <thrust/device_free.h>
+#include <thrust/detail/allocator/destroy_range.h>
+
+namespace thrust
+{
+namespace detail
+{
+
+// define an empty allocator class to use below
+struct device_delete_allocator {};
+
+}
+
+template<typename T>
+  void device_delete(device_ptr<T> ptr,
+                     const size_t n)
+{
+  // we can use device_allocator to destroy the range
+  thrust::detail::device_delete_allocator a;
+  thrust::detail::destroy_range(a, ptr, n);
+  thrust::device_free(ptr);
+} // end device_delete()
+
+} // end thrust
+
diff --git a/compat/thrust/detail/device_free.inl b/compat/thrust/detail/device_free.inl
new file mode 100644
index 0000000000..ab8db9f09c
--- /dev/null
+++ b/compat/thrust/detail/device_free.inl
@@ -0,0 +1,44 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file device_free.inl
+ *  \brief Inline file for device_free.h.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/device_free.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/system/detail/generic/select_system.h>
+#include <thrust/detail/malloc_and_free.h>
+
+namespace thrust
+{
+
+void device_free(thrust::device_ptr<void> ptr)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef thrust::iterator_system< thrust::device_ptr<void> >::type system;
+
+  // XXX lower to select_system(system) here
+  system s;
+
+  thrust::free(s, ptr);
+} // end device_free()
+
+} // end thrust
+
diff --git a/compat/thrust/detail/device_malloc.inl b/compat/thrust/detail/device_malloc.inl
new file mode 100644
index 0000000000..76d0029993
--- /dev/null
+++ b/compat/thrust/detail/device_malloc.inl
@@ -0,0 +1,60 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file device_malloc.inl
+ *  \brief Inline file for device_malloc.h.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/device_malloc.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/system/detail/generic/select_system.h>
+#include <thrust/detail/malloc_and_free.h>
+
+namespace thrust
+{
+
+
+thrust::device_ptr<void> device_malloc(const std::size_t n)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef thrust::iterator_system< thrust::device_ptr<void> >::type system;
+
+  // XXX lower to select_system(system) here
+  system s;
+
+  return thrust::device_ptr<void>(thrust::malloc(s, n).get());
+} // end device_malloc()
+
+
+template<typename T>
+  thrust::device_ptr<T> device_malloc(const std::size_t n)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef thrust::iterator_system< thrust::device_ptr<void> >::type system;
+
+  // XXX lower to select_system(system) here
+  system s;
+
+  return thrust::device_ptr<T>(thrust::malloc<T>(s,n).get());
+} // end device_malloc()
+
+
+} // end thrust
+
diff --git a/compat/thrust/detail/device_new.inl b/compat/thrust/detail/device_new.inl
new file mode 100644
index 0000000000..1f00a97a94
--- /dev/null
+++ b/compat/thrust/detail/device_new.inl
@@ -0,0 +1,60 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file device_new.inl
+ *  \brief Inline file for device_new.h.
+ */
+
+#include <thrust/device_new.h>
+#include <thrust/device_malloc.h>
+#include <thrust/uninitialized_fill.h>
+
+namespace thrust
+{
+
+template<typename T>
+  device_ptr<T> device_new(device_ptr<void> p,
+                           const size_t n)
+{
+  // XXX TODO dispatch n null device constructors at p here
+  // in the meantime, dispatch 1 null host constructor here
+  // and dispatch n copy constructors
+  return device_new<T>(p, T(), n);
+} // end device_new()
+
+template<typename T>
+  device_ptr<T> device_new(device_ptr<void> p,
+                           const T &exemplar,
+                           const size_t n)
+{
+  device_ptr<T> result(reinterpret_cast<T*>(p.get()));
+
+  // run copy constructors at p here
+  thrust::uninitialized_fill(result, result + n, exemplar);
+  
+  return result;
+} // end device_new()
+
+template<typename T>
+  device_ptr<T> device_new(const size_t n)
+{
+  // call placement new
+  return device_new<T>(thrust::device_malloc<T>(n));
+} // end device_new()
+
+} // thrust
+
diff --git a/compat/thrust/detail/device_ptr.inl b/compat/thrust/detail/device_ptr.inl
new file mode 100644
index 0000000000..0afe8a19c1
--- /dev/null
+++ b/compat/thrust/detail/device_ptr.inl
@@ -0,0 +1,74 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file device_ptr.inl
+ *  \brief Inline file for device_ptr.h.
+ */
+
+#include <thrust/device_ptr.h>
+#include <thrust/device_reference.h>
+#include <iostream>
+
+#include <thrust/detail/type_traits.h>
+#include <thrust/iterator/iterator_traits.h>
+
+namespace thrust
+{
+
+template<typename T>
+  device_ptr<T> device_pointer_cast(T *ptr)
+{
+  return device_ptr<T>(ptr);
+} // end device_pointer_cast()
+
+template<typename T>
+  device_ptr<T> device_pointer_cast(const device_ptr<T> &ptr)
+{
+  return ptr;
+} // end device_pointer_cast()
+
+// output to ostream
+template<class E, class T, class Y>
+  std::basic_ostream<E, T> &operator<<(std::basic_ostream<E, T> &os, const device_ptr<Y> &p)
+{
+  return os << p.get();
+} // end operator<<()
+
+
+namespace detail
+{
+
+template<typename T>
+  struct is_device_ptr< thrust::device_ptr<T> >
+    : public true_type
+{
+}; // end is_device_ptr
+
+#if (THRUST_HOST_COMPILER == THRUST_HOST_COMPILER_MSVC) && (_MSC_VER <= 1400)
+// XXX WAR MSVC 2005 problem with correctly implementing
+//     pointer_raw_pointer for device_ptr by specializing it here
+template<typename T>
+  struct pointer_raw_pointer< thrust::device_ptr<T> >
+{
+  typedef typename device_ptr<T>::raw_pointer type;
+}; // end pointer_raw_pointer
+#endif
+
+
+} // end namespace detail
+} // end namespace thrust
+
diff --git a/compat/thrust/detail/device_reference.inl b/compat/thrust/detail/device_reference.inl
new file mode 100644
index 0000000000..ad5cb76688
--- /dev/null
+++ b/compat/thrust/detail/device_reference.inl
@@ -0,0 +1,53 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file device_reference.inl
+ *  \brief Inline file for device_reference.h.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/device_reference.h>
+
+namespace thrust
+{
+
+template<typename T>
+  template<typename OtherT>
+    device_reference<T> &
+      device_reference<T>
+        ::operator=(const device_reference<OtherT> &other)
+{
+  return super_t::operator=(other);
+} // end operator=()
+
+template<typename T>
+  device_reference<T> &
+    device_reference<T>
+      ::operator=(const value_type &x)
+{
+  return super_t::operator=(x);
+} // end operator=()
+
+template<typename T>
+__host__ __device__
+void swap(device_reference<T> &a, device_reference<T> &b)
+{
+  a.swap(b);
+} // end swap()
+
+} // end thrust
+
diff --git a/compat/thrust/detail/device_vector.inl b/compat/thrust/detail/device_vector.inl
new file mode 100644
index 0000000000..f6bafbaa5f
--- /dev/null
+++ b/compat/thrust/detail/device_vector.inl
@@ -0,0 +1,37 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file device_vector.inl
+ *  \brief Inline file for device_vector.h.
+ */
+
+#include <thrust/host_vector.h>
+
+namespace thrust
+{
+
+template<typename T, typename Alloc>
+  template<typename OtherT, typename OtherAlloc>
+    device_vector<T,Alloc>
+      ::device_vector(const host_vector<OtherT,OtherAlloc> &v)
+        :Parent(v)
+{
+  ;
+} // end device_vector::device_vector()
+
+} // end namespace thrust
+
diff --git a/compat/thrust/detail/dispatch/is_trivial_copy.h b/compat/thrust/detail/dispatch/is_trivial_copy.h
new file mode 100644
index 0000000000..2bedf1f716
--- /dev/null
+++ b/compat/thrust/detail/dispatch/is_trivial_copy.h
@@ -0,0 +1,59 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file trivial_copy.h
+ *  \brief Device implementations for copying memory between host and device.
+ */
+
+#pragma once
+
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/detail/type_traits.h>
+#include <thrust/iterator/detail/is_trivial_iterator.h>
+
+namespace thrust
+{
+
+namespace detail
+{
+
+namespace dispatch
+{
+
+
+// a trivial copy's iterator's value_types match,
+// the iterators themselves are normal_iterators
+// and the ToIterator's value_type has_trivial_assign
+template<typename FromIterator, typename ToIterator>
+  struct is_trivial_copy :
+    integral_constant<
+      bool,
+      is_same<
+        typename thrust::iterator_value<FromIterator>::type,
+        typename thrust::iterator_value<ToIterator>::type
+      >::value
+      && is_trivial_iterator<FromIterator>::value
+      && is_trivial_iterator<ToIterator>::value
+      && has_trivial_assign<typename thrust::iterator_value<ToIterator>::type>::value
+    > {};
+
+} // end namespace dispatch
+
+} // end namespace detail
+
+} // end namespace thrust
+
diff --git a/compat/thrust/detail/distance.inl b/compat/thrust/detail/distance.inl
new file mode 100644
index 0000000000..f37595f324
--- /dev/null
+++ b/compat/thrust/detail/distance.inl
@@ -0,0 +1,39 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file distance.inl
+ *  \brief Inline file for distance.h
+ */
+
+#include <thrust/advance.h>
+#include <thrust/system/detail/generic/distance.h>
+#include <thrust/iterator/iterator_traits.h>
+
+namespace thrust
+{
+
+
+template<typename InputIterator>
+  inline typename thrust::iterator_traits<InputIterator>::difference_type
+    distance(InputIterator first, InputIterator last)
+{
+  return thrust::system::detail::generic::distance(first, last);
+} // end distance()
+
+
+} // end namespace thrust
+
diff --git a/compat/thrust/detail/equal.inl b/compat/thrust/detail/equal.inl
new file mode 100644
index 0000000000..ca6fecccf1
--- /dev/null
+++ b/compat/thrust/detail/equal.inl
@@ -0,0 +1,82 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file equal.inl
+ *  \brief Inline file for equal.h.
+ */
+
+#include <thrust/equal.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/system/detail/generic/select_system.h>
+#include <thrust/system/detail/generic/equal.h>
+#include <thrust/system/detail/adl/equal.h>
+
+namespace thrust
+{
+
+
+template<typename System, typename InputIterator1, typename InputIterator2>
+bool equal(const thrust::detail::execution_policy_base<System> &system, InputIterator1 first1, InputIterator1 last1, InputIterator2 first2)
+{
+  using thrust::system::detail::generic::equal;
+  return equal(thrust::detail::derived_cast(thrust::detail::strip_const(system)), first1, last1, first2);
+} // end equal()
+
+
+template<typename System, typename InputIterator1, typename InputIterator2, typename BinaryPredicate>
+bool equal(const thrust::detail::execution_policy_base<System> &system, InputIterator1 first1, InputIterator1 last1, InputIterator2 first2, BinaryPredicate binary_pred)
+{
+  using thrust::system::detail::generic::equal;
+  return equal(thrust::detail::derived_cast(thrust::detail::strip_const(system)), first1, last1, first2, binary_pred);
+} // end equal()
+
+
+template <typename InputIterator1, typename InputIterator2>
+bool equal(InputIterator1 first1, InputIterator1 last1,
+           InputIterator2 first2)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator1>::type System1;
+  typedef typename thrust::iterator_system<InputIterator2>::type System2;
+
+  System1 system1;
+  System2 system2;
+
+  return thrust::equal(select_system(system1,system2), first1, last1, first2);
+}
+
+
+template <typename InputIterator1, typename InputIterator2, 
+          typename BinaryPredicate>
+bool equal(InputIterator1 first1, InputIterator1 last1,
+           InputIterator2 first2, BinaryPredicate binary_pred)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator1>::type System1;
+  typedef typename thrust::iterator_system<InputIterator2>::type System2;
+
+  System1 system1;
+  System2 system2;
+
+  return thrust::equal(select_system(system1,system2), first1, last1, first2, binary_pred);
+}
+
+
+} // end namespace thrust
+
diff --git a/compat/thrust/detail/execute_with_allocator.h b/compat/thrust/detail/execute_with_allocator.h
new file mode 100644
index 0000000000..9d3c1ba29b
--- /dev/null
+++ b/compat/thrust/detail/execute_with_allocator.h
@@ -0,0 +1,84 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/raw_pointer_cast.h>
+#include <thrust/detail/type_traits/pointer_traits.h>
+#include <thrust/detail/allocator/allocator_traits.h>
+#include <thrust/detail/util/blocking.h>
+#include <thrust/pair.h>
+
+namespace thrust
+{
+namespace detail
+{
+
+template<typename ToPointer, typename FromPointer>
+__host__ __device__
+ToPointer reinterpret_pointer_cast(FromPointer ptr)
+{
+  typedef typename thrust::detail::pointer_element<ToPointer>::type to_element;
+  return ToPointer(reinterpret_cast<to_element*>(thrust::raw_pointer_cast(ptr)));
+}
+
+
+template<typename Allocator, template <typename> class BaseSystem>
+  struct execute_with_allocator
+    : BaseSystem<execute_with_allocator<Allocator, BaseSystem> >
+{
+  Allocator &m_alloc;
+
+  execute_with_allocator(Allocator &alloc)
+    : m_alloc(alloc)
+  {}
+
+  template<typename T>
+    friend thrust::pair<T*,std::ptrdiff_t>
+      get_temporary_buffer(execute_with_allocator &system, std::ptrdiff_t n)
+  {
+    typedef typename thrust::detail::allocator_traits<Allocator> alloc_traits;
+    typedef typename alloc_traits::void_pointer                  void_pointer;
+    typedef typename alloc_traits::size_type                     size_type;
+    typedef typename alloc_traits::value_type                    value_type;
+
+    // how many elements of type value_type do we need to accomodate n elements of type T?
+    size_type num_elements = thrust::detail::util::divide_ri(sizeof(T) * n, sizeof(value_type));
+
+    // allocate that many
+    void_pointer ptr = alloc_traits::allocate(system.m_alloc, num_elements);
+
+    // return the pointer and the number of elements of type T allocated
+    return thrust::make_pair(thrust::detail::reinterpret_pointer_cast<T*>(ptr),n);
+  }
+
+  template<typename Pointer>
+    friend void return_temporary_buffer(execute_with_allocator &system, Pointer p)
+  {
+    typedef typename thrust::detail::allocator_traits<Allocator> alloc_traits;
+    typedef typename alloc_traits::pointer                       pointer;
+
+    // return the pointer to the allocator
+    pointer to_ptr = thrust::detail::reinterpret_pointer_cast<pointer>(p);
+    alloc_traits::deallocate(system.m_alloc, to_ptr, 0);
+  }
+};
+
+
+} // end detail
+} // end thrust
+
diff --git a/compat/thrust/detail/execution_policy.h b/compat/thrust/detail/execution_policy.h
new file mode 100644
index 0000000000..28e77f2e91
--- /dev/null
+++ b/compat/thrust/detail/execution_policy.h
@@ -0,0 +1,78 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+namespace thrust
+{
+namespace detail
+{
+
+
+// execution_policy_base serves as a guard against
+// inifinite recursion in thrust entry points:
+//
+// template<typename DerivedPolicy>
+// void foo(const thrust::detail::execution_policy_base<DerivedPolicy> &s)
+// {
+//   using thrust::system::detail::generic::foo;
+//
+//   foo(thrust::detail::derived_cast(thrust::detail::strip_const(s));
+// }
+//
+// foo is not recursive when
+// 1. DerivedPolicy is derived from thrust::execution_policy below
+// 2. generic::foo takes thrust::execution_policy as a parameter
+template<typename DerivedPolicy> struct execution_policy_base {};
+
+
+template<typename DerivedPolicy>
+__host__ __device__
+inline execution_policy_base<DerivedPolicy> &strip_const(const execution_policy_base<DerivedPolicy> &x)
+{
+  return const_cast<execution_policy_base<DerivedPolicy>&>(x);
+}
+
+
+template<typename DerivedPolicy>
+__host__ __device__
+inline DerivedPolicy &derived_cast(execution_policy_base<DerivedPolicy> &x)
+{
+  return static_cast<DerivedPolicy&>(x);
+}
+
+
+template<typename DerivedPolicy>
+__host__ __device__
+inline const DerivedPolicy &derived_cast(const execution_policy_base<DerivedPolicy> &x)
+{
+  return static_cast<const DerivedPolicy&>(x);
+}
+
+
+} // end detail
+
+
+template<typename DerivedPolicy>
+  struct execution_policy
+    : thrust::detail::execution_policy_base<DerivedPolicy>
+{};
+
+
+} // end thrust
+
diff --git a/compat/thrust/detail/extrema.inl b/compat/thrust/detail/extrema.inl
new file mode 100644
index 0000000000..4bcd0bde56
--- /dev/null
+++ b/compat/thrust/detail/extrema.inl
@@ -0,0 +1,160 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+#include <thrust/detail/config.h>
+#include <thrust/extrema.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/system/detail/generic/select_system.h>
+#include <thrust/system/detail/generic/extrema.h>
+#include <thrust/system/detail/adl/extrema.h>
+
+namespace thrust
+{
+
+
+template<typename DerivedPolicy, typename ForwardIterator>
+ForwardIterator min_element(const thrust::detail::execution_policy_base<DerivedPolicy> &exec, ForwardIterator first, ForwardIterator last)
+{
+  using thrust::system::detail::generic::min_element;
+  return min_element(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last);
+} // end min_element()
+
+
+template<typename DerivedPolicy, typename ForwardIterator, typename BinaryPredicate>
+ForwardIterator min_element(const thrust::detail::execution_policy_base<DerivedPolicy> &exec, ForwardIterator first, ForwardIterator last, BinaryPredicate comp)
+{
+  using thrust::system::detail::generic::min_element;
+  return min_element(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, comp);
+} // end min_element()
+
+
+template<typename DerivedPolicy, typename ForwardIterator>
+ForwardIterator max_element(const thrust::detail::execution_policy_base<DerivedPolicy> &exec, ForwardIterator first, ForwardIterator last)
+{
+  using thrust::system::detail::generic::max_element;
+  return max_element(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last);
+} // end max_element()
+
+
+template<typename DerivedPolicy, typename ForwardIterator, typename BinaryPredicate>
+ForwardIterator max_element(const thrust::detail::execution_policy_base<DerivedPolicy> &exec, ForwardIterator first, ForwardIterator last, BinaryPredicate comp)
+{
+  using thrust::system::detail::generic::max_element;
+  return max_element(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, comp);
+} // end max_element()
+
+
+template<typename DerivedPolicy, typename ForwardIterator>
+thrust::pair<ForwardIterator,ForwardIterator> minmax_element(const thrust::detail::execution_policy_base<DerivedPolicy> &exec, ForwardIterator first, ForwardIterator last)
+{
+  using thrust::system::detail::generic::minmax_element;
+  return minmax_element(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last);
+} // end minmax_element()
+
+
+template<typename DerivedPolicy, typename ForwardIterator, typename BinaryPredicate>
+thrust::pair<ForwardIterator,ForwardIterator> minmax_element(const thrust::detail::execution_policy_base<DerivedPolicy> &exec, ForwardIterator first, ForwardIterator last, BinaryPredicate comp)
+{
+  using thrust::system::detail::generic::minmax_element;
+  return minmax_element(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, comp);
+} // end minmax_element()
+
+
+template <typename ForwardIterator>
+ForwardIterator min_element(ForwardIterator first, ForwardIterator last)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<ForwardIterator>::type System;
+
+  System system;
+
+  return thrust::min_element(select_system(system), first, last);
+} // end min_element()
+
+
+template <typename ForwardIterator, typename BinaryPredicate>
+ForwardIterator min_element(ForwardIterator first, ForwardIterator last,
+                            BinaryPredicate comp)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<ForwardIterator>::type System;
+
+  System system;
+
+  return thrust::min_element(select_system(system), first, last, comp);
+} // end min_element()
+
+
+template <typename ForwardIterator>
+ForwardIterator max_element(ForwardIterator first, ForwardIterator last)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<ForwardIterator>::type System;
+
+  System system;
+
+  return thrust::max_element(select_system(system), first, last);
+} // end max_element()
+
+
+template <typename ForwardIterator, typename BinaryPredicate>
+ForwardIterator max_element(ForwardIterator first, ForwardIterator last,
+                            BinaryPredicate comp)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<ForwardIterator>::type System;
+
+  System system;
+
+  return thrust::max_element(select_system(system), first, last, comp);
+} // end max_element()
+
+
+template <typename ForwardIterator>
+thrust::pair<ForwardIterator,ForwardIterator> 
+minmax_element(ForwardIterator first, ForwardIterator last)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<ForwardIterator>::type System;
+
+  System system;
+
+  return thrust::minmax_element(select_system(system), first, last);
+} // end minmax_element()
+
+
+template <typename ForwardIterator, typename BinaryPredicate>
+thrust::pair<ForwardIterator,ForwardIterator> 
+minmax_element(ForwardIterator first, ForwardIterator last, BinaryPredicate comp)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<ForwardIterator>::type System;
+
+  System system;
+
+  return thrust::minmax_element(select_system(system), first, last, comp);
+} // end minmax_element()
+
+
+} // end namespace thrust
+
diff --git a/compat/thrust/detail/fill.inl b/compat/thrust/detail/fill.inl
new file mode 100644
index 0000000000..c60e4a059e
--- /dev/null
+++ b/compat/thrust/detail/fill.inl
@@ -0,0 +1,85 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file fill.inl
+ *  \brief Inline file for fill.h.
+ */
+
+#include <thrust/fill.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/system/detail/generic/select_system.h>
+#include <thrust/system/detail/generic/fill.h>
+#include <thrust/system/detail/adl/fill.h>
+
+namespace thrust
+{
+
+
+template<typename DerivedPolicy, typename ForwardIterator, typename T>
+  void fill(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+            ForwardIterator first,
+            ForwardIterator last,
+            const T &value)
+{
+  using thrust::system::detail::generic::fill;
+  return fill(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, value);
+} // end fill()
+
+
+template<typename DerivedPolicy, typename OutputIterator, typename Size, typename T>
+  OutputIterator fill_n(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                        OutputIterator first,
+                        Size n,
+                        const T &value)
+{
+  using thrust::system::detail::generic::fill_n;
+  return fill_n(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, n, value);
+} // end fill_n()
+
+
+template<typename ForwardIterator, typename T>
+  void fill(ForwardIterator first,
+            ForwardIterator last,
+            const T &value)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<ForwardIterator>::type System;
+
+  System system;
+
+  thrust::fill(select_system(system), first, last, value);
+} // end fill()
+
+
+template<typename OutputIterator, typename Size, typename T>
+  OutputIterator fill_n(OutputIterator first,
+                        Size n,
+                        const T &value)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<OutputIterator>::type System;
+
+  System system;
+
+  return thrust::fill_n(select_system(system), first, n, value);
+} // end fill()
+
+
+} // end namespace thrust
+
diff --git a/compat/thrust/detail/find.inl b/compat/thrust/detail/find.inl
new file mode 100644
index 0000000000..465c937395
--- /dev/null
+++ b/compat/thrust/detail/find.inl
@@ -0,0 +1,109 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file find.inl
+ *  \brief Inline file for find.h
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/system/detail/generic/select_system.h>
+#include <thrust/system/detail/generic/find.h>
+#include <thrust/system/detail/adl/find.h>
+
+namespace thrust
+{
+
+
+template<typename DerivedPolicy, typename InputIterator, typename T>
+InputIterator find(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                   InputIterator first,
+                   InputIterator last,
+                   const T& value)
+{
+  using thrust::system::detail::generic::find;
+  return find(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, value);
+} // end find()
+
+
+template<typename DerivedPolicy, typename InputIterator, typename Predicate>
+InputIterator find_if(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                      InputIterator first,
+                      InputIterator last,
+                      Predicate pred)
+{
+  using thrust::system::detail::generic::find_if;
+  return find_if(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, pred);
+} // end find_if()
+
+
+template<typename DerivedPolicy, typename InputIterator, typename Predicate>
+InputIterator find_if_not(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                          InputIterator first,
+                          InputIterator last,
+                          Predicate pred)
+{
+  using thrust::system::detail::generic::find_if_not;
+  return find_if_not(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, pred);
+} // end find_if_not()
+
+
+template <typename InputIterator, typename T>
+InputIterator find(InputIterator first,
+                   InputIterator last,
+                   const T& value)
+{
+    using thrust::system::detail::generic::select_system;
+
+    typedef typename thrust::iterator_system<InputIterator>::type System;
+
+    System system;
+
+    return thrust::find(select_system(system), first, last, value);
+}
+
+template <typename InputIterator, typename Predicate>
+InputIterator find_if(InputIterator first,
+                      InputIterator last,
+                      Predicate pred)
+{
+    using thrust::system::detail::generic::select_system;
+
+    typedef typename thrust::iterator_system<InputIterator>::type System;
+
+    System system;
+
+    return thrust::find_if(select_system(system), first, last, pred);
+}
+
+template <typename InputIterator, typename Predicate>
+InputIterator find_if_not(InputIterator first,
+                          InputIterator last,
+                          Predicate pred)
+{
+    using thrust::system::detail::generic::select_system;
+
+    typedef typename thrust::iterator_system<InputIterator>::type System;
+
+    System system;
+
+    return thrust::find_if_not(select_system(system), first, last, pred);
+}
+
+
+} // end namespace thrust
+
diff --git a/compat/thrust/detail/for_each.inl b/compat/thrust/detail/for_each.inl
new file mode 100644
index 0000000000..7c9dc172e3
--- /dev/null
+++ b/compat/thrust/detail/for_each.inl
@@ -0,0 +1,90 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file for_each.inl
+ *  \brief Inline file for for_each.h.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/for_each.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/system/detail/generic/select_system.h>
+#include <thrust/system/detail/generic/for_each.h>
+#include <thrust/system/detail/adl/for_each.h>
+
+namespace thrust
+{
+
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename UnaryFunction>
+  InputIterator for_each(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                         InputIterator first,
+                         InputIterator last,
+                         UnaryFunction f)
+{
+  using thrust::system::detail::generic::for_each;
+
+  return for_each(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, f);
+}
+
+
+template<typename InputIterator,
+         typename UnaryFunction>
+InputIterator for_each(InputIterator first,
+                       InputIterator last,
+                       UnaryFunction f)
+{
+  using thrust::system::detail::generic::select_system;
+  typedef typename thrust::iterator_system<InputIterator>::type System;
+
+  System system;
+  return thrust::for_each(select_system(system), first, last, f);
+} // end for_each()
+
+
+template<typename DerivedPolicy, typename InputIterator, typename Size, typename UnaryFunction>
+  InputIterator for_each_n(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                           InputIterator first,
+                           Size n,
+                           UnaryFunction f)
+{
+  using thrust::system::detail::generic::for_each_n;
+
+  return for_each_n(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, n, f);
+} // end for_each_n()
+
+
+template<typename InputIterator,
+         typename Size,
+         typename UnaryFunction>
+InputIterator for_each_n(InputIterator first,
+                         Size n,
+                         UnaryFunction f)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator>::type System;
+
+  System system;
+  return thrust::for_each_n(select_system(system), first, n, f);
+} // end for_each_n()
+
+
+} // end namespace thrust
+
diff --git a/compat/thrust/detail/function.h b/compat/thrust/detail/function.h
new file mode 100644
index 0000000000..36b76c286f
--- /dev/null
+++ b/compat/thrust/detail/function.h
@@ -0,0 +1,226 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/raw_reference_cast.h>
+
+namespace thrust
+{
+namespace detail
+{
+
+
+template<typename Function, typename Result>
+  struct host_function
+{
+  // mutable because Function::operator() might be const
+  mutable Function m_f;
+
+  inline host_function()
+    : m_f()
+  {}
+
+  inline host_function(const Function &f)
+    : m_f(f)
+  {}
+
+  template<typename Argument>
+    inline Result operator()(Argument &x) const
+  {
+    // we static cast to Result to handle void Result without error
+    // in case Function's result is non-void
+    return static_cast<Result>(m_f(thrust::raw_reference_cast(x)));
+  }
+
+  template<typename Argument>
+    inline Result operator()(const Argument &x) const
+  {
+    // we static cast to Result to handle void Result without error
+    // in case Function's result is non-void
+    return static_cast<Result>(m_f(thrust::raw_reference_cast(x)));
+  }
+
+  template<typename Argument1, typename Argument2>
+    inline Result operator()(Argument1 &x, Argument2 &y) const
+  {
+    // we static cast to Result to handle void Result without error
+    // in case Function's result is non-void
+    return static_cast<Result>(m_f(thrust::raw_reference_cast(x), thrust::raw_reference_cast(y)));
+  }
+
+  template<typename Argument1, typename Argument2>
+    inline Result operator()(const Argument1 &x, Argument2 &y) const
+  {
+    // we static cast to Result to handle void Result without error
+    // in case Function's result is non-void
+    return static_cast<Result>(m_f(thrust::raw_reference_cast(x), thrust::raw_reference_cast(y)));
+  }
+
+  template<typename Argument1, typename Argument2>
+    inline Result operator()(const Argument1 &x, const Argument2 &y) const
+  {
+    // we static cast to Result to handle void Result without error
+    // in case Function's result is non-void
+    return static_cast<Result>(m_f(thrust::raw_reference_cast(x), thrust::raw_reference_cast(y)));
+  }
+
+  template<typename Argument1, typename Argument2>
+    inline Result operator()(Argument1 &x, const Argument2 &y) const
+  {
+    // we static cast to Result to handle void Result without error
+    // in case Function's result is non-void
+    return static_cast<Result>(m_f(thrust::raw_reference_cast(x), thrust::raw_reference_cast(y)));
+  }
+}; // end host_function
+
+
+template<typename Function, typename Result>
+  struct device_function
+{
+  // mutable because Function::operator() might be const
+  mutable Function m_f;
+
+  inline __device__ device_function()
+    : m_f()
+  {}
+
+  inline __device__ device_function(const Function &f)
+    : m_f(f)
+  {}
+
+  template<typename Argument>
+    inline __device__ Result operator()(Argument &x) const
+  {
+    // we static cast to Result to handle void Result without error
+    // in case Function's result is non-void
+    return static_cast<Result>(m_f(thrust::raw_reference_cast(x)));
+  }
+
+  template<typename Argument>
+    inline __device__ Result operator()(const Argument &x) const
+  {
+    // we static cast to Result to handle void Result without error
+    // in case Function's result is non-void
+    return static_cast<Result>(m_f(thrust::raw_reference_cast(x)));
+  }
+
+  template<typename Argument1, typename Argument2>
+    inline __device__ Result operator()(Argument1 &x, Argument2 &y) const
+  {
+    // we static cast to Result to handle void Result without error
+    // in case Function's result is non-void
+    return static_cast<Result>(m_f(thrust::raw_reference_cast(x), thrust::raw_reference_cast(y)));
+  }
+
+  template<typename Argument1, typename Argument2>
+    inline __device__ Result operator()(const Argument1 &x, Argument2 &y) const
+  {
+    // we static cast to Result to handle void Result without error
+    // in case Function's result is non-void
+    return static_cast<Result>(m_f(thrust::raw_reference_cast(x), thrust::raw_reference_cast(y)));
+  }
+
+  template<typename Argument1, typename Argument2>
+    inline __device__ Result operator()(const Argument1 &x, const Argument2 &y) const
+  {
+    // we static cast to Result to handle void Result without error
+    // in case Function's result is non-void
+    return static_cast<Result>(m_f(thrust::raw_reference_cast(x), thrust::raw_reference_cast(y)));
+  }
+
+  template<typename Argument1, typename Argument2>
+    inline __device__ Result operator()(Argument1 &x, const Argument2 &y) const
+  {
+    // we static cast to Result to handle void Result without error
+    // in case Function's result is non-void
+    return static_cast<Result>(m_f(thrust::raw_reference_cast(x), thrust::raw_reference_cast(y)));
+  }
+}; // end device_function
+
+
+template<typename Function, typename Result>
+  struct host_device_function
+{
+  // mutable because Function::operator() might be const
+  mutable Function m_f;
+
+  inline __host__ __device__
+  host_device_function()
+    : m_f()
+  {}
+
+  inline __host__ __device__
+  host_device_function(const Function &f)
+    : m_f(f)
+  {}
+
+  __thrust_hd_warning_disable__
+  template<typename Argument>
+  inline __host__ __device__
+    Result operator()(Argument &x) const
+  {
+    // we static cast to Result to handle void Result without error
+    // in case Function's result is non-void
+    return static_cast<Result>(m_f(thrust::raw_reference_cast(x)));
+  }
+
+  template<typename Argument>
+    inline __host__ __device__ Result operator()(const Argument &x) const
+  {
+    // we static cast to Result to handle void Result without error
+    // in case Function's result is non-void
+    return static_cast<Result>(m_f(thrust::raw_reference_cast(x)));
+  }
+
+  template<typename Argument1, typename Argument2>
+    inline __host__ __device__ Result operator()(Argument1 &x, Argument2 &y) const
+  {
+    // we static cast to Result to handle void Result without error
+    // in case Function's result is non-void
+    return static_cast<Result>(m_f(thrust::raw_reference_cast(x), thrust::raw_reference_cast(y)));
+  }
+
+  template<typename Argument1, typename Argument2>
+    inline __host__ __device__ Result operator()(const Argument1 &x, Argument2 &y) const
+  {
+    // we static cast to Result to handle void Result without error
+    // in case Function's result is non-void
+    return static_cast<Result>(m_f(thrust::raw_reference_cast(x), thrust::raw_reference_cast(y)));
+  }
+
+  template<typename Argument1, typename Argument2>
+    inline __host__ __device__ Result operator()(const Argument1 &x, const Argument2 &y) const
+  {
+    // we static cast to Result to handle void Result without error
+    // in case Function's result is non-void
+    return static_cast<Result>(m_f(thrust::raw_reference_cast(x), thrust::raw_reference_cast(y)));
+  }
+
+  template<typename Argument1, typename Argument2>
+    inline __host__ __device__ Result operator()(Argument1 &x, const Argument2 &y) const
+  {
+    // we static cast to Result to handle void Result without error
+    // in case Function's result is non-void
+    return static_cast<Result>(m_f(thrust::raw_reference_cast(x), thrust::raw_reference_cast(y)));
+  }
+}; // end host_device_function
+
+
+} // end detail
+} // end thrust
+
diff --git a/compat/thrust/detail/functional.inl b/compat/thrust/detail/functional.inl
new file mode 100644
index 0000000000..4024585935
--- /dev/null
+++ b/compat/thrust/detail/functional.inl
@@ -0,0 +1,122 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/functional.h>
+
+namespace thrust
+{
+
+namespace detail
+{
+
+template<typename Operation>
+  struct unary_traits_imp;
+
+template<typename Operation>
+  struct unary_traits_imp<Operation*>
+{
+  typedef Operation                         function_type;
+  typedef const function_type &             param_type;
+  typedef typename Operation::result_type   result_type;
+  typedef typename Operation::argument_type argument_type;
+}; // end unary_traits_imp
+
+template<typename Result, typename Argument>
+  struct unary_traits_imp<Result(*)(Argument)>
+{
+  typedef Result   (*function_type)(Argument);
+  typedef Result   (*param_type)(Argument);
+  typedef Result   result_type;
+  typedef Argument argument_type;
+}; // end unary_traits_imp
+
+template<typename Operation>
+  struct binary_traits_imp;
+
+template<typename Operation>
+  struct binary_traits_imp<Operation*>
+{
+  typedef Operation                                function_type;
+  typedef const function_type &                    param_type;
+  typedef typename Operation::result_type          result_type;
+  typedef typename Operation::first_argument_type  first_argument_type;
+  typedef typename Operation::second_argument_type second_argument_type;
+}; // end binary_traits_imp
+
+template<typename Result, typename Argument1, typename Argument2>
+  struct binary_traits_imp<Result(*)(Argument1, Argument2)>
+{
+  typedef Result (*function_type)(Argument1, Argument2);
+  typedef Result (*param_type)(Argument1, Argument2);
+  typedef Result result_type;
+  typedef Argument1 first_argument_type;
+  typedef Argument2 second_argument_type;
+}; // end binary_traits_imp
+
+} // end detail
+
+template<typename Operation>
+  struct unary_traits
+{
+  typedef typename detail::unary_traits_imp<Operation*>::function_type function_type;
+  typedef typename detail::unary_traits_imp<Operation*>::param_type    param_type;
+  typedef typename detail::unary_traits_imp<Operation*>::result_type   result_type;
+  typedef typename detail::unary_traits_imp<Operation*>::argument_type argument_type;
+}; // end unary_traits
+
+template<typename Result, typename Argument>
+  struct unary_traits<Result(*)(Argument)>
+{
+  typedef Result   (*function_type)(Argument);
+  typedef Result   (*param_type)(Argument);
+  typedef Result   result_type;
+  typedef Argument argument_type;
+}; // end unary_traits
+
+template<typename Operation>
+  struct binary_traits
+{
+  typedef typename detail::binary_traits_imp<Operation*>::function_type        function_type;
+  typedef typename detail::binary_traits_imp<Operation*>::param_type           param_type;
+  typedef typename detail::binary_traits_imp<Operation*>::result_type          result_type;
+  typedef typename detail::binary_traits_imp<Operation*>::first_argument_type  first_argument_type;
+  typedef typename detail::binary_traits_imp<Operation*>::second_argument_type second_argument_type;
+}; // end binary_traits
+
+template<typename Result, typename Argument1, typename Argument2>
+  struct binary_traits<Result(*)(Argument1, Argument2)>
+{
+  typedef Result (*function_type)(Argument1, Argument2);
+  typedef Result (*param_type)(Argument1, Argument2);
+  typedef Result result_type;
+  typedef Argument1 first_argument_type;
+  typedef Argument2 second_argument_type;
+}; // end binary_traits
+
+template<typename Predicate>
+  unary_negate<Predicate> not1(const Predicate &pred)
+{
+  return unary_negate<Predicate>(pred);
+} // end not1()
+
+template<typename BinaryPredicate>
+  binary_negate<BinaryPredicate> not2(const BinaryPredicate &pred)
+{
+  return binary_negate<BinaryPredicate>(pred);
+} // end not2()
+
+} // end thrust
+
diff --git a/compat/thrust/detail/functional/actor.h b/compat/thrust/detail/functional/actor.h
new file mode 100644
index 0000000000..0b95a6b894
--- /dev/null
+++ b/compat/thrust/detail/functional/actor.h
@@ -0,0 +1,192 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+// Portions of this code are derived from
+//
+// Manjunath Kudlur's Carbon library
+//
+// and
+//
+// Based on Boost.Phoenix v1.2
+// Copyright (c) 2001-2002 Joel de Guzman
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/tuple.h>
+#include <thrust/detail/functional/value.h>
+#include <thrust/detail/functional/composite.h>
+#include <thrust/detail/functional/operators/assignment_operator.h>
+#include <thrust/detail/type_traits/result_of.h>
+
+namespace thrust
+{
+namespace detail
+{
+namespace functional
+{
+
+template<typename Action, typename Env>
+  struct apply_actor
+{
+  typedef typename Action::template result<Env>::type type;
+};
+
+template<typename Eval>
+  struct actor
+    : Eval
+{
+  typedef Eval eval_type;
+
+  __host__ __device__
+  actor(void);
+
+  __host__ __device__
+  actor(const Eval &base);
+
+  __host__ __device__
+  typename apply_actor<eval_type, thrust::null_type >::type
+  operator()(void) const;
+
+  template<typename T0>
+  __host__ __device__
+  typename apply_actor<eval_type, thrust::tuple<T0&> >::type
+  operator()(T0 &_0) const;
+
+  template<typename T0, typename T1>
+  __host__ __device__
+  typename apply_actor<eval_type, thrust::tuple<T0&,T1&> >::type
+  operator()(T0 &_0, T1 &_1) const;
+
+  template<typename T0, typename T1, typename T2>
+  __host__ __device__
+  typename apply_actor<eval_type, thrust::tuple<T0&,T1&,T2&> >::type
+  operator()(T0 &_0, T1 &_1, T2 &_2) const;
+
+  template<typename T0, typename T1, typename T2, typename T3>
+  __host__ __device__
+  typename apply_actor<eval_type, thrust::tuple<T0&,T1&,T2&,T3&> >::type
+  operator()(T0 &_0, T1 &_1, T2 &_2, T3 &_3) const;
+
+  template<typename T0, typename T1, typename T2, typename T3, typename T4>
+  __host__ __device__
+  typename apply_actor<eval_type, thrust::tuple<T0&,T1&,T2&,T3&,T4&> >::type
+  operator()(T0 &_0, T1 &_1, T2 &_2, T3 &_3, T4 &_4) const;
+
+  template<typename T0, typename T1, typename T2, typename T3, typename T4, typename T5>
+  __host__ __device__
+  typename apply_actor<eval_type, thrust::tuple<T0&,T1&,T2&,T3&,T4&,T5&> >::type
+  operator()(T0 &_0, T1 &_1, T2 &_2, T3 &_3, T4 &_4, T5 &_5) const;
+
+  template<typename T0, typename T1, typename T2, typename T3, typename T4, typename T5, typename T6>
+  __host__ __device__
+  typename apply_actor<eval_type, thrust::tuple<T0&,T1&,T2&,T3&,T4&,T5&,T6&> >::type
+  operator()(T0 &_0, T1 &_1, T2 &_2, T3 &_3, T4 &_4, T5 &_5, T6 &_6) const;
+
+  template<typename T0, typename T1, typename T2, typename T3, typename T4, typename T5, typename T6, typename T7>
+  __host__ __device__
+  typename apply_actor<eval_type, thrust::tuple<T0&,T1&,T2&,T3&,T4&,T5&,T6&,T7&> >::type
+  operator()(T0 &_0, T1 &_1, T2 &_2, T3 &_3, T4 &_4, T5 &_5, T6 &_6, T7 &_7) const;
+
+  template<typename T0, typename T1, typename T2, typename T3, typename T4, typename T5, typename T6, typename T7, typename T8>
+  __host__ __device__
+  typename apply_actor<eval_type, thrust::tuple<T0&,T1&,T2&,T3&,T4&,T5&,T6&,T7&,T8&> >::type
+  operator()(T0 &_0, T1 &_1, T2 &_2, T3 &_3, T4 &_4, T5 &_5, T6 &_6, T7 &_7, T8 &_8) const;
+
+  template<typename T0, typename T1, typename T2, typename T3, typename T4, typename T5, typename T6, typename T7, typename T8, typename T9>
+  __host__ __device__
+  typename apply_actor<eval_type, thrust::tuple<T0&,T1&,T2&,T3&,T4&,T5&,T6&,T7&,T8&,T9&> >::type
+  operator()(T0 &_0, T1 &_1, T2 &_2, T3 &_3, T4 &_4, T5 &_5, T6 &_6, T7 &_7, T8 &_8, T9 &_9) const;
+
+  template<typename T>
+  __host__ __device__
+  typename assign_result<Eval,T>::type
+  operator=(const T &_1) const;
+}; // end actor
+
+// in general, as_actor should turn things into values
+template<typename T>
+  struct as_actor
+{
+  typedef value<T> type;
+
+  static inline __host__ __device__ type convert(const T &x)
+  {
+    return val(x);
+  } // end convert()
+}; // end as_actor
+
+// specialization for things which are already actors
+template<typename Eval>
+  struct as_actor<actor<Eval> >
+{
+  typedef actor<Eval> type;
+
+  static inline __host__ __device__ const type &convert(const actor<Eval> &x)
+  {
+    return x;
+  } // end convert()
+}; // end as_actor
+
+template<typename T>
+  typename as_actor<T>::type
+  __host__ __device__
+    make_actor(const T &x)
+{
+  return as_actor<T>::convert(x);
+} // end make_actor()
+
+} // end functional
+
+// provide specializations for result_of for nullary, unary, and binary invocations of actor
+template<typename Eval>
+  struct result_of<
+    thrust::detail::functional::actor<Eval>()
+  >
+{
+  typedef typename thrust::detail::functional::apply_actor<
+    thrust::detail::functional::actor<Eval>,
+    thrust::null_type
+  >::type type;
+}; // end result_of
+
+template<typename Eval, typename Arg1>
+  struct result_of<
+    thrust::detail::functional::actor<Eval>(Arg1)
+  >
+{
+  typedef typename thrust::detail::functional::apply_actor<
+    thrust::detail::functional::actor<Eval>,
+    thrust::tuple<Arg1>
+  >::type type;
+}; // end result_of
+
+template<typename Eval, typename Arg1, typename Arg2>
+  struct result_of<
+    thrust::detail::functional::actor<Eval>(Arg1,Arg2)
+  >
+{
+  typedef typename thrust::detail::functional::apply_actor<
+    thrust::detail::functional::actor<Eval>,
+    thrust::tuple<Arg1,Arg2>
+  >::type type;
+}; // end result_of
+
+} // end detail
+} // end thrust
+
+#include <thrust/detail/functional/actor.inl>
+
diff --git a/compat/thrust/detail/functional/actor.inl b/compat/thrust/detail/functional/actor.inl
new file mode 100644
index 0000000000..84347be7b8
--- /dev/null
+++ b/compat/thrust/detail/functional/actor.inl
@@ -0,0 +1,194 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+// Portions of this code are derived from
+//
+// Manjunath Kudlur's Carbon library
+//
+// and
+//
+// Based on Boost.Phoenix v1.2
+// Copyright (c) 2001-2002 Joel de Guzman
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/functional/composite.h>
+#include <thrust/detail/functional/operators/assignment_operator.h>
+#include <thrust/functional.h>
+
+namespace thrust
+{
+
+namespace detail
+{
+namespace functional
+{
+
+template<typename Eval>
+  actor<Eval>
+    ::actor(void)
+      : eval_type()
+{}
+
+template<typename Eval>
+  actor<Eval>
+    ::actor(const Eval &base)
+      : eval_type(base)
+{}
+
+template<typename Eval>
+  typename apply_actor<
+    typename actor<Eval>::eval_type,
+    typename thrust::null_type
+  >::type
+    actor<Eval>
+      ::operator()(void) const
+{
+  return eval_type::eval(thrust::null_type());
+} // end basic_environment::operator()
+
+template<typename Eval>
+  template<typename T0>
+    typename apply_actor<
+      typename actor<Eval>::eval_type,
+      typename thrust::tuple<T0&>
+    >::type
+      actor<Eval>
+        ::operator()(T0 &_0) const
+{
+  return eval_type::eval(thrust::tie(_0));
+} // end basic_environment::operator()
+
+template<typename Eval>
+  template<typename T0, typename T1>
+    typename apply_actor<
+      typename actor<Eval>::eval_type,
+      typename thrust::tuple<T0&,T1&>
+    >::type
+      actor<Eval>
+        ::operator()(T0 &_0, T1 &_1) const
+{
+  return eval_type::eval(thrust::tie(_0,_1));
+} // end basic_environment::operator()
+
+template<typename Eval>
+  template<typename T0, typename T1, typename T2>
+    typename apply_actor<
+      typename actor<Eval>::eval_type,
+      typename thrust::tuple<T0&,T1&,T2&>
+    >::type
+      actor<Eval>
+        ::operator()(T0 &_0, T1 &_1, T2 &_2) const
+{
+  return eval_type::eval(thrust::tie(_0,_1,_2));
+} // end basic_environment::operator()
+
+template<typename Eval>
+  template<typename T0, typename T1, typename T2, typename T3>
+    typename apply_actor<
+      typename actor<Eval>::eval_type,
+      typename thrust::tuple<T0&,T1&,T2&,T3&>
+    >::type
+      actor<Eval>
+        ::operator()(T0 &_0, T1 &_1, T2 &_2, T3 &_3) const
+{
+  return eval_type::eval(thrust::tie(_0,_1,_2,_3));
+} // end basic_environment::operator()
+
+template<typename Eval>
+  template<typename T0, typename T1, typename T2, typename T3, typename T4>
+    typename apply_actor<
+      typename actor<Eval>::eval_type,
+      typename thrust::tuple<T0&,T1&,T2&,T3&,T4&>
+    >::type
+      actor<Eval>
+        ::operator()(T0 &_0, T1 &_1, T2 &_2, T3 &_3, T4 &_4) const
+{
+  return eval_type::eval(thrust::tie(_0,_1,_2,_3,_4));
+} // end basic_environment::operator()
+
+template<typename Eval>
+  template<typename T0, typename T1, typename T2, typename T3, typename T4, typename T5>
+    typename apply_actor<
+      typename actor<Eval>::eval_type,
+      typename thrust::tuple<T0&,T1&,T2&,T3&,T4&,T5&>
+    >::type
+      actor<Eval>
+        ::operator()(T0 &_0, T1 &_1, T2 &_2, T3 &_3, T4 &_4, T5 &_5) const
+{
+  return eval_type::eval(thrust::tie(_0,_1,_2,_3,_4,_5));
+} // end basic_environment::operator()
+
+template<typename Eval>
+  template<typename T0, typename T1, typename T2, typename T3, typename T4, typename T5, typename T6>
+    typename apply_actor<
+      typename actor<Eval>::eval_type,
+      typename thrust::tuple<T0&,T1&,T2&,T3&,T4&,T5&,T6&>
+    >::type
+      actor<Eval>
+        ::operator()(T0 &_0, T1 &_1, T2 &_2, T3 &_3, T4 &_4, T5 &_5, T6 &_6) const
+{
+  return eval_type::eval(thrust::tie(_0,_1,_2,_3,_4,_5,_6));
+} // end basic_environment::operator()
+
+template<typename Eval>
+  template<typename T0, typename T1, typename T2, typename T3, typename T4, typename T5, typename T6, typename T7>
+    typename apply_actor<
+      typename actor<Eval>::eval_type,
+      typename thrust::tuple<T0&,T1&,T2&,T3&,T4&,T5&,T6&,T7&>
+    >::type
+      actor<Eval>
+        ::operator()(T0 &_0, T1 &_1, T2 &_2, T3 &_3, T4 &_4, T5 &_5, T6 &_6, T7 &_7) const
+{
+  return eval_type::eval(thrust::tie(_0,_1,_2,_3,_4,_5,_6,_7));
+} // end basic_environment::operator()
+
+template<typename Eval>
+  template<typename T0, typename T1, typename T2, typename T3, typename T4, typename T5, typename T6, typename T7, typename T8>
+    typename apply_actor<
+      typename actor<Eval>::eval_type,
+      typename thrust::tuple<T0&,T1&,T2&,T3&,T4&,T5&,T6&,T7&,T8&>
+    >::type
+      actor<Eval>
+        ::operator()(T0 &_0, T1 &_1, T2 &_2, T3 &_3, T4 &_4, T5 &_5, T6 &_6, T7 &_7, T8 &_8) const
+{
+  return eval_type::eval(thrust::tie(_0,_1,_2,_3,_4,_5,_6,_7,_8));
+} // end basic_environment::operator()
+
+template<typename Eval>
+  template<typename T0, typename T1, typename T2, typename T3, typename T4, typename T5, typename T6, typename T7, typename T8, typename T9>
+    typename apply_actor<
+      typename actor<Eval>::eval_type,
+      typename thrust::tuple<T0&,T1&,T2&,T3&,T4&,T5&,T6&,T7&,T8&,T9&>
+    >::type
+      actor<Eval>
+        ::operator()(T0 &_0, T1 &_1, T2 &_2, T3 &_3, T4 &_4, T5 &_5, T6 &_6, T7 &_7, T8 &_8, T9 &_9) const
+{
+  return eval_type::eval(thrust::tie(_0,_1,_2,_3,_4,_5,_6,_7,_8,_9));
+} // end basic_environment::operator()
+
+template<typename Eval>
+  template<typename T>
+    typename assign_result<Eval,T>::type
+      actor<Eval>
+        ::operator=(const T& _1) const
+{
+  return do_assign(*this,_1);
+} // end actor::operator=()
+
+} // end functional
+} // end detail
+} // end thrust
+
diff --git a/compat/thrust/detail/functional/argument.h b/compat/thrust/detail/functional/argument.h
new file mode 100644
index 0000000000..96a20bed1f
--- /dev/null
+++ b/compat/thrust/detail/functional/argument.h
@@ -0,0 +1,75 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+// Portions of this code are derived from
+//
+// Manjunath Kudlur's Carbon library
+//
+// and
+//
+// Based on Boost.Phoenix v1.2
+// Copyright (c) 2001-2002 Joel de Guzman
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/tuple.h>
+
+namespace thrust
+{
+namespace detail
+{
+namespace functional
+{
+
+template<unsigned int i, typename Env>
+  struct argument_helper
+{
+  typedef typename thrust::tuple_element<i,Env>::type type;
+};
+
+template<unsigned int i>
+  struct argument_helper<i,thrust::null_type>
+{
+  typedef thrust::null_type type;
+};
+
+
+template<unsigned int i>
+  class argument
+{
+  public:
+    template<typename Env>
+      struct result
+        : argument_helper<i,Env>
+    {
+    };
+
+    __host__ __device__
+    argument(void){}
+
+    template<typename Env>
+    __host__ __device__
+    typename result<Env>::type eval(const Env &e) const
+    {
+      return thrust::get<i>(e);
+    } // end eval()
+}; // end argument
+
+} // end functional
+} // end detail
+} // end thrust
+
diff --git a/compat/thrust/detail/functional/composite.h b/compat/thrust/detail/functional/composite.h
new file mode 100644
index 0000000000..1d5fde3152
--- /dev/null
+++ b/compat/thrust/detail/functional/composite.h
@@ -0,0 +1,163 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+// Portions of this code are derived from
+//
+// Manjunath Kudlur's Carbon library
+//
+// and
+//
+// Based on Boost.Phoenix v1.2
+// Copyright (c) 2001-2002 Joel de Guzman
+
+#pragma once
+
+#include <thrust/detail/functional/actor.h>
+#include <thrust/tuple.h>
+
+namespace thrust
+{
+namespace detail
+{
+namespace functional
+{
+
+// XXX we should just take a single EvalTuple
+template<typename Eval0,
+         typename Eval1  = thrust::null_type,
+         typename Eval2  = thrust::null_type,
+         typename Eval3  = thrust::null_type,
+         typename Eval4  = thrust::null_type,
+         typename Eval5  = thrust::null_type,
+         typename Eval6  = thrust::null_type,
+         typename Eval7  = thrust::null_type,
+         typename Eval8  = thrust::null_type,
+         typename Eval9  = thrust::null_type,
+         typename Eval10 = thrust::null_type>
+  class composite;
+
+template<typename Eval0, typename Eval1>
+  class composite<
+    Eval0,
+    Eval1,
+    thrust::null_type,
+    thrust::null_type,
+    thrust::null_type,
+    thrust::null_type,
+    thrust::null_type,
+    thrust::null_type,
+    thrust::null_type,
+    thrust::null_type
+  >
+{
+  public:
+    template<typename Env>
+      struct result
+    {
+      typedef typename Eval0::template result<
+        thrust::tuple<
+          typename Eval1::template result<Env>::type
+        >
+      >::type type;
+    };
+
+    __host__ __device__
+    composite(const Eval0 &e0, const Eval1 &e1)
+      : m_eval0(e0),
+        m_eval1(e1)
+    {}
+
+    template<typename Env>
+    __host__ __device__
+    typename result<Env>::type
+    eval(const Env &x) const
+    {
+      typename Eval1::template result<Env>::type result1 = m_eval1.eval(x);
+      return m_eval0.eval(thrust::tie(result1));
+    }
+
+  private:
+    Eval0 m_eval0;
+    Eval1 m_eval1;
+}; // end composite<Eval0,Eval1>
+
+template<typename Eval0, typename Eval1, typename Eval2>
+  class composite<
+    Eval0,
+    Eval1,
+    Eval2,
+    thrust::null_type,
+    thrust::null_type,
+    thrust::null_type,
+    thrust::null_type,
+    thrust::null_type,
+    thrust::null_type,
+    thrust::null_type
+  >
+{
+  public:
+    template<typename Env>
+      struct result
+    {
+      typedef typename Eval0::template result<
+        thrust::tuple<
+          typename Eval1::template result<Env>::type,
+          typename Eval2::template result<Env>::type
+        >
+      >::type type;
+    };
+
+    __host__ __device__
+    composite(const Eval0 &e0, const Eval1 &e1, const Eval2 &e2)
+      : m_eval0(e0),
+        m_eval1(e1),
+        m_eval2(e2)
+    {}
+
+    template<typename Env>
+    __host__ __device__
+    typename result<Env>::type
+    eval(const Env &x) const
+    {
+      typename Eval1::template result<Env>::type result1 = m_eval1.eval(x);
+      typename Eval2::template result<Env>::type result2 = m_eval2.eval(x);
+      return m_eval0.eval(thrust::tie(result1,result2));
+    }
+
+  private:
+    Eval0 m_eval0;
+    Eval1 m_eval1;
+    Eval2 m_eval2;
+}; // end composite<Eval0,Eval1,Eval2>
+
+template<typename Eval0, typename Eval1>
+__host__ __device__
+  actor<composite<Eval0,Eval1> > compose(const Eval0 &e0, const Eval1 &e1)
+{
+  return actor<composite<Eval0,Eval1> >(composite<Eval0,Eval1>(e0,e1));
+}
+
+template<typename Eval0, typename Eval1, typename Eval2>
+__host__ __device__
+  actor<composite<Eval0,Eval1,Eval2> > compose(const Eval0 &e0, const Eval1 &e1, const Eval2 &e2)
+{
+  return actor<composite<Eval0,Eval1,Eval2> >(composite<Eval0,Eval1,Eval2>(e0,e1,e2));
+}
+
+} // end functional
+} // end detail
+} // end thrust
+
diff --git a/compat/thrust/detail/functional/operators.h b/compat/thrust/detail/functional/operators.h
new file mode 100644
index 0000000000..0fc3539cb9
--- /dev/null
+++ b/compat/thrust/detail/functional/operators.h
@@ -0,0 +1,25 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/functional/operators/arithmetic_operators.h>
+#include <thrust/detail/functional/operators/relational_operators.h>
+#include <thrust/detail/functional/operators/logical_operators.h>
+#include <thrust/detail/functional/operators/bitwise_operators.h>
+#include <thrust/detail/functional/operators/compound_assignment_operators.h>
+
diff --git a/compat/thrust/detail/functional/operators/arithmetic_operators.h b/compat/thrust/detail/functional/operators/arithmetic_operators.h
new file mode 100644
index 0000000000..a11e7acdd1
--- /dev/null
+++ b/compat/thrust/detail/functional/operators/arithmetic_operators.h
@@ -0,0 +1,394 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/functional/actor.h>
+#include <thrust/detail/functional/composite.h>
+#include <thrust/detail/functional/operators/operator_adaptors.h>
+#include <thrust/functional.h>
+
+namespace thrust
+{
+namespace detail
+{
+namespace functional
+{
+
+template<typename Eval>
+__host__ __device__
+actor<
+  composite<
+    unary_operator<thrust::negate>,
+    actor<Eval>
+  >
+>
+__host__ __device__
+operator-(const actor<Eval> &_1)
+{
+  return compose(unary_operator<thrust::negate>(), _1);
+} // end operator-()
+
+// there's no standard unary_plus functional, so roll an ad hoc one here
+template<typename T>
+  struct unary_plus
+    : public thrust::unary_function<T,T>
+{
+  __host__ __device__ T operator()(const T &x) const {return +x;}
+}; // end unary_plus
+
+template<typename Eval>
+__host__ __device__
+actor<
+  composite<
+    unary_operator<unary_plus>,
+    actor<Eval>
+  >
+>
+operator+(const actor<Eval> &_1)
+{
+  return compose(unary_operator<unary_plus>(), _1);
+} // end operator+()
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<thrust::plus>,
+    actor<T1>,
+    typename as_actor<T2>::type
+  >
+>
+operator+(const actor<T1> &_1, const T2 &_2)
+{
+  return compose(binary_operator<thrust::plus>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator+()
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<thrust::plus>,
+    typename as_actor<T1>::type,
+    actor<T2>
+  >
+>
+operator+(const T1 &_1, const actor<T2> &_2)
+{
+  return compose(binary_operator<thrust::plus>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator+()
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<thrust::plus>,
+    actor<T1>,
+    actor<T2>
+  >
+>
+operator+(const actor<T1> &_1, const actor<T2> &_2)
+{
+  return compose(binary_operator<thrust::plus>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator+()
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<thrust::minus>,
+    typename as_actor<T1>::type,
+    actor<T2>
+  >
+>
+operator-(const T1 &_1, const actor<T2> &_2)
+{
+  return compose(binary_operator<thrust::minus>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator-()
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<thrust::minus>,
+    actor<T1>,
+    typename as_actor<T2>::type
+  >
+>
+operator-(const actor<T1> &_1, const T2 &_2)
+{
+  return compose(binary_operator<thrust::minus>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator-()
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<thrust::minus>,
+    actor<T1>,
+    actor<T2>
+  >
+>
+operator-(const actor<T1> &_1, const actor<T2> &_2)
+{
+  return compose(binary_operator<thrust::minus>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator-()
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<thrust::multiplies>,
+    typename as_actor<T1>::type,
+    actor<T2>
+  >
+>
+operator*(const T1 &_1, const actor<T2> &_2)
+{
+  return compose(binary_operator<thrust::multiplies>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator*()
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<thrust::multiplies>,
+    actor<T1>,
+    typename as_actor<T2>::type
+  >
+>
+operator*(const actor<T1> &_1, const T2 &_2)
+{
+  return compose(binary_operator<thrust::multiplies>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator*()
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<thrust::multiplies>,
+    actor<T1>,
+    actor<T2>
+  >
+>
+operator*(const actor<T1> &_1, const actor<T2> &_2)
+{
+  return compose(binary_operator<thrust::multiplies>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator*()
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<thrust::divides>,
+    actor<T1>,
+    typename as_actor<T2>::type
+  >
+>
+operator/(const actor<T1> &_1, const T2 &_2)
+{
+  return compose(binary_operator<thrust::divides>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator/()
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<thrust::divides>,
+    typename as_actor<T1>::type,
+    actor<T2>
+  >
+>
+operator/(const T1 &_1, const actor<T2> &_2)
+{
+  return compose(binary_operator<thrust::divides>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator/()
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<thrust::divides>,
+    actor<T1>,
+    actor<T2>
+  >
+>
+operator/(const actor<T1> &_1, const actor<T2> &_2)
+{
+  return compose(binary_operator<thrust::divides>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator/()
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<thrust::modulus>,
+    actor<T1>,
+    typename as_actor<T2>::type
+  >
+>
+operator%(const actor<T1> &_1, const T2 &_2)
+{
+  return compose(binary_operator<thrust::modulus>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator%()
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<thrust::modulus>,
+    typename as_actor<T1>::type,
+    actor<T2>
+  >
+>
+operator%(const T1 &_1, const actor<T2> &_2)
+{
+  return compose(binary_operator<thrust::modulus>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator%()
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<thrust::modulus>,
+    actor<T1>,
+    actor<T2>
+  >
+>
+operator%(const actor<T1> &_1, const actor<T2> &_2)
+{
+  return compose(binary_operator<thrust::modulus>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator%()
+
+// there's no standard prefix_increment functional, so roll an ad hoc one here
+template<typename T>
+  struct prefix_increment
+    : public thrust::unary_function<T&,T&>
+{
+  __host__ __device__ T& operator()(T &x) const { return ++x; }
+}; // end prefix_increment
+
+template<typename Eval>
+__host__ __device__
+actor<
+  composite<
+    unary_operator<prefix_increment>,
+    actor<Eval>
+  >
+>
+operator++(const actor<Eval> &_1)
+{
+  return compose(unary_operator<prefix_increment>(), _1);
+} // end operator++()
+
+// there's no standard suffix_increment functional, so roll an ad hoc one here
+template<typename T>
+  struct suffix_increment
+    : public thrust::unary_function<T&,T>
+{
+  __host__ __device__ T operator()(T &x) const { return x++; }
+}; // end suffix_increment
+
+template<typename Eval>
+__host__ __device__
+actor<
+  composite<
+    unary_operator<suffix_increment>,
+    actor<Eval>
+  >
+>
+operator++(const actor<Eval> &_1, int)
+{
+  return compose(unary_operator<suffix_increment>(), _1);
+} // end operator++()
+
+// there's no standard prefix_decrement functional, so roll an ad hoc one here
+template<typename T>
+  struct prefix_decrement
+    : public thrust::unary_function<T&,T&>
+{
+  __host__ __device__ T& operator()(T &x) const { return --x; }
+}; // end prefix_decrement
+
+template<typename Eval>
+__host__ __device__
+actor<
+  composite<
+    unary_operator<prefix_decrement>,
+    actor<Eval>
+  >
+>
+operator--(const actor<Eval> &_1)
+{
+  return compose(unary_operator<prefix_decrement>(), _1);
+} // end operator--()
+
+// there's no standard suffix_decrement functional, so roll an ad hoc one here
+template<typename T>
+  struct suffix_decrement
+    : public thrust::unary_function<T&,T>
+{
+  __host__ __device__ T operator()(T &x) const { return x--; }
+}; // end suffix_decrement
+
+template<typename Eval>
+__host__ __device__
+actor<
+  composite<
+    unary_operator<suffix_decrement>,
+    actor<Eval>
+  >
+>
+operator--(const actor<Eval> &_1, int)
+{
+  return compose(unary_operator<suffix_decrement>(), _1);
+} // end operator--()
+
+} // end functional
+} // end detail
+} // end thrust
+
diff --git a/compat/thrust/detail/functional/operators/assignment_operator.h b/compat/thrust/detail/functional/operators/assignment_operator.h
new file mode 100644
index 0000000000..e5d66202bf
--- /dev/null
+++ b/compat/thrust/detail/functional/operators/assignment_operator.h
@@ -0,0 +1,72 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/functional/actor.h>
+#include <thrust/detail/functional/composite.h>
+#include <thrust/detail/functional/operators/operator_adaptors.h>
+#include <thrust/functional.h>
+
+namespace thrust
+{
+
+// XXX WAR circular inclusion with this forward declaration
+template<typename,typename,typename> struct binary_function;
+
+namespace detail
+{
+namespace functional
+{
+
+// XXX WAR circular inclusion with this forward declaration
+template<typename> struct as_actor;
+
+// there's no standard assign functional, so roll an ad hoc one here
+template<typename T>
+  struct assign
+    : thrust::binary_function<T&,T,T&>
+{
+  __host__ __device__ T& operator()(T &lhs, const T &rhs) const { return lhs = rhs; }
+}; // end assign
+
+template<typename Eval, typename T>
+  struct assign_result
+{
+  typedef actor<
+    composite<
+      binary_operator<assign>,
+      actor<Eval>,
+      typename as_actor<T>::type
+    >
+  > type;
+}; // end assign_result
+
+template<typename Eval, typename T>
+  __host__ __device__
+    typename assign_result<Eval,T>::type
+      do_assign(const actor<Eval> &_1, const T &_2)
+{
+  return compose(binary_operator<assign>(),
+                 _1,
+                 as_actor<T>::convert(_2));
+} // end do_assign()
+
+} // end functional
+} // end detail
+} // end thrust
+
diff --git a/compat/thrust/detail/functional/operators/bitwise_operators.h b/compat/thrust/detail/functional/operators/bitwise_operators.h
new file mode 100644
index 0000000000..c89c5d4f83
--- /dev/null
+++ b/compat/thrust/detail/functional/operators/bitwise_operators.h
@@ -0,0 +1,313 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/functional/actor.h>
+#include <thrust/detail/functional/composite.h>
+#include <thrust/detail/functional/operators/operator_adaptors.h>
+#include <thrust/functional.h>
+
+namespace thrust
+{
+namespace detail
+{
+namespace functional
+{
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<thrust::bit_and>,
+    actor<T1>,
+    typename as_actor<T2>::type
+  >
+>
+operator&(const actor<T1> &_1, const T2 &_2)
+{
+  return compose(binary_operator<thrust::bit_and>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator&()
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<thrust::bit_and>,
+    typename as_actor<T1>::type,
+    actor<T2>
+  >
+>
+operator&(const T1 &_1, const actor<T2> &_2)
+{
+  return compose(binary_operator<thrust::bit_and>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator&()
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<thrust::bit_and>,
+    actor<T1>,
+    actor<T2>
+  >
+>
+operator&(const actor<T1> &_1, const actor<T2> &_2)
+{
+  return compose(binary_operator<thrust::bit_and>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator&()
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<thrust::bit_or>,
+    actor<T1>,
+    typename as_actor<T2>::type
+  >
+>
+operator|(const actor<T1> &_1, const T2 &_2)
+{
+  return compose(binary_operator<thrust::bit_or>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator|()
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<thrust::bit_or>,
+    typename as_actor<T1>::type,
+    actor<T2>
+  >
+>
+operator|(const T1 &_1, const actor<T2> &_2)
+{
+  return compose(binary_operator<thrust::bit_or>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator|()
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<thrust::bit_or>,
+    actor<T1>,
+    actor<T2>
+  >
+>
+operator|(const actor<T1> &_1, const actor<T2> &_2)
+{
+  return compose(binary_operator<thrust::bit_or>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator|()
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<thrust::bit_xor>,
+    actor<T1>,
+    typename as_actor<T2>::type
+  >
+>
+operator^(const actor<T1> &_1, const T2 &_2)
+{
+  return compose(binary_operator<thrust::bit_xor>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator^()
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<thrust::bit_xor>,
+    typename as_actor<T1>::type,
+    actor<T2>
+  >
+>
+operator^(const T1 &_1, const actor<T2> &_2)
+{
+  return compose(binary_operator<thrust::bit_xor>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator^()
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<thrust::bit_xor>,
+    actor<T1>,
+    actor<T2>
+  >
+>
+operator^(const actor<T1> &_1, const actor<T2> &_2)
+{
+  return compose(binary_operator<thrust::bit_xor>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator^()
+
+// there's no standard bit_not functional, so roll an ad hoc one here
+template<typename T>
+  struct bit_not
+    : public thrust::unary_function<T,T>
+{
+  __host__ __device__ T operator()(const T &x) const {return ~x;}
+}; // end bit_not
+
+template<typename Eval>
+__host__ __device__
+actor<
+  composite<
+    unary_operator<bit_not>,
+    actor<Eval>
+  >
+>
+__host__ __device__
+operator~(const actor<Eval> &_1)
+{
+  return compose(unary_operator<bit_not>(), _1);
+} // end operator~()
+
+// there's no standard bit_lshift functional, so roll an ad hoc one here
+template<typename T>
+  struct bit_lshift
+    : public thrust::binary_function<T,T,T>
+{
+  __host__ __device__ T operator()(const T &lhs, const T &rhs) const {return lhs << rhs;}
+}; // end bit_lshift
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<bit_lshift>,
+    actor<T1>,
+    typename as_actor<T2>::type
+  >
+>
+operator<<(const actor<T1> &_1, const T2 &_2)
+{
+  return compose(binary_operator<bit_lshift>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator<<()
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<bit_lshift>,
+    typename as_actor<T1>::type,
+    actor<T2>
+  >
+>
+operator<<(const T1 &_1, const actor<T2> &_2)
+{
+  return compose(binary_operator<bit_lshift>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator<<()
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<bit_lshift>,
+    actor<T1>,
+    actor<T2>
+  >
+>
+operator<<(const actor<T1> &_1, const actor<T2> &_2)
+{
+  return compose(binary_operator<bit_lshift>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator<<()
+
+// there's no standard bit_rshift functional, so roll an ad hoc one here
+template<typename T>
+  struct bit_rshift
+    : public thrust::binary_function<T,T,T>
+{
+  __host__ __device__ T operator()(const T &lhs, const T &rhs) const {return lhs >> rhs;}
+}; // end bit_rshift
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<bit_rshift>,
+    actor<T1>,
+    typename as_actor<T2>::type
+  >
+>
+operator>>(const actor<T1> &_1, const T2 &_2)
+{
+  return compose(binary_operator<bit_rshift>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator>>()
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<bit_rshift>,
+    typename as_actor<T1>::type,
+    actor<T2>
+  >
+>
+operator>>(const T1 &_1, const actor<T2> &_2)
+{
+  return compose(binary_operator<bit_rshift>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator>>()
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<bit_rshift>,
+    actor<T1>,
+    actor<T2>
+  >
+>
+operator>>(const actor<T1> &_1, const actor<T2> &_2)
+{
+  return compose(binary_operator<bit_rshift>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator>>()
+
+} // end functional
+} // end detail
+} // end thrust
+
diff --git a/compat/thrust/detail/functional/operators/compound_assignment_operators.h b/compat/thrust/detail/functional/operators/compound_assignment_operators.h
new file mode 100644
index 0000000000..ef7389b55c
--- /dev/null
+++ b/compat/thrust/detail/functional/operators/compound_assignment_operators.h
@@ -0,0 +1,424 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/functional/actor.h>
+#include <thrust/detail/functional/composite.h>
+#include <thrust/detail/functional/operators/operator_adaptors.h>
+
+namespace thrust
+{
+namespace detail
+{
+namespace functional
+{
+
+template<typename T>
+  struct plus_equal
+    : public thrust::binary_function<T&,T,T&>
+{
+  __host__ __device__ T& operator()(T &lhs, const T &rhs) const { return lhs += rhs; }
+}; // end plus_equal
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<plus_equal>,
+    actor<T1>,
+    typename as_actor<T2>::type
+  >
+>
+operator+=(const actor<T1> &_1, const T2 &_2)
+{
+  return compose(binary_operator<plus_equal>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator+=()
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<plus_equal>,
+    actor<T1>,
+    actor<T2>
+  >
+>
+operator+=(const actor<T1> &_1, const actor<T2> &_2)
+{
+  return compose(binary_operator<plus_equal>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator+=()
+
+template<typename T>
+  struct minus_equal
+    : public thrust::binary_function<T&,T,T&>
+{
+  __host__ __device__ T& operator()(T &lhs, const T &rhs) const { return lhs -= rhs; }
+}; // end minus_equal
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<minus_equal>,
+    actor<T1>,
+    typename as_actor<T2>::type
+  >
+>
+operator-=(const actor<T1> &_1, const T2 &_2)
+{
+  return compose(binary_operator<minus_equal>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator-=()
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<minus_equal>,
+    actor<T1>,
+    actor<T2>
+  >
+>
+operator-=(const actor<T1> &_1, const actor<T2> &_2)
+{
+  return compose(binary_operator<minus_equal>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator-=()
+
+template<typename T>
+  struct multiplies_equal
+    : public thrust::binary_function<T&,T,T&>
+{
+  __host__ __device__ T& operator()(T &lhs, const T&rhs) const { return lhs *= rhs; }
+}; // end multiplies_equal
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<multiplies_equal>,
+    actor<T1>,
+    typename as_actor<T2>::type
+  >
+>
+operator*=(const actor<T1> &_1, const T2 &_2)
+{
+  return compose(binary_operator<multiplies_equal>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator*=()
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<multiplies_equal>,
+    actor<T1>,
+    actor<T2>
+  >
+>
+operator*=(const actor<T1> &_1, const actor<T2> &_2)
+{
+  return compose(binary_operator<multiplies_equal>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator*=()
+
+template<typename T>
+  struct divides_equal
+    : public thrust::binary_function<T&,T,T&>
+{
+  __host__ __device__ T& operator()(T &lhs, const T&rhs) const { return lhs /= rhs; }
+}; // end divides_equal
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<divides_equal>,
+    actor<T1>,
+    typename as_actor<T2>::type
+  >
+>
+operator/=(const actor<T1> &_1, const T2 &_2)
+{
+  return compose(binary_operator<divides_equal>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator/=()
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<divides_equal>,
+    actor<T1>,
+    actor<T2>
+  >
+>
+operator/=(const actor<T1> &_1, const actor<T2> &_2)
+{
+  return compose(binary_operator<divides_equal>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator/=()
+
+template<typename T>
+  struct modulus_equal
+    : public thrust::binary_function<T&,T,T&>
+{
+  __host__ __device__ T& operator()(T &lhs, const T&rhs) const { return lhs %= rhs; }
+}; // end modulus_equal
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<modulus_equal>,
+    actor<T1>,
+    typename as_actor<T2>::type
+  >
+>
+operator%=(const actor<T1> &_1, const T2 &_2)
+{
+  return compose(binary_operator<modulus_equal>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator%=()
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<modulus_equal>,
+    actor<T1>,
+    actor<T2>
+  >
+>
+operator%=(const actor<T1> &_1, const actor<T2> &_2)
+{
+  return compose(binary_operator<modulus_equal>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator%=()
+
+template<typename T>
+  struct bit_and_equal
+    : public thrust::binary_function<T&,T,T&>
+{
+  __host__ __device__ T& operator()(T &lhs, const T&rhs) const { return lhs &= rhs; }
+}; // end bit_and_equal
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<bit_and_equal>,
+    actor<T1>,
+    typename as_actor<T2>::type
+  >
+>
+operator&=(const actor<T1> &_1, const T2 &_2)
+{
+  return compose(binary_operator<bit_and_equal>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator&=()
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<bit_and_equal>,
+    actor<T1>,
+    actor<T2>
+  >
+>
+operator&=(const actor<T1> &_1, const actor<T2> &_2)
+{
+  return compose(binary_operator<bit_and_equal>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator&=()
+
+template<typename T>
+  struct bit_or_equal
+    : public thrust::binary_function<T&,T,T&>
+{
+  __host__ __device__ T& operator()(T &lhs, const T&rhs) const { return lhs |= rhs; }
+}; // end bit_or_equal
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<bit_or_equal>,
+    actor<T1>,
+    typename as_actor<T2>::type
+  >
+>
+operator|=(const actor<T1> &_1, const T2 &_2)
+{
+  return compose(binary_operator<bit_or_equal>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator|=()
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<bit_or_equal>,
+    actor<T1>,
+    actor<T2>
+  >
+>
+operator|=(const actor<T1> &_1, const actor<T2> &_2)
+{
+  return compose(binary_operator<bit_or_equal>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator|=()
+
+template<typename T>
+  struct bit_xor_equal
+    : public thrust::binary_function<T&,T,T&>
+{
+  __host__ __device__ T& operator()(T &lhs, const T&rhs) const { return lhs ^= rhs; }
+}; // end bit_xor_equal
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<bit_xor_equal>,
+    actor<T1>,
+    typename as_actor<T2>::type
+  >
+>
+operator^=(const actor<T1> &_1, const T2 &_2)
+{
+  return compose(binary_operator<bit_xor_equal>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator|=()
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<bit_xor_equal>,
+    actor<T1>,
+    actor<T2>
+  >
+>
+operator^=(const actor<T1> &_1, const actor<T2> &_2)
+{
+  return compose(binary_operator<bit_xor_equal>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator|=()
+
+template<typename T>
+  struct bit_lshift_equal
+    : public thrust::binary_function<T&,T,T&>
+{
+  __host__ __device__ T& operator()(T &lhs, const T&rhs) const { return lhs <<= rhs; }
+}; // end bit_lshift_equal
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<bit_lshift_equal>,
+    actor<T1>,
+    typename as_actor<T2>::type
+  >
+>
+operator<<=(const actor<T1> &_1, const T2 &_2)
+{
+  return compose(binary_operator<bit_lshift_equal>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator<<=()
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<bit_lshift_equal>,
+    actor<T1>,
+    actor<T2>
+  >
+>
+operator<<=(const actor<T1> &_1, const actor<T2> &_2)
+{
+  return compose(binary_operator<bit_lshift_equal>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator<<=()
+
+template<typename T>
+  struct bit_rshift_equal
+    : public thrust::binary_function<T&,T,T&>
+{
+  __host__ __device__ T& operator()(T &lhs, const T&rhs) const { return lhs >>= rhs; }
+}; // end bit_rshift_equal
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<bit_rshift_equal>,
+    actor<T1>,
+    typename as_actor<T2>::type
+  >
+>
+operator>>=(const actor<T1> &_1, const T2 &_2)
+{
+  return compose(binary_operator<bit_rshift_equal>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator>>=()
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<bit_rshift_equal>,
+    actor<T1>,
+    actor<T2>
+  >
+>
+operator>>=(const actor<T1> &_1, const actor<T2> &_2)
+{
+  return compose(binary_operator<bit_rshift_equal>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator>>=()
+
+} // end functional
+} // end detail
+} // end thrust
+
diff --git a/compat/thrust/detail/functional/operators/logical_operators.h b/compat/thrust/detail/functional/operators/logical_operators.h
new file mode 100644
index 0000000000..9c952620db
--- /dev/null
+++ b/compat/thrust/detail/functional/operators/logical_operators.h
@@ -0,0 +1,144 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/functional/actor.h>
+#include <thrust/detail/functional/composite.h>
+#include <thrust/detail/functional/operators/operator_adaptors.h>
+#include <thrust/functional.h>
+
+namespace thrust
+{
+namespace detail
+{
+namespace functional
+{
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<thrust::logical_and>,
+    actor<T1>,
+    typename as_actor<T2>::type
+  >
+>
+operator&&(const actor<T1> &_1, const T2 &_2)
+{
+  return compose(binary_operator<thrust::logical_and>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator&&()
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<thrust::logical_and>,
+    typename as_actor<T1>::type,
+    actor<T2>
+  >
+>
+operator&&(const T1 &_1, const actor<T2> &_2)
+{
+  return compose(binary_operator<thrust::logical_and>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator&&()
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<thrust::logical_and>,
+    actor<T1>,
+    actor<T2>
+  >
+>
+operator&&(const actor<T1> &_1, const actor<T2> &_2)
+{
+  return compose(binary_operator<thrust::logical_and>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator&&()
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<thrust::logical_or>,
+    actor<T1>,
+    typename as_actor<T2>::type
+  >
+>
+operator||(const actor<T1> &_1, const T2 &_2)
+{
+  return compose(binary_operator<thrust::logical_or>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator&&()
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<thrust::logical_or>,
+    typename as_actor<T1>::type,
+    actor<T2>
+  >
+>
+operator||(const T1 &_1, const actor<T2> &_2)
+{
+  return compose(binary_operator<thrust::logical_or>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator&&()
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<thrust::logical_or>,
+    actor<T1>,
+    actor<T2>
+  >
+>
+operator||(const actor<T1> &_1, const actor<T2> &_2)
+{
+  return compose(binary_operator<thrust::logical_or>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator&&()
+
+template<typename Eval>
+__host__ __device__
+actor<
+  composite<
+    unary_operator<thrust::logical_not>,
+    actor<Eval>
+  >
+>
+operator!(const actor<Eval> &_1)
+{
+  return compose(unary_operator<thrust::logical_not>(), _1);
+} // end operator!()
+
+} // end functional
+} // end detail
+} // end thrust
+
diff --git a/compat/thrust/detail/functional/operators/operator_adaptors.h b/compat/thrust/detail/functional/operators/operator_adaptors.h
new file mode 100644
index 0000000000..d35fe9726b
--- /dev/null
+++ b/compat/thrust/detail/functional/operators/operator_adaptors.h
@@ -0,0 +1,115 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/tuple.h>
+#include <thrust/detail/type_traits.h>
+
+namespace thrust
+{
+namespace detail
+{
+namespace functional
+{
+
+// this thing (which models Eval) is an adaptor for the unary
+// functors inside functional.h
+template<template<typename> class UnaryOperator>
+  struct unary_operator
+{
+  template<typename Env>
+    struct argument
+      : thrust::detail::eval_if<
+          (thrust::tuple_size<Env>::value == 0),
+          thrust::detail::identity_<thrust::null_type>,
+          thrust::tuple_element<0,Env>
+        >
+  {
+  };
+
+  template<typename Env>
+    struct operator_type
+  {
+    typedef UnaryOperator<
+      typename thrust::detail::remove_reference<
+        typename argument<Env>::type
+      >::type
+    > type;
+  };
+
+  template<typename Env>
+    struct result
+  {
+    typedef typename operator_type<Env>::type op_type;
+    typedef typename op_type::result_type type;
+  };
+
+  template<typename Env>
+  __host__ __device__
+  typename result<Env>::type eval(const Env &e) const
+  {
+    typename operator_type<Env>::type op;
+    return op(thrust::get<0>(e));
+  } // end eval()
+}; // end unary_operator
+
+// this thing (which models Eval) is an adaptor for the binary
+// functors inside functional.h
+template<template<typename> class BinaryOperator>
+  struct binary_operator
+{
+  template<typename Env>
+    struct first_argument
+      : thrust::detail::eval_if<
+          (thrust::tuple_size<Env>::value == 0),
+          thrust::detail::identity_<thrust::null_type>,
+          thrust::tuple_element<0,Env>
+        >
+  {
+  };
+
+  template<typename Env>
+    struct operator_type
+  {
+    typedef BinaryOperator<
+      typename thrust::detail::remove_reference<
+        typename first_argument<Env>::type
+      >::type
+    > type;
+  };
+
+  template<typename Env>
+    struct result
+  {
+    typedef typename operator_type<Env>::type op_type;
+    typedef typename op_type::result_type type;
+  };
+
+  template<typename Env>
+  __host__ __device__
+  typename result<Env>::type eval(const Env &e) const
+  {
+    typename operator_type<Env>::type op;
+    return op(thrust::get<0>(e), thrust::get<1>(e));
+  } // end eval()
+}; // end binary_operator
+
+} // end functional
+} // end detail
+} // end thrust
+
diff --git a/compat/thrust/detail/functional/operators/relational_operators.h b/compat/thrust/detail/functional/operators/relational_operators.h
new file mode 100644
index 0000000000..6b26534430
--- /dev/null
+++ b/compat/thrust/detail/functional/operators/relational_operators.h
@@ -0,0 +1,323 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/functional/actor.h>
+#include <thrust/detail/functional/composite.h>
+#include <thrust/detail/functional/operators/operator_adaptors.h>
+#include <thrust/functional.h>
+
+namespace thrust
+{
+namespace detail
+{
+namespace functional
+{
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<thrust::equal_to>,
+    actor<T1>,
+    typename as_actor<T2>::type
+  >
+>
+operator==(const actor<T1> &_1, const T2 &_2)
+{
+  return compose(binary_operator<thrust::equal_to>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator==()
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<thrust::equal_to>,
+    typename as_actor<T1>::type,
+    actor<T2>
+  >
+>
+operator==(const T1 &_1, const actor<T2> &_2)
+{
+  return compose(binary_operator<thrust::equal_to>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator==()
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<thrust::equal_to>,
+    actor<T1>,
+    actor<T2>
+  >
+>
+operator==(const actor<T1> &_1, const actor<T2> &_2)
+{
+  return compose(binary_operator<thrust::equal_to>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator==()
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<thrust::not_equal_to>,
+    actor<T1>,
+    typename as_actor<T2>::type
+  >
+>
+operator!=(const actor<T1> &_1, const T2 &_2)
+{
+  return compose(binary_operator<thrust::not_equal_to>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator!=()
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<thrust::not_equal_to>,
+    typename as_actor<T1>::type,
+    actor<T2>
+  >
+>
+operator!=(const T1 &_1, const actor<T2> &_2)
+{
+  return compose(binary_operator<thrust::not_equal_to>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator!=()
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<thrust::not_equal_to>,
+    actor<T1>,
+    actor<T2>
+  >
+>
+operator!=(const actor<T1> &_1, const actor<T2> &_2)
+{
+  return compose(binary_operator<thrust::not_equal_to>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator!=()
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<thrust::greater>,
+    actor<T1>,
+    typename as_actor<T2>::type
+  >
+>
+operator>(const actor<T1> &_1, const T2 &_2)
+{
+  return compose(binary_operator<thrust::greater>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator>()
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<thrust::greater>,
+    typename as_actor<T1>::type,
+    actor<T2>
+  >
+>
+operator>(const T1 &_1, const actor<T2> &_2)
+{
+  return compose(binary_operator<thrust::greater>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator>()
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<thrust::greater>,
+    actor<T1>,
+    actor<T2>
+  >
+>
+operator>(const actor<T1> &_1, const actor<T2> &_2)
+{
+  return compose(binary_operator<thrust::greater>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator>()
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<thrust::less>,
+    actor<T1>,
+    typename as_actor<T2>::type
+  >
+>
+operator<(const actor<T1> &_1, const T2 &_2)
+{
+  return compose(binary_operator<thrust::less>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator<()
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<thrust::less>,
+    typename as_actor<T1>::type,
+    actor<T2>
+  >
+>
+operator<(const T1 &_1, const actor<T2> &_2)
+{
+  return compose(binary_operator<thrust::less>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator<()
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<thrust::less>,
+    actor<T1>,
+    actor<T2>
+  >
+>
+operator<(const actor<T1> &_1, const actor<T2> &_2)
+{
+  return compose(binary_operator<thrust::less>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator<()
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<thrust::greater_equal>,
+    actor<T1>,
+    typename as_actor<T2>::type
+  >
+>
+operator>=(const actor<T1> &_1, const T2 &_2)
+{
+  return compose(binary_operator<thrust::greater_equal>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator>=()
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<thrust::greater_equal>,
+    typename as_actor<T1>::type,
+    actor<T2>
+  >
+>
+operator>=(const T1 &_1, const actor<T2> &_2)
+{
+  return compose(binary_operator<thrust::greater_equal>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator>=()
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<thrust::greater_equal>,
+    actor<T1>,
+    actor<T2>
+  >
+>
+operator>=(const actor<T1> &_1, const actor<T2> &_2)
+{
+  return compose(binary_operator<thrust::greater_equal>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator>=()
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<thrust::less_equal>,
+    actor<T1>,
+    typename as_actor<T2>::type
+  >
+>
+operator<=(const actor<T1> &_1, const T2 &_2)
+{
+  return compose(binary_operator<thrust::less_equal>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator<=()
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<thrust::less_equal>,
+    typename as_actor<T1>::type,
+    actor<T2>
+  >
+>
+operator<=(const T1 &_1, const actor<T2> &_2)
+{
+  return compose(binary_operator<thrust::less_equal>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator<=()
+
+template<typename T1, typename T2>
+__host__ __device__
+actor<
+  composite<
+    binary_operator<thrust::less_equal>,
+    actor<T1>,
+    actor<T2>
+  >
+>
+operator<=(const actor<T1> &_1, const actor<T2> &_2)
+{
+  return compose(binary_operator<thrust::less_equal>(),
+                 make_actor(_1),
+                 make_actor(_2));
+} // end operator<=()
+
+} // end functional
+} // end detail
+} // end thrust
+
diff --git a/compat/thrust/detail/functional/placeholder.h b/compat/thrust/detail/functional/placeholder.h
new file mode 100644
index 0000000000..9acf6da803
--- /dev/null
+++ b/compat/thrust/detail/functional/placeholder.h
@@ -0,0 +1,39 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/functional/actor.h>
+#include <thrust/detail/functional/argument.h>
+
+namespace thrust
+{
+namespace detail
+{
+namespace functional
+{
+
+template<unsigned int i>
+  struct placeholder
+{
+  typedef actor<argument<i> > type;
+};
+
+} // end functional
+} // end detail
+} // end thrust
+
diff --git a/compat/thrust/detail/functional/value.h b/compat/thrust/detail/functional/value.h
new file mode 100644
index 0000000000..27e2802e3a
--- /dev/null
+++ b/compat/thrust/detail/functional/value.h
@@ -0,0 +1,80 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+// Portions of this code are derived from
+//
+// Manjunath Kudlur's Carbon library
+//
+// and
+//
+// Based on Boost.Phoenix v1.2
+// Copyright (c) 2001-2002 Joel de Guzman
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/functional/actor.h>
+
+namespace thrust
+{
+namespace detail
+{
+namespace functional
+{
+
+
+template<typename Eval> struct actor;
+
+
+template<typename T>
+  class value
+{
+  public:
+
+    template<typename Env>
+      struct result
+    {
+      typedef T type;
+    };
+
+    __host__ __device__
+    value(const T &arg)
+      : m_val(arg)
+    {}
+
+    template<typename Env>
+    __host__ __device__
+      T eval(const Env &) const
+    {
+      return m_val;
+    }
+
+  private:
+    T m_val;
+}; // end value
+
+template<typename T>
+__host__ __device__
+actor<value<T> > val(const T &x)
+{
+  return value<T>(x);
+} // end val()
+
+
+} // end functional
+} // end detail
+} // end thrust
+
diff --git a/compat/thrust/detail/gather.inl b/compat/thrust/detail/gather.inl
new file mode 100644
index 0000000000..4edecd038a
--- /dev/null
+++ b/compat/thrust/detail/gather.inl
@@ -0,0 +1,160 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file gather.inl
+ *  \brief Inline file for gather.h.
+ */
+
+#include <thrust/gather.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/system/detail/generic/select_system.h>
+#include <thrust/system/detail/generic/gather.h>
+#include <thrust/system/detail/adl/gather.h>
+
+namespace thrust
+{
+
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename RandomAccessIterator,
+         typename OutputIterator>
+  OutputIterator gather(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                        InputIterator                                               map_first,
+                        InputIterator                                               map_last,
+                        RandomAccessIterator                                        input_first,
+                        OutputIterator                                              result)
+{
+  using thrust::system::detail::generic::gather;
+  return gather(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), map_first, map_last, input_first, result);
+} // end gather()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename RandomAccessIterator,
+         typename OutputIterator>
+  OutputIterator gather_if(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                           InputIterator1                                              map_first,
+                           InputIterator1                                              map_last,
+                           InputIterator2                                              stencil,
+                           RandomAccessIterator                                        input_first,
+                           OutputIterator                                              result)
+{
+  using thrust::system::detail::generic::gather_if;
+  return gather_if(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), map_first, map_last, stencil, input_first, result);
+} // end gather_if()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename RandomAccessIterator,
+         typename OutputIterator,
+         typename Predicate>
+  OutputIterator gather_if(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                           InputIterator1                                              map_first,
+                           InputIterator1                                              map_last,
+                           InputIterator2                                              stencil,
+                           RandomAccessIterator                                        input_first,
+                           OutputIterator                                              result,
+                           Predicate                                                   pred)
+{
+  using thrust::system::detail::generic::gather_if;
+  return gather_if(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), map_first, map_last, stencil, input_first, result, pred);
+} // end gather_if()
+
+
+template<typename InputIterator,
+         typename RandomAccessIterator,
+         typename OutputIterator>
+  OutputIterator gather(InputIterator        map_first,
+                        InputIterator        map_last,
+                        RandomAccessIterator input_first,
+                        OutputIterator       result)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator>::type        System1; 
+  typedef typename thrust::iterator_system<RandomAccessIterator>::type System2; 
+  typedef typename thrust::iterator_system<OutputIterator>::type       System3; 
+
+  System1 system1;
+  System2 system2;
+  System3 system3;
+
+  return thrust::gather(select_system(system1,system2,system3), map_first, map_last, input_first, result);
+} // end gather()
+
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename RandomAccessIterator,
+         typename OutputIterator>
+  OutputIterator gather_if(InputIterator1       map_first,
+                           InputIterator1       map_last,
+                           InputIterator2       stencil,
+                           RandomAccessIterator input_first,
+                           OutputIterator       result)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator1>::type       System1; 
+  typedef typename thrust::iterator_system<InputIterator2>::type       System2; 
+  typedef typename thrust::iterator_system<RandomAccessIterator>::type System3; 
+  typedef typename thrust::iterator_system<OutputIterator>::type       System4; 
+
+  System1 system1;
+  System2 system2;
+  System3 system3;
+  System4 system4;
+
+  return thrust::gather_if(select_system(system1,system2,system3,system4), map_first, map_last, stencil, input_first, result);
+} // end gather_if()
+
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename RandomAccessIterator,
+         typename OutputIterator,
+         typename Predicate>
+  OutputIterator gather_if(InputIterator1       map_first,
+                           InputIterator1       map_last,
+                           InputIterator2       stencil,
+                           RandomAccessIterator input_first,
+                           OutputIterator       result,
+                           Predicate            pred)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator1>::type       System1; 
+  typedef typename thrust::iterator_system<InputIterator2>::type       System2; 
+  typedef typename thrust::iterator_system<RandomAccessIterator>::type System3; 
+  typedef typename thrust::iterator_system<OutputIterator>::type       System4; 
+
+  System1 system1;
+  System2 system2;
+  System3 system3;
+  System4 system4;
+
+  return thrust::gather_if(select_system(system1,system2,system3,system4), map_first, map_last, stencil, input_first, result, pred);
+} // end gather_if()
+
+
+} // end namespace thrust
+
diff --git a/compat/thrust/detail/generate.inl b/compat/thrust/detail/generate.inl
new file mode 100644
index 0000000000..c12580452e
--- /dev/null
+++ b/compat/thrust/detail/generate.inl
@@ -0,0 +1,94 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file generate.inl
+ *  \author Jared Hoberock
+ *  \brief Inline file for generate.h.
+ */
+
+#include <thrust/generate.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/system/detail/generic/select_system.h>
+#include <thrust/system/detail/generic/generate.h>
+#include <thrust/system/detail/adl/generate.h>
+
+namespace thrust
+{
+
+
+template<typename DerivedPolicy,
+         typename ForwardIterator,
+         typename Generator>
+  void generate(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                ForwardIterator first,
+                ForwardIterator last,
+                Generator gen)
+{
+  using thrust::system::detail::generic::generate;
+  return generate(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, gen);
+} // end generate()
+
+
+template<typename DerivedPolicy,
+         typename OutputIterator,
+         typename Size,
+         typename Generator>
+  OutputIterator generate_n(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                            OutputIterator first,
+                            Size n,
+                            Generator gen)
+{
+  using thrust::system::detail::generic::generate_n;
+  return generate_n(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, n, gen);
+} // end generate_n()
+
+
+template<typename ForwardIterator,
+         typename Generator>
+  void generate(ForwardIterator first,
+                ForwardIterator last,
+                Generator gen)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<ForwardIterator>::type System;
+
+  System system;
+
+  return thrust::generate(select_system(system), first, last, gen);
+} // end generate()
+
+
+template<typename OutputIterator,
+         typename Size,
+         typename Generator>
+  OutputIterator generate_n(OutputIterator first,
+                            Size n,
+                            Generator gen)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<OutputIterator>::type System;
+
+  System system;
+
+  return thrust::generate_n(select_system(system), first, n, gen);
+} // end generate_n()
+
+
+} // end thrust
+
diff --git a/compat/thrust/detail/host_vector.inl b/compat/thrust/detail/host_vector.inl
new file mode 100644
index 0000000000..e5c60ab973
--- /dev/null
+++ b/compat/thrust/detail/host_vector.inl
@@ -0,0 +1,37 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file host_vector.inl
+ *  \brief Inline file for host_vector.h.
+ */
+
+#include <thrust/host_vector.h>
+
+namespace thrust
+{
+
+template<typename T, typename Alloc>
+  template<typename OtherT, typename OtherAlloc>
+    host_vector<T,Alloc>
+      ::host_vector(const device_vector<OtherT,OtherAlloc> &v)
+        :Parent(v)
+{
+  ;
+} // end host_vector::host_vector()
+
+} // end namespace thrust
+
diff --git a/compat/thrust/detail/inner_product.inl b/compat/thrust/detail/inner_product.inl
new file mode 100644
index 0000000000..f7773d8d2c
--- /dev/null
+++ b/compat/thrust/detail/inner_product.inl
@@ -0,0 +1,104 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file inner_product.inl
+ *  \brief Inline file for inner_product.h.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/inner_product.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/system/detail/generic/select_system.h>
+#include <thrust/system/detail/generic/inner_product.h>
+#include <thrust/system/detail/adl/inner_product.h>
+
+namespace thrust
+{
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputType>
+OutputType inner_product(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                         InputIterator1 first1,
+                         InputIterator1 last1,
+                         InputIterator2 first2,
+                         OutputType init)
+{
+  using thrust::system::detail::generic::inner_product;
+  return inner_product(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first1, last1, first2, init);
+} // end inner_product()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputType,
+         typename BinaryFunction1,
+         typename BinaryFunction2>
+OutputType inner_product(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                         InputIterator1 first1,
+                         InputIterator1 last1,
+                         InputIterator2 first2,
+                         OutputType init, 
+                         BinaryFunction1 binary_op1,
+                         BinaryFunction2 binary_op2)
+{
+  using thrust::system::detail::generic::inner_product;
+  return inner_product(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first1, last1, first2, init, binary_op1, binary_op2);
+} // end inner_product()
+
+
+template <typename InputIterator1, typename InputIterator2, typename OutputType>
+OutputType 
+inner_product(InputIterator1 first1, InputIterator1 last1,
+              InputIterator2 first2, OutputType init)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator1>::type System1;
+  typedef typename thrust::iterator_system<InputIterator2>::type System2;
+
+  System1 system1;
+  System2 system2;
+
+  return thrust::inner_product(select_system(system1,system2), first1, last1, first2, init);
+} // end inner_product()
+
+
+template <typename InputIterator1, typename InputIterator2, typename OutputType,
+          typename BinaryFunction1, typename BinaryFunction2>
+OutputType
+inner_product(InputIterator1 first1, InputIterator1 last1,
+              InputIterator2 first2, OutputType init, 
+              BinaryFunction1 binary_op1, BinaryFunction2 binary_op2)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator1>::type System1;
+  typedef typename thrust::iterator_system<InputIterator2>::type System2;
+
+  System1 system1;
+  System2 system2;
+
+  return thrust::inner_product(select_system(system1,system2), first1, last1, first2, init, binary_op1, binary_op2);
+} // end inner_product()
+
+
+} // end namespace thrust
+
diff --git a/compat/thrust/detail/integer_traits.h b/compat/thrust/detail/integer_traits.h
new file mode 100644
index 0000000000..e4cf5d159f
--- /dev/null
+++ b/compat/thrust/detail/integer_traits.h
@@ -0,0 +1,132 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <limits>
+#include <limits.h>
+
+namespace thrust
+{
+
+namespace detail
+{
+
+template<typename T>
+  class integer_traits
+{
+  public:
+    static const bool is_integral = false;
+};
+
+template<typename T, T min_val, T max_val>
+  class integer_traits_base
+{
+  public:
+    static const bool is_integral = true;
+    static const T const_min = min_val;
+    static const T const_max = max_val;
+};
+
+
+template<>
+  class integer_traits<bool>
+    : public std::numeric_limits<bool>,
+      public integer_traits_base<bool, false, true>
+{};
+
+
+template<>
+  class integer_traits<char>
+    : public std::numeric_limits<char>,
+      public integer_traits_base<char, CHAR_MIN, CHAR_MAX>
+{};
+
+
+template<>
+  class integer_traits<signed char>
+    : public std::numeric_limits<signed char>,
+      public integer_traits_base<signed char, SCHAR_MIN, SCHAR_MAX>
+{};
+
+
+template<>
+  class integer_traits<unsigned char>
+    : public std::numeric_limits<unsigned char>,
+      public integer_traits_base<unsigned char, 0, UCHAR_MAX>
+{};
+
+
+template<>
+  class integer_traits<short>
+    : public std::numeric_limits<short>,
+      public integer_traits_base<short, SHRT_MIN, SHRT_MAX>
+{};
+
+
+template<>
+  class integer_traits<unsigned short>
+    : public std::numeric_limits<unsigned short>,
+      public integer_traits_base<unsigned short, 0, USHRT_MAX>
+{};
+
+
+template<>
+  class integer_traits<int>
+    : public std::numeric_limits<int>,
+      public integer_traits_base<int, INT_MIN, INT_MAX>
+{};
+
+
+template<>
+  class integer_traits<unsigned int>
+    : public std::numeric_limits<unsigned int>,
+      public integer_traits_base<unsigned int, 0, UINT_MAX>
+{};
+
+
+template<>
+  class integer_traits<long>
+    : public std::numeric_limits<long>,
+      public integer_traits_base<long, LONG_MIN, LONG_MAX>
+{};
+
+
+template<>
+  class integer_traits<unsigned long>
+    : public std::numeric_limits<unsigned long>,
+      public integer_traits_base<unsigned long, 0, ULONG_MAX>
+{};
+
+
+template<>
+  class integer_traits<long long>
+    : public std::numeric_limits<long long>,
+      public integer_traits_base<long long, LLONG_MIN, LLONG_MAX>
+{};
+
+
+template<>
+  class integer_traits<unsigned long long>
+    : public std::numeric_limits<unsigned long long>,
+      public integer_traits_base<unsigned long long, 0, ULLONG_MAX>
+{};
+
+} // end detail
+
+} // end thrust
+
diff --git a/compat/thrust/detail/internal_functional.h b/compat/thrust/detail/internal_functional.h
new file mode 100644
index 0000000000..6d5264ae35
--- /dev/null
+++ b/compat/thrust/detail/internal_functional.h
@@ -0,0 +1,678 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file internal_functional.inl
+ *  \brief Non-public functionals used to implement algorithm internals.
+ */
+
+#pragma once
+
+#include <thrust/tuple.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/detail/type_traits.h>
+#include <thrust/iterator/detail/tuple_of_iterator_references.h>
+#include <thrust/detail/raw_reference_cast.h>
+#include <memory> // for ::new
+
+namespace thrust
+{
+namespace detail
+{
+
+// unary_negate does not need to know argument_type
+template <typename Predicate>
+struct unary_negate
+{
+    typedef bool result_type;
+
+    Predicate pred;
+
+    __host__ __device__
+    explicit unary_negate(const Predicate& pred) : pred(pred) {}
+
+    template <typename T>
+    __host__ __device__
+    bool operator()(const T& x)
+    {
+        return !bool(pred(x));
+    }
+};
+
+// binary_negate does not need to know first_argument_type or second_argument_type
+template <typename Predicate>
+struct binary_negate
+{
+    typedef bool result_type;
+
+    Predicate pred;
+
+    __host__ __device__
+    explicit binary_negate(const Predicate& pred) : pred(pred) {}
+
+    template <typename T1, typename T2>
+        __host__ __device__
+        bool operator()(const T1& x, const T2& y)
+        {
+            return !bool(pred(x,y));
+        }
+};
+
+template<typename Predicate>
+  __host__ __device__
+  thrust::detail::unary_negate<Predicate> not1(const Predicate &pred)
+{
+    return thrust::detail::unary_negate<Predicate>(pred);
+}
+
+template<typename Predicate>
+  __host__ __device__
+  thrust::detail::binary_negate<Predicate> not2(const Predicate &pred)
+{
+    return thrust::detail::binary_negate<Predicate>(pred);
+}
+
+
+// convert a predicate to a 0 or 1 integral value
+template <typename Predicate, typename IntegralType>
+struct predicate_to_integral
+{
+    Predicate pred;
+
+    __host__ __device__
+    explicit predicate_to_integral(const Predicate& pred) : pred(pred) {}
+
+    template <typename T>
+        __host__ __device__
+        bool operator()(const T& x)
+        {
+            return pred(x) ? IntegralType(1) : IntegralType(0);
+        }
+};
+
+
+// note that detail::equal_to does not force conversion from T2 -> T1 as equal_to does
+template <typename T1>
+struct equal_to
+{
+    typedef bool result_type;
+
+    template <typename T2>
+        __host__ __device__
+        bool operator()(const T1& lhs, const T2& rhs) const
+        {
+            return lhs == rhs;
+        }
+};
+
+// note that equal_to_value does not force conversion from T2 -> T1 as equal_to does
+template <typename T2>
+struct equal_to_value
+{
+    T2 rhs;
+
+    equal_to_value(const T2& rhs) : rhs(rhs) {}
+
+    template <typename T1>
+        __host__ __device__
+        bool operator()(const T1& lhs) const
+        {
+            return lhs == rhs;
+        }
+};
+
+template <typename Predicate>
+struct tuple_binary_predicate
+{
+    typedef bool result_type;
+
+    __host__ __device__
+        tuple_binary_predicate(const Predicate& p) : pred(p) {}
+
+    template<typename Tuple>
+        __host__ __device__
+        bool operator()(const Tuple& t) const
+        { 
+            return pred(thrust::get<0>(t), thrust::get<1>(t));
+        }
+
+    Predicate pred;
+};
+
+template <typename Predicate>
+struct tuple_not_binary_predicate
+{
+    typedef bool result_type;
+
+    __host__ __device__
+        tuple_not_binary_predicate(const Predicate& p) : pred(p) {}
+
+    template<typename Tuple>
+        __host__ __device__
+        bool operator()(const Tuple& t) const
+        { 
+            return !pred(thrust::get<0>(t), thrust::get<1>(t));
+        }
+
+    Predicate pred;
+};
+
+template<typename Generator>
+  struct host_generate_functor
+{
+  typedef void result_type;
+
+  __host__ __device__
+  host_generate_functor(Generator g)
+    : gen(g) {}
+
+  // operator() does not take an lvalue reference because some iterators
+  // produce temporary proxy references when dereferenced. for example,
+  // consider the temporary tuple of references produced by zip_iterator.
+  // such temporaries cannot bind to an lvalue reference.
+  //
+  // to WAR this, accept a const reference (which is bindable to a temporary),
+  // and const_cast in the implementation.
+  //
+  // XXX change to an rvalue reference upon c++0x (which either a named variable
+  //     or temporary can bind to)
+  template<typename T>
+  __host__
+  void operator()(const T &x)
+  {
+    // we have to be naughty and const_cast this to get it to work
+    T &lvalue = const_cast<T&>(x);
+
+    // this assigns correctly whether x is a true reference or proxy
+    lvalue = gen();
+  }
+
+  Generator gen;
+};
+
+template<typename Generator>
+  struct device_generate_functor
+{
+  typedef void result_type;
+
+  __host__ __device__
+  device_generate_functor(Generator g)
+    : gen(g) {}
+
+  // operator() does not take an lvalue reference because some iterators
+  // produce temporary proxy references when dereferenced. for example,
+  // consider the temporary tuple of references produced by zip_iterator.
+  // such temporaries cannot bind to an lvalue reference.
+  //
+  // to WAR this, accept a const reference (which is bindable to a temporary),
+  // and const_cast in the implementation.
+  //
+  // XXX change to an rvalue reference upon c++0x (which either a named variable
+  //     or temporary can bind to)
+  template<typename T>
+  __host__ __device__
+  void operator()(const T &x)
+  {
+    // we have to be naughty and const_cast this to get it to work
+    T &lvalue = const_cast<T&>(x);
+
+    // this assigns correctly whether x is a true reference or proxy
+    lvalue = gen();
+  }
+
+  Generator gen;
+};
+
+template<typename System, typename Generator>
+  struct generate_functor
+    : thrust::detail::eval_if<
+        thrust::detail::is_convertible<System, thrust::host_system_tag>::value,
+        thrust::detail::identity_<host_generate_functor<Generator> >,
+        thrust::detail::identity_<device_generate_functor<Generator> >
+      >
+{};
+
+
+template<typename ResultType, typename BinaryFunction>
+  struct zipped_binary_op
+{
+  typedef ResultType result_type;
+
+  __host__ __device__
+  zipped_binary_op(BinaryFunction binary_op)
+    : m_binary_op(binary_op) {}
+
+  template<typename Tuple>
+  __host__ __device__
+  inline result_type operator()(Tuple t)
+  {
+    return m_binary_op(thrust::get<0>(t), thrust::get<1>(t));
+  }
+
+  BinaryFunction m_binary_op;
+};
+
+
+template<typename T>
+  struct is_non_const_reference
+    : thrust::detail::and_<
+        thrust::detail::not_<thrust::detail::is_const<T> >,
+        thrust::detail::is_reference<T>
+      >
+{};
+
+template<typename T> struct is_tuple_of_iterator_references : thrust::detail::false_type {};
+
+template<typename T1, typename T2, typename T3,
+         typename T4, typename T5, typename T6,
+         typename T7, typename T8, typename T9,
+         typename T10>
+  struct is_tuple_of_iterator_references<
+    thrust::detail::tuple_of_iterator_references<
+      T1,T2,T3,T4,T5,T6,T7,T8,T9,T10
+    >
+  >
+    : thrust::detail::true_type
+{};
+
+// use this enable_if to avoid assigning to temporaries in the transform functors below
+// XXX revisit this problem with c++11 perfect forwarding
+template<typename T>
+  struct enable_if_non_const_reference_or_tuple_of_iterator_references
+    : thrust::detail::enable_if<
+        is_non_const_reference<T>::value || is_tuple_of_iterator_references<T>::value
+      >
+{};
+
+
+template<typename UnaryFunction>
+  struct host_unary_transform_functor
+{
+  typedef void result_type;
+
+  UnaryFunction f;
+
+  host_unary_transform_functor(UnaryFunction f_)
+    :f(f_) {}
+
+  template<typename Tuple>
+  inline __host__
+  typename enable_if_non_const_reference_or_tuple_of_iterator_references<
+    typename thrust::tuple_element<1,Tuple>::type
+  >::type
+    operator()(Tuple t)
+  {
+    thrust::get<1>(t) = f(thrust::get<0>(t));
+  }
+};
+
+template<typename UnaryFunction>
+  struct device_unary_transform_functor
+{
+  typedef void result_type;
+
+  UnaryFunction f;
+
+  device_unary_transform_functor(UnaryFunction f_)
+    :f(f_) {}
+
+  // add __host__ to allow the omp backend compile with nvcc
+  template<typename Tuple>
+  inline __host__ __device__
+  typename enable_if_non_const_reference_or_tuple_of_iterator_references<
+    typename thrust::tuple_element<1,Tuple>::type
+  >::type
+    operator()(Tuple t)
+  {
+    thrust::get<1>(t) = f(thrust::get<0>(t));
+  }
+};
+
+
+template<typename System, typename UnaryFunction>
+  struct unary_transform_functor
+    : thrust::detail::eval_if<
+        thrust::detail::is_convertible<System, thrust::host_system_tag>::value,
+        thrust::detail::identity_<host_unary_transform_functor<UnaryFunction> >,
+        thrust::detail::identity_<device_unary_transform_functor<UnaryFunction> >
+      >
+{};
+
+
+template <typename BinaryFunction>
+  struct host_binary_transform_functor
+{
+  BinaryFunction f;
+
+  host_binary_transform_functor(BinaryFunction f_)
+    :f(f_)
+  {}
+
+  template <typename Tuple>
+  __host__
+  void operator()(Tuple t)
+  { 
+    thrust::get<2>(t) = f(thrust::get<0>(t), thrust::get<1>(t));
+  }
+}; // end binary_transform_functor
+
+
+template <typename BinaryFunction>
+  struct device_binary_transform_functor
+{
+  BinaryFunction f;
+
+  device_binary_transform_functor(BinaryFunction f_)
+    :f(f_)
+  {}
+
+  // add __host__ to allow the omp backend compile with nvcc
+  template <typename Tuple>
+  inline __host__ __device__
+  typename enable_if_non_const_reference_or_tuple_of_iterator_references<
+    typename thrust::tuple_element<2,Tuple>::type
+  >::type
+    operator()(Tuple t)
+  { 
+    thrust::get<2>(t) = f(thrust::get<0>(t), thrust::get<1>(t));
+  }
+}; // end binary_transform_functor
+
+
+template<typename System, typename BinaryFunction>
+  struct binary_transform_functor
+    : thrust::detail::eval_if<
+        thrust::detail::is_convertible<System, thrust::host_system_tag>::value,
+        thrust::detail::identity_<host_binary_transform_functor<BinaryFunction> >,
+        thrust::detail::identity_<device_binary_transform_functor<BinaryFunction> >
+      >
+{};
+
+
+template <typename UnaryFunction, typename Predicate>
+struct host_unary_transform_if_functor
+{
+  UnaryFunction unary_op;
+  Predicate pred;
+
+  host_unary_transform_if_functor(UnaryFunction unary_op_, Predicate pred_)
+    : unary_op(unary_op_), pred(pred_) {}
+
+  template<typename Tuple>
+  inline __host__
+  typename enable_if_non_const_reference_or_tuple_of_iterator_references<
+    typename thrust::tuple_element<1,Tuple>::type
+  >::type
+    operator()(Tuple t)
+  {
+    if(pred(thrust::get<0>(t)))
+    {
+      thrust::get<1>(t) = unary_op(thrust::get<0>(t));
+    }
+  }
+}; // end host_unary_transform_if_functor
+
+
+template <typename UnaryFunction, typename Predicate>
+struct device_unary_transform_if_functor
+{
+  UnaryFunction unary_op;
+  Predicate pred;
+
+  device_unary_transform_if_functor(UnaryFunction unary_op_, Predicate pred_)
+    : unary_op(unary_op_), pred(pred_) {}
+
+  template<typename Tuple>
+  inline __host__ __device__
+  typename enable_if_non_const_reference_or_tuple_of_iterator_references<
+    typename thrust::tuple_element<1,Tuple>::type
+  >::type
+    operator()(Tuple t)
+  {
+    if(pred(thrust::get<0>(t)))
+    {
+      thrust::get<1>(t) = unary_op(thrust::get<0>(t));
+    }
+  }
+}; // end device_unary_transform_if_functor
+
+
+template<typename System, typename UnaryFunction, typename Predicate>
+  struct unary_transform_if_functor
+    : thrust::detail::eval_if<
+        thrust::detail::is_convertible<System, thrust::host_system_tag>::value,
+        thrust::detail::identity_<host_unary_transform_if_functor<UnaryFunction,Predicate> >,
+        thrust::detail::identity_<device_unary_transform_if_functor<UnaryFunction,Predicate> >
+      >
+{};
+
+
+template <typename UnaryFunction, typename Predicate>
+struct host_unary_transform_if_with_stencil_functor
+{
+  UnaryFunction unary_op;
+  Predicate pred;
+  
+  host_unary_transform_if_with_stencil_functor(UnaryFunction _unary_op, Predicate _pred)
+    : unary_op(_unary_op), pred(_pred) {} 
+  
+  template <typename Tuple>
+  inline __host__
+  typename enable_if_non_const_reference_or_tuple_of_iterator_references<
+    typename thrust::tuple_element<2,Tuple>::type
+  >::type
+    operator()(Tuple t)
+  {
+    if(pred(thrust::get<1>(t)))
+      thrust::get<2>(t) = unary_op(thrust::get<0>(t));
+  }
+}; // end host_unary_transform_if_with_stencil_functor
+
+
+template <typename UnaryFunction, typename Predicate>
+struct device_unary_transform_if_with_stencil_functor
+{
+  UnaryFunction unary_op;
+  Predicate pred;
+  
+  device_unary_transform_if_with_stencil_functor(UnaryFunction _unary_op, Predicate _pred)
+    : unary_op(_unary_op), pred(_pred) {} 
+  
+  // add __host__ to allow the omp backend compile with nvcc
+  template <typename Tuple>
+  inline __host__ __device__
+  typename enable_if_non_const_reference_or_tuple_of_iterator_references<
+    typename thrust::tuple_element<2,Tuple>::type
+  >::type
+    operator()(Tuple t)
+  {
+    if(pred(thrust::get<1>(t)))
+      thrust::get<2>(t) = unary_op(thrust::get<0>(t));
+  }
+}; // end device_unary_transform_if_with_stencil_functor
+
+
+template<typename System, typename UnaryFunction, typename Predicate>
+  struct unary_transform_if_with_stencil_functor
+    : thrust::detail::eval_if<
+        thrust::detail::is_convertible<System, thrust::host_system_tag>::value,
+        thrust::detail::identity_<host_unary_transform_if_with_stencil_functor<UnaryFunction,Predicate> >,
+        thrust::detail::identity_<device_unary_transform_if_with_stencil_functor<UnaryFunction,Predicate> >
+      >
+{};
+
+
+template <typename BinaryFunction, typename Predicate>
+struct host_binary_transform_if_functor
+{
+  BinaryFunction binary_op;
+  Predicate pred;
+
+  host_binary_transform_if_functor(BinaryFunction _binary_op, Predicate _pred)
+    : binary_op(_binary_op), pred(_pred) {} 
+
+  template <typename Tuple>
+  inline __host__
+  typename enable_if_non_const_reference_or_tuple_of_iterator_references<
+    typename thrust::tuple_element<3,Tuple>::type
+  >::type
+    operator()(Tuple t)
+  {
+    if(pred(thrust::get<2>(t)))
+      thrust::get<3>(t) = binary_op(thrust::get<0>(t), thrust::get<1>(t));
+  }
+}; // end host_binary_transform_if_functor
+
+
+template <typename BinaryFunction, typename Predicate>
+struct device_binary_transform_if_functor
+{
+  BinaryFunction binary_op;
+  Predicate pred;
+
+  device_binary_transform_if_functor(BinaryFunction _binary_op, Predicate _pred)
+    : binary_op(_binary_op), pred(_pred) {} 
+
+  // add __host__ to allow the omp backend compile with nvcc
+  template <typename Tuple>
+  inline __host__ __device__
+  typename enable_if_non_const_reference_or_tuple_of_iterator_references<
+    typename thrust::tuple_element<3,Tuple>::type
+  >::type
+    operator()(Tuple t)
+  {
+    if(pred(thrust::get<2>(t)))
+      thrust::get<3>(t) = binary_op(thrust::get<0>(t), thrust::get<1>(t));
+  }
+}; // end device_binary_transform_if_functor
+
+
+template<typename System, typename BinaryFunction, typename Predicate>
+  struct binary_transform_if_functor
+    : thrust::detail::eval_if<
+        thrust::detail::is_convertible<System, thrust::host_system_tag>::value,
+        thrust::detail::identity_<host_binary_transform_if_functor<BinaryFunction,Predicate> >,
+        thrust::detail::identity_<device_binary_transform_if_functor<BinaryFunction,Predicate> >
+      >
+{};
+
+
+template<typename T>
+  struct host_destroy_functor
+{
+  __host__
+  void operator()(T &x) const
+  {
+    x.~T();
+  } // end operator()()
+}; // end host_destroy_functor
+
+
+template<typename T>
+  struct device_destroy_functor
+{
+  // add __host__ to allow the omp backend to compile with nvcc
+  __host__ __device__
+  void operator()(T &x) const
+  {
+    x.~T();
+  } // end operator()()
+}; // end device_destroy_functor
+
+
+template<typename System, typename T>
+  struct destroy_functor
+    : thrust::detail::eval_if<
+        thrust::detail::is_convertible<System, thrust::host_system_tag>::value,
+        thrust::detail::identity_<host_destroy_functor<T> >,
+        thrust::detail::identity_<device_destroy_functor<T> >
+      >
+{};
+
+
+template <typename T>
+struct fill_functor
+{
+  const T exemplar;
+
+  fill_functor(const T& _exemplar) 
+    : exemplar(_exemplar) {}
+
+  __host__ __device__
+  T operator()(void) const
+  { 
+    return exemplar;
+  }
+};
+
+
+template<typename T>
+  struct uninitialized_fill_functor
+{
+  T exemplar;
+
+  uninitialized_fill_functor(T x):exemplar(x){}
+
+  __host__ __device__
+  void operator()(T &x)
+  {
+    ::new(static_cast<void*>(&x)) T(exemplar);
+  } // end operator()()
+}; // end uninitialized_fill_functor
+
+
+// this predicate tests two two-element tuples
+// we first use a Compare for the first element
+// if the first elements are equivalent, we use
+// < for the second elements
+template<typename Compare>
+  struct compare_first_less_second
+{
+  compare_first_less_second(Compare c)
+    : comp(c) {}
+
+  template<typename T1, typename T2>
+  __host__ __device__
+  bool operator()(T1 lhs, T2 rhs)
+  {
+    return comp(thrust::get<0>(lhs), thrust::get<0>(rhs)) || (!comp(thrust::get<0>(rhs), thrust::get<0>(lhs)) && thrust::get<1>(lhs) < thrust::get<1>(rhs));
+  }
+
+  Compare comp;
+}; // end compare_first_less_second
+
+
+template<typename Compare>
+  struct compare_first
+{
+  Compare comp;
+
+  compare_first(Compare comp)
+    : comp(comp)
+  {}
+
+  template<typename Tuple1, typename Tuple2>
+  __host__ __device__
+  bool operator()(const Tuple1 &x, const Tuple2 &y)
+  {
+    return comp(thrust::raw_reference_cast(thrust::get<0>(x)), thrust::raw_reference_cast(thrust::get<0>(y)));
+  }
+}; // end compare_first
+
+
+} // end namespace detail
+} // end namespace thrust
+
diff --git a/compat/thrust/detail/logical.inl b/compat/thrust/detail/logical.inl
new file mode 100644
index 0000000000..126a3e3fb1
--- /dev/null
+++ b/compat/thrust/detail/logical.inl
@@ -0,0 +1,96 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file logical.inl
+ *  \brief Inline file for logical.h.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/system/detail/generic/select_system.h>
+#include <thrust/system/detail/generic/logical.h>
+#include <thrust/system/detail/adl/logical.h>
+
+namespace thrust
+{
+
+
+template <typename DerivedPolicy, typename InputIterator, typename Predicate>
+bool all_of(const thrust::detail::execution_policy_base<DerivedPolicy> &exec, InputIterator first, InputIterator last, Predicate pred)
+{
+  using thrust::system::detail::generic::all_of;
+  return all_of(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, pred);
+} // end all_of()
+
+
+template <typename DerivedPolicy, typename InputIterator, typename Predicate>
+bool any_of(const thrust::detail::execution_policy_base<DerivedPolicy> &exec, InputIterator first, InputIterator last, Predicate pred)
+{
+  using thrust::system::detail::generic::any_of;
+  return any_of(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, pred);
+} // end any_of()
+
+
+template <typename DerivedPolicy, typename InputIterator, typename Predicate>
+bool none_of(const thrust::detail::execution_policy_base<DerivedPolicy> &exec, InputIterator first, InputIterator last, Predicate pred)
+{
+  using thrust::system::detail::generic::none_of;
+  return none_of(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, pred);
+} // end none_of()
+
+
+template <typename InputIterator, typename Predicate>
+bool all_of(InputIterator first, InputIterator last, Predicate pred)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator>::type System;
+
+  System system;
+
+  return thrust::all_of(select_system(system), first, last, pred);
+}
+
+
+template <typename InputIterator, typename Predicate>
+bool any_of(InputIterator first, InputIterator last, Predicate pred)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator>::type System;
+
+  System system;
+
+  return thrust::any_of(select_system(system), first, last, pred);
+}
+
+
+template <typename InputIterator, typename Predicate>
+bool none_of(InputIterator first, InputIterator last, Predicate pred)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator>::type System;
+
+  System system;
+
+  return thrust::none_of(select_system(system), first, last, pred);
+}
+
+
+} // end namespace thrust
+
diff --git a/compat/thrust/detail/malloc_and_free.h b/compat/thrust/detail/malloc_and_free.h
new file mode 100644
index 0000000000..57b1685476
--- /dev/null
+++ b/compat/thrust/detail/malloc_and_free.h
@@ -0,0 +1,79 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/execution_policy.h>
+#include <thrust/detail/pointer.h>
+#include <thrust/detail/raw_pointer_cast.h>
+#include <thrust/system/detail/generic/memory.h>
+#include <thrust/system/detail/adl/malloc_and_free.h>
+
+namespace thrust
+{
+
+template<typename DerivedPolicy>
+pointer<void,DerivedPolicy> malloc(const thrust::detail::execution_policy_base<DerivedPolicy> &exec, std::size_t n)
+{
+  using thrust::system::detail::generic::malloc;
+
+  // XXX should use a hypothetical thrust::static_pointer_cast here
+  void *raw_ptr = static_cast<void*>(thrust::raw_pointer_cast(malloc(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), n)));
+
+  return pointer<void,DerivedPolicy>(raw_ptr);
+}
+
+template<typename T, typename DerivedPolicy>
+pointer<T,DerivedPolicy> malloc(const thrust::detail::execution_policy_base<DerivedPolicy> &exec, std::size_t n)
+{
+  using thrust::system::detail::generic::malloc;
+
+  T *raw_ptr = static_cast<T*>(thrust::raw_pointer_cast(malloc<T>(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), n)));
+
+  return pointer<T,DerivedPolicy>(raw_ptr);
+}
+
+
+// XXX WAR nvbug 992955
+#if THRUST_DEVICE_COMPILER == THRUST_DEVICE_COMPILER_NVCC
+#if CUDA_VERSION < 5000
+
+// cudafe generates unqualified calls to free(int *volatile)
+// which get confused with thrust::free
+// spoof a thrust::free which simply maps to ::free
+inline __host__ __device__
+void free(int *volatile ptr)
+{
+  ::free(ptr);
+}
+
+#endif // CUDA_VERSION
+#endif // THRUST_DEVICE_COMPILER
+
+template<typename DerivedPolicy, typename Pointer>
+void free(const thrust::detail::execution_policy_base<DerivedPolicy> &exec, Pointer ptr)
+{
+  using thrust::system::detail::generic::free;
+
+  free(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), ptr);
+}
+
+// XXX consider another form of free which does not take a system argument and
+// instead infers the system from the pointer
+
+} // end namespace thrust
+
diff --git a/compat/thrust/detail/merge.inl b/compat/thrust/detail/merge.inl
new file mode 100644
index 0000000000..77f09f5bed
--- /dev/null
+++ b/compat/thrust/detail/merge.inl
@@ -0,0 +1,217 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+/*! \file merge.inl
+ *  \brief Inline file for merge.h.
+ */
+
+#include <thrust/merge.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/system/detail/generic/select_system.h>
+#include <thrust/system/detail/generic/merge.h>
+#include <thrust/system/detail/adl/merge.h>
+
+namespace thrust
+{
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator>
+  OutputIterator merge(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                       InputIterator1 first1,
+                       InputIterator1 last1,
+                       InputIterator2 first2,
+                       InputIterator2 last2,
+                       OutputIterator result)
+{
+  using thrust::system::detail::generic::merge;
+  return merge(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first1, last1, first2, last2, result);
+} // end merge()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename StrictWeakCompare>
+  OutputIterator merge(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                       InputIterator1 first1,
+                       InputIterator1 last1,
+                       InputIterator2 first2,
+                       InputIterator2 last2,
+                       OutputIterator result,
+                       StrictWeakCompare comp)
+{
+  using thrust::system::detail::generic::merge;
+  return merge(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first1, last1, first2, last2, result, comp);
+} // end merge()
+
+
+template<typename DerivedPolicy, typename InputIterator1, typename InputIterator2, typename InputIterator3, typename InputIterator4, typename OutputIterator1, typename OutputIterator2>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    merge_by_key(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                 InputIterator1 keys_first1, InputIterator1 keys_last1,
+                 InputIterator2 keys_first2, InputIterator2 keys_last2,
+                 InputIterator3 values_first1, InputIterator4 values_first2,
+                 OutputIterator1 keys_result,
+                 OutputIterator2 values_result)
+{
+  using thrust::system::detail::generic::merge_by_key;
+  return merge_by_key(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), keys_first1, keys_last1, keys_first2, keys_last2, values_first1, values_first2, keys_result, values_result);
+} // end merge_by_key()
+
+
+template<typename DerivedPolicy, typename InputIterator1, typename InputIterator2, typename InputIterator3, typename InputIterator4, typename OutputIterator1, typename OutputIterator2, typename Compare>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    merge_by_key(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                 InputIterator1 keys_first1, InputIterator1 keys_last1,
+                 InputIterator2 keys_first2, InputIterator2 keys_last2,
+                 InputIterator3 values_first1, InputIterator4 values_first2,
+                 OutputIterator1 keys_result,
+                 OutputIterator2 values_result,
+                 Compare comp)
+{
+  using thrust::system::detail::generic::merge_by_key;
+  return merge_by_key(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), keys_first1, keys_last1, keys_first2, keys_last2, values_first1, values_first2, keys_result, values_result, comp);
+} // end merge_by_key()
+
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename StrictWeakOrdering>
+  OutputIterator merge(InputIterator1 first1,
+                       InputIterator1 last1,
+                       InputIterator2 first2,
+                       InputIterator2 last2,
+                       OutputIterator result,
+                       StrictWeakOrdering comp)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator1>::type System1;
+  typedef typename thrust::iterator_system<InputIterator2>::type System2;
+  typedef typename thrust::iterator_system<OutputIterator>::type System3;
+
+  System1 system1;
+  System2 system2;
+  System3 system3;
+
+  return thrust::merge(select_system(system1,system2,system3), first1, last1, first2, last2, result, comp);
+} // end merge()
+
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator>
+  OutputIterator merge(InputIterator1 first1,
+                       InputIterator1 last1,
+                       InputIterator2 first2,
+                       InputIterator2 last2,
+                       OutputIterator result)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator1>::type System1;
+  typedef typename thrust::iterator_system<InputIterator2>::type System2;
+  typedef typename thrust::iterator_system<OutputIterator>::type System3;
+
+  System1 system1;
+  System2 system2;
+  System3 system3;
+
+  return thrust::merge(select_system(system1,system2,system3), first1, last1, first2, last2, result);
+} // end merge()
+
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename InputIterator4,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename StrictWeakOrdering>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    merge_by_key(InputIterator1 keys_first1,
+                 InputIterator1 keys_last1,
+                 InputIterator2 keys_first2,
+                 InputIterator2 keys_last2,
+                 InputIterator3 values_first1,
+                 InputIterator4 values_first2,
+                 OutputIterator1 keys_result,
+                 OutputIterator2 values_result,
+                 StrictWeakOrdering comp)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator1>::type  System1;
+  typedef typename thrust::iterator_system<InputIterator2>::type  System2;
+  typedef typename thrust::iterator_system<InputIterator3>::type  System3;
+  typedef typename thrust::iterator_system<InputIterator4>::type  System4;
+  typedef typename thrust::iterator_system<OutputIterator1>::type System5;
+  typedef typename thrust::iterator_system<OutputIterator2>::type System6;
+
+  System1 system1;
+  System2 system2;
+  System3 system3;
+  System4 system4;
+  System5 system5;
+  System6 system6;
+
+  return thrust::merge_by_key(select_system(system1,system2,system3,system4,system5,system6), keys_first1, keys_last1, keys_first2, keys_last2, values_first1, values_first2, keys_result, values_result, comp);
+} // end merge_by_key()
+
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename InputIterator4,
+         typename OutputIterator1,
+         typename OutputIterator2>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    merge_by_key(InputIterator1 keys_first1,
+                 InputIterator1 keys_last1,
+                 InputIterator2 keys_first2,
+                 InputIterator2 keys_last2,
+                 InputIterator3 values_first1,
+                 InputIterator4 values_first2,
+                 OutputIterator1 keys_result,
+                 OutputIterator2 values_result)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator1>::type  System1;
+  typedef typename thrust::iterator_system<InputIterator2>::type  System2;
+  typedef typename thrust::iterator_system<InputIterator3>::type  System3;
+  typedef typename thrust::iterator_system<InputIterator4>::type  System4;
+  typedef typename thrust::iterator_system<OutputIterator1>::type System5;
+  typedef typename thrust::iterator_system<OutputIterator2>::type System6;
+
+  System1 system1;
+  System2 system2;
+  System3 system3;
+  System4 system4;
+  System5 system5;
+  System6 system6;
+
+  return thrust::merge_by_key(select_system(system1,system2,system3,system4,system5,system6), keys_first1, keys_last1, keys_first2, keys_last2, values_first1, values_first2, keys_result, values_result);
+} // end merge_by_key()
+
+
+} // end thrust
+
diff --git a/compat/thrust/detail/minmax.h b/compat/thrust/detail/minmax.h
new file mode 100644
index 0000000000..a560ea1fe5
--- /dev/null
+++ b/compat/thrust/detail/minmax.h
@@ -0,0 +1,55 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+namespace thrust
+{
+
+
+template<typename T, typename BinaryPredicate>
+__host__ __device__
+  T min THRUST_PREVENT_MACRO_SUBSTITUTION (const T &lhs, const T &rhs, BinaryPredicate comp)
+{
+  return comp(rhs, lhs) ? rhs : lhs;
+} // end min()
+
+template<typename T>
+__host__ __device__
+  T min THRUST_PREVENT_MACRO_SUBSTITUTION (const T &lhs, const T &rhs)
+{
+  return rhs < lhs ? rhs : lhs;
+} // end min()
+
+template<typename T, typename BinaryPredicate>
+__host__ __device__
+  T max THRUST_PREVENT_MACRO_SUBSTITUTION (const T &lhs, const T &rhs, BinaryPredicate comp)
+{
+  return comp(lhs,rhs) ? rhs : lhs;
+} // end max()
+
+template<typename T>
+__host__ __device__
+  T max THRUST_PREVENT_MACRO_SUBSTITUTION (const T &lhs, const T &rhs)
+{
+  return lhs < rhs ? rhs : lhs;
+} // end max()
+
+
+} // end thrust
+
diff --git a/compat/thrust/detail/mismatch.inl b/compat/thrust/detail/mismatch.inl
new file mode 100644
index 0000000000..37ac663bbe
--- /dev/null
+++ b/compat/thrust/detail/mismatch.inl
@@ -0,0 +1,93 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file mismatch.inl
+ *  \brief Inline file for mismatch.h
+ */
+
+
+#include <thrust/detail/config.h>
+#include <thrust/mismatch.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/system/detail/generic/select_system.h>
+#include <thrust/system/detail/generic/mismatch.h>
+#include <thrust/system/detail/adl/mismatch.h>
+
+namespace thrust
+{
+
+
+template<typename DerivedPolicy, typename InputIterator1, typename InputIterator2>
+thrust::pair<InputIterator1, InputIterator2> mismatch(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                                      InputIterator1 first1,
+                                                      InputIterator1 last1,
+                                                      InputIterator2 first2)
+{
+  using thrust::system::detail::generic::mismatch;
+  return mismatch(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first1, last1, first2);
+} // end mismatch()
+
+
+template<typename DerivedPolicy, typename InputIterator1, typename InputIterator2, typename BinaryPredicate>
+thrust::pair<InputIterator1, InputIterator2> mismatch(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                                      InputIterator1 first1,
+                                                      InputIterator1 last1,
+                                                      InputIterator2 first2,
+                                                      BinaryPredicate pred)
+{
+  using thrust::system::detail::generic::mismatch;
+  return mismatch(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first1, last1, first2, pred);
+} // end mismatch()
+
+
+template <typename InputIterator1, typename InputIterator2>
+thrust::pair<InputIterator1, InputIterator2> mismatch(InputIterator1 first1,
+                                                      InputIterator1 last1,
+                                                      InputIterator2 first2)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator1>::type System1;
+  typedef typename thrust::iterator_system<InputIterator2>::type System2;
+
+  System1 system1;
+  System2 system2;
+
+  return thrust::mismatch(select_system(system1,system2), first1, last1, first2);
+} // end mismatch()
+
+
+template <typename InputIterator1, typename InputIterator2, typename BinaryPredicate>
+thrust::pair<InputIterator1, InputIterator2> mismatch(InputIterator1 first1,
+                                                      InputIterator1 last1,
+                                                      InputIterator2 first2,
+                                                      BinaryPredicate pred)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator1>::type System1;
+  typedef typename thrust::iterator_system<InputIterator2>::type System2;
+
+  System1 system1;
+  System2 system2;
+
+  return thrust::mismatch(select_system(system1,system2), first1, last1, first2, pred);
+} // end mismatch()
+
+
+} // end namespace thrust
+
diff --git a/compat/thrust/detail/mpl/math.h b/compat/thrust/detail/mpl/math.h
new file mode 100644
index 0000000000..80adfc1e88
--- /dev/null
+++ b/compat/thrust/detail/mpl/math.h
@@ -0,0 +1,174 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file math.h
+ *  \brief Math-related metaprogramming functionality.
+ */
+
+
+#pragma once
+
+namespace thrust
+{
+
+namespace detail
+{
+
+namespace mpl
+{
+
+namespace math
+{
+
+namespace detail
+{
+
+// compute the log base-2 of an integer at compile time
+template <unsigned int N, unsigned int Cur>
+struct log2
+{
+    static const unsigned int value = log2<N / 2,Cur+1>::value;
+};
+
+template <unsigned int Cur>
+struct log2<1, Cur>
+{
+    static const unsigned int value = Cur;
+};
+
+template <unsigned int Cur>
+struct log2<0, Cur>
+{
+    // undefined
+};
+
+} // end namespace detail
+
+
+template <unsigned int N>
+struct log2
+{
+    static const unsigned int value = detail::log2<N,0>::value;
+};
+
+
+template <typename T, T lhs, T rhs>
+struct min
+{
+  static const T value = (lhs < rhs) ? lhs : rhs;
+};
+
+
+template <typename T, T lhs, T rhs>
+struct max
+{
+  static const T value = (!(lhs < rhs)) ? lhs : rhs;
+};
+
+
+template<typename result_type, result_type x, result_type y>
+  struct mul
+{
+  static const result_type value = x * y;
+};
+
+
+template<typename result_type, result_type x, result_type y>
+  struct mod
+{
+  static const result_type value = x % y;
+};
+
+
+template<typename result_type, result_type x, result_type y>
+  struct div
+{
+  static const result_type value = x / y;
+};
+
+
+template<typename result_type, result_type x, result_type y>
+  struct geq
+{
+  static const bool value = x >= y;
+};
+
+
+template<typename result_type, result_type x, result_type y>
+  struct lt
+{
+  static const bool value = x < y;
+};
+
+
+template<typename result_type, result_type x, result_type y>
+  struct gt
+{
+  static const bool value = x > y;
+};
+
+
+template<bool x, bool y>
+  struct or_
+{
+  static const bool value = (x || y);
+};
+
+
+template<typename result_type, result_type x, result_type y>
+  struct bit_and
+{
+  static const result_type value = x & y;
+};
+
+
+template<typename result_type, result_type x, result_type y>
+  struct plus
+{
+  static const result_type value = x + y;
+};
+
+
+template<typename result_type, result_type x, result_type y>
+  struct minus
+{
+  static const result_type value = x - y;
+};
+
+
+template<typename result_type, result_type x, result_type y>
+  struct equal
+{
+  static const bool value = x == y;
+};
+
+
+template<typename result_type, result_type x>
+  struct is_odd
+{
+  static const bool value = x & 1;
+};
+
+
+} // end namespace math
+
+} // end namespace mpl
+
+} // end namespace detail
+
+} // end namespace thrust
+
diff --git a/compat/thrust/detail/numeric_traits.h b/compat/thrust/detail/numeric_traits.h
new file mode 100644
index 0000000000..a3bc56c211
--- /dev/null
+++ b/compat/thrust/detail/numeric_traits.h
@@ -0,0 +1,130 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/type_traits.h>
+#include <limits>
+
+//#include <stdint.h> // for intmax_t (not provided on MSVS 2005)
+
+namespace thrust
+{
+
+namespace detail
+{
+
+// XXX good enough for the platforms we care about
+typedef long long intmax_t;
+
+template<typename Number>
+  struct is_signed
+    : integral_constant<bool, std::numeric_limits<Number>::is_signed>
+{}; // end is_signed
+
+
+template<typename T>
+  struct num_digits
+    : eval_if<
+        std::numeric_limits<T>::is_specialized,
+        integral_constant<
+          int,
+          std::numeric_limits<T>::digits
+        >,
+        integral_constant<
+          int,
+          sizeof(T) * std::numeric_limits<unsigned char>::digits - (is_signed<T>::value ? 1 : 0)  
+        >
+      >::type
+{}; // end num_digits
+
+
+template<typename Integer>
+  struct integer_difference
+    //: eval_if<
+    //    sizeof(Integer) >= sizeof(intmax_t),
+    //    eval_if<
+    //      is_signed<Integer>::value,
+    //      identity_<Integer>,
+    //      identity_<intmax_t>
+    //    >,
+    //    eval_if<
+    //      sizeof(Integer) < sizeof(std::ptrdiff_t),
+    //      identity_<std::ptrdiff_t>,
+    //      identity_<intmax_t>
+    //    >
+    //  >
+{
+  private:
+    // XXX workaround a pedantic warning in old versions of g++
+    //     which complains about &&ing with a constant value
+    template<bool x, bool y>
+      struct and_
+    {
+      static const bool value = false;
+    };
+
+    template<bool y>
+      struct and_<true,y>
+    {
+      static const bool value = y;
+    };
+
+  public:
+    typedef typename
+      eval_if<
+        and_<
+          std::numeric_limits<Integer>::is_signed,
+          // digits is the number of no-sign bits
+          (!std::numeric_limits<Integer>::is_bounded || (int(std::numeric_limits<Integer>::digits) + 1 >= num_digits<intmax_t>::value))
+        >::value,
+        identity_<Integer>,
+        eval_if<
+          int(std::numeric_limits<Integer>::digits) + 1 < num_digits<signed int>::value,
+          identity_<signed int>,
+          eval_if<
+            int(std::numeric_limits<Integer>::digits) + 1 < num_digits<signed long>::value,
+            identity_<signed long>,
+            identity_<intmax_t>
+          >
+        >
+      >::type type;
+}; // end integer_difference
+
+
+template<typename Number>
+  struct numeric_difference
+    : eval_if<
+      is_integral<Number>::value,
+      integer_difference<Number>,
+      identity_<Number>
+    >
+{}; // end numeric_difference
+
+
+template<typename Number>
+__host__ __device__
+typename numeric_difference<Number>::type
+numeric_distance(Number x, Number y)
+{
+  typedef typename numeric_difference<Number>::type difference_type;
+  return difference_type(y) - difference_type(x);
+} // end numeric_distance
+
+} // end detail
+
+} // end thrust
+
diff --git a/compat/thrust/detail/overlapped_copy.h b/compat/thrust/detail/overlapped_copy.h
new file mode 100644
index 0000000000..a5540b8643
--- /dev/null
+++ b/compat/thrust/detail/overlapped_copy.h
@@ -0,0 +1,131 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/iterator/detail/minimum_system.h>
+#include <thrust/detail/copy.h>
+#include <thrust/detail/temporary_array.h>
+#include <thrust/system/cpp/detail/execution_policy.h>
+
+namespace thrust
+{
+namespace detail
+{
+
+
+template<typename InputIterator,
+         typename OutputIterator>
+  OutputIterator sequential_copy(InputIterator first,
+                                 InputIterator last,
+                                 OutputIterator result)
+{
+  for(; first != last; ++first, ++result)
+  {
+    *result = *first;
+  } // end for
+
+  return result;
+} // end sequential_copy()
+
+
+template<typename BidirectionalIterator1,
+         typename BidirectionalIterator2>
+  BidirectionalIterator2 sequential_copy_backward(BidirectionalIterator1 first,
+                                                  BidirectionalIterator1 last,
+                                                  BidirectionalIterator2 result)
+{
+  // yes, we preincrement
+  // the ranges are open on the right, i.e. [first, last)
+  while(first != last)
+  {
+    *--result = *--last;
+  } // end while
+
+  return result;
+} // end sequential_copy_backward()
+
+
+namespace dispatch
+{
+
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2>
+  RandomAccessIterator2 overlapped_copy(thrust::system::cpp::detail::execution_policy<DerivedPolicy> &,
+                                        RandomAccessIterator1 first,
+                                        RandomAccessIterator1 last,
+                                        RandomAccessIterator2 result)
+{
+  if(first < last && first <= result && result < last)
+  {
+    // result lies in [first, last)
+    // it's safe to use std::copy_backward here
+    thrust::detail::sequential_copy_backward(first, last, result + (last - first));
+    result += (last - first);
+  } // end if
+  else
+  {
+    // result + (last - first) lies in [first, last)
+    // it's safe to use sequential_copy here
+    result = thrust::detail::sequential_copy(first, last, result);
+  } // end else
+
+  return result;
+} // end overlapped_copy()
+
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2>
+  RandomAccessIterator2 overlapped_copy(thrust::execution_policy<DerivedPolicy> &exec,
+                                        RandomAccessIterator1 first,
+                                        RandomAccessIterator1 last,
+                                        RandomAccessIterator2 result)
+{
+  typedef typename thrust::iterator_value<RandomAccessIterator1>::type value_type;
+
+  // make a temporary copy of [first,last), and copy into it first
+  thrust::detail::temporary_array<value_type, DerivedPolicy> temp(exec, first, last);
+  return thrust::copy(exec, temp.begin(), temp.end(), result);
+} // end overlapped_copy()
+
+} // end dispatch
+
+
+template<typename RandomAccessIterator1,
+         typename RandomAccessIterator2>
+  RandomAccessIterator2 overlapped_copy(RandomAccessIterator1 first,
+                                        RandomAccessIterator1 last,
+                                        RandomAccessIterator2 result)
+{
+  typedef typename thrust::iterator_system<RandomAccessIterator2>::type System1;
+  typedef typename thrust::iterator_system<RandomAccessIterator2>::type System2;
+
+  typedef typename thrust::detail::minimum_system<System1, System2>::type System;
+
+  // XXX presumes System is default constructible
+  System system;
+
+  return thrust::detail::dispatch::overlapped_copy(system, first, last, result);
+} // end overlapped_copy()
+
+} // end detail
+} // end thrust
+
diff --git a/compat/thrust/detail/pair.inl b/compat/thrust/detail/pair.inl
new file mode 100644
index 0000000000..776bdc2315
--- /dev/null
+++ b/compat/thrust/detail/pair.inl
@@ -0,0 +1,225 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/pair.h>
+#include <thrust/detail/swap.h>
+
+namespace thrust
+{
+
+template <typename T1, typename T2>
+  pair<T1,T2>
+    ::pair(void)
+      :first(),second()
+{
+  ;
+} // end pair::pair()
+
+
+template <typename T1, typename T2>
+  pair<T1,T2>
+    ::pair(const T1 &x, const T2 &y)
+      :first(x),second(y)
+{
+  ;
+} // end pair::pair()
+
+
+template <typename T1, typename T2>
+  template <typename U1, typename U2>
+    pair<T1,T2>
+      ::pair(const pair<U1,U2> &p)
+        :first(p.first),second(p.second)
+{
+  ;
+} // end pair::pair()
+
+
+template <typename T1, typename T2>
+  template <typename U1, typename U2>
+    pair<T1,T2>
+      ::pair(const std::pair<U1,U2> &p)
+        :first(p.first),second(p.second)
+{
+  ;
+} // end pair::pair()
+
+
+template<typename T1, typename T2>
+  inline __host__ __device__
+    void pair<T1,T2>
+      ::swap(thrust::pair<T1,T2> &p)
+{
+  using thrust::swap;
+
+  swap(first, p.first);
+  swap(second, p.second);
+} // end pair::swap()
+
+
+template <typename T1, typename T2>
+  inline __host__ __device__
+    bool operator==(const pair<T1,T2> &x, const pair<T1,T2> &y)
+{
+  return x.first == y.first && x.second == y.second;
+} // end operator==()
+
+
+template <typename T1, typename T2>
+  inline __host__ __device__
+    bool operator<(const pair<T1,T2> &x, const pair<T1,T2> &y)
+{
+  return x.first < y.first || (!(y.first < x.first) && x.second < y.second);
+} // end operator<()
+
+
+template <typename T1, typename T2>
+  inline __host__ __device__
+    bool operator!=(const pair<T1,T2> &x, const pair<T1,T2> &y)
+{
+  return !(x == y);
+} // end operator==()
+
+
+template <typename T1, typename T2>
+  inline __host__ __device__
+    bool operator>(const pair<T1,T2> &x, const pair<T1,T2> &y)
+{
+  return y < x;
+} // end operator<()
+
+
+template <typename T1, typename T2>
+  inline __host__ __device__
+    bool operator<=(const pair<T1,T2> &x, const pair<T1,T2> &y)
+{
+  return !(y < x);
+} // end operator<=()
+
+
+template <typename T1, typename T2>
+  inline __host__ __device__
+    bool operator>=(const pair<T1,T2> &x, const pair<T1,T2> &y)
+{
+  return !(x < y);
+} // end operator>=()
+
+
+template <typename T1, typename T2>
+  inline __host__ __device__
+    void swap(pair<T1,T2> &x, pair<T1,T2> &y)
+{
+  return x.swap(y);
+} // end swap()
+
+
+template <typename T1, typename T2>
+  inline __host__ __device__
+    pair<T1,T2> make_pair(T1 x, T2 y)
+{
+  return pair<T1,T2>(x,y);
+} // end make_pair()
+
+
+// specializations of tuple_element for pair
+template<typename T1, typename T2>
+  struct tuple_element<0, pair<T1,T2> >
+{
+  typedef T1 type;
+}; // end tuple_element
+
+template<typename T1, typename T2>
+  struct tuple_element<1, pair<T1,T2> >
+{
+  typedef T2 type;
+}; // end tuple_element
+
+
+// specialization of tuple_size for pair
+template<typename T1, typename T2>
+  struct tuple_size< pair<T1,T2 > >
+{
+  static const unsigned int value = 2;
+}; // end tuple_size
+
+
+
+namespace detail
+{
+
+
+template<int N, typename Pair> struct pair_get {};
+
+template<typename Pair>
+  struct pair_get<0, Pair>
+{
+  inline __host__ __device__
+    const typename tuple_element<0, Pair>::type &
+      operator()(const Pair &p) const
+  {
+    return p.first;
+  } // end operator()()
+
+  inline __host__ __device__
+    typename tuple_element<0, Pair>::type &
+      operator()(Pair &p) const
+  {
+    return p.first;
+  } // end operator()()
+}; // end pair_get
+
+
+template<typename Pair>
+  struct pair_get<1, Pair>
+{
+  inline __host__ __device__
+    const typename tuple_element<1, Pair>::type &
+      operator()(const Pair &p) const
+  {
+    return p.second;
+  } // end operator()()
+
+  inline __host__ __device__
+    typename tuple_element<1, Pair>::type &
+      operator()(Pair &p) const
+  {
+    return p.second;
+  } // end operator()()
+}; // end pair_get
+
+} // end detail
+
+
+
+template<unsigned int N, typename T1, typename T2>
+  inline __host__ __device__
+    typename tuple_element<N, pair<T1,T2> >::type &
+      get(pair<T1,T2> &p)
+{
+  return detail::pair_get<N, pair<T1,T2> >()(p);
+} // end get()
+
+template<unsigned int N, typename T1, typename T2>
+  inline __host__ __device__
+    const typename tuple_element<N, pair<T1,T2> >::type &
+      get(const pair<T1,T2> &p)
+{
+  return detail::pair_get<N, pair<T1,T2> >()(p);
+} // end get()
+
+
+} // end thrust
+
diff --git a/compat/thrust/detail/partition.inl b/compat/thrust/detail/partition.inl
new file mode 100644
index 0000000000..19ef08a73c
--- /dev/null
+++ b/compat/thrust/detail/partition.inl
@@ -0,0 +1,398 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file partition.inl
+ *  \brief Inline file for partition.h.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/partition.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/system/detail/generic/select_system.h>
+#include <thrust/system/detail/generic/partition.h>
+#include <thrust/system/detail/adl/partition.h>
+
+namespace thrust
+{
+
+
+template<typename DerivedPolicy,
+         typename ForwardIterator,
+         typename Predicate>
+  ForwardIterator partition(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                            ForwardIterator first,
+                            ForwardIterator last,
+                            Predicate pred)
+{
+  using thrust::system::detail::generic::partition;
+  return partition(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, pred);
+} // end partition()
+
+
+template<typename DerivedPolicy,
+         typename ForwardIterator,
+         typename InputIterator,
+         typename Predicate>
+  ForwardIterator partition(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                            ForwardIterator first,
+                            ForwardIterator last,
+                            InputIterator stencil,
+                            Predicate pred)
+{
+  using thrust::system::detail::generic::partition;
+  return partition(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, stencil, pred);
+} // end partition()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename Predicate>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    partition_copy(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                   InputIterator first,
+                   InputIterator last,
+                   OutputIterator1 out_true,
+                   OutputIterator2 out_false,
+                   Predicate pred)
+{
+  using thrust::system::detail::generic::partition_copy;
+  return partition_copy(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, out_true, out_false, pred);
+} // end partition_copy()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename Predicate>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    partition_copy(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                   InputIterator1 first,
+                   InputIterator1 last,
+                   InputIterator2 stencil,
+                   OutputIterator1 out_true,
+                   OutputIterator2 out_false,
+                   Predicate pred)
+{
+  using thrust::system::detail::generic::partition_copy;
+  return partition_copy(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, stencil, out_true, out_false, pred);
+} // end partition_copy()
+
+
+template<typename DerivedPolicy,
+         typename ForwardIterator,
+         typename Predicate>
+  ForwardIterator stable_partition(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                   ForwardIterator first,
+                                   ForwardIterator last,
+                                   Predicate pred)
+{
+  using thrust::system::detail::generic::stable_partition;
+  return stable_partition(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, pred);
+} // end stable_partition()
+
+
+template<typename DerivedPolicy,
+         typename ForwardIterator,
+         typename InputIterator,
+         typename Predicate>
+  ForwardIterator stable_partition(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                   ForwardIterator first,
+                                   ForwardIterator last,
+                                   InputIterator stencil,
+                                   Predicate pred)
+{
+  using thrust::system::detail::generic::stable_partition;
+  return stable_partition(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, stencil, pred);
+} // end stable_partition()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename Predicate>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    stable_partition_copy(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                          InputIterator first,
+                          InputIterator last,
+                          OutputIterator1 out_true,
+                          OutputIterator2 out_false,
+                          Predicate pred)
+{
+  using thrust::system::detail::generic::stable_partition_copy;
+  return stable_partition_copy(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, out_true, out_false, pred);
+} // end stable_partition_copy()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename Predicate>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    stable_partition_copy(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                          InputIterator1 first,
+                          InputIterator1 last,
+                          InputIterator2 stencil,
+                          OutputIterator1 out_true,
+                          OutputIterator2 out_false,
+                          Predicate pred)
+{
+  using thrust::system::detail::generic::stable_partition_copy;
+  return stable_partition_copy(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, stencil, out_true, out_false, pred);
+} // end stable_partition_copy()
+
+
+template<typename DerivedPolicy, typename ForwardIterator, typename Predicate>
+  ForwardIterator partition_point(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                  ForwardIterator first,
+                                  ForwardIterator last,
+                                  Predicate pred)
+{
+  using thrust::system::detail::generic::partition_point;
+  return partition_point(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, pred);
+} // end partition_point()
+
+
+template<typename DerivedPolicy, typename InputIterator, typename Predicate>
+  bool is_partitioned(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                      InputIterator first,
+                      InputIterator last,
+                      Predicate pred)
+{
+  using thrust::system::detail::generic::is_partitioned;
+  return is_partitioned(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, pred);
+} // end is_partitioned()
+
+
+template<typename ForwardIterator,
+         typename Predicate>
+  ForwardIterator partition(ForwardIterator first,
+                            ForwardIterator last,
+                            Predicate pred)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<ForwardIterator>::type System;
+
+  System system;
+
+  return thrust::partition(select_system(system), first, last, pred);
+} // end partition()
+
+
+template<typename ForwardIterator,
+         typename InputIterator,
+         typename Predicate>
+  ForwardIterator partition(ForwardIterator first,
+                            ForwardIterator last,
+                            InputIterator stencil,
+                            Predicate pred)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<ForwardIterator>::type System1;
+  typedef typename thrust::iterator_system<InputIterator>::type System2;
+
+  System1 system1;
+  System2 system2;
+
+  return thrust::partition(select_system(system1,system2), first, last, stencil, pred);
+} // end partition()
+
+
+template<typename ForwardIterator,
+         typename Predicate>
+  ForwardIterator stable_partition(ForwardIterator first,
+                                   ForwardIterator last,
+                                   Predicate pred)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<ForwardIterator>::type System;
+
+  System system;
+
+  return thrust::stable_partition(select_system(system), first, last, pred);
+} // end stable_partition()
+
+
+template<typename ForwardIterator,
+         typename InputIterator,
+         typename Predicate>
+  ForwardIterator stable_partition(ForwardIterator first,
+                                   ForwardIterator last,
+                                   InputIterator stencil,
+                                   Predicate pred)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<ForwardIterator>::type System1;
+  typedef typename thrust::iterator_system<InputIterator>::type   System2;
+
+  System1 system1;
+  System2 system2;
+
+  return thrust::stable_partition(select_system(system1,system2), first, last, stencil, pred);
+} // end stable_partition()
+
+
+template<typename InputIterator,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename Predicate>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    partition_copy(InputIterator first,
+                   InputIterator last,
+                   OutputIterator1 out_true,
+                   OutputIterator2 out_false,
+                   Predicate pred)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator>::type   System1;
+  typedef typename thrust::iterator_system<OutputIterator1>::type System2;
+  typedef typename thrust::iterator_system<OutputIterator2>::type System3;
+
+  System1 system1;
+  System2 system2;
+  System3 system3;
+
+  return thrust::partition_copy(select_system(system1,system2,system3), first, last, out_true, out_false, pred);
+} // end partition_copy()
+
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename Predicate>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    partition_copy(InputIterator1 first,
+                   InputIterator1 last,
+                   InputIterator2 stencil,
+                   OutputIterator1 out_true,
+                   OutputIterator2 out_false,
+                   Predicate pred)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator1>::type  System1;
+  typedef typename thrust::iterator_system<InputIterator1>::type  System2;
+  typedef typename thrust::iterator_system<OutputIterator1>::type System3;
+  typedef typename thrust::iterator_system<OutputIterator2>::type System4;
+
+  System1 system1;
+  System2 system2;
+  System3 system3;
+  System4 system4;
+
+  return thrust::partition_copy(select_system(system1,system2,system3,system4), first, last, stencil, out_true, out_false, pred);
+} // end partition_copy()
+
+
+template<typename InputIterator,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename Predicate>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    stable_partition_copy(InputIterator first,
+                          InputIterator last,
+                          OutputIterator1 out_true,
+                          OutputIterator2 out_false,
+                          Predicate pred)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator>::type   System1;
+  typedef typename thrust::iterator_system<OutputIterator1>::type System2;
+  typedef typename thrust::iterator_system<OutputIterator2>::type System3;
+
+  System1 system1;
+  System2 system2;
+  System3 system3;
+
+  return thrust::stable_partition_copy(select_system(system1,system2,system3), first, last, out_true, out_false, pred);
+} // end stable_partition_copy()
+
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename Predicate>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    stable_partition_copy(InputIterator1 first,
+                          InputIterator1 last,
+                          InputIterator2 stencil,
+                          OutputIterator1 out_true,
+                          OutputIterator2 out_false,
+                          Predicate pred)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator1>::type   System1;
+  typedef typename thrust::iterator_system<InputIterator2>::type   System2;
+  typedef typename thrust::iterator_system<OutputIterator1>::type  System3;
+  typedef typename thrust::iterator_system<OutputIterator2>::type  System4;
+
+  System1 system1;
+  System2 system2;
+  System3 system3;
+  System4 system4;
+
+  return thrust::stable_partition_copy(select_system(system1,system2,system3,system4), first, last, stencil, out_true, out_false, pred);
+} // end stable_partition_copy()
+
+
+template<typename ForwardIterator, typename Predicate>
+  ForwardIterator partition_point(ForwardIterator first,
+                                  ForwardIterator last,
+                                  Predicate pred)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<ForwardIterator>::type System;
+
+  System system;
+
+  return thrust::partition_point(select_system(system), first, last, pred);
+} // end partition_point()
+
+
+template<typename InputIterator, typename Predicate>
+  bool is_partitioned(InputIterator first,
+                      InputIterator last,
+                      Predicate pred)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator>::type System;
+
+  System system;
+
+  return thrust::is_partitioned(select_system(system), first, last, pred);
+} // end is_partitioned()
+
+
+} // end thrust
+
diff --git a/compat/thrust/detail/pointer.h b/compat/thrust/detail/pointer.h
new file mode 100644
index 0000000000..bc97939c77
--- /dev/null
+++ b/compat/thrust/detail/pointer.h
@@ -0,0 +1,184 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/iterator/iterator_adaptor.h>
+#include <thrust/iterator/detail/iterator_traversal_tags.h>
+#include <thrust/detail/type_traits/pointer_traits.h>
+#include <thrust/detail/type_traits.h>
+#include <thrust/detail/reference_forward_declaration.h>
+
+namespace thrust
+{
+
+// declare pointer with default values of template parameters
+template<typename Element, typename Tag, typename Reference = use_default, typename Derived = use_default> class pointer;
+
+} // end thrust
+
+
+// specialize std::iterator_traits to avoid problems with the name of
+// pointer's constructor shadowing its nested pointer type
+// do this before pointer is defined so the specialization is correctly
+// used inside the definition
+namespace std
+{
+
+template<typename Element, typename Tag, typename Reference, typename Derived>
+  struct iterator_traits<thrust::pointer<Element,Tag,Reference,Derived> >
+{
+  private:
+    typedef thrust::pointer<Element,Tag,Reference,Derived> ptr;
+
+  public:
+    typedef typename ptr::iterator_category iterator_category;
+    typedef typename ptr::value_type        value_type;
+    typedef typename ptr::difference_type   difference_type;
+    // XXX implement this type (the result of operator->) later
+    typedef void                             pointer;
+    typedef typename ptr::reference         reference;
+}; // end iterator_traits
+
+} // end std
+
+
+namespace thrust
+{
+
+namespace detail
+{
+
+// this metafunction computes the type of iterator_adaptor thrust::pointer should inherit from
+template<typename Element, typename Tag, typename Reference, typename Derived>
+  struct pointer_base
+{
+  // void pointers should have no element type
+  // note that we remove_cv from the Element type to get the value_type
+  typedef typename thrust::detail::eval_if<
+    thrust::detail::is_void<typename thrust::detail::remove_const<Element>::type>::value,
+    thrust::detail::identity_<void>,
+    thrust::detail::remove_cv<Element>
+  >::type value_type;
+
+  // if no Derived type is given, just use pointer
+  typedef typename thrust::detail::eval_if<
+    thrust::detail::is_same<Derived,use_default>::value,
+    thrust::detail::identity_<pointer<Element,Tag,Reference,Derived> >,
+    thrust::detail::identity_<Derived>
+  >::type derived_type;
+
+  // void pointers should have no reference type
+  // if no Reference type is given, just use reference
+  typedef typename thrust::detail::eval_if<
+    thrust::detail::is_void<typename thrust::detail::remove_const<Element>::type>::value,
+    thrust::detail::identity_<void>,
+    thrust::detail::eval_if<
+      thrust::detail::is_same<Reference,use_default>::value,
+      thrust::detail::identity_<reference<Element,derived_type> >,
+      thrust::detail::identity_<Reference>
+    >
+  >::type reference_arg;
+
+  typedef thrust::iterator_adaptor<
+    derived_type,                        // pass along the type of our Derived class to iterator_adaptor
+    Element *,                           // we adapt a raw pointer
+    value_type,                          // the value type
+    Tag,                                 // system tag
+    thrust::random_access_traversal_tag, // pointers have random access traversal
+    reference_arg,                       // pass along our Reference type
+    std::ptrdiff_t
+  > type;
+}; // end pointer_base
+
+
+} // end detail
+
+
+// the base type for all of thrust's tagged pointers.
+// for reasonable pointer-like semantics, derived types should reimplement the following:
+// 1. no-argument constructor
+// 2. constructor from OtherElement *
+// 3. constructor from OtherPointer related by convertibility
+// 4. assignment from OtherPointer related by convertibility
+// These should just call the corresponding members of pointer.
+template<typename Element, typename Tag, typename Reference, typename Derived>
+  class pointer
+    : public thrust::detail::pointer_base<Element,Tag,Reference,Derived>::type
+{
+  private:
+    typedef typename thrust::detail::pointer_base<Element,Tag,Reference,Derived>::type         super_t;
+
+    typedef typename thrust::detail::pointer_base<Element,Tag,Reference,Derived>::derived_type derived_type;
+
+    // friend iterator_core_access to give it access to dereference
+    friend class thrust::iterator_core_access;
+
+    __host__ __device__
+    typename super_t::reference dereference() const;
+
+    // don't provide access to this part of super_t's interface
+    using super_t::base;
+    using typename super_t::base_type;
+
+  public:
+    typedef typename super_t::base_type raw_pointer;
+
+    // constructors
+    
+    __host__ __device__
+    pointer();
+
+    // OtherValue shall be convertible to Value
+    // XXX consider making the pointer implementation a template parameter which defaults to Element *
+    template<typename OtherElement>
+    __host__ __device__
+    explicit pointer(OtherElement *ptr);
+
+    // OtherPointer's element_type shall be convertible to Element
+    // OtherPointer's system shall be convertible to Tag
+    template<typename OtherPointer>
+    __host__ __device__
+    pointer(const OtherPointer &other,
+            typename thrust::detail::enable_if_pointer_is_convertible<
+              OtherPointer,
+              pointer<Element,Tag,Reference,Derived>
+            >::type * = 0);
+
+    // assignment
+    
+    // OtherPointer's element_type shall be convertible to Element
+    // OtherPointer's system shall be convertible to Tag
+    template<typename OtherPointer>
+    __host__ __device__
+    typename thrust::detail::enable_if_pointer_is_convertible<
+      OtherPointer,
+      pointer,
+      derived_type &
+    >::type
+    operator=(const OtherPointer &other);
+
+    // observers
+
+    __host__ __device__
+    Element *get() const;
+}; // end pointer
+
+} // end thrust
+
+#include <thrust/detail/pointer.inl>
+
diff --git a/compat/thrust/detail/pointer.inl b/compat/thrust/detail/pointer.inl
new file mode 100644
index 0000000000..1d066b041c
--- /dev/null
+++ b/compat/thrust/detail/pointer.inl
@@ -0,0 +1,143 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/pointer.h>
+
+
+namespace thrust
+{
+
+
+template<typename Element, typename Tag, typename Reference, typename Derived>
+  pointer<Element,Tag,Reference,Derived>
+    ::pointer()
+      : super_t(static_cast<Element*>(0))
+{} // end pointer::pointer
+
+
+template<typename Element, typename Tag, typename Reference, typename Derived>
+  template<typename OtherElement>
+    pointer<Element,Tag,Reference,Derived>
+      ::pointer(OtherElement *other)
+        : super_t(other)
+{} // end pointer::pointer
+
+
+template<typename Element, typename Tag, typename Reference, typename Derived>
+  template<typename OtherPointer>
+    pointer<Element,Tag,Reference,Derived>
+      ::pointer(const OtherPointer &other,
+                typename thrust::detail::enable_if_pointer_is_convertible<
+                  OtherPointer,
+                  pointer<Element,Tag,Reference,Derived>
+                 >::type *)
+        : super_t(thrust::detail::pointer_traits<OtherPointer>::get(other))
+{} // end pointer::pointer
+
+
+template<typename Element, typename Tag, typename Reference, typename Derived>
+  template<typename OtherPointer>
+    typename thrust::detail::enable_if_pointer_is_convertible<
+      OtherPointer,
+      pointer<Element,Tag,Reference,Derived>,
+      typename pointer<Element,Tag,Reference,Derived>::derived_type &
+    >::type
+      pointer<Element,Tag,Reference,Derived>
+        ::operator=(const OtherPointer &other)
+{
+  super_t::base_reference() = thrust::detail::pointer_traits<OtherPointer>::get(other);
+  return static_cast<derived_type&>(*this);
+} // end pointer::operator=
+
+
+template<typename Element, typename Tag, typename Reference, typename Derived>
+  typename pointer<Element,Tag,Reference,Derived>::super_t::reference
+    pointer<Element,Tag,Reference,Derived>
+      ::dereference() const
+{
+  return typename super_t::reference(static_cast<const derived_type&>(*this));
+} // end pointer::dereference
+
+
+template<typename Element, typename Tag, typename Reference, typename Derived>
+  Element *pointer<Element,Tag,Reference,Derived>
+    ::get() const
+{
+  return super_t::base();
+} // end pointer::get
+
+
+namespace detail
+{
+
+#if (THRUST_HOST_COMPILER == THRUST_HOST_COMPILER_MSVC) && (_MSC_VER <= 1400)
+// XXX WAR MSVC 2005 problem with correctly implementing
+//     pointer_raw_pointer for pointer by specializing it here
+template<typename Element, typename Tag, typename Reference, typename Derived>
+  struct pointer_raw_pointer< thrust::pointer<Element,Tag,Reference,Derived> >
+{
+  typedef typename pointer<Element,Tag,Reference,Derived>::raw_pointer type;
+}; // end pointer_raw_pointer
+#endif
+
+
+#if (THRUST_HOST_COMPILER == THRUST_HOST_COMPILER_GCC) && (THRUST_GCC_VERSION < 40200)
+// XXX WAR g++-4.1 problem with correctly implementing
+//     pointer_element for pointer by specializing it here
+template<typename Element, typename Tag>
+  struct pointer_element< thrust::pointer<Element,Tag> >
+{
+  typedef Element type;
+}; // end pointer_element
+
+template<typename Element, typename Tag, typename Reference>
+  struct pointer_element< thrust::pointer<Element,Tag,Reference> >
+    : pointer_element< thrust::pointer<Element,Tag> >
+{}; // end pointer_element
+
+template<typename Element, typename Tag, typename Reference, typename Derived>
+  struct pointer_element< thrust::pointer<Element,Tag,Reference,Derived> >
+    : pointer_element< thrust::pointer<Element,Tag,Reference> >
+{}; // end pointer_element
+
+
+
+// XXX WAR g++-4.1 problem with correctly implementing
+//     rebind_pointer for pointer by specializing it here
+template<typename Element, typename Tag, typename NewElement>
+  struct rebind_pointer<thrust::pointer<Element,Tag>, NewElement>
+{
+  // XXX note we don't attempt to rebind the pointer's Reference type (or Derived)
+  typedef thrust::pointer<NewElement,Tag> type;
+};
+
+template<typename Element, typename Tag, typename Reference, typename NewElement>
+  struct rebind_pointer<thrust::pointer<Element,Tag,Reference>, NewElement>
+    : rebind_pointer<thrust::pointer<Element,Tag>, NewElement>
+{};
+
+template<typename Element, typename Tag, typename Reference, typename Derived, typename NewElement>
+  struct rebind_pointer<thrust::pointer<Element,Tag,Reference,Derived>, NewElement>
+    : rebind_pointer<thrust::pointer<Element,Tag,Reference>, NewElement>
+{};
+#endif
+
+} // end namespace detail
+
+
+} // end thrust
+
diff --git a/compat/thrust/detail/range/tail_flags.h b/compat/thrust/detail/range/tail_flags.h
new file mode 100644
index 0000000000..06fd9f8a00
--- /dev/null
+++ b/compat/thrust/detail/range/tail_flags.h
@@ -0,0 +1,124 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/iterator/transform_iterator.h>
+#include <thrust/iterator/zip_iterator.h>
+#include <thrust/iterator/counting_iterator.h>
+#include <thrust/tuple.h>
+#include <thrust/functional.h>
+
+namespace thrust
+{
+namespace detail
+{
+
+
+template<typename RandomAccessIterator,
+         typename BinaryPredicate = thrust::equal_to<typename thrust::iterator_value<RandomAccessIterator>::type>,
+         typename IndexType = typename thrust::iterator_difference<RandomAccessIterator>::type>
+  class tail_flags
+{
+  private:
+    struct tail_flag_functor
+    {
+      BinaryPredicate binary_pred; // this must be the first member for performance reasons
+      IndexType n;
+
+      typedef bool result_type;
+
+      tail_flag_functor(IndexType n)
+        : binary_pred(), n(n)
+      {}
+
+      tail_flag_functor(IndexType n, BinaryPredicate binary_pred)
+        : binary_pred(binary_pred), n(n)
+      {}
+
+      template<typename Tuple>
+      __host__ __device__ __thrust_forceinline__
+      result_type operator()(const Tuple &t)
+      {
+        const IndexType i = thrust::get<0>(t);
+
+        // note that we do not dereference the tuple's 2nd element when i >= n
+        // and therefore do not dereference a bad location at the boundary
+        return (i == (n - 1) || !binary_pred(thrust::get<1>(t), thrust::get<2>(t)));
+      }
+    };
+
+    typedef thrust::counting_iterator<IndexType> counting_iterator;
+
+  public:
+    typedef thrust::transform_iterator<
+      tail_flag_functor,
+      thrust::zip_iterator<thrust::tuple<counting_iterator,RandomAccessIterator,RandomAccessIterator> >
+    > iterator;
+
+    tail_flags(RandomAccessIterator first, RandomAccessIterator last)
+      : m_begin(thrust::make_transform_iterator(thrust::make_zip_iterator(thrust::make_tuple(thrust::counting_iterator<IndexType>(0), first, first + 1)),
+                                                tail_flag_functor(last - first))),
+        m_end(m_begin + (last - first))
+    {}
+
+    tail_flags(RandomAccessIterator first, RandomAccessIterator last, BinaryPredicate binary_pred)
+      : m_begin(thrust::make_transform_iterator(thrust::make_zip_iterator(thrust::make_tuple(thrust::counting_iterator<IndexType>(0), first, first + 1)),
+                                                tail_flag_functor(last - first, binary_pred))),
+        m_end(m_begin + (last - first))
+    {}
+
+    iterator begin() const
+    {
+      return m_begin;
+    }
+
+    iterator end() const
+    {
+      return m_end;
+    }
+
+    template<typename OtherIndex>
+    typename iterator::reference operator[](OtherIndex i)
+    {
+      return *(begin() + i);
+    }
+
+  private:
+    iterator m_begin, m_end;
+};
+
+
+template<typename RandomAccessIterator, typename BinaryPredicate>
+  tail_flags<RandomAccessIterator, BinaryPredicate>
+    make_tail_flags(RandomAccessIterator first, RandomAccessIterator last, BinaryPredicate binary_pred)
+{
+  return tail_flags<RandomAccessIterator, BinaryPredicate>(first, last, binary_pred);
+}
+
+
+template<typename RandomAccessIterator>
+  tail_flags<RandomAccessIterator>
+    make_tail_flags(RandomAccessIterator first, RandomAccessIterator last)
+{
+  return tail_flags<RandomAccessIterator>(first, last);
+}
+
+
+} // end detail
+} // end thrust
+
diff --git a/compat/thrust/detail/raw_pointer_cast.h b/compat/thrust/detail/raw_pointer_cast.h
new file mode 100644
index 0000000000..05e1e6bc45
--- /dev/null
+++ b/compat/thrust/detail/raw_pointer_cast.h
@@ -0,0 +1,33 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/type_traits/pointer_traits.h>
+
+namespace thrust
+{
+
+template<typename Pointer>
+  inline __host__ __device__ typename thrust::detail::pointer_traits<Pointer>::raw_pointer
+    raw_pointer_cast(const Pointer &ptr)
+{
+  return thrust::detail::pointer_traits<Pointer>::get(ptr);
+} // end raw_pointer_cast()
+
+} // end thrust
+
diff --git a/compat/thrust/detail/raw_reference_cast.h b/compat/thrust/detail/raw_reference_cast.h
new file mode 100644
index 0000000000..1ffd7e5701
--- /dev/null
+++ b/compat/thrust/detail/raw_reference_cast.h
@@ -0,0 +1,121 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/type_traits/has_nested_type.h>
+#include <thrust/detail/raw_pointer_cast.h>
+#include <thrust/detail/type_traits.h>
+#include <thrust/iterator/detail/tuple_of_iterator_references.h>
+
+namespace thrust
+{
+namespace detail
+{
+
+
+__THRUST_DEFINE_HAS_NESTED_TYPE(is_wrapped_reference, wrapped_reference_hint)
+
+namespace raw_reference_detail
+{
+
+template<typename T, typename Enable = void>
+  struct raw_reference
+    : add_reference<T>
+{};
+
+
+// XXX consider making raw_reference<T&> an error
+
+
+template<typename T>
+  struct raw_reference<
+    T,
+    typename thrust::detail::enable_if<
+      is_wrapped_reference<
+        typename remove_cv<T>::type
+      >::value
+    >::type
+  >
+{
+  typedef typename add_reference<
+    typename pointer_element<typename T::pointer>::type
+  >::type type;
+};
+
+} // end raw_reference_ns
+
+template<typename T>
+  struct raw_reference : 
+    raw_reference_detail::raw_reference<T>
+{};
+
+
+// wrapped reference-like things which aren't strictly wrapped references
+// (e.g. tuples of wrapped references) are considered unwrappable
+template<typename T>
+  struct is_unwrappable
+    : is_wrapped_reference<T>
+{};
+
+
+template<typename T, typename Result = void>
+  struct enable_if_unwrappable
+    : enable_if<
+        is_unwrappable<T>::value,
+        Result
+      >
+{};
+
+
+} // end detail
+
+
+template<typename T>
+  inline __host__ __device__ typename detail::raw_reference<T>::type raw_reference_cast(T &ref)
+{
+  return *thrust::raw_pointer_cast(&ref);
+} // end raw_reference_cast
+
+
+template<typename T>
+  inline __host__ __device__ typename detail::raw_reference<const T>::type raw_reference_cast(const T &ref)
+{
+  return *thrust::raw_pointer_cast(&ref);
+} // end raw_reference_cast
+
+
+template<
+  typename T0, typename T1, typename T2,
+  typename T3, typename T4, typename T5,
+  typename T6, typename T7, typename T8,
+  typename T9
+>
+inline __host__ __device__
+typename detail::enable_if_unwrappable<
+  thrust::detail::tuple_of_iterator_references<T0,T1,T2,T3,T4,T5,T6,T7,T8,T9>,
+  typename detail::raw_reference<
+    thrust::detail::tuple_of_iterator_references<T0,T1,T2,T3,T4,T5,T6,T7,T8,T9>
+  >::type
+>::type
+raw_reference_cast(detail::tuple_of_iterator_references<T0,T1,T2,T3,T4,T5,T6,T7,T8,T9> t);
+
+
+} // end thrust
+
+#include <thrust/detail/raw_reference_cast.inl>
+
diff --git a/compat/thrust/detail/raw_reference_cast.inl b/compat/thrust/detail/raw_reference_cast.inl
new file mode 100644
index 0000000000..ea619ec028
--- /dev/null
+++ b/compat/thrust/detail/raw_reference_cast.inl
@@ -0,0 +1,277 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/raw_reference_cast.h>
+#include <thrust/detail/type_traits.h>
+#include <thrust/detail/tuple_transform.h>
+
+namespace thrust
+{
+namespace detail
+{
+
+// specialize is_unwrappable
+// a tuple is_unwrappable if any of its elements is_unwrappable
+template<
+  typename T0, typename T1, typename T2,
+  typename T3, typename T4, typename T5,
+  typename T6, typename T7, typename T8,
+  typename T9
+>
+  struct is_unwrappable<
+    thrust::tuple<T0,T1,T2,T3,T4,T5,T6,T7,T8,T9>
+  >
+    : or_<
+        is_unwrappable<T0>,
+        is_unwrappable<T1>,
+        is_unwrappable<T2>,
+        is_unwrappable<T3>,
+        is_unwrappable<T4>,
+        is_unwrappable<T5>,
+        is_unwrappable<T6>,
+        is_unwrappable<T7>,
+        is_unwrappable<T8>,
+        is_unwrappable<T9>
+      >
+{};
+
+
+// specialize is_unwrappable
+// a tuple_of_iterator_references is_unwrappable if any of its elements is_unwrappable
+template<
+  typename T0, typename T1, typename T2,
+  typename T3, typename T4, typename T5,
+  typename T6, typename T7, typename T8,
+  typename T9
+>
+  struct is_unwrappable<
+    thrust::detail::tuple_of_iterator_references<T0,T1,T2,T3,T4,T5,T6,T7,T8,T9>
+  >
+    : or_<
+        is_unwrappable<T0>,
+        is_unwrappable<T1>,
+        is_unwrappable<T2>,
+        is_unwrappable<T3>,
+        is_unwrappable<T4>,
+        is_unwrappable<T5>,
+        is_unwrappable<T6>,
+        is_unwrappable<T7>,
+        is_unwrappable<T8>,
+        is_unwrappable<T9>
+      >
+{};
+
+
+namespace raw_reference_detail
+{
+
+// unlike raw_reference,
+// raw_reference_tuple_helper needs to return a value
+// when it encounters one, rather than a reference
+// upon encountering tuple, recurse
+//
+// we want the following behavior:
+//  1. T                                -> T
+//  2. T&                               -> T&
+//  3. null_type                        -> null_type
+//  4. reference<T>                     -> T&
+//  5. tuple_of_iterator_references<T>  -> tuple_of_iterator_references<raw_reference_tuple_helper<T>::type>
+
+
+// wrapped references are unwrapped using raw_reference, otherwise, return T
+template<typename T>
+  struct raw_reference_tuple_helper
+    : eval_if<
+        is_unwrappable<
+          typename remove_cv<T>::type
+        >::value,
+        raw_reference<T>,
+        identity_<T>
+      >
+{};
+
+
+// recurse on tuples
+template <
+  typename T0, typename T1, typename T2,
+  typename T3, typename T4, typename T5,
+  typename T6, typename T7, typename T8,
+  typename T9
+>
+  struct raw_reference_tuple_helper<
+    thrust::tuple<T0,T1,T2,T3,T4,T5,T6,T7,T8,T9>
+  >
+{
+  typedef thrust::tuple<
+    typename raw_reference_tuple_helper<T0>::type,
+    typename raw_reference_tuple_helper<T1>::type,
+    typename raw_reference_tuple_helper<T2>::type,
+    typename raw_reference_tuple_helper<T3>::type,
+    typename raw_reference_tuple_helper<T4>::type,
+    typename raw_reference_tuple_helper<T5>::type,
+    typename raw_reference_tuple_helper<T6>::type,
+    typename raw_reference_tuple_helper<T7>::type,
+    typename raw_reference_tuple_helper<T8>::type,
+    typename raw_reference_tuple_helper<T9>::type
+  > type;
+};
+
+
+template <
+  typename T0, typename T1, typename T2,
+  typename T3, typename T4, typename T5,
+  typename T6, typename T7, typename T8,
+  typename T9
+>
+  struct raw_reference_tuple_helper<
+    thrust::detail::tuple_of_iterator_references<T0,T1,T2,T3,T4,T5,T6,T7,T8,T9>
+  >
+{
+  typedef thrust::detail::tuple_of_iterator_references<
+    typename raw_reference_tuple_helper<T0>::type,
+    typename raw_reference_tuple_helper<T1>::type,
+    typename raw_reference_tuple_helper<T2>::type,
+    typename raw_reference_tuple_helper<T3>::type,
+    typename raw_reference_tuple_helper<T4>::type,
+    typename raw_reference_tuple_helper<T5>::type,
+    typename raw_reference_tuple_helper<T6>::type,
+    typename raw_reference_tuple_helper<T7>::type,
+    typename raw_reference_tuple_helper<T8>::type,
+    typename raw_reference_tuple_helper<T9>::type
+  > type;
+};
+
+
+} // end raw_reference_detail
+
+
+// if a tuple "tuple_type" is_unwrappable,
+//   then the raw_reference of tuple_type is a tuple of its members' raw_references
+//   else the raw_reference of tuple_type is tuple_type &
+template <
+  typename T0, typename T1, typename T2,
+  typename T3, typename T4, typename T5,
+  typename T6, typename T7, typename T8,
+  typename T9
+>
+  struct raw_reference<
+    thrust::tuple<T0,T1,T2,T3,T4,T5,T6,T7,T8,T9>
+  >
+{
+  private:
+    typedef thrust::tuple<T0,T1,T2,T3,T4,T5,T6,T7,T8,T9> tuple_type;
+
+  public:
+    typedef typename eval_if<
+      is_unwrappable<tuple_type>::value,
+      raw_reference_detail::raw_reference_tuple_helper<tuple_type>,
+      add_reference<tuple_type>
+    >::type type;
+};
+
+
+template <
+  typename T0, typename T1, typename T2,
+  typename T3, typename T4, typename T5,
+  typename T6, typename T7, typename T8,
+  typename T9
+>
+  struct raw_reference<
+    thrust::detail::tuple_of_iterator_references<T0,T1,T2,T3,T4,T5,T6,T7,T8,T9>
+  >
+{
+  private:
+    typedef detail::tuple_of_iterator_references<T0,T1,T2,T3,T4,T5,T6,T7,T8,T9> tuple_type;
+
+  public:
+    typedef typename raw_reference_detail::raw_reference_tuple_helper<tuple_type>::type type;
+
+    // XXX figure out why is_unwrappable seems to be broken for tuple_of_iterator_references
+    //typedef typename eval_if<
+    //  is_unwrappable<tuple_type>::value,
+    //  raw_reference_detail::raw_reference_tuple_helper<tuple_type>,
+    //  add_reference<tuple_type>
+    //>::type type;
+};
+
+
+struct raw_reference_caster
+{
+  template<typename T>
+  __host__ __device__
+  typename detail::raw_reference<T>::type operator()(T &ref)
+  {
+    return thrust::raw_reference_cast(ref);
+  }
+
+  template<typename T>
+  __host__ __device__
+  typename detail::raw_reference<const T>::type operator()(const T &ref)
+  {
+    return thrust::raw_reference_cast(ref);
+  }
+
+  template<
+    typename T0, typename T1, typename T2,
+    typename T3, typename T4, typename T5,
+    typename T6, typename T7, typename T8,
+    typename T9
+  >
+  __host__ __device__
+  typename detail::raw_reference<
+    thrust::detail::tuple_of_iterator_references<T0,T1,T2,T3,T4,T5,T6,T7,T8,T9>
+  >::type
+  operator()(thrust::detail::tuple_of_iterator_references<T0,T1,T2,T3,T4,T5,T6,T7,T8,T9> t,
+             typename enable_if<
+               is_unwrappable<thrust::detail::tuple_of_iterator_references<T0,T1,T2,T3,T4,T5,T6,T7,T8,T9> >::value
+             >::type * = 0)
+  {
+    return thrust::raw_reference_cast(t);
+  }
+}; // end raw_reference_caster
+
+
+} // end detail
+
+
+template<
+  typename T0, typename T1, typename T2,
+  typename T3, typename T4, typename T5,
+  typename T6, typename T7, typename T8,
+  typename T9
+>
+__host__ __device__
+typename detail::enable_if_unwrappable<
+  thrust::detail::tuple_of_iterator_references<T0,T1,T2,T3,T4,T5,T6,T7,T8,T9>,
+  typename detail::raw_reference<
+    thrust::detail::tuple_of_iterator_references<T0,T1,T2,T3,T4,T5,T6,T7,T8,T9>
+  >::type
+>::type
+raw_reference_cast(thrust::detail::tuple_of_iterator_references<T0,T1,T2,T3,T4,T5,T6,T7,T8,T9> t)
+{
+  thrust::detail::raw_reference_caster f;
+
+  // note that we pass raw_reference_tuple_helper, not raw_reference as the unary metafunction
+  // the subtle difference is important
+  return thrust::detail::tuple_host_device_transform<detail::raw_reference_detail::raw_reference_tuple_helper>(t, f);
+} // end raw_reference_cast
+
+
+} // end thrust
+
diff --git a/compat/thrust/detail/reduce.inl b/compat/thrust/detail/reduce.inl
new file mode 100644
index 0000000000..ba84423475
--- /dev/null
+++ b/compat/thrust/detail/reduce.inl
@@ -0,0 +1,261 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file reduce.inl
+ *  \brief Inline file for reduce.h.
+ */
+
+#include <thrust/reduce.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/system/detail/generic/select_system.h>
+#include <thrust/system/detail/generic/reduce.h>
+#include <thrust/system/detail/generic/reduce_by_key.h>
+#include <thrust/system/detail/adl/reduce.h>
+#include <thrust/system/detail/adl/reduce_by_key.h>
+
+namespace thrust
+{
+
+
+template<typename DerivedPolicy, typename InputIterator>
+  typename thrust::iterator_traits<InputIterator>::value_type
+    reduce(const thrust::detail::execution_policy_base<DerivedPolicy> &exec, InputIterator first, InputIterator last)
+{
+  using thrust::system::detail::generic::reduce;
+  return reduce(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last);
+} // end reduce()
+
+
+template<typename DerivedPolicy, typename InputIterator, typename T>
+  T reduce(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+           InputIterator first,
+           InputIterator last,
+           T init)
+{
+  using thrust::system::detail::generic::reduce;
+  return reduce(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, init);
+} // end reduce()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename T,
+         typename BinaryFunction>
+  T reduce(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+           InputIterator first,
+           InputIterator last,
+           T init,
+           BinaryFunction binary_op)
+{
+  using thrust::system::detail::generic::reduce;
+  return reduce(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, init, binary_op);
+} // end reduce()
+
+
+template <typename DerivedPolicy,
+          typename InputIterator1,
+          typename InputIterator2,
+          typename OutputIterator1,
+          typename OutputIterator2>
+  thrust::pair<OutputIterator1,OutputIterator2>
+  reduce_by_key(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                InputIterator1 keys_first, 
+                InputIterator1 keys_last,
+                InputIterator2 values_first,
+                OutputIterator1 keys_output,
+                OutputIterator2 values_output)
+{
+  using thrust::system::detail::generic::reduce_by_key;
+  return reduce_by_key(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), keys_first, keys_last, values_first, keys_output, values_output);
+} // end reduce_by_key()
+
+
+template <typename DerivedPolicy,
+          typename InputIterator1,
+          typename InputIterator2,
+          typename OutputIterator1,
+          typename OutputIterator2,
+          typename BinaryPredicate>
+  thrust::pair<OutputIterator1,OutputIterator2>
+  reduce_by_key(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                InputIterator1 keys_first, 
+                InputIterator1 keys_last,
+                InputIterator2 values_first,
+                OutputIterator1 keys_output,
+                OutputIterator2 values_output,
+                BinaryPredicate binary_pred)
+{
+  using thrust::system::detail::generic::reduce_by_key;
+  return reduce_by_key(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), keys_first, keys_last, values_first, keys_output, values_output, binary_pred);
+} // end reduce_by_key()
+
+
+template <typename DerivedPolicy,
+          typename InputIterator1,
+          typename InputIterator2,
+          typename OutputIterator1,
+          typename OutputIterator2,
+          typename BinaryPredicate,
+          typename BinaryFunction>
+  thrust::pair<OutputIterator1,OutputIterator2>
+  reduce_by_key(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                InputIterator1 keys_first, 
+                InputIterator1 keys_last,
+                InputIterator2 values_first,
+                OutputIterator1 keys_output,
+                OutputIterator2 values_output,
+                BinaryPredicate binary_pred,
+                BinaryFunction binary_op)
+{
+  using thrust::system::detail::generic::reduce_by_key;
+  return reduce_by_key(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), keys_first, keys_last, values_first, keys_output, values_output, binary_pred, binary_op);
+} // end reduce_by_key()
+
+
+template<typename InputIterator>
+typename thrust::iterator_traits<InputIterator>::value_type
+  reduce(InputIterator first,
+         InputIterator last)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator>::type System;
+
+  System system;
+
+  return thrust::reduce(select_system(system), first, last);
+}
+
+template<typename InputIterator,
+         typename T>
+   T reduce(InputIterator first,
+            InputIterator last,
+            T init)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator>::type System;
+
+  System system;
+
+  return thrust::reduce(select_system(system), first, last, init);
+}
+
+
+template<typename InputIterator,
+         typename T,
+         typename BinaryFunction>
+   T reduce(InputIterator first,
+            InputIterator last,
+            T init,
+            BinaryFunction binary_op)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator>::type System;
+
+  System system;
+
+  return thrust::reduce(select_system(system), first, last, init, binary_op);
+}
+
+template <typename InputIterator1,
+          typename InputIterator2,
+          typename OutputIterator1,
+          typename OutputIterator2>
+  thrust::pair<OutputIterator1,OutputIterator2>
+  reduce_by_key(InputIterator1 keys_first, 
+                InputIterator1 keys_last,
+                InputIterator2 values_first,
+                OutputIterator1 keys_output,
+                OutputIterator2 values_output)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator1>::type  System1;
+  typedef typename thrust::iterator_system<InputIterator2>::type  System2;
+  typedef typename thrust::iterator_system<OutputIterator1>::type System3;
+  typedef typename thrust::iterator_system<OutputIterator2>::type System4;
+
+  System1 system1;
+  System2 system2;
+  System3 system3;
+  System4 system4;
+
+  return thrust::reduce_by_key(select_system(system1,system2,system3,system4), keys_first, keys_last, values_first, keys_output, values_output);
+}
+
+template <typename InputIterator1,
+          typename InputIterator2,
+          typename OutputIterator1,
+          typename OutputIterator2,
+          typename BinaryPredicate>
+  thrust::pair<OutputIterator1,OutputIterator2>
+  reduce_by_key(InputIterator1 keys_first, 
+                InputIterator1 keys_last,
+                InputIterator2 values_first,
+                OutputIterator1 keys_output,
+                OutputIterator2 values_output,
+                BinaryPredicate binary_pred)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator1>::type  System1;
+  typedef typename thrust::iterator_system<InputIterator2>::type  System2;
+  typedef typename thrust::iterator_system<OutputIterator1>::type System3;
+  typedef typename thrust::iterator_system<OutputIterator2>::type System4;
+
+  System1 system1;
+  System2 system2;
+  System3 system3;
+  System4 system4;
+
+  return thrust::reduce_by_key(select_system(system1,system2,system3,system4), keys_first, keys_last, values_first, keys_output, values_output, binary_pred);
+}
+
+template <typename InputIterator1,
+          typename InputIterator2,
+          typename OutputIterator1,
+          typename OutputIterator2,
+          typename BinaryPredicate,
+          typename BinaryFunction>
+  thrust::pair<OutputIterator1,OutputIterator2>
+  reduce_by_key(InputIterator1 keys_first, 
+                InputIterator1 keys_last,
+                InputIterator2 values_first,
+                OutputIterator1 keys_output,
+                OutputIterator2 values_output,
+                BinaryPredicate binary_pred,
+                BinaryFunction binary_op)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator1>::type  System1;
+  typedef typename thrust::iterator_system<InputIterator2>::type  System2;
+  typedef typename thrust::iterator_system<OutputIterator1>::type System3;
+  typedef typename thrust::iterator_system<OutputIterator2>::type System4;
+
+  System1 system1;
+  System2 system2;
+  System3 system3;
+  System4 system4;
+
+  return thrust::reduce_by_key(select_system(system1,system2,system3,system4), keys_first, keys_last, values_first, keys_output, values_output, binary_pred, binary_op);
+}
+
+} // end namespace thrust
+
diff --git a/compat/thrust/detail/reference.h b/compat/thrust/detail/reference.h
new file mode 100644
index 0000000000..8c0b06186f
--- /dev/null
+++ b/compat/thrust/detail/reference.h
@@ -0,0 +1,167 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/type_traits.h>
+#include <thrust/detail/use_default.h>
+#include <thrust/detail/reference_forward_declaration.h>
+
+
+namespace thrust
+{
+namespace detail
+{
+
+template<typename> struct is_wrapped_reference;
+
+}
+
+// the base type for all of thrust's system-annotated references.
+// for reasonable reference-like semantics, derived types must reimplement the following:
+// 1. constructor from pointer
+// 2. copy constructor
+// 3. templated copy constructor from other reference
+// 4. templated assignment from other reference
+// 5. assignment from value_type
+template<typename Element, typename Pointer, typename Derived>
+  class reference
+{
+  private:
+    typedef typename thrust::detail::eval_if<
+      thrust::detail::is_same<Derived,use_default>::value,
+      thrust::detail::identity_<reference>,
+      thrust::detail::identity_<Derived>
+    >::type derived_type;
+
+    // hint for is_wrapped_reference lets it know that this type (or a derived type)
+    // is a wrapped reference
+    struct wrapped_reference_hint {};
+    template<typename> friend struct thrust::detail::is_wrapped_reference;
+
+  public:
+    typedef Pointer                                              pointer;
+    typedef typename thrust::detail::remove_const<Element>::type value_type;
+
+    __host__ __device__
+    explicit reference(const pointer &ptr);
+
+    template<typename OtherElement, typename OtherPointer, typename OtherDerived>
+    __host__ __device__
+    reference(const reference<OtherElement,OtherPointer,OtherDerived> &other,
+              typename thrust::detail::enable_if_convertible<
+                typename reference<OtherElement,OtherPointer,OtherDerived>::pointer,
+                pointer
+              >::type * = 0);
+
+    __host__ __device__
+    derived_type &operator=(const reference &other);
+
+    // XXX this may need an enable_if
+    template<typename OtherElement, typename OtherPointer, typename OtherDerived>
+    __host__ __device__
+    derived_type &operator=(const reference<OtherElement,OtherPointer,OtherDerived> &other);
+
+    __host__ __device__
+    derived_type &operator=(const value_type &x);
+
+    __host__ __device__
+    pointer operator&() const;
+
+    __host__ __device__
+    operator value_type () const;
+
+    __host__ __device__
+    void swap(derived_type &other);
+
+    derived_type &operator++();
+
+    value_type operator++(int);
+
+    // XXX parameterize the type of rhs
+    derived_type &operator+=(const value_type &rhs);
+
+    derived_type &operator--();
+
+    value_type operator--(int);
+
+    // XXX parameterize the type of rhs
+    derived_type &operator-=(const value_type &rhs);
+
+    // XXX parameterize the type of rhs
+    derived_type &operator*=(const value_type &rhs);
+
+    // XXX parameterize the type of rhs
+    derived_type &operator/=(const value_type &rhs);
+
+    // XXX parameterize the type of rhs
+    derived_type &operator%=(const value_type &rhs);
+
+    // XXX parameterize the type of rhs
+    derived_type &operator<<=(const value_type &rhs);
+
+    // XXX parameterize the type of rhs
+    derived_type &operator>>=(const value_type &rhs);
+
+    // XXX parameterize the type of rhs
+    derived_type &operator&=(const value_type &rhs);
+
+    // XXX parameterize the type of rhs
+    derived_type &operator|=(const value_type &rhs);
+
+    // XXX parameterize the type of rhs
+    derived_type &operator^=(const value_type &rhs);
+
+  private:
+    const pointer m_ptr;
+
+    // allow access to m_ptr for other references
+    template <typename OtherElement, typename OtherPointer, typename OtherDerived> friend class reference;
+
+    template<typename System>
+    __host__ __device__
+    inline value_type strip_const_get_value(const System &system) const;
+
+    template<typename OtherPointer>
+    __host__ __device__
+    inline void assign_from(OtherPointer src);
+
+    // XXX this helper exists only to avoid warnings about null references from the other assign_from
+    template<typename System1, typename System2, typename OtherPointer>
+    inline __host__ __device__
+    void assign_from(System1 *system1, System2 *system2, OtherPointer src);
+
+    template<typename System, typename OtherPointer>
+    __host__ __device__
+    inline void strip_const_assign_value(const System &system, OtherPointer src);
+
+    // XXX this helper exists only to avoid warnings about null references from the other swap
+    template<typename System>
+    inline __host__ __device__
+    void swap(System *system, derived_type &other);
+
+    // XXX this helper exists only to avoid warnings about null references from operator value_type ()
+    template<typename System>
+    inline __host__ __device__
+    value_type convert_to_value_type(System *system) const;
+}; // end reference
+
+  
+} // end thrust
+
+#include <thrust/detail/reference.inl>
+
diff --git a/compat/thrust/detail/reference.inl b/compat/thrust/detail/reference.inl
new file mode 100644
index 0000000000..8b55edb712
--- /dev/null
+++ b/compat/thrust/detail/reference.inl
@@ -0,0 +1,361 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/reference.h>
+#include <thrust/detail/type_traits.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/system/detail/generic/select_system.h>
+#include <thrust/system/detail/generic/memory.h>
+#include <thrust/system/detail/adl/get_value.h>
+#include <thrust/system/detail/adl/assign_value.h>
+#include <thrust/system/detail/adl/iter_swap.h>
+
+
+namespace thrust
+{
+
+
+template<typename Element, typename Pointer, typename Derived>
+  template<typename OtherElement, typename OtherPointer, typename OtherDerived>
+    reference<Element,Pointer,Derived>
+      ::reference(const reference<OtherElement,OtherPointer,OtherDerived> &other,
+                  typename thrust::detail::enable_if_convertible<
+                    typename reference<OtherElement,OtherPointer,OtherDerived>::pointer,
+                    pointer
+                  >::type *)
+        : m_ptr(other.m_ptr)
+{}
+
+
+template<typename Element, typename Pointer, typename Derived>
+  reference<Element,Pointer,Derived>
+    ::reference(const pointer &ptr)
+      : m_ptr(ptr)
+{}
+
+
+template<typename Element, typename Pointer, typename Derived>
+  typename reference<Element,Pointer,Derived>::pointer
+    reference<Element,Pointer,Derived>
+      ::operator&() const
+{
+  return m_ptr;
+} // end reference::operator&()
+
+
+template<typename Element, typename Pointer, typename Derived>
+  typename reference<Element,Pointer,Derived>::derived_type &
+    reference<Element,Pointer,Derived>
+      ::operator=(const value_type &v)
+{
+  assign_from(&v);
+  return static_cast<derived_type&>(*this);
+} // end reference::operator=()
+
+
+template<typename Element, typename Pointer, typename Derived>
+  typename reference<Element,Pointer,Derived>::derived_type &
+    reference<Element,Pointer,Derived>
+      ::operator=(const reference &other)
+{
+  assign_from(&other); 
+  return static_cast<derived_type&>(*this);
+} // end reference::operator=()
+
+
+template<typename Element, typename Pointer, typename Derived>
+  template<typename OtherElement, typename OtherPointer, typename OtherDerived>
+    typename reference<Element,Pointer,Derived>::derived_type &
+      reference<Element,Pointer,Derived>
+        ::operator=(const reference<OtherElement,OtherPointer,OtherDerived> &other)
+{
+  assign_from(&other);
+  return static_cast<derived_type&>(*this);
+} // end reference::operator=()
+
+
+template<typename Element, typename Pointer, typename Derived>
+  template<typename System>
+    typename reference<Element,Pointer,Derived>::value_type
+      reference<Element,Pointer,Derived>
+        ::convert_to_value_type(System *system) const
+{
+  using thrust::system::detail::generic::select_system;
+  return strip_const_get_value(select_system(*system));
+} // end convert_to_value_type()
+
+
+template<typename Element, typename Pointer, typename Derived>
+  reference<Element,Pointer,Derived>
+    ::operator typename reference<Element,Pointer,Derived>::value_type () const
+{
+  typedef typename thrust::iterator_system<pointer>::type System;
+
+  // XXX avoid default-constructing a system
+  // XXX use null a reference for dispatching
+  // XXX this assumes that the eventual invocation of
+  // XXX get_value will not access system state
+  System *system = 0;
+
+  return convert_to_value_type(system);
+} // end reference::operator value_type ()
+
+
+template<typename Element, typename Pointer, typename Derived>
+  template<typename System>
+    typename reference<Element,Pointer,Derived>::value_type
+      reference<Element,Pointer,Derived>
+        ::strip_const_get_value(const System &system) const
+{
+  System &non_const_system = const_cast<System&>(system);
+
+  using thrust::system::detail::generic::get_value;
+
+  return get_value(thrust::detail::derived_cast(non_const_system), m_ptr);
+} // end reference::strip_const_get_value()
+
+
+template<typename Element, typename Pointer, typename Derived>
+  template<typename System1, typename System2, typename OtherPointer>
+    void reference<Element,Pointer,Derived>
+      ::assign_from(System1 *system1, System2 *system2, OtherPointer src)
+{
+  using thrust::system::detail::generic::select_system;
+
+  strip_const_assign_value(select_system(*system1, *system2), src);
+} // end assign_from()
+
+
+template<typename Element, typename Pointer, typename Derived>
+  template<typename OtherPointer>
+    void reference<Element,Pointer,Derived>
+      ::assign_from(OtherPointer src)
+{
+  typedef typename thrust::iterator_system<pointer>::type      System1;
+  typedef typename thrust::iterator_system<OtherPointer>::type System2;
+
+  // XXX avoid default-constructing a system
+  // XXX use null references for dispatching
+  // XXX this assumes that the eventual invocation of
+  // XXX assign_value will not access system state
+  System1 *system1 = 0;
+  System2 *system2 = 0;
+
+  assign_from(system1, system2, src);
+} // end assign_from()
+
+
+template<typename Element, typename Pointer, typename Derived>
+  template<typename System, typename OtherPointer>
+    void reference<Element,Pointer,Derived>
+      ::strip_const_assign_value(const System &system, OtherPointer src)
+{
+  System &non_const_system = const_cast<System&>(system);
+
+  using thrust::system::detail::generic::assign_value;
+
+  assign_value(thrust::detail::derived_cast(non_const_system), m_ptr, src);
+} // end strip_const_assign_value()
+
+
+template<typename Element, typename Pointer, typename Derived>
+  template<typename System>
+    void reference<Element,Pointer,Derived>
+      ::swap(System *system, derived_type &other)
+{
+  using thrust::system::detail::generic::select_system;
+  using thrust::system::detail::generic::iter_swap;
+
+  iter_swap(select_system(*system, *system), m_ptr, other.m_ptr);
+} // end reference::swap()
+
+
+template<typename Element, typename Pointer, typename Derived>
+  void reference<Element,Pointer,Derived>
+    ::swap(derived_type &other)
+{
+  typedef typename thrust::iterator_system<pointer>::type System;
+
+  // XXX avoid default-constructing a system
+  // XXX use null references for dispatching
+  // XXX this assumes that the eventual invocation
+  // XXX of iter_swap will not access system state
+  System *system = 0;
+
+  swap(system, other);
+} // end reference::swap()
+
+
+template<typename Element, typename Pointer, typename Derived>
+  typename reference<Element,Pointer,Derived>::derived_type &
+    reference<Element,Pointer,Derived>
+      ::operator++(void)
+{
+  value_type temp = *this;
+  ++temp;
+  *this = temp;
+  return static_cast<derived_type&>(*this);
+} // end reference::operator++()
+
+
+template<typename Element, typename Pointer, typename Derived>
+  typename reference<Element,Pointer,Derived>::value_type
+    reference<Element,Pointer,Derived>
+      ::operator++(int)
+{
+  value_type temp = *this;
+  value_type result = temp++;
+  *this = temp;
+  return result;
+} // end reference::operator++()
+
+
+template<typename Element, typename Pointer, typename Derived>
+  typename reference<Element,Pointer,Derived>::derived_type &
+    reference<Element,Pointer,Derived>
+      ::operator+=(const value_type &rhs)
+{
+  value_type temp = *this;
+  temp += rhs;
+  *this = temp;
+  return static_cast<derived_type&>(*this);
+} // end reference::operator+=()
+
+template<typename Element, typename Pointer, typename Derived>
+  typename reference<Element,Pointer,Derived>::derived_type &
+    reference<Element,Pointer,Derived>
+      ::operator--(void)
+{
+  value_type temp = *this;
+  --temp;
+  *this = temp;
+  return static_cast<derived_type&>(*this);
+} // end reference::operator--()
+
+template<typename Element, typename Pointer, typename Derived>
+  typename reference<Element,Pointer,Derived>::value_type
+    reference<Element,Pointer,Derived>
+      ::operator--(int)
+{
+  value_type temp = *this;
+  value_type result = temp--;
+  *this = temp;
+  return result;
+} // end reference::operator--()
+
+template<typename Element, typename Pointer, typename Derived>
+  typename reference<Element,Pointer,Derived>::derived_type &
+    reference<Element,Pointer,Derived>
+      ::operator-=(const value_type &rhs)
+{
+  value_type temp = *this;
+  temp -= rhs;
+  *this = temp;
+  return static_cast<derived_type&>(*this);
+} // end reference::operator-=()
+
+template<typename Element, typename Pointer, typename Derived>
+  typename reference<Element,Pointer,Derived>::derived_type &
+    reference<Element,Pointer,Derived>
+      ::operator*=(const value_type &rhs)
+{
+  value_type temp = *this;
+  temp *= rhs;
+  *this = temp;
+  return static_cast<derived_type&>(*this);
+} // end reference::operator*=()
+
+template<typename Element, typename Pointer, typename Derived>
+  typename reference<Element,Pointer,Derived>::derived_type &
+    reference<Element,Pointer,Derived>
+      ::operator/=(const value_type &rhs)
+{
+  value_type temp = *this;
+  temp /= rhs;
+  *this = temp;
+  return static_cast<derived_type&>(*this);
+} // end reference::operator/=()
+
+template<typename Element, typename Pointer, typename Derived>
+  typename reference<Element,Pointer,Derived>::derived_type &
+    reference<Element,Pointer,Derived>
+      ::operator%=(const value_type &rhs)
+{
+  value_type temp = *this;
+  temp %= rhs;
+  *this = temp;
+  return static_cast<derived_type&>(*this);
+} // end reference::operator%=()
+
+template<typename Element, typename Pointer, typename Derived>
+  typename reference<Element,Pointer,Derived>::derived_type &
+    reference<Element,Pointer,Derived>
+      ::operator<<=(const value_type &rhs)
+{
+  value_type temp = *this;
+  temp <<= rhs;
+  *this = temp;
+  return static_cast<derived_type&>(*this);
+} // end reference::operator<<=()
+
+template<typename Element, typename Pointer, typename Derived>
+  typename reference<Element,Pointer,Derived>::derived_type &
+    reference<Element,Pointer,Derived>
+      ::operator>>=(const value_type &rhs)
+{
+  value_type temp = *this;
+  temp >>= rhs;
+  *this = temp;
+  return static_cast<derived_type&>(*this);
+} // end reference::operator>>=()
+
+template<typename Element, typename Pointer, typename Derived>
+  typename reference<Element,Pointer,Derived>::derived_type &
+    reference<Element,Pointer,Derived>
+      ::operator&=(const value_type &rhs)
+{
+  value_type temp = *this;
+  temp &= rhs;
+  *this = temp;
+  return static_cast<derived_type&>(*this);
+} // end reference::operator&=()
+
+template<typename Element, typename Pointer, typename Derived>
+  typename reference<Element,Pointer,Derived>::derived_type &
+    reference<Element,Pointer,Derived>
+      ::operator|=(const value_type &rhs)
+{
+  value_type temp = *this;
+  temp |= rhs;
+  *this = temp;
+  return static_cast<derived_type&>(*this);
+} // end reference::operator|=()
+
+template<typename Element, typename Pointer, typename Derived>
+  typename reference<Element,Pointer,Derived>::derived_type &
+    reference<Element,Pointer,Derived>
+      ::operator^=(const value_type &rhs)
+{
+  value_type temp = *this;
+  temp ^= rhs;
+  *this = temp;
+  return static_cast<derived_type&>(*this);
+} // end reference::operator^=()
+
+  
+} // end thrust
+
diff --git a/compat/thrust/detail/reference_forward_declaration.h b/compat/thrust/detail/reference_forward_declaration.h
new file mode 100644
index 0000000000..60524d3ec2
--- /dev/null
+++ b/compat/thrust/detail/reference_forward_declaration.h
@@ -0,0 +1,28 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/use_default.h>
+
+namespace thrust
+{
+
+template<typename Element, typename Pointer, typename Derived = use_default> class reference;
+
+} // end thrust
+
diff --git a/compat/thrust/detail/remove.inl b/compat/thrust/detail/remove.inl
new file mode 100644
index 0000000000..5675243fbd
--- /dev/null
+++ b/compat/thrust/detail/remove.inl
@@ -0,0 +1,238 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file remove.inl
+ *  \brief Inline file for remove.h.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/remove.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/system/detail/generic/select_system.h>
+#include <thrust/system/detail/generic/remove.h>
+#include <thrust/system/detail/adl/remove.h>
+
+namespace thrust
+{
+
+
+template<typename DerivedPolicy,
+         typename ForwardIterator,
+         typename T>
+  ForwardIterator remove(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                         ForwardIterator first,
+                         ForwardIterator last,
+                         const T &value)
+{
+  using thrust::system::detail::generic::remove;
+  return remove(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, value);
+} // end remove()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator,
+         typename T>
+  OutputIterator remove_copy(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                             InputIterator first,
+                             InputIterator last,
+                             OutputIterator result,
+                             const T &value)
+{
+  using thrust::system::detail::generic::remove_copy;
+  return remove_copy(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, result, value);
+} // end remove_copy()
+
+
+template<typename DerivedPolicy,
+         typename ForwardIterator,
+         typename Predicate>
+  ForwardIterator remove_if(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                            ForwardIterator first,
+                            ForwardIterator last,
+                            Predicate pred)
+{
+  using thrust::system::detail::generic::remove_if;
+  return remove_if(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, pred);
+} // end remove_if()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator,
+         typename Predicate>
+  OutputIterator remove_copy_if(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                InputIterator first,
+                                InputIterator last,
+                                OutputIterator result,
+                                Predicate pred)
+{
+  using thrust::system::detail::generic::remove_copy_if;
+  return remove_copy_if(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, result, pred);
+} // end remove_copy_if()
+
+
+template<typename DerivedPolicy,
+         typename ForwardIterator,
+         typename InputIterator,
+         typename Predicate>
+  ForwardIterator remove_if(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                            ForwardIterator first,
+                            ForwardIterator last,
+                            InputIterator stencil,
+                            Predicate pred)
+{
+  using thrust::system::detail::generic::remove_if;
+  return remove_if(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, stencil, pred);
+} // end remove_if()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename Predicate>
+  OutputIterator remove_copy_if(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                InputIterator1 first,
+                                InputIterator1 last,
+                                InputIterator2 stencil,
+                                OutputIterator result,
+                                Predicate pred)
+{
+  using thrust::system::detail::generic::remove_copy_if;
+  return remove_copy_if(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, stencil, result, pred);
+} // end remove_copy_if()
+
+
+template<typename ForwardIterator,
+         typename T>
+  ForwardIterator remove(ForwardIterator first,
+                         ForwardIterator last,
+                         const T &value)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<ForwardIterator>::type System;
+
+  System system;
+
+  return thrust::remove(select_system(system), first, last, value);
+} // end remove()
+
+
+template<typename InputIterator,
+         typename OutputIterator,
+         typename T>
+  OutputIterator remove_copy(InputIterator first,
+                             InputIterator last,
+                             OutputIterator result,
+                             const T &value)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator>::type  System1;
+  typedef typename thrust::iterator_system<OutputIterator>::type System2;
+
+  System1 system1;
+  System2 system2;
+
+  return thrust::remove_copy(select_system(system1,system2), first, last, result, value);
+} // end remove_copy()
+
+
+template<typename ForwardIterator,
+         typename Predicate>
+  ForwardIterator remove_if(ForwardIterator first,
+                            ForwardIterator last,
+                            Predicate pred)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<ForwardIterator>::type System;
+
+  System system;
+
+  return thrust::remove_if(select_system(system), first, last, pred);
+} // end remove_if()
+
+
+template<typename ForwardIterator,
+         typename InputIterator,
+         typename Predicate>
+  ForwardIterator remove_if(ForwardIterator first,
+                            ForwardIterator last,
+                            InputIterator stencil,
+                            Predicate pred)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<ForwardIterator>::type System1;
+  typedef typename thrust::iterator_system<InputIterator>::type   System2;
+
+  System1 system1;
+  System2 system2;
+
+  return thrust::remove_if(select_system(system1,system2), first, last, stencil, pred);
+} // end remove_if()
+
+
+template<typename InputIterator,
+         typename OutputIterator,
+         typename Predicate>
+  OutputIterator remove_copy_if(InputIterator first,
+                                InputIterator last,
+                                OutputIterator result,
+                                Predicate pred)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator>::type  System1;
+  typedef typename thrust::iterator_system<OutputIterator>::type System2;
+
+  System1 system1;
+  System2 system2;
+
+  return thrust::remove_copy_if(select_system(system1,system2), first, last, result, pred);
+} // end remove_copy_if()
+
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename Predicate>
+  OutputIterator remove_copy_if(InputIterator1 first,
+                                InputIterator1 last,
+                                InputIterator2 stencil,
+                                OutputIterator result,
+                                Predicate pred)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator1>::type System1;
+  typedef typename thrust::iterator_system<InputIterator2>::type System2;
+  typedef typename thrust::iterator_system<OutputIterator>::type System3;
+
+  System1 system1;
+  System2 system2;
+  System3 system3;
+
+  return thrust::remove_copy_if(select_system(system1,system2,system3), first, last, stencil, result, pred);
+} // end remove_copy_if()
+
+
+} // end namespace thrust
+
diff --git a/compat/thrust/detail/replace.inl b/compat/thrust/detail/replace.inl
new file mode 100644
index 0000000000..1eaf24d621
--- /dev/null
+++ b/compat/thrust/detail/replace.inl
@@ -0,0 +1,210 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file replace.inl
+ *  \brief Inline file for replace.h.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/replace.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/system/detail/generic/select_system.h>
+#include <thrust/system/detail/generic/replace.h>
+#include <thrust/system/detail/adl/replace.h>
+
+namespace thrust
+{
+
+
+template<typename DerivedPolicy, typename ForwardIterator, typename T>
+  void replace(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+               ForwardIterator first, ForwardIterator last,
+               const T &old_value,
+               const T &new_value)
+{
+  using thrust::system::detail::generic::replace;
+  return replace(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, old_value, new_value);
+} // end replace()
+
+
+template<typename DerivedPolicy, typename ForwardIterator, typename Predicate, typename T>
+  void replace_if(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                  ForwardIterator first, ForwardIterator last,
+                  Predicate pred,
+                  const T &new_value)
+{
+  using thrust::system::detail::generic::replace_if;
+  return replace_if(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, pred, new_value);
+} // end replace_if()
+
+
+template<typename DerivedPolicy, typename ForwardIterator, typename InputIterator, typename Predicate, typename T>
+  void replace_if(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                  ForwardIterator first, ForwardIterator last,
+                  InputIterator stencil,
+                  Predicate pred,
+                  const T &new_value)
+{
+  using thrust::system::detail::generic::replace_if;
+  return replace_if(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, stencil, pred, new_value);
+} // end replace_if()
+
+
+template<typename DerivedPolicy, typename InputIterator, typename OutputIterator, typename T>
+  OutputIterator replace_copy(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                              InputIterator first, InputIterator last,
+                              OutputIterator result,
+                              const T &old_value,
+                              const T &new_value)
+{
+  using thrust::system::detail::generic::replace_copy;
+  return replace_copy(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, result, old_value, new_value);
+} // end replace_copy()
+
+
+template<typename DerivedPolicy, typename InputIterator, typename OutputIterator, typename Predicate, typename T>
+  OutputIterator replace_copy_if(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                 InputIterator first, InputIterator last,
+                                 OutputIterator result,
+                                 Predicate pred,
+                                 const T &new_value)
+{
+  using thrust::system::detail::generic::replace_copy_if;
+  return replace_copy_if(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, result, pred, new_value);
+} // end replace_copy_if()
+
+
+template<typename DerivedPolicy, typename InputIterator1, typename InputIterator2, typename OutputIterator, typename Predicate, typename T>
+  OutputIterator replace_copy_if(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                 InputIterator1 first, InputIterator1 last,
+                                 InputIterator2 stencil,
+                                 OutputIterator result,
+                                 Predicate pred,
+                                 const T &new_value)
+{
+  using thrust::system::detail::generic::replace_copy_if;
+  return replace_copy_if(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, stencil, result, pred, new_value);
+} // end replace_copy_if()
+
+
+template<typename InputIterator, typename OutputIterator, typename Predicate, typename T>
+  OutputIterator replace_copy_if(InputIterator first, InputIterator last,
+                                 OutputIterator result,
+                                 Predicate pred,
+                                 const T &new_value)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator>::type  System1;
+  typedef typename thrust::iterator_system<OutputIterator>::type System2;
+
+  System1 system1;
+  System2 system2;
+
+  return thrust::replace_copy_if(select_system(system1,system2), first, last, result, pred, new_value);
+} // end replace_copy_if()
+
+
+template<typename InputIterator1, typename InputIterator2, typename OutputIterator, typename Predicate, typename T>
+  OutputIterator replace_copy_if(InputIterator1 first, InputIterator1 last,
+                                 InputIterator2 stencil,
+                                 OutputIterator result,
+                                 Predicate pred,
+                                 const T &new_value)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator1>::type System1;
+  typedef typename thrust::iterator_system<InputIterator2>::type System2;
+  typedef typename thrust::iterator_system<OutputIterator>::type System3;
+
+  System1 system1;
+  System2 system2;
+  System3 system3;
+
+  return thrust::replace_copy_if(select_system(system1,system2,system3), first, last, stencil, result, pred, new_value);
+} // end replace_copy_if()
+
+
+template<typename InputIterator, typename OutputIterator, typename T>
+  OutputIterator replace_copy(InputIterator first, InputIterator last,
+                              OutputIterator result,
+                              const T &old_value,
+                              const T &new_value)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator>::type  System1;
+  typedef typename thrust::iterator_system<OutputIterator>::type System2;
+
+  System1 system1;
+  System2 system2;
+
+  return thrust::replace_copy(select_system(system1,system2), first, last, result, old_value, new_value);
+} // end replace_copy()
+
+
+template<typename ForwardIterator, typename Predicate, typename T>
+  void replace_if(ForwardIterator first, ForwardIterator last,
+                  Predicate pred,
+                  const T &new_value)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<ForwardIterator>::type System;
+
+  System system;
+
+  return thrust::replace_if(select_system(system), first, last, pred, new_value);
+} // end replace_if()
+
+
+template<typename ForwardIterator, typename InputIterator, typename Predicate, typename T>
+  void replace_if(ForwardIterator first, ForwardIterator last,
+                  InputIterator stencil,
+                  Predicate pred,
+                  const T &new_value)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<ForwardIterator>::type System1;
+  typedef typename thrust::iterator_system<InputIterator>::type   System2;
+
+  System1 system1;
+  System2 system2;
+
+  return thrust::replace_if(select_system(system1,system2), first, last, stencil, pred, new_value);
+} // end replace_if()
+
+
+template<typename ForwardIterator, typename T>
+  void replace(ForwardIterator first, ForwardIterator last,
+               const T &old_value,
+               const T &new_value)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<ForwardIterator>::type System;
+
+  System system;
+
+  return thrust::replace(select_system(system), first, last, old_value, new_value);
+} // end replace()
+
+
+} // end namespace thrust
+
diff --git a/compat/thrust/detail/reverse.inl b/compat/thrust/detail/reverse.inl
new file mode 100644
index 0000000000..18c26c00e6
--- /dev/null
+++ b/compat/thrust/detail/reverse.inl
@@ -0,0 +1,87 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file reverse.inl
+ *  \brief Inline file for reverse.h.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/reverse.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/system/detail/generic/select_system.h>
+#include <thrust/system/detail/generic/reverse.h>
+#include <thrust/system/detail/adl/reverse.h>
+
+namespace thrust
+{
+
+
+template<typename DerivedPolicy, typename BidirectionalIterator>
+  void reverse(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+               BidirectionalIterator first,
+               BidirectionalIterator last)
+{
+  using thrust::system::detail::generic::reverse;
+  return reverse(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last);
+} // end reverse()
+
+
+template<typename DerivedPolicy, typename BidirectionalIterator, typename OutputIterator>
+  OutputIterator reverse_copy(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                              BidirectionalIterator first,
+                              BidirectionalIterator last,
+                              OutputIterator result)
+{
+  using thrust::system::detail::generic::reverse_copy;
+  return reverse_copy(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, result);
+} // end reverse_copy()
+
+
+template<typename BidirectionalIterator>
+  void reverse(BidirectionalIterator first,
+               BidirectionalIterator last)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<BidirectionalIterator>::type System;
+
+  System system;
+
+  return thrust::reverse(select_system(system), first, last);
+} // end reverse()
+
+
+template<typename BidirectionalIterator,
+         typename OutputIterator>
+  OutputIterator reverse_copy(BidirectionalIterator first,
+                              BidirectionalIterator last,
+                              OutputIterator result)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<BidirectionalIterator>::type System1;
+  typedef typename thrust::iterator_system<OutputIterator>::type        System2;
+
+  System1 system1;
+  System2 system2;
+
+  return thrust::reverse_copy(select_system(system1,system2), first, last, result);
+} // end reverse_copy()
+
+
+} // end thrust
+
diff --git a/compat/thrust/detail/scan.inl b/compat/thrust/detail/scan.inl
new file mode 100644
index 0000000000..3e5fd9b4f7
--- /dev/null
+++ b/compat/thrust/detail/scan.inl
@@ -0,0 +1,502 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file scan.inl
+ *  \brief Inline file for scan.h.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/scan.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/system/detail/generic/select_system.h>
+#include <thrust/system/detail/generic/scan.h>
+#include <thrust/system/detail/generic/scan_by_key.h>
+#include <thrust/system/detail/adl/scan.h>
+#include <thrust/system/detail/adl/scan_by_key.h>
+
+namespace thrust
+{
+
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator>
+  OutputIterator inclusive_scan(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                InputIterator first,
+                                InputIterator last,
+                                OutputIterator result)
+{
+  using thrust::system::detail::generic::inclusive_scan;
+  return inclusive_scan(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, result);
+} // end inclusive_scan() 
+
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator,
+         typename AssociativeOperator>
+  OutputIterator inclusive_scan(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                InputIterator first,
+                                InputIterator last,
+                                OutputIterator result,
+                                AssociativeOperator binary_op)
+{
+  using thrust::system::detail::generic::inclusive_scan;
+  return inclusive_scan(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, result, binary_op);
+} // end inclusive_scan()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator>
+  OutputIterator exclusive_scan(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                InputIterator first,
+                                InputIterator last,
+                                OutputIterator result)
+{
+  using thrust::system::detail::generic::exclusive_scan;
+  return exclusive_scan(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, result);
+} // end exclusive_scan()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator,
+         typename T>
+  OutputIterator exclusive_scan(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                InputIterator first,
+                                InputIterator last,
+                                OutputIterator result,
+                                T init)
+{
+  using thrust::system::detail::generic::exclusive_scan;
+  return exclusive_scan(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, result, init);
+} // end exclusive_scan()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator,
+         typename T,
+         typename AssociativeOperator>
+  OutputIterator exclusive_scan(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                InputIterator first,
+                                InputIterator last,
+                                OutputIterator result,
+                                T init,
+                                AssociativeOperator binary_op)
+{
+  using thrust::system::detail::generic::exclusive_scan;
+  return exclusive_scan(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, result, init, binary_op);
+} // end exclusive_scan()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator>
+  OutputIterator inclusive_scan_by_key(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                       InputIterator1 first1,
+                                       InputIterator1 last1,
+                                       InputIterator2 first2,
+                                       OutputIterator result)
+{
+  using thrust::system::detail::generic::inclusive_scan_by_key;
+  return inclusive_scan_by_key(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first1, last1, first2, result);
+} // end inclusive_scan_by_key()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename BinaryPredicate>
+  OutputIterator inclusive_scan_by_key(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                       InputIterator1 first1,
+                                       InputIterator1 last1,
+                                       InputIterator2 first2,
+                                       OutputIterator result,
+                                       BinaryPredicate binary_pred)
+{
+  using thrust::system::detail::generic::inclusive_scan_by_key;
+  return inclusive_scan_by_key(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first1, last1, first2, result, binary_pred);
+} // end inclusive_scan_by_key()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename BinaryPredicate,
+         typename AssociativeOperator>
+  OutputIterator inclusive_scan_by_key(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                       InputIterator1 first1,
+                                       InputIterator1 last1,
+                                       InputIterator2 first2,
+                                       OutputIterator result,
+                                       BinaryPredicate binary_pred,
+                                       AssociativeOperator binary_op)
+{
+  using thrust::system::detail::generic::inclusive_scan_by_key;
+  return inclusive_scan_by_key(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first1, last1, first2, result, binary_pred, binary_op);
+} // end inclusive_scan_by_key()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator>
+  OutputIterator exclusive_scan_by_key(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                       InputIterator1 first1,
+                                       InputIterator1 last1,
+                                       InputIterator2 first2,
+                                       OutputIterator result)
+{
+  using thrust::system::detail::generic::exclusive_scan_by_key;
+  return exclusive_scan_by_key(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first1, last1, first2, result);
+} // end exclusive_scan_by_key()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename T>
+  OutputIterator exclusive_scan_by_key(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                       InputIterator1 first1,
+                                       InputIterator1 last1,
+                                       InputIterator2 first2,
+                                       OutputIterator result,
+                                       T init)
+{
+  using thrust::system::detail::generic::exclusive_scan_by_key;
+  return exclusive_scan_by_key(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first1, last1, first2, result, init);
+} // end exclusive_scan_by_key()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename T,
+         typename BinaryPredicate>
+  OutputIterator exclusive_scan_by_key(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                       InputIterator1 first1,
+                                       InputIterator1 last1,
+                                       InputIterator2 first2,
+                                       OutputIterator result,
+                                       T init,
+                                       BinaryPredicate binary_pred)
+{
+  using thrust::system::detail::generic::exclusive_scan_by_key;
+  return exclusive_scan_by_key(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first1, last1, first2, result, init, binary_pred);
+} // end exclusive_scan_by_key()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename T,
+         typename BinaryPredicate,
+         typename AssociativeOperator>
+  OutputIterator exclusive_scan_by_key(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                       InputIterator1 first1,
+                                       InputIterator1 last1,
+                                       InputIterator2 first2,
+                                       OutputIterator result,
+                                       T init,
+                                       BinaryPredicate binary_pred,
+                                       AssociativeOperator binary_op)
+{
+  using thrust::system::detail::generic::exclusive_scan_by_key;
+  return exclusive_scan_by_key(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first1, last1, first2, result, init, binary_pred, binary_op);
+} // end exclusive_scan_by_key()
+
+
+template<typename InputIterator,
+         typename OutputIterator>
+  OutputIterator inclusive_scan(InputIterator first,
+                                InputIterator last,
+                                OutputIterator result)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator>::type  System1;
+  typedef typename thrust::iterator_system<OutputIterator>::type System2;
+
+  System1 system1;
+  System2 system2;
+
+  return thrust::inclusive_scan(select_system(system1,system2), first, last, result);
+} // end inclusive_scan()
+
+
+template<typename InputIterator,
+         typename OutputIterator,
+         typename BinaryFunction>
+  OutputIterator inclusive_scan(InputIterator first,
+                                InputIterator last,
+                                OutputIterator result,
+                                BinaryFunction binary_op)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator>::type  System1;
+  typedef typename thrust::iterator_system<OutputIterator>::type System2;
+
+  System1 system1;
+  System2 system2;
+
+  return thrust::inclusive_scan(select_system(system1,system2), first, last, result, binary_op);
+} // end inclusive_scan()
+
+
+template<typename InputIterator,
+         typename OutputIterator>
+  OutputIterator exclusive_scan(InputIterator first,
+                                InputIterator last,
+                                OutputIterator result)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator>::type  System1;
+  typedef typename thrust::iterator_system<OutputIterator>::type System2;
+
+  System1 system1;
+  System2 system2;
+
+  return thrust::exclusive_scan(select_system(system1,system2), first, last, result);
+} // end exclusive_scan()
+
+
+template<typename InputIterator,
+         typename OutputIterator,
+         typename T>
+  OutputIterator exclusive_scan(InputIterator first,
+                                InputIterator last,
+                                OutputIterator result,
+                                T init)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator>::type  System1;
+  typedef typename thrust::iterator_system<OutputIterator>::type System2;
+
+  System1 system1;
+  System2 system2;
+
+  return thrust::exclusive_scan(select_system(system1,system2), first, last, result, init);
+} // end exclusive_scan()
+
+
+template<typename InputIterator,
+         typename OutputIterator,
+         typename T,
+         typename BinaryFunction>
+  OutputIterator exclusive_scan(InputIterator first,
+                                InputIterator last,
+                                OutputIterator result,
+                                T init,
+                                BinaryFunction binary_op)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator>::type  System1;
+  typedef typename thrust::iterator_system<OutputIterator>::type System2;
+
+  System1 system1;
+  System2 system2;
+
+  return thrust::exclusive_scan(select_system(system1,system2), first, last, result, init, binary_op);
+} // end exclusive_scan()
+
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator>
+  OutputIterator inclusive_scan_by_key(InputIterator1 first1,
+                                       InputIterator1 last1,
+                                       InputIterator2 first2,
+                                       OutputIterator result)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator1>::type System1;
+  typedef typename thrust::iterator_system<InputIterator2>::type System2;
+  typedef typename thrust::iterator_system<OutputIterator>::type System3;
+
+  System1 system1;
+  System2 system2;
+  System3 system3;
+
+  return thrust::inclusive_scan_by_key(select_system(system1,system2,system3), first1, last1, first2, result);
+}
+
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename BinaryPredicate>
+  OutputIterator inclusive_scan_by_key(InputIterator1 first1,
+                                       InputIterator1 last1,
+                                       InputIterator2 first2,
+                                       OutputIterator result,
+                                       BinaryPredicate binary_pred)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator1>::type System1;
+  typedef typename thrust::iterator_system<InputIterator2>::type System2;
+  typedef typename thrust::iterator_system<OutputIterator>::type System3;
+
+  System1 system1;
+  System2 system2;
+  System3 system3;
+
+  return thrust::inclusive_scan_by_key(select_system(system1,system2,system3), first1, last1, first2, result, binary_pred);
+}
+
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename BinaryPredicate,
+         typename AssociativeOperator>
+  OutputIterator inclusive_scan_by_key(InputIterator1 first1,
+                                       InputIterator1 last1,
+                                       InputIterator2 first2,
+                                       OutputIterator result,
+                                       BinaryPredicate binary_pred,
+                                       AssociativeOperator binary_op)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator1>::type System1;
+  typedef typename thrust::iterator_system<InputIterator2>::type System2;
+  typedef typename thrust::iterator_system<OutputIterator>::type System3;
+
+  System1 system1;
+  System2 system2;
+  System3 system3;
+
+  return thrust::inclusive_scan_by_key(select_system(system1,system2,system3), first1, last1, first2, result, binary_pred, binary_op);
+}
+
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator>
+  OutputIterator exclusive_scan_by_key(InputIterator1 first1,
+                                       InputIterator1 last1,
+                                       InputIterator2 first2,
+                                       OutputIterator result)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator1>::type System1;
+  typedef typename thrust::iterator_system<InputIterator2>::type System2;
+  typedef typename thrust::iterator_system<OutputIterator>::type System3;
+
+  System1 system1;
+  System2 system2;
+  System3 system3;
+
+  return thrust::exclusive_scan_by_key(select_system(system1,system2,system3), first1, last1, first2, result);
+}
+
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename T>
+  OutputIterator exclusive_scan_by_key(InputIterator1 first1,
+                                       InputIterator1 last1,
+                                       InputIterator2 first2,
+                                       OutputIterator result,
+                                       T init)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator1>::type System1;
+  typedef typename thrust::iterator_system<InputIterator2>::type System2;
+  typedef typename thrust::iterator_system<OutputIterator>::type System3;
+
+  System1 system1;
+  System2 system2;
+  System3 system3;
+
+  return thrust::exclusive_scan_by_key(select_system(system1,system2,system3), first1, last1, first2, result, init);
+}
+
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename T,
+         typename BinaryPredicate>
+  OutputIterator exclusive_scan_by_key(InputIterator1 first1,
+                                       InputIterator1 last1,
+                                       InputIterator2 first2,
+                                       OutputIterator result,
+                                       T init,
+                                       BinaryPredicate binary_pred)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator1>::type System1;
+  typedef typename thrust::iterator_system<InputIterator2>::type System2;
+  typedef typename thrust::iterator_system<OutputIterator>::type System3;
+
+  System1 system1;
+  System2 system2;
+  System3 system3;
+
+  return thrust::exclusive_scan_by_key(select_system(system1,system2,system3), first1, last1, first2, result, init, binary_pred);
+}
+
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename T,
+         typename BinaryPredicate,
+         typename AssociativeOperator>
+  OutputIterator exclusive_scan_by_key(InputIterator1 first1,
+                                       InputIterator1 last1,
+                                       InputIterator2 first2,
+                                       OutputIterator result,
+                                       T init,
+                                       BinaryPredicate binary_pred,
+                                       AssociativeOperator binary_op)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator1>::type System1;
+  typedef typename thrust::iterator_system<InputIterator2>::type System2;
+  typedef typename thrust::iterator_system<OutputIterator>::type System3;
+
+  System1 system1;
+  System2 system2;
+  System3 system3;
+
+  return thrust::exclusive_scan_by_key(select_system(system1,system2,system3), first1, last1, first2, result, init, binary_pred, binary_op);
+}
+
+
+} // end namespace thrust
+
diff --git a/compat/thrust/detail/scatter.inl b/compat/thrust/detail/scatter.inl
new file mode 100644
index 0000000000..934addb727
--- /dev/null
+++ b/compat/thrust/detail/scatter.inl
@@ -0,0 +1,159 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file scatter.inl
+ *  \brief Inline file for scatter.h.
+ */
+
+#include <thrust/scatter.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/system/detail/generic/select_system.h>
+#include <thrust/system/detail/generic/scatter.h>
+#include <thrust/system/detail/adl/scatter.h>
+
+namespace thrust
+{
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename RandomAccessIterator>
+  void scatter(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+               InputIterator1 first,
+               InputIterator1 last,
+               InputIterator2 map,
+               RandomAccessIterator output)
+{
+  using thrust::system::detail::generic::scatter;
+  return scatter(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, map, output);
+} // end scatter()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename RandomAccessIterator>
+  void scatter_if(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                  InputIterator1 first,
+                  InputIterator1 last,
+                  InputIterator2 map,
+                  InputIterator3 stencil,
+                  RandomAccessIterator output)
+{
+  using thrust::system::detail::generic::scatter_if;
+  return scatter_if(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, map, stencil, output);
+} // end scatter_if()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename RandomAccessIterator,
+         typename Predicate>
+  void scatter_if(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                  InputIterator1 first,
+                  InputIterator1 last,
+                  InputIterator2 map,
+                  InputIterator3 stencil,
+                  RandomAccessIterator output,
+                  Predicate pred)
+{
+  using thrust::system::detail::generic::scatter_if;
+  return scatter_if(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, map, stencil, output, pred);
+} // end scatter_if()
+
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename RandomAccessIterator>
+  void scatter(InputIterator1 first,
+               InputIterator1 last,
+               InputIterator2 map,
+               RandomAccessIterator output)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator1>::type       System1; 
+  typedef typename thrust::iterator_system<InputIterator2>::type       System2; 
+  typedef typename thrust::iterator_system<RandomAccessIterator>::type System3; 
+
+  System1 system1;
+  System2 system2;
+  System3 system3;
+
+  return thrust::scatter(select_system(system1,system2,system3), first, last, map, output);
+} // end scatter()
+
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename RandomAccessIterator>
+  void scatter_if(InputIterator1 first,
+                  InputIterator1 last,
+                  InputIterator2 map,
+                  InputIterator3 stencil,
+                  RandomAccessIterator output)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator1>::type       System1; 
+  typedef typename thrust::iterator_system<InputIterator2>::type       System2; 
+  typedef typename thrust::iterator_system<InputIterator3>::type       System3; 
+  typedef typename thrust::iterator_system<RandomAccessIterator>::type System4; 
+
+  System1 system1;
+  System2 system2;
+  System3 system3;
+  System4 system4;
+
+  return thrust::scatter_if(select_system(system1,system2,system3,system4), first, last, map, stencil, output);
+} // end scatter_if()
+
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename RandomAccessIterator,
+         typename Predicate>
+  void scatter_if(InputIterator1 first,
+                  InputIterator1 last,
+                  InputIterator2 map,
+                  InputIterator3 stencil,
+                  RandomAccessIterator output,
+                  Predicate pred)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator1>::type       System1; 
+  typedef typename thrust::iterator_system<InputIterator2>::type       System2; 
+  typedef typename thrust::iterator_system<InputIterator3>::type       System3; 
+  typedef typename thrust::iterator_system<RandomAccessIterator>::type System4; 
+
+  System1 system1;
+  System2 system2;
+  System3 system3;
+  System4 system4;
+
+  return thrust::scatter_if(select_system(system1,system2,system3,system4), first, last, map, stencil, output, pred);
+} // end scatter_if()
+
+} // end namespace thrust
+
diff --git a/compat/thrust/detail/sequence.inl b/compat/thrust/detail/sequence.inl
new file mode 100644
index 0000000000..f1741877f8
--- /dev/null
+++ b/compat/thrust/detail/sequence.inl
@@ -0,0 +1,112 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file sequence.inl
+ *  \brief Inline file for sequence.h.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/sequence.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/system/detail/generic/select_system.h>
+#include <thrust/system/detail/generic/sequence.h>
+#include <thrust/system/detail/adl/sequence.h>
+
+namespace thrust
+{
+
+
+template<typename DerivedPolicy, typename ForwardIterator>
+  void sequence(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                ForwardIterator first,
+                ForwardIterator last)
+{
+  using thrust::system::detail::generic::sequence;
+  return sequence(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last);
+} // end sequence()
+
+
+template<typename DerivedPolicy, typename ForwardIterator, typename T>
+  void sequence(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                ForwardIterator first,
+                ForwardIterator last,
+                T init)
+{
+  using thrust::system::detail::generic::sequence;
+  return sequence(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, init);
+} // end sequence()
+
+
+template<typename DerivedPolicy, typename ForwardIterator, typename T>
+  void sequence(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                ForwardIterator first,
+                ForwardIterator last,
+                T init,
+                T step)
+{
+  using thrust::system::detail::generic::sequence;
+  return sequence(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, init, step);
+} // end sequence()
+
+
+template<typename ForwardIterator>
+  void sequence(ForwardIterator first,
+                ForwardIterator last)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<ForwardIterator>::type System;
+
+  System system;
+
+  return thrust::sequence(select_system(system), first, last);
+} // end sequence()
+
+
+template<typename ForwardIterator, typename T>
+  void sequence(ForwardIterator first,
+                ForwardIterator last,
+                T init)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<ForwardIterator>::type System;
+
+  System system;
+
+  return thrust::sequence(select_system(system), first, last, init);
+} // end sequence()
+
+
+template<typename ForwardIterator, typename T>
+  void sequence(ForwardIterator first,
+                ForwardIterator last,
+                T init,
+                T step)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<ForwardIterator>::type System;
+
+  System system;
+
+  return thrust::sequence(select_system(system), first, last, init, step);
+} // end sequence()
+
+
+} // end namespace thrust
+
diff --git a/compat/thrust/detail/set_operations.inl b/compat/thrust/detail/set_operations.inl
new file mode 100644
index 0000000000..daec46156d
--- /dev/null
+++ b/compat/thrust/detail/set_operations.inl
@@ -0,0 +1,836 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+/*! \file set_operations.inl
+ *  \brief Inline file for set_operations.h.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/system/detail/generic/select_system.h>
+#include <thrust/system/detail/generic/set_operations.h>
+#include <thrust/system/detail/adl/set_operations.h>
+
+namespace thrust
+{
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator>
+  OutputIterator set_difference(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                InputIterator1                                              first1,
+                                InputIterator1                                              last1,
+                                InputIterator2                                              first2,
+                                InputIterator2                                              last2,
+                                OutputIterator                                              result)
+{
+  using thrust::system::detail::generic::set_difference;
+  return set_difference(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first1, last1, first2, last2, result);
+} // end set_difference()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename StrictWeakCompare>
+  OutputIterator set_difference(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                InputIterator1                                              first1,
+                                InputIterator1                                              last1,
+                                InputIterator2                                              first2,
+                                InputIterator2                                              last2,
+                                OutputIterator                                              result,
+                                StrictWeakCompare                                           comp)
+{
+  using thrust::system::detail::generic::set_difference;
+  return set_difference(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first1, last1, first2, last2, result, comp);
+} // end set_difference()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename InputIterator4,
+         typename OutputIterator1,
+         typename OutputIterator2>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    set_difference_by_key(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                          InputIterator1                                              keys_first1,
+                          InputIterator1                                              keys_last1,
+                          InputIterator2                                              keys_first2,
+                          InputIterator2                                              keys_last2,
+                          InputIterator3                                              values_first1,
+                          InputIterator4                                              values_first2,
+                          OutputIterator1                                             keys_result,
+                          OutputIterator2                                             values_result)
+{
+  using thrust::system::detail::generic::set_difference_by_key;
+  return set_difference_by_key(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), keys_first1, keys_last1, keys_first2, keys_last2, values_first1, values_first2, keys_result, values_result);
+} // end set_difference_by_key()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename InputIterator4,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename StrictWeakCompare>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    set_difference_by_key(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                          InputIterator1                                              keys_first1,
+                          InputIterator1                                              keys_last1,
+                          InputIterator2                                              keys_first2,
+                          InputIterator2                                              keys_last2,
+                          InputIterator3                                              values_first1,
+                          InputIterator4                                              values_first2,
+                          OutputIterator1                                             keys_result,
+                          OutputIterator2                                             values_result,
+                          StrictWeakCompare                                           comp)
+{
+  using thrust::system::detail::generic::set_difference_by_key;
+  return set_difference_by_key(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), keys_first1, keys_last1, keys_first2, keys_last2, values_first1, values_first2, keys_result, values_result, comp);
+} // end set_difference_by_key()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator>
+  OutputIterator set_intersection(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                  InputIterator1                                              first1,
+                                  InputIterator1                                              last1,
+                                  InputIterator2                                              first2,
+                                  InputIterator2                                              last2,
+                                  OutputIterator                                              result)
+{
+  using thrust::system::detail::generic::set_intersection;
+  return set_intersection(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first1, last1, first2, last2, result);
+} // end set_intersection()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename StrictWeakCompare>
+  OutputIterator set_intersection(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                  InputIterator1                                              first1,
+                                  InputIterator1                                              last1,
+                                  InputIterator2                                              first2,
+                                  InputIterator2                                              last2,
+                                  OutputIterator                                              result,
+                                  StrictWeakCompare                                           comp)
+{
+  using thrust::system::detail::generic::set_intersection;
+  return set_intersection(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first1, last1, first2, last2, result, comp);
+} // end set_intersection()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename OutputIterator1,
+         typename OutputIterator2>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    set_intersection_by_key(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                            InputIterator1                                              keys_first1,
+                            InputIterator1                                              keys_last1,
+                            InputIterator2                                              keys_first2,
+                            InputIterator2                                              keys_last2,
+                            InputIterator3                                              values_first1,
+                            OutputIterator1                                             keys_result,
+                            OutputIterator2                                             values_result)
+{
+  using thrust::system::detail::generic::set_intersection_by_key;
+  return set_intersection_by_key(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), keys_first1, keys_last1, keys_first2, keys_last2, values_first1, keys_result, values_result);
+} // end set_intersection_by_key()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename StrictWeakCompare>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    set_intersection_by_key(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                            InputIterator1                                              keys_first1,
+                            InputIterator1                                              keys_last1,
+                            InputIterator2                                              keys_first2,
+                            InputIterator2                                              keys_last2,
+                            InputIterator3                                              values_first1,
+                            OutputIterator1                                             keys_result,
+                            OutputIterator2                                             values_result,
+                            StrictWeakCompare                                           comp)
+{
+  using thrust::system::detail::generic::set_intersection_by_key;
+  return set_intersection_by_key(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), keys_first1, keys_last1, keys_first2, keys_last2, values_first1, keys_result, values_result, comp);
+} // end set_intersection_by_key()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator>
+  OutputIterator set_symmetric_difference(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                          InputIterator1                                              first1,
+                                          InputIterator1                                              last1,
+                                          InputIterator2                                              first2,
+                                          InputIterator2                                              last2,
+                                          OutputIterator                                              result)
+{
+  using thrust::system::detail::generic::set_symmetric_difference;
+  return set_symmetric_difference(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first1, last1, first2, last2, result);
+} // end set_symmetric_difference()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename StrictWeakCompare>
+  OutputIterator set_symmetric_difference(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                          InputIterator1                                              first1,
+                                          InputIterator1                                              last1,
+                                          InputIterator2                                              first2,
+                                          InputIterator2                                              last2,
+                                          OutputIterator                                              result,
+                                          StrictWeakCompare                                           comp)
+{
+  using thrust::system::detail::generic::set_symmetric_difference;
+  return set_symmetric_difference(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first1, last1, first2, last2, result, comp);
+} // end set_symmetric_difference()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename InputIterator4,
+         typename OutputIterator1,
+         typename OutputIterator2>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    set_symmetric_difference_by_key(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                    InputIterator1                                              keys_first1,
+                                    InputIterator1                                              keys_last1,
+                                    InputIterator2                                              keys_first2,
+                                    InputIterator2                                              keys_last2,
+                                    InputIterator3                                              values_first1,
+                                    InputIterator4                                              values_first2,
+                                    OutputIterator1                                             keys_result,
+                                    OutputIterator2                                             values_result)
+{
+  using thrust::system::detail::generic::set_symmetric_difference_by_key;
+  return set_symmetric_difference_by_key(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), keys_first1, keys_last1, keys_first2, keys_last2, values_first1, values_first2, keys_result, values_result);
+} // end set_symmetric_difference_by_key()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename InputIterator4,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename StrictWeakCompare>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    set_symmetric_difference_by_key(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                    InputIterator1                                              keys_first1,
+                                    InputIterator1                                              keys_last1,
+                                    InputIterator2                                              keys_first2,
+                                    InputIterator2                                              keys_last2,
+                                    InputIterator3                                              values_first1,
+                                    InputIterator4                                              values_first2,
+                                    OutputIterator1                                             keys_result,
+                                    OutputIterator2                                             values_result,
+                                    StrictWeakCompare                                           comp)
+{
+  using thrust::system::detail::generic::set_symmetric_difference_by_key;
+  return set_symmetric_difference_by_key(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), keys_first1, keys_last1, keys_first2, keys_last2, values_first1, values_first2, keys_result, values_result, comp);
+} // end set_symmetric_difference_by_key()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator>
+  OutputIterator set_union(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                           InputIterator1                                              first1,
+                           InputIterator1                                              last1,
+                           InputIterator2                                              first2,
+                           InputIterator2                                              last2,
+                           OutputIterator                                              result)
+{
+  using thrust::system::detail::generic::set_union;
+  return set_union(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first1, last1, first2, last2, result);
+} // end set_union()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename StrictWeakCompare>
+  OutputIterator set_union(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                           InputIterator1                                              first1,
+                           InputIterator1                                              last1,
+                           InputIterator2                                              first2,
+                           InputIterator2                                              last2,
+                           OutputIterator                                              result,
+                           StrictWeakCompare                                           comp)
+{
+  using thrust::system::detail::generic::set_union;
+  return set_union(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first1, last1, first2, last2, result, comp);
+} // end set_union()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename InputIterator4,
+         typename OutputIterator1,
+         typename OutputIterator2>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    set_union_by_key(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                     InputIterator1                                              keys_first1,
+                     InputIterator1                                              keys_last1,
+                     InputIterator2                                              keys_first2,
+                     InputIterator2                                              keys_last2,
+                     InputIterator3                                              values_first1,
+                     InputIterator4                                              values_first2,
+                     OutputIterator1                                             keys_result,
+                     OutputIterator2                                             values_result)
+{
+  using thrust::system::detail::generic::set_union_by_key;
+  return set_union_by_key(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), keys_first1, keys_last1, keys_first2, keys_last2, values_first1, values_first2, keys_result, values_result);
+} // end set_union_by_key()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename InputIterator4,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename StrictWeakCompare>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    set_union_by_key(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                     InputIterator1                                              keys_first1,
+                     InputIterator1                                              keys_last1,
+                     InputIterator2                                              keys_first2,
+                     InputIterator2                                              keys_last2,
+                     InputIterator3                                              values_first1,
+                     InputIterator4                                              values_first2,
+                     OutputIterator1                                             keys_result,
+                     OutputIterator2                                             values_result,
+                     StrictWeakCompare                                           comp)
+{
+  using thrust::system::detail::generic::set_union_by_key;
+  return set_union_by_key(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), keys_first1, keys_last1, keys_first2, keys_last2, values_first1, values_first2, keys_result, values_result, comp);
+} // end set_union_by_key()
+
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename StrictWeakOrdering>
+  OutputIterator set_difference(InputIterator1 first1,
+                                InputIterator1 last1,
+                                InputIterator2 first2,
+                                InputIterator2 last2,
+                                OutputIterator result,
+                                StrictWeakOrdering comp)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator1>::type System1;
+  typedef typename thrust::iterator_system<InputIterator2>::type System2;
+  typedef typename thrust::iterator_system<OutputIterator>::type System3;
+
+  System1 system1;
+  System2 system2;
+  System3 system3;
+
+  return thrust::set_difference(select_system(system1,system2,system3), first1, last1, first2, last2, result, comp);
+} // end set_difference()
+
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator>
+  OutputIterator set_difference(InputIterator1 first1,
+                                InputIterator1 last1,
+                                InputIterator2 first2,
+                                InputIterator2 last2,
+                                OutputIterator result)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator1>::type System1;
+  typedef typename thrust::iterator_system<InputIterator2>::type System2;
+  typedef typename thrust::iterator_system<OutputIterator>::type System3;
+
+  System1 system1;
+  System2 system2;
+  System3 system3;
+
+  return thrust::set_difference(select_system(system1,system2,system3), first1, last1, first2, last2, result);
+} // end set_difference()
+
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename InputIterator4,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename StrictWeakOrdering>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    set_difference_by_key(InputIterator1 keys_first1,
+                          InputIterator1 keys_last1,
+                          InputIterator2 keys_first2,
+                          InputIterator2 keys_last2,
+                          InputIterator3 values_first1,
+                          InputIterator4 values_first2,
+                          OutputIterator1 keys_result,
+                          OutputIterator2 values_result,
+                          StrictWeakOrdering comp)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator1>::type  System1;
+  typedef typename thrust::iterator_system<InputIterator2>::type  System2;
+  typedef typename thrust::iterator_system<InputIterator3>::type  System3;
+  typedef typename thrust::iterator_system<InputIterator4>::type  System4;
+  typedef typename thrust::iterator_system<OutputIterator1>::type System5;
+  typedef typename thrust::iterator_system<OutputIterator2>::type System6;
+
+  System1 system1;
+  System2 system2;
+  System3 system3;
+  System4 system4;
+  System5 system5;
+  System6 system6;
+
+  return thrust::set_difference_by_key(select_system(system1,system2,system3,system4,system5,system6), keys_first1, keys_last1, keys_first2, keys_last2, values_first1, values_first2, keys_result, values_result, comp);
+} // end set_difference_by_key()
+
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename InputIterator4,
+         typename OutputIterator1,
+         typename OutputIterator2>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    set_difference_by_key(InputIterator1 keys_first1,
+                          InputIterator1 keys_last1,
+                          InputIterator2 keys_first2,
+                          InputIterator2 keys_last2,
+                          InputIterator3 values_first1,
+                          InputIterator4 values_first2,
+                          OutputIterator1 keys_result,
+                          OutputIterator2 values_result)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator1>::type  System1;
+  typedef typename thrust::iterator_system<InputIterator2>::type  System2;
+  typedef typename thrust::iterator_system<InputIterator3>::type  System3;
+  typedef typename thrust::iterator_system<InputIterator4>::type  System4;
+  typedef typename thrust::iterator_system<OutputIterator1>::type System5;
+  typedef typename thrust::iterator_system<OutputIterator2>::type System6;
+
+  System1 system1;
+  System2 system2;
+  System3 system3;
+  System4 system4;
+  System5 system5;
+  System6 system6;
+
+  return thrust::set_difference_by_key(select_system(system1,system2,system3,system4,system5,system6), keys_first1, keys_last1, keys_first2, keys_last2, values_first1, values_first2, keys_result, values_result);
+} // end set_difference_by_key()
+
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename StrictWeakOrdering>
+  OutputIterator set_intersection(InputIterator1 first1,
+                                  InputIterator1 last1,
+                                  InputIterator2 first2,
+                                  InputIterator2 last2,
+                                  OutputIterator result,
+                                  StrictWeakOrdering comp)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator1>::type System1;
+  typedef typename thrust::iterator_system<InputIterator2>::type System2;
+  typedef typename thrust::iterator_system<OutputIterator>::type System3;
+
+  System1 system1;
+  System2 system2;
+  System3 system3;
+
+  return thrust::set_intersection(select_system(system1,system2,system3), first1, last1, first2, last2, result, comp);
+} // end set_intersection()
+
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator>
+  OutputIterator set_intersection(InputIterator1 first1,
+                                  InputIterator1 last1,
+                                  InputIterator2 first2,
+                                  InputIterator2 last2,
+                                  OutputIterator result)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator1>::type System1;
+  typedef typename thrust::iterator_system<InputIterator2>::type System2;
+  typedef typename thrust::iterator_system<OutputIterator>::type System3;
+
+  System1 system1;
+  System2 system2;
+  System3 system3;
+
+  return thrust::set_intersection(select_system(system1,system2,system3), first1, last1, first2, last2, result);
+} // end set_intersection()
+
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename StrictWeakOrdering>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    set_intersection_by_key(InputIterator1 keys_first1,
+                            InputIterator1 keys_last1,
+                            InputIterator2 keys_first2,
+                            InputIterator2 keys_last2,
+                            InputIterator3 values_first1,
+                            OutputIterator1 keys_result,
+                            OutputIterator2 values_result,
+                            StrictWeakOrdering comp)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator1>::type  System1;
+  typedef typename thrust::iterator_system<InputIterator2>::type  System2;
+  typedef typename thrust::iterator_system<InputIterator3>::type  System3;
+  typedef typename thrust::iterator_system<OutputIterator1>::type System4;
+  typedef typename thrust::iterator_system<OutputIterator2>::type System5;
+
+  System1 system1;
+  System2 system2;
+  System3 system3;
+  System4 system4;
+  System5 system5;
+
+  return thrust::set_intersection_by_key(select_system(system1,system2,system3,system4,system5), keys_first1, keys_last1, keys_first2, keys_last2, values_first1, keys_result, values_result, comp);
+} // end set_intersection_by_key()
+
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename OutputIterator1,
+         typename OutputIterator2>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    set_intersection_by_key(InputIterator1 keys_first1,
+                            InputIterator1 keys_last1,
+                            InputIterator2 keys_first2,
+                            InputIterator2 keys_last2,
+                            InputIterator3 values_first1,
+                            OutputIterator1 keys_result,
+                            OutputIterator2 values_result)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator1>::type  System1;
+  typedef typename thrust::iterator_system<InputIterator2>::type  System2;
+  typedef typename thrust::iterator_system<InputIterator3>::type  System3;
+  typedef typename thrust::iterator_system<OutputIterator1>::type System4;
+  typedef typename thrust::iterator_system<OutputIterator2>::type System5;
+
+  System1 system1;
+  System2 system2;
+  System3 system3;
+  System4 system4;
+  System5 system5;
+
+  return thrust::set_intersection_by_key(select_system(system1,system2,system3,system4,system5), keys_first1, keys_last1, keys_first2, keys_last2, values_first1, keys_result, values_result);
+} // end set_intersection_by_key()
+
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename StrictWeakOrdering>
+  OutputIterator set_symmetric_difference(InputIterator1 first1,
+                                          InputIterator1 last1,
+                                          InputIterator2 first2,
+                                          InputIterator2 last2,
+                                          OutputIterator result,
+                                          StrictWeakOrdering comp)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator1>::type System1;
+  typedef typename thrust::iterator_system<InputIterator2>::type System2;
+  typedef typename thrust::iterator_system<OutputIterator>::type System3;
+
+  System1 system1;
+  System2 system2;
+  System3 system3;
+
+  return thrust::set_symmetric_difference(select_system(system1,system2,system3), first1, last1, first2, last2, result, comp);
+} // end set_symmetric_difference()
+
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator>
+  OutputIterator set_symmetric_difference(InputIterator1 first1,
+                                          InputIterator1 last1,
+                                          InputIterator2 first2,
+                                          InputIterator2 last2,
+                                          OutputIterator result)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator1>::type System1;
+  typedef typename thrust::iterator_system<InputIterator2>::type System2;
+  typedef typename thrust::iterator_system<OutputIterator>::type System3;
+
+  System1 system1;
+  System2 system2;
+  System3 system3;
+
+  return thrust::set_symmetric_difference(select_system(system1,system2,system3), first1, last1, first2, last2, result);
+} // end set_symmetric_difference()
+
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename InputIterator4,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename StrictWeakOrdering>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    set_symmetric_difference_by_key(InputIterator1 keys_first1,
+                                    InputIterator1 keys_last1,
+                                    InputIterator2 keys_first2,
+                                    InputIterator2 keys_last2,
+                                    InputIterator3 values_first1,
+                                    InputIterator4 values_first2,
+                                    OutputIterator1 keys_result,
+                                    OutputIterator2 values_result,
+                                    StrictWeakOrdering comp)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator1>::type  System1;
+  typedef typename thrust::iterator_system<InputIterator2>::type  System2;
+  typedef typename thrust::iterator_system<InputIterator3>::type  System3;
+  typedef typename thrust::iterator_system<InputIterator4>::type  System4;
+  typedef typename thrust::iterator_system<OutputIterator1>::type System5;
+  typedef typename thrust::iterator_system<OutputIterator2>::type System6;
+
+  System1 system1;
+  System2 system2;
+  System3 system3;
+  System4 system4;
+  System5 system5;
+  System6 system6;
+
+  return thrust::set_symmetric_difference_by_key(select_system(system1,system2,system3,system4,system5,system6), keys_first1, keys_last1, keys_first2, keys_last2, values_first1, values_first2, keys_result, values_result, comp);
+} // end set_symmetric_difference_by_key()
+
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename InputIterator4,
+         typename OutputIterator1,
+         typename OutputIterator2>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    set_symmetric_difference_by_key(InputIterator1 keys_first1,
+                                    InputIterator1 keys_last1,
+                                    InputIterator2 keys_first2,
+                                    InputIterator2 keys_last2,
+                                    InputIterator3 values_first1,
+                                    InputIterator4 values_first2,
+                                    OutputIterator1 keys_result,
+                                    OutputIterator2 values_result)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator1>::type  System1;
+  typedef typename thrust::iterator_system<InputIterator2>::type  System2;
+  typedef typename thrust::iterator_system<InputIterator3>::type  System3;
+  typedef typename thrust::iterator_system<InputIterator4>::type  System4;
+  typedef typename thrust::iterator_system<OutputIterator1>::type System5;
+  typedef typename thrust::iterator_system<OutputIterator2>::type System6;
+
+  System1 system1;
+  System2 system2;
+  System3 system3;
+  System4 system4;
+  System5 system5;
+  System6 system6;
+
+  return thrust::set_symmetric_difference_by_key(select_system(system1,system2,system3,system4,system5,system6), keys_first1, keys_last1, keys_first2, keys_last2, values_first1, values_first2, keys_result, values_result);
+} // end set_symmetric_difference_by_key()
+
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename StrictWeakOrdering>
+  OutputIterator set_union(InputIterator1 first1,
+                           InputIterator1 last1,
+                           InputIterator2 first2,
+                           InputIterator2 last2,
+                           OutputIterator result,
+                           StrictWeakOrdering comp)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator1>::type System1;
+  typedef typename thrust::iterator_system<InputIterator2>::type System2;
+  typedef typename thrust::iterator_system<OutputIterator>::type System3;
+
+  System1 system1;
+  System2 system2;
+  System3 system3;
+
+  return thrust::set_union(select_system(system1,system2,system3), first1, last1, first2, last2, result, comp);
+} // end set_union()
+
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator>
+  OutputIterator set_union(InputIterator1 first1,
+                           InputIterator1 last1,
+                           InputIterator2 first2,
+                           InputIterator2 last2,
+                           OutputIterator result)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator1>::type System1;
+  typedef typename thrust::iterator_system<InputIterator2>::type System2;
+  typedef typename thrust::iterator_system<OutputIterator>::type System3;
+
+  System1 system1;
+  System2 system2;
+  System3 system3;
+
+  return thrust::set_union(select_system(system1,system2,system3), first1, last1, first2, last2, result);
+} // end set_union()
+
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename InputIterator4,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename StrictWeakOrdering>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    set_union_by_key(InputIterator1 keys_first1,
+                     InputIterator1 keys_last1,
+                     InputIterator2 keys_first2,
+                     InputIterator2 keys_last2,
+                     InputIterator3 values_first1,
+                     InputIterator4 values_first2,
+                     OutputIterator1 keys_result,
+                     OutputIterator2 values_result,
+                     StrictWeakOrdering comp)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator1>::type  System1;
+  typedef typename thrust::iterator_system<InputIterator2>::type  System2;
+  typedef typename thrust::iterator_system<InputIterator3>::type  System3;
+  typedef typename thrust::iterator_system<InputIterator4>::type  System4;
+  typedef typename thrust::iterator_system<OutputIterator1>::type System5;
+  typedef typename thrust::iterator_system<OutputIterator2>::type System6;
+
+  System1 system1;
+  System2 system2;
+  System3 system3;
+  System4 system4;
+  System5 system5;
+  System6 system6;
+
+  return thrust::set_union_by_key(select_system(system1,system2,system3,system4,system5,system6), keys_first1, keys_last1, keys_first2, keys_last2, values_first1, values_first2, keys_result, values_result, comp);
+} // end set_union_by_key()
+
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename InputIterator4,
+         typename OutputIterator1,
+         typename OutputIterator2>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    set_union_by_key(InputIterator1 keys_first1,
+                     InputIterator1 keys_last1,
+                     InputIterator2 keys_first2,
+                     InputIterator2 keys_last2,
+                     InputIterator3 values_first1,
+                     InputIterator4 values_first2,
+                     OutputIterator1 keys_result,
+                     OutputIterator2 values_result)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator1>::type  System1;
+  typedef typename thrust::iterator_system<InputIterator2>::type  System2;
+  typedef typename thrust::iterator_system<InputIterator3>::type  System3;
+  typedef typename thrust::iterator_system<InputIterator4>::type  System4;
+  typedef typename thrust::iterator_system<OutputIterator1>::type System5;
+  typedef typename thrust::iterator_system<OutputIterator2>::type System6;
+
+  System1 system1;
+  System2 system2;
+  System3 system3;
+  System4 system4;
+  System5 system5;
+  System6 system6;
+
+  return thrust::set_union_by_key(select_system(system1,system2,system3,system4,system5,system6), keys_first1, keys_last1, keys_first2, keys_last2, values_first1, values_first2, keys_result, values_result);
+} // end set_union_by_key()
+
+
+} // end thrust
+
diff --git a/compat/thrust/detail/sort.inl b/compat/thrust/detail/sort.inl
new file mode 100644
index 0000000000..08be55a8ce
--- /dev/null
+++ b/compat/thrust/detail/sort.inl
@@ -0,0 +1,383 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file sort.inl
+ *  \brief Inline file for sort.h.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/sort.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/system/detail/generic/select_system.h>
+#include <thrust/system/detail/generic/sort.h>
+#include <thrust/system/detail/adl/sort.h>
+
+namespace thrust
+{
+
+
+template<typename DerivedPolicy, typename RandomAccessIterator>
+  void sort(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+            RandomAccessIterator first,
+            RandomAccessIterator last)
+{
+  using thrust::system::detail::generic::sort;
+  return sort(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last);
+} // end sort()
+
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator,
+         typename StrictWeakOrdering>
+  void sort(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+            RandomAccessIterator first,
+            RandomAccessIterator last,
+            StrictWeakOrdering comp)
+{
+  using thrust::system::detail::generic::sort;
+  return sort(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, comp);
+} // end sort()
+
+
+template<typename DerivedPolicy, typename RandomAccessIterator>
+  void stable_sort(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                   RandomAccessIterator first,
+                   RandomAccessIterator last)
+{
+  using thrust::system::detail::generic::stable_sort;
+  return stable_sort(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last);
+} // end stable_sort()
+
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator,
+         typename StrictWeakOrdering>
+  void stable_sort(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                   RandomAccessIterator first,
+                   RandomAccessIterator last,
+                   StrictWeakOrdering comp)
+{
+  using thrust::system::detail::generic::stable_sort;
+  return stable_sort(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, comp);
+} // end stable_sort()
+
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2>
+  void sort_by_key(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                   RandomAccessIterator1 keys_first,
+                   RandomAccessIterator1 keys_last,
+                   RandomAccessIterator2 values_first)
+{
+  using thrust::system::detail::generic::sort_by_key;
+  return sort_by_key(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), keys_first, keys_last, values_first);
+} // end sort_by_key()
+
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2,
+         typename StrictWeakOrdering>
+  void sort_by_key(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                   RandomAccessIterator1 keys_first,
+                   RandomAccessIterator1 keys_last,
+                   RandomAccessIterator2 values_first,
+                   StrictWeakOrdering comp)
+{
+  using thrust::system::detail::generic::sort_by_key;
+  return sort_by_key(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), keys_first, keys_last, values_first, comp);
+} // end sort_by_key()
+
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2>
+  void stable_sort_by_key(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                          RandomAccessIterator1 keys_first,
+                          RandomAccessIterator1 keys_last,
+                          RandomAccessIterator2 values_first)
+{
+  using thrust::system::detail::generic::stable_sort_by_key;
+  return stable_sort_by_key(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), keys_first, keys_last, values_first);
+} // end stable_sort_by_key()
+
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2,
+         typename StrictWeakOrdering>
+  void stable_sort_by_key(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                          RandomAccessIterator1 keys_first,
+                          RandomAccessIterator1 keys_last,
+                          RandomAccessIterator2 values_first,
+                          StrictWeakOrdering comp)
+{
+  using thrust::system::detail::generic::stable_sort_by_key;
+  return stable_sort_by_key(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), keys_first, keys_last, values_first, comp);
+} // end stable_sort_by_key()
+
+
+template<typename DerivedPolicy, typename ForwardIterator>
+  bool is_sorted(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                 ForwardIterator first,
+                 ForwardIterator last)
+{
+  using thrust::system::detail::generic::is_sorted;
+  return is_sorted(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last);
+} // end is_sorted()
+
+
+template<typename DerivedPolicy, typename ForwardIterator, typename Compare>
+  bool is_sorted(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                 ForwardIterator first,
+                 ForwardIterator last,
+                 Compare comp)
+{
+  using thrust::system::detail::generic::is_sorted;
+  return is_sorted(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, comp);
+} // end is_sorted()
+
+
+template<typename DerivedPolicy, typename ForwardIterator>
+  ForwardIterator is_sorted_until(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                  ForwardIterator first,
+                                  ForwardIterator last)
+{
+  using thrust::system::detail::generic::is_sorted_until;
+  return is_sorted_until(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last);
+} // end is_sorted_until()
+
+
+template<typename DerivedPolicy, typename ForwardIterator, typename Compare>
+  ForwardIterator is_sorted_until(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                  ForwardIterator first,
+                                  ForwardIterator last,
+                                  Compare comp)
+{
+  using thrust::system::detail::generic::is_sorted_until;
+  return is_sorted_until(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, comp);
+} // end is_sorted_until()
+
+
+///////////////
+// Key Sorts //
+///////////////
+
+template<typename RandomAccessIterator>
+  void sort(RandomAccessIterator first,
+            RandomAccessIterator last)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<RandomAccessIterator>::type System;
+
+  System system;
+
+  return thrust::sort(select_system(system), first, last);
+} // end sort()
+
+
+template<typename RandomAccessIterator,
+         typename StrictWeakOrdering>
+  void sort(RandomAccessIterator first,
+            RandomAccessIterator last,
+            StrictWeakOrdering comp)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<RandomAccessIterator>::type System;
+
+  System system;
+
+  return thrust::sort(select_system(system), first, last, comp);
+} // end sort()
+
+
+template<typename RandomAccessIterator>
+  void stable_sort(RandomAccessIterator first,
+                   RandomAccessIterator last)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<RandomAccessIterator>::type System;
+
+  System system;
+
+  return thrust::stable_sort(select_system(system), first, last);
+} // end stable_sort() 
+
+
+template<typename RandomAccessIterator,
+         typename StrictWeakOrdering>
+  void stable_sort(RandomAccessIterator first,
+                   RandomAccessIterator last,
+                   StrictWeakOrdering comp)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<RandomAccessIterator>::type System;
+
+  System system;
+
+  return thrust::stable_sort(select_system(system), first, last, comp);
+} // end stable_sort()
+
+
+
+/////////////////////
+// Key-Value Sorts //
+/////////////////////
+
+template<typename RandomAccessIterator1,
+         typename RandomAccessIterator2>
+  void sort_by_key(RandomAccessIterator1 keys_first,
+                   RandomAccessIterator1 keys_last,
+                   RandomAccessIterator2 values_first)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<RandomAccessIterator1>::type System1;
+  typedef typename thrust::iterator_system<RandomAccessIterator2>::type System2;
+
+  System1 system1;
+  System2 system2;
+
+  return thrust::sort_by_key(select_system(system1,system2), keys_first, keys_last, values_first);
+} // end sort_by_key()
+
+
+template<typename RandomAccessIterator1,
+         typename RandomAccessIterator2,
+         typename StrictWeakOrdering>
+  void sort_by_key(RandomAccessIterator1 keys_first,
+                   RandomAccessIterator1 keys_last,
+                   RandomAccessIterator2 values_first,
+                   StrictWeakOrdering comp)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<RandomAccessIterator1>::type System1;
+  typedef typename thrust::iterator_system<RandomAccessIterator2>::type System2;
+
+  System1 system1;
+  System2 system2;
+
+  return thrust::sort_by_key(select_system(system1,system2), keys_first, keys_last, values_first, comp);
+} // end sort_by_key()
+
+
+template<typename RandomAccessIterator1,
+         typename RandomAccessIterator2>
+  void stable_sort_by_key(RandomAccessIterator1 keys_first,
+                          RandomAccessIterator1 keys_last,
+                          RandomAccessIterator2 values_first)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<RandomAccessIterator1>::type System1;
+  typedef typename thrust::iterator_system<RandomAccessIterator2>::type System2;
+
+  System1 system1;
+  System2 system2;
+
+  return thrust::stable_sort_by_key(select_system(system1,system2), keys_first, keys_last, values_first);
+} // end stable_sort_by_key()
+
+
+template<typename RandomAccessIterator1,
+         typename RandomAccessIterator2,
+         typename StrictWeakOrdering>
+  void stable_sort_by_key(RandomAccessIterator1 keys_first,
+                          RandomAccessIterator1 keys_last,
+                          RandomAccessIterator2 values_first,
+                          StrictWeakOrdering comp)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<RandomAccessIterator1>::type System1;
+  typedef typename thrust::iterator_system<RandomAccessIterator2>::type System2;
+
+  System1 system1;
+  System2 system2;
+
+  return thrust::stable_sort_by_key(select_system(system1,system2), keys_first, keys_last, values_first, comp);
+} // end stable_sort_by_key()
+
+
+template<typename ForwardIterator>
+  bool is_sorted(ForwardIterator first,
+                 ForwardIterator last)
+{
+  using thrust::system::detail::generic::select_system;
+  
+  typedef typename thrust::iterator_system<ForwardIterator>::type System;
+
+  System system;
+
+  return thrust::is_sorted(select_system(system), first, last);
+} // end is_sorted()
+
+
+template<typename ForwardIterator,
+         typename Compare>
+  bool is_sorted(ForwardIterator first,
+                 ForwardIterator last,
+                 Compare comp)
+{
+  using thrust::system::detail::generic::select_system;
+  
+  typedef typename thrust::iterator_system<ForwardIterator>::type System;
+
+  System system;
+
+  return thrust::is_sorted(select_system(system), first, last, comp);
+} // end is_sorted()
+
+
+template<typename ForwardIterator>
+  ForwardIterator is_sorted_until(ForwardIterator first,
+                                  ForwardIterator last)
+{
+  using thrust::system::detail::generic::select_system;
+  
+  typedef typename thrust::iterator_system<ForwardIterator>::type System;
+
+  System system;
+
+  return thrust::is_sorted_until(select_system(system), first, last);
+} // end is_sorted_until()
+
+
+template<typename ForwardIterator,
+         typename Compare>
+  ForwardIterator is_sorted_until(ForwardIterator first,
+                                  ForwardIterator last,
+                                  Compare comp)
+{
+  using thrust::system::detail::generic::select_system;
+  
+  typedef typename thrust::iterator_system<ForwardIterator>::type System;
+
+  System system;
+
+  return thrust::is_sorted_until(select_system(system), first, last, comp);
+} // end is_sorted_until()
+
+
+} // end namespace thrust
+
diff --git a/compat/thrust/detail/static_assert.h b/compat/thrust/detail/static_assert.h
new file mode 100644
index 0000000000..ccc084286c
--- /dev/null
+++ b/compat/thrust/detail/static_assert.h
@@ -0,0 +1,71 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/type_traits.h>
+
+/*
+ * (C) Copyright John Maddock 2000.
+ * 
+ * Distributed under the Boost Software License, Version 1.0.
+ * (See accompanying NOTICE file for the complete license)
+ *
+ * For more information, see http://www.boost.org
+ */
+
+//
+// Helper macro THRUST_JOIN (based on BOOST_JOIN):
+// The following piece of macro magic joins the two
+// arguments together, even when one of the arguments is
+// itself a macro (see 16.3.1 in C++ standard).  The key
+// is that macro expansion of macro arguments does not
+// occur in THRUST_DO_JOIN2 but does in THRUST_DO_JOIN.
+//
+#define THRUST_JOIN( X, Y ) THRUST_DO_JOIN( X, Y )
+#define THRUST_DO_JOIN( X, Y ) THRUST_DO_JOIN2(X,Y)
+#define THRUST_DO_JOIN2( X, Y ) X##Y
+
+namespace thrust
+{
+
+namespace detail
+{
+
+// HP aCC cannot deal with missing names for template value parameters
+template <bool x> struct STATIC_ASSERTION_FAILURE;
+
+template <> struct STATIC_ASSERTION_FAILURE<true> { enum { value = 1 }; };
+
+// HP aCC cannot deal with missing names for template value parameters
+template<int x> struct static_assert_test{};
+
+template<typename, bool x>
+  struct depend_on_instantiation
+{
+  static const bool value = x;
+};
+
+} // end detail
+
+} // end thrust
+
+#define THRUST_STATIC_ASSERT( B ) \
+   typedef ::thrust::detail::static_assert_test<\
+      sizeof(::thrust::detail::STATIC_ASSERTION_FAILURE< (bool)( B ) >)>\
+         THRUST_JOIN(thrust_static_assert_typedef_, __LINE__)
+
diff --git a/compat/thrust/detail/swap.h b/compat/thrust/detail/swap.h
new file mode 100644
index 0000000000..9f82ac247a
--- /dev/null
+++ b/compat/thrust/detail/swap.h
@@ -0,0 +1,35 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+namespace thrust
+{
+
+template<typename Assignable1, typename Assignable2>
+__host__ __device__
+inline void swap(Assignable1 &a, Assignable2 &b)
+{
+  Assignable1 temp = a;
+  a = b;
+  b = temp;
+} // end swap()
+
+} // end namespace thrust
+
diff --git a/compat/thrust/detail/swap.inl b/compat/thrust/detail/swap.inl
new file mode 100644
index 0000000000..eafd70ae6a
--- /dev/null
+++ b/compat/thrust/detail/swap.inl
@@ -0,0 +1,21 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+#include <thrust/swap.h>
+#include <thrust/detail/swap.h>
+#include <thrust/detail/swap_ranges.inl>
+
diff --git a/compat/thrust/detail/swap_ranges.inl b/compat/thrust/detail/swap_ranges.inl
new file mode 100644
index 0000000000..e3b06deb01
--- /dev/null
+++ b/compat/thrust/detail/swap_ranges.inl
@@ -0,0 +1,64 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file swap_ranges.inl
+ *  \brief Inline file for swap_ranges.h.
+ */
+
+#include <thrust/swap.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/system/detail/generic/select_system.h>
+#include <thrust/system/detail/generic/swap_ranges.h>
+#include <thrust/system/detail/adl/swap_ranges.h>
+
+namespace thrust
+{
+
+
+template<typename DerivedPolicy,
+         typename ForwardIterator1,
+         typename ForwardIterator2>
+  ForwardIterator2 swap_ranges(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                               ForwardIterator1 first1,
+                               ForwardIterator1 last1,
+                               ForwardIterator2 first2)
+{
+  using thrust::system::detail::generic::swap_ranges;
+  return swap_ranges(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first1, last1, first2);
+} // end swap_ranges()
+
+
+template<typename ForwardIterator1,
+         typename ForwardIterator2>
+  ForwardIterator2 swap_ranges(ForwardIterator1 first1,
+                               ForwardIterator1 last1,
+                               ForwardIterator2 first2)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<ForwardIterator1>::type System1;
+  typedef typename thrust::iterator_system<ForwardIterator2>::type System2;
+
+  System1 system1;
+  System2 system2;
+
+  return thrust::swap_ranges(select_system(system1,system2), first1, last1, first2);
+} // end swap_ranges()
+
+
+} // end namespace thrust
+
diff --git a/compat/thrust/detail/tabulate.inl b/compat/thrust/detail/tabulate.inl
new file mode 100644
index 0000000000..961c76e9f6
--- /dev/null
+++ b/compat/thrust/detail/tabulate.inl
@@ -0,0 +1,55 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/tabulate.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/system/detail/generic/select_system.h>
+#include <thrust/system/detail/generic/tabulate.h>
+#include <thrust/system/detail/adl/tabulate.h>
+
+namespace thrust
+{
+
+
+template<typename DerivedPolicy, typename ForwardIterator, typename UnaryOperation>
+  void tabulate(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                ForwardIterator first,
+                ForwardIterator last,
+                UnaryOperation unary_op)
+{
+  using thrust::system::detail::generic::tabulate;
+  return tabulate(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, unary_op);
+} // end tabulate()
+
+
+template<typename ForwardIterator, typename UnaryOperation>
+  void tabulate(ForwardIterator first,
+                ForwardIterator last,
+                UnaryOperation unary_op)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<ForwardIterator>::type System;
+
+  System system;
+
+  return thrust::tabulate(select_system(system), first, last, unary_op);
+} // end tabulate()
+
+
+} // end namespace thrust
+
diff --git a/compat/thrust/detail/temporary_array.h b/compat/thrust/detail/temporary_array.h
new file mode 100644
index 0000000000..3a9e08481c
--- /dev/null
+++ b/compat/thrust/detail/temporary_array.h
@@ -0,0 +1,158 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+/*! \file temporary_array.h
+ *  \brief Container-like class temporary storage inside algorithms.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/iterator/retag.h>
+#include <thrust/detail/contiguous_storage.h>
+#include <thrust/detail/allocator/temporary_allocator.h>
+#include <thrust/detail/allocator/no_throw_allocator.h>
+#include <memory>
+
+namespace thrust
+{
+namespace detail
+{
+
+
+template<typename T, typename System>
+  class temporary_array
+    : public contiguous_storage<
+               T,
+               no_throw_allocator<
+                 temporary_allocator<T,System>
+               >
+             >
+{
+  private:
+    typedef contiguous_storage<
+      T,
+      no_throw_allocator<
+        temporary_allocator<T,System>
+      >
+    > super_t;
+
+    // to help out the constructor
+    typedef no_throw_allocator<temporary_allocator<T,System> > alloc_type;
+
+  public:
+    typedef typename super_t::size_type size_type;
+
+    temporary_array(thrust::execution_policy<System> &system, size_type n);
+
+    // provide a kill-switch to explicitly avoid initialization
+    temporary_array(int uninit, thrust::execution_policy<System> &system, size_type n);
+
+    template<typename InputIterator>
+    temporary_array(thrust::execution_policy<System> &system,
+                    InputIterator first,
+                    size_type n);
+
+    template<typename InputIterator, typename InputSystem>
+    temporary_array(thrust::execution_policy<System> &system,
+                    thrust::execution_policy<InputSystem> &input_system,
+                    InputIterator first,
+                    size_type n);
+
+    template<typename InputIterator>
+    temporary_array(thrust::execution_policy<System> &system,
+                    InputIterator first,
+                    InputIterator last);
+
+    template<typename InputSystem, typename InputIterator>
+    temporary_array(thrust::execution_policy<System> &system,
+                    thrust::execution_policy<InputSystem> &input_system,
+                    InputIterator first,
+                    InputIterator last);
+
+    ~temporary_array();
+}; // end temporary_array
+
+
+// XXX eliminate this when we do ranges for real
+template<typename Iterator, typename System>
+  class tagged_iterator_range
+{
+  public:
+    typedef thrust::detail::tagged_iterator<Iterator,System> iterator;
+
+    template<typename Ignored1, typename Ignored2>
+    tagged_iterator_range(const Ignored1 &, const Ignored2 &, Iterator first, Iterator last)
+      : m_begin(reinterpret_tag<System>(first)),
+        m_end(reinterpret_tag<System>(last))
+    {}
+
+    iterator begin(void) const { return m_begin; }
+    iterator end(void) const { return m_end; }
+
+  private:
+    iterator m_begin, m_end;
+};
+
+
+// if FromSystem is convertible to ToSystem, then just make a shallow
+// copy of the range. else, use a temporary_array
+// note that the resulting iterator is explicitly tagged with ToSystem either way
+template<typename Iterator, typename FromSystem, typename ToSystem>
+  struct move_to_system_base
+    : public eval_if<
+        is_convertible<
+          FromSystem,
+          ToSystem
+        >::value,
+        identity_<
+          tagged_iterator_range<Iterator,ToSystem>
+        >,
+        identity_<
+          temporary_array<
+            typename thrust::iterator_value<Iterator>::type,
+            ToSystem
+          >
+        >
+      >
+{};
+
+
+template<typename Iterator, typename FromSystem, typename ToSystem>
+  class move_to_system
+    : public move_to_system_base<
+        Iterator,
+        FromSystem,
+        ToSystem
+      >::type
+{
+  typedef typename move_to_system_base<Iterator,FromSystem,ToSystem>::type super_t;
+
+  public:
+    move_to_system(thrust::execution_policy<FromSystem> &from_system,
+                   thrust::execution_policy<ToSystem> &to_system,
+                   Iterator first,
+                   Iterator last)
+      : super_t(to_system, from_system, first, last) {}
+};
+
+
+} // end detail
+} // end thrust
+
+#include <thrust/detail/temporary_array.inl>
+
diff --git a/compat/thrust/detail/temporary_array.inl b/compat/thrust/detail/temporary_array.inl
new file mode 100644
index 0000000000..36ed16736e
--- /dev/null
+++ b/compat/thrust/detail/temporary_array.inl
@@ -0,0 +1,148 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/detail/temporary_array.h>
+#include <thrust/distance.h>
+#include <thrust/system/detail/generic/select_system.h>
+#include <thrust/detail/type_traits.h>
+
+
+namespace thrust
+{
+
+namespace detail
+{
+namespace temporary_array_detail
+{
+
+
+template<typename T> struct avoid_initialization : thrust::detail::has_trivial_copy_constructor<T> {};
+
+
+template<typename T, typename TemporaryArray, typename Size>
+typename thrust::detail::enable_if<
+  avoid_initialization<T>::value
+>::type
+  construct_values(TemporaryArray &,
+                   Size)
+{
+  // avoid the overhead of initialization
+} // end construct_values()
+
+
+template<typename T, typename TemporaryArray, typename Size>
+typename thrust::detail::disable_if<
+  avoid_initialization<T>::value
+>::type
+  construct_values(TemporaryArray &a,
+                   Size n)
+{
+  a.default_construct_n(a.begin(), n);
+} // end construct_values()
+
+
+} // end temporary_array_detail
+
+
+template<typename T, typename System>
+  temporary_array<T,System>
+    ::temporary_array(thrust::execution_policy<System> &system, size_type n)
+      :super_t(n, alloc_type(temporary_allocator<T,System>(system)))
+{
+  temporary_array_detail::construct_values<T>(*this, n);
+} // end temporary_array::temporary_array()
+
+
+template<typename T, typename System>
+  temporary_array<T,System>
+    ::temporary_array(int, thrust::execution_policy<System> &system, size_type n)
+      :super_t(n, alloc_type(temporary_allocator<T,System>(system)))
+{
+  // avoid initialization
+  ;
+} // end temporary_array::temporary_array()
+
+
+template<typename T, typename System>
+  template<typename InputIterator>
+    temporary_array<T,System>
+      ::temporary_array(thrust::execution_policy<System> &system,
+                        InputIterator first,
+                        size_type n)
+        : super_t(alloc_type(temporary_allocator<T,System>(system)))
+{
+  super_t::allocate(n);
+
+  super_t::uninitialized_copy_n(system, first, n, super_t::begin());
+} // end temporary_array::temporary_array()
+
+
+template<typename T, typename System>
+  template<typename InputIterator, typename InputSystem>
+    temporary_array<T,System>
+      ::temporary_array(thrust::execution_policy<System> &system,
+                        thrust::execution_policy<InputSystem> &input_system,
+                        InputIterator first,
+                        size_type n)
+        : super_t(alloc_type(temporary_allocator<T,System>(system)))
+{
+  super_t::allocate(n);
+
+  super_t::uninitialized_copy_n(input_system, first, n, super_t::begin());
+} // end temporary_array::temporary_array()
+
+
+template<typename T, typename System>
+  template<typename InputIterator>
+    temporary_array<T,System>
+      ::temporary_array(thrust::execution_policy<System> &system,
+                        InputIterator first,
+                        InputIterator last)
+        : super_t(alloc_type(temporary_allocator<T,System>(system)))
+{
+  super_t::allocate(thrust::distance(first,last));
+
+  super_t::uninitialized_copy(system, first, last, super_t::begin());
+} // end temporary_array::temporary_array()
+
+
+template<typename T, typename System>
+  template<typename InputSystem, typename InputIterator>
+    temporary_array<T,System>
+      ::temporary_array(thrust::execution_policy<System> &system,
+                        thrust::execution_policy<InputSystem> &input_system,
+                        InputIterator first,
+                        InputIterator last)
+        : super_t(alloc_type(temporary_allocator<T,System>(system)))
+{
+  super_t::allocate(thrust::distance(first,last));
+
+  super_t::uninitialized_copy(input_system, first, last, super_t::begin());
+} // end temporary_array::temporary_array()
+
+
+template<typename T, typename System>
+  temporary_array<T,System>
+    ::~temporary_array()
+{
+  // note that super_t::destroy will ignore trivial destructors automatically
+  super_t::destroy(super_t::begin(), super_t::end());
+} // end temporary_array::~temporary_array()
+
+} // end detail
+
+} // end thrust
+
diff --git a/compat/thrust/detail/temporary_buffer.h b/compat/thrust/detail/temporary_buffer.h
new file mode 100644
index 0000000000..046a3b363a
--- /dev/null
+++ b/compat/thrust/detail/temporary_buffer.h
@@ -0,0 +1,71 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/execution_policy.h>
+#include <thrust/pair.h>
+#include <thrust/detail/pointer.h>
+#include <thrust/detail/raw_pointer_cast.h>
+#include <thrust/system/detail/generic/temporary_buffer.h>
+#include <thrust/system/detail/adl/temporary_buffer.h>
+
+namespace thrust
+{
+namespace detail
+{
+namespace get_temporary_buffer_detail
+{
+
+
+template<typename T, typename DerivedPolicy, typename Pair>
+  thrust::pair<thrust::pointer<T,DerivedPolicy>, typename thrust::pointer<T,DerivedPolicy>::difference_type>
+    down_cast_pair(Pair p)
+{
+  // XXX should use a hypothetical thrust::static_pointer_cast here
+  thrust::pointer<T,DerivedPolicy> ptr = thrust::pointer<T,DerivedPolicy>(static_cast<T*>(thrust::raw_pointer_cast(p.first)));
+
+  typedef thrust::pair<thrust::pointer<T,DerivedPolicy>, typename thrust::pointer<T,DerivedPolicy>::difference_type> result_type;
+  return result_type(ptr, p.second);
+} // end down_cast_pair()
+
+
+} // end get_temporary_buffer_detail
+} // end detail
+
+
+template<typename T, typename DerivedPolicy>
+  thrust::pair<thrust::pointer<T,DerivedPolicy>, typename thrust::pointer<T,DerivedPolicy>::difference_type>
+    get_temporary_buffer(const thrust::detail::execution_policy_base<DerivedPolicy> &exec, typename thrust::pointer<T,DerivedPolicy>::difference_type n)
+{
+  using thrust::system::detail::generic::get_temporary_buffer;
+
+  return thrust::detail::get_temporary_buffer_detail::down_cast_pair<T,DerivedPolicy>(get_temporary_buffer<T>(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), n));
+} // end get_temporary_buffer()
+
+
+template<typename DerivedPolicy, typename Pointer>
+  void return_temporary_buffer(const thrust::detail::execution_policy_base<DerivedPolicy> &exec, Pointer p)
+{
+  using thrust::system::detail::generic::return_temporary_buffer;
+
+  return return_temporary_buffer(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), p);
+} // end return_temporary_buffer()
+
+
+} // end thrust
+
diff --git a/compat/thrust/detail/transform.inl b/compat/thrust/detail/transform.inl
new file mode 100644
index 0000000000..ae303bcc0d
--- /dev/null
+++ b/compat/thrust/detail/transform.inl
@@ -0,0 +1,239 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file transform.inl
+ *  \brief Inline file for transform.h.
+ */
+
+#include <thrust/transform.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/system/detail/generic/select_system.h>
+#include <thrust/system/detail/generic/transform.h>
+#include <thrust/system/detail/adl/transform.h>
+
+namespace thrust
+{
+
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator,
+         typename UnaryFunction>
+  OutputIterator transform(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                           InputIterator first, InputIterator last,
+                           OutputIterator result,
+                           UnaryFunction op)
+{
+  using thrust::system::detail::generic::transform;
+  return transform(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, result, op);
+} // end transform()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename BinaryFunction>
+  OutputIterator transform(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                           InputIterator1 first1, InputIterator1 last1,
+                           InputIterator2 first2,
+                           OutputIterator result,
+                           BinaryFunction op)
+{
+  using thrust::system::detail::generic::transform;
+  return transform(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first1, last1, first2, result, op);
+} // end transform()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename ForwardIterator,
+         typename UnaryFunction,
+         typename Predicate>
+  ForwardIterator transform_if(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                               InputIterator first, InputIterator last,
+                               ForwardIterator result,
+                               UnaryFunction op,
+                               Predicate pred)
+{
+  using thrust::system::detail::generic::transform_if;
+  return transform_if(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, result, op, pred);
+} // end transform_if()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename ForwardIterator,
+         typename UnaryFunction,
+         typename Predicate>
+  ForwardIterator transform_if(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                               InputIterator1 first, InputIterator1 last,
+                               InputIterator2 stencil,
+                               ForwardIterator result,
+                               UnaryFunction op,
+                               Predicate pred)
+{
+  using thrust::system::detail::generic::transform_if;
+  return transform_if(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, stencil, result, op, pred);
+} // end transform_if()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename ForwardIterator,
+         typename BinaryFunction,
+         typename Predicate>
+  ForwardIterator transform_if(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                               InputIterator1 first1, InputIterator1 last1,
+                               InputIterator2 first2,
+                               InputIterator3 stencil,
+                               ForwardIterator result,
+                               BinaryFunction binary_op,
+                               Predicate pred)
+{
+  using thrust::system::detail::generic::transform_if;
+  return transform_if(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first1, last1, first2, stencil, result, binary_op, pred);
+} // end transform_if()
+
+
+template<typename InputIterator,
+         typename OutputIterator,
+         typename UnaryFunction>
+  OutputIterator transform(InputIterator first,
+                           InputIterator last,
+                           OutputIterator result,
+                           UnaryFunction op)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator>::type  System1;
+  typedef typename thrust::iterator_system<OutputIterator>::type System2;
+
+  System1 system1;
+  System2 system2;
+
+  return thrust::transform(select_system(system1,system2), first, last, result, op);
+} // end transform()
+
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename BinaryFunction>
+  OutputIterator transform(InputIterator1 first1,
+                           InputIterator1 last1,
+                           InputIterator2 first2,
+                           OutputIterator result,
+                           BinaryFunction op)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator1>::type System1;
+  typedef typename thrust::iterator_system<InputIterator2>::type System2;
+  typedef typename thrust::iterator_system<OutputIterator>::type System3;
+
+  System1 system1;
+  System2 system2;
+  System3 system3;
+
+  return thrust::transform(select_system(system1,system2,system3), first1, last1, first2, result, op);
+} // end transform()
+
+
+template<typename InputIterator,
+         typename ForwardIterator,
+         typename UnaryFunction,
+         typename Predicate>
+  ForwardIterator transform_if(InputIterator first,
+                               InputIterator last,
+                               ForwardIterator result,
+                               UnaryFunction unary_op,
+                               Predicate pred)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator>::type   System1;
+  typedef typename thrust::iterator_system<ForwardIterator>::type System2;
+
+  System1 system1;
+  System2 system2;
+
+  return thrust::transform_if(select_system(system1,system2), first, last, result, unary_op, pred);
+} // end transform_if()
+
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename ForwardIterator,
+         typename UnaryFunction,
+         typename Predicate>
+  ForwardIterator transform_if(InputIterator1 first,
+                               InputIterator1 last,
+                               InputIterator2 stencil,
+                               ForwardIterator result,
+                               UnaryFunction unary_op,
+                               Predicate pred)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator1>::type  System1;
+  typedef typename thrust::iterator_system<InputIterator2>::type  System2;
+  typedef typename thrust::iterator_system<ForwardIterator>::type System3;
+
+  System1 system1;
+  System2 system2;
+  System3 system3;
+
+  return thrust::transform_if(select_system(system1,system2,system3), first, last, stencil, result, unary_op, pred);
+} // end transform_if()
+
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename ForwardIterator,
+         typename BinaryFunction,
+         typename Predicate>
+  ForwardIterator transform_if(InputIterator1 first1,
+                               InputIterator1 last1,
+                               InputIterator2 first2,
+                               InputIterator3 stencil,
+                               ForwardIterator result,
+                               BinaryFunction binary_op,
+                               Predicate pred)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator1>::type  System1;
+  typedef typename thrust::iterator_system<InputIterator2>::type  System2;
+  typedef typename thrust::iterator_system<InputIterator3>::type  System3;
+  typedef typename thrust::iterator_system<ForwardIterator>::type System4;
+
+  System1 system1;
+  System2 system2;
+  System3 system3;
+  System4 system4;
+
+  return thrust::transform_if(select_system(system1,system2,system3,system4), first1, last1, first2, stencil, result, binary_op, pred);
+} // end transform_if()
+
+
+} // end namespace thrust
+
diff --git a/compat/thrust/detail/transform_reduce.inl b/compat/thrust/detail/transform_reduce.inl
new file mode 100644
index 0000000000..ede65030b1
--- /dev/null
+++ b/compat/thrust/detail/transform_reduce.inl
@@ -0,0 +1,70 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file transform_reduce.inl
+ *  \brief Inline file for transform_reduce.h.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/system/detail/generic/select_system.h>
+#include <thrust/system/detail/generic/transform_reduce.h>
+#include <thrust/system/detail/adl/transform_reduce.h>
+
+namespace thrust
+{
+
+
+template<typename DerivedPolicy,
+         typename InputIterator, 
+         typename UnaryFunction, 
+         typename OutputType,
+         typename BinaryFunction>
+  OutputType transform_reduce(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                              InputIterator first,
+                              InputIterator last,
+                              UnaryFunction unary_op,
+                              OutputType init,
+                              BinaryFunction binary_op)
+{
+  using thrust::system::detail::generic::transform_reduce;
+  return transform_reduce(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, unary_op, init, binary_op);
+} // end transform_reduce()
+
+
+template<typename InputIterator, 
+         typename UnaryFunction, 
+         typename OutputType,
+         typename BinaryFunction>
+  OutputType transform_reduce(InputIterator first,
+                              InputIterator last,
+                              UnaryFunction unary_op,
+                              OutputType init,
+                              BinaryFunction binary_op)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator>::type System;
+
+  System system;
+
+  return thrust::transform_reduce(select_system(system), first, last, unary_op, init, binary_op);
+} // end transform_reduce()
+
+
+} // end namespace thrust
+
diff --git a/compat/thrust/detail/transform_scan.inl b/compat/thrust/detail/transform_scan.inl
new file mode 100644
index 0000000000..0187c4b98d
--- /dev/null
+++ b/compat/thrust/detail/transform_scan.inl
@@ -0,0 +1,115 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file transform_scan.inl
+ *  \brief Inline file for transform_scan.h.
+ */
+
+#include <thrust/scan.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/system/detail/generic/select_system.h>
+#include <thrust/system/detail/generic/transform_scan.h>
+#include <thrust/system/detail/adl/transform_scan.h>
+
+namespace thrust
+{
+
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator,
+         typename UnaryFunction,
+         typename AssociativeOperator>
+  OutputIterator transform_inclusive_scan(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                          InputIterator first,
+                                          InputIterator last,
+                                          OutputIterator result,
+                                          UnaryFunction unary_op,
+                                          AssociativeOperator binary_op)
+{
+  using thrust::system::detail::generic::transform_inclusive_scan;
+  return transform_inclusive_scan(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, result, unary_op, binary_op);
+} // end transform_inclusive_scan()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator,
+         typename UnaryFunction,
+         typename T,
+         typename AssociativeOperator>
+  OutputIterator transform_exclusive_scan(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                          InputIterator first,
+                                          InputIterator last,
+                                          OutputIterator result,
+                                          UnaryFunction unary_op,
+                                          T init,
+                                          AssociativeOperator binary_op)
+{
+  using thrust::system::detail::generic::transform_exclusive_scan;
+  return transform_exclusive_scan(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, result, unary_op, init, binary_op);
+} // end transform_exclusive_scan()
+
+
+template<typename InputIterator,
+         typename OutputIterator,
+         typename UnaryFunction,
+         typename BinaryFunction>
+  OutputIterator transform_inclusive_scan(InputIterator first,
+                                          InputIterator last,
+                                          OutputIterator result,
+                                          UnaryFunction unary_op,
+                                          BinaryFunction binary_op)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator>::type  System1;
+  typedef typename thrust::iterator_system<OutputIterator>::type System2;
+
+  System1 system1;
+  System2 system2;
+
+  return thrust::transform_inclusive_scan(select_system(system1,system2), first, last, result, unary_op, binary_op);
+} // end transform_inclusive_scan()
+
+
+template<typename InputIterator,
+         typename OutputIterator,
+         typename UnaryFunction,
+         typename T,
+         typename AssociativeOperator>
+  OutputIterator transform_exclusive_scan(InputIterator first,
+                                          InputIterator last,
+                                          OutputIterator result,
+                                          UnaryFunction unary_op,
+                                          T init,
+                                          AssociativeOperator binary_op)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator>::type  System1;
+  typedef typename thrust::iterator_system<OutputIterator>::type System2;
+
+  System1 system1;
+  System2 system2;
+
+  return thrust::transform_exclusive_scan(select_system(system1,system2), first, last, result, unary_op, init, binary_op);
+} // end transform_exclusive_scan()
+
+
+} // end namespace thrust
+
diff --git a/compat/thrust/detail/trivial_sequence.h b/compat/thrust/detail/trivial_sequence.h
new file mode 100644
index 0000000000..cc7e32be27
--- /dev/null
+++ b/compat/thrust/detail/trivial_sequence.h
@@ -0,0 +1,87 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+/*! \file trivial_sequence.h
+ *  \brief Container-like class for wrapping sequences.  The wrapped
+ *         sequence always has trivial iterators, even when the input
+ *         sequence does not.
+ */
+
+
+#pragma once
+
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/detail/type_traits.h>
+#include <thrust/detail/execution_policy.h>
+#include <thrust/detail/temporary_array.h>
+
+namespace thrust
+{
+
+namespace detail
+{
+
+// never instantiated
+template<typename Iterator, typename DerivedPolicy, typename is_trivial> struct _trivial_sequence { };
+
+// trivial case
+template<typename Iterator, typename DerivedPolicy>
+struct _trivial_sequence<Iterator, DerivedPolicy, thrust::detail::true_type>
+{
+    typedef Iterator iterator_type;
+    Iterator first, last;
+
+    _trivial_sequence(thrust::execution_policy<DerivedPolicy> &, Iterator _first, Iterator _last) : first(_first), last(_last)
+    {
+//        std::cout << "trivial case" << std::endl;
+    }
+
+    iterator_type begin() { return first; }
+    iterator_type end()   { return last; }
+};
+
+// non-trivial case
+template<typename Iterator, typename DerivedPolicy>
+struct _trivial_sequence<Iterator, DerivedPolicy, thrust::detail::false_type>
+{
+    typedef typename thrust::iterator_value<Iterator>::type iterator_value;
+    typedef typename thrust::detail::temporary_array<iterator_value, DerivedPolicy>::iterator iterator_type;
+    
+    thrust::detail::temporary_array<iterator_value, DerivedPolicy> buffer;
+
+    _trivial_sequence(thrust::execution_policy<DerivedPolicy> &exec, Iterator first, Iterator last)
+      : buffer(exec, first, last)
+    {
+//        std::cout << "non-trivial case" << std::endl;
+    }
+
+    iterator_type begin() { return buffer.begin(); }
+    iterator_type end()   { return buffer.end(); }
+};
+
+template <typename Iterator, typename DerivedPolicy>
+struct trivial_sequence
+  : detail::_trivial_sequence<Iterator, DerivedPolicy, typename thrust::detail::is_trivial_iterator<Iterator>::type>
+{
+    typedef _trivial_sequence<Iterator, DerivedPolicy, typename thrust::detail::is_trivial_iterator<Iterator>::type> super_t;
+
+    trivial_sequence(thrust::execution_policy<DerivedPolicy> &exec, Iterator first, Iterator last) : super_t(exec, first, last) { }
+};
+
+} // end namespace detail
+
+} // end namespace thrust
+
diff --git a/compat/thrust/detail/tuple.inl b/compat/thrust/detail/tuple.inl
new file mode 100644
index 0000000000..067ad636c4
--- /dev/null
+++ b/compat/thrust/detail/tuple.inl
@@ -0,0 +1,948 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/detail/type_traits.h>
+#include <thrust/detail/swap.h>
+
+namespace thrust
+{
+
+// define null_type
+struct null_type {};
+
+// null_type comparisons
+__host__ __device__ inline
+bool operator==(const null_type&, const null_type&) { return true; }
+
+__host__ __device__ inline
+bool operator>=(const null_type&, const null_type&) { return true; }
+
+__host__ __device__ inline
+bool operator<=(const null_type&, const null_type&) { return true; }
+
+__host__ __device__ inline
+bool operator!=(const null_type&, const null_type&) { return false; }
+
+__host__ __device__ inline
+bool operator<(const null_type&, const null_type&) { return false; }
+
+__host__ __device__ inline
+bool operator>(const null_type&, const null_type&) { return false; }
+
+// forward declaration for tuple
+template <
+  class T0 = null_type, class T1 = null_type, class T2 = null_type,
+  class T3 = null_type, class T4 = null_type, class T5 = null_type,
+  class T6 = null_type, class T7 = null_type, class T8 = null_type,
+  class T9 = null_type>
+class tuple;
+
+// forward declaration of tuple_element
+template<int i, typename T> struct tuple_element;
+
+// specializations for tuple_element
+template<class T>
+  struct tuple_element<0,T>
+{
+  typedef typename T::head_type type;
+}; // end tuple_element<0,T>
+
+template<int N, class T>
+  struct tuple_element<N, const T>
+{
+  private:
+    typedef typename T::tail_type Next;
+    typedef typename tuple_element<N-1, Next>::type unqualified_type;
+
+  public:
+    typedef typename thrust::detail::add_const<unqualified_type>::type type;
+}; // end tuple_element<N, const T>
+
+template<class T>
+  struct tuple_element<0,const T>
+{
+  typedef typename thrust::detail::add_const<typename T::head_type>::type type;
+}; // end tuple_element<0,const T>
+
+
+
+// forward declaration of tuple_size
+template<class T> struct tuple_size;
+
+// specializations for tuple_size
+template<>
+  struct tuple_size< tuple<> >
+{
+  static const int value = 0;
+}; // end tuple_size< tuple<> >
+
+template<>
+  struct tuple_size<null_type>
+{
+  static const int value = 0;
+}; // end tuple_size<null_type>
+
+
+
+// forward declaration of detail::cons
+namespace detail
+{
+
+template <class HT, class TT> struct cons;
+
+} // end detail
+
+
+// -- some traits classes for get functions
+template <class T> struct access_traits
+{
+  typedef const T& const_type;
+  typedef T& non_const_type;
+
+  typedef const typename thrust::detail::remove_cv<T>::type& parameter_type;
+
+// used as the tuple constructors parameter types
+// Rationale: non-reference tuple element types can be cv-qualified.
+// It should be possible to initialize such types with temporaries,
+// and when binding temporaries to references, the reference must
+// be non-volatile and const. 8.5.3. (5)
+}; // end access_traits
+
+template <class T> struct access_traits<T&>
+{
+  typedef T& const_type;
+  typedef T& non_const_type;
+
+  typedef T& parameter_type;
+}; // end access_traits<T&>
+
+// forward declarations of get()
+template<int N, class HT, class TT>
+__host__ __device__
+inline typename access_traits<
+                  typename tuple_element<N, detail::cons<HT, TT> >::type
+                >::non_const_type
+// XXX we probably don't need to do this for any compiler we care about -jph
+//get(cons<HT, TT>& c BOOST_APPEND_EXPLICIT_TEMPLATE_NON_TYPE(int, N));
+get(detail::cons<HT, TT>& c);
+
+template<int N, class HT, class TT>
+__host__ __device__
+inline typename access_traits<
+                  typename tuple_element<N, detail::cons<HT, TT> >::type
+                >::const_type
+// XXX we probably don't need to do this for any compiler we care about -jph
+//get(const cons<HT, TT>& c BOOST_APPEND_EXPLICIT_TEMPLATE_NON_TYPE(int, N));
+get(const detail::cons<HT, TT>& c);
+
+namespace detail
+{
+
+// -- generate error template, referencing to non-existing members of this
+// template is used to produce compilation errors intentionally
+template<class T>
+class generate_error;
+
+// - cons getters --------------------------------------------------------
+// called: get_class<N>::get<RETURN_TYPE>(aTuple)
+
+template< int N >
+struct get_class
+{
+  template<class RET, class HT, class TT >
+  __host__ __device__
+  inline static RET get(const cons<HT, TT>& t)
+  {
+    // XXX we may not need to deal with this for any compiler we care about -jph
+    //return get_class<N-1>::BOOST_NESTED_TEMPLATE get<RET>(t.tail);
+    return get_class<N-1>::template get<RET>(t.tail);
+    
+    // gcc 4.3 couldn't compile this:
+    //return get_class<N-1>::get<RET>(t.tail);
+  }
+
+  template<class RET, class HT, class TT >
+  __host__ __device__
+  inline static RET get(cons<HT, TT>& t)
+  {
+    // XXX we may not need to deal with this for any compiler we care about -jph
+    //return get_class<N-1>::BOOST_NESTED_TEMPLATE get<RET>(t.tail);
+    return get_class<N-1>::template get<RET>(t.tail);
+
+    // gcc 4.3 couldn't compile this:
+    //return get_class<N-1>::get<RET>(t.tail);
+  }
+}; // end get_class
+
+template<>
+struct get_class<0>
+{
+  template<class RET, class HT, class TT>
+  __host__ __device__
+  inline static RET get(const cons<HT, TT>& t)
+  {
+    return t.head;
+  }
+
+  template<class RET, class HT, class TT>
+  __host__ __device__
+  inline static RET get(cons<HT, TT>& t)
+  {
+    return t.head;
+  }
+}; // get get_class<0>
+
+
+template <bool If, class Then, class Else> struct IF
+{
+  typedef Then RET;
+};
+
+template <class Then, class Else> struct IF<false, Then, Else>
+{
+  typedef Else RET;
+};
+
+//  These helper templates wrap void types and plain function types.
+//  The rationale is to allow one to write tuple types with those types
+//  as elements, even though it is not possible to instantiate such object.
+//  E.g: typedef tuple<void> some_type; // ok
+//  but: some_type x; // fails
+
+template <class T> class non_storeable_type
+{
+  __host__ __device__
+  non_storeable_type();
+};
+
+template <class T> struct wrap_non_storeable_type
+{
+  // XXX is_function looks complicated; punt for now -jph
+  //typedef typename IF<
+  //  ::thrust::detail::is_function<T>::value, non_storeable_type<T>, T
+  //>::RET type;
+
+  typedef T type;
+};
+
+template <> struct wrap_non_storeable_type<void>
+{
+  typedef non_storeable_type<void> type;
+};
+
+
+template <class HT, class TT>
+  struct cons
+{
+  typedef HT head_type;
+  typedef TT tail_type;
+
+  typedef typename
+    wrap_non_storeable_type<head_type>::type stored_head_type;
+
+  stored_head_type head;
+  tail_type tail;
+
+  inline __host__ __device__
+  typename access_traits<stored_head_type>::non_const_type
+  get_head() { return head; }
+
+  inline __host__ __device__
+  typename access_traits<tail_type>::non_const_type
+  get_tail() { return tail; }
+
+  inline __host__ __device__
+  typename access_traits<stored_head_type>::const_type
+  get_head() const { return head; }
+
+  inline __host__ __device__
+  typename access_traits<tail_type>::const_type
+  get_tail() const { return tail; }
+
+  inline __host__ __device__
+  cons(void) : head(), tail() {}
+  //  cons() : head(detail::default_arg<HT>::f()), tail() {}
+
+  // the argument for head is not strictly needed, but it prevents
+  // array type elements. This is good, since array type elements
+  // cannot be supported properly in any case (no assignment,
+  // copy works only if the tails are exactly the same type, ...)
+
+  inline __host__ __device__
+  cons(typename access_traits<stored_head_type>::parameter_type h,
+       const tail_type& t)
+    : head (h), tail(t) {}
+
+  template <class T1, class T2, class T3, class T4, class T5,
+            class T6, class T7, class T8, class T9, class T10>
+  inline __host__ __device__
+  cons( T1& t1, T2& t2, T3& t3, T4& t4, T5& t5,
+        T6& t6, T7& t7, T8& t8, T9& t9, T10& t10 )
+    : head (t1),
+      tail (t2, t3, t4, t5, t6, t7, t8, t9, t10, static_cast<const null_type&>(null_type()))
+      {}
+
+  template <class T2, class T3, class T4, class T5,
+            class T6, class T7, class T8, class T9, class T10>
+  inline __host__ __device__
+  cons( const null_type& /*t1*/, T2& t2, T3& t3, T4& t4, T5& t5,
+        T6& t6, T7& t7, T8& t8, T9& t9, T10& t10 )
+    : head (),
+      tail (t2, t3, t4, t5, t6, t7, t8, t9, t10, static_cast<const null_type&>(null_type()))
+      {}
+
+
+  template <class HT2, class TT2>
+  inline __host__ __device__
+  cons( const cons<HT2, TT2>& u ) : head(u.head), tail(u.tail) {}
+
+  template <class HT2, class TT2>
+  inline __host__ __device__
+  cons& operator=( const cons<HT2, TT2>& u ) {
+    head=u.head; tail=u.tail; return *this;
+  }
+
+  // must define assignment operator explicitly, implicit version is
+  // illformed if HT is a reference (12.8. (12))
+  inline __host__ __device__
+  cons& operator=(const cons& u) {
+    head = u.head; tail = u.tail;  return *this;
+  }
+
+  // XXX enable when we support std::pair -jph
+  //template <class T1, class T2>
+  //__host__ __device__
+  //cons& operator=( const std::pair<T1, T2>& u ) {
+  //  //BOOST_STATIC_ASSERT(length<cons>::value == 2); // check length = 2
+  //  head = u.first; tail.head = u.second; return *this;
+  //}
+
+  // get member functions (non-const and const)
+  template <int N>
+  __host__ __device__
+  typename access_traits<
+             typename tuple_element<N, cons<HT, TT> >::type
+           >::non_const_type
+  get() {
+    return thrust::get<N>(*this); // delegate to non-member get
+  }
+
+  template <int N>
+  __host__ __device__
+  typename access_traits<
+             typename tuple_element<N, cons<HT, TT> >::type
+           >::const_type
+  get() const {
+    return thrust::get<N>(*this); // delegate to non-member get
+  }
+
+  inline __host__ __device__
+  void swap(cons &c)
+  {
+    using thrust::swap;
+
+    swap(head, c.head);
+    tail.swap(c.tail);
+  }
+};
+
+template <class HT>
+  struct cons<HT, null_type>
+{
+  typedef HT head_type;
+  typedef null_type tail_type;
+  typedef cons<HT, null_type> self_type;
+
+  typedef typename
+    wrap_non_storeable_type<head_type>::type stored_head_type;
+  stored_head_type head;
+
+  typename access_traits<stored_head_type>::non_const_type
+  inline __host__ __device__
+  get_head() { return head; }
+
+  inline __host__ __device__
+  null_type get_tail() { return null_type(); }
+
+  inline __host__ __device__
+  typename access_traits<stored_head_type>::const_type
+  get_head() const { return head; }
+
+  inline __host__ __device__
+  null_type get_tail() const { return null_type(); }
+
+  inline __host__ __device__
+  cons() : head() {}
+
+  inline __host__ __device__
+  cons(typename access_traits<stored_head_type>::parameter_type h,
+       const null_type& = null_type())
+    : head (h) {}
+
+  template<class T1>
+  inline __host__ __device__
+  cons(T1& t1, const null_type&, const null_type&, const null_type&,
+       const null_type&, const null_type&, const null_type&,
+       const null_type&, const null_type&, const null_type&)
+  : head (t1) {}
+
+  inline __host__ __device__
+  cons(const null_type&,
+       const null_type&, const null_type&, const null_type&,
+       const null_type&, const null_type&, const null_type&,
+       const null_type&, const null_type&, const null_type&)
+  : head () {}
+
+  template <class HT2>
+  inline __host__ __device__
+  cons( const cons<HT2, null_type>& u ) : head(u.head) {}
+
+  template <class HT2>
+  inline __host__ __device__
+  cons& operator=(const cons<HT2, null_type>& u )
+  {
+    head = u.head;
+    return *this;
+  }
+
+  // must define assignment operator explicitly, implicit version
+  // is illformed if HT is a reference
+  inline __host__ __device__
+  cons& operator=(const cons& u) { head = u.head; return *this; }
+
+  template <int N>
+  inline __host__ __device__
+  typename access_traits<
+             typename tuple_element<N, self_type>::type
+            >::non_const_type
+  // XXX we probably don't need this for the compilers we care about -jph
+  //get(BOOST_EXPLICIT_TEMPLATE_NON_TYPE(int, N))
+  get(void)
+  {
+    return thrust::get<N>(*this);
+  }
+
+  template <int N>
+  inline __host__ __device__
+  typename access_traits<
+             typename tuple_element<N, self_type>::type
+           >::const_type
+  // XXX we probably don't need this for the compilers we care about -jph
+  //get(BOOST_EXPLICIT_TEMPLATE_NON_TYPE(int, N)) const
+  get(void) const
+  {
+    return thrust::get<N>(*this);
+  }
+
+  inline __host__ __device__
+  void swap(cons &c)
+  {
+    using thrust::swap;
+
+    swap(head, c.head);
+  }
+}; // end cons
+
+template <class T0, class T1, class T2, class T3, class T4,
+          class T5, class T6, class T7, class T8, class T9>
+  struct map_tuple_to_cons
+{
+  typedef cons<T0,
+               typename map_tuple_to_cons<T1, T2, T3, T4, T5,
+                                          T6, T7, T8, T9, null_type>::type
+              > type;
+}; // end map_tuple_to_cons
+
+// The empty tuple is a null_type
+template <>
+  struct map_tuple_to_cons<null_type, null_type, null_type, null_type, null_type, null_type, null_type, null_type, null_type, null_type>
+{
+  typedef null_type type;
+}; // end map_tuple_to_cons<...>
+
+
+
+// ---------------------------------------------------------------------------
+// The call_traits for make_tuple
+
+// Must be instantiated with plain or const plain types (not with references)
+
+// from template<class T> foo(const T& t) : make_tuple_traits<const T>::type
+// from template<class T> foo(T& t) : make_tuple_traits<T>::type
+
+// Conversions:
+// T -> T,
+// references -> compile_time_error
+// array -> const ref array
+
+
+template<class T>
+struct make_tuple_traits {
+  typedef T type;
+
+  // commented away, see below  (JJ)
+  //  typedef typename IF<
+  //  boost::is_function<T>::value,
+  //  T&,
+  //  T>::RET type;
+
+};
+
+// The is_function test was there originally for plain function types,
+// which can't be stored as such (we must either store them as references or
+// pointers). Such a type could be formed if make_tuple was called with a
+// reference to a function.
+// But this would mean that a const qualified function type was formed in
+// the make_tuple function and hence make_tuple can't take a function
+// reference as a parameter, and thus T can't be a function type.
+// So is_function test was removed.
+// (14.8.3. says that type deduction fails if a cv-qualified function type
+// is created. (It only applies for the case of explicitly specifying template
+// args, though?)) (JJ)
+
+template<class T>
+struct make_tuple_traits<T&> {
+  typedef typename
+     detail::generate_error<T&>::
+       do_not_use_with_reference_type error;
+};
+
+// Arrays can't be stored as plain types; convert them to references.
+// All arrays are converted to const. This is because make_tuple takes its
+// parameters as const T& and thus the knowledge of the potential
+// non-constness of actual argument is lost.
+template<class T, int n>  struct make_tuple_traits <T[n]> {
+  typedef const T (&type)[n];
+};
+
+template<class T, int n>
+struct make_tuple_traits<const T[n]> {
+  typedef const T (&type)[n];
+};
+
+template<class T, int n>  struct make_tuple_traits<volatile T[n]> {
+  typedef const volatile T (&type)[n];
+};
+
+template<class T, int n>
+struct make_tuple_traits<const volatile T[n]> {
+  typedef const volatile T (&type)[n];
+};
+
+// XXX enable these if we ever care about reference_wrapper -jph
+//template<class T>
+//struct make_tuple_traits<reference_wrapper<T> >{
+//  typedef T& type;
+//};
+//
+//template<class T>
+//struct make_tuple_traits<const reference_wrapper<T> >{
+//  typedef T& type;
+//};
+
+
+// a helper traits to make the make_tuple functions shorter (Vesa Karvonen's
+// suggestion)
+template <
+  class T0 = null_type, class T1 = null_type, class T2 = null_type,
+  class T3 = null_type, class T4 = null_type, class T5 = null_type,
+  class T6 = null_type, class T7 = null_type, class T8 = null_type,
+  class T9 = null_type
+>
+struct make_tuple_mapper {
+  typedef
+    tuple<typename make_tuple_traits<T0>::type,
+          typename make_tuple_traits<T1>::type,
+          typename make_tuple_traits<T2>::type,
+          typename make_tuple_traits<T3>::type,
+          typename make_tuple_traits<T4>::type,
+          typename make_tuple_traits<T5>::type,
+          typename make_tuple_traits<T6>::type,
+          typename make_tuple_traits<T7>::type,
+          typename make_tuple_traits<T8>::type,
+          typename make_tuple_traits<T9>::type> type;
+};
+
+} // end detail
+
+
+template<int N, class HT, class TT>
+__host__ __device__
+inline typename access_traits<
+                  typename tuple_element<N, detail::cons<HT, TT> >::type
+                >::non_const_type
+get(detail::cons<HT, TT>& c)
+{
+  //return detail::get_class<N>::BOOST_NESTED_TEMPLATE
+  
+  // gcc 4.3 couldn't compile this:
+  //return detail::get_class<N>::
+
+  return detail::get_class<N>::template
+         get<
+           typename access_traits<
+             typename tuple_element<N, detail::cons<HT, TT> >::type
+           >::non_const_type,
+           HT,TT
+         >(c);
+}
+
+
+// get function for const cons-lists, returns a const reference to
+// the element. If the element is a reference, returns the reference
+// as such (that is, can return a non-const reference)
+template<int N, class HT, class TT>
+__host__ __device__
+inline typename access_traits<
+                  typename tuple_element<N, detail::cons<HT, TT> >::type
+                >::const_type
+get(const detail::cons<HT, TT>& c)
+{
+  //return detail::get_class<N>::BOOST_NESTED_TEMPLATE
+
+  // gcc 4.3 couldn't compile this:
+  //return detail::get_class<N>::
+
+  return detail::get_class<N>::template
+         get<
+           typename access_traits<
+             typename tuple_element<N, detail::cons<HT, TT> >::type
+           >::const_type,
+           HT,TT
+         >(c);
+}
+
+
+template<class T0>
+__host__ __device__ inline
+  typename detail::make_tuple_mapper<T0>::type
+    make_tuple(const T0& t0)
+{
+  typedef typename detail::make_tuple_mapper<T0>::type t;
+  return t(t0);
+} // end make_tuple()
+
+template<class T0, class T1>
+__host__ __device__ inline
+  typename detail::make_tuple_mapper<T0, T1>::type
+    make_tuple(const T0& t0, const T1& t1)
+{
+  typedef typename detail::make_tuple_mapper<T0,T1>::type t;
+  return t(t0,t1);
+} // end make_tuple()
+
+template<class T0, class T1, class T2>
+__host__ __device__ inline
+  typename detail::make_tuple_mapper<T0, T1, T2>::type
+    make_tuple(const T0& t0, const T1& t1, const T2& t2)
+{
+  typedef typename detail::make_tuple_mapper<T0,T1,T2>::type t;
+  return t(t0,t1,t2);
+} // end make_tuple()
+
+template<class T0, class T1, class T2, class T3>
+__host__ __device__ inline
+  typename detail::make_tuple_mapper<T0, T1, T2, T3>::type
+    make_tuple(const T0& t0, const T1& t1, const T2& t2, const T3& t3)
+{
+  typedef typename detail::make_tuple_mapper<T0,T1,T2,T3>::type t;
+  return t(t0,t1,t2,t3);
+} // end make_tuple()
+
+template<class T0, class T1, class T2, class T3, class T4>
+__host__ __device__ inline
+  typename detail::make_tuple_mapper<T0, T1, T2, T3, T4>::type
+    make_tuple(const T0& t0, const T1& t1, const T2& t2, const T3& t3, const T4& t4)
+{
+  typedef typename detail::make_tuple_mapper<T0,T1,T2,T3,T4>::type t;
+  return t(t0,t1,t2,t3,t4);
+} // end make_tuple()
+
+template<class T0, class T1, class T2, class T3, class T4, class T5>
+__host__ __device__ inline
+  typename detail::make_tuple_mapper<T0, T1, T2, T3, T4, T5>::type
+    make_tuple(const T0& t0, const T1& t1, const T2& t2, const T3& t3, const T4& t4, const T5& t5)
+{
+  typedef typename detail::make_tuple_mapper<T0,T1,T2,T3,T4,T5>::type t;
+  return t(t0,t1,t2,t3,t4,t5);
+} // end make_tuple()
+
+template<class T0, class T1, class T2, class T3, class T4, class T5, class T6>
+__host__ __device__ inline
+  typename detail::make_tuple_mapper<T0, T1, T2, T3, T4, T5, T6>::type
+    make_tuple(const T0& t0, const T1& t1, const T2& t2, const T3& t3, const T4& t4, const T5& t5, const T6& t6)
+{
+  typedef typename detail::make_tuple_mapper<T0,T1,T2,T3,T4,T5,T6>::type t;
+  return t(t0,t1,t2,t3,t4,t5,t6);
+} // end make_tuple()
+
+template<class T0, class T1, class T2, class T3, class T4, class T5, class T6, class T7>
+__host__ __device__ inline
+  typename detail::make_tuple_mapper<T0, T1, T2, T3, T4, T5, T6, T7>::type
+    make_tuple(const T0& t0, const T1& t1, const T2& t2, const T3& t3, const T4& t4, const T5& t5, const T6& t6, const T7& t7)
+{
+  typedef typename detail::make_tuple_mapper<T0,T1,T2,T3,T4,T5,T6,T7>::type t;
+  return t(t0,t1,t2,t3,t4,t5,t6,t7);
+} // end make_tuple()
+
+template<class T0, class T1, class T2, class T3, class T4, class T5, class T6, class T7, class T8>
+__host__ __device__ inline
+  typename detail::make_tuple_mapper<T0, T1, T2, T3, T4, T5, T6, T7, T8>::type
+    make_tuple(const T0& t0, const T1& t1, const T2& t2, const T3& t3, const T4& t4, const T5& t5, const T6& t6, const T7& t7, const T8& t8)
+{
+  typedef typename detail::make_tuple_mapper<T0,T1,T2,T3,T4,T5,T6,T7,T8>::type t;
+  return t(t0,t1,t2,t3,t4,t5,t6,t7,t8);
+} // end make_tuple()
+
+template<class T0, class T1, class T2, class T3, class T4, class T5, class T6, class T7, class T8, class T9>
+__host__ __device__ inline
+  typename detail::make_tuple_mapper<T0, T1, T2, T3, T4, T5, T6, T7, T8, T9>::type
+    make_tuple(const T0& t0, const T1& t1, const T2& t2, const T3& t3, const T4& t4, const T5& t5, const T6& t6, const T7& t7, const T8& t8, const T9& t9)
+{
+  typedef typename detail::make_tuple_mapper<T0,T1,T2,T3,T4,T5,T6,T7,T8,T9>::type t;
+  return t(t0,t1,t2,t3,t4,t5,t6,t7,t8,t9);
+} // end make_tuple()
+
+
+template<typename T0>
+__host__ __device__ inline
+tuple<T0&> tie(T0 &t0)
+{
+  return tuple<T0&>(t0);
+}
+
+template<typename T0,typename T1>
+__host__ __device__ inline
+tuple<T0&,T1&> tie(T0 &t0, T1 &t1)
+{
+  return tuple<T0&,T1&>(t0,t1);
+}
+
+template<typename T0,typename T1, typename T2>
+__host__ __device__ inline
+tuple<T0&,T1&,T2&> tie(T0 &t0, T1 &t1, T2 &t2)
+{
+  return tuple<T0&,T1&,T2&>(t0,t1,t2);
+}
+
+template<typename T0,typename T1, typename T2, typename T3>
+__host__ __device__ inline
+tuple<T0&,T1&,T2&,T3&> tie(T0 &t0, T1 &t1, T2 &t2, T3 &t3)
+{
+  return tuple<T0&,T1&,T2&,T3&>(t0,t1,t2,t3);
+}
+
+template<typename T0,typename T1, typename T2, typename T3, typename T4>
+__host__ __device__ inline
+tuple<T0&,T1&,T2&,T3&,T4&> tie(T0 &t0, T1 &t1, T2 &t2, T3 &t3, T4 &t4)
+{
+  return tuple<T0&,T1&,T2&,T3&,T4&>(t0,t1,t2,t3,t4);
+}
+
+template<typename T0,typename T1, typename T2, typename T3, typename T4, typename T5>
+__host__ __device__ inline
+tuple<T0&,T1&,T2&,T3&,T4&,T5&> tie(T0 &t0, T1 &t1, T2 &t2, T3 &t3, T4 &t4, T5 &t5)
+{
+  return tuple<T0&,T1&,T2&,T3&,T4&,T5&>(t0,t1,t2,t3,t4,t5);
+}
+
+template<typename T0,typename T1, typename T2, typename T3, typename T4, typename T5, typename T6>
+__host__ __device__ inline
+tuple<T0&,T1&,T2&,T3&,T4&,T5&,T6&> tie(T0 &t0, T1 &t1, T2 &t2, T3 &t3, T4 &t4, T5 &t5, T6 &t6)
+{
+  return tuple<T0&,T1&,T2&,T3&,T4&,T5&,T6&>(t0,t1,t2,t3,t4,t5,t6);
+}
+
+template<typename T0,typename T1, typename T2, typename T3, typename T4, typename T5, typename T6, typename T7>
+__host__ __device__ inline
+tuple<T0&,T1&,T2&,T3&,T4&,T5&,T6&,T7&> tie(T0 &t0, T1 &t1, T2 &t2, T3 &t3, T4 &t4, T5 &t5, T6 &t6, T7 &t7)
+{
+  return tuple<T0&,T1&,T2&,T3&,T4&,T5&,T6&,T7&>(t0,t1,t2,t3,t4,t5,t6,t7);
+}
+
+template<typename T0,typename T1, typename T2, typename T3, typename T4, typename T5, typename T6, typename T7, typename T8>
+__host__ __device__ inline
+tuple<T0&,T1&,T2&,T3&,T4&,T5&,T6&,T7&,T8&> tie(T0 &t0, T1 &t1, T2 &t2, T3 &t3, T4 &t4, T5 &t5, T6 &t6, T7 &t7, T8 &t8)
+{
+  return tuple<T0&,T1&,T2&,T3&,T4&,T5&,T6&,T7&,T8&>(t0,t1,t2,t3,t4,t5,t6,t7,t8);
+}
+
+template<typename T0,typename T1, typename T2, typename T3, typename T4, typename T5, typename T6, typename T7, typename T8, typename T9>
+__host__ __device__ inline
+tuple<T0&,T1&,T2&,T3&,T4&,T5&,T6&,T7&,T8&,T9&> tie(T0 &t0, T1 &t1, T2 &t2, T3 &t3, T4 &t4, T5 &t5, T6 &t6, T7 &t7, T8 &t8, T9 &t9)
+{
+  return tuple<T0&,T1&,T2&,T3&,T4&,T5&,T6&,T7&,T8&,T9&>(t0,t1,t2,t3,t4,t5,t6,t7,t8,t9);
+}
+
+template<
+  typename T0, typename T1, typename T2, typename T3, typename T4, typename T5, typename T6, typename T7, typename T8, typename T9,
+  typename U0, typename U1, typename U2, typename U3, typename U4, typename U5, typename U6, typename U7, typename U8, typename U9
+>
+__host__ __device__ inline
+void swap(thrust::tuple<T0,T1,T2,T3,T4,T5,T6,T7,T8,T9> &x,
+          thrust::tuple<U0,U1,U2,U3,U4,U5,U6,U7,U8,U9> &y)
+{
+  return x.swap(y);
+}
+
+
+
+namespace detail
+{
+
+template<class T1, class T2>
+__host__ __device__
+inline bool eq(const T1& lhs, const T2& rhs) {
+  return lhs.get_head() == rhs.get_head() &&
+         eq(lhs.get_tail(), rhs.get_tail());
+}
+template<>
+inline bool eq<null_type,null_type>(const null_type&, const null_type&) { return true; }
+
+template<class T1, class T2>
+__host__ __device__
+inline bool neq(const T1& lhs, const T2& rhs) {
+  return lhs.get_head() != rhs.get_head()  ||
+         neq(lhs.get_tail(), rhs.get_tail());
+}
+template<>
+__host__ __device__
+inline bool neq<null_type,null_type>(const null_type&, const null_type&) { return false; }
+
+template<class T1, class T2>
+__host__ __device__
+inline bool lt(const T1& lhs, const T2& rhs) {
+  return (lhs.get_head() < rhs.get_head())  ||
+            (!(rhs.get_head() < lhs.get_head()) &&
+             lt(lhs.get_tail(), rhs.get_tail()));
+}
+template<>
+__host__ __device__
+inline bool lt<null_type,null_type>(const null_type&, const null_type&) { return false; }
+
+template<class T1, class T2>
+__host__ __device__
+inline bool gt(const T1& lhs, const T2& rhs) {
+  return (lhs.get_head() > rhs.get_head())  ||
+            (!(rhs.get_head() > lhs.get_head()) &&
+             gt(lhs.get_tail(), rhs.get_tail()));
+}
+template<>
+__host__ __device__
+inline bool gt<null_type,null_type>(const null_type&, const null_type&) { return false; }
+
+template<class T1, class T2>
+__host__ __device__
+inline bool lte(const T1& lhs, const T2& rhs) {
+  return lhs.get_head() <= rhs.get_head()  &&
+          ( !(rhs.get_head() <= lhs.get_head()) ||
+            lte(lhs.get_tail(), rhs.get_tail()));
+}
+template<>
+__host__ __device__
+inline bool lte<null_type,null_type>(const null_type&, const null_type&) { return true; }
+
+template<class T1, class T2>
+__host__ __device__
+inline bool gte(const T1& lhs, const T2& rhs) {
+  return lhs.get_head() >= rhs.get_head()  &&
+          ( !(rhs.get_head() >= lhs.get_head()) ||
+            gte(lhs.get_tail(), rhs.get_tail()));
+}
+template<>
+__host__ __device__
+inline bool gte<null_type,null_type>(const null_type&, const null_type&) { return true; }
+
+} // end detail
+
+
+
+// equal ----
+
+template<class T1, class T2, class S1, class S2>
+__host__ __device__
+inline bool operator==(const detail::cons<T1, T2>& lhs, const detail::cons<S1, S2>& rhs)
+{
+  // XXX support this eventually -jph
+  //// check that tuple lengths are equal
+  //BOOST_STATIC_ASSERT(tuple_size<T2>::value == tuple_size<S2>::value);
+
+  return  detail::eq(lhs, rhs);
+} // end operator==()
+
+// not equal -----
+
+template<class T1, class T2, class S1, class S2>
+__host__ __device__
+inline bool operator!=(const detail::cons<T1, T2>& lhs, const detail::cons<S1, S2>& rhs)
+{
+  // XXX support this eventually -jph
+  //// check that tuple lengths are equal
+  //BOOST_STATIC_ASSERT(tuple_size<T2>::value == tuple_size<S2>::value);
+
+  return detail::neq(lhs, rhs);
+} // end operator!=()
+
+// <
+template<class T1, class T2, class S1, class S2>
+__host__ __device__
+inline bool operator<(const detail::cons<T1, T2>& lhs, const detail::cons<S1, S2>& rhs)
+{
+  // XXX support this eventually -jph
+  //// check that tuple lengths are equal
+  //BOOST_STATIC_ASSERT(tuple_size<T2>::value == tuple_size<S2>::value);
+
+  return detail::lt(lhs, rhs);
+} // end operator<()
+
+// >
+template<class T1, class T2, class S1, class S2>
+__host__ __device__
+inline bool operator>(const detail::cons<T1, T2>& lhs, const detail::cons<S1, S2>& rhs)
+{
+  // XXX support this eventually -jph
+  //// check that tuple lengths are equal
+  //BOOST_STATIC_ASSERT(tuple_size<T2>::value == tuple_size<S2>::value);
+
+  return detail::gt(lhs, rhs);
+} // end operator>()
+
+// <=
+template<class T1, class T2, class S1, class S2>
+__host__ __device__
+inline bool operator<=(const detail::cons<T1, T2>& lhs, const detail::cons<S1, S2>& rhs)
+{
+  // XXX support this eventually -jph
+  //// check that tuple lengths are equal
+  //BOOST_STATIC_ASSERT(tuple_size<T2>::value == tuple_size<S2>::value);
+
+  return detail::lte(lhs, rhs);
+} // end operator<=()
+
+// >=
+template<class T1, class T2, class S1, class S2>
+__host__ __device__
+inline bool operator>=(const detail::cons<T1, T2>& lhs, const detail::cons<S1, S2>& rhs)
+{
+  // XXX support this eventually -jph
+  //// check that tuple lengths are equal
+  //BOOST_STATIC_ASSERT(tuple_size<T2>::value == tuple_size<S2>::value);
+
+  return detail::gte(lhs, rhs);
+} // end operator>=()
+
+} // end thrust
+
diff --git a/compat/thrust/detail/tuple_meta_transform.h b/compat/thrust/detail/tuple_meta_transform.h
new file mode 100644
index 0000000000..ff99709b6d
--- /dev/null
+++ b/compat/thrust/detail/tuple_meta_transform.h
@@ -0,0 +1,177 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/tuple.h>
+
+namespace thrust
+{
+
+namespace detail
+{
+
+template<typename Tuple,
+         template<typename> class UnaryMetaFunction,
+         unsigned int sz = thrust::tuple_size<Tuple>::value>
+  struct tuple_meta_transform;
+
+template<typename Tuple,
+         template<typename> class UnaryMetaFunction>
+  struct tuple_meta_transform<Tuple,UnaryMetaFunction,0>
+{
+  typedef null_type type;
+};
+
+template<typename Tuple,
+         template<typename> class UnaryMetaFunction>
+  struct tuple_meta_transform<Tuple,UnaryMetaFunction,1>
+{
+  typedef thrust::tuple<
+    typename UnaryMetaFunction<typename thrust::tuple_element<0,Tuple>::type>::type
+  > type;
+};
+
+template<typename Tuple,
+         template<typename> class UnaryMetaFunction>
+  struct tuple_meta_transform<Tuple,UnaryMetaFunction,2>
+{
+  typedef thrust::tuple<
+    typename UnaryMetaFunction<typename thrust::tuple_element<0,Tuple>::type>::type,
+    typename UnaryMetaFunction<typename thrust::tuple_element<1,Tuple>::type>::type
+  > type;
+};
+
+template<typename Tuple,
+         template<typename> class UnaryMetaFunction>
+  struct tuple_meta_transform<Tuple,UnaryMetaFunction,3>
+{
+  typedef thrust::tuple<
+    typename UnaryMetaFunction<typename thrust::tuple_element<0,Tuple>::type>::type,
+    typename UnaryMetaFunction<typename thrust::tuple_element<1,Tuple>::type>::type,
+    typename UnaryMetaFunction<typename thrust::tuple_element<2,Tuple>::type>::type
+  > type;
+};
+
+template<typename Tuple,
+         template<typename> class UnaryMetaFunction>
+  struct tuple_meta_transform<Tuple,UnaryMetaFunction,4>
+{
+  typedef thrust::tuple<
+    typename UnaryMetaFunction<typename thrust::tuple_element<0,Tuple>::type>::type,
+    typename UnaryMetaFunction<typename thrust::tuple_element<1,Tuple>::type>::type,
+    typename UnaryMetaFunction<typename thrust::tuple_element<2,Tuple>::type>::type,
+    typename UnaryMetaFunction<typename thrust::tuple_element<3,Tuple>::type>::type
+  > type;
+};
+
+template<typename Tuple,
+         template<typename> class UnaryMetaFunction>
+  struct tuple_meta_transform<Tuple,UnaryMetaFunction,5>
+{
+  typedef thrust::tuple<
+    typename UnaryMetaFunction<typename thrust::tuple_element<0,Tuple>::type>::type,
+    typename UnaryMetaFunction<typename thrust::tuple_element<1,Tuple>::type>::type,
+    typename UnaryMetaFunction<typename thrust::tuple_element<2,Tuple>::type>::type,
+    typename UnaryMetaFunction<typename thrust::tuple_element<3,Tuple>::type>::type,
+    typename UnaryMetaFunction<typename thrust::tuple_element<4,Tuple>::type>::type
+  > type;
+};
+
+template<typename Tuple,
+         template<typename> class UnaryMetaFunction>
+  struct tuple_meta_transform<Tuple,UnaryMetaFunction,6>
+{
+  typedef thrust::tuple<
+    typename UnaryMetaFunction<typename thrust::tuple_element<0,Tuple>::type>::type,
+    typename UnaryMetaFunction<typename thrust::tuple_element<1,Tuple>::type>::type,
+    typename UnaryMetaFunction<typename thrust::tuple_element<2,Tuple>::type>::type,
+    typename UnaryMetaFunction<typename thrust::tuple_element<3,Tuple>::type>::type,
+    typename UnaryMetaFunction<typename thrust::tuple_element<4,Tuple>::type>::type,
+    typename UnaryMetaFunction<typename thrust::tuple_element<5,Tuple>::type>::type
+  > type;
+};
+
+template<typename Tuple,
+         template<typename> class UnaryMetaFunction>
+  struct tuple_meta_transform<Tuple,UnaryMetaFunction,7>
+{
+  typedef thrust::tuple<
+    typename UnaryMetaFunction<typename thrust::tuple_element<0,Tuple>::type>::type,
+    typename UnaryMetaFunction<typename thrust::tuple_element<1,Tuple>::type>::type,
+    typename UnaryMetaFunction<typename thrust::tuple_element<2,Tuple>::type>::type,
+    typename UnaryMetaFunction<typename thrust::tuple_element<3,Tuple>::type>::type,
+    typename UnaryMetaFunction<typename thrust::tuple_element<4,Tuple>::type>::type,
+    typename UnaryMetaFunction<typename thrust::tuple_element<5,Tuple>::type>::type,
+    typename UnaryMetaFunction<typename thrust::tuple_element<6,Tuple>::type>::type
+  > type;
+};
+
+template<typename Tuple,
+         template<typename> class UnaryMetaFunction>
+  struct tuple_meta_transform<Tuple,UnaryMetaFunction,8>
+{
+  typedef thrust::tuple<
+    typename UnaryMetaFunction<typename thrust::tuple_element<0,Tuple>::type>::type,
+    typename UnaryMetaFunction<typename thrust::tuple_element<1,Tuple>::type>::type,
+    typename UnaryMetaFunction<typename thrust::tuple_element<2,Tuple>::type>::type,
+    typename UnaryMetaFunction<typename thrust::tuple_element<3,Tuple>::type>::type,
+    typename UnaryMetaFunction<typename thrust::tuple_element<4,Tuple>::type>::type,
+    typename UnaryMetaFunction<typename thrust::tuple_element<5,Tuple>::type>::type,
+    typename UnaryMetaFunction<typename thrust::tuple_element<6,Tuple>::type>::type,
+    typename UnaryMetaFunction<typename thrust::tuple_element<7,Tuple>::type>::type
+  > type;
+};
+
+template<typename Tuple,
+         template<typename> class UnaryMetaFunction>
+  struct tuple_meta_transform<Tuple,UnaryMetaFunction,9>
+{
+  typedef thrust::tuple<
+    typename UnaryMetaFunction<typename thrust::tuple_element<0,Tuple>::type>::type,
+    typename UnaryMetaFunction<typename thrust::tuple_element<1,Tuple>::type>::type,
+    typename UnaryMetaFunction<typename thrust::tuple_element<2,Tuple>::type>::type,
+    typename UnaryMetaFunction<typename thrust::tuple_element<3,Tuple>::type>::type,
+    typename UnaryMetaFunction<typename thrust::tuple_element<4,Tuple>::type>::type,
+    typename UnaryMetaFunction<typename thrust::tuple_element<5,Tuple>::type>::type,
+    typename UnaryMetaFunction<typename thrust::tuple_element<6,Tuple>::type>::type,
+    typename UnaryMetaFunction<typename thrust::tuple_element<7,Tuple>::type>::type,
+    typename UnaryMetaFunction<typename thrust::tuple_element<8,Tuple>::type>::type
+  > type;
+};
+
+template<typename Tuple,
+         template<typename> class UnaryMetaFunction>
+  struct tuple_meta_transform<Tuple,UnaryMetaFunction,10>
+{
+  typedef thrust::tuple<
+    typename UnaryMetaFunction<typename thrust::tuple_element<0,Tuple>::type>::type,
+    typename UnaryMetaFunction<typename thrust::tuple_element<1,Tuple>::type>::type,
+    typename UnaryMetaFunction<typename thrust::tuple_element<2,Tuple>::type>::type,
+    typename UnaryMetaFunction<typename thrust::tuple_element<3,Tuple>::type>::type,
+    typename UnaryMetaFunction<typename thrust::tuple_element<4,Tuple>::type>::type,
+    typename UnaryMetaFunction<typename thrust::tuple_element<5,Tuple>::type>::type,
+    typename UnaryMetaFunction<typename thrust::tuple_element<6,Tuple>::type>::type,
+    typename UnaryMetaFunction<typename thrust::tuple_element<7,Tuple>::type>::type,
+    typename UnaryMetaFunction<typename thrust::tuple_element<8,Tuple>::type>::type,
+    typename UnaryMetaFunction<typename thrust::tuple_element<9,Tuple>::type>::type
+  > type;
+};
+
+} // end detail
+
+} // end thrust
+
diff --git a/compat/thrust/detail/tuple_transform.h b/compat/thrust/detail/tuple_transform.h
new file mode 100644
index 0000000000..f18b8727e1
--- /dev/null
+++ b/compat/thrust/detail/tuple_transform.h
@@ -0,0 +1,418 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/tuple.h>
+#include <thrust/detail/tuple_meta_transform.h>
+
+namespace thrust
+{
+
+namespace detail
+{
+
+template<typename Tuple,
+         template<typename> class UnaryMetaFunction,
+         typename UnaryFunction,
+         unsigned int sz = thrust::tuple_size<Tuple>::value>
+  struct tuple_transform_functor;
+
+
+template<typename Tuple,
+         template<typename> class UnaryMetaFunction,
+         typename UnaryFunction>
+  struct tuple_transform_functor<Tuple,UnaryMetaFunction,UnaryFunction,0>
+{
+  static __host__
+  typename tuple_meta_transform<Tuple,UnaryMetaFunction>::type
+  do_it_on_the_host(const Tuple &t, UnaryFunction f)
+  {
+    return thrust::null_type();
+  }
+
+  static __host__ __device__
+  typename tuple_meta_transform<Tuple,UnaryMetaFunction>::type
+  do_it_on_the_host_or_device(const Tuple &t, UnaryFunction f)
+  {
+    return thrust::null_type();
+  }
+};
+
+
+template<typename Tuple,
+         template<typename> class UnaryMetaFunction,
+         typename UnaryFunction>
+  struct tuple_transform_functor<Tuple,UnaryMetaFunction,UnaryFunction,1>
+{
+  static __host__
+  typename tuple_meta_transform<Tuple,UnaryMetaFunction>::type
+  do_it_on_the_host(const Tuple &t, UnaryFunction f)
+  {
+    typedef typename tuple_meta_transform<Tuple,UnaryMetaFunction>::type XfrmTuple;
+
+    return XfrmTuple(f(thrust::get<0>(t)));
+  }
+
+  static __host__ __device__
+  typename tuple_meta_transform<Tuple,UnaryMetaFunction>::type
+  do_it_on_the_host_or_device(const Tuple &t, UnaryFunction f)
+  {
+    typedef typename tuple_meta_transform<Tuple,UnaryMetaFunction>::type XfrmTuple;
+
+    return XfrmTuple(f(thrust::get<0>(t)));
+  }
+};
+
+
+template<typename Tuple,
+         template<typename> class UnaryMetaFunction,
+         typename UnaryFunction>
+  struct tuple_transform_functor<Tuple,UnaryMetaFunction,UnaryFunction,2>
+{
+  static __host__
+  typename tuple_meta_transform<Tuple,UnaryMetaFunction>::type
+  do_it_on_the_host(const Tuple &t, UnaryFunction f)
+  {
+    typedef typename tuple_meta_transform<Tuple,UnaryMetaFunction>::type XfrmTuple;
+
+    return XfrmTuple(f(thrust::get<0>(t)),
+                     f(thrust::get<1>(t)));
+  }
+
+  static __host__ __device__
+  typename tuple_meta_transform<Tuple,UnaryMetaFunction>::type
+  do_it_on_the_host_or_device(const Tuple &t, UnaryFunction f)
+  {
+    typedef typename tuple_meta_transform<Tuple,UnaryMetaFunction>::type XfrmTuple;
+
+    return XfrmTuple(f(thrust::get<0>(t)),
+                     f(thrust::get<1>(t)));
+  }
+};
+
+
+template<typename Tuple,
+         template<typename> class UnaryMetaFunction,
+         typename UnaryFunction>
+  struct tuple_transform_functor<Tuple,UnaryMetaFunction,UnaryFunction,3>
+{
+  static __host__
+  typename tuple_meta_transform<Tuple,UnaryMetaFunction>::type
+  do_it_on_the_host(const Tuple &t, UnaryFunction f)
+  {
+    typedef typename tuple_meta_transform<Tuple,UnaryMetaFunction>::type XfrmTuple;
+
+    return XfrmTuple(f(thrust::get<0>(t)),
+                     f(thrust::get<1>(t)),
+                     f(thrust::get<2>(t)));
+  }
+
+  static __host__ __device__
+  typename tuple_meta_transform<Tuple,UnaryMetaFunction>::type
+  do_it_on_the_host_or_device(const Tuple &t, UnaryFunction f)
+  {
+    typedef typename tuple_meta_transform<Tuple,UnaryMetaFunction>::type XfrmTuple;
+
+    return XfrmTuple(f(thrust::get<0>(t)),
+                     f(thrust::get<1>(t)),
+                     f(thrust::get<2>(t)));
+  }
+};
+
+
+template<typename Tuple,
+         template<typename> class UnaryMetaFunction,
+         typename UnaryFunction>
+  struct tuple_transform_functor<Tuple,UnaryMetaFunction,UnaryFunction,4>
+{
+  static __host__
+  typename tuple_meta_transform<Tuple,UnaryMetaFunction>::type
+  do_it_on_the_host(const Tuple &t, UnaryFunction f)
+  {
+    typedef typename tuple_meta_transform<Tuple,UnaryMetaFunction>::type XfrmTuple;
+
+    return XfrmTuple(f(thrust::get<0>(t)),
+                     f(thrust::get<1>(t)),
+                     f(thrust::get<2>(t)),
+                     f(thrust::get<3>(t)));
+  }
+
+  static __host__ __device__
+  typename tuple_meta_transform<Tuple,UnaryMetaFunction>::type
+  do_it_on_the_host_or_device(const Tuple &t, UnaryFunction f)
+  {
+    typedef typename tuple_meta_transform<Tuple,UnaryMetaFunction>::type XfrmTuple;
+
+    return XfrmTuple(f(thrust::get<0>(t)),
+                     f(thrust::get<1>(t)),
+                     f(thrust::get<2>(t)),
+                     f(thrust::get<3>(t)));
+  }
+};
+
+
+template<typename Tuple,
+         template<typename> class UnaryMetaFunction,
+         typename UnaryFunction>
+  struct tuple_transform_functor<Tuple,UnaryMetaFunction,UnaryFunction,5>
+{
+  static __host__
+  typename tuple_meta_transform<Tuple,UnaryMetaFunction>::type
+  do_it_on_the_host(const Tuple &t, UnaryFunction f)
+  {
+    typedef typename tuple_meta_transform<Tuple,UnaryMetaFunction>::type XfrmTuple;
+
+    return XfrmTuple(f(thrust::get<0>(t)),
+                     f(thrust::get<1>(t)),
+                     f(thrust::get<2>(t)),
+                     f(thrust::get<3>(t)),
+                     f(thrust::get<4>(t)));
+  }
+
+  static __host__ __device__
+  typename tuple_meta_transform<Tuple,UnaryMetaFunction>::type
+  do_it_on_the_host_or_device(const Tuple &t, UnaryFunction f)
+  {
+    typedef typename tuple_meta_transform<Tuple,UnaryMetaFunction>::type XfrmTuple;
+
+    return XfrmTuple(f(thrust::get<0>(t)),
+                     f(thrust::get<1>(t)),
+                     f(thrust::get<2>(t)),
+                     f(thrust::get<3>(t)),
+                     f(thrust::get<4>(t)));
+  }
+};
+
+
+template<typename Tuple,
+         template<typename> class UnaryMetaFunction,
+         typename UnaryFunction>
+  struct tuple_transform_functor<Tuple,UnaryMetaFunction,UnaryFunction,6>
+{
+  static __host__
+  typename tuple_meta_transform<Tuple,UnaryMetaFunction>::type
+  do_it_on_the_host(const Tuple &t, UnaryFunction f)
+  {
+    typedef typename tuple_meta_transform<Tuple,UnaryMetaFunction>::type XfrmTuple;
+
+    return XfrmTuple(f(thrust::get<0>(t)),
+                     f(thrust::get<1>(t)),
+                     f(thrust::get<2>(t)),
+                     f(thrust::get<3>(t)),
+                     f(thrust::get<4>(t)),
+                     f(thrust::get<5>(t)));
+  }
+
+  static __host__ __device__
+  typename tuple_meta_transform<Tuple,UnaryMetaFunction>::type
+  do_it_on_the_host_or_device(const Tuple &t, UnaryFunction f)
+  {
+    typedef typename tuple_meta_transform<Tuple,UnaryMetaFunction>::type XfrmTuple;
+
+    return XfrmTuple(f(thrust::get<0>(t)),
+                     f(thrust::get<1>(t)),
+                     f(thrust::get<2>(t)),
+                     f(thrust::get<3>(t)),
+                     f(thrust::get<4>(t)),
+                     f(thrust::get<5>(t)));
+  }
+};
+
+
+template<typename Tuple,
+         template<typename> class UnaryMetaFunction,
+         typename UnaryFunction>
+  struct tuple_transform_functor<Tuple,UnaryMetaFunction,UnaryFunction,7>
+{
+  static __host__
+  typename tuple_meta_transform<Tuple,UnaryMetaFunction>::type
+  do_it_on_the_host(const Tuple &t, UnaryFunction f)
+  {
+    typedef typename tuple_meta_transform<Tuple,UnaryMetaFunction>::type XfrmTuple;
+
+    return XfrmTuple(f(thrust::get<0>(t)),
+                     f(thrust::get<1>(t)),
+                     f(thrust::get<2>(t)),
+                     f(thrust::get<3>(t)),
+                     f(thrust::get<4>(t)),
+                     f(thrust::get<5>(t)),
+                     f(thrust::get<6>(t)));
+  }
+
+  static __host__ __device__
+  typename tuple_meta_transform<Tuple,UnaryMetaFunction>::type
+  do_it_on_the_host_or_device(const Tuple &t, UnaryFunction f)
+  {
+    typedef typename tuple_meta_transform<Tuple,UnaryMetaFunction>::type XfrmTuple;
+
+    return XfrmTuple(f(thrust::get<0>(t)),
+                     f(thrust::get<1>(t)),
+                     f(thrust::get<2>(t)),
+                     f(thrust::get<3>(t)),
+                     f(thrust::get<4>(t)),
+                     f(thrust::get<5>(t)),
+                     f(thrust::get<6>(t)));
+  }
+};
+
+
+template<typename Tuple,
+         template<typename> class UnaryMetaFunction,
+         typename UnaryFunction>
+  struct tuple_transform_functor<Tuple,UnaryMetaFunction,UnaryFunction,8>
+{
+  static __host__
+  typename tuple_meta_transform<Tuple,UnaryMetaFunction>::type
+  do_it_on_the_host(const Tuple &t, UnaryFunction f)
+  {
+    typedef typename tuple_meta_transform<Tuple,UnaryMetaFunction>::type XfrmTuple;
+
+    return XfrmTuple(f(thrust::get<0>(t)),
+                     f(thrust::get<1>(t)),
+                     f(thrust::get<2>(t)),
+                     f(thrust::get<3>(t)),
+                     f(thrust::get<4>(t)),
+                     f(thrust::get<5>(t)),
+                     f(thrust::get<6>(t)),
+                     f(thrust::get<7>(t)));
+  }
+
+  static __host__ __device__
+  typename tuple_meta_transform<Tuple,UnaryMetaFunction>::type
+  do_it_on_the_host_or_device(const Tuple &t, UnaryFunction f)
+  {
+    typedef typename tuple_meta_transform<Tuple,UnaryMetaFunction>::type XfrmTuple;
+
+    return XfrmTuple(f(thrust::get<0>(t)),
+                     f(thrust::get<1>(t)),
+                     f(thrust::get<2>(t)),
+                     f(thrust::get<3>(t)),
+                     f(thrust::get<4>(t)),
+                     f(thrust::get<5>(t)),
+                     f(thrust::get<6>(t)),
+                     f(thrust::get<7>(t)));
+  }
+};
+
+
+template<typename Tuple,
+         template<typename> class UnaryMetaFunction,
+         typename UnaryFunction>
+  struct tuple_transform_functor<Tuple,UnaryMetaFunction,UnaryFunction,9>
+{
+  static __host__
+  typename tuple_meta_transform<Tuple,UnaryMetaFunction>::type
+  do_it_on_the_host(const Tuple &t, UnaryFunction f)
+  {
+    typedef typename tuple_meta_transform<Tuple,UnaryMetaFunction>::type XfrmTuple;
+
+    return XfrmTuple(f(thrust::get<0>(t)),
+                     f(thrust::get<1>(t)),
+                     f(thrust::get<2>(t)),
+                     f(thrust::get<3>(t)),
+                     f(thrust::get<4>(t)),
+                     f(thrust::get<5>(t)),
+                     f(thrust::get<6>(t)),
+                     f(thrust::get<7>(t)),
+                     f(thrust::get<8>(t)));
+  }
+
+  static __host__ __device__
+  typename tuple_meta_transform<Tuple,UnaryMetaFunction>::type
+  do_it_on_the_host_or_device(const Tuple &t, UnaryFunction f)
+  {
+    typedef typename tuple_meta_transform<Tuple,UnaryMetaFunction>::type XfrmTuple;
+
+    return XfrmTuple(f(thrust::get<0>(t)),
+                     f(thrust::get<1>(t)),
+                     f(thrust::get<2>(t)),
+                     f(thrust::get<3>(t)),
+                     f(thrust::get<4>(t)),
+                     f(thrust::get<5>(t)),
+                     f(thrust::get<6>(t)),
+                     f(thrust::get<7>(t)),
+                     f(thrust::get<8>(t)));
+  }
+};
+
+
+template<typename Tuple,
+         template<typename> class UnaryMetaFunction,
+         typename UnaryFunction>
+  struct tuple_transform_functor<Tuple,UnaryMetaFunction,UnaryFunction,10>
+{
+  static __host__
+  typename tuple_meta_transform<Tuple,UnaryMetaFunction>::type
+  do_it_on_the_host(const Tuple &t, UnaryFunction f)
+  {
+    typedef typename tuple_meta_transform<Tuple,UnaryMetaFunction>::type XfrmTuple;
+
+    return XfrmTuple(f(thrust::get<0>(t)),
+                     f(thrust::get<1>(t)),
+                     f(thrust::get<2>(t)),
+                     f(thrust::get<3>(t)),
+                     f(thrust::get<4>(t)),
+                     f(thrust::get<5>(t)),
+                     f(thrust::get<6>(t)),
+                     f(thrust::get<7>(t)),
+                     f(thrust::get<8>(t)),
+                     f(thrust::get<9>(t)));
+  }
+
+  static __host__ __device__
+  typename tuple_meta_transform<Tuple,UnaryMetaFunction>::type
+  do_it_on_the_host_or_device(const Tuple &t, UnaryFunction f)
+  {
+    typedef typename tuple_meta_transform<Tuple,UnaryMetaFunction>::type XfrmTuple;
+
+    return XfrmTuple(f(thrust::get<0>(t)),
+                     f(thrust::get<1>(t)),
+                     f(thrust::get<2>(t)),
+                     f(thrust::get<3>(t)),
+                     f(thrust::get<4>(t)),
+                     f(thrust::get<5>(t)),
+                     f(thrust::get<6>(t)),
+                     f(thrust::get<7>(t)),
+                     f(thrust::get<8>(t)),
+                     f(thrust::get<9>(t)));
+  }
+};
+
+
+template<template<typename> class UnaryMetaFunction,
+         typename Tuple,
+         typename UnaryFunction>
+typename tuple_meta_transform<Tuple,UnaryMetaFunction>::type
+tuple_host_transform(const Tuple &t, UnaryFunction f)
+{
+  return tuple_transform_functor<Tuple,UnaryMetaFunction,UnaryFunction>::do_it_on_the_host(t,f);
+}
+
+template<template<typename> class UnaryMetaFunction,
+         typename Tuple,
+         typename UnaryFunction>
+typename tuple_meta_transform<Tuple,UnaryMetaFunction>::type
+__host__ __device__
+tuple_host_device_transform(const Tuple &t, UnaryFunction f)
+{
+  return tuple_transform_functor<Tuple,UnaryMetaFunction,UnaryFunction>::do_it_on_the_host_or_device(t,f);
+}
+
+} // end detail
+
+} // end thrust
+
diff --git a/compat/thrust/detail/type_traits.h b/compat/thrust/detail/type_traits.h
new file mode 100644
index 0000000000..5dbeb906eb
--- /dev/null
+++ b/compat/thrust/detail/type_traits.h
@@ -0,0 +1,641 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file type_traits.h
+ *  \brief Temporarily define some type traits
+ *         until nvcc can compile tr1::type_traits.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// XXX nvcc 2.2 closed beta can't compile type_traits
+//// find type_traits
+//
+//#ifdef __GNUC__
+//
+//#if __GNUC__ == 4 && __GNUC_MINOR__ == 2
+//#include <tr1/type_traits>
+//#elif __GNUC__ == 4 && __GNUC_MINOR__ > 2
+//#include <type_traits>
+//#endif // GCC version
+//
+//#endif // GCC
+//
+//#ifdef _MSC_VER
+//#include <type_traits>
+//#endif // MSVC
+
+
+namespace thrust
+{
+
+// forward declaration of device_reference
+template<typename T> class device_reference;
+
+namespace detail
+{
+ /// helper classes [4.3].
+ template<typename _Tp, _Tp __v>
+   struct integral_constant
+   {
+     static const _Tp                      value = __v;
+     typedef _Tp                           value_type;
+     typedef integral_constant<_Tp, __v>   type;
+   };
+ 
+ /// typedef for true_type
+ typedef integral_constant<bool, true>     true_type;
+
+ /// typedef for true_type
+ typedef integral_constant<bool, false>    false_type;
+
+//template<typename T> struct is_integral : public std::tr1::is_integral<T> {};
+template<typename T> struct is_integral                           : public false_type {};
+template<>           struct is_integral<bool>                     : public true_type {};
+template<>           struct is_integral<char>                     : public true_type {};
+template<>           struct is_integral<signed char>              : public true_type {};
+template<>           struct is_integral<unsigned char>            : public true_type {};
+template<>           struct is_integral<short>                    : public true_type {};
+template<>           struct is_integral<unsigned short>           : public true_type {};
+template<>           struct is_integral<int>                      : public true_type {};
+template<>           struct is_integral<unsigned int>             : public true_type {};
+template<>           struct is_integral<long>                     : public true_type {};
+template<>           struct is_integral<unsigned long>            : public true_type {};
+template<>           struct is_integral<long long>                : public true_type {};
+template<>           struct is_integral<unsigned long long>       : public true_type {};
+template<>           struct is_integral<const bool>               : public true_type {};
+template<>           struct is_integral<const char>               : public true_type {};
+template<>           struct is_integral<const unsigned char>      : public true_type {};
+template<>           struct is_integral<const short>              : public true_type {};
+template<>           struct is_integral<const unsigned short>     : public true_type {};
+template<>           struct is_integral<const int>                : public true_type {};
+template<>           struct is_integral<const unsigned int>       : public true_type {};
+template<>           struct is_integral<const long>               : public true_type {};
+template<>           struct is_integral<const unsigned long>      : public true_type {};
+template<>           struct is_integral<const long long>          : public true_type {};
+template<>           struct is_integral<const unsigned long long> : public true_type {};
+
+template<typename T> struct is_floating_point              : public false_type {};
+template<>           struct is_floating_point<float>       : public true_type {};
+template<>           struct is_floating_point<double>      : public true_type {};
+template<>           struct is_floating_point<long double> : public true_type {};
+
+template<typename T> struct is_arithmetic               : public is_integral<T> {};
+template<>           struct is_arithmetic<float>        : public true_type {};
+template<>           struct is_arithmetic<double>       : public true_type {};
+template<>           struct is_arithmetic<const float>  : public true_type {};
+template<>           struct is_arithmetic<const double> : public true_type {};
+
+template<typename T> struct is_pointer      : public false_type {};
+template<typename T> struct is_pointer<T *> : public true_type  {};
+
+template<typename T> struct is_device_ptr  : public false_type {};
+
+template<typename T> struct is_void             : public false_type {};
+template<>           struct is_void<void>       : public true_type {};
+template<>           struct is_void<const void> : public true_type {};
+
+
+namespace tt_detail
+{
+
+
+} // end tt_detail
+
+template<typename T> struct is_pod
+   : public integral_constant<
+       bool,
+       is_void<T>::value || is_pointer<T>::value || is_arithmetic<T>::value
+#if THRUST_HOST_COMPILER   == THRUST_HOST_COMPILER_MSVC
+// use intrinsic type traits
+       || __is_pod(T)
+#elif THRUST_HOST_COMPILER == THRUST_HOST_COMPILER_GCC
+// only use the intrinsic for >= 4.3
+#if (__GNUC__ >= 4) && (__GNUC_MINOR__ >= 3)
+       || __is_pod(T)
+#endif // GCC VERSION
+#endif // THRUST_HOST_COMPILER
+     >
+ {};
+
+
+template<typename T> struct has_trivial_constructor
+  : public integral_constant<
+      bool,
+      is_pod<T>::value
+#if THRUST_HOST_COMPILER   == THRUST_HOST_COMPILER_MSVC
+      || __has_trivial_constructor(T)
+#elif THRUST_HOST_COMPILER == THRUST_HOST_COMPILER_GCC
+// only use the intrinsic for >= 4.3
+#if (__GNUC__ >= 4) && (__GNUC_MINOR__ >= 3)
+      || __has_trivial_constructor(T)
+#endif // GCC VERSION
+#endif // THRUST_HOST_COMPILER
+      >
+{};
+
+template<typename T> struct has_trivial_copy_constructor
+  : public integral_constant<
+      bool,
+      is_pod<T>::value
+#if THRUST_HOST_COMPILER == THRUST_HOST_COMPILER_MSVC
+      || __has_trivial_copy(T)
+#elif THRUST_HOST_COMPILER == THRUST_HOST_COMPILER_GCC
+// only use the intrinsic for >= 4.3
+#if (__GNUC__ >= 4) && (__GNUC_MINOR__ >= 3)
+      || __has_trivial_copy(T)
+#endif // GCC VERSION
+#endif // THRUST_HOST_COMPILER
+    >
+{};
+
+template<typename T> struct has_trivial_destructor : public is_pod<T> {};
+
+template<typename T> struct is_const          : public false_type {};
+template<typename T> struct is_const<const T> : public true_type {};
+
+template<typename T> struct is_volatile             : public false_type {};
+template<typename T> struct is_volatile<volatile T> : public true_type {};
+
+template<typename T>
+  struct add_const
+{
+  typedef T const type;
+}; // end add_const
+
+template<typename T>
+  struct remove_const
+{
+  typedef T type;
+}; // end remove_const
+
+template<typename T>
+  struct remove_const<const T>
+{
+  typedef T type;
+}; // end remove_const
+
+template<typename T>
+  struct add_volatile
+{
+  typedef volatile T type;
+}; // end add_volatile
+
+template<typename T>
+  struct remove_volatile
+{
+  typedef T type;
+}; // end remove_volatile
+
+template<typename T>
+  struct remove_volatile<volatile T>
+{
+  typedef T type;
+}; // end remove_volatile
+
+template<typename T>
+  struct add_cv
+{
+  typedef const volatile T type;
+}; // end add_cv
+
+template<typename T>
+  struct remove_cv
+{
+  typedef typename remove_const<typename remove_volatile<T>::type>::type type;
+}; // end remove_cv
+
+
+template<typename T> struct is_reference     : public false_type {};
+template<typename T> struct is_reference<T&> : public true_type {};
+
+template<typename T> struct is_device_reference                                : public false_type {};
+template<typename T> struct is_device_reference< thrust::device_reference<T> > : public true_type {};
+
+
+// NB: Careful with reference to void.
+template<typename _Tp, bool = (is_void<_Tp>::value || is_reference<_Tp>::value)>
+  struct __add_reference_helper
+  { typedef _Tp&    type; };
+
+template<typename _Tp>
+  struct __add_reference_helper<_Tp, true>
+  { typedef _Tp     type; };
+
+template<typename _Tp>
+  struct add_reference
+    : public __add_reference_helper<_Tp>{};
+
+template<typename T>
+  struct remove_reference
+{
+  typedef T type;
+}; // end remove_reference
+
+template<typename T>
+  struct remove_reference<T&>
+{
+  typedef T type;
+}; // end remove_reference
+
+template<typename T1, typename T2>
+  struct is_same
+    : public false_type
+{
+}; // end is_same
+
+template<typename T>
+  struct is_same<T,T>
+    : public true_type
+{
+}; // end is_same
+
+template<typename T1, typename T2>
+  struct lazy_is_same
+    : is_same<typename T1::type, typename T2::type>
+{
+}; // end lazy_is_same
+
+template<typename T1, typename T2>
+  struct is_different
+    : public true_type
+{
+}; // end is_different
+
+template<typename T>
+  struct is_different<T,T>
+    : public false_type
+{
+}; // end is_different
+
+template<typename T1, typename T2>
+  struct lazy_is_different
+    : is_different<typename T1::type, typename T2::type>
+{
+}; // end lazy_is_different
+
+namespace tt_detail
+{
+
+template<typename T>
+  struct is_int_or_cref
+{
+  typedef typename remove_reference<T>::type type_sans_ref;
+  static const bool value = (is_integral<T>::value
+                             || (is_integral<type_sans_ref>::value
+                                 && is_const<type_sans_ref>::value
+                                 && !is_volatile<type_sans_ref>::value));
+}; // end is_int_or_cref
+
+
+__THRUST_DISABLE_MSVC_POSSIBLE_LOSS_OF_DATA_WARNING_BEGIN
+__THRUST_DISABLE_MSVC_FORCING_VALUE_TO_BOOL_BEGIN
+
+
+template<typename From, typename To>
+  struct is_convertible_sfinae
+{
+  private:
+    typedef char                          one_byte;
+    typedef struct { char two_chars[2]; } two_bytes;
+
+    static one_byte  test(To);
+    static two_bytes test(...);
+    static From      m_from;
+
+  public:
+    static const bool value = sizeof(test(m_from)) == sizeof(one_byte);
+}; // end is_convertible_sfinae
+
+
+__THRUST_DISABLE_MSVC_FORCING_VALUE_TO_BOOL_END
+__THRUST_DISABLE_MSVC_POSSIBLE_LOSS_OF_DATA_WARNING_END
+
+
+template<typename From, typename To>
+  struct is_convertible_needs_simple_test
+{
+  static const bool from_is_void      = is_void<From>::value;
+  static const bool to_is_void        = is_void<To>::value;
+  static const bool from_is_float     = is_floating_point<typename remove_reference<From>::type>::value;
+  static const bool to_is_int_or_cref = is_int_or_cref<To>::value;
+
+  static const bool value = (from_is_void || to_is_void || (from_is_float && to_is_int_or_cref));
+}; // end is_convertible_needs_simple_test
+
+
+template<typename From, typename To,
+         bool = is_convertible_needs_simple_test<From,To>::value>
+  struct is_convertible
+{
+  static const bool value = (is_void<To>::value
+                             || (is_int_or_cref<To>::value
+                                 && !is_void<From>::value));
+}; // end is_convertible
+
+
+template<typename From, typename To>
+  struct is_convertible<From, To, false>
+{
+  static const bool value = (is_convertible_sfinae<typename
+                             add_reference<From>::type, To>::value);
+}; // end is_convertible
+
+
+} // end tt_detail
+
+template<typename From, typename To>
+  struct is_convertible
+    : public integral_constant<bool, tt_detail::is_convertible<From, To>::value>
+{
+}; // end is_convertible
+
+
+template<typename T1, typename T2>
+  struct is_one_convertible_to_the_other
+    : public integral_constant<
+        bool,
+        is_convertible<T1,T2>::value || is_convertible<T2,T1>::value
+      >
+{};
+
+
+// mpl stuff
+
+template <typename Condition1,               typename Condition2,              typename Condition3 = false_type,
+          typename Condition4  = false_type, typename Condition5 = false_type, typename Condition6 = false_type,
+          typename Condition7  = false_type, typename Condition8 = false_type, typename Condition9 = false_type,
+          typename Condition10 = false_type>
+  struct or_
+    : public integral_constant<
+        bool,
+        Condition1::value || Condition2::value || Condition3::value || Condition4::value || Condition5::value || Condition6::value || Condition7::value || Condition8::value || Condition9::value || Condition10::value
+      >
+{
+}; // end or_
+
+template <typename Condition1, typename Condition2, typename Condition3 = true_type>
+  struct and_
+    : public integral_constant<bool, Condition1::value && Condition2::value && Condition3::value>
+{
+}; // end and_
+
+template <typename Boolean>
+  struct not_
+    : public integral_constant<bool, !Boolean::value>
+{
+}; // end not_
+
+template <bool, typename Then, typename Else>
+  struct eval_if
+{
+}; // end eval_if
+
+template<typename Then, typename Else>
+  struct eval_if<true, Then, Else>
+{
+  typedef typename Then::type type;
+}; // end eval_if
+
+template<typename Then, typename Else>
+  struct eval_if<false, Then, Else>
+{
+  typedef typename Else::type type;
+}; // end eval_if
+
+template<typename T>
+//  struct identity
+//  XXX WAR nvcc's confusion with thrust::identity
+  struct identity_
+{
+  typedef T type;
+}; // end identity
+
+template<bool, typename T = void> struct enable_if {};
+template<typename T>              struct enable_if<true, T> {typedef T type;};
+
+template<bool, typename T> struct lazy_enable_if {};
+template<typename T>       struct lazy_enable_if<true, T> {typedef typename T::type type;};
+
+template<bool condition, typename T = void> struct disable_if : enable_if<!condition, T> {};
+template<bool condition, typename T>        struct lazy_disable_if : lazy_enable_if<!condition, T> {};
+
+
+template<typename T1, typename T2, typename T = void>
+  struct enable_if_convertible
+    : enable_if< is_convertible<T1,T2>::value, T >
+{};
+
+
+template<typename T1, typename T2, typename T = void>
+  struct disable_if_convertible
+    : disable_if< is_convertible<T1,T2>::value, T >
+{};
+
+
+template<typename T1, typename T2, typename Result = void>
+  struct enable_if_different
+    : enable_if<is_different<T1,T2>::value, Result>
+{};
+
+
+template<typename T>
+  struct is_numeric
+    : and_<
+        is_convertible<int,T>,
+        is_convertible<T,int>
+      >
+{
+}; // end is_numeric
+
+
+template<typename> struct is_reference_to_const             : false_type {};
+template<typename T> struct is_reference_to_const<const T&> : true_type {};
+
+
+// make_unsigned follows
+
+namespace tt_detail
+{
+
+template<typename T> struct make_unsigned_simple;
+
+template<> struct make_unsigned_simple<char>                   { typedef unsigned char          type; };
+template<> struct make_unsigned_simple<signed char>            { typedef signed   char          type; };
+template<> struct make_unsigned_simple<unsigned char>          { typedef unsigned char          type; };
+template<> struct make_unsigned_simple<short>                  { typedef unsigned short         type; };
+template<> struct make_unsigned_simple<unsigned short>         { typedef unsigned short         type; };
+template<> struct make_unsigned_simple<int>                    { typedef unsigned int           type; };
+template<> struct make_unsigned_simple<unsigned int>           { typedef unsigned int           type; };
+template<> struct make_unsigned_simple<long int>               { typedef unsigned long int      type; };
+template<> struct make_unsigned_simple<unsigned long int>      { typedef unsigned long int      type; };
+template<> struct make_unsigned_simple<long long int>          { typedef unsigned long long int type; };
+template<> struct make_unsigned_simple<unsigned long long int> { typedef unsigned long long int type; };
+
+template<typename T>
+  struct make_unsigned_base
+{
+  // remove cv
+  typedef typename remove_cv<T>::type remove_cv_t;
+
+  // get the simple unsigned type
+  typedef typename make_unsigned_simple<remove_cv_t>::type unsigned_remove_cv_t;
+
+  // add back const, volatile, both, or neither to the simple result
+  typedef typename eval_if<
+    is_const<T>::value && is_volatile<T>::value,
+    // add cv back
+    add_cv<unsigned_remove_cv_t>,
+    // check const & volatile individually
+    eval_if<
+      is_const<T>::value,
+      // add c back
+      add_const<unsigned_remove_cv_t>,
+      eval_if<
+        is_volatile<T>::value,
+        // add v back
+        add_volatile<unsigned_remove_cv_t>,
+        // original type was neither cv, return the simple unsigned result
+        identity_<unsigned_remove_cv_t>
+      >
+    >
+  >::type type;
+};
+
+} // end tt_detail
+
+template<typename T>
+  struct make_unsigned
+    : tt_detail::make_unsigned_base<T>
+{};
+
+struct largest_available_float
+{
+#if defined(__CUDA_ARCH__)
+#  if (__CUDA_ARCH__ < 130)
+  typedef float type;
+#  else
+  typedef double type;
+#  endif
+#else
+  typedef double type;
+#endif
+};
+
+// T1 wins if they are both the same size
+template<typename T1, typename T2>
+  struct larger_type
+    : thrust::detail::eval_if<
+        (sizeof(T2) > sizeof(T1)),
+        thrust::detail::identity_<T2>,
+        thrust::detail::identity_<T1>
+      >
+{};
+
+
+namespace is_base_of_ns
+{
+
+typedef char                          yes;
+typedef struct { char two_chars[2]; } no;
+
+template<typename Base, typename Derived>
+  struct host
+{
+  operator Base*() const;
+  operator Derived*();
+}; // end host
+
+template<typename Base, typename Derived>
+  struct impl
+{
+  template<typename T> static yes check(Derived *, T);
+  static no check(Base*, int);
+
+  static const bool value = sizeof(check(host<Base,Derived>(), int())) == sizeof(yes);
+}; // end impl
+
+} // end is_base_of_ns
+
+
+template<typename Base, typename Derived>
+  struct is_base_of
+    : integral_constant<
+        bool,
+        is_base_of_ns::impl<Base,Derived>::value
+      >
+{};
+
+template<typename Base, typename Derived, typename Result = void>
+  struct enable_if_base_of
+    : enable_if<
+        is_base_of<Base,Derived>::value,
+        Result
+      >
+{};
+
+
+namespace is_assignable_ns
+{
+
+template<typename T1, typename T2>
+  class is_assignable
+{
+  typedef char                      yes_type;
+  typedef struct { char array[2]; } no_type;
+
+  template<typename T> static typename add_reference<T>::type declval();
+  
+  template<unsigned int> struct helper { typedef void * type; };
+
+  template<typename U1, typename U2> static yes_type test(typename helper<sizeof(declval<U1>() = declval<U2>())>::type);
+
+  template<typename,typename> static no_type test(...);
+
+  public:
+    static const bool value = sizeof(test<T1,T2>(0)) == 1;
+}; // end is_assignable
+
+} // end is_assignable_ns
+
+
+template<typename T1, typename T2>
+  struct is_assignable
+    : integral_constant<
+        bool,
+        is_assignable_ns::is_assignable<T1,T2>::value
+      >
+{};
+
+
+template<typename T>
+  struct is_copy_assignable
+    : is_assignable<
+        typename add_reference<T>::type,
+        typename add_reference<typename add_const<T>::type>::type
+      >
+{};
+
+
+} // end detail
+
+} // end thrust
+
+#include <thrust/detail/type_traits/has_trivial_assign.h>
+
diff --git a/compat/thrust/detail/type_traits/algorithm/intermediate_type_from_function_and_iterators.h b/compat/thrust/detail/type_traits/algorithm/intermediate_type_from_function_and_iterators.h
new file mode 100644
index 0000000000..92767b5497
--- /dev/null
+++ b/compat/thrust/detail/type_traits/algorithm/intermediate_type_from_function_and_iterators.h
@@ -0,0 +1,61 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/type_traits.h>
+#include <thrust/detail/type_traits/function_traits.h>
+#include <thrust/detail/type_traits/iterator/is_output_iterator.h>
+
+namespace thrust
+{
+
+namespace detail
+{
+
+// this trait reports what type should be used as a temporary in certain algorithms
+// which aggregate intermediate results from a function before writing to an output iterator
+
+// the pseudocode for deducing the type of the temporary used below:
+// 
+// if Function is an AdaptableFunction
+//   result = Function::result_type
+// else if OutputIterator2 is a "pure" output iterator
+//   result = InputIterator2::value_type
+// else
+//   result = OutputIterator2::value_type
+//
+// XXX upon c++0x, TemporaryType needs to be:
+// result_of<BinaryFunction>::type
+template<typename InputIterator, typename OutputIterator, typename Function>
+  struct intermediate_type_from_function_and_iterators
+    : eval_if<
+        has_result_type<Function>::value,
+        result_type<Function>,
+        eval_if<
+          is_output_iterator<OutputIterator>::value,
+          thrust::iterator_value<InputIterator>,
+          thrust::iterator_value<OutputIterator>
+        >
+      >
+{
+}; // end intermediate_type_from_function_and_iterators
+
+} // end detail
+
+} // end thrust
+
diff --git a/compat/thrust/detail/type_traits/function_traits.h b/compat/thrust/detail/type_traits/function_traits.h
new file mode 100644
index 0000000000..39015c608d
--- /dev/null
+++ b/compat/thrust/detail/type_traits/function_traits.h
@@ -0,0 +1,96 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/type_traits.h>
+#include <thrust/detail/type_traits/has_nested_type.h>
+
+namespace thrust
+{
+
+// forward definitions for is_commutative
+template <typename T> struct plus;
+template <typename T> struct multiplies;
+template <typename T> struct minimum;
+template <typename T> struct maximum;
+template <typename T> struct logical_or;
+template <typename T> struct logical_and;
+template <typename T> struct bit_or;
+template <typename T> struct bit_and;
+template <typename T> struct bit_xor;
+
+namespace detail
+{
+
+
+// some metafunctions which check for the nested types of the adaptable functions
+
+__THRUST_DEFINE_HAS_NESTED_TYPE(has_result_type, result_type)
+
+__THRUST_DEFINE_HAS_NESTED_TYPE(has_argument_type, argument_type)
+
+__THRUST_DEFINE_HAS_NESTED_TYPE(has_first_argument_type, first_argument_type)
+
+__THRUST_DEFINE_HAS_NESTED_TYPE(has_second_argument_type, second_argument_type)
+
+
+template<typename AdaptableBinaryFunction>
+  struct result_type
+{
+  typedef typename AdaptableBinaryFunction::result_type type;
+};
+
+
+template<typename T>
+  struct is_adaptable_unary_function
+    : thrust::detail::and_<
+        has_result_type<T>,
+        has_argument_type<T>
+      >
+{};
+
+
+template<typename T>
+  struct is_adaptable_binary_function
+    : thrust::detail::and_<
+        has_result_type<T>,
+        thrust::detail::and_<
+          has_first_argument_type<T>,
+          has_second_argument_type<T>
+        >
+      >
+{};
+
+
+template<typename BinaryFunction>
+  struct is_commutative
+    : public thrust::detail::false_type
+{};
+
+template<typename T> struct is_commutative< typename thrust::plus<T>        > : public thrust::detail::is_arithmetic<T> {};
+template<typename T> struct is_commutative< typename thrust::multiplies<T>  > : public thrust::detail::is_arithmetic<T> {};
+template<typename T> struct is_commutative< typename thrust::minimum<T>     > : public thrust::detail::is_arithmetic<T> {};
+template<typename T> struct is_commutative< typename thrust::maximum<T>     > : public thrust::detail::is_arithmetic<T> {};
+template<typename T> struct is_commutative< typename thrust::logical_or<T>  > : public thrust::detail::is_arithmetic<T> {};
+template<typename T> struct is_commutative< typename thrust::logical_and<T> > : public thrust::detail::is_arithmetic<T> {};
+template<typename T> struct is_commutative< typename thrust::bit_or<T>      > : public thrust::detail::is_arithmetic<T> {};
+template<typename T> struct is_commutative< typename thrust::bit_and<T>     > : public thrust::detail::is_arithmetic<T> {};
+template<typename T> struct is_commutative< typename thrust::bit_xor<T>     > : public thrust::detail::is_arithmetic<T> {};
+
+} // end namespace detail
+} // end namespace thrust
+
diff --git a/compat/thrust/detail/type_traits/has_member_function.h b/compat/thrust/detail/type_traits/has_member_function.h
new file mode 100644
index 0000000000..117f4cb9bf
--- /dev/null
+++ b/compat/thrust/detail/type_traits/has_member_function.h
@@ -0,0 +1,118 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/type_traits.h>
+
+#define __THRUST_DEFINE_HAS_MEMBER_FUNCTION(trait_name, member_function_name)                                \
+template<typename T, typename Signature> class trait_name;                                                   \
+                                                                                                             \
+template<typename T, typename Result>                                                                        \
+class trait_name<T, Result(void)>                                                                            \
+{                                                                                                            \
+   class yes { char m; };                                                                                    \
+   class no { yes m[2]; };                                                                                   \
+   struct base_mixin                                                                                         \
+   {                                                                                                         \
+     Result member_function_name();                                                                          \
+   };                                                                                                        \
+   struct base : public T, public base_mixin {};                                                             \
+   template <typename U, U t>  class helper{};                                                               \
+   template <typename U>                                                                                     \
+   static no deduce(U*, helper<Result (base_mixin::*)(), &U::member_function_name>* = 0);                    \
+   static yes deduce(...);                                                                                   \
+public:                                                                                                      \
+   static const bool value = sizeof(yes) == sizeof(deduce(static_cast<base*>(0)));                           \
+   typedef thrust::detail::integral_constant<bool,value> type;                                               \
+};                                                                                                           \
+                                                                                                             \
+template<typename T, typename Result, typename Arg>                                                          \
+class trait_name<T, Result(Arg)>                                                                             \
+{                                                                                                            \
+   class yes { char m; };                                                                                    \
+   class no { yes m[2]; };                                                                                   \
+   struct base_mixin                                                                                         \
+   {                                                                                                         \
+     Result member_function_name(Arg);                                                                       \
+   };                                                                                                        \
+   struct base : public T, public base_mixin {};                                                             \
+   template <typename U, U t>  class helper{};                                                               \
+   template <typename U>                                                                                     \
+   static no deduce(U*, helper<Result (base_mixin::*)(Arg), &U::member_function_name>* = 0);                 \
+   static yes deduce(...);                                                                                   \
+public:                                                                                                      \
+   static const bool value = sizeof(yes) == sizeof(deduce(static_cast<base*>(0)));                           \
+   typedef thrust::detail::integral_constant<bool,value> type;                                               \
+};                                                                                                           \
+                                                                                                             \
+template<typename T, typename Result, typename Arg1, typename Arg2>                                          \
+class trait_name<T, Result(Arg1,Arg2)>                                                                       \
+{                                                                                                            \
+   class yes { char m; };                                                                                    \
+   class no { yes m[2]; };                                                                                   \
+   struct base_mixin                                                                                         \
+   {                                                                                                         \
+     Result member_function_name(Arg1,Arg2);                                                                 \
+   };                                                                                                        \
+   struct base : public T, public base_mixin {};                                                             \
+   template <typename U, U t>  class helper{};                                                               \
+   template <typename U>                                                                                     \
+   static no deduce(U*, helper<Result (base_mixin::*)(Arg1,Arg2), &U::member_function_name>* = 0);           \
+   static yes deduce(...);                                                                                   \
+public:                                                                                                      \
+   static const bool value = sizeof(yes) == sizeof(deduce(static_cast<base*>(0)));                           \
+   typedef thrust::detail::integral_constant<bool,value> type;                                               \
+};                                                                                                           \
+                                                                                                             \
+template<typename T, typename Result, typename Arg1, typename Arg2, typename Arg3>                           \
+class trait_name<T, Result(Arg1,Arg2,Arg3)>                                                                  \
+{                                                                                                            \
+   class yes { char m; };                                                                                    \
+   class no { yes m[2]; };                                                                                   \
+   struct base_mixin                                                                                         \
+   {                                                                                                         \
+     Result member_function_name(Arg1,Arg2,Arg3);                                                            \
+   };                                                                                                        \
+   struct base : public T, public base_mixin {};                                                             \
+   template <typename U, U t>  class helper{};                                                               \
+   template <typename U>                                                                                     \
+   static no deduce(U*, helper<Result (base_mixin::*)(Arg1,Arg2,Arg3), &U::member_function_name>* = 0);      \
+   static yes deduce(...);                                                                                   \
+public:                                                                                                      \
+   static const bool value = sizeof(yes) == sizeof(deduce(static_cast<base*>(0)));                           \
+   typedef thrust::detail::integral_constant<bool,value> type;                                               \
+};                                                                                                           \
+                                                                                                             \
+template<typename T, typename Result, typename Arg1, typename Arg2, typename Arg3, typename Arg4>            \
+class trait_name<T, Result(Arg1,Arg2,Arg3,Arg4)>                                                             \
+{                                                                                                            \
+   class yes { char m; };                                                                                    \
+   class no { yes m[2]; };                                                                                   \
+   struct base_mixin                                                                                         \
+   {                                                                                                         \
+     Result member_function_name(Arg1,Arg2,Arg3,Arg4);                                                       \
+   };                                                                                                        \
+   struct base : public T, public base_mixin {};                                                             \
+   template <typename U, U t>  class helper{};                                                               \
+   template <typename U>                                                                                     \
+   static no deduce(U*, helper<Result (base_mixin::*)(Arg1,Arg2,Arg3,Arg4), &U::member_function_name>* = 0); \
+   static yes deduce(...);                                                                                   \
+public:                                                                                                      \
+   static const bool value = sizeof(yes) == sizeof(deduce(static_cast<base*>(0)));                           \
+   typedef thrust::detail::integral_constant<bool,value> type;                                               \
+};                                                                                                           
+
diff --git a/compat/thrust/detail/type_traits/has_nested_type.h b/compat/thrust/detail/type_traits/has_nested_type.h
new file mode 100644
index 0000000000..98c9460500
--- /dev/null
+++ b/compat/thrust/detail/type_traits/has_nested_type.h
@@ -0,0 +1,32 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/type_traits.h>
+
+#define __THRUST_DEFINE_HAS_NESTED_TYPE(trait_name, nested_type_name) \
+template<typename T> \
+  struct trait_name  \
+{                    \
+  typedef char yes_type; \
+  typedef int  no_type;  \
+  template<typename S> static yes_type test(typename S::nested_type_name *); \
+  template<typename S> static no_type  test(...); \
+  static bool const value = sizeof(test<T>(0)) == sizeof(yes_type);\
+  typedef thrust::detail::integral_constant<bool, value> type;\
+};
+
diff --git a/compat/thrust/detail/type_traits/has_trivial_assign.h b/compat/thrust/detail/type_traits/has_trivial_assign.h
new file mode 100644
index 0000000000..d248245e84
--- /dev/null
+++ b/compat/thrust/detail/type_traits/has_trivial_assign.h
@@ -0,0 +1,52 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file type_traits.h
+ *  \brief Temporarily define some type traits
+ *         until nvcc can compile tr1::type_traits.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/type_traits.h>
+
+namespace thrust
+{
+
+namespace detail
+{
+
+template<typename T> struct has_trivial_assign
+  : public integral_constant<
+      bool,
+      (is_pod<T>::value && !is_const<T>::value)
+#if THRUST_HOST_COMPILER == THRUST_HOST_COMPILER_MSVC
+      || __has_trivial_assign(T)
+#elif THRUST_HOST_COMPILER == THRUST_HOST_COMPILER_GCC
+// only use the intrinsic for >= 4.3
+#if (__GNUC__ >= 4) && (__GNUC_MINOR__ >= 3)
+      || __has_trivial_assign(T)
+#endif // GCC VERSION
+#endif // THRUST_HOST_COMPILER
+    >
+{};
+
+} // end detail
+
+} // end thrust
+
diff --git a/compat/thrust/detail/type_traits/is_call_possible.h b/compat/thrust/detail/type_traits/is_call_possible.h
new file mode 100644
index 0000000000..41b9539e19
--- /dev/null
+++ b/compat/thrust/detail/type_traits/is_call_possible.h
@@ -0,0 +1,161 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/type_traits.h>
+#include <thrust/detail/type_traits/has_member_function.h>
+
+// inspired by Roman Perepelitsa's presentation from comp.lang.c++.moderated
+// based on the implementation here: http://www.rsdn.ru/forum/cpp/2759773.1.aspx
+
+namespace thrust
+{
+namespace detail
+{
+namespace is_call_possible_detail
+{
+
+template<typename T> class void_exp_result {}; 
+
+template<typename T, typename U> 
+U const& operator,(U const&, void_exp_result<T>); 
+
+template<typename T, typename U> 
+U& operator,(U&, void_exp_result<T>); 
+
+template<typename src_type, typename dest_type> 
+struct clone_constness 
+{
+  typedef dest_type type; 
+}; 
+
+template<typename src_type, typename dest_type> 
+struct clone_constness<const src_type, dest_type> 
+{ 
+  typedef const dest_type type; 
+};
+
+} // end is_call_possible_detail
+} // end detail
+} // end thrust
+
+#define __THRUST_DEFINE_IS_CALL_POSSIBLE(trait_name, member_function_name)                                                                \
+__THRUST_DEFINE_HAS_MEMBER_FUNCTION(trait_name##_has_member, member_function_name)                                                        \
+                                                                                                                                          \
+template <typename T, typename Signature>                                                                                                 \
+struct trait_name                                                                                                                         \
+{                                                                                                                                         \
+  private:                                                                                                                                \
+    struct yes {};                                                                                                                        \
+    struct no { yes m[2]; };                                                                                                              \
+    struct derived : public T                                                                                                             \
+    {                                                                                                                                     \
+      using T::member_function_name;                                                                                                      \
+      no member_function_name(...) const;                                                                                                 \
+    };                                                                                                                                    \
+                                                                                                                                          \
+    typedef typename thrust::detail::is_call_possible_detail::clone_constness<T, derived>::type derived_type;                             \
+                                                                                                                                          \
+    template<typename U, typename Result>                                                                                                 \
+    struct return_value_check                                                                                                             \
+    {                                                                                                                                     \
+      static yes deduce(Result);                                                                                                          \
+      static no deduce(...);                                                                                                              \
+      static no deduce(no);                                                                                                               \
+      static no deduce(thrust::detail::is_call_possible_detail::void_exp_result<T>);                                                      \
+    };                                                                                                                                    \
+                                                                                                                                          \
+    template<typename U>                                                                                                                  \
+    struct return_value_check<U, void>                                                                                                    \
+    {                                                                                                                                     \
+      static yes deduce(...);                                                                                                             \
+      static no deduce(no);                                                                                                               \
+    };                                                                                                                                    \
+                                                                                                                                          \
+    template<bool has_the_member_of_interest, typename F>                                                                                 \
+    struct impl                                                                                                                           \
+    {                                                                                                                                     \
+      static const bool value = false;                                                                                                    \
+    };                                                                                                                                    \
+                                                                                                                                          \
+    template<typename Result, typename Arg>                                                                                               \
+    struct impl<true, Result(Arg)>                                                                                                        \
+    {                                                                                                                                     \
+      static typename add_reference<derived_type>::type test_me;                                                                          \
+      static typename add_reference<Arg>::type          arg;                                                                              \
+                                                                                                                                          \
+      static const bool value =                                                                                                           \
+        sizeof(                                                                                                                           \
+                return_value_check<T, Result>::deduce(                                                                                    \
+                  (test_me.member_function_name(arg), thrust::detail::is_call_possible_detail::void_exp_result<T>())                      \
+                )                                                                                                                         \
+              ) == sizeof(yes);                                                                                                           \
+    };                                                                                                                                    \
+                                                                                                                                          \
+    template<typename Result, typename Arg1, typename Arg2>                                                                               \
+    struct impl<true, Result(Arg1,Arg2)>                                                                                                  \
+    {                                                                                                                                     \
+      static typename add_reference<derived_type>::type test_me;                                                                          \
+      static typename add_reference<Arg1>::type         arg1;                                                                             \
+      static typename add_reference<Arg2>::type         arg2;                                                                             \
+                                                                                                                                          \
+      static const bool value =                                                                                                           \
+        sizeof(                                                                                                                           \
+                return_value_check<T, Result>::deduce(                                                                                    \
+                  (test_me.member_function_name(arg1,arg2), thrust::detail::is_call_possible_detail::void_exp_result<T>())                \
+                )                                                                                                                         \
+              ) == sizeof(yes);                                                                                                           \
+    };                                                                                                                                    \
+                                                                                                                                          \
+    template<typename Result, typename Arg1, typename Arg2, typename Arg3>                                                                \
+    struct impl<true, Result(Arg1,Arg2,Arg3)>                                                                                             \
+    {                                                                                                                                     \
+      static typename add_reference<derived_type>::type test_me;                                                                          \
+      static typename add_reference<Arg1>::type         arg1;                                                                             \
+      static typename add_reference<Arg2>::type         arg2;                                                                             \
+      static typename add_reference<Arg3>::type         arg3;                                                                             \
+                                                                                                                                          \
+      static const bool value =                                                                                                           \
+        sizeof(                                                                                                                           \
+                return_value_check<T, Result>::deduce(                                                                                    \
+                  (test_me.member_function_name(arg1,arg2,arg3), thrust::detail::is_call_possible_detail::void_exp_result<T>())           \
+                )                                                                                                                         \
+              ) == sizeof(yes);                                                                                                           \
+    };                                                                                                                                    \
+                                                                                                                                          \
+    template<typename Result, typename Arg1, typename Arg2, typename Arg3, typename Arg4>                                                 \
+    struct impl<true, Result(Arg1,Arg2,Arg3,Arg4)>                                                                                        \
+    {                                                                                                                                     \
+      static typename add_reference<derived_type>::type test_me;                                                                          \
+      static typename add_reference<Arg1>::type         arg1;                                                                             \
+      static typename add_reference<Arg2>::type         arg2;                                                                             \
+      static typename add_reference<Arg3>::type         arg3;                                                                             \
+      static typename add_reference<Arg4>::type         arg4;                                                                             \
+                                                                                                                                          \
+      static const bool value =                                                                                                           \
+        sizeof(                                                                                                                           \
+                return_value_check<T, Result>::deduce(                                                                                    \
+                  (test_me.member_function_name(arg1,arg2,arg3,arg4), thrust::detail::is_call_possible_detail::void_exp_result<T>())      \
+                )                                                                                                                         \
+              ) == sizeof(yes);                                                                                                           \
+    };                                                                                                                                    \
+                                                                                                                                          \
+  public:                                                                                                                                 \
+    static const bool value = impl<trait_name##_has_member<T,Signature>::value, Signature>::value;                                        \
+    typedef thrust::detail::integral_constant<bool,value> type;                                                                           \
+}; 
+
diff --git a/compat/thrust/detail/type_traits/is_metafunction_defined.h b/compat/thrust/detail/type_traits/is_metafunction_defined.h
new file mode 100644
index 0000000000..fba0811fb3
--- /dev/null
+++ b/compat/thrust/detail/type_traits/is_metafunction_defined.h
@@ -0,0 +1,41 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/type_traits/has_nested_type.h>
+#include <thrust/detail/type_traits.h>
+
+namespace thrust
+{
+
+namespace detail
+{
+
+__THRUST_DEFINE_HAS_NESTED_TYPE(is_metafunction_defined, type)
+
+template<typename Metafunction>
+  struct enable_if_defined
+    : thrust::detail::lazy_enable_if<
+        is_metafunction_defined<Metafunction>::value,
+        Metafunction
+      >
+{};
+
+} // end detail
+
+} // end thrust
+
diff --git a/compat/thrust/detail/type_traits/iterator/is_discard_iterator.h b/compat/thrust/detail/type_traits/iterator/is_discard_iterator.h
new file mode 100644
index 0000000000..cca59da045
--- /dev/null
+++ b/compat/thrust/detail/type_traits/iterator/is_discard_iterator.h
@@ -0,0 +1,40 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/type_traits.h>
+#include <thrust/iterator/discard_iterator.h>
+
+namespace thrust
+{
+namespace detail
+{
+
+template <typename Iterator>
+struct is_discard_iterator
+  : public thrust::detail::false_type
+{};
+
+template <typename System>
+struct is_discard_iterator< thrust::discard_iterator<System> >
+ : public thrust::detail::true_type
+{};
+
+} // end namespace detail
+} // end namespace thrust
+
diff --git a/compat/thrust/detail/type_traits/iterator/is_output_iterator.h b/compat/thrust/detail/type_traits/iterator/is_output_iterator.h
new file mode 100644
index 0000000000..4cefe6353a
--- /dev/null
+++ b/compat/thrust/detail/type_traits/iterator/is_output_iterator.h
@@ -0,0 +1,66 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/type_traits.h>
+#include <thrust/detail/type_traits/is_metafunction_defined.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/iterator/detail/any_assign.h>
+
+namespace thrust
+{
+
+namespace detail
+{
+
+
+template<typename T>
+  struct is_void_like
+    : thrust::detail::or_<
+        thrust::detail::is_void<T>,
+        thrust::detail::is_same<T,thrust::detail::any_assign>
+      >
+{}; // end is_void_like
+
+
+template<typename T>
+  struct lazy_is_void_like
+    : is_void_like<typename T::type>
+{}; // end lazy_is_void_like
+
+
+// XXX this meta function should first check that T is actually an iterator
+//
+//     if thrust::iterator_value<T> is defined and thrust::iterator_value<T>::type == void
+//       return false
+//     else
+//       return true
+template<typename T>
+  struct is_output_iterator
+    : eval_if<
+        is_metafunction_defined<thrust::iterator_value<T> >::value,
+        lazy_is_void_like<thrust::iterator_value<T> >,
+        thrust::detail::true_type
+      >::type
+{
+}; // end is_output_iterator
+
+} // end detail
+
+} // end thrust
+
diff --git a/compat/thrust/detail/type_traits/minimum_type.h b/compat/thrust/detail/type_traits/minimum_type.h
new file mode 100644
index 0000000000..aaa011ec8d
--- /dev/null
+++ b/compat/thrust/detail/type_traits/minimum_type.h
@@ -0,0 +1,162 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/type_traits.h>
+
+namespace thrust
+{
+
+namespace detail
+{ 
+
+namespace minimum_type_detail
+{
+
+//
+// Returns the minimum type or is empty
+// if T1 and T2 are unrelated.
+//
+template <typename T1, typename T2, bool GreaterEqual, bool LessEqual> struct minimum_type_impl {};
+  
+template <typename T1, typename T2>
+struct minimum_type_impl<T1,T2,true,false>
+{
+  typedef T2 type;
+}; // end minimum_type_impl
+
+template <typename T1, typename T2>
+struct minimum_type_impl<T1,T2,false,true>
+{
+  typedef T1 type;
+}; // end minimum_type_impl
+
+template <typename T1, typename T2>
+struct minimum_type_impl<T1,T2,true,true>
+{
+  typedef T1 type;
+}; // end minimum_type_impl
+
+template <typename T1, typename T2>
+struct primitive_minimum_type
+  : minimum_type_detail::minimum_type_impl<
+      T1,
+      T2,
+      ::thrust::detail::is_convertible<T1,T2>::value,
+      ::thrust::detail::is_convertible<T2,T1>::value
+    >
+{
+}; // end primitive_minimum_type
+
+// because some types are not convertible (even to themselves)
+// specialize primitive_minimum_type for when both types are identical
+template <typename T>
+struct primitive_minimum_type<T,T>
+{
+  typedef T type;
+}; // end primitive_minimum_type
+
+// XXX this belongs somewhere more general
+struct any_conversion
+{
+  template<typename T> operator T (void);
+};
+
+} // end minimum_type_detail
+
+template<typename T1,
+         typename T2  = minimum_type_detail::any_conversion,
+         typename T3  = minimum_type_detail::any_conversion,
+         typename T4  = minimum_type_detail::any_conversion,
+         typename T5  = minimum_type_detail::any_conversion,
+         typename T6  = minimum_type_detail::any_conversion,
+         typename T7  = minimum_type_detail::any_conversion,
+         typename T8  = minimum_type_detail::any_conversion,
+         typename T9  = minimum_type_detail::any_conversion,
+         typename T10 = minimum_type_detail::any_conversion,
+         typename T11 = minimum_type_detail::any_conversion,
+         typename T12 = minimum_type_detail::any_conversion,
+         typename T13 = minimum_type_detail::any_conversion,
+         typename T14 = minimum_type_detail::any_conversion,
+         typename T15 = minimum_type_detail::any_conversion,
+         typename T16 = minimum_type_detail::any_conversion>
+  struct minimum_type;
+
+// base case
+template<typename T1, typename T2>
+  struct minimum_type<T1,T2>
+    : minimum_type_detail::primitive_minimum_type<T1,T2>
+{};
+
+template<typename T1, typename T2>
+  struct lazy_minimum_type
+    : minimum_type<
+        typename T1::type,
+        typename T2::type
+      >
+{};
+
+// carefully avoid referring to a nested ::type which may not exist
+template<typename T1,  typename T2,  typename T3,  typename T4,
+         typename T5,  typename T6,  typename T7,  typename T8,
+         typename T9,  typename T10, typename T11, typename T12,
+         typename T13, typename T14, typename T15, typename T16>
+  struct minimum_type
+    : lazy_minimum_type<
+        lazy_minimum_type<
+          lazy_minimum_type<
+            minimum_type<
+              T1,T2
+            >,
+            minimum_type<
+              T3,T4
+            >
+          >,
+          lazy_minimum_type<
+            minimum_type<
+              T5,T6
+            >,
+            minimum_type<
+              T7,T8
+            >
+          >
+        >,
+        lazy_minimum_type<
+          lazy_minimum_type<
+            minimum_type<
+              T9,T10
+            >,
+            minimum_type<
+              T11,T12
+            >
+          >,
+          lazy_minimum_type<
+            minimum_type<
+              T13,T14
+            >,
+            minimum_type<
+              T15,T16
+            >
+          >
+        >
+      >
+{};
+
+} // end detail
+
+} // end thrust
+
diff --git a/compat/thrust/detail/type_traits/pointer_traits.h b/compat/thrust/detail/type_traits/pointer_traits.h
new file mode 100644
index 0000000000..a0b5dc625c
--- /dev/null
+++ b/compat/thrust/detail/type_traits/pointer_traits.h
@@ -0,0 +1,276 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/type_traits.h>
+#include <thrust/detail/type_traits/is_metafunction_defined.h>
+#include <thrust/detail/type_traits/has_nested_type.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <cstddef>
+
+namespace thrust
+{
+namespace detail
+{
+
+template<typename Ptr> struct pointer_element;
+
+template<template<typename> class Ptr, typename Arg>
+  struct pointer_element<Ptr<Arg> >
+{
+  typedef Arg type;
+};
+
+template<template<typename,typename> class Ptr, typename Arg1, typename Arg2>
+  struct pointer_element<Ptr<Arg1,Arg2> >
+{
+  typedef Arg1 type;
+};
+
+template<template<typename,typename,typename> class Ptr, typename Arg1, typename Arg2, typename Arg3>
+  struct pointer_element<Ptr<Arg1,Arg2,Arg3> >
+{
+  typedef Arg1 type;
+};
+
+template<template<typename,typename,typename,typename> class Ptr, typename Arg1, typename Arg2, typename Arg3, typename Arg4>
+  struct pointer_element<Ptr<Arg1,Arg2,Arg3,Arg4> >
+{
+  typedef Arg1 type;
+};
+
+template<template<typename,typename,typename,typename,typename> class Ptr, typename Arg1, typename Arg2, typename Arg3, typename Arg4, typename Arg5>
+  struct pointer_element<Ptr<Arg1,Arg2,Arg3,Arg4,Arg5> >
+{
+  typedef Arg1 type;
+};
+
+template<typename T>
+  struct pointer_element<T*>
+{
+  typedef T type;
+};
+
+template<typename Ptr>
+  struct pointer_difference
+{
+  typedef typename Ptr::difference_type type;
+};
+
+template<typename T>
+  struct pointer_difference<T*>
+{
+  typedef std::ptrdiff_t type;
+};
+
+template<typename Ptr, typename T> struct rebind_pointer;
+
+template<typename T, typename U>
+  struct rebind_pointer<T*,U>
+{
+  typedef U* type;
+};
+
+template<template<typename> class Ptr, typename Arg, typename T>
+  struct rebind_pointer<Ptr<Arg>,T>
+{
+  typedef Ptr<T> type;
+};
+
+template<template<typename, typename> class Ptr, typename Arg1, typename Arg2, typename T>
+  struct rebind_pointer<Ptr<Arg1,Arg2>,T>
+{
+  typedef Ptr<T,Arg2> type;
+};
+
+template<template<typename, typename, typename> class Ptr, typename Arg1, typename Arg2, typename Arg3, typename T>
+  struct rebind_pointer<Ptr<Arg1,Arg2,Arg3>,T>
+{
+  typedef Ptr<T,Arg2,Arg3> type;
+};
+
+template<template<typename, typename, typename, typename> class Ptr, typename Arg1, typename Arg2, typename Arg3, typename Arg4, typename T>
+  struct rebind_pointer<Ptr<Arg1,Arg2,Arg3,Arg4>,T>
+{
+  typedef Ptr<T,Arg2,Arg3,Arg4> type;
+};
+
+// XXX this should probably be renamed native_type or similar
+__THRUST_DEFINE_HAS_NESTED_TYPE(has_raw_pointer, raw_pointer)
+
+namespace pointer_traits_detail
+{
+
+template<typename Ptr, typename Enable = void> struct pointer_raw_pointer_impl {};
+
+template<typename T>
+  struct pointer_raw_pointer_impl<T*>
+{
+  typedef T* type;
+};
+
+template<typename Ptr>
+  struct pointer_raw_pointer_impl<Ptr, typename enable_if<has_raw_pointer<Ptr>::value>::type>
+{
+  typedef typename Ptr::raw_pointer type;
+};
+
+} // end pointer_traits_detail
+
+template<typename T>
+  struct pointer_raw_pointer
+    : pointer_traits_detail::pointer_raw_pointer_impl<T>
+{};
+
+namespace pointer_traits_detail
+{
+
+template<typename Void>
+  struct capture_address
+{
+  template<typename T>
+  __host__ __device__
+  capture_address(T &r)
+    : m_addr(&r)
+  {}
+
+  inline __host__ __device__
+  Void *operator&() const
+  {
+    return m_addr;
+  }
+
+  Void *m_addr;
+};
+
+// metafunction to compute the type of pointer_to's parameter below
+template<typename T>
+  struct pointer_to_param
+    : thrust::detail::eval_if<
+        thrust::detail::is_void<T>::value,
+        thrust::detail::identity_<capture_address<T> >,
+        thrust::detail::add_reference<T>
+      >
+{};
+
+}
+
+template<typename Ptr>
+  struct pointer_traits
+{
+  typedef Ptr                                    pointer;
+  typedef typename pointer_element<Ptr>::type    element_type;
+  typedef typename pointer_difference<Ptr>::type difference_type;
+
+  template<typename U>
+    struct rebind 
+  {
+    typedef typename rebind_pointer<Ptr,U>::type other;
+  };
+
+  __host__ __device__
+  inline static pointer pointer_to(typename pointer_traits_detail::pointer_to_param<element_type>::type r)
+  {
+    // XXX this is supposed to be pointer::pointer_to(&r); (i.e., call a static member function of pointer called pointer_to)
+    //     assume that pointer has a constructor from raw pointer instead
+    
+    return pointer(&r);
+  }
+
+  // thrust additions follow
+  typedef typename pointer_raw_pointer<Ptr>::type raw_pointer;
+
+  __host__ __device__
+  inline static raw_pointer get(pointer ptr)
+  {
+    return ptr.get();
+  }
+};
+
+template<typename T>
+  struct pointer_traits<T*>
+{
+  typedef T*                                    pointer;
+  typedef T                                     element_type;
+  typedef typename pointer_difference<T*>::type difference_type;
+
+  template<typename U>
+    struct rebind
+  {
+    typedef U* other;
+  };
+
+  __host__ __device__
+  inline static pointer pointer_to(typename pointer_traits_detail::pointer_to_param<element_type>::type r)
+  {
+    return &r;
+  }
+
+  // thrust additions follow
+  typedef typename pointer_raw_pointer<T*>::type raw_pointer;
+
+  __host__ __device__
+  inline static raw_pointer get(pointer ptr)
+  {
+    return ptr;
+  }
+};
+
+template<typename FromPtr, typename ToPtr>
+  struct is_pointer_convertible
+    : thrust::detail::and_<
+        thrust::detail::is_convertible<
+          typename pointer_element<FromPtr>::type *,
+          typename pointer_element<ToPtr>::type *
+        >,
+        thrust::detail::is_convertible<
+          typename iterator_system<FromPtr>::type,
+          typename iterator_system<ToPtr>::type
+        >
+      >
+{};
+
+// this could be a lot better, but for our purposes, it's probably
+// sufficient just to check if pointer_raw_pointer<T> has meaning
+template<typename T>
+  struct is_thrust_pointer
+    : is_metafunction_defined<pointer_raw_pointer<T> >
+{};
+
+// avoid inspecting traits of the arguments if they aren't known to be pointers
+template<typename FromPtr, typename ToPtr>
+  struct lazy_is_pointer_convertible
+    : thrust::detail::eval_if<
+        is_thrust_pointer<FromPtr>::value && is_thrust_pointer<ToPtr>::value,
+        is_pointer_convertible<FromPtr,ToPtr>,
+        thrust::detail::identity_<thrust::detail::false_type>
+      >
+{};
+
+template<typename FromPtr, typename ToPtr, typename T = void>
+  struct enable_if_pointer_is_convertible
+    : thrust::detail::enable_if<
+        lazy_is_pointer_convertible<FromPtr,ToPtr>::type::value,
+        T
+      >
+{};
+
+
+} // end detail
+} // end thrust
+
diff --git a/compat/thrust/detail/type_traits/result_of.h b/compat/thrust/detail/type_traits/result_of.h
new file mode 100644
index 0000000000..e30b4fda3b
--- /dev/null
+++ b/compat/thrust/detail/type_traits/result_of.h
@@ -0,0 +1,52 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/type_traits.h>
+#include <thrust/detail/type_traits/function_traits.h>
+
+namespace thrust
+{
+namespace detail
+{
+
+template<typename Signature, typename Enable = void> struct result_of;
+
+// specialization for unary invocations of things which have result_type
+template<typename Functor, typename Arg1>
+  struct result_of<
+    Functor(Arg1),
+    typename thrust::detail::enable_if<thrust::detail::has_result_type<Functor>::value>::type
+  >
+{
+  typedef typename Functor::result_type type;
+}; // end result_of
+
+// specialization for binary invocations of things which have result_type
+template<typename Functor, typename Arg1, typename Arg2>
+  struct result_of<
+    Functor(Arg1,Arg2),
+    typename thrust::detail::enable_if<thrust::detail::has_result_type<Functor>::value>::type
+  >
+{
+  typedef typename Functor::result_type type;
+};
+
+} // end detail
+} // end thrust
+
diff --git a/compat/thrust/detail/uninitialized_copy.inl b/compat/thrust/detail/uninitialized_copy.inl
new file mode 100644
index 0000000000..a01dca53d6
--- /dev/null
+++ b/compat/thrust/detail/uninitialized_copy.inl
@@ -0,0 +1,93 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file uninitialized_copy.inl
+ *  \brief Inline file for uninitialized_copy.h.
+ */
+
+#include <thrust/uninitialized_copy.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/system/detail/generic/select_system.h>
+#include <thrust/system/detail/generic/uninitialized_copy.h>
+#include <thrust/system/detail/adl/uninitialized_copy.h>
+
+namespace thrust
+{
+
+
+template<typename DerivedPolicy, typename InputIterator, typename ForwardIterator>
+  ForwardIterator uninitialized_copy(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                     InputIterator first,
+                                     InputIterator last,
+                                     ForwardIterator result)
+{
+  using thrust::system::detail::generic::uninitialized_copy;
+  return uninitialized_copy(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, result);
+} // end uninitialized_copy()
+
+
+template<typename DerivedPolicy, typename InputIterator, typename Size, typename ForwardIterator>
+  ForwardIterator uninitialized_copy_n(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                       InputIterator first,
+                                       Size n,
+                                       ForwardIterator result)
+{
+  using thrust::system::detail::generic::uninitialized_copy_n;
+  return uninitialized_copy_n(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, n, result);
+} // end uninitialized_copy_n()
+
+
+template<typename InputIterator,
+         typename ForwardIterator>
+  ForwardIterator uninitialized_copy(InputIterator first,
+                                     InputIterator last,
+                                     ForwardIterator result)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator>::type   System1;
+  typedef typename thrust::iterator_system<ForwardIterator>::type System2;
+
+  System1 system1;
+  System2 system2;
+
+  return thrust::uninitialized_copy(select_system(system1,system2), first, last, result);
+} // end uninitialized_copy()
+
+
+template<typename InputIterator,
+         typename Size,
+         typename ForwardIterator>
+  ForwardIterator uninitialized_copy_n(InputIterator first,
+                                       Size n,
+                                       ForwardIterator result)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator>::type   System1;
+  typedef typename thrust::iterator_system<ForwardIterator>::type System2;
+
+  System1 system1;
+  System2 system2;
+
+  return thrust::uninitialized_copy_n(select_system(system1,system2), first, n, result);
+} // end uninitialized_copy_n()
+
+
+} // end thrust
+
+
diff --git a/compat/thrust/detail/uninitialized_fill.inl b/compat/thrust/detail/uninitialized_fill.inl
new file mode 100644
index 0000000000..3545de56ee
--- /dev/null
+++ b/compat/thrust/detail/uninitialized_fill.inl
@@ -0,0 +1,88 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file uninitialized_fill.inl
+ *  \brief Inline file for uninitialized_fill.h.
+ */
+
+#include <thrust/uninitialized_fill.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/system/detail/generic/select_system.h>
+#include <thrust/system/detail/generic/uninitialized_fill.h>
+#include <thrust/system/detail/adl/uninitialized_fill.h>
+
+namespace thrust
+{
+
+
+template<typename DerivedPolicy, typename ForwardIterator, typename T>
+  void uninitialized_fill(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                          ForwardIterator first,
+                          ForwardIterator last,
+                          const T &x)
+{
+  using thrust::system::detail::generic::uninitialized_fill;
+  return uninitialized_fill(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, x);
+} // end uninitialized_fill()
+
+
+template<typename DerivedPolicy, typename ForwardIterator, typename Size, typename T>
+  ForwardIterator uninitialized_fill_n(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                       ForwardIterator first,
+                                       Size n,
+                                       const T &x)
+{
+  using thrust::system::detail::generic::uninitialized_fill_n;
+  return uninitialized_fill_n(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, n, x);
+} // end uninitialized_fill_n()
+
+
+template<typename ForwardIterator,
+         typename T>
+  void uninitialized_fill(ForwardIterator first,
+                          ForwardIterator last,
+                          const T &x)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<ForwardIterator>::type System;
+
+  System system;
+
+  thrust::uninitialized_fill(select_system(system), first, last, x);
+} // end uninitialized_fill()
+
+
+template<typename ForwardIterator,
+         typename Size,
+         typename T>
+  ForwardIterator uninitialized_fill_n(ForwardIterator first,
+                                       Size n,
+                                       const T &x)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<ForwardIterator>::type System;
+
+  System system;
+
+  return thrust::uninitialized_fill_n(select_system(system), first, n, x);
+} // end uninitialized_fill_n()
+
+
+} // end thrust
+
diff --git a/compat/thrust/detail/unique.inl b/compat/thrust/detail/unique.inl
new file mode 100644
index 0000000000..e90187d919
--- /dev/null
+++ b/compat/thrust/detail/unique.inl
@@ -0,0 +1,320 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file unique.inl
+ *  \brief Inline file for unique.h.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/unique.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/system/detail/generic/select_system.h>
+#include <thrust/system/detail/generic/unique.h>
+#include <thrust/system/detail/generic/unique_by_key.h>
+#include <thrust/system/detail/adl/unique.h>
+#include <thrust/system/detail/adl/unique_by_key.h>
+
+namespace thrust
+{
+
+
+template <typename DerivedPolicy,
+          typename ForwardIterator>
+ForwardIterator unique(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                       ForwardIterator first,
+                       ForwardIterator last)
+{
+  using thrust::system::detail::generic::unique;
+  return unique(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last);
+} // end unique()
+
+
+template <typename DerivedPolicy,
+          typename ForwardIterator,
+          typename BinaryPredicate>
+ForwardIterator unique(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                       ForwardIterator first,
+                       ForwardIterator last,
+                       BinaryPredicate binary_pred)
+{
+  using thrust::system::detail::generic::unique;
+  return unique(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, binary_pred);
+} // end unique()
+
+
+template <typename DerivedPolicy,
+          typename InputIterator,
+          typename OutputIterator>
+OutputIterator unique_copy(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                           InputIterator first,
+                           InputIterator last,
+                           OutputIterator output)
+{
+  using thrust::system::detail::generic::unique_copy;
+  return unique_copy(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, output);
+} // end unique_copy()
+
+
+template <typename DerivedPolicy,
+          typename InputIterator,
+          typename OutputIterator,
+          typename BinaryPredicate>
+OutputIterator unique_copy(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                           InputIterator first,
+                           InputIterator last,
+                           OutputIterator output,
+                           BinaryPredicate binary_pred)
+{
+  using thrust::system::detail::generic::unique_copy;
+  return unique_copy(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, output, binary_pred);
+} // end unique_copy()
+
+
+template <typename DerivedPolicy,
+          typename ForwardIterator1,
+          typename ForwardIterator2>
+  thrust::pair<ForwardIterator1,ForwardIterator2>
+  unique_by_key(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                ForwardIterator1 keys_first, 
+                ForwardIterator1 keys_last,
+                ForwardIterator2 values_first)
+{
+  using thrust::system::detail::generic::unique_by_key;
+  return unique_by_key(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), keys_first, keys_last, values_first);
+} // end unique_by_key()
+
+
+template <typename DerivedPolicy,
+          typename ForwardIterator1,
+          typename ForwardIterator2,
+          typename BinaryPredicate>
+  thrust::pair<ForwardIterator1,ForwardIterator2>
+  unique_by_key(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                ForwardIterator1 keys_first, 
+                ForwardIterator1 keys_last,
+                ForwardIterator2 values_first,
+                BinaryPredicate binary_pred)
+{
+  using thrust::system::detail::generic::unique_by_key;
+  return unique_by_key(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), keys_first, keys_last, values_first, binary_pred);
+} // end unique_by_key()
+
+
+template <typename DerivedPolicy,
+          typename InputIterator1,
+          typename InputIterator2,
+          typename OutputIterator1,
+          typename OutputIterator2>
+  thrust::pair<OutputIterator1,OutputIterator2>
+  unique_by_key_copy(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                     InputIterator1 keys_first, 
+                     InputIterator1 keys_last,
+                     InputIterator2 values_first,
+                     OutputIterator1 keys_output,
+                     OutputIterator2 values_output)
+{
+  using thrust::system::detail::generic::unique_by_key_copy;
+  return unique_by_key_copy(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), keys_first, keys_last, values_first, keys_output, values_output);
+} // end unique_by_key_copy()
+
+
+template <typename DerivedPolicy,
+          typename InputIterator1,
+          typename InputIterator2,
+          typename OutputIterator1,
+          typename OutputIterator2,
+          typename BinaryPredicate>
+  thrust::pair<OutputIterator1,OutputIterator2>
+  unique_by_key_copy(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                     InputIterator1 keys_first, 
+                     InputIterator1 keys_last,
+                     InputIterator2 values_first,
+                     OutputIterator1 keys_output,
+                     OutputIterator2 values_output,
+                     BinaryPredicate binary_pred)
+{
+  using thrust::system::detail::generic::unique_by_key_copy;
+  return unique_by_key_copy(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), keys_first, keys_last, values_first, keys_output, values_output, binary_pred);
+} // end unique_by_key_copy()
+
+
+template<typename ForwardIterator>
+  ForwardIterator unique(ForwardIterator first,
+                         ForwardIterator last)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<ForwardIterator>::type System;
+
+  System system;
+
+  return thrust::unique(select_system(system), first, last);
+} // end unique()
+
+
+template<typename ForwardIterator,
+         typename BinaryPredicate>
+  ForwardIterator unique(ForwardIterator first,
+                         ForwardIterator last,
+                         BinaryPredicate binary_pred)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<ForwardIterator>::type System;
+
+  System system;
+
+  return thrust::unique(select_system(system), first, last, binary_pred);
+} // end unique()
+
+
+template<typename InputIterator,
+         typename OutputIterator>
+  OutputIterator unique_copy(InputIterator first,
+                             InputIterator last,
+                             OutputIterator output)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator>::type  System1;
+  typedef typename thrust::iterator_system<OutputIterator>::type System2;
+
+  System1 system1;
+  System2 system2;
+
+  return thrust::unique_copy(select_system(system1,system2), first, last, output);
+} // end unique_copy()
+
+
+template<typename InputIterator,
+         typename OutputIterator,
+         typename BinaryPredicate>
+  OutputIterator unique_copy(InputIterator first,
+                             InputIterator last,
+                             OutputIterator output,
+                             BinaryPredicate binary_pred)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator>::type  System1;
+  typedef typename thrust::iterator_system<OutputIterator>::type System2;
+
+  System1 system1;
+  System2 system2;
+
+  return thrust::unique_copy(select_system(system1,system2), first, last, output, binary_pred);
+} // end unique_copy()
+
+
+template<typename ForwardIterator1,
+         typename ForwardIterator2>
+  thrust::pair<ForwardIterator1,ForwardIterator2>
+    unique_by_key(ForwardIterator1 keys_first, 
+                  ForwardIterator1 keys_last,
+                  ForwardIterator2 values_first)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<ForwardIterator1>::type System1;
+  typedef typename thrust::iterator_system<ForwardIterator2>::type System2;
+
+  System1 system1;
+  System2 system2;
+
+  return thrust::unique_by_key(select_system(system1,system2), keys_first, keys_last, values_first);
+} // end unique_by_key()
+
+
+template<typename ForwardIterator1,
+         typename ForwardIterator2,
+         typename BinaryPredicate>
+  thrust::pair<ForwardIterator1,ForwardIterator2>
+    unique_by_key(ForwardIterator1 keys_first, 
+                  ForwardIterator1 keys_last,
+                  ForwardIterator2 values_first,
+                  BinaryPredicate binary_pred)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<ForwardIterator1>::type System1;
+  typedef typename thrust::iterator_system<ForwardIterator2>::type System2;
+
+  System1 system1;
+  System2 system2;
+
+  return thrust::unique_by_key(select_system(system1,system2), keys_first, keys_last, values_first, binary_pred);
+} // end unique_by_key()
+
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator1,
+         typename OutputIterator2>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    unique_by_key_copy(InputIterator1 keys_first, 
+                       InputIterator1 keys_last,
+                       InputIterator2 values_first,
+                       OutputIterator1 keys_output,
+                       OutputIterator2 values_output)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator1>::type  System1;
+  typedef typename thrust::iterator_system<InputIterator2>::type  System2;
+  typedef typename thrust::iterator_system<OutputIterator1>::type System3;
+  typedef typename thrust::iterator_system<OutputIterator2>::type System4;
+
+  System1 system1;
+  System2 system2;
+  System3 system3;
+  System4 system4;
+
+  return thrust::unique_by_key_copy(select_system(system1,system2,system3,system4), keys_first, keys_last, values_first, keys_output, values_output);
+} // end unique_by_key_copy()
+
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename BinaryPredicate>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    unique_by_key_copy(InputIterator1 keys_first, 
+                       InputIterator1 keys_last,
+                       InputIterator2 values_first,
+                       OutputIterator1 keys_output,
+                       OutputIterator2 values_output,
+                       BinaryPredicate binary_pred)
+{
+  using thrust::system::detail::generic::select_system;
+
+  typedef typename thrust::iterator_system<InputIterator1>::type  System1;
+  typedef typename thrust::iterator_system<InputIterator2>::type  System2;
+  typedef typename thrust::iterator_system<OutputIterator1>::type System3;
+  typedef typename thrust::iterator_system<OutputIterator2>::type System4;
+
+  System1 system1;
+  System2 system2;
+  System3 system3;
+  System4 system4;
+
+  return thrust::unique_by_key_copy(select_system(system1,system2,system3,system4), keys_first, keys_last, values_first, keys_output, values_output, binary_pred);
+} // end unique_by_key_copy()
+
+
+} // end namespace thrust
+
diff --git a/compat/thrust/detail/use_default.h b/compat/thrust/detail/use_default.h
new file mode 100644
index 0000000000..c6eb66ef05
--- /dev/null
+++ b/compat/thrust/detail/use_default.h
@@ -0,0 +1,27 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+namespace thrust
+{
+
+struct use_default {};
+
+} // end thrust
+
diff --git a/compat/thrust/detail/util/align.h b/compat/thrust/detail/util/align.h
new file mode 100644
index 0000000000..10f107a95b
--- /dev/null
+++ b/compat/thrust/detail/util/align.h
@@ -0,0 +1,52 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+#pragma once
+
+#include <thrust/detail/cstdint.h>
+
+// functions to handle memory alignment
+
+namespace thrust
+{
+namespace detail
+{
+namespace util
+{
+
+template <typename T>
+T * align_up(T * ptr, detail::uintptr_t bytes)
+{
+    return (T *) ( bytes * (((detail::uintptr_t) ptr + (bytes - 1)) / bytes) );
+}
+
+template <typename T>
+T * align_down(T * ptr, detail::uintptr_t bytes)
+{
+    return (T *) ( bytes * (detail::uintptr_t(ptr) / bytes) );
+}
+
+template <typename T>
+bool is_aligned(T * ptr, detail::uintptr_t bytes = sizeof(T))
+{
+    return detail::uintptr_t(ptr) % bytes == 0;
+}
+
+} // end namespace util
+} // end namespace detail
+} // end namespace thrust
+
diff --git a/compat/thrust/detail/util/blocking.h b/compat/thrust/detail/util/blocking.h
new file mode 100644
index 0000000000..3bb78a637b
--- /dev/null
+++ b/compat/thrust/detail/util/blocking.h
@@ -0,0 +1,58 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+#pragma once
+
+//functions to support blocking
+
+namespace thrust
+{
+
+namespace detail
+{
+
+namespace util
+{
+
+// x/y rounding towards +infinity for integers, used to determine # of blocks/warps etc.
+template<typename L, typename R>
+  inline __host__ __device__ L divide_ri(const L x, const R y)
+{
+    return (x + (y - 1)) / y;
+}
+
+// x/y rounding towards zero for integers, used to determine # of blocks/warps etc.
+template<typename L, typename R>
+  inline __host__ __device__ L divide_rz(const L x, const R y)
+{
+    return x / y;
+}
+
+// round x towards infinity to the next multiple of y
+template<typename L, typename R>
+  inline __host__ __device__ L round_i(const L x, const R y){ return y * divide_ri(x, y); }
+
+// round x towards zero to the next multiple of y
+template<typename L, typename R>
+  inline __host__ __device__ L round_z(const L x, const R y){ return y * divide_rz(x, y); }
+
+} // end namespace util
+
+} // end namespace detail
+
+} // end namespace thrust
+
diff --git a/compat/thrust/detail/vector_base.h b/compat/thrust/detail/vector_base.h
new file mode 100644
index 0000000000..6974eab554
--- /dev/null
+++ b/compat/thrust/detail/vector_base.h
@@ -0,0 +1,534 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file vector_base.h
+ *  \brief Defines the interface to a base class for
+ *         host_vector & device_vector.
+ */
+
+#pragma once
+
+#include <thrust/iterator/detail/normal_iterator.h>
+#include <thrust/iterator/reverse_iterator.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/detail/type_traits.h>
+#include <thrust/detail/contiguous_storage.h>
+#include <vector>
+
+namespace thrust
+{
+
+namespace detail
+{
+
+template<typename T, typename Alloc>
+  class vector_base
+{
+  private:
+    typedef thrust::detail::contiguous_storage<T,Alloc> storage_type;
+
+  public:
+    // typedefs
+    typedef typename storage_type::value_type      value_type;
+    typedef typename storage_type::pointer         pointer;
+    typedef typename storage_type::const_pointer   const_pointer;
+    typedef typename storage_type::reference       reference;
+    typedef typename storage_type::const_reference const_reference;
+    typedef typename storage_type::size_type       size_type;
+    typedef typename storage_type::difference_type difference_type;
+    typedef typename storage_type::allocator_type  allocator_type;
+
+    typedef typename storage_type::iterator        iterator;
+    typedef typename storage_type::const_iterator  const_iterator;
+
+    typedef thrust::reverse_iterator<iterator>       reverse_iterator;
+    typedef thrust::reverse_iterator<const_iterator> const_reverse_iterator;
+
+    /*! This constructor creates an empty vector_base.
+     */
+    vector_base(void);
+
+    /*! This constructor creates a vector_base with default-constructed
+     *  elements.
+     *  \param n The number of elements to create.
+     */
+    explicit vector_base(size_type n);
+
+    /*! This constructor creates a vector_base with copies
+     *  of an exemplar element.
+     *  \param n The number of elements to initially create.
+     *  \param value An element to copy.
+     */
+    explicit vector_base(size_type n, const value_type &value);
+
+    /*! Copy constructor copies from an exemplar vector_base.
+     *  \param v The vector_base to copy.
+     */
+    vector_base(const vector_base &v);
+
+    /*! assign operator makes a copy of an exemplar vector_base.
+     *  \param v The vector_base to copy.
+     */
+    vector_base &operator=(const vector_base &v);
+
+    /*! Copy constructor copies from an exemplar vector_base with different
+     *  type.
+     *  \param v The vector_base to copy.
+     */
+    template<typename OtherT, typename OtherAlloc>
+    vector_base(const vector_base<OtherT, OtherAlloc> &v);
+
+    /*! assign operator makes a copy of an exemplar vector_base with different
+     *  type.
+     *  \param v The vector_base to copy.
+     */
+    template<typename OtherT, typename OtherAlloc>
+    vector_base &operator=(const vector_base<OtherT,OtherAlloc> &v);
+
+    /*! Copy constructor copies from an exemplar std::vector.
+     *  \param v The std::vector to copy.
+     *  XXX TODO: Make this method redundant with a properly templatized constructor.
+     *            We would like to copy from a vector whose element type is anything
+     *            assignable to value_type.
+     */
+    template<typename OtherT, typename OtherAlloc>
+    vector_base(const std::vector<OtherT, OtherAlloc> &v);
+
+    /*! assign operator makes a copy of an exemplar std::vector.
+     *  \param v The vector to copy.
+     *  XXX TODO: Templatize this assign on the type of the vector to copy from.
+     *            We would like to copy from a vector whose element type is anything
+     *            assignable to value_type.
+     */
+    template<typename OtherT, typename OtherAlloc>
+    vector_base &operator=(const std::vector<OtherT,OtherAlloc> &v);
+
+    /*! This constructor builds a vector_base from a range.
+     *  \param first The beginning of the range.
+     *  \param last The end of the range.
+     */
+    template<typename InputIterator>
+    vector_base(InputIterator first, InputIterator last);
+
+    /*! The destructor erases the elements.
+     */
+    ~vector_base(void);
+
+    /*! \brief Resizes this vector_base to the specified number of elements.
+     *  \param new_size Number of elements this vector_base should contain.
+     *  \throw std::length_error If n exceeds max_size9).
+     *
+     *  This method will resize this vector_base to the specified number of
+     *  elements. If the number is smaller than this vector_base's current
+     *  size this vector_base is truncated, otherwise this vector_base is
+     *  extended and new elements are default constructed.
+     */
+    void resize(size_type new_size);
+
+    /*! \brief Resizes this vector_base to the specified number of elements.
+     *  \param new_size Number of elements this vector_base should contain.
+     *  \param x Data with which new elements should be populated.
+     *  \throw std::length_error If n exceeds max_size().
+     *
+     *  This method will resize this vector_base to the specified number of
+     *  elements.  If the number is smaller than this vector_base's current
+     *  size this vector_base is truncated, otherwise this vector_base is
+     *  extended and new elements are populated with given data.
+     */
+    void resize(size_type new_size, const value_type &x);
+
+    /*! Returns the number of elements in this vector_base.
+     */
+    size_type size(void) const;
+
+    /*! Returns the size() of the largest possible vector_base.
+     *  \return The largest possible return value of size().
+     */
+    size_type max_size(void) const;
+
+    /*! \brief If n is less than or equal to capacity(), this call has no effect.
+     *         Otherwise, this method is a request for allocation of additional memory. If
+     *         the request is successful, then capacity() is greater than or equal to
+     *         n; otherwise, capacity() is unchanged. In either case, size() is unchanged.
+     *  \throw std::length_error If n exceeds max_size().
+     */
+    void reserve(size_type n);
+
+    /*! Returns the number of elements which have been reserved in this
+     *  vector_base.
+     */
+    size_type capacity(void) const;
+
+    /*! This method shrinks the capacity of this vector_base to exactly
+     *  fit its elements.
+     */
+    void shrink_to_fit(void);
+
+    /*! \brief Subscript access to the data contained in this vector_dev.
+     *  \param n The index of the element for which data should be accessed.
+     *  \return Read/write reference to data.
+     *
+     *  This operator allows for easy, array-style, data access.
+     *  Note that data access with this operator is unchecked and
+     *  out_of_range lookups are not defined.
+     */
+    reference operator[](size_type n);
+
+    /*! \brief Subscript read access to the data contained in this vector_dev.
+     *  \param n The index of the element for which data should be accessed.
+     *  \return Read reference to data.
+     *
+     *  This operator allows for easy, array-style, data access.
+     *  Note that data access with this operator is unchecked and
+     *  out_of_range lookups are not defined.
+     */
+    const_reference operator[](size_type n) const;
+
+    /*! This method returns an iterator pointing to the beginning of
+     *  this vector_base.
+     *  \return mStart
+     */
+    iterator begin(void);
+
+    /*! This method returns a const_iterator pointing to the beginning
+     *  of this vector_base.
+     *  \return mStart
+     */
+    const_iterator begin(void) const;
+
+    /*! This method returns a const_iterator pointing to the beginning
+     *  of this vector_base.
+     *  \return mStart
+     */
+    const_iterator cbegin(void) const;
+
+    /*! This method returns a reverse_iterator pointing to the beginning of
+     *  this vector_base's reversed sequence.
+     *  \return A reverse_iterator pointing to the beginning of this
+     *          vector_base's reversed sequence.
+     */
+    reverse_iterator rbegin(void);
+
+    /*! This method returns a const_reverse_iterator pointing to the beginning of
+     *  this vector_base's reversed sequence.
+     *  \return A const_reverse_iterator pointing to the beginning of this
+     *          vector_base's reversed sequence.
+     */
+    const_reverse_iterator rbegin(void) const;
+
+    /*! This method returns a const_reverse_iterator pointing to the beginning of
+     *  this vector_base's reversed sequence.
+     *  \return A const_reverse_iterator pointing to the beginning of this
+     *          vector_base's reversed sequence.
+     */
+    const_reverse_iterator crbegin(void) const;
+
+    /*! This method returns an iterator pointing to one element past the
+     *  last of this vector_base.
+     *  \return begin() + size().
+     */
+    iterator end(void);
+
+    /*! This method returns a const_iterator pointing to one element past the
+     *  last of this vector_base.
+     *  \return begin() + size().
+     */
+    const_iterator end(void) const;
+
+    /*! This method returns a const_iterator pointing to one element past the
+     *  last of this vector_base.
+     *  \return begin() + size().
+     */
+    const_iterator cend(void) const;
+
+    /*! This method returns a reverse_iterator pointing to one element past the
+     *  last of this vector_base's reversed sequence.
+     *  \return rbegin() + size().
+     */
+    reverse_iterator rend(void);
+
+    /*! This method returns a const_reverse_iterator pointing to one element past the
+     *  last of this vector_base's reversed sequence.
+     *  \return rbegin() + size().
+     */
+    const_reverse_iterator rend(void) const;
+
+    /*! This method returns a const_reverse_iterator pointing to one element past the
+     *  last of this vector_base's reversed sequence.
+     *  \return rbegin() + size().
+     */
+    const_reverse_iterator crend(void) const;
+
+    /*! This method returns a const_reference referring to the first element of this
+     *  vector_base.
+     *  \return The first element of this vector_base.
+     */
+    const_reference front(void) const;
+
+    /*! This method returns a reference pointing to the first element of this
+     *  vector_base.
+     *  \return The first element of this vector_base.
+     */
+    reference front(void);
+
+    /*! This method returns a const reference pointing to the last element of
+     *  this vector_base.
+     *  \return The last element of this vector_base.
+     */
+    const_reference back(void) const;
+
+    /*! This method returns a reference referring to the last element of
+     *  this vector_dev.
+     *  \return The last element of this vector_base.
+     */
+    reference back(void);
+
+    /*! This method returns a pointer to this vector_base's first element.
+     *  \return A pointer to the first element of this vector_base.
+     */
+    pointer data(void);
+
+    /*! This method returns a const_pointer to this vector_base's first element.
+     *  \return a const_pointer to the first element of this vector_base.
+     */
+    const_pointer data(void) const;
+
+    /*! This method resizes this vector_base to 0.
+     */
+    void clear(void);
+
+    /*! This method returns true iff size() == 0.
+     *  \return true if size() == 0; false, otherwise.
+     */
+    bool empty(void) const;
+
+    /*! This method appends the given element to the end of this vector_base.
+     *  \param x The element to append.
+     */
+    void push_back(const value_type &x);
+
+    /*! This method erases the last element of this vector_base, invalidating
+     *  all iterators and references to it.
+     */
+    void pop_back(void);
+
+    /*! This method swaps the contents of this vector_base with another vector_base.
+     *  \param v The vector_base with which to swap.
+     */
+    void swap(vector_base &v);
+
+    /*! This method removes the element at position pos.
+     *  \param pos The position of the element of interest.
+     *  \return An iterator pointing to the new location of the element that followed the element
+     *          at position pos.
+     */
+    iterator erase(iterator pos);
+
+    /*! This method removes the range of elements [first,last) from this vector_base.
+     *  \param first The beginning of the range of elements to remove.
+     *  \param last The end of the range of elements to remove.
+     *  \return An iterator pointing to the new location of the element that followed the last
+     *          element in the sequence [first,last).
+     */
+    iterator erase(iterator first, iterator last);
+
+    /*! This method inserts a single copy of a given exemplar value at the
+     *  specified position in this vector_base.
+     *  \param position The insertion position.
+     *  \param x The exemplar element to copy & insert.
+     *  \return An iterator pointing to the newly inserted element.
+     */
+    iterator insert(iterator position, const T &x); 
+
+    /*! This method inserts a copy of an exemplar value to a range at the
+     *  specified position in this vector_base.
+     *  \param position The insertion position
+     *  \param n The number of insertions to perform.
+     *  \param x The value to replicate and insert.
+     */
+    void insert(iterator position, size_type n, const T &x);
+
+    /*! This method inserts a copy of an input range at the specified position
+     *  in this vector_base.
+     *  \param position The insertion position.
+     *  \param first The beginning of the range to copy.
+     *  \param last  The end of the range to copy.
+     *
+     *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html>Input Iterator</a>,
+     *                        and \p InputIterator's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/Assignable.html">Assignable</a>.
+     */
+    template<typename InputIterator>
+    void insert(iterator position, InputIterator first, InputIterator last);
+
+    /*! This version of \p assign replicates a given exemplar
+     *  \p n times into this vector_base.
+     *  \param n The number of times to copy \p x.
+     *  \param x The exemplar element to replicate.
+     */
+    void assign(size_type n, const T &x);
+
+    /*! This version of \p assign makes this vector_base a copy of a given input range.
+     *  \param first The beginning of the range to copy.
+     *  \param last  The end of the range to copy.
+     *
+     *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator">Input Iterator</a>.
+     */
+    template<typename InputIterator>
+    void assign(InputIterator first, InputIterator last);
+
+    /*! This method returns a copy of this vector's allocator.
+     *  \return A copy of the alloctor used by this vector.
+     */
+    allocator_type get_allocator(void) const;
+
+  protected:
+    // Our storage
+    storage_type m_storage;
+
+    // The size of this vector_base, in number of elements.
+    size_type m_size;
+
+  private:
+    // these methods resolve the ambiguity of the constructor template of form (Iterator, Iterator)
+    template<typename IteratorOrIntegralType>
+      void init_dispatch(IteratorOrIntegralType begin, IteratorOrIntegralType end, false_type); 
+
+    template<typename IteratorOrIntegralType>
+      void init_dispatch(IteratorOrIntegralType n, IteratorOrIntegralType value, true_type); 
+
+    template<typename InputIterator>
+      void range_init(InputIterator first, InputIterator last);
+
+    template<typename InputIterator>
+      void range_init(InputIterator first, InputIterator last, thrust::incrementable_traversal_tag);
+
+    template<typename ForwardIterator>
+      void range_init(ForwardIterator first, ForwardIterator last, thrust::random_access_traversal_tag);
+
+    void default_init(size_type n);
+
+    void fill_init(size_type n, const T &x);
+
+    // these methods resolve the ambiguity of the insert() template of form (iterator, InputIterator, InputIterator)
+    template<typename InputIteratorOrIntegralType>
+      void insert_dispatch(iterator position, InputIteratorOrIntegralType first, InputIteratorOrIntegralType last, false_type);
+
+    // these methods resolve the ambiguity of the insert() template of form (iterator, InputIterator, InputIterator)
+    template<typename InputIteratorOrIntegralType>
+      void insert_dispatch(iterator position, InputIteratorOrIntegralType n, InputIteratorOrIntegralType x, true_type);
+
+    // this method appends n default-constructed elements at the end
+    void append(size_type n);
+
+    // this method performs insertion from a fill value
+    void fill_insert(iterator position, size_type n, const T &x);
+
+    // this method performs insertion from a range
+    template<typename InputIterator>
+      void copy_insert(iterator position, InputIterator first, InputIterator last);
+
+    // these methods resolve the ambiguity of the assign() template of form (InputIterator, InputIterator)
+    template<typename InputIterator>
+      void assign_dispatch(InputIterator first, InputIterator last, false_type);
+
+    // these methods resolve the ambiguity of the assign() template of form (InputIterator, InputIterator)
+    template<typename Integral>
+      void assign_dispatch(Integral n, Integral x, true_type);
+
+    // this method performs assignment from a range
+    template<typename InputIterator>
+      void range_assign(InputIterator first, InputIterator last);
+
+    // this method performs assignment from a range of RandomAccessIterators
+    template<typename RandomAccessIterator>
+      void range_assign(RandomAccessIterator first, RandomAccessIterator last, thrust::random_access_traversal_tag);
+
+    // this method performs assignment from a range of InputIterators
+    template<typename InputIterator>
+      void range_assign(InputIterator first, InputIterator last, thrust::incrementable_traversal_tag);
+
+    // this method performs assignment from a fill value
+    void fill_assign(size_type n, const T &x);
+
+    // this method allocates new storage and construct copies the given range
+    template<typename ForwardIterator>
+    void allocate_and_copy(size_type requested_size,
+                           ForwardIterator first, ForwardIterator last,
+                           storage_type &new_storage);
+}; // end vector_base
+
+} // end detail
+
+/*! This function assigns the contents of vector a to vector b and the
+ *  contents of vector b to vector a.
+ *
+ *  \param a The first vector of interest. After completion, the contents
+ *           of b will be returned here.
+ *  \param b The second vector of interest. After completion, the contents
+ *           of a will be returned here.
+ */
+template<typename T, typename Alloc>
+  void swap(detail::vector_base<T,Alloc> &a,
+            detail::vector_base<T,Alloc> &b);
+
+
+/*! This operator allows comparison between two vectors.
+ *  \param lhs The first \p vector to compare.
+ *  \param rhs The second \p vector to compare.
+ *  \return \c true if and only if each corresponding element in either
+ *          \p vector equals the other; \c false, otherwise.
+ */
+template<typename T1, typename Alloc1,
+         typename T2, typename Alloc2>
+bool operator==(const detail::vector_base<T1,Alloc1>& lhs,
+                const detail::vector_base<T2,Alloc2>& rhs);
+    
+template<typename T1, typename Alloc1,
+         typename T2, typename Alloc2>
+bool operator==(const detail::vector_base<T1,Alloc1>& lhs,
+                const std::vector<T2,Alloc2>&         rhs);
+
+template<typename T1, typename Alloc1,
+         typename T2, typename Alloc2>
+bool operator==(const std::vector<T1,Alloc1>&         lhs,
+                const detail::vector_base<T2,Alloc2>& rhs);
+
+/*! This operator allows comparison between two vectors.
+ *  \param lhs The first \p vector to compare.
+ *  \param rhs The second \p vector to compare.
+ *  \return \c false if and only if each corresponding element in either
+ *          \p vector equals the other; \c true, otherwise.
+ */
+template<typename T1, typename Alloc1,
+         typename T2, typename Alloc2>
+bool operator!=(const detail::vector_base<T1,Alloc1>& lhs,
+                const detail::vector_base<T2,Alloc2>& rhs);
+    
+template<typename T1, typename Alloc1,
+         typename T2, typename Alloc2>
+bool operator!=(const detail::vector_base<T1,Alloc1>& lhs,
+                const std::vector<T2,Alloc2>&         rhs);
+
+template<typename T1, typename Alloc1,
+         typename T2, typename Alloc2>
+bool operator!=(const std::vector<T1,Alloc1>&         lhs,
+                const detail::vector_base<T2,Alloc2>& rhs);
+
+} // end thrust
+
+#include <thrust/detail/vector_base.inl>
+
diff --git a/compat/thrust/detail/vector_base.inl b/compat/thrust/detail/vector_base.inl
new file mode 100644
index 0000000000..24e6466c18
--- /dev/null
+++ b/compat/thrust/detail/vector_base.inl
@@ -0,0 +1,1203 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file vector_base.inl
+ *  \brief Inline file for vector_base.h.
+ */
+
+#include <thrust/detail/vector_base.h>
+#include <thrust/detail/copy.h>
+#include <thrust/detail/overlapped_copy.h>
+#include <thrust/equal.h>
+#include <thrust/distance.h>
+#include <thrust/advance.h>
+#include <thrust/detail/type_traits.h>
+#include <thrust/detail/minmax.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/detail/temporary_array.h>
+
+#include <stdexcept>
+
+namespace thrust
+{
+
+namespace detail
+{
+
+template<typename T, typename Alloc>
+  vector_base<T,Alloc>
+    ::vector_base(void)
+      :m_storage(),
+       m_size(0)
+{
+  ;
+} // end vector_base::vector_base()
+
+template<typename T, typename Alloc>
+  vector_base<T,Alloc>
+    ::vector_base(size_type n)
+      :m_storage(),
+       m_size(0)
+{
+  default_init(n);
+} // end vector_base::vector_base()
+
+template<typename T, typename Alloc>
+  vector_base<T,Alloc>
+    ::vector_base(size_type n, const value_type &value)
+      :m_storage(),
+       m_size(0)
+{
+  fill_init(n,value);
+} // end vector_base::vector_base()
+
+template<typename T, typename Alloc>
+  vector_base<T,Alloc>
+    ::vector_base(const vector_base &v)
+      :m_storage(),
+       m_size(0)
+{
+  range_init(v.begin(), v.end());
+} // end vector_base::vector_base()
+
+template<typename T, typename Alloc>
+  vector_base<T,Alloc> &
+    vector_base<T,Alloc>
+      ::operator=(const vector_base &v)
+{
+  if(this != &v)
+  {
+    assign(v.begin(), v.end());
+  } // end if
+
+  return *this;
+} // end vector_base::operator=()
+
+template<typename T, typename Alloc>
+  template<typename OtherT, typename OtherAlloc>
+    vector_base<T,Alloc>
+      ::vector_base(const vector_base<OtherT,OtherAlloc> &v)
+        :m_storage(),
+         m_size(0)
+{
+  range_init(v.begin(), v.end());
+} // end vector_base::vector_base()
+
+template<typename T, typename Alloc>
+  template<typename OtherT, typename OtherAlloc>
+    vector_base<T,Alloc> &
+      vector_base<T,Alloc>
+        ::operator=(const vector_base<OtherT,OtherAlloc> &v)
+{
+  assign(v.begin(), v.end());
+
+  return *this;
+} // end vector_base::operator=()
+
+template<typename T, typename Alloc>
+  template<typename OtherT, typename OtherAlloc>
+    vector_base<T,Alloc>
+      ::vector_base(const std::vector<OtherT,OtherAlloc> &v)
+        :m_storage(),
+         m_size(0)
+{
+  range_init(v.begin(), v.end());
+} // end vector_base::vector_base()
+
+template<typename T, typename Alloc>
+  template<typename OtherT, typename OtherAlloc>
+    vector_base<T,Alloc> &
+      vector_base<T,Alloc>
+        ::operator=(const std::vector<OtherT,OtherAlloc> &v)
+{
+  assign(v.begin(), v.end());
+
+  return *this;
+} // end vector_base::operator=()
+
+template<typename T, typename Alloc>
+  template<typename IteratorOrIntegralType>
+    void vector_base<T,Alloc>
+      ::init_dispatch(IteratorOrIntegralType n,
+                      IteratorOrIntegralType value,
+                      true_type)
+{
+  fill_init(n,value);
+} // end vector_base::init_dispatch()
+
+template<typename T, typename Alloc>
+  void vector_base<T,Alloc>
+    ::default_init(size_type n)
+{
+  if(n > 0)
+  {
+    m_storage.allocate(n);
+    m_size = n;
+
+    m_storage.default_construct_n(begin(), size());
+  } // end if
+} // end vector_base::default_init()
+
+template<typename T, typename Alloc>
+  void vector_base<T,Alloc>
+    ::fill_init(size_type n, const T &x)
+{
+  if(n > 0)
+  {
+    m_storage.allocate(n);
+    m_size = n;
+
+    m_storage.uninitialized_fill_n(begin(), size(), x);
+  } // end if
+} // end vector_base::fill_init()
+
+template<typename T, typename Alloc>
+  template<typename InputIterator>
+    void vector_base<T,Alloc>
+      ::init_dispatch(InputIterator first,
+                      InputIterator last,
+                      false_type)
+{
+  range_init(first, last);
+} // end vector_base::init_dispatch()
+
+template<typename T, typename Alloc>
+  template<typename InputIterator>
+    void vector_base<T,Alloc>
+      ::range_init(InputIterator first,
+                   InputIterator last)
+{
+  range_init(first, last,
+    typename thrust::iterator_traversal<InputIterator>::type());
+} // end vector_base::range_init()
+
+template<typename T, typename Alloc>
+  template<typename InputIterator>
+    void vector_base<T,Alloc>
+      ::range_init(InputIterator first,
+                   InputIterator last,
+                   thrust::incrementable_traversal_tag)
+{
+  for(; first != last; ++first)
+    push_back(*first);
+} // end vector_base::range_init()
+
+template<typename T, typename Alloc>
+  template<typename ForwardIterator>
+    void vector_base<T,Alloc>
+      ::range_init(ForwardIterator first,
+                   ForwardIterator last,
+                   thrust::random_access_traversal_tag)
+{
+  size_type new_size = thrust::distance(first, last);
+
+  allocate_and_copy(new_size, first, last, m_storage);
+  m_size    = new_size;
+} // end vector_base::range_init()
+
+template<typename T, typename Alloc>
+  template<typename InputIterator>
+    vector_base<T,Alloc>
+      ::vector_base(InputIterator first,
+                    InputIterator last)
+        :m_storage(),
+         m_size(0)
+{
+  // check the type of InputIterator: if it's an integral type,
+  // we need to interpret this call as (size_type, value_type)
+  typedef thrust::detail::is_integral<InputIterator> Integer;
+
+  init_dispatch(first, last, Integer());
+} // end vector_basee::vector_base()
+
+template<typename T, typename Alloc>
+  void vector_base<T,Alloc>
+    ::resize(size_type new_size)
+{
+  if(new_size < size())
+  {
+    iterator new_end = begin();
+    thrust::advance(new_end, new_size);
+    erase(new_end, end());
+  } // end if
+  else
+  {
+    append(new_size - size());
+  } // end else
+} // end vector_base::resize()
+
+template<typename T, typename Alloc>
+  void vector_base<T,Alloc>
+    ::resize(size_type new_size, const value_type &x)
+{
+  if(new_size < size())
+  {
+    iterator new_end = begin();
+    thrust::advance(new_end, new_size);
+    erase(new_end, end());
+  } // end if
+  else
+  {
+    insert(end(), new_size - size(), x);
+  } // end else
+} // end vector_base::resize()
+
+template<typename T, typename Alloc>
+  typename vector_base<T,Alloc>::size_type
+    vector_base<T,Alloc>
+      ::size(void) const
+{
+  return m_size;
+} // end vector_base::size()
+
+template<typename T, typename Alloc>
+  typename vector_base<T,Alloc>::size_type
+    vector_base<T,Alloc>
+      ::max_size(void) const
+{
+  return m_storage.max_size();
+} // end vector_base::max_size()
+
+template<typename T, typename Alloc>
+  void vector_base<T,Alloc>
+    ::reserve(size_type n)
+{
+  if(n > capacity())
+  {
+    allocate_and_copy(n, begin(), end(), m_storage);
+  } // end if
+} // end vector_base::reserve()
+
+template<typename T, typename Alloc>
+  typename vector_base<T,Alloc>::size_type
+    vector_base<T,Alloc>
+      ::capacity(void) const
+{
+  return m_storage.size();
+} // end vector_base::capacity()
+
+template<typename T, typename Alloc>
+  void vector_base<T,Alloc>
+    ::shrink_to_fit(void)
+{
+  // use the swap trick
+  vector_base(*this).swap(*this);
+} // end vector_base::shrink_to_fit()
+
+template<typename T, typename Alloc>
+  typename vector_base<T,Alloc>::reference
+    vector_base<T,Alloc>
+      ::operator[](const size_type n)
+{
+  return m_storage[n];
+} // end vector_base::operator[]
+
+template<typename T, typename Alloc>
+  typename vector_base<T,Alloc>::const_reference 
+    vector_base<T,Alloc>
+      ::operator[](const size_type n) const
+{
+  return m_storage[n];
+} // end vector_base::operator[]
+
+template<typename T, typename Alloc>
+  typename vector_base<T,Alloc>::iterator
+    vector_base<T,Alloc>
+      ::begin(void)
+{
+  return m_storage.begin();
+} // end vector_base::begin()
+
+template<typename T, typename Alloc>
+  typename vector_base<T,Alloc>::const_iterator
+    vector_base<T,Alloc>
+      ::begin(void) const
+{
+  return m_storage.begin();
+} // end vector_base::begin()
+
+template<typename T, typename Alloc>
+  typename vector_base<T,Alloc>::const_iterator
+    vector_base<T,Alloc>
+      ::cbegin(void) const
+{
+  return begin();
+} // end vector_base::cbegin()
+
+template<typename T, typename Alloc>
+  typename vector_base<T,Alloc>::reverse_iterator
+    vector_base<T,Alloc>
+      ::rbegin(void)
+{
+  return reverse_iterator(end());
+} // end vector_base::rbegin()
+
+template<typename T, typename Alloc>
+  typename vector_base<T,Alloc>::const_reverse_iterator
+    vector_base<T,Alloc>
+      ::rbegin(void) const
+{
+  return const_reverse_iterator(end());
+} // end vector_base::rbegin()
+
+template<typename T, typename Alloc>
+  typename vector_base<T,Alloc>::const_reverse_iterator
+    vector_base<T,Alloc>
+      ::crbegin(void) const
+{
+  return rbegin();
+} // end vector_base::crbegin()
+
+template<typename T, typename Alloc>
+  typename vector_base<T,Alloc>::iterator
+    vector_base<T,Alloc>
+      ::end(void)
+{
+  iterator result = begin();
+  thrust::advance(result, size());
+  return result;
+} // end vector_base::end()
+
+template<typename T, typename Alloc>
+  typename vector_base<T,Alloc>::const_iterator
+    vector_base<T,Alloc>
+      ::end(void) const
+{
+  const_iterator result = begin();
+  thrust::advance(result, size());
+  return result;
+} // end vector_base::end()
+
+template<typename T, typename Alloc>
+  typename vector_base<T,Alloc>::const_iterator
+    vector_base<T,Alloc>
+      ::cend(void) const
+{
+  return end();
+} // end vector_base::cend()
+
+template<typename T, typename Alloc>
+  typename vector_base<T,Alloc>::reverse_iterator
+    vector_base<T,Alloc>
+      ::rend(void)
+{
+  return reverse_iterator(begin());
+} // end vector_base::rend()
+
+template<typename T, typename Alloc>
+  typename vector_base<T,Alloc>::const_reverse_iterator
+    vector_base<T,Alloc>
+      ::rend(void) const
+{
+  return const_reverse_iterator(begin());
+} // end vector_base::rend()
+
+template<typename T, typename Alloc>
+  typename vector_base<T,Alloc>::const_reverse_iterator
+    vector_base<T,Alloc>
+      ::crend(void) const
+{
+  return rend();
+} // end vector_base::crend()
+
+template<typename T, typename Alloc>
+  typename vector_base<T,Alloc>::const_reference
+    vector_base<T,Alloc>
+      ::front(void) const
+{
+  return *begin();
+} // end vector_base::front()
+
+template<typename T, typename Alloc>
+  typename vector_base<T,Alloc>::reference
+    vector_base<T,Alloc>
+      ::front(void)
+{
+  return *begin();
+} // end vector_base::front()
+
+template<typename T, typename Alloc>
+  typename vector_base<T,Alloc>::const_reference
+    vector_base<T,Alloc>
+      ::back(void) const
+{
+  const_iterator ptr_to_back = end();
+  --ptr_to_back;
+  return *ptr_to_back;
+} // end vector_base::vector_base
+
+template<typename T, typename Alloc>
+  typename vector_base<T,Alloc>::reference
+    vector_base<T,Alloc>
+      ::back(void)
+{
+  iterator ptr_to_back = end();
+  --ptr_to_back;
+  return *ptr_to_back;
+} // end vector_base::vector_base
+
+template<typename T, typename Alloc>
+  typename vector_base<T,Alloc>::pointer
+    vector_base<T,Alloc>
+      ::data(void)
+{
+  return &front();
+} // end vector_base::data()
+
+template<typename T, typename Alloc>
+  typename vector_base<T,Alloc>::const_pointer
+    vector_base<T,Alloc>
+      ::data(void) const
+{
+  return &front();
+} // end vector_base::data()
+
+template<typename T, typename Alloc>
+  vector_base<T,Alloc>
+    ::~vector_base(void)
+{
+  // destroy every living thing
+  m_storage.destroy(begin(),end());
+} // end vector_base::~vector_base()
+
+template<typename T, typename Alloc>
+  void vector_base<T,Alloc>
+    ::clear(void)
+{
+  resize(0);
+} // end vector_base::~vector_dev()
+
+template<typename T, typename Alloc>
+  bool vector_base<T,Alloc>
+    ::empty(void) const
+{
+  return size() == 0;
+} // end vector_base::empty();
+
+template<typename T, typename Alloc>
+  void vector_base<T,Alloc>
+    ::push_back(const value_type &x)
+{
+  insert(end(), x);
+} // end vector_base::push_back()
+
+template<typename T, typename Alloc>
+  void vector_base<T,Alloc>
+    ::pop_back(void)
+{
+  iterator e = end();
+  iterator ptr_to_back = e;
+  --ptr_to_back;
+  m_storage.destroy(ptr_to_back, e);
+  --m_size;
+} // end vector_base::pop_back()
+
+template<typename T, typename Alloc>
+  typename vector_base<T,Alloc>::iterator vector_base<T,Alloc>
+    ::erase(iterator pos)
+{
+  iterator end = pos;
+  ++end;
+  return erase(pos,end);
+} // end vector_base::erase()
+
+template<typename T, typename Alloc>
+  typename vector_base<T,Alloc>::iterator vector_base<T,Alloc>
+    ::erase(iterator first, iterator last)
+{
+  // overlap copy the range [last,end()) to first
+  // XXX this copy only potentially overlaps
+  iterator i = thrust::detail::overlapped_copy(last, end(), first);
+
+  // destroy everything after i
+  m_storage.destroy(i, end());
+
+  // modify our size
+  m_size -= (last - first);
+
+  // return an iterator pointing to the position of the first element
+  // following the erased range
+  return first;
+} // end vector_base::erase()
+
+template<typename T, typename Alloc>
+  void vector_base<T,Alloc>
+    ::swap(vector_base &v)
+{
+  thrust::swap(m_storage,  v.m_storage);
+  thrust::swap(m_size,     v.m_size);
+} // end vector_base::swap()
+
+template<typename T, typename Alloc>
+  void vector_base<T,Alloc>
+    ::assign(size_type n, const T &x)
+{
+  fill_assign(n, x);
+} // end vector_base::assign()
+
+template<typename T, typename Alloc>
+  template<typename InputIterator>
+    void vector_base<T,Alloc>
+      ::assign(InputIterator first, InputIterator last)
+{
+  // we could have received assign(n, x), so disambiguate on the
+  // type of InputIterator
+  typedef typename thrust::detail::is_integral<InputIterator> integral;
+
+  assign_dispatch(first, last, integral());
+} // end vector_base::assign()
+
+template<typename T, typename Alloc>
+  typename vector_base<T,Alloc>::allocator_type
+    vector_base<T,Alloc>
+      ::get_allocator(void) const
+{
+  return m_storage.get_allocator();
+} // end vector_base::get_allocator()
+
+template<typename T, typename Alloc>
+  typename vector_base<T,Alloc>::iterator
+    vector_base<T,Alloc>
+      ::insert(iterator position, const T &x)
+{
+  // find the index of the insertion
+  size_type index = thrust::distance(begin(), position);
+
+  // make the insertion
+  insert(position, 1, x);
+
+  // return an iterator pointing back to position
+  iterator result = begin();
+  thrust::advance(result, index);
+  return result;
+} // end vector_base::insert()
+
+template<typename T, typename Alloc>
+  void vector_base<T,Alloc>
+    ::insert(iterator position, size_type n, const T &x)
+{
+  fill_insert(position, n, x);
+} // end vector_base::insert()
+
+template<typename T, typename Alloc>
+  template<typename InputIterator>
+    void vector_base<T,Alloc>
+      ::insert(iterator position, InputIterator first, InputIterator last)
+{
+  // we could have received insert(position, n, x), so disambiguate on the
+  // type of InputIterator
+  typedef typename thrust::detail::is_integral<InputIterator> integral;
+
+  insert_dispatch(position, first, last, integral());
+} // end vector_base::insert()
+
+template<typename T, typename Alloc>
+  template<typename InputIterator>
+    void vector_base<T,Alloc>
+      ::assign_dispatch(InputIterator first, InputIterator last, false_type)
+{
+  range_assign(first, last);
+} // end vector_base::assign_dispatch()
+
+template<typename T, typename Alloc>
+  template<typename Integral>
+    void vector_base<T,Alloc>
+      ::assign_dispatch(Integral n, Integral x, true_type)
+{
+  fill_assign(n, x);
+} // end vector_base::assign_dispatch()
+
+template<typename T, typename Alloc>
+  template<typename InputIterator>
+    void vector_base<T,Alloc>
+      ::insert_dispatch(iterator position, InputIterator first, InputIterator last, false_type)
+{
+  copy_insert(position, first, last);
+} // end vector_base::insert_dispatch()
+
+template<typename T, typename Alloc>
+  template<typename Integral>
+    void vector_base<T,Alloc>
+      ::insert_dispatch(iterator position, Integral n, Integral x, true_type)
+{
+  fill_insert(position, n, x);
+} // end vector_base::insert_dispatch()
+
+template<typename T, typename Alloc>
+  template<typename ForwardIterator>
+    void vector_base<T,Alloc>
+      ::copy_insert(iterator position,
+                    ForwardIterator first,
+                    ForwardIterator last)
+{
+  if(first != last)
+  {
+    // how many new elements will we create?
+    const size_type num_new_elements = thrust::distance(first, last);
+    if(capacity() - size() >= num_new_elements)
+    {
+      // we've got room for all of them
+      // how many existing elements will we displace?
+      const size_type num_displaced_elements = end() - position;
+      iterator old_end = end();
+
+      if(num_displaced_elements > num_new_elements)
+      {
+        // construct copy n displaced elements to new elements
+        // following the insertion
+        m_storage.uninitialized_copy(end() - num_new_elements, end(), end());
+
+        // extend the size
+        m_size += num_new_elements;
+
+        // copy num_displaced_elements - num_new_elements elements to existing elements
+        // this copy overlaps
+        const size_type copy_length = (old_end - num_new_elements) - position;
+        thrust::detail::overlapped_copy(position, old_end - num_new_elements, old_end - copy_length);
+
+        // finally, copy the range to the insertion point
+        thrust::copy(first, last, position);
+      } // end if
+      else
+      {
+        ForwardIterator mid = first;
+        thrust::advance(mid, num_displaced_elements);
+
+        // construct copy new elements at the end of the vector
+        m_storage.uninitialized_copy(mid, last, end());
+
+        // extend the size
+        m_size += num_new_elements - num_displaced_elements;
+
+        // construct copy the displaced elements
+        m_storage.uninitialized_copy(position, old_end, end());
+
+        // extend the size
+        m_size += num_displaced_elements;
+
+        // copy to elements which already existed
+        thrust::copy(first, mid, position);
+      } // end else
+    } // end if
+    else
+    {
+      const size_type old_size = size();
+
+      // compute the new capacity after the allocation
+      size_type new_capacity = old_size + thrust::max THRUST_PREVENT_MACRO_SUBSTITUTION (old_size, num_new_elements);
+
+      // allocate exponentially larger new storage
+      new_capacity = thrust::max THRUST_PREVENT_MACRO_SUBSTITUTION <size_type>(new_capacity, 2 * capacity());
+
+      // do not exceed maximum storage
+      new_capacity = thrust::min THRUST_PREVENT_MACRO_SUBSTITUTION <size_type>(new_capacity, max_size());
+
+      if(new_capacity > max_size())
+      {
+        throw std::length_error("insert(): insertion exceeds max_size().");
+      } // end if
+
+      storage_type new_storage(new_capacity);
+
+      // record how many constructors we invoke in the try block below
+      iterator new_end = new_storage.begin();
+
+      try
+      {
+        // construct copy elements before the insertion to the beginning of the newly
+        // allocated storage
+        new_end = m_storage.uninitialized_copy(begin(), position, new_storage.begin());
+
+        // construct copy elements to insert
+        new_end = m_storage.uninitialized_copy(first, last, new_end);
+
+        // construct copy displaced elements from the old storage to the new storage
+        // remember [position, end()) refers to the old storage
+        new_end = m_storage.uninitialized_copy(position, end(), new_end);
+      } // end try
+      catch(...)
+      {
+        // something went wrong, so destroy & deallocate the new storage 
+        m_storage.destroy(new_storage.begin(), new_end);
+        new_storage.deallocate();
+
+        // rethrow
+        throw;
+      } // end catch
+
+      // call destructors on the elements in the old storage
+      m_storage.destroy(begin(), end());
+
+      // record the vector's new state
+      m_storage.swap(new_storage);
+      m_size = old_size + num_new_elements;
+    } // end else
+  } // end if
+} // end vector_base::copy_insert()
+
+template<typename T, typename Alloc>
+  void vector_base<T,Alloc>
+    ::append(size_type n)
+{
+  if(n != 0)
+  {
+    if(capacity() - size() >= n)
+    {
+      // we've got room for all of them
+
+      // default construct new elements at the end of the vector
+      m_storage.default_construct_n(end(), n);
+
+      // extend the size
+      m_size += n;
+    } // end if
+    else
+    {
+      const size_type old_size = size();
+
+      // compute the new capacity after the allocation
+      size_type new_capacity = old_size + thrust::max THRUST_PREVENT_MACRO_SUBSTITUTION (old_size, n);
+
+      // allocate exponentially larger new storage
+      new_capacity = thrust::max THRUST_PREVENT_MACRO_SUBSTITUTION <size_type>(new_capacity, 2 * capacity());
+
+      // do not exceed maximum storage
+      new_capacity = thrust::min THRUST_PREVENT_MACRO_SUBSTITUTION <size_type>(new_capacity, max_size());
+
+      // create new storage
+      storage_type new_storage(new_capacity);
+
+      // record how many constructors we invoke in the try block below
+      iterator new_end = new_storage.begin();
+
+      try
+      {
+        // construct copy all elements into the newly allocated storage
+        new_end = m_storage.uninitialized_copy(begin(), end(), new_storage.begin());
+
+        // construct new elements to insert
+        m_storage.default_construct_n(new_end, n);
+        new_end += n;
+      } // end try
+      catch(...)
+      {
+        // something went wrong, so destroy & deallocate the new storage 
+        m_storage.destroy(new_storage.begin(), new_end);
+        new_storage.deallocate();
+
+        // rethrow
+        throw;
+      } // end catch
+
+      // call destructors on the elements in the old storage
+      m_storage.destroy(begin(), end());
+
+      // record the vector's new state
+      m_storage.swap(new_storage);
+      m_size    = old_size + n;
+    } // end else
+  } // end if
+} // end vector_base::append()
+
+template<typename T, typename Alloc>
+  void vector_base<T,Alloc>
+    ::fill_insert(iterator position, size_type n, const T &x)
+{
+  if(n != 0)
+  {
+    if(capacity() - size() >= n)
+    {
+      // we've got room for all of them
+      // how many existing elements will we displace?
+      const size_type num_displaced_elements = end() - position;
+      iterator old_end = end();
+
+      if(num_displaced_elements > n)
+      {
+        // construct copy n displaced elements to new elements
+        // following the insertion
+        m_storage.uninitialized_copy(end() - n, end(), end());
+
+        // extend the size
+        m_size += n;
+
+        // copy num_displaced_elements - n elements to existing elements
+        // this copy overlaps
+        const size_type copy_length = (old_end - n) - position;
+        thrust::detail::overlapped_copy(position, old_end - n, old_end - copy_length);
+
+        // finally, fill the range to the insertion point
+        thrust::fill_n(position, n, x);
+      } // end if
+      else
+      {
+        // construct new elements at the end of the vector
+        m_storage.uninitialized_fill_n(end(), n - num_displaced_elements, x);
+
+        // extend the size
+        m_size += n - num_displaced_elements;
+
+        // construct copy the displaced elements
+        m_storage.uninitialized_copy(position, old_end, end());
+
+        // extend the size
+        m_size += num_displaced_elements;
+
+        // fill to elements which already existed
+        thrust::fill(position, old_end, x);
+      } // end else
+    } // end if
+    else
+    {
+      const size_type old_size = size();
+
+      // compute the new capacity after the allocation
+      size_type new_capacity = old_size + thrust::max THRUST_PREVENT_MACRO_SUBSTITUTION (old_size, n);
+
+      // allocate exponentially larger new storage
+      new_capacity = thrust::max THRUST_PREVENT_MACRO_SUBSTITUTION <size_type>(new_capacity, 2 * capacity());
+
+      // do not exceed maximum storage
+      new_capacity = thrust::min THRUST_PREVENT_MACRO_SUBSTITUTION <size_type>(new_capacity, max_size());
+
+      if(new_capacity > max_size())
+      {
+        throw std::length_error("insert(): insertion exceeds max_size().");
+      } // end if
+
+      storage_type new_storage(new_capacity);
+
+      // record how many constructors we invoke in the try block below
+      iterator new_end = new_storage.begin();
+
+      try
+      {
+        // construct copy elements before the insertion to the beginning of the newly
+        // allocated storage
+        new_end = m_storage.uninitialized_copy(begin(), position, new_storage.begin());
+
+        // construct new elements to insert
+        m_storage.uninitialized_fill_n(new_end, n, x);
+        new_end += n;
+
+        // construct copy displaced elements from the old storage to the new storage
+        // remember [position, end()) refers to the old storage
+        new_end = m_storage.uninitialized_copy(position, end(), new_end);
+      } // end try
+      catch(...)
+      {
+        // something went wrong, so destroy & deallocate the new storage 
+        m_storage.destroy(new_storage.begin(), new_end);
+        new_storage.deallocate();
+
+        // rethrow
+        throw;
+      } // end catch
+
+      // call destructors on the elements in the old storage
+      m_storage.destroy(begin(), end());
+
+      // record the vector's new state
+      m_storage.swap(new_storage);
+      m_size    = old_size + n;
+    } // end else
+  } // end if
+} // end vector_base::fill_insert()
+
+template<typename T, typename Alloc>
+  template<typename InputIterator>
+    void vector_base<T,Alloc>
+      ::range_assign(InputIterator first,
+                     InputIterator last)
+{
+  // dispatch on traversal
+  range_assign(first, last,
+    typename thrust::iterator_traversal<InputIterator>::type());
+} // end range_assign()
+
+template<typename T, typename Alloc>
+  template<typename InputIterator>
+    void vector_base<T,Alloc>
+      ::range_assign(InputIterator first,
+                     InputIterator last,
+                     thrust::incrementable_traversal_tag)
+{
+  iterator current(begin());
+
+  // assign to elements which already exist
+  for(; first != last && current != end(); ++current, ++first)
+  {
+    *current = *first;
+  } // end for
+  
+  // either just the input was exhausted or both
+  // the input and vector elements were exhausted
+  if(first == last)
+  {
+    // if we exhausted the input, erase leftover elements
+    erase(current, end());
+  } // end if
+  else
+  {
+    // insert the rest of the input at the end of the vector
+    insert(end(), first, last);
+  } // end else
+} // end vector_base::range_assign()
+
+template<typename T, typename Alloc>
+  template<typename RandomAccessIterator>
+    void vector_base<T,Alloc>
+      ::range_assign(RandomAccessIterator first,
+                     RandomAccessIterator last,
+                     thrust::random_access_traversal_tag)
+{
+  const size_type n = thrust::distance(first, last);
+
+  if(n > capacity())
+  {
+    storage_type new_storage;
+    allocate_and_copy(n, first, last, new_storage);
+
+    // call destructors on the elements in the old storage
+    m_storage.destroy(begin(), end());
+
+    // record the vector's new state
+    m_storage.swap(new_storage);
+    m_size = n;
+  } // end if
+  else if(size() >= n)
+  {
+    // we can already accomodate the new range
+    iterator new_end = thrust::copy(first, last, begin());
+
+    // destroy the elements we don't need
+    m_storage.destroy(new_end, end());
+
+    // update size
+    m_size = n;
+  } // end else if
+  else
+  {
+    // range fits inside allocated storage, but some elements
+    // have not been constructed yet
+    
+    // XXX TODO we could possibly implement this with one call
+    // to transform rather than copy + uninitialized_copy
+
+    // copy to elements which already exist
+    RandomAccessIterator mid = first;
+    thrust::advance(mid, size());
+    thrust::copy(first, mid, begin());
+
+    // uninitialize_copy to elements which must be constructed
+    m_storage.uninitialized_copy(mid, last, end());
+
+    // update size
+    m_size = n;
+  } // end else
+} // end vector_base::assign()
+
+template<typename T, typename Alloc>
+  void vector_base<T,Alloc>
+    ::fill_assign(size_type n, const T &x)
+{
+  if(n > capacity())
+  {
+    // XXX we should also include a copy of the allocator:
+    // vector_base<T,Alloc> temp(n, x, get_allocator());
+    vector_base<T,Alloc> temp(n, x);
+    temp.swap(*this);
+  } // end if
+  else if(n > size())
+  {
+    // fill to existing elements
+    thrust::fill(begin(), end(), x);
+
+    // construct uninitialized elements
+    m_storage.uninitialized_fill_n(end(), n - size(), x);
+
+    // adjust size
+    m_size += (n - size());
+  } // end else if
+  else
+  {
+    // fill to existing elements
+    iterator new_end = thrust::fill_n(begin(), n, x);
+
+    // erase the elements after the fill
+    erase(new_end, end());
+  } // end else
+} // end vector_base::fill_assign()
+
+template<typename T, typename Alloc>
+  template<typename ForwardIterator>
+    void vector_base<T,Alloc>
+      ::allocate_and_copy(size_type requested_size,
+                          ForwardIterator first, ForwardIterator last,
+                          storage_type &new_storage)
+{
+  if(requested_size == 0)
+  {
+    new_storage.deallocate();
+    return;
+  } // end if
+
+  // allocate exponentially larger new storage
+  size_type allocated_size = thrust::max<size_type>(requested_size, 2 * capacity());
+
+  // do not exceed maximum storage
+  allocated_size = thrust::min<size_type>(allocated_size, max_size());
+
+  if(requested_size > allocated_size)
+  {
+    throw std::length_error("assignment exceeds max_size().");
+  } // end if
+
+  new_storage.allocate(allocated_size);
+
+  try
+  {
+    // construct the range to the newly allocated storage
+    m_storage.uninitialized_copy(first, last, new_storage.begin());
+  } // end try
+  catch(...)
+  {
+    // something went wrong, so destroy & deallocate the new storage 
+    // XXX seems like this destroys too many elements -- should just be last - first instead of requested_size
+    iterator new_storage_end = new_storage.begin();
+    thrust::advance(new_storage_end, requested_size);
+    m_storage.destroy(new_storage.begin(), new_storage_end);
+    new_storage.deallocate();
+
+    // rethrow
+    throw;
+  } // end catch
+} // end vector_base::allocate_and_copy()
+
+
+} // end detail
+
+template<typename T, typename Alloc>
+  void swap(detail::vector_base<T,Alloc> &a,
+            detail::vector_base<T,Alloc> &b)
+{
+  a.swap(b);
+} // end swap()
+
+
+
+namespace detail
+{
+    
+// iterator tags match
+template <typename InputIterator1, typename InputIterator2>
+bool vector_equal(InputIterator1 first1, InputIterator1 last1,
+                  InputIterator2 first2,
+                  thrust::detail::true_type)
+{
+  return thrust::equal(first1, last1, first2);
+}
+
+// iterator tags differ
+template <typename InputIterator1, typename InputIterator2>
+bool vector_equal(InputIterator1 first1, InputIterator1 last1,
+                  InputIterator2 first2,
+                  thrust::detail::false_type)
+{
+  typename thrust::iterator_difference<InputIterator1>::type n = thrust::distance(first1,last1);
+
+  typedef typename thrust::iterator_system<InputIterator1>::type FromSystem1;
+  typedef typename thrust::iterator_system<InputIterator2>::type FromSystem2;
+
+  // bring both ranges to the host system
+  // note that these copies are no-ops if the range is already convertible to the host system
+  FromSystem1 from_system1;
+  FromSystem2 from_system2;
+  thrust::host_system_tag to_system;
+  thrust::detail::move_to_system<InputIterator1, FromSystem1, thrust::host_system_tag> rng1(from_system1, to_system, first1, last1);
+  thrust::detail::move_to_system<InputIterator2, FromSystem2, thrust::host_system_tag> rng2(from_system2, to_system, first2, first2 + n);
+
+  return thrust::equal(rng1.begin(), rng1.end(), rng2.begin());
+}
+
+template <typename InputIterator1, typename InputIterator2>
+bool vector_equal(InputIterator1 first1, InputIterator1 last1,
+                  InputIterator2 first2)
+{
+  typedef typename thrust::iterator_system<InputIterator1>::type system1;
+  typedef typename thrust::iterator_system<InputIterator2>::type system2;
+
+  // dispatch on the sameness of the two systems
+  return vector_equal(first1, last1, first2,
+    thrust::detail::is_same<system1,system2>());
+}
+
+} // end namespace detail
+
+
+
+
+template<typename T1, typename Alloc1,
+         typename T2, typename Alloc2>
+bool operator==(const detail::vector_base<T1,Alloc1>& lhs,
+                const detail::vector_base<T2,Alloc2>& rhs)
+{
+    return lhs.size() == rhs.size() && detail::vector_equal(lhs.begin(), lhs.end(), rhs.begin());
+}
+    
+template<typename T1, typename Alloc1,
+         typename T2, typename Alloc2>
+bool operator==(const detail::vector_base<T1,Alloc1>& lhs,
+                const std::vector<T2,Alloc2>&         rhs)
+{
+    return lhs.size() == rhs.size() && detail::vector_equal(lhs.begin(), lhs.end(), rhs.begin());
+}
+
+template<typename T1, typename Alloc1,
+         typename T2, typename Alloc2>
+bool operator==(const std::vector<T1,Alloc1>&         lhs,
+                const detail::vector_base<T2,Alloc2>& rhs)
+{
+    return lhs.size() == rhs.size() && detail::vector_equal(lhs.begin(), lhs.end(), rhs.begin());
+}
+
+template<typename T1, typename Alloc1,
+         typename T2, typename Alloc2>
+bool operator!=(const detail::vector_base<T1,Alloc1>& lhs,
+                const detail::vector_base<T2,Alloc2>& rhs)
+{
+    return !(lhs == rhs);
+}
+    
+template<typename T1, typename Alloc1,
+         typename T2, typename Alloc2>
+bool operator!=(const detail::vector_base<T1,Alloc1>& lhs,
+                const std::vector<T2,Alloc2>&         rhs)
+{
+    return !(lhs == rhs);
+}
+
+template<typename T1, typename Alloc1,
+         typename T2, typename Alloc2>
+bool operator!=(const std::vector<T1,Alloc1>&         lhs,
+                const detail::vector_base<T2,Alloc2>& rhs)
+{
+    return !(lhs == rhs);
+}
+
+} // end thrust
+
diff --git a/compat/thrust/device_allocator.h b/compat/thrust/device_allocator.h
new file mode 100644
index 0000000000..a5462d1a28
--- /dev/null
+++ b/compat/thrust/device_allocator.h
@@ -0,0 +1,123 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file device_allocator.h
+ *  \brief An allocator which creates new elements in device memory
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/device_new_allocator.h>
+#include <limits>
+#include <stdexcept>
+
+namespace thrust
+{
+
+/*! \addtogroup memory_management_classes Memory Management Classes
+ *  \{
+ */
+
+template<typename T> class device_allocator;
+
+/*! \p device_allocator<void> is a device memory allocator.
+ *  This class is a specialization for \c void.
+ *
+ *  \see device_ptr
+ *  \see http://www.sgi.com/tech/stl/Allocators.html
+ */
+template<>
+  class device_allocator<void>
+{
+  public:
+    /*! Type of element allocated, \c void. */
+    typedef void                              value_type;
+
+    /*! Pointer to allocation, \c device_ptr<void>. */
+    typedef device_ptr<void>                  pointer;
+
+    /*! \c const pointer to allocation, \c device_ptr<const void>. */
+    typedef device_ptr<const void>            const_pointer;
+
+    /*! Type of allocation size, \c std::size_t. */
+    typedef std::size_t                       size_type;
+
+    /*! Type of allocation difference, \c pointer::difference_type. */
+    typedef pointer::difference_type difference_type;
+
+    /*! The \p rebind metafunction provides the type of a \p device_allocator
+     *  instantiated with another type.
+     *
+     *  \tparam U The other type to use for instantiation.
+     */
+    template<typename U>
+      struct rebind
+    {
+      /*! The typedef \p other gives the type of the rebound \p device_allocator.
+       */
+      typedef device_allocator<U> other;
+    }; // end rebind
+}; // end device_allocator<void>
+
+/*! \p device_allocator is a device memory allocator.
+ *  This implementation inherits from \p device_new_allocator.
+ *
+ *  \see device_ptr
+ *  \see device_new_allocator
+ *  \see http://www.sgi.com/tech/stl/Allocators.html
+ */
+template<typename T>
+  class device_allocator
+    : public device_new_allocator<T>
+{
+  public:
+    /*! The \p rebind metafunction provides the type of a \p device_allocator
+     *  instantiated with another type.
+     *
+     *  \tparam U The other type to use for instantiation.
+     */
+    template<typename U>
+      struct rebind
+    {
+      /*! The typedef \p other gives the type of the rebound \p device_allocator.
+       */
+      typedef device_allocator<U> other;
+    }; // end rebind
+
+    /*! No-argument constructor has no effect.
+     */
+    __host__ __device__
+    inline device_allocator() {}
+
+    /*! Copy constructor has no effect.
+     */
+    __host__ __device__
+    inline device_allocator(device_allocator const&) {}
+
+    /*! Constructor from other \p allocator has no effect.
+     */
+    template<typename U>
+    __host__ __device__
+    inline device_allocator(device_allocator<U> const&) {}
+}; // end device_allocator
+
+/*! \}
+ */
+
+} // end thrust
+
diff --git a/compat/thrust/device_delete.h b/compat/thrust/device_delete.h
new file mode 100644
index 0000000000..1df3bb6f46
--- /dev/null
+++ b/compat/thrust/device_delete.h
@@ -0,0 +1,56 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file device_delete.h
+ *  \brief Deletes variables in device memory
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/device_ptr.h>
+
+namespace thrust
+{
+
+/*! \addtogroup deallocation_functions Deallocation Functions
+ *  \ingroup memory_management_functions
+ *  \{
+ */
+
+/*! \p device_delete deletes a \p device_ptr allocated with
+ *  \p device_new.
+ *
+ *  \param ptr The \p device_ptr to delete, assumed to have
+ *         been allocated with \p device_new.
+ *  \param n The number of objects to destroy at \p ptr. Defaults to \c 1
+ *         similar to \p device_new.
+ *
+ *  \see device_ptr
+ *  \see device_new
+ */
+template<typename T>
+  inline void device_delete(thrust::device_ptr<T> ptr,
+                            const size_t n = 1);
+
+/*! \}
+ */
+
+} // end thrust
+
+#include <thrust/detail/device_delete.inl>
+
diff --git a/compat/thrust/device_free.h b/compat/thrust/device_free.h
new file mode 100644
index 0000000000..a734418e58
--- /dev/null
+++ b/compat/thrust/device_free.h
@@ -0,0 +1,68 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file device_free.h
+ *  \brief Deallocates storage allocated by \p device_malloc
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/device_ptr.h>
+
+namespace thrust
+{
+
+/*! \addtogroup deallocation_functions Deallocation Functions
+ *  \ingroup memory_management_functions
+ *  \{
+ */
+
+/*! \p device_free deallocates memory allocated by the function \p device_malloc.
+ *
+ *  \param ptr A \p device_ptr pointing to memory to be deallocated.
+ *
+ *  The following code snippet demonstrates how to use \p device_free to
+ *  deallocate memory allocated by \p device_malloc.
+ *
+ *  \code
+ *  #include <thrust/device_malloc.h>
+ *  #include <thrust/device_free.h>
+ *  ...
+ *  // allocate some integers with device_malloc
+ *  const int N = 100;
+ *  thrust::device_ptr<int> int_array = thrust::device_malloc<int>(N);
+ *
+ *  // manipulate integers
+ *  ...
+ *
+ *  // deallocate with device_free
+ *  thrust::device_free(int_array);
+ *  \endcode
+ *
+ *  \see device_ptr
+ *  \see device_malloc
+ */
+inline void device_free(thrust::device_ptr<void> ptr);
+
+/*! \}
+ */
+
+} // end thrust
+
+#include <thrust/detail/device_free.inl>
+
diff --git a/compat/thrust/device_malloc.h b/compat/thrust/device_malloc.h
new file mode 100644
index 0000000000..a3b07234f9
--- /dev/null
+++ b/compat/thrust/device_malloc.h
@@ -0,0 +1,103 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file device_malloc.h
+ *  \brief Allocates storage in device memory
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/device_ptr.h>
+#include <cstddef> // for std::size_t
+
+namespace thrust
+{
+
+/*! \addtogroup allocation_functions Allocation Functions
+ *  \ingroup memory_management_functions
+ *  \{
+ */
+
+/*! This version of \p device_malloc allocates sequential device storage
+ *  for bytes.
+ *
+ *  \param n The number of bytes to allocate sequentially
+ *           in device memory.
+ *  \return A \p device_ptr to the newly allocated memory.
+ *
+ *  The following code snippet demonstrates how to use \p device_malloc to
+ *  allocate a range of device memory.
+ *
+ *  \code
+ *  #include <thrust/device_malloc.h>
+ *  #include <thrust/device_free.h>
+ *  ...
+ *  // allocate some memory with device_malloc
+ *  const int N = 100;
+ *  thrust::device_ptr<void> void_ptr = thrust::device_malloc(N);
+ *
+ *  // manipulate memory
+ *  ...
+ *
+ *  // deallocate with device_free
+ *  thrust::device_free(void_ptr);
+ *  \endcode
+ *
+ *  \see device_ptr
+ *  \see device_free
+ */
+inline thrust::device_ptr<void> device_malloc(const std::size_t n);
+
+/*! This version of \p device_malloc allocates sequential device storage for
+ *  new objects of the given type.
+ *
+ *  \param n The number of objects of type T to allocate
+ *           sequentially in device memory.
+ *  \return A \p device_ptr to the newly allocated memory.
+ *
+ *  The following code snippet demonstrates how to use \p device_malloc to
+ *  allocate a range of device memory.
+ *
+ *  \code
+ *  #include <thrust/device_malloc.h>
+ *  #include <thrust/device_free.h>
+ *  ...
+ *  // allocate some integers with device_malloc
+ *  const int N = 100;
+ *  thrust::device_ptr<int> int_array = thrust::device_malloc<int>(N);
+ *
+ *  // manipulate integers
+ *  ...
+ *
+ *  // deallocate with device_free
+ *  thrust::device_free(int_array);
+ *  \endcode
+ *
+ *  \see device_ptr
+ *  \see device_free
+ */
+template<typename T>
+  inline thrust::device_ptr<T> device_malloc(const std::size_t n);
+
+/*! \}
+ */
+
+} // end thrust
+
+#include <thrust/detail/device_malloc.inl>
+
diff --git a/compat/thrust/device_malloc_allocator.h b/compat/thrust/device_malloc_allocator.h
new file mode 100644
index 0000000000..404a6d297a
--- /dev/null
+++ b/compat/thrust/device_malloc_allocator.h
@@ -0,0 +1,174 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file device_malloc_allocator.h
+ *  \brief An allocator which allocates storage with \p device_malloc
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/device_ptr.h>
+#include <thrust/device_reference.h>
+#include <thrust/device_malloc.h>
+#include <thrust/device_free.h>
+#include <limits>
+#include <stdexcept>
+
+namespace thrust
+{
+
+// forward declarations to WAR circular #includes
+template<typename> class device_ptr;
+template<typename T> device_ptr<T> device_malloc(const std::size_t n);
+
+/*! \addtogroup memory_management Memory Management
+ *  \addtogroup memory_management_classes Memory Management Classes
+ *  \ingroup memory_management
+ *  \{
+ */
+
+/*! \p device_malloc_allocator is a device memory allocator that employs the
+ *  \p device_malloc function for allocation.
+ *
+ *  \see device_malloc
+ *  \see device_ptr
+ *  \see http://www.sgi.com/tech/stl/Allocators.html
+ */
+template<typename T>
+  class device_malloc_allocator
+{
+  public:
+    /*! Type of element allocated, \c T. */
+    typedef T                                 value_type;
+
+    /*! Pointer to allocation, \c device_ptr<T>. */
+    typedef device_ptr<T>                     pointer;
+
+    /*! \c const pointer to allocation, \c device_ptr<const T>. */
+    typedef device_ptr<const T>               const_pointer;
+
+    /*! Reference to allocated element, \c device_reference<T>. */
+    typedef device_reference<T>               reference;
+
+    /*! \c const reference to allocated element, \c device_reference<const T>. */
+    typedef device_reference<const T>         const_reference;
+
+    /*! Type of allocation size, \c std::size_t. */
+    typedef std::size_t                       size_type;
+
+    /*! Type of allocation difference, \c pointer::difference_type. */
+    typedef typename pointer::difference_type difference_type;
+
+    /*! The \p rebind metafunction provides the type of a \p device_malloc_allocator
+     *  instantiated with another type.
+     *
+     *  \tparam U The other type to use for instantiation.
+     */
+    template<typename U>
+      struct rebind
+    {
+      /*! The typedef \p other gives the type of the rebound \p device_malloc_allocator.
+       */
+      typedef device_malloc_allocator<U> other;
+    }; // end rebind
+
+    /*! No-argument constructor has no effect. */
+    __host__ __device__
+    inline device_malloc_allocator() {}
+
+    /*! No-argument destructor has no effect. */
+    __host__ __device__
+    inline ~device_malloc_allocator() {}
+
+    /*! Copy constructor has no effect. */
+    __host__ __device__
+    inline device_malloc_allocator(device_malloc_allocator const&) {}
+
+    /*! Constructor from other \p device_malloc_allocator has no effect. */
+    template<typename U>
+    __host__ __device__
+    inline device_malloc_allocator(device_malloc_allocator<U> const&) {}
+
+    /*! Returns the address of an allocated object.
+     *  \return <tt>&r</tt>.
+     */
+    __host__ __device__
+    inline pointer address(reference r) { return &r; }
+    
+    /*! Returns the address an allocated object.
+     *  \return <tt>&r</tt>.
+     */
+    __host__ __device__
+    inline const_pointer address(const_reference r) { return &r; }
+
+    /*! Allocates storage for \p cnt objects.
+     *  \param cnt The number of objects to allocate.
+     *  \return A \p pointer to uninitialized storage for \p cnt objects.
+     *  \note Memory allocated by this function must be deallocated with \p deallocate.
+     */
+    __host__
+    inline pointer allocate(size_type cnt,
+                            const_pointer = const_pointer(static_cast<T*>(0)))
+    {
+      if(cnt > this->max_size())
+      {
+        throw std::bad_alloc();
+      } // end if
+
+      return pointer(device_malloc<T>(cnt));
+    } // end allocate()
+
+    /*! Deallocates storage for objects allocated with \p allocate.
+     *  \param p A \p pointer to the storage to deallocate.
+     *  \param cnt The size of the previous allocation.
+     *  \note Memory deallocated by this function must previously have been
+     *        allocated with \p allocate.
+     */
+    __host__
+    inline void deallocate(pointer p, size_type cnt)
+    {
+      device_free(p);
+    } // end deallocate()
+
+    /*! Returns the largest value \c n for which <tt>allocate(n)</tt> might succeed.
+     *  \return The largest value \c n for which <tt>allocate(n)</tt> might succeed.
+     */
+    inline size_type max_size() const
+    {
+      return (std::numeric_limits<size_type>::max)() / sizeof(T);
+    } // end max_size()
+
+    /*! Compares against another \p device_malloc_allocator for equality.
+     *  \return \c true
+     */
+    __host__ __device__
+    inline bool operator==(device_malloc_allocator const&) { return true; }
+
+    /*! Compares against another \p device_malloc_allocator for inequality.
+     *  \return \c false
+     */
+    __host__ __device__
+    inline bool operator!=(device_malloc_allocator const &a) {return !operator==(a); }
+}; // end device_malloc_allocator
+
+/*! \}
+ */
+
+} // end thrust
+
+
diff --git a/compat/thrust/device_new.h b/compat/thrust/device_new.h
new file mode 100644
index 0000000000..001d476896
--- /dev/null
+++ b/compat/thrust/device_new.h
@@ -0,0 +1,88 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file device_new.h
+ *  \brief Constructs new elements in device memory
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// #include this for size_t
+#include <cstddef>
+#include <thrust/device_ptr.h>
+
+namespace thrust
+{
+
+/*!
+ *  \addtogroup allocation_functions Allocation Functions
+ *  \{
+ */
+
+/*! \p device_new implements the placement \c new operator for types
+ *  resident in device memory. \p device_new calls <tt>T</tt>'s null
+ *  constructor on a array of objects in device memory.
+ *  No memory is allocated by this function.
+ *
+ *  \param  p A \p device_ptr to a region of device memory into which
+ *          to construct one or many <tt>T</tt>s.
+ *  \param  n The number of objects to construct at \p p.
+ *  \return p, casted to <tt>T</tt>'s type.
+ *
+ *  \see device_ptr
+ */
+template <typename T>
+  device_ptr<T> device_new(device_ptr<void> p,
+                           const size_t n = 1);
+
+/*! \p device_new implements the placement new operator for types
+ *  resident in device memory. \p device_new calls <tt>T</tt>'s copy
+ *  constructor on a array of objects in device memory. No memory is
+ *  allocated by this function.
+ *
+ *  \param  p A \p device_ptr to a region of device memory into which to
+ *          construct one or many <tt>T</tt>s.
+ *  \param exemplar The value from which to copy.
+ *  \param  n The number of objects to construct at \p p.
+ *  \return p, casted to <tt>T</tt>'s type.
+ *
+ *  \see device_ptr
+ *  \see fill
+ */
+template <typename T>
+  device_ptr<T> device_new(device_ptr<void> p,
+                           const T &exemplar,
+                           const size_t n = 1);
+
+/*! \p device_new implements the new operator for types resident in device memory.
+ *  It allocates device memory large enough to hold \p n new objects of type \c T.
+ *
+ *  \param n The number of objects to allocate. Defaults to \c 1.
+ *  \return A \p device_ptr to the newly allocated region of device memory.
+ */
+template <typename T>
+  device_ptr<T> device_new(const size_t n = 1);
+
+/*! \}
+ */
+
+} // end thrust
+
+#include <thrust/detail/device_new.inl>
+
diff --git a/compat/thrust/device_new_allocator.h b/compat/thrust/device_new_allocator.h
new file mode 100644
index 0000000000..527d1fd7b5
--- /dev/null
+++ b/compat/thrust/device_new_allocator.h
@@ -0,0 +1,172 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file device_new_allocator.h
+ *  \brief An allocator which allocates storage with \p device_new
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/device_ptr.h>
+#include <thrust/device_reference.h>
+#include <thrust/device_new.h>
+#include <thrust/device_delete.h>
+#include <limits>
+#include <stdexcept>
+
+namespace thrust
+{
+
+/*! \addtogroup memory_management Memory Management
+ *  \addtogroup memory_management_classes Memory Management Classes
+ *  \ingroup memory_management
+ *  \{
+ */
+
+/*! \p device_new_allocator is a device memory allocator that employs the
+ *  \p device_new function for allocation.
+ *
+ *  \see device_new
+ *  \see device_ptr
+ *  \see http://www.sgi.com/tech/stl/Allocators.html
+ */
+template<typename T>
+  class device_new_allocator
+{
+  public:
+    /*! Type of element allocated, \c T. */
+    typedef T                                 value_type;
+
+    /*! Pointer to allocation, \c device_ptr<T>. */
+    typedef device_ptr<T>                     pointer;
+
+    /*! \c const pointer to allocation, \c device_ptr<const T>. */
+    typedef device_ptr<const T>               const_pointer;
+
+    /*! Reference to allocated element, \c device_reference<T>. */
+    typedef device_reference<T>               reference;
+
+    /*! \c const reference to allocated element, \c device_reference<const T>. */
+    typedef device_reference<const T>         const_reference;
+
+    /*! Type of allocation size, \c std::size_t. */
+    typedef std::size_t                       size_type;
+
+    /*! Type of allocation difference, \c pointer::difference_type. */
+    typedef typename pointer::difference_type difference_type;
+
+    /*! The \p rebind metafunction provides the type of a \p device_new_allocator
+     *  instantiated with another type.
+     *
+     *  \tparam U The other type to use for instantiation.
+     */
+    template<typename U>
+      struct rebind
+    {
+      /*! The typedef \p other gives the type of the rebound \p device_new_allocator.
+       */
+      typedef device_new_allocator<U> other;
+    }; // end rebind
+
+    /*! No-argument constructor has no effect. */
+    __host__ __device__
+    inline device_new_allocator() {}
+
+    /*! No-argument destructor has no effect. */
+    __host__ __device__
+    inline ~device_new_allocator() {}
+
+    /*! Copy constructor has no effect. */
+    __host__ __device__
+    inline device_new_allocator(device_new_allocator const&) {}
+
+    /*! Constructor from other \p device_malloc_allocator has no effect. */
+    template<typename U>
+    __host__ __device__
+    inline device_new_allocator(device_new_allocator<U> const&) {}
+
+    /*! Returns the address of an allocated object.
+     *  \return <tt>&r</tt>.
+     */
+    __host__ __device__
+    inline pointer address(reference r) { return &r; }
+    
+    /*! Returns the address an allocated object.
+     *  \return <tt>&r</tt>.
+     */
+    __host__ __device__
+    inline const_pointer address(const_reference r) { return &r; }
+
+    /*! Allocates storage for \p cnt objects.
+     *  \param cnt The number of objects to allocate.
+     *  \return A \p pointer to uninitialized storage for \p cnt objects.
+     *  \note Memory allocated by this function must be deallocated with \p deallocate.
+     */
+    __host__
+    inline pointer allocate(size_type cnt,
+                            const_pointer = const_pointer(static_cast<T*>(0)))
+    {
+      if(cnt > this->max_size())
+      {
+        throw std::bad_alloc();
+      } // end if
+
+      // use "::operator new" rather than keyword new
+      return pointer(device_new<T>(cnt));
+    } // end allocate()
+
+    /*! Deallocates storage for objects allocated with \p allocate.
+     *  \param p A \p pointer to the storage to deallocate.
+     *  \param cnt The size of the previous allocation.
+     *  \note Memory deallocated by this function must previously have been
+     *        allocated with \p allocate.
+     */
+    __host__
+    inline void deallocate(pointer p, size_type cnt)
+    {
+      // use "::operator delete" rather than keyword delete
+      device_delete(p);
+    } // end deallocate()
+
+    /*! Returns the largest value \c n for which <tt>allocate(n)</tt> might succeed.
+     *  \return The largest value \c n for which <tt>allocate(n)</tt> might succeed.
+     */
+    __host__ __device__
+    inline size_type max_size() const
+    {
+      return std::numeric_limits<size_type>::max THRUST_PREVENT_MACRO_SUBSTITUTION () / sizeof(T);
+    } // end max_size()
+
+    /*! Compares against another \p device_malloc_allocator for equality.
+     *  \return \c true
+     */
+    __host__ __device__
+    inline bool operator==(device_new_allocator const&) { return true; }
+
+    /*! Compares against another \p device_malloc_allocator for inequality.
+     *  \return \c false
+     */
+    __host__ __device__
+    inline bool operator!=(device_new_allocator const &a) {return !operator==(a); }
+}; // end device_new_allocator
+
+/*! \}
+ */
+
+} // end thrust
+
diff --git a/compat/thrust/device_ptr.h b/compat/thrust/device_ptr.h
new file mode 100644
index 0000000000..dfc7e90dc3
--- /dev/null
+++ b/compat/thrust/device_ptr.h
@@ -0,0 +1,170 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file device_ptr.h
+ *  \brief A pointer to a variable which resides in the "device" system's memory space
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/memory.h>
+#include <ostream>
+
+namespace thrust
+{
+
+/*! \addtogroup memory_management Memory Management
+ *  \addtogroup memory_management_classes Memory Management Classes
+ *  \ingroup memory_management
+ *  \{
+ */
+
+// forward declarations
+template<typename T> class device_reference;
+
+/*! \p device_ptr stores a pointer to an object allocated in device memory. This type
+ *  provides type safety when dispatching standard algorithms on ranges resident in
+ *  device memory.
+ *
+ *  \p device_ptr has pointer semantics: it may be dereferenced safely from the host and
+ *  may be manipulated with pointer arithmetic.
+ *
+ *  \p device_ptr can be created with the functions device_malloc, device_new, or
+ *  device_pointer_cast, or by explicitly calling its constructor with a raw pointer.
+ *
+ *  The raw pointer encapsulated by a \p device_ptr may be obtained by either its <tt>get</tt>
+ *  method or the \p raw_pointer_cast free function.
+ *
+ *  \note \p device_ptr is not a smart pointer; it is the programmer's responsibility to
+ *  deallocate memory pointed to by \p device_ptr.
+ *
+ *  \see device_malloc
+ *  \see device_new
+ *  \see device_pointer_cast
+ *  \see raw_pointer_cast
+ */
+template<typename T>
+  class device_ptr
+    : public thrust::pointer<
+               T,
+               thrust::device_system_tag,
+               thrust::device_reference<T>,
+               thrust::device_ptr<T>
+             >
+{
+  private:
+    typedef thrust::pointer<
+      T,
+      thrust::device_system_tag,
+      thrust::device_reference<T>,
+      thrust::device_ptr<T>
+    > super_t;
+
+  public:
+    /*! \p device_ptr's null constructor initializes its raw pointer to \c 0.
+     */
+    __host__ __device__
+    device_ptr() : super_t() {}
+
+    /*! \p device_ptr's copy constructor is templated to allow copying to a
+     *  <tt>device_ptr<const T></tt> from a <tt>T *</tt>.
+     *  
+     *  \param ptr A raw pointer to copy from, presumed to point to a location in
+     *         device memory.
+     */
+    template<typename OtherT>
+    __host__ __device__
+    explicit device_ptr(OtherT *ptr) : super_t(ptr) {}
+
+    /*! \p device_ptr's copy constructor allows copying from another device_ptr with related type.
+     *  \param other The \p device_ptr to copy from.
+     */
+    template<typename OtherT>
+    __host__ __device__
+    device_ptr(const device_ptr<OtherT> &other) : super_t(other) {}
+
+    /*! \p device_ptr's assignment operator allows assigning from another \p device_ptr with related type.
+     *  \param other The other \p device_ptr to copy from.
+     *  \return <tt>*this</tt>
+     */
+    template<typename OtherT>
+    __host__ __device__
+    device_ptr &operator=(const device_ptr<OtherT> &other)
+    {
+      super_t::operator=(other);
+      return *this;
+    }
+
+// declare these members for the purpose of Doxygenating them
+// they actually exist in a derived-from class
+#if 0
+    /*! This method returns this \p device_ptr's raw pointer.
+     *  \return This \p device_ptr's raw pointer.
+     */
+    __host__ __device__
+    T *get(void) const;
+#endif // end doxygen-only members
+}; // end device_ptr
+
+/*! This operator outputs the value of a \p device_ptr's raw pointer to a \p std::basic_ostream.
+ *
+ *  \param os The std::basic_ostream of interest.
+ *  \param p The device_ptr of interest.
+ *  \return os.
+ */
+template<class E, class T, class Y>
+inline std::basic_ostream<E, T> &operator<<(std::basic_ostream<E, T> &os, const device_ptr<Y> &p);
+
+/*! \}
+ */
+
+
+/*!
+ *  \addtogroup memory_management_functions Memory Management Functions
+ *  \ingroup memory_management
+ *  \{
+ */
+
+/*! \p device_pointer_cast creates a device_ptr from a raw pointer which is presumed to point
+ *  to a location in device memory.
+ *
+ *  \param ptr A raw pointer, presumed to point to a location in device memory.
+ *  \return A device_ptr wrapping ptr.
+ */
+template<typename T>
+__host__ __device__
+inline device_ptr<T> device_pointer_cast(T *ptr);
+
+/*! This version of \p device_pointer_cast creates a copy of a device_ptr from another device_ptr.
+ *  This version is included for symmetry with \p raw_pointer_cast.
+ *
+ *  \param ptr A device_ptr.
+ *  \return A copy of \p ptr.
+ */
+template<typename T>
+__host__ __device__
+inline device_ptr<T> device_pointer_cast(const device_ptr<T> &ptr);
+
+/*! \}
+ */
+
+} // end thrust
+
+#include <thrust/detail/device_ptr.inl>
+#include <thrust/detail/raw_pointer_cast.h>
+
diff --git a/compat/thrust/device_reference.h b/compat/thrust/device_reference.h
new file mode 100644
index 0000000000..edae2b59af
--- /dev/null
+++ b/compat/thrust/device_reference.h
@@ -0,0 +1,969 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file device_reference.h
+ *  \brief A reference to a variable which resides in the "device" system's memory space
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/device_ptr.h>
+#include <thrust/detail/type_traits.h>
+#include <thrust/detail/reference.h>
+
+namespace thrust
+{
+
+/*! \addtogroup memory_management_classes Memory Management Classes
+ *  \ingroup memory_management
+ *  \{
+ */
+
+/*! \p device_reference acts as a reference-like object to an object stored in device memory.
+ *  \p device_reference is not intended to be used directly; rather, this type
+ *  is the result of deferencing a \p device_ptr. Similarly, taking the address of
+ *  a \p device_reference yields a \p device_ptr.
+ *  
+ *  \p device_reference may often be used from host code in place of operations defined on
+ *  its associated \c value_type. For example, when \p device_reference refers to an
+ *  arithmetic type, arithmetic operations on it are legal:
+ *
+ *  \code
+ *  #include <thrust/device_vector.h>
+ *
+ *  int main(void)
+ *  {
+ *    thrust::device_vector<int> vec(1, 13);
+ *
+ *    thrust::device_reference<int> ref_to_thirteen = vec[0];
+ *
+ *    int x = ref_to_thirteen + 1;
+ *
+ *    // x is 14
+ *
+ *    return 0;
+ *  }
+ *  \endcode
+ *
+ *  Similarly, we can print the value of \c ref_to_thirteen in the above code by using an
+ *  \c iostream:
+ *
+ *  \code
+ *  #include <thrust/device_vector.h>
+ *  #include <iostream>
+ *
+ *  int main(void)
+ *  {
+ *    thrust::device_vector<int> vec(1, 13);
+ *
+ *    thrust::device_reference<int> ref_to_thirteen = vec[0];
+ *
+ *    std::cout << ref_to_thirteen << std::endl;
+ *
+ *    // 13 is printed
+ *
+ *    return 0;
+ *  }
+ *  \endcode
+ *
+ *  Of course, we needn't explicitly create a \p device_reference in the previous
+ *  example, because one is returned by \p device_vector's bracket operator. A more natural
+ *  way to print the value of a \p device_vector element might be:
+ *
+ *  \code
+ *  #include <thrust/device_vector.h>
+ *  #include <iostream>
+ *
+ *  int main(void)
+ *  {
+ *    thrust::device_vector<int> vec(1, 13);
+ *
+ *    std::cout << vec[0] << std::endl;
+ *
+ *    // 13 is printed
+ *
+ *    return 0;
+ *  }
+ *  \endcode
+ *
+ *  These kinds of operations should be used sparingly in performance-critical code, because
+ *  they imply a potentially expensive copy between host and device space.
+ *
+ *  Some operations which are possible with regular objects are impossible with their
+ *  corresponding \p device_reference objects due to the requirements of the C++ language. For
+ *  example, because the member access operator cannot be overloaded, member variables and functions
+ *  of a referent object cannot be directly accessed through its \p device_reference.
+ *
+ *  The following code, which generates a compiler error, illustrates:
+ *
+ *  \code
+ *  #include <thrust/device_vector.h>
+ *
+ *  struct foo
+ *  {
+ *    int x;
+ *  };
+ *
+ *  int main(void)
+ *  {
+ *    thrust::device_vector<foo> foo_vec(1);
+ *
+ *    thrust::device_reference<foo> foo_ref = foo_vec[0];
+ *
+ *    foo_ref.x = 13; // ERROR: x cannot be accessed through foo_ref
+ *
+ *    return 0;
+ *  }
+ *  \endcode
+ *
+ *  Instead, a host space copy must be created to access \c foo's \c x member:
+ *
+ *  \code
+ *  #include <thrust/device_vector.h>
+ *
+ *  struct foo
+ *  {
+ *    int x;
+ *  };
+ *
+ *  int main(void)
+ *  {
+ *    thrust::device_vector<foo> foo_vec(1);
+ *
+ *    // create a local host-side foo object
+ *    foo host_foo;
+ *    host_foo.x = 13;
+ *
+ *    thrust::device_reference<foo> foo_ref = foo_vec[0];
+ *
+ *    foo_ref = host_foo;
+ *
+ *    // foo_ref's x member is 13
+ *
+ *    return 0;
+ *  }
+ *  \endcode
+ *  
+ *  Another common case where a \p device_reference cannot directly be used in place of
+ *  its referent object occurs when passing them as parameters to functions like \c printf
+ *  which have varargs parameters. Because varargs parameters must be Plain Old Data, a
+ *  \p device_reference to a POD type requires a cast when passed to \c printf:
+ *
+ *  \code
+ *  #include <stdio.h>
+ *  #include <thrust/device_vector.h>
+ *
+ *  int main(void)
+ *  {
+ *    thrust::device_vector<int> vec(1,13);
+ *
+ *    // vec[0] must be cast to int when passing to printf
+ *    printf("%d\n", (int) vec[0]);
+ *
+ *    return 0;
+ *  }
+ *  \endcode
+ *
+ *  \see device_ptr
+ *  \see device_vector
+ */
+template<typename T>
+  class device_reference
+    : public thrust::reference<
+               T,
+               thrust::device_ptr<T>,
+               thrust::device_reference<T>
+             >
+{
+  private:
+    typedef thrust::reference<
+      T,
+      thrust::device_ptr<T>,
+      thrust::device_reference<T>
+    > super_t;
+
+  public:
+    /*! The type of the value referenced by this type of \p device_reference.
+     */
+    typedef typename super_t::value_type value_type;
+
+    /*! The type of the expression <tt>&ref</tt>, where <tt>ref</tt> is a \p device_reference.
+     */
+    typedef typename super_t::pointer    pointer;
+
+    /*! This copy constructor accepts a const reference to another
+     *  \p device_reference. After this \p device_reference is constructed,
+     *  it shall refer to the same object as \p other.
+     *  
+     *  \param other A \p device_reference to copy from.
+     *
+     *  The following code snippet demonstrates the semantics of this
+     *  copy constructor.
+     *
+     *  \code
+     *  #include <thrust/device_vector.h>
+     *  #include <assert.h>
+     *  ...
+     *  thrust::device_vector<int> v(1,0);
+     *  thrust::device_reference<int> ref = v[0];
+     *
+     *  // ref equals the object at v[0]
+     *  assert(ref == v[0]);
+     *
+     *  // the address of ref equals the address of v[0]
+     *  assert(&ref == &v[0]);
+     *
+     *  // modifying v[0] modifies ref
+     *  v[0] = 13;
+     *  assert(ref == 13);
+     *  \endcode
+     *
+     *  \note This constructor is templated primarily to allow initialization of 
+     *  <tt>device_reference<const T></tt> from <tt>device_reference<T></tt>.
+     */
+    template<typename OtherT>
+    __host__ __device__
+    device_reference(const device_reference<OtherT> &other,
+                     typename thrust::detail::enable_if_convertible<
+                       typename device_reference<OtherT>::pointer,
+                       pointer
+                     >::type * = 0)
+      : super_t(other)
+    {}
+
+    /*! This copy constructor initializes this \p device_reference
+     *  to refer to an object pointed to by the given \p device_ptr. After
+     *  this \p device_reference is constructed, it shall refer to the
+     *  object pointed to by \p ptr.
+     *
+     *  \param ptr A \p device_ptr to copy from.
+     *
+     *  The following code snippet demonstrates the semantic of this
+     *  copy constructor.
+     *
+     *  \code
+     *  #include <thrust/device_vector.h>
+     *  #include <assert.h>
+     *  ...
+     *  thrust::device_vector<int> v(1,0);
+     *  thrust::device_ptr<int> ptr = &v[0];
+     *  thrust::device_reference<int> ref(ptr);
+     *
+     *  // ref equals the object pointed to by ptr
+     *  assert(ref == *ptr);
+     *
+     *  // the address of ref equals ptr
+     *  assert(&ref == ptr);
+     *
+     *  // modifying *ptr modifies ref
+     *  *ptr = 13;
+     *  assert(ref == 13);
+     *  \endcode
+     */
+    __host__ __device__
+    explicit device_reference(const pointer &ptr)
+      : super_t(ptr)
+    {}
+
+    /*! This assignment operator assigns the value of the object referenced by
+     *  the given \p device_reference to the object referenced by this
+     *  \p device_reference.
+     *
+     *  \param other The \p device_reference to assign from.
+     *  \return <tt>*this</tt>
+     */
+    template<typename OtherT>
+    __host__ __device__
+    device_reference &operator=(const device_reference<OtherT> &other);
+
+    /*! Assignment operator assigns the value of the given value to the
+     *  value referenced by this \p device_reference.
+     *  
+     *  \param x The value to assign from.
+     *  \return <tt>*this</tt>
+     */
+    __host__ __device__
+    device_reference &operator=(const value_type &x);
+
+// declare these members for the purpose of Doxygenating them
+// they actually exist in a derived-from class
+#if 0
+    /*! Address-of operator returns a \p device_ptr pointing to the object
+     *  referenced by this \p device_reference. It does not return the
+     *  address of this \p device_reference.
+     *
+     *  \return A \p device_ptr pointing to the object this
+     *  \p device_reference references.
+     */
+    __host__ __device__
+    pointer operator&(void) const;
+
+    /*! Conversion operator converts this \p device_reference to T
+     *  by returning a copy of the object referenced by this
+     *  \p device_reference.
+     *
+     *  \return A copy of the object referenced by this \p device_reference.
+     */
+    __host__ __device__
+    operator value_type (void) const;
+
+    /*! swaps the value this \p device_reference references with another.
+     *  \p other The other \p device_reference with which to swap.
+     */
+    __host__ __device__
+    void swap(device_reference &other);
+
+    /*! Prefix increment operator increments the object referenced by this
+     *  \p device_reference.
+     *
+     *  \return <tt>*this</tt>
+     *  
+     *  The following code snippet demonstrates the semantics of
+     *  \p device_reference's prefix increment operator.
+     *
+     *  \code
+     *  #include <thrust/device_vector.h>
+     *  #include <assert.h>
+     *  ...
+     *  thrust::device_vector<int> v(1,0);
+     *  thrust::device_ptr<int> ptr = &v[0];
+     *  thrust::device_reference<int> ref(ptr);
+     *
+     *  // ref equals 0
+     *  assert(ref == 0);
+     *
+     *  // the object pointed to by ptr equals 1
+     *  assert(*ptr == 1);
+     *
+     *  // v[0] equals 1
+     *  assert(v[0] == 1);
+     *
+     *  // increment ref
+     *  ++ref;
+     *
+     *  // ref equals 1
+     *  assert(ref == 1);
+     *
+     *  // the object pointed to by ptr equals 1
+     *  assert(*ptr == 1);
+     *
+     *  // v[0] equals 1
+     *  assert(v[0] == 1);
+     *  \endcode
+     *
+     *  \note The increment executes as if it were executed on the host.
+     *  This may change in a later version.
+     */
+    device_reference &operator++(void);
+
+    /*! Postfix increment operator copies the object referenced by this
+     *  \p device_reference, increments the object referenced by this
+     *  \p device_reference, and returns the copy.
+     *
+     *  \return A copy of the object referenced by this \p device_reference
+     *          before being incremented.
+     *
+     *  The following code snippet demonstrates the semantics of
+     *  \p device_reference's postfix increment operator.
+     *
+     *  \code
+     *  #include <thrust/device_vector.h>
+     *  #include <assert.h>
+     *  ...
+     *  thrust::device_vector<int> v(1,0);
+     *  thrust::device_ptr<int> ptr = &v[0];
+     *  thrust::device_reference<int> ref(ptr);
+     *
+     *  // ref equals 0
+     *  assert(ref == 0);
+     *
+     *  // the object pointed to by ptr equals 0
+     *  assert(*ptr == 0);
+     *
+     *  // v[0] equals 0
+     *  assert(v[0] == 0);
+     *
+     *  // increment ref
+     *  int x = ref++;
+     *
+     *  // x equals 0
+     *  assert(x == 0)
+     *
+     *  // ref equals 1
+     *  assert(ref == 1);
+     *
+     *  // the object pointed to by ptr equals 1
+     *  assert(*ptr == 1);
+     *
+     *  // v[0] equals 1
+     *  assert(v[0] == 1);
+     *  \endcode
+     *
+     *  \note The increment executes as if it were executed on the host.
+     *  This may change in a later version.
+     */
+    value_type operator++(int);
+
+    /*! Addition assignment operator add-assigns the object referenced by this
+     *  \p device_reference and returns this \p device_reference.
+     *
+     *  \param rhs The right hand side of the add-assignment.
+     *  \return <tt>*this</tt>.
+     *
+     *  The following code snippet demonstrates the semantics of
+     *  \p device_reference's addition assignment operator.
+     *
+     *  \code
+     *  #include <thrust/device_vector.h>
+     *  #include <assert.h>
+     *  ...
+     *  thrust::device_vector<int> v(1,0);
+     *  thrust::device_ptr<int> ptr = &v[0];
+     *  thrust::device_reference<int> ref(ptr);
+     *
+     *  // ref equals 0
+     *  assert(ref == 0);
+     *
+     *  // the object pointed to by ptr equals 0
+     *  assert(*ptr == 0);
+     *
+     *  // v[0] equals 0
+     *  assert(v[0] == 0);
+     *
+     *  // add-assign ref
+     *  ref += 5;
+     *
+     *  // ref equals 5
+     *  assert(ref == 5);
+     *
+     *  // the object pointed to by ptr equals 5
+     *  assert(*ptr == 5);
+     *
+     *  // v[0] equals 5
+     *  assert(v[0] == 5);
+     *  \endcode
+     *
+     *  \note The add-assignment executes as as if it were executed on the host.
+     *  This may change in a later version.
+     */
+    device_reference &operator+=(const T &rhs);
+
+    /*! Prefix decrement operator decrements the object referenced by this
+     *  \p device_reference.
+     *
+     *  \return <tt>*this</tt>
+     *  
+     *  The following code snippet demonstrates the semantics of
+     *  \p device_reference's prefix decrement operator.
+     *
+     *  \code
+     *  #include <thrust/device_vector.h>
+     *  #include <assert.h>
+     *  ...
+     *  thrust::device_vector<int> v(1,0);
+     *  thrust::device_ptr<int> ptr = &v[0];
+     *  thrust::device_reference<int> ref(ptr);
+     *
+     *  // ref equals 0
+     *  assert(ref == 0);
+     *
+     *  // the object pointed to by ptr equals 0
+     *  assert(*ptr == 0);
+     *
+     *  // v[0] equals 0
+     *  assert(v[0] == 0);
+     *
+     *  // decrement ref
+     *  --ref;
+     *
+     *  // ref equals -1
+     *  assert(ref == -1);
+     *
+     *  // the object pointed to by ptr equals -1
+     *  assert(*ptr == -1);
+     *
+     *  // v[0] equals -1
+     *  assert(v[0] == -1);
+     *  \endcode
+     *
+     *  \note The decrement executes as if it were executed on the host.
+     *  This may change in a later version.
+     */
+    device_reference &operator--(void);
+
+    /*! Postfix decrement operator copies the object referenced by this
+     *  \p device_reference, decrements the object referenced by this
+     *  \p device_reference, and returns the copy.
+     *
+     *  \return A copy of the object referenced by this \p device_reference
+     *          before being decremented.
+     *
+     *  The following code snippet demonstrates the semantics of
+     *  \p device_reference's postfix decrement operator.
+     *
+     *  \code
+     *  #include <thrust/device_vector.h>
+     *  #include <assert.h>
+     *  ...
+     *  thrust::device_vector<int> v(1,0);
+     *  thrust::device_ptr<int> ptr = &v[0];
+     *  thrust::device_reference<int> ref(ptr);
+     *
+     *  // ref equals 0
+     *  assert(ref == 0);
+     *
+     *  // the object pointed to by ptr equals 0
+     *  assert(*ptr == 0);
+     *
+     *  // v[0] equals 0
+     *  assert(v[0] == 0);
+     *
+     *  // decrement ref
+     *  int x = ref--;
+     *
+     *  // x equals 0
+     *  assert(x == 0)
+     *
+     *  // ref equals -1
+     *  assert(ref == -1);
+     *
+     *  // the object pointed to by ptr equals -1
+     *  assert(*ptr == -1);
+     *
+     *  // v[0] equals -1
+     *  assert(v[0] == -1);
+     *  \endcode
+     *
+     *  \note The decrement executes as if it were executed on the host.
+     *  This may change in a later version.
+     */
+    value_type operator--(int);
+
+    /*! Subtraction assignment operator subtract-assigns the object referenced by this
+     *  \p device_reference and returns this \p device_reference.
+     *
+     *  \param rhs The right hand side of the subtraction-assignment.
+     *  \return <tt>*this</tt>.
+     *
+     *  The following code snippet demonstrates the semantics of
+     *  \p device_reference's addition assignment operator.
+     *
+     *  \code
+     *  #include <thrust/device_vector.h>
+     *  #include <assert.h>
+     *  ...
+     *  thrust::device_vector<int> v(1,0);
+     *  thrust::device_ptr<int> ptr = &v[0];
+     *  thrust::device_reference<int> ref(ptr);
+     *
+     *  // ref equals 0
+     *  assert(ref == 0);
+     *
+     *  // the object pointed to by ptr equals 0
+     *  assert(*ptr == 0);
+     *
+     *  // v[0] equals 0
+     *  assert(v[0] == 0);
+     *
+     *  // subtract-assign ref
+     *  ref -= 5;
+     *
+     *  // ref equals -5
+     *  assert(ref == -5);
+     *
+     *  // the object pointed to by ptr equals -5
+     *  assert(*ptr == -5);
+     *
+     *  // v[0] equals -5
+     *  assert(v[0] == -5);
+     *  \endcode
+     *
+     *  \note The subtract-assignment executes as as if it were executed on the host.
+     *  This may change in a later version.
+     */
+    device_reference &operator-=(const T &rhs);
+
+    /*! Multiplication assignment operator multiply-assigns the object referenced by this
+     *  \p device_reference and returns this \p device_reference.
+     *
+     *  \param rhs The right hand side of the multiply-assignment.
+     *  \return <tt>*this</tt>.
+     *
+     *  The following code snippet demonstrates the semantics of
+     *  \p device_reference's multiply assignment operator.
+     *
+     *  \code
+     *  #include <thrust/device_vector.h>
+     *  #include <assert.h>
+     *  ...
+     *  thrust::device_vector<int> v(1,1);
+     *  thrust::device_ptr<int> ptr = &v[0];
+     *  thrust::device_reference<int> ref(ptr);
+     *
+     *  // ref equals 1
+     *  assert(ref == 1);
+     *
+     *  // the object pointed to by ptr equals 1
+     *  assert(*ptr == 1);
+     *
+     *  // v[0] equals 1
+     *  assert(v[0] == 1);
+     *
+     *  // multiply-assign ref
+     *  ref *= 5;
+     *
+     *  // ref equals 5
+     *  assert(ref == 5);
+     *
+     *  // the object pointed to by ptr equals 5
+     *  assert(*ptr == 5);
+     *
+     *  // v[0] equals 5
+     *  assert(v[0] == 5);
+     *  \endcode
+     *
+     *  \note The multiply-assignment executes as as if it were executed on the host.
+     *  This may change in a later version.
+     */
+    device_reference &operator*=(const T &rhs);
+
+    /*! Division assignment operator divide-assigns the object referenced by this
+     *  \p device_reference and returns this \p device_reference.
+     *
+     *  \param rhs The right hand side of the divide-assignment.
+     *  \return <tt>*this</tt>.
+     *
+     *  The following code snippet demonstrates the semantics of
+     *  \p device_reference's divide assignment operator.
+     *
+     *  \code
+     *  #include <thrust/device_vector.h>
+     *  #include <assert.h>
+     *  ...
+     *  thrust::device_vector<int> v(1,5);
+     *  thrust::device_ptr<int> ptr = &v[0];
+     *  thrust::device_reference<int> ref(ptr);
+     *
+     *  // ref equals 5
+     *  assert(ref == 5);
+     *
+     *  // the object pointed to by ptr equals 5
+     *  assert(*ptr == 5);
+     *
+     *  // v[0] equals 5
+     *  assert(v[0] == 5);
+     *
+     *  // divide-assign ref
+     *  ref /= 5;
+     *
+     *  // ref equals 1
+     *  assert(ref == 1);
+     *
+     *  // the object pointed to by ptr equals 1
+     *  assert(*ptr == 1);
+     *
+     *  // v[0] equals 1
+     *  assert(v[0] == 1);
+     *  \endcode
+     *
+     *  \note The divide-assignment executes as as if it were executed on the host.
+     *  This may change in a later version.
+     */
+    device_reference &operator/=(const T &rhs);
+
+    /*! Modulation assignment operator modulus-assigns the object referenced by this
+     *  \p device_reference and returns this \p device_reference.
+     *
+     *  \param rhs The right hand side of the divide-assignment.
+     *  \return <tt>*this</tt>.
+     *
+     *  The following code snippet demonstrates the semantics of
+     *  \p device_reference's divide assignment operator.
+     *
+     *  \code
+     *  #include <thrust/device_vector.h>
+     *  #include <assert.h>
+     *  ...
+     *  thrust::device_vector<int> v(1,5);
+     *  thrust::device_ptr<int> ptr = &v[0];
+     *  thrust::device_reference<int> ref(ptr);
+     *
+     *  // ref equals 5
+     *  assert(ref == 5);
+     *
+     *  // the object pointed to by ptr equals 5
+     *  assert(*ptr == 5);
+     *
+     *  // v[0] equals 5
+     *  assert(v[0] == 5);
+     *
+     *  // modulus-assign ref
+     *  ref %= 5;
+     *
+     *  // ref equals 0
+     *  assert(ref == 0);
+     *
+     *  // the object pointed to by ptr equals 0
+     *  assert(*ptr == 0);
+     *
+     *  // v[0] equals 0
+     *  assert(v[0] == 0);
+     *  \endcode
+     *
+     *  \note The modulus-assignment executes as as if it were executed on the host.
+     *  This may change in a later version.
+     */
+    device_reference &operator%=(const T &rhs);
+
+    /*! Bitwise left shift assignment operator left shift-assigns the object referenced by this
+     *  \p device_reference and returns this \p device_reference.
+     *
+     *  \param rhs The right hand side of the left shift-assignment.
+     *  \return <tt>*this</tt>.
+     *
+     *  The following code snippet demonstrates the semantics of
+     *  \p device_reference's left shift assignment operator.
+     *
+     *  \code
+     *  #include <thrust/device_vector.h>
+     *  #include <assert.h>
+     *  ...
+     *  thrust::device_vector<int> v(1,1);
+     *  thrust::device_ptr<int> ptr = &v[0];
+     *  thrust::device_reference<int> ref(ptr);
+     *
+     *  // ref equals 1
+     *  assert(ref == 1);
+     *
+     *  // the object pointed to by ptr equals 1
+     *  assert(*ptr == 1);
+     *
+     *  // v[0] equals 1
+     *  assert(v[0] == 1);
+     *
+     *  // left shift-assign ref
+     *  ref <<= 1;
+     *
+     *  // ref equals 2
+     *  assert(ref == 2);
+     *
+     *  // the object pointed to by ptr equals 2
+     *  assert(*ptr == 2);
+     *
+     *  // v[0] equals 2
+     *  assert(v[0] == 2);
+     *  \endcode
+     *
+     *  \note The left shift-assignment executes as as if it were executed on the host.
+     *  This may change in a later version.
+     */
+    device_reference &operator<<=(const T &rhs);
+
+    /*! Bitwise right shift assignment operator right shift-assigns the object referenced by this
+     *  \p device_reference and returns this \p device_reference.
+     *
+     *  \param rhs The right hand side of the right shift-assignment.
+     *  \return <tt>*this</tt>.
+     *
+     *  The following code snippet demonstrates the semantics of
+     *  \p device_reference's right shift assignment operator.
+     *
+     *  \code
+     *  #include <thrust/device_vector.h>
+     *  #include <assert.h>
+     *  ...
+     *  thrust::device_vector<int> v(1,2);
+     *  thrust::device_ptr<int> ptr = &v[0];
+     *  thrust::device_reference<int> ref(ptr);
+     *
+     *  // ref equals 2
+     *  assert(ref == 2);
+     *
+     *  // the object pointed to by ptr equals 2
+     *  assert(*ptr == 2);
+     *
+     *  // v[0] equals 2
+     *  assert(v[0] == 2);
+     *
+     *  // right shift-assign ref
+     *  ref >>= 1;
+     *
+     *  // ref equals 1
+     *  assert(ref == 1);
+     *
+     *  // the object pointed to by ptr equals 1
+     *  assert(*ptr == 1);
+     *
+     *  // v[0] equals 1
+     *  assert(v[0] == 1);
+     *  \endcode
+     *
+     *  \note The right shift-assignment executes as as if it were executed on the host.
+     *  This may change in a later version.
+     */
+    device_reference &operator>>=(const T &rhs);
+
+    /*! Bitwise AND assignment operator AND-assigns the object referenced by this
+     *  \p device_reference and returns this \p device_reference.
+     *
+     *  \param rhs The right hand side of the AND-assignment.
+     *  \return <tt>*this</tt>.
+     *
+     *  The following code snippet demonstrates the semantics of
+     *  \p device_reference's AND assignment operator.
+     *
+     *  \code
+     *  #include <thrust/device_vector.h>
+     *  #include <assert.h>
+     *  ...
+     *  thrust::device_vector<int> v(1,1);
+     *  thrust::device_ptr<int> ptr = &v[0];
+     *  thrust::device_reference<int> ref(ptr);
+     *
+     *  // ref equals 1
+     *  assert(ref == 1);
+     *
+     *  // the object pointed to by ptr equals 1
+     *  assert(*ptr == 1);
+     *
+     *  // v[0] equals 1
+     *  assert(v[0] == 1);
+     *
+     *  // right AND-assign ref
+     *  ref &= 0;
+     *
+     *  // ref equals 0
+     *  assert(ref == 0);
+     *
+     *  // the object pointed to by ptr equals 0
+     *  assert(*ptr == 0);
+     *
+     *  // v[0] equals 0
+     *  assert(v[0] == 0);
+     *  \endcode
+     *
+     *  \note The AND-assignment executes as as if it were executed on the host.
+     *  This may change in a later version.
+     */
+    device_reference &operator&=(const T &rhs);
+
+    /*! Bitwise OR assignment operator OR-assigns the object referenced by this
+     *  \p device_reference and returns this \p device_reference.
+     *
+     *  \param rhs The right hand side of the OR-assignment.
+     *  \return <tt>*this</tt>.
+     *
+     *  The following code snippet demonstrates the semantics of
+     *  \p device_reference's OR assignment operator.
+     *
+     *  \code
+     *  #include <thrust/device_vector.h>
+     *  #include <assert.h>
+     *  ...
+     *  thrust::device_vector<int> v(1,0);
+     *  thrust::device_ptr<int> ptr = &v[0];
+     *  thrust::device_reference<int> ref(ptr);
+     *
+     *  // ref equals 0
+     *  assert(ref == 0);
+     *
+     *  // the object pointed to by ptr equals 0
+     *  assert(*ptr == 0);
+     *
+     *  // v[0] equals 0
+     *  assert(v[0] == 0);
+     *
+     *  // right OR-assign ref
+     *  ref |= 1;
+     *
+     *  // ref equals 1
+     *  assert(ref == 1);
+     *
+     *  // the object pointed to by ptr equals 1
+     *  assert(*ptr == 1);
+     *
+     *  // v[0] equals 1
+     *  assert(v[0] == 1);
+     *  \endcode
+     *
+     *  \note The OR-assignment executes as as if it were executed on the host.
+     *  This may change in a later version.
+     */
+    device_reference &operator|=(const T &rhs);
+
+    /*! Bitwise XOR assignment operator XOR-assigns the object referenced by this
+     *  \p device_reference and returns this \p device_reference.
+     *
+     *  \param rhs The right hand side of the XOR-assignment.
+     *  \return <tt>*this</tt>.
+     *
+     *  The following code snippet demonstrates the semantics of
+     *  \p device_reference's XOR assignment operator.
+     *
+     *  \code
+     *  #include <thrust/device_vector.h>
+     *  #include <assert.h>
+     *  ...
+     *  thrust::device_vector<int> v(1,1);
+     *  thrust::device_ptr<int> ptr = &v[0];
+     *  thrust::device_reference<int> ref(ptr);
+     *
+     *  // ref equals 1
+     *  assert(ref == 1);
+     *
+     *  // the object pointed to by ptr equals 1
+     *  assert(*ptr == 1);
+     *
+     *  // v[0] equals 1
+     *  assert(v[0] == 1);
+     *
+     *  // right XOR-assign ref
+     *  ref ^= 1;
+     *
+     *  // ref equals 0
+     *  assert(ref == 0);
+     *
+     *  // the object pointed to by ptr equals 0
+     *  assert(*ptr == 0);
+     *
+     *  // v[0] equals 0
+     *  assert(v[0] == 0);
+     *  \endcode
+     *
+     *  \note The XOR-assignment executes as as if it were executed on the host.
+     *  This may change in a later version.
+     */
+    device_reference &operator^=(const T &rhs);
+#endif // end doxygen-only members
+}; // end device_reference
+
+/*! swaps the value of one \p device_reference with another.
+ *  \p x The first \p device_reference of interest.
+ *  \p y The second \p device_reference of interest.
+ */
+template<typename T>
+__host__ __device__
+void swap(device_reference<T> &x, device_reference<T> &y);
+
+/*! \}
+ */
+
+} // end thrust
+
+#include <thrust/detail/device_reference.inl>
+
diff --git a/compat/thrust/device_vector.h b/compat/thrust/device_vector.h
new file mode 100644
index 0000000000..8c9d0051a2
--- /dev/null
+++ b/compat/thrust/device_vector.h
@@ -0,0 +1,418 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file device_vector.h
+ *  \brief A dynamically-sizable array of elements which reside in the "device" memory space
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/device_malloc_allocator.h>
+#include <thrust/detail/vector_base.h>
+#include <vector>
+
+namespace thrust
+{
+
+// forward declaration of host_vector
+template<typename T, typename Alloc> class host_vector;
+
+/*! \addtogroup container_classes Container Classes
+ *  \addtogroup device_containers Device Containers
+ *  \ingroup container_classes
+ *  \{
+ */
+
+/*! A \p device_vector is a container that supports random access to elements,
+ *  constant time removal of elements at the end, and linear time insertion
+ *  and removal of elements at the beginning or in the middle. The number of
+ *  elements in a \p device_vector may vary dynamically; memory management is
+ *  automatic. The memory associated with a \p device_vector resides in the memory
+ *  space of a parallel device.
+ *
+ *  \see http://www.sgi.com/tech/stl/Vector.html
+ *  \see host_vector
+ */
+template<typename T, typename Alloc = thrust::device_malloc_allocator<T> >
+  class device_vector
+    : public detail::vector_base<T,Alloc>
+{
+  private:
+    typedef detail::vector_base<T,Alloc> Parent;
+
+  public:
+    /*! \cond */
+    typedef typename Parent::size_type  size_type;
+    typedef typename Parent::value_type value_type;
+    /*! \endcond */
+
+    /*! This constructor creates an empty \p device_vector.
+     */
+    __host__
+    device_vector(void)
+      :Parent() {}
+
+    /*! This constructor creates a \p device_vector with the given
+     *  size.
+     *  \param n The number of elements to initially craete.
+     */
+    __host__
+    explicit device_vector(size_type n)
+      :Parent(n) {}
+
+    /*! This constructor creates a \p device_vector with copies
+     *  of an exemplar element.
+     *  \param n The number of elements to initially create.
+     *  \param value An element to copy.
+     */
+    __host__
+    explicit device_vector(size_type n, const value_type &value)
+      :Parent(n,value) {}
+
+    /*! Copy constructor copies from an exemplar \p device_vector.
+     *  \param v The \p device_vector to copy.
+     */
+    __host__
+    device_vector(const device_vector &v)
+      :Parent(v) {}
+
+    /*! Copy constructor copies from an exemplar \p device_vector with different type.
+     *  \param v The \p device_vector to copy.
+     */
+    template<typename OtherT, typename OtherAlloc>
+    __device__
+    device_vector(const device_vector<OtherT,OtherAlloc> &v)
+      :Parent(v) {}
+
+    /*! Assign operator copies from an exemplar \p device_vector with different type.
+     *  \param v The \p device_vector to copy.
+     */
+    template<typename OtherT, typename OtherAlloc>
+    __device__
+    device_vector &operator=(const device_vector<OtherT,OtherAlloc> &v)
+    { Parent::operator=(v); return *this; }
+
+    /*! Copy constructor copies from an exemplar \c std::vector.
+     *  \param v The <tt>std::vector</tt> to copy.
+     */
+    template<typename OtherT, typename OtherAlloc>
+    __host__
+    device_vector(const std::vector<OtherT,OtherAlloc> &v)
+      :Parent(v) {}
+
+    /*! Assign operator copies from an exemplar <tt>std::vector</tt>.
+     *  \param v The <tt>std::vector</tt> to copy.
+     */
+    template<typename OtherT, typename OtherAlloc>
+    __host__
+    device_vector &operator=(const std::vector<OtherT,OtherAlloc> &v)
+    { Parent::operator=(v); return *this;}
+
+    /*! Copy constructor copies from an exemplar \p host_vector with possibly different type.
+     *  \param v The \p host_vector to copy.
+     */
+    template<typename OtherT, typename OtherAlloc>
+    __host__
+    device_vector(const host_vector<OtherT,OtherAlloc> &v);
+
+    /*! Assign operator copies from an examplar \p host_vector.
+     *  \param v The \p host_vector to copy.
+     */
+    template<typename OtherT, typename OtherAlloc>
+    __host__
+    device_vector &operator=(const host_vector<OtherT,OtherAlloc> &v)
+    { Parent::operator=(v); return *this; }
+
+    /*! This constructor builds a \p device_vector from a range.
+     *  \param first The beginning of the range.
+     *  \param last The end of the range.
+     */
+    template<typename InputIterator>
+    __host__
+    device_vector(InputIterator first, InputIterator last)
+      :Parent(first,last) {}
+
+// declare these members for the purpose of Doxygenating them
+// they actually exist in a derived-from class
+#if 0
+    /*! \brief Resizes this vector to the specified number of elements.
+     *  \param new_size Number of elements this vector should contain.
+     *  \param x Data with which new elements should be populated.
+     *  \throw std::length_error If n exceeds max_size().
+     *
+     *  This method will resize this vector to the specified number of
+     *  elements.  If the number is smaller than this vector's current
+     *  size this vector is truncated, otherwise this vector is
+     *  extended and new elements are populated with given data.
+     */
+    void resize(size_type new_size, const value_type &x = value_type());
+
+    /*! Returns the number of elements in this vector.
+     */
+    size_type size(void) const;
+
+    /*! Returns the size() of the largest possible vector.
+     *  \return The largest possible return value of size().
+     */
+    size_type max_size(void) const;
+
+    /*! \brief If n is less than or equal to capacity(), this call has no effect.
+     *         Otherwise, this method is a request for allocation of additional memory. If
+     *         the request is successful, then capacity() is greater than or equal to
+     *         n; otherwise, capacity() is unchanged. In either case, size() is unchanged.
+     *  \throw std::length_error If n exceeds max_size().
+     */
+    void reserve(size_type n);
+
+    /*! Returns the number of elements which have been reserved in this
+     *  vector.
+     */
+    size_type capacity(void) const;
+
+    /*! This method shrinks the capacity of this vector to exactly
+     *  fit its elements.
+     */
+    void shrink_to_fit(void);
+
+    /*! \brief Subscript access to the data contained in this vector_dev.
+     *  \param n The index of the element for which data should be accessed.
+     *  \return Read/write reference to data.
+     *
+     *  This operator allows for easy, array-style, data access.
+     *  Note that data access with this operator is unchecked and
+     *  out_of_range lookups are not defined.
+     */
+    reference operator[](size_type n);
+
+    /*! \brief Subscript read access to the data contained in this vector_dev.
+     *  \param n The index of the element for which data should be accessed.
+     *  \return Read reference to data.
+     *
+     *  This operator allows for easy, array-style, data access.
+     *  Note that data access with this operator is unchecked and
+     *  out_of_range lookups are not defined.
+     */
+    const_reference operator[](size_type n) const;
+
+    /*! This method returns an iterator pointing to the beginning of
+     *  this vector.
+     *  \return mStart
+     */
+    iterator begin(void);
+
+    /*! This method returns a const_iterator pointing to the beginning
+     *  of this vector.
+     *  \return mStart
+     */
+    const_iterator begin(void) const;
+
+    /*! This method returns a const_iterator pointing to the beginning
+     *  of this vector.
+     *  \return mStart
+     */
+    const_iterator cbegin(void) const;
+
+    /*! This method returns a reverse_iterator pointing to the beginning of
+     *  this vector's reversed sequence.
+     *  \return A reverse_iterator pointing to the beginning of this
+     *          vector's reversed sequence.
+     */
+    reverse_iterator rbegin(void);
+
+    /*! This method returns a const_reverse_iterator pointing to the beginning of
+     *  this vector's reversed sequence.
+     *  \return A const_reverse_iterator pointing to the beginning of this
+     *          vector's reversed sequence.
+     */
+    const_reverse_iterator rbegin(void) const;
+
+    /*! This method returns a const_reverse_iterator pointing to the beginning of
+     *  this vector's reversed sequence.
+     *  \return A const_reverse_iterator pointing to the beginning of this
+     *          vector's reversed sequence.
+     */
+    const_reverse_iterator crbegin(void) const;
+
+    /*! This method returns an iterator pointing to one element past the
+     *  last of this vector.
+     *  \return begin() + size().
+     */
+    iterator end(void);
+
+    /*! This method returns a const_iterator pointing to one element past the
+     *  last of this vector.
+     *  \return begin() + size().
+     */
+    const_iterator end(void) const;
+
+    /*! This method returns a const_iterator pointing to one element past the
+     *  last of this vector.
+     *  \return begin() + size().
+     */
+    const_iterator cend(void) const;
+
+    /*! This method returns a reverse_iterator pointing to one element past the
+     *  last of this vector's reversed sequence.
+     *  \return rbegin() + size().
+     */
+    reverse_iterator rend(void);
+
+    /*! This method returns a const_reverse_iterator pointing to one element past the
+     *  last of this vector's reversed sequence.
+     *  \return rbegin() + size().
+     */
+    const_reverse_iterator rend(void) const;
+
+    /*! This method returns a const_reverse_iterator pointing to one element past the
+     *  last of this vector's reversed sequence.
+     *  \return rbegin() + size().
+     */
+    const_reverse_iterator crend(void) const;
+
+    /*! This method returns a const_reference referring to the first element of this
+     *  vector.
+     *  \return The first element of this vector.
+     */
+    const_reference front(void) const;
+
+    /*! This method returns a reference pointing to the first element of this
+     *  vector.
+     *  \return The first element of this vector.
+     */
+    reference front(void);
+
+    /*! This method returns a const reference pointing to the last element of
+     *  this vector.
+     *  \return The last element of this vector.
+     */
+    const_reference back(void) const;
+
+    /*! This method returns a reference referring to the last element of
+     *  this vector_dev.
+     *  \return The last element of this vector.
+     */
+    reference back(void);
+
+    /*! This method returns a pointer to this vector's first element.
+     *  \return A pointer to the first element of this vector.
+     */
+    pointer data(void);
+
+    /*! This method returns a const_pointer to this vector's first element.
+     *  \return a const_pointer to the first element of this vector.
+     */
+    const_pointer data(void) const;
+
+    /*! This method resizes this vector to 0.
+     */
+    void clear(void);
+
+    /*! This method returns true iff size() == 0.
+     *  \return true if size() == 0; false, otherwise.
+     */
+    bool empty(void) const;
+
+    /*! This method appends the given element to the end of this vector.
+     *  \param x The element to append.
+     */
+    void push_back(const value_type &x);
+
+    /*! This method erases the last element of this vector, invalidating
+     *  all iterators and references to it.
+     */
+    void pop_back(void);
+
+    /*! This method swaps the contents of this vector_base with another vector.
+     *  \param v The vector with which to swap.
+     */
+    void swap(device_vector &v);
+
+    /*! This method removes the element at position pos.
+     *  \param pos The position of the element of interest.
+     *  \return An iterator pointing to the new location of the element that followed the element
+     *          at position pos.
+     */
+    iterator erase(iterator pos);
+
+    /*! This method removes the range of elements [first,last) from this vector.
+     *  \param first The beginning of the range of elements to remove.
+     *  \param last The end of the range of elements to remove.
+     *  \return An iterator pointing to the new location of the element that followed the last
+     *          element in the sequence [first,last).
+     */
+    iterator erase(iterator first, iterator last);
+
+    /*! This method inserts a single copy of a given exemplar value at the
+     *  specified position in this vector.
+     *  \param position The insertion position.
+     *  \param x The exemplar element to copy & insert.
+     *  \return An iterator pointing to the newly inserted element.
+     */
+    iterator insert(iterator position, const T &x); 
+
+    /*! This method inserts a copy of an exemplar value to a range at the
+     *  specified position in this vector.
+     *  \param position The insertion position
+     *  \param n The number of insertions to perform.
+     *  \param x The value to replicate and insert.
+     */
+    void insert(iterator position, size_type n, const T &x);
+
+    /*! This method inserts a copy of an input range at the specified position
+     *  in this vector.
+     *  \param position The insertion position.
+     *  \param first The beginning of the range to copy.
+     *  \param last  The end of the range to copy.
+     *
+     *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html>Input Iterator</a>,
+     *                        and \p InputIterator's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/Assignable.html">Assignable</a>.
+     */
+    template<typename InputIterator>
+    void insert(iterator position, InputIterator first, InputIterator last);
+
+    /*! This version of \p assign replicates a given exemplar
+     *  \p n times into this vector.
+     *  \param n The number of times to copy \p x.
+     *  \param x The exemplar element to replicate.
+     */
+    void assign(size_type n, const T &x);
+
+    /*! This version of \p assign makes this vector a copy of a given input range.
+     *  \param first The beginning of the range to copy.
+     *  \param last  The end of the range to copy.
+     *
+     *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator">Input Iterator</a>.
+     */
+    template<typename InputIterator>
+    void assign(InputIterator first, InputIterator last);
+
+    /*! This method returns a copy of this vector's allocator.
+     *  \return A copy of the alloctor used by this vector.
+     */
+    allocator_type get_allocator(void) const;
+#endif // end doxygen-only members
+}; // end device_vector
+
+/*! \}
+ */
+
+} // end thrust
+
+#include <thrust/detail/device_vector.inl>
+
+
diff --git a/compat/thrust/distance.h b/compat/thrust/distance.h
new file mode 100644
index 0000000000..67b41946bf
--- /dev/null
+++ b/compat/thrust/distance.h
@@ -0,0 +1,76 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file distance.h
+ *  \brief Computes the size of a range
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/iterator/iterator_traits.h>
+
+namespace thrust
+{
+
+
+/*! \addtogroup iterators
+ *  \{
+ */
+
+/*! \p distance finds the distance between \p first and \p last, i.e. the
+ *  number of times that \p first must be incremented until it is equal to
+ *  \p last.
+ *
+ *  \param first The beginning of an input range of interest.
+ *  \param last The end of an input range of interest.
+ *  \return The distance between the beginning and end of the input range.
+ *
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>.
+ *
+ *  \pre If \c InputIterator meets the requirements of random access iterator, \p last shall be reachable from \p first or
+ *       \p first shall be reachable from \p last; otherwise, \p last shall be reachable from \p first.
+ *
+ *  The following code snippet demonstrates how to use \p distance to compute
+ *  the distance to one iterator from another.
+ *
+ *  \code
+ *  #include <thrust/distance.h>
+ *  #include <thrust/device_vector.h>
+ *  ...
+ *  thrust::device_vector<int> vec(13);
+ *  thrust::device_vector<int>::iterator iter1 = vec.begin();
+ *  thrust::device_vector<int>::iterator iter2 = iter1 + 7;
+ *
+ *  int d = thrust::distance(iter1, iter2);
+ *
+ *  // d is 7
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/distance.html
+ */
+template<typename InputIterator>
+  inline typename thrust::iterator_traits<InputIterator>::difference_type
+    distance(InputIterator first, InputIterator last);
+
+/*! \} // end iterators
+ */
+
+} // end thrust
+
+#include <thrust/detail/distance.inl>
+
diff --git a/compat/thrust/equal.h b/compat/thrust/equal.h
new file mode 100644
index 0000000000..e96946fcf7
--- /dev/null
+++ b/compat/thrust/equal.h
@@ -0,0 +1,236 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file equal.h
+ *  \brief Equality between ranges
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/execution_policy.h>
+
+namespace thrust
+{
+
+
+/*! \addtogroup reductions
+ *  \{
+ *  \addtogroup comparisons
+ *  \ingroup reductions
+ *  \{
+ */
+
+
+/*! \p equal returns \c true if the two ranges <tt>[first1, last1)</tt>
+ *  and <tt>[first2, first2 + (last1 - first1))</tt> are identical when
+ *  compared element-by-element, and otherwise returns \c false.
+ *
+ *  This version of \p equal returns \c true if and only if for every
+ *  iterator \c i in <tt>[first1, last1)</tt>, <tt>*i == *(first2 + (i - first1))</tt>.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first1 The beginning of the first sequence.
+ *  \param last1  The end of the first sequence.
+ *  \param first2 The beginning of the second sequence.
+ *  \return \c true, if the sequences are equal; \c false, otherwise.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator1's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/EqualityComparable.html">Equality Comparable</a>,
+ *          and \p InputIterator1's \c value_type can be compared for equality with \c InputIterator2's \c value_type.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator2's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/EqualityComparable.html">Equality Comparable</a>,
+ *          and \p InputIterator2's \c value_type can be compared for equality with \c InputIterator1's \c value_type.
+ *
+ *  The following code snippet demonstrates how to use \p equal to test
+ *  two ranges for equality using the \p thrust::host execution policy:
+ *
+ *  \code
+ *  #include <thrust/equal.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  int A1[7] = {3, 1, 4, 1, 5, 9, 3};
+ *  int A2[7] = {3, 1, 4, 2, 8, 5, 7};
+ *  ...
+ *  bool result = thrust::equal(thrust::host, A1, A1 + 7, A2);
+ *
+ *  // result == false
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/equal.html
+ */
+template<typename DerivedPolicy, typename InputIterator1, typename InputIterator2>
+bool equal(const thrust::detail::execution_policy_base<DerivedPolicy> &exec, InputIterator1 first1, InputIterator1 last1, InputIterator2 first2);
+
+
+/*! \p equal returns \c true if the two ranges <tt>[first1, last1)</tt>
+ *  and <tt>[first2, first2 + (last1 - first1))</tt> are identical when
+ *  compared element-by-element, and otherwise returns \c false.
+ *
+ *  This version of \p equal returns \c true if and only if for every
+ *  iterator \c i in <tt>[first1, last1)</tt>, <tt>*i == *(first2 + (i - first1))</tt>.
+ *
+ *  \param first1 The beginning of the first sequence.
+ *  \param last1  The end of the first sequence.
+ *  \param first2 The beginning of the second sequence.
+ *  \return \c true, if the sequences are equal; \c false, otherwise.
+ *
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator1's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/EqualityComparable.html">Equality Comparable</a>,
+ *          and \p InputIterator1's \c value_type can be compared for equality with \c InputIterator2's \c value_type.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator2's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/EqualityComparable.html">Equality Comparable</a>,
+ *          and \p InputIterator2's \c value_type can be compared for equality with \c InputIterator1's \c value_type.
+ *
+ *  The following code snippet demonstrates how to use \p equal to test
+ *  two ranges for equality.
+ *
+ *  \code
+ *  #include <thrust/equal.h>
+ *  ...
+ *  int A1[7] = {3, 1, 4, 1, 5, 9, 3};
+ *  int A2[7] = {3, 1, 4, 2, 8, 5, 7};
+ *  ...
+ *  bool result = thrust::equal(A1, A1 + 7, A2);
+ *
+ *  // result == false
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/equal.html
+ */
+template <typename InputIterator1, typename InputIterator2>
+bool equal(InputIterator1 first1, InputIterator1 last1,
+           InputIterator2 first2);
+
+
+/*! \p equal returns \c true if the two ranges <tt>[first1, last1)</tt>
+ *  and <tt>[first2, first2 + (last1 - first1))</tt> are identical when
+ *  compared element-by-element, and otherwise returns \c false.
+ *
+ *  This version of \p equal returns \c true if and only if for every
+ *  iterator \c i in <tt>[first1, last1)</tt>,
+ *  <tt>binary_pred(*i, *(first2 + (i - first1)))</tt> is \c true.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first1 The beginning of the first sequence.
+ *  \param last1  The end of the first sequence.
+ *  \param first2 The beginning of the second sequence.
+ *  \param binary_pred Binary predicate used to test element equality.
+ *  \return \c true, if the sequences are equal; \c false, otherwise.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator1's \c value_type is convertible to \p BinaryPredicate's \c first_argument_type.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator2's \c value_type is convertible to \p BinaryPredicate's \c second_argument_type.
+ *  \tparam BinaryPredicate is a model of <a href="http://www.sgi.com/tech/stl/BinaryPredicate.html">Binary Predicate</a>.
+ *
+ *  The following code snippet demonstrates how to use \p equal to compare the
+ *  elements in two ranges modulo 2 using the \p thrust::host execution policy.
+ *
+ *  \code
+ *  #include <thrust/equal.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  
+ *  __host__ __device__
+ *  struct compare_modulo_two
+ *  {
+ *    bool operator()(int x, int y)
+ *    {
+ *      return (x % 2) == (y % 2);
+ *    }
+ *  };
+ *  ...
+ *  int x[5] = {0, 2, 4, 6, 8, 10};
+ *  int y[5] = {1, 3, 5, 7, 9, 11};
+ *
+ *  bool result = thrust::equal(x, x + 5, y, compare_modulo_two());
+ *
+ *  // result is true
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/equal.html
+ */
+template<typename DerivedPolicy, typename InputIterator1, typename InputIterator2, typename BinaryPredicate>
+bool equal(const thrust::detail::execution_policy_base<DerivedPolicy> &exec, InputIterator1 first1, InputIterator1 last1, InputIterator2 first2, BinaryPredicate binary_pred);
+
+
+/*! \p equal returns \c true if the two ranges <tt>[first1, last1)</tt>
+ *  and <tt>[first2, first2 + (last1 - first1))</tt> are identical when
+ *  compared element-by-element, and otherwise returns \c false.
+ *
+ *  This version of \p equal returns \c true if and only if for every
+ *  iterator \c i in <tt>[first1, last1)</tt>,
+ *  <tt>binary_pred(*i, *(first2 + (i - first1)))</tt> is \c true.
+ *
+ *  \param first1 The beginning of the first sequence.
+ *  \param last1  The end of the first sequence.
+ *  \param first2 The beginning of the second sequence.
+ *  \param binary_pred Binary predicate used to test element equality.
+ *  \return \c true, if the sequences are equal; \c false, otherwise.
+ *
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator1's \c value_type is convertible to \p BinaryPredicate's \c first_argument_type.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator2's \c value_type is convertible to \p BinaryPredicate's \c second_argument_type.
+ *  \tparam BinaryPredicate is a model of <a href="http://www.sgi.com/tech/stl/BinaryPredicate.html">Binary Predicate</a>.
+ *
+ *  The following code snippet demonstrates how to use \p equal to compare the
+ *  elements in two ranges modulo 2.
+ *
+ *  \code
+ *  #include <thrust/equal.h>
+ *  
+ *  __host__ __device__
+ *  struct compare_modulo_two
+ *  {
+ *    bool operator()(int x, int y)
+ *    {
+ *      return (x % 2) == (y % 2);
+ *    }
+ *  };
+ *  ...
+ *  int x[5] = {0, 2, 4, 6, 8, 10};
+ *  int y[5] = {1, 3, 5, 7, 9, 11};
+ *
+ *  bool result = thrust::equal(x, x + 5, y, compare_modulo_two());
+ *
+ *  // result is true
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/equal.html
+ */
+template <typename InputIterator1, typename InputIterator2, 
+          typename BinaryPredicate>
+bool equal(InputIterator1 first1, InputIterator1 last1,
+           InputIterator2 first2, BinaryPredicate binary_pred);
+
+
+/*! \} // end comparisons
+ *  \} // end reductions
+ */
+
+} // end namespace thrust
+
+#include <thrust/detail/equal.inl>
+
diff --git a/compat/thrust/execution_policy.h b/compat/thrust/execution_policy.h
new file mode 100644
index 0000000000..a5b61e95b4
--- /dev/null
+++ b/compat/thrust/execution_policy.h
@@ -0,0 +1,351 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+/*! \file thrust/execution_policy.h
+ *  \brief Thrust execution policies.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// get the definition of thrust::execution_policy
+#include <thrust/detail/execution_policy.h>
+
+// #include the host system's execution_policy header
+#define __THRUST_HOST_SYSTEM_EXECUTION_POLICY_HEADER <__THRUST_HOST_SYSTEM_ROOT/execution_policy.h>
+#include __THRUST_HOST_SYSTEM_EXECUTION_POLICY_HEADER
+#undef __THRUST_HOST_SYSTEM_EXECUTION_POLICY_HEADER
+
+// #include the device system's execution_policy.h header
+#define __THRUST_DEVICE_SYSTEM_EXECUTION_POLICY_HEADER <__THRUST_DEVICE_SYSTEM_ROOT/execution_policy.h>
+#include __THRUST_DEVICE_SYSTEM_EXECUTION_POLICY_HEADER
+#undef __THRUST_DEVICE_SYSTEM_EXECUTION_POLICY_HEADER
+
+namespace thrust
+{
+
+
+/*! \cond
+ */
+
+
+namespace detail
+{
+
+
+typedef thrust::system::__THRUST_HOST_SYSTEM_NAMESPACE::detail::par_t host_t;
+
+
+typedef thrust::system::__THRUST_DEVICE_SYSTEM_NAMESPACE::detail::par_t device_t;
+
+
+} // end detail
+
+
+/*! \endcond
+ */
+
+
+/*! \addtogroup execution_policies Parallel Execution Policies
+ *  \{
+ */
+
+
+// define execution_policy for the purpose of Doxygenating it
+// it is actually defined elsewhere
+#if 0
+/*! \p execution_policy is the base class for all Thrust parallel execution policies
+ *  like \p thrust::host, \p thrust::device, and each backend system's tag type.
+ *
+ *  Custom user-defined backends should derive a policy from this type in order to
+ *  interoperate with Thrust algorithm dispatch.
+ *
+ *  The following code snippet demonstrates how to derive a standalone custom execution policy
+ *  from \p thrust::execution_policy to implement a backend which only implements \p for_each:
+ *
+ *  \code
+ *  #include <thrust/execution_policy.h>
+ *  #include <iostream>
+ *
+ *  // define a type derived from thrust::execution_policy to distinguish our custom execution policy:
+ *  struct my_policy : thrust::execution_policy<my_policy> {};
+ *
+ *  // overload for_each on my_policy
+ *  template<typename Iterator, typename Function>
+ *  Iterator for_each(my_policy, Iterator first, Iterator last, Function f)
+ *  {
+ *    std::cout << "Hello, world from for_each(my_policy)!" << std::endl;
+ *
+ *    for(; first < last; ++first)
+ *    {
+ *      f(*first);
+ *    }
+ *
+ *    return first;
+ *  }
+ *
+ *  struct ignore_argument
+ *  {
+ *    void operator()(int) {}
+ *  };
+ *
+ *  int main()
+ *  {
+ *    int data[4];
+ *
+ *    // dispatch thrust::for_each using our custom policy:
+ *    my_policy exec;
+ *    thrust::for_each(exec, data, data + 4, ignore_argument());
+ *
+ *    // can't dispatch thrust::transform because no overload exists for my_policy:
+ *    //thrust::transform(exec, data, data, + 4, data, thrust::identity<int>()); // error!
+ *
+ *    return 0;
+ *  }
+ *  \endcode
+ *
+ *  \see host_execution_policy
+ *  \see device_execution_policy
+ */
+template<typename DerivedPolicy>
+struct execution_policy : thrust::detail::execution_policy_base<DerivedPolicy>
+{};
+#endif
+
+
+/*! \p host_execution_policy is the base class for all Thrust parallel execution policies
+ *  which are derived from Thrust's default host backend system configured with the \p THRUST_HOST_SYSTEM
+ *  macro.
+ *
+ *  Custom user-defined backends which wish to inherit the functionality of Thrust's host backend system
+ *  should derive a policy from this type in order to interoperate with Thrust algorithm dispatch.
+ *
+ *  The following code snippet demonstrates how to derive a standalone custom execution policy from
+ *  \p thrust::host_execution_policy to implement a backend which specializes \p for_each while inheriting
+ *  the behavior of every other algorithm from the host system:
+ *
+ *  \code
+ *  #include <thrust/execution_policy.h>
+ *  #include <iostream>
+ *
+ *  // define a type derived from thrust::host_execution_policy to distinguish our custom execution policy:
+ *  struct my_policy : thrust::host_execution_policy<my_policy> {};
+ *
+ *  // overload for_each on my_policy
+ *  template<typename Iterator, typename Function>
+ *  Iterator for_each(my_policy, Iterator first, Iterator last, Function f)
+ *  {
+ *    std::cout << "Hello, world from for_each(my_policy)!" << std::endl;
+ *
+ *    for(; first < last; ++first)
+ *    {
+ *      f(*first);
+ *    }
+ *
+ *    return first;
+ *  }
+ *
+ *  struct ignore_argument
+ *  {
+ *    void operator()(int) {}
+ *  };
+ *
+ *  int main()
+ *  {
+ *    int data[4];
+ *
+ *    // dispatch thrust::for_each using our custom policy:
+ *    my_policy exec;
+ *    thrust::for_each(exec, data, data + 4, ignore_argument());
+ *
+ *    // dispatch thrust::transform whose behavior our policy inherits
+ *    thrust::transform(exec, data, data, + 4, data, thrust::identity<int>());
+ *
+ *    return 0;
+ *  }
+ *  \endcode
+ *
+ *  \see execution_policy
+ *  \see device_execution_policy
+ */
+template<typename DerivedPolicy>
+  struct host_execution_policy
+    : thrust::system::__THRUST_HOST_SYSTEM_NAMESPACE::execution_policy<DerivedPolicy>
+{};
+
+
+/*! \p device_execution_policy is the base class for all Thrust parallel execution policies
+ *  which are derived from Thrust's default device backend system configured with the \p THRUST_DEVICE_SYSTEM
+ *  macro.
+ *
+ *  Custom user-defined backends which wish to inherit the functionality of Thrust's device backend system
+ *  should derive a policy from this type in order to interoperate with Thrust algorithm dispatch.
+ *
+ *  The following code snippet demonstrates how to derive a standalone custom execution policy from
+ *  \p thrust::device_execution_policy to implement a backend which specializes \p for_each while inheriting
+ *  the behavior of every other algorithm from the device system:
+ *
+ *  \code
+ *  #include <thrust/execution_policy.h>
+ *  #include <iostream>
+ *
+ *  // define a type derived from thrust::device_execution_policy to distinguish our custom execution policy:
+ *  struct my_policy : thrust::device_execution_policy<my_policy> {};
+ *
+ *  // overload for_each on my_policy
+ *  template<typename Iterator, typename Function>
+ *  Iterator for_each(my_policy, Iterator first, Iterator last, Function f)
+ *  {
+ *    std::cout << "Hello, world from for_each(my_policy)!" << std::endl;
+ *
+ *    for(; first < last; ++first)
+ *    {
+ *      f(*first);
+ *    }
+ *
+ *    return first;
+ *  }
+ *
+ *  struct ignore_argument
+ *  {
+ *    void operator()(int) {}
+ *  };
+ *
+ *  int main()
+ *  {
+ *    int data[4];
+ *
+ *    // dispatch thrust::for_each using our custom policy:
+ *    my_policy exec;
+ *    thrust::for_each(exec, data, data + 4, ignore_argument());
+ *
+ *    // dispatch thrust::transform whose behavior our policy inherits
+ *    thrust::transform(exec, data, data, + 4, data, thrust::identity<int>());
+ *
+ *    return 0;
+ *  }
+ *  \endcode
+ *
+ *  \see execution_policy
+ *  \see host_execution_policy
+ */
+template<typename DerivedPolicy>
+  struct device_execution_policy
+    : thrust::system::__THRUST_DEVICE_SYSTEM_NAMESPACE::execution_policy<DerivedPolicy>
+{};
+
+
+/*! \p thrust::host is the default parallel execution policy associated with Thrust's host backend system
+ *  configured by the \p THRUST_HOST_SYSTEM macro.
+ *
+ *  Instead of relying on implicit algorithm dispatch through iterator system tags, users may directly target
+ *  algorithm dispatch at Thrust's host system by providing \p thrust::host as an algorithm parameter.
+ *
+ *  Explicit dispatch can be useful in avoiding the introduction of data copies into containers such as
+ *  \p thrust::host_vector.
+ *
+ *  Note that even though \p thrust::host targets the host CPU, it is a parallel execution policy. That is,
+ *  the order that an algorithm invokes functors or dereferences iterators is not defined.
+ *
+ *  The type of \p thrust::host is implementation-defined.
+ *
+ *  The following code snippet demonstrates how to use \p thrust::host to explicitly dispatch an invocation
+ *  of \p thrust::for_each to the host backend system:
+ *
+ *  \code
+ *  #include <thrust/for_each.h>
+ *  #include <thrust/execution_policy.h>
+ *  #include <cstdio>
+ *
+ *  struct printf_functor
+ *  {
+ *    __host__ __device__
+ *    void operator()(int x)
+ *    {
+ *      printf("%d\n");
+ *    }
+ *  };
+ *  ...
+ *  int vec[3];
+ *  vec[0] = 0; vec[1] = 1; vec[2] = 2;
+ *
+ *  thrust::for_each(thrust::host, vec.begin(), vec.end(), printf_functor());
+ *
+ *  // 0 1 2 is printed to standard output in some unspecified order
+ *  \endcode
+ *
+ *  \see host_execution_policy
+ *  \see thrust::device
+ */
+static const detail::host_t host;
+
+
+/*! \p thrust::device is the default parallel execution policy associated with Thrust's device backend system
+ *  configured by the \p THRUST_DEVICE_SYSTEM macro.
+ *
+ *  Instead of relying on implicit algorithm dispatch through iterator system tags, users may directly target
+ *  algorithm dispatch at Thrust's device system by providing \p thrust::device as an algorithm parameter.
+ *
+ *  Explicit dispatch can be useful in avoiding the introduction of data copies into containers such as
+ *  \p thrust::device_vector or to avoid wrapping e.g. raw pointers allocated by the CUDA API with types
+ *  such as \p thrust::device_ptr.
+ *
+ *  The user must take care to guarantee that the iterators provided to an algorithm are compatible with
+ *  the device backend system. For example, raw pointers allocated by <tt>std::malloc</tt> typically
+ *  cannot be dereferenced by a GPU. For this reason, raw pointers allocated by host APIs should not be mixed
+ *  with a \p thrust::device algorithm invocation when the device backend is CUDA.
+ *
+ *  The type of \p thrust::device is implementation-defined.
+ *
+ *  The following code snippet demonstrates how to use \p thrust::device to explicitly dispatch an invocation
+ *  of \p thrust::for_each to the device backend system:
+ *
+ *  \code
+ *  #include <thrust/for_each.h>
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/execution_policy.h>
+ *  #include <cstdio>
+ *
+ *  struct printf_functor
+ *  {
+ *    __host__ __device__
+ *    void operator()(int x)
+ *    {
+ *      printf("%d\n");
+ *    }
+ *  };
+ *  ...
+ *  thrust::device_vector<int> d_vec[3];
+ *  d_vec[0] = 0; d_vec[1] = 1; d_vec[2] = 2;
+ *
+ *  thrust::for_each(thrust::device, vec.begin(), vec.end(), printf_functor());
+ *
+ *  // 0 1 2 is printed to standard output in some unspecified order
+ *  \endcode
+ *
+ *  \see host_execution_policy
+ *  \see thrust::device
+ */
+static const detail::device_t device;
+
+
+/*! \}
+ */
+
+
+} // end thrust
+
diff --git a/compat/thrust/extrema.h b/compat/thrust/extrema.h
new file mode 100644
index 0000000000..335bcd1e6b
--- /dev/null
+++ b/compat/thrust/extrema.h
@@ -0,0 +1,798 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+/*! \file extrema.h
+ *  \brief Functions for computing computing extremal values
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/execution_policy.h>
+#include <thrust/pair.h>
+
+namespace thrust
+{
+
+
+/*! This version of \p min returns the smaller of two values, given a comparison operation.
+ *  \param lhs The first value to compare.
+ *  \param rhs The second value to compare.
+ *  \param comp A comparison operation.
+ *  \return The smaller element.
+ *
+ *  \tparam T is convertible to \p BinaryPredicate's first argument type and to its second argument type.
+ *  \tparam BinaryPredicate is a model of <a href="http://www.sgi.com/tech/stl/BinaryPredicate.html">BinaryPredicate</a>.
+ *
+ *  The following code snippet demonstrates how to use \p min to compute the smaller of two
+ *  key-value objects.
+ *
+ *  \code
+ *  #include <thrust/extrema.h>
+ *  ...
+ *  struct key_value
+ *  {
+ *    int key;
+ *    int value;
+ *  };
+ *
+ *  struct compare_key_value
+ *  {
+ *    __host__ __device__
+ *    bool operator()(key_value lhs, key_value rhs)
+ *    {
+ *      return lhs.key < rhs.key;
+ *    }
+ *  };
+ *
+ *  ...
+ *  key_value a = {13, 0};
+ *  key_value b = { 7, 1);
+ *
+ *  key_value smaller = thrust::min(a, b, compare_key_value());
+ *
+ *  // smaller is {7, 1}
+ *  \endcode
+ *
+ *  \note Returns the first argument when the arguments are equivalent.
+ *  \see max
+ */
+template<typename T, typename BinaryPredicate>
+__host__ __device__
+  T min THRUST_PREVENT_MACRO_SUBSTITUTION (const T &lhs, const T &rhs, BinaryPredicate comp);
+
+
+/*! This version of \p min returns the smaller of two values.
+ *  \param lhs The first value to compare.
+ *  \param rhs The second value to compare.
+ *  \return The smaller element.
+ *
+ *  \tparam T is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable.html">LessThan Comparable</a>.
+ *
+ *  The following code snippet demonstrates how to use \p min to compute the smaller of two
+ *  integers.
+ *
+ *  \code
+ *  #include <thrust/extrema.h>
+ *  ...
+ *  int a = 13;
+ *  int b = 7;
+ *
+ *  int smaller = thrust::min(a, b);
+ *
+ *  // smaller is 7
+ *  \endcode
+ *
+ *  \note Returns the first argument when the arguments are equivalent.
+ *  \see max
+ */
+template<typename T>
+__host__ __device__
+  T min THRUST_PREVENT_MACRO_SUBSTITUTION (const T &lhs, const T &rhs);
+
+
+/*! This version of \p max returns the larger of two values, given a comparison operation.
+ *  \param lhs The first value to compare.
+ *  \param rhs The second value to compare.
+ *  \param comp A comparison operation.
+ *  \return The larger element.
+ *
+ *  \tparam T is convertible to \p BinaryPredicate's first argument type and to its second argument type.
+ *  \tparam BinaryPredicate is a model of <a href="http://www.sgi.com/tech/stl/BinaryPredicate.html">BinaryPredicate</a>.
+ *
+ *  The following code snippet demonstrates how to use \p max to compute the larger of two
+ *  key-value objects.
+ *
+ *  \code
+ *  #include <thrust/extrema.h>
+ *  ...
+ *  struct key_value
+ *  {
+ *    int key;
+ *    int value;
+ *  };
+ *
+ *  struct compare_key_value
+ *  {
+ *    __host__ __device__
+ *    bool operator()(key_value lhs, key_value rhs)
+ *    {
+ *      return lhs.key < rhs.key;
+ *    }
+ *  };
+ *
+ *  ...
+ *  key_value a = {13, 0};
+ *  key_value b = { 7, 1);
+ *
+ *  key_value larger = thrust::max(a, b, compare_key_value());
+ *
+ *  // larger is {13, 0}
+ *  \endcode
+ *
+ *  \note Returns the first argument when the arguments are equivalent.
+ *  \see min
+ */
+template<typename T, typename BinaryPredicate>
+__host__ __device__
+  T max THRUST_PREVENT_MACRO_SUBSTITUTION (const T &lhs, const T &rhs, BinaryPredicate comp);
+
+
+/*! This version of \p max returns the larger of two values.
+ *  \param lhs The first value to compare.
+ *  \param rhs The second value to compare.
+ *  \return The larger element.
+ *
+ *  \tparam T is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable.html">LessThan Comparable</a>.
+ *
+ *  The following code snippet demonstrates how to use \p max to compute the larger of two
+ *  integers.
+ *
+ *  \code
+ *  #include <thrust/extrema.h>
+ *  ...
+ *  int a = 13;
+ *  int b = 7;
+ *
+ *  int larger = thrust::min(a, b);
+ *
+ *  // larger is 13
+ *  \endcode
+ *
+ *  \note Returns the first argument when the arguments are equivalent.
+ *  \see min
+ */
+template<typename T>
+__host__ __device__
+  T max THRUST_PREVENT_MACRO_SUBSTITUTION (const T &lhs, const T &rhs);
+
+
+/*! \addtogroup reductions
+ *  \{
+ *  \addtogroup extrema
+ *  \ingroup reductions
+ *  \{
+ */
+
+/*! \p min_element finds the smallest element in the range <tt>[first, last)</tt>.
+ *  It returns the first iterator \c i in <tt>[first, last)</tt>
+ *  such that no other iterator in <tt>[first, last)</tt> points to a value smaller
+ *  than \c *i. The return value is \p last if and only if <tt>[first, last)</tt> is an
+ *  empty range.
+ *
+ *  The two versions of \p min_element differ in how they define whether one element is
+ *  less than another. This version compares objects using \c operator<. Specifically,
+ *  this version of \p min_element returns the first iterator \c i in <tt>[first, last)</tt>
+ *  such that, for every iterator \c j in <tt>[first, last)</tt>, <tt>*j < *i</tt> is
+ *  \c false.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the sequence.
+ *  \param last  The end of the sequence.
+ *  \return An iterator pointing to the smallest element of the range <tt>[first, last)</tt>,
+ *          if it is not an empty range; \p last, otherwise.
+ *
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator">Forward Iterator</a>,
+ *          and \c ForwardIterator's \c value_type is a model of
+ *          <a href="http://www.sgi.com/tech/stl/LessThanComparable.html">LessThan Comparable</a>.
+ *
+ *  \code
+ *  #include <thrust/extrema.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  int data[6] = {1, 0, 2, 2, 1, 3};
+ *  int *result = thrust::min_element(thrust::host, data, data + 6);
+ *
+ *  // result is data + 1
+ *  // *result is 0
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/min_element.html 
+ */
+template<typename DerivedPolicy, typename ForwardIterator>
+ForwardIterator min_element(const thrust::detail::execution_policy_base<DerivedPolicy> &exec, ForwardIterator first, ForwardIterator last);
+
+
+/*! \p min_element finds the smallest element in the range <tt>[first, last)</tt>.
+ *  It returns the first iterator \c i in <tt>[first, last)</tt>
+ *  such that no other iterator in <tt>[first, last)</tt> points to a value smaller
+ *  than \c *i. The return value is \p last if and only if <tt>[first, last)</tt> is an
+ *  empty range.
+ *
+ *  The two versions of \p min_element differ in how they define whether one element is
+ *  less than another. This version compares objects using \c operator<. Specifically,
+ *  this version of \p min_element returns the first iterator \c i in <tt>[first, last)</tt>
+ *  such that, for every iterator \c j in <tt>[first, last)</tt>, <tt>*j < *i</tt> is
+ *  \c false.
+ *
+ *  \param first The beginning of the sequence.
+ *  \param last  The end of the sequence.
+ *  \return An iterator pointing to the smallest element of the range <tt>[first, last)</tt>,
+ *          if it is not an empty range; \p last, otherwise.
+ *
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator">Forward Iterator</a>,
+ *          and \c ForwardIterator's \c value_type is a model of
+ *          <a href="http://www.sgi.com/tech/stl/LessThanComparable.html">LessThan Comparable</a>.
+ *
+ *  \code
+ *  #include <thrust/extrema.h>
+ *  ...
+ *  int data[6] = {1, 0, 2, 2, 1, 3};
+ *  int *result = thrust::min_element(data, data + 6);
+ *
+ *  // result is data + 1
+ *  // *result is 0
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/min_element.html 
+ */
+template <typename ForwardIterator>
+ForwardIterator min_element(ForwardIterator first, ForwardIterator last);
+
+
+/*! \p min_element finds the smallest element in the range <tt>[first, last)</tt>.
+ *  It returns the first iterator \c i in <tt>[first, last)</tt>
+ *  such that no other iterator in <tt>[first, last)</tt> points to a value smaller
+ *  than \c *i. The return value is \p last if and only if <tt>[first, last)</tt> is an
+ *  empty range.
+ *
+ *  The two versions of \p min_element differ in how they define whether one element is
+ *  less than another. This version compares objects using a function object \p comp.
+ *  Specifically, this version of \p min_element returns the first iterator \c i in <tt>[first, last)</tt>
+ *  such that, for every iterator \c j in <tt>[first, last)</tt>, <tt>comp(*j, *i)</tt> is
+ *  \c false.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the sequence.
+ *  \param last  The end of the sequence.
+ *  \param comp A binary predicate used for comparison.
+ *  \return An iterator pointing to the smallest element of the range <tt>[first, last)</tt>,
+ *          if it is not an empty range; \p last, otherwise.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator">Forward Iterator</a>,
+ *          and \p ForwardIterator's \c value_type is convertible to both \p comp's
+ *          \c first_argument_type and \c second_argument_type.
+ *  \tparam BinaryPredicate is a model of <a href="http://www.sgi.com/tech/stl/BinaryPredicate">Binary Predicate</a>.
+ *
+ *  The following code snippet demonstrates how to use \p min_element to find the smallest element
+ *  of a collection of key-value pairs using the \p thrust::host execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/extrema.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *
+ *  struct key_value
+ *  {
+ *    int key;
+ *    int value;
+ *  };
+ *
+ *  struct compare_key_value
+ *  {
+ *    __host__ __device__
+ *    bool operator()(key_value lhs, key_value rhs)
+ *    {
+ *      return lhs.key < rhs.key;
+ *    }
+ *  };
+ *
+ *  ...
+ *  key_value data[4] = { {4,5}, {0,7}, {2,3}, {6,1} };
+ *
+ *  key_value *smallest = thrust::min_element(thrust::host, data, data + 4, compare_key_value());
+ *
+ *  // smallest == data + 1
+ *  // *smallest == {0,7}
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/min_element.html 
+ */
+template<typename DerivedPolicy, typename ForwardIterator, typename BinaryPredicate>
+ForwardIterator min_element(const thrust::detail::execution_policy_base<DerivedPolicy> &exec, ForwardIterator first, ForwardIterator last, BinaryPredicate comp);
+
+
+/*! \p min_element finds the smallest element in the range <tt>[first, last)</tt>.
+ *  It returns the first iterator \c i in <tt>[first, last)</tt>
+ *  such that no other iterator in <tt>[first, last)</tt> points to a value smaller
+ *  than \c *i. The return value is \p last if and only if <tt>[first, last)</tt> is an
+ *  empty range.
+ *
+ *  The two versions of \p min_element differ in how they define whether one element is
+ *  less than another. This version compares objects using a function object \p comp.
+ *  Specifically, this version of \p min_element returns the first iterator \c i in <tt>[first, last)</tt>
+ *  such that, for every iterator \c j in <tt>[first, last)</tt>, <tt>comp(*j, *i)</tt> is
+ *  \c false.
+ *
+ *  \param first The beginning of the sequence.
+ *  \param last  The end of the sequence.
+ *  \param comp A binary predicate used for comparison.
+ *  \return An iterator pointing to the smallest element of the range <tt>[first, last)</tt>,
+ *          if it is not an empty range; \p last, otherwise.
+ *
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator">Forward Iterator</a>,
+ *          and \p ForwardIterator's \c value_type is convertible to both \p comp's
+ *          \c first_argument_type and \c second_argument_type.
+ *  \tparam BinaryPredicate is a model of <a href="http://www.sgi.com/tech/stl/BinaryPredicate">Binary Predicate</a>.
+ *
+ *  The following code snippet demonstrates how to use \p min_element to find the smallest element
+ *  of a collection of key-value pairs.
+ *
+ *  \code
+ *  #include <thrust/extrema.h>
+ *
+ *  struct key_value
+ *  {
+ *    int key;
+ *    int value;
+ *  };
+ *
+ *  struct compare_key_value
+ *  {
+ *    __host__ __device__
+ *    bool operator()(key_value lhs, key_value rhs)
+ *    {
+ *      return lhs.key < rhs.key;
+ *    }
+ *  };
+ *
+ *  ...
+ *  key_value data[4] = { {4,5}, {0,7}, {2,3}, {6,1} };
+ *
+ *  key_value *smallest = thrust::min_element(data, data + 4, compare_key_value());
+ *
+ *  // smallest == data + 1
+ *  // *smallest == {0,7}
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/min_element.html 
+ */
+template <typename ForwardIterator, typename BinaryPredicate>
+ForwardIterator min_element(ForwardIterator first, ForwardIterator last,
+                            BinaryPredicate comp);
+
+
+/*! \p max_element finds the largest element in the range <tt>[first, last)</tt>.
+ *  It returns the first iterator \c i in <tt>[first, last)</tt>
+ *  such that no other iterator in <tt>[first, last)</tt> points to a value larger
+ *  than \c *i. The return value is \p last if and only if <tt>[first, last)</tt> is an
+ *  empty range.
+ *
+ *  The two versions of \p max_element differ in how they define whether one element is
+ *  greater than another. This version compares objects using \c operator<. Specifically,
+ *  this version of \p max_element returns the first iterator \c i in <tt>[first, last)</tt>
+ *  such that, for every iterator \c j in <tt>[first, last)</tt>, <tt>*i < *j</tt> is
+ *  \c false.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the sequence.
+ *  \param last  The end of the sequence.
+ *  \return An iterator pointing to the largest element of the range <tt>[first, last)</tt>,
+ *          if it is not an empty range; \p last, otherwise.
+ *
+ *  \tparam A Thrust backend system.
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>,
+ *          and \c ForwardIterator's \c value_type is a model of
+ *          <a href="http://www.sgi.com/tech/stl/LessThanComparable.html">LessThan Comparable</a>.
+ *
+ *  \code
+ *  #include <thrust/extrema.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  int data[6] = {1, 0, 2, 2, 1, 3};
+ *  int *result = thrust::max_element(thrust::host, data, data + 6);
+ *
+ *  // *result == 3
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/max_element.html 
+ */
+template<typename DerivedPolicy, typename ForwardIterator>
+ForwardIterator max_element(const thrust::detail::execution_policy_base<DerivedPolicy> &exec, ForwardIterator first, ForwardIterator last);
+
+
+/*! \p max_element finds the largest element in the range <tt>[first, last)</tt>.
+ *  It returns the first iterator \c i in <tt>[first, last)</tt>
+ *  such that no other iterator in <tt>[first, last)</tt> points to a value larger
+ *  than \c *i. The return value is \p last if and only if <tt>[first, last)</tt> is an
+ *  empty range.
+ *
+ *  The two versions of \p max_element differ in how they define whether one element is
+ *  greater than another. This version compares objects using \c operator<. Specifically,
+ *  this version of \p max_element returns the first iterator \c i in <tt>[first, last)</tt>
+ *  such that, for every iterator \c j in <tt>[first, last)</tt>, <tt>*i < *j</tt> is
+ *  \c false.
+ *
+ *  \param first The beginning of the sequence.
+ *  \param last  The end of the sequence.
+ *  \return An iterator pointing to the largest element of the range <tt>[first, last)</tt>,
+ *          if it is not an empty range; \p last, otherwise.
+ *
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>,
+ *          and \c ForwardIterator's \c value_type is a model of
+ *          <a href="http://www.sgi.com/tech/stl/LessThanComparable.html">LessThan Comparable</a>.
+ *
+ *  \code
+ *  #include <thrust/extrema.h>
+ *  ...
+ *  int data[6] = {1, 0, 2, 2, 1, 3};
+ *  int *result = thrust::max_element(data, data + 6);
+ *
+ *  // *result == 3
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/max_element.html 
+ */
+template <typename ForwardIterator>
+ForwardIterator max_element(ForwardIterator first, ForwardIterator last);
+
+
+/*! \p max_element finds the largest element in the range <tt>[first, last)</tt>.
+ *  It returns the first iterator \c i in <tt>[first, last)</tt>
+ *  such that no other iterator in <tt>[first, last)</tt> points to a value larger
+ *  than \c *i. The return value is \p last if and only if <tt>[first, last)</tt> is an
+ *  empty range.
+ *
+ *  The two versions of \p max_element differ in how they define whether one element is
+ *  less than another. This version compares objects using a function object \p comp.
+ *  Specifically, this version of \p max_element returns the first iterator \c i in <tt>[first, last)</tt>
+ *  such that, for every iterator \c j in <tt>[first, last)</tt>, <tt>comp(*i, *j)</tt> is
+ *  \c false.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the sequence.
+ *  \param last  The end of the sequence.
+ *  \param comp A binary predicate used for comparison.
+ *  \return An iterator pointing to the largest element of the range <tt>[first, last)</tt>,
+ *          if it is not an empty range; \p last, otherwise.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>,
+ *          and \p ForwardIterator's \c value_type is convertible to both \p comp's
+ *          \c first_argument_type and \c second_argument_type.
+ *  \tparam BinaryPredicate is a model of <a href="http://www.sgi.com/tech/stl/BinaryPredicate.html">Binary Predicate</a>.
+ *
+ *  The following code snippet demonstrates how to use \p max_element to find the largest element
+ *  of a collection of key-value pairs using the \p thrust::host execution policy for parallelization.
+ *
+ *  \code
+ *  #include <thrust/extrema.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *
+ *  struct key_value
+ *  {
+ *    int key;
+ *    int value;
+ *  };
+ *
+ *  struct compare_key_value
+ *  {
+ *    __host__ __device__
+ *    bool operator()(key_value lhs, key_value rhs)
+ *    {
+ *      return lhs.key < rhs.key;
+ *    }
+ *  };
+ *
+ *  ...
+ *  key_value data[4] = { {4,5}, {0,7}, {2,3}, {6,1} };
+ *
+ *  key_value *largest = thrust::max_element(thrust::host, data, data + 4, compare_key_value());
+ *
+ *  // largest == data + 3
+ *  // *largest == {6,1}
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/max_element.html 
+ */
+template<typename DerivedPolicy, typename ForwardIterator, typename BinaryPredicate>
+ForwardIterator max_element(const thrust::detail::execution_policy_base<DerivedPolicy> &exec, ForwardIterator first, ForwardIterator last, BinaryPredicate comp);
+
+
+/*! \p max_element finds the largest element in the range <tt>[first, last)</tt>.
+ *  It returns the first iterator \c i in <tt>[first, last)</tt>
+ *  such that no other iterator in <tt>[first, last)</tt> points to a value larger
+ *  than \c *i. The return value is \p last if and only if <tt>[first, last)</tt> is an
+ *  empty range.
+ *
+ *  The two versions of \p max_element differ in how they define whether one element is
+ *  less than another. This version compares objects using a function object \p comp.
+ *  Specifically, this version of \p max_element returns the first iterator \c i in <tt>[first, last)</tt>
+ *  such that, for every iterator \c j in <tt>[first, last)</tt>, <tt>comp(*i, *j)</tt> is
+ *  \c false.
+ *
+ *  \param first The beginning of the sequence.
+ *  \param last  The end of the sequence.
+ *  \param comp A binary predicate used for comparison.
+ *  \return An iterator pointing to the largest element of the range <tt>[first, last)</tt>,
+ *          if it is not an empty range; \p last, otherwise.
+ *
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>,
+ *          and \p ForwardIterator's \c value_type is convertible to both \p comp's
+ *          \c first_argument_type and \c second_argument_type.
+ *  \tparam BinaryPredicate is a model of <a href="http://www.sgi.com/tech/stl/BinaryPredicate.html">Binary Predicate</a>.
+ *
+ *  The following code snippet demonstrates how to use \p max_element to find the largest element
+ *  of a collection of key-value pairs.
+ *
+ *  \code
+ *  #include <thrust/extrema.h>
+ *
+ *  struct key_value
+ *  {
+ *    int key;
+ *    int value;
+ *  };
+ *
+ *  struct compare_key_value
+ *  {
+ *    __host__ __device__
+ *    bool operator()(key_value lhs, key_value rhs)
+ *    {
+ *      return lhs.key < rhs.key;
+ *    }
+ *  };
+ *
+ *  ...
+ *  key_value data[4] = { {4,5}, {0,7}, {2,3}, {6,1} };
+ *
+ *  key_value *largest = thrust::max_element(data, data + 4, compare_key_value());
+ *
+ *  // largest == data + 3
+ *  // *largest == {6,1}
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/max_element.html 
+ */
+template <typename ForwardIterator, typename BinaryPredicate>
+ForwardIterator max_element(ForwardIterator first, ForwardIterator last,
+                            BinaryPredicate comp);
+
+
+/*! \p minmax_element finds the smallest and largest elements in the range <tt>[first, last)</tt>.
+ *  It returns a pair of iterators <tt>(imin, imax)</tt> where \c imin is the same iterator
+ *  returned by \p min_element and \c imax is the same iterator returned by \p max_element.
+ *  This function is potentially more efficient than separate calls to \p min_element and \p max_element.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the sequence.
+ *  \param last  The end of the sequence.
+ *  \return A pair of iterator pointing to the smallest and largest elements of the range <tt>[first, last)</tt>,
+ *          if it is not an empty range; \p last, otherwise.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>,
+ *          and \c ForwardIterator's \c value_type is a model of
+ *          <a href="http://www.sgi.com/tech/stl/LessThanComparable.html">LessThan Comparable</a>.
+ *
+ *  \code
+ *  #include <thrust/extrema.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  int data[6] = {1, 0, 2, 2, 1, 3};
+ *  thrust::pair<int *, int *> result = thrust::minmax_element(thrust::host, data, data + 6);
+ *
+ *  // result.first is data + 1
+ *  // result.second is data + 5
+ *  // *result.first is 0
+ *  // *result.second is 3
+ *  \endcode
+ *
+ *  \see min_element
+ *  \see max_element
+ *  \see http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2005/n1840.pdf
+ */
+template<typename DerivedPolicy, typename ForwardIterator>
+thrust::pair<ForwardIterator,ForwardIterator> minmax_element(const thrust::detail::execution_policy_base<DerivedPolicy> &exec, ForwardIterator first, ForwardIterator last);
+
+
+/*! \p minmax_element finds the smallest and largest elements in the range <tt>[first, last)</tt>.
+ *  It returns a pair of iterators <tt>(imin, imax)</tt> where \c imin is the same iterator
+ *  returned by \p min_element and \c imax is the same iterator returned by \p max_element.
+ *  This function is potentially more efficient than separate calls to \p min_element and \p max_element.
+ *
+ *  \param first The beginning of the sequence.
+ *  \param last  The end of the sequence.
+ *  \return A pair of iterator pointing to the smallest and largest elements of the range <tt>[first, last)</tt>,
+ *          if it is not an empty range; \p last, otherwise.
+ *
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>,
+ *          and \c ForwardIterator's \c value_type is a model of
+ *          <a href="http://www.sgi.com/tech/stl/LessThanComparable.html">LessThan Comparable</a>.
+ *
+ *  \code
+ *  #include <thrust/extrema.h>
+ *  ...
+ *  int data[6] = {1, 0, 2, 2, 1, 3};
+ *  thrust::pair<int *, int *> result = thrust::minmax_element(data, data + 6);
+ *
+ *  // result.first is data + 1
+ *  // result.second is data + 5
+ *  // *result.first is 0
+ *  // *result.second is 3
+ *  \endcode
+ *
+ *  \see min_element
+ *  \see max_element
+ *  \see http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2005/n1840.pdf
+ */
+template <typename ForwardIterator>
+thrust::pair<ForwardIterator,ForwardIterator> minmax_element(ForwardIterator first, 
+                                                             ForwardIterator last);
+
+
+/*! \p minmax_element finds the smallest and largest elements in the range <tt>[first, last)</tt>.
+ *  It returns a pair of iterators <tt>(imin, imax)</tt> where \c imin is the same iterator
+ *  returned by \p min_element and \c imax is the same iterator returned by \p max_element.
+ *  This function is potentially more efficient than separate calls to \p min_element and \p max_element.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the sequence.
+ *  \param last  The end of the sequence.
+ *  \param comp A binary predicate used for comparison.
+ *  \return A pair of iterator pointing to the smallest and largest elements of the range <tt>[first, last)</tt>,
+ *          if it is not an empty range; \p last, otherwise.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator">Forward Iterator</a>,
+ *          and \p ForwardIterator's \c value_type is convertible to both \p comp's
+ *          \c first_argument_type and \c second_argument_type.
+ *  \tparam BinaryPredicate is a model of <a href="http://www.sgi.com/tech/stl/BinaryPredicate">Binary Predicate</a>.
+ *
+ *  The following code snippet demonstrates how to use \p minmax_element to find the smallest and largest elements
+ *  of a collection of key-value pairs using the \p thrust::host execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/extrema.h>
+ *  #include <thrust/pair.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *
+ *  struct key_value
+ *  {
+ *    int key;
+ *    int value;
+ *  };
+ *
+ *  struct compare_key_value
+ *  {
+ *    __host__ __device__
+ *    bool operator()(key_value lhs, key_value rhs)
+ *    {
+ *      return lhs.key < rhs.key;
+ *    }
+ *  };
+ *
+ *  ...
+ *  key_value data[4] = { {4,5}, {0,7}, {2,3}, {6,1} };
+ *
+ *  thrust::pair<key_value*,key_value*> extrema = thrust::minmax_element(thrust::host, data, data + 4, compare_key_value());
+ *
+ *  // extrema.first   == data + 1
+ *  // *extrema.first  == {0,7}
+ *  // extrema.second  == data + 3
+ *  // *extrema.second == {6,1}
+ *  \endcode
+ *
+ *  \see min_element
+ *  \see max_element
+ *  \see http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2005/n1840.pdf
+ */
+template<typename DerivedPolicy, typename ForwardIterator, typename BinaryPredicate>
+thrust::pair<ForwardIterator,ForwardIterator> minmax_element(const thrust::detail::execution_policy_base<DerivedPolicy> &exec, ForwardIterator first, ForwardIterator last, BinaryPredicate comp);
+
+
+/*! \p minmax_element finds the smallest and largest elements in the range <tt>[first, last)</tt>.
+ *  It returns a pair of iterators <tt>(imin, imax)</tt> where \c imin is the same iterator
+ *  returned by \p min_element and \c imax is the same iterator returned by \p max_element.
+ *  This function is potentially more efficient than separate calls to \p min_element and \p max_element.
+ *
+ *  \param first The beginning of the sequence.
+ *  \param last  The end of the sequence.
+ *  \param comp A binary predicate used for comparison.
+ *  \return A pair of iterator pointing to the smallest and largest elements of the range <tt>[first, last)</tt>,
+ *          if it is not an empty range; \p last, otherwise.
+ *
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator">Forward Iterator</a>,
+ *          and \p ForwardIterator's \c value_type is convertible to both \p comp's
+ *          \c first_argument_type and \c second_argument_type.
+ *  \tparam BinaryPredicate is a model of <a href="http://www.sgi.com/tech/stl/BinaryPredicate">Binary Predicate</a>.
+ *
+ *  The following code snippet demonstrates how to use \p minmax_element to find the smallest and largest elements
+ *  of a collection of key-value pairs.
+ *
+ *  \code
+ *  #include <thrust/extrema.h>
+ *  #include <thrust/pair.h>
+ *
+ *  struct key_value
+ *  {
+ *    int key;
+ *    int value;
+ *  };
+ *
+ *  struct compare_key_value
+ *  {
+ *    __host__ __device__
+ *    bool operator()(key_value lhs, key_value rhs)
+ *    {
+ *      return lhs.key < rhs.key;
+ *    }
+ *  };
+ *
+ *  ...
+ *  key_value data[4] = { {4,5}, {0,7}, {2,3}, {6,1} };
+ *
+ *  thrust::pair<key_value*,key_value*> extrema = thrust::minmax_element(data, data + 4, compare_key_value());
+ *
+ *  // extrema.first   == data + 1
+ *  // *extrema.first  == {0,7}
+ *  // extrema.second  == data + 3
+ *  // *extrema.second == {6,1}
+ *  \endcode
+ *
+ *  \see min_element
+ *  \see max_element
+ *  \see http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2005/n1840.pdf
+ */
+template <typename ForwardIterator, typename BinaryPredicate>
+thrust::pair<ForwardIterator,ForwardIterator> minmax_element(ForwardIterator first, 
+                                                             ForwardIterator last,
+                                                             BinaryPredicate comp);
+
+/*! \} // end extrema
+ *  \} // end reductions
+ */
+
+} // end thrust
+
+#include <thrust/detail/extrema.inl>
+#include <thrust/detail/minmax.h>
+
diff --git a/compat/thrust/fill.h b/compat/thrust/fill.h
new file mode 100644
index 0000000000..b492cec9dd
--- /dev/null
+++ b/compat/thrust/fill.h
@@ -0,0 +1,205 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file fill.h
+ *  \brief Fills a range with a constant value
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/execution_policy.h>
+
+namespace thrust
+{
+
+
+/*! \addtogroup transformations
+ *  \addtogroup filling
+ *  \ingroup transformations
+ *  \{
+ */
+
+
+/*! \p fill assigns the value \p value to every element in
+ *  the range <tt>[first, last)</tt>. That is, for every
+ *  iterator \c i in <tt>[first, last)</tt>, it performs
+ *  the assignment <tt>*i = value</tt>.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the sequence.
+ *  \param last The end of the sequence.
+ *  \param value The value to be copied.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>,
+ *          and \p ForwardIterator is mutable.
+ *  \tparam T is a model of <a href="http://www.sgi.com/tech/stl/Assignable.html">Assignable</a>,
+ *          and \p T's \c value_type is convertible to \p ForwardIterator's \c value_type.
+ *
+ *  The following code snippet demonstrates how to use \p fill to set a thrust::device_vector's
+ *  elements to a given value using the \p thrust::device execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/fill.h>
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  thrust::device_vector<int> v(4);
+ *  thrust::fill(thrust::device, v.begin(), v.end(), 137);
+ *
+ *  // v[0] == 137, v[1] == 137, v[2] == 137, v[3] == 137
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/fill.html
+ *  \see \c fill_n
+ *  \see \c uninitialized_fill
+ */
+template<typename DerivedPolicy, typename ForwardIterator, typename T>
+  void fill(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+            ForwardIterator first,
+            ForwardIterator last,
+            const T &value);
+
+
+/*! \p fill assigns the value \p value to every element in
+ *  the range <tt>[first, last)</tt>. That is, for every
+ *  iterator \c i in <tt>[first, last)</tt>, it performs
+ *  the assignment <tt>*i = value</tt>.
+ *
+ *  \param first The beginning of the sequence.
+ *  \param last The end of the sequence.
+ *  \param value The value to be copied.
+ *
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>,
+ *          and \p ForwardIterator is mutable.
+ *  \tparam T is a model of <a href="http://www.sgi.com/tech/stl/Assignable.html">Assignable</a>,
+ *          and \p T's \c value_type is convertible to \p ForwardIterator's \c value_type.
+ *
+ *  The following code snippet demonstrates how to use \p fill to set a thrust::device_vector's
+ *  elements to a given value.
+ *
+ *  \code
+ *  #include <thrust/fill.h>
+ *  #include <thrust/device_vector.h>
+ *  ...
+ *  thrust::device_vector<int> v(4);
+ *  thrust::fill(v.begin(), v.end(), 137);
+ *
+ *  // v[0] == 137, v[1] == 137, v[2] == 137, v[3] == 137
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/fill.html
+ *  \see \c fill_n
+ *  \see \c uninitialized_fill
+ */
+template<typename ForwardIterator, typename T>
+  void fill(ForwardIterator first,
+            ForwardIterator last,
+            const T &value);
+
+
+/*! \p fill_n assigns the value \p value to every element in
+ *  the range <tt>[first, first+n)</tt>. That is, for every
+ *  iterator \c i in <tt>[first, first+n)</tt>, it performs
+ *  the assignment <tt>*i = value</tt>.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the sequence.
+ *  \param n The size of the sequence.
+ *  \param value The value to be copied.
+ *  \return <tt>first + n</tt>
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam T is a model of <a href="http://www.sgi.com/tech/stl/Assignable.html">Assignable</a>,
+ *          and \p T's \c value_type is convertible to a type in \p OutputIterator's set of \c value_type.
+ *
+ *  The following code snippet demonstrates how to use \p fill to set a thrust::device_vector's
+ *  elements to a given value using the \p thrust::device execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/fill.h>
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  thrust::device_vector<int> v(4);
+ *  thrust::fill_n(thrust::device, v.begin(), v.size(), 137);
+ *
+ *  // v[0] == 137, v[1] == 137, v[2] == 137, v[3] == 137
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/fill_n.html
+ *  \see \c fill
+ *  \see \c uninitialized_fill_n
+ */
+template<typename DerivedPolicy, typename OutputIterator, typename Size, typename T>
+  OutputIterator fill_n(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                        OutputIterator first,
+                        Size n,
+                        const T &value);
+
+
+/*! \p fill_n assigns the value \p value to every element in
+ *  the range <tt>[first, first+n)</tt>. That is, for every
+ *  iterator \c i in <tt>[first, first+n)</tt>, it performs
+ *  the assignment <tt>*i = value</tt>.
+ *
+ *  \param first The beginning of the sequence.
+ *  \param n The size of the sequence.
+ *  \param value The value to be copied.
+ *  \return <tt>first + n</tt>
+ *
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam T is a model of <a href="http://www.sgi.com/tech/stl/Assignable.html">Assignable</a>,
+ *          and \p T's \c value_type is convertible to a type in \p OutputIterator's set of \c value_type.
+ *
+ *  The following code snippet demonstrates how to use \p fill to set a thrust::device_vector's
+ *  elements to a given value.
+ *
+ *  \code
+ *  #include <thrust/fill.h>
+ *  #include <thrust/device_vector.h>
+ *  ...
+ *  thrust::device_vector<int> v(4);
+ *  thrust::fill_n(v.begin(), v.size(), 137);
+ *
+ *  // v[0] == 137, v[1] == 137, v[2] == 137, v[3] == 137
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/fill_n.html
+ *  \see \c fill
+ *  \see \c uninitialized_fill_n
+ */
+template<typename OutputIterator, typename Size, typename T>
+  OutputIterator fill_n(OutputIterator first,
+                        Size n,
+                        const T &value);
+
+
+/*! \} // end filling
+ *  \} // transformations
+ */
+
+} // end namespace thrust
+
+#include <thrust/detail/fill.inl>
+
diff --git a/compat/thrust/find.h b/compat/thrust/find.h
new file mode 100644
index 0000000000..fa01ded500
--- /dev/null
+++ b/compat/thrust/find.h
@@ -0,0 +1,382 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file find.h
+ *  \brief Locating values in (unsorted) ranges
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/execution_policy.h>
+
+namespace thrust
+{
+
+
+/*! \addtogroup algorithms
+ */
+
+/*! \addtogroup searching
+ *  \ingroup algorithms
+ *  \{
+ */
+
+
+/*! \p find returns the first iterator \c i in the range 
+ *  <tt>[first, last)</tt> such that <tt>*i == value</tt>
+ *  or \c last if no such iterator exists.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first Beginning of the sequence to search.
+ *  \param last End of the sequence to search.
+ *  \param value The value to find.
+ *  \return The first iterator \c i such that <tt>*i == value</tt> or \c last.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>
+ *          and \p InputIterator's \c value_type is equality comparable to type \c T.
+ *  \tparam T is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable.html">EqualityComparable</a>. 
+ *
+ *  \code
+ *  #include <thrust/find.h>
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  thrust::device_vector<int> input(4);
+ *
+ *  input[0] = 0;
+ *  input[1] = 5;
+ *  input[2] = 3;
+ *  input[3] = 7;
+ *
+ *  thrust::device_vector<int>::iterator iter;
+ *
+ *  iter = thrust::find(thrust::device, input.begin(), input.end(), 3); // returns input.first() + 2
+ *  iter = thrust::find(thrust::device, input.begin(), input.end(), 5); // returns input.first() + 1
+ *  iter = thrust::find(thrust::device, input.begin(), input.end(), 9); // returns input.end()
+ *  \endcode
+ *
+ *  \see find_if
+ *  \see mismatch
+ */
+template<typename DerivedPolicy, typename InputIterator, typename T>
+InputIterator find(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                   InputIterator first,
+                   InputIterator last,
+                   const T& value);
+
+
+/*! \p find returns the first iterator \c i in the range 
+ *  <tt>[first, last)</tt> such that <tt>*i == value</tt>
+ *  or \c last if no such iterator exists.
+ *
+ *  \param first Beginning of the sequence to search.
+ *  \param last End of the sequence to search.
+ *  \param value The value to find.
+ *  \return The first iterator \c i such that <tt>*i == value</tt> or \c last.
+ *
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>
+ *          and \p InputIterator's \c value_type is equality comparable to type \c T.
+ *  \tparam T is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable.html">EqualityComparable</a>. 
+ *
+ *  \code
+ *  #include <thrust/find.h>
+ *  #include <thrust/device_vector.h>
+ *  ...
+ *  thrust::device_vector<int> input(4);
+ *
+ *  input[0] = 0;
+ *  input[1] = 5;
+ *  input[2] = 3;
+ *  input[3] = 7;
+ *
+ *  thrust::device_vector<int>::iterator iter;
+ *
+ *  iter = thrust::find(input.begin(), input.end(), 3); // returns input.first() + 2
+ *  iter = thrust::find(input.begin(), input.end(), 5); // returns input.first() + 1
+ *  iter = thrust::find(input.begin(), input.end(), 9); // returns input.end()
+ *  \endcode
+ *
+ *  \see find_if
+ *  \see mismatch
+ */
+template <typename InputIterator, typename T>
+InputIterator find(InputIterator first,
+                   InputIterator last,
+                   const T& value);
+
+
+/*! \p find_if returns the first iterator \c i in the range 
+ *  <tt>[first, last)</tt> such that <tt>pred(*i)</tt> is \c true
+ *  or \c last if no such iterator exists.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first Beginning of the sequence to search.
+ *  \param last End of the sequence to search.
+ *  \param pred A predicate used to test range elements.
+ *  \return The first iterator \c i such that <tt>pred(*i)</tt> is \c true, or \c last.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>.
+ *  \tparam Predicate is a model of <a href="http://www.sgi.com/tech/stl/Predicate.html">Predicate</a>.
+ *
+ *  \code
+ *  #include <thrust/find.h>
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *
+ *  struct greater_than_four
+ *  {
+ *    __host__ __device__
+ *    bool operator()(int x)
+ *    {
+ *      return x > 4;
+ *    }
+ *  };
+ *
+ *  struct greater_than_ten
+ *  {
+ *    __host__ __device__
+ *    bool operator()(int x)
+ *    {
+ *      return x > 10;
+ *    }
+ *  };
+ *
+ *  ...
+ *  thrust::device_vector<int> input(4);
+ *
+ *  input[0] = 0;
+ *  input[1] = 5;
+ *  input[2] = 3;
+ *  input[3] = 7;
+ *
+ *  thrust::device_vector<int>::iterator iter;
+ *
+ *  iter = thrust::find_if(thrust::device, input.begin(), input.end(), greater_than_four()); // returns input.first() + 1
+ *
+ *  iter = thrust::find_if(thrust::device, input.begin(), input.end(), greater_than_ten());  // returns input.end()
+ *  \endcode
+ *
+ *  \see find
+ *  \see find_if_not
+ *  \see mismatch
+ */
+template<typename DerivedPolicy, typename InputIterator, typename Predicate>
+InputIterator find_if(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                      InputIterator first,
+                      InputIterator last,
+                      Predicate pred);
+
+
+/*! \p find_if returns the first iterator \c i in the range 
+ *  <tt>[first, last)</tt> such that <tt>pred(*i)</tt> is \c true
+ *  or \c last if no such iterator exists.
+ *
+ *  \param first Beginning of the sequence to search.
+ *  \param last End of the sequence to search.
+ *  \param pred A predicate used to test range elements.
+ *  \return The first iterator \c i such that <tt>pred(*i)</tt> is \c true, or \c last.
+ *
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>.
+ *  \tparam Predicate is a model of <a href="http://www.sgi.com/tech/stl/Predicate.html">Predicate</a>.
+ *
+ *  \code
+ *  #include <thrust/find.h>
+ *  #include <thrust/device_vector.h>
+ *
+ *  struct greater_than_four
+ *  {
+ *    __host__ __device__
+ *    bool operator()(int x)
+ *    {
+ *      return x > 4;
+ *    }
+ *  };
+ *
+ *  struct greater_than_ten
+ *  {
+ *    __host__ __device__
+ *    bool operator()(int x)
+ *    {
+ *      return x > 10;
+ *    }
+ *  };
+ *
+ *  ...
+ *  thrust::device_vector<int> input(4);
+ *
+ *  input[0] = 0;
+ *  input[1] = 5;
+ *  input[2] = 3;
+ *  input[3] = 7;
+ *
+ *  thrust::device_vector<int>::iterator iter;
+ *
+ *  iter = thrust::find_if(input.begin(), input.end(), greater_than_four()); // returns input.first() + 1
+ *
+ *  iter = thrust::find_if(input.begin(), input.end(), greater_than_ten());  // returns input.end()
+ *  \endcode
+ *
+ *  \see find
+ *  \see find_if_not
+ *  \see mismatch
+ */
+template <typename InputIterator, typename Predicate>
+InputIterator find_if(InputIterator first,
+                      InputIterator last,
+                      Predicate pred);
+
+
+/*! \p find_if_not returns the first iterator \c i in the range 
+ *  <tt>[first, last)</tt> such that <tt>pred(*i)</tt> is \c false
+ *  or \c last if no such iterator exists.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first Beginning of the sequence to search.
+ *  \param last End of the sequence to search.
+ *  \param pred A predicate used to test range elements.
+ *  \return The first iterator \c i such that <tt>pred(*i)</tt> is \c false, or \c last.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>.
+ *  \tparam Predicate is a model of <a href="http://www.sgi.com/tech/stl/Predicate.html">Predicate</a>.
+ *
+ *  \code
+ *  #include <thrust/find.h>
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *
+ *  struct greater_than_four
+ *  {
+ *    __host__ __device__
+ *    bool operator()(int x)
+ *    {
+ *      return x > 4;
+ *    }
+ *  };
+ *
+ *  struct greater_than_ten
+ *  {
+ *    __host__ __device__
+ *    bool operator()(int x)
+ *    {
+ *      return x > 10;
+ *    }
+ *  };
+ *
+ *  ...
+ *  thrust::device_vector<int> input(4);
+ *
+ *  input[0] = 0;
+ *  input[1] = 5;
+ *  input[2] = 3;
+ *  input[3] = 7;
+ *
+ *  thrust::device_vector<int>::iterator iter;
+ *
+ *  iter = thrust::find_if_not(thrust::device, input.begin(), input.end(), greater_than_four()); // returns input.first()
+ *
+ *  iter = thrust::find_if_not(thrust::device, input.begin(), input.end(), greater_than_ten());  // returns input.first()
+ *  \endcode
+ *
+ *  \see find
+ *  \see find_if
+ *  \see mismatch
+ */
+template<typename DerivedPolicy, typename InputIterator, typename Predicate>
+InputIterator find_if_not(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                          InputIterator first,
+                          InputIterator last,
+                          Predicate pred);
+
+
+/*! \p find_if_not returns the first iterator \c i in the range 
+ *  <tt>[first, last)</tt> such that <tt>pred(*i)</tt> is \c false
+ *  or \c last if no such iterator exists.
+ *
+ *  \param first Beginning of the sequence to search.
+ *  \param last End of the sequence to search.
+ *  \param pred A predicate used to test range elements.
+ *  \return The first iterator \c i such that <tt>pred(*i)</tt> is \c false, or \c last.
+ *
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>.
+ *  \tparam Predicate is a model of <a href="http://www.sgi.com/tech/stl/Predicate.html">Predicate</a>.
+ *
+ *  \code
+ *  #include <thrust/find.h>
+ *  #include <thrust/device_vector.h>
+ *
+ *  struct greater_than_four
+ *  {
+ *    __host__ __device__
+ *    bool operator()(int x)
+ *    {
+ *      return x > 4;
+ *    }
+ *  };
+ *
+ *  struct greater_than_ten
+ *  {
+ *    __host__ __device__
+ *    bool operator()(int x)
+ *    {
+ *      return x > 10;
+ *    }
+ *  };
+ *
+ *  ...
+ *  thrust::device_vector<int> input(4);
+ *
+ *  input[0] = 0;
+ *  input[1] = 5;
+ *  input[2] = 3;
+ *  input[3] = 7;
+ *
+ *  thrust::device_vector<int>::iterator iter;
+ *
+ *  iter = thrust::find_if_not(input.begin(), input.end(), greater_than_four()); // returns input.first()
+ *
+ *  iter = thrust::find_if_not(input.begin(), input.end(), greater_than_ten());  // returns input.first()
+ *  \endcode
+ *
+ *  \see find
+ *  \see find_if
+ *  \see mismatch
+ */
+template <typename InputIterator, typename Predicate>
+InputIterator find_if_not(InputIterator first,
+                          InputIterator last,
+                          Predicate pred);
+
+/*! \} // end searching
+ */
+
+
+} // end namespace thrust
+
+#include <thrust/detail/find.inl>
+
diff --git a/compat/thrust/for_each.h b/compat/thrust/for_each.h
new file mode 100644
index 0000000000..efab9d8fab
--- /dev/null
+++ b/compat/thrust/for_each.h
@@ -0,0 +1,278 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ * *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file for_each.h
+ *  \brief Applies a function to each element in a range
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/type_traits.h>
+#include <thrust/detail/execution_policy.h>
+
+namespace thrust
+{
+
+
+/*! \addtogroup modifying
+ *  \ingroup transformations
+ *  \{
+ */
+
+
+/*! \p for_each applies the function object \p f to each element
+ *  in the range <tt>[first, last)</tt>; \p f's return value, if any,
+ *  is ignored. Unlike the C++ Standard Template Library function
+ *  <tt>std::for_each</tt>, this version offers no guarantee on
+ *  order of execution. For this reason, this version of \p for_each
+ *  does not return a copy of the function object.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the sequence.
+ *  \param last The end of the sequence.
+ *  \param f The function object to apply to the range <tt>[first, last)</tt>.
+ *  \return last
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator">Input Iterator</a>,
+ *          and \p InputIterator's \c value_type is convertible to \p UnaryFunction's \c argument_type.
+ *  \tparam UnaryFunction is a model of <a href="http://www.sgi.com/tech/stl/UnaryFunction">Unary Function</a>,
+ *          and \p UnaryFunction does not apply any non-constant operation through its argument.
+ *
+ *  The following code snippet demonstrates how to use \p for_each to print the elements
+ *  of a \p std::device_vector using the \p thrust::device parallelization policy:
+ *
+ *  \code
+ *  #include <thrust/for_each.h>
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/execution_policy.h>
+ *  #include <cstdio>
+ *  ...
+ *
+ *  struct printf_functor
+ *  {
+ *    __host__ __device__
+ *    void operator()(int x)
+ *    {
+ *      // note that using printf in a __device__ function requires
+ *      // code compiled for a GPU with compute capability 2.0 or
+ *      // higher (nvcc --arch=sm_20)
+ *      printf("%d\n");
+ *    }
+ *  };
+ *  ...
+ *  thrust::device_vector<int> d_vec(3);
+ *  d_vec[0] = 0; d_vec[1] = 1; d_vec[2] = 2;
+ *
+ *  thrust::for_each(thrust::device, d_vec.begin(), d_vec.end(), printf_functor());
+ *
+ *  // 0 1 2 is printed to standard output in some unspecified order
+ *  \endcode
+ *
+ *  \see for_each_n
+ *  \see http://www.sgi.com/tech/stl/for_each.html
+ */
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename UnaryFunction>
+InputIterator for_each(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                       InputIterator first,
+                       InputIterator last,
+                       UnaryFunction f);
+
+
+/*! \p for_each_n applies the function object \p f to each element
+ *  in the range <tt>[first, first + n)</tt>; \p f's return value, if any,
+ *  is ignored. Unlike the C++ Standard Template Library function
+ *  <tt>std::for_each</tt>, this version offers no guarantee on
+ *  order of execution.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the sequence.
+ *  \param n The size of the input sequence.
+ *  \param f The function object to apply to the range <tt>[first, first + n)</tt>.
+ *  \return <tt>first + n</tt>
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator">Input Iterator</a>,
+ *          and \p InputIterator's \c value_type is convertible to \p UnaryFunction's \c argument_type.
+ *  \tparam Size is an integral type.
+ *  \tparam UnaryFunction is a model of <a href="http://www.sgi.com/tech/stl/UnaryFunction">Unary Function</a>,
+ *          and \p UnaryFunction does not apply any non-constant operation through its argument.
+ *
+ *  The following code snippet demonstrates how to use \p for_each_n to print the elements
+ *  of a \p device_vector using the \p thrust::device parallelization policy.
+ *
+ *  \code
+ *  #include <thrust/for_each.h>
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/execution_policy.h>
+ *  #include <cstdio>
+ *
+ *  struct printf_functor
+ *  {
+ *    __host__ __device__
+ *    void operator()(int x)
+ *    {
+ *      // note that using printf in a __device__ function requires
+ *      // code compiled for a GPU with compute capability 2.0 or
+ *      // higher (nvcc --arch=sm_20)
+ *      printf("%d\n");
+ *    }
+ *  };
+ *  ...
+ *  thrust::device_vector<int> d_vec(3);
+ *  d_vec[0] = 0; d_vec[1] = 1; d_vec[2] = 2;
+ *
+ *  thrust::for_each_n(thrust::device, d_vec.begin(), d_vec.size(), printf_functor());
+ *
+ *  // 0 1 2 is printed to standard output in some unspecified order
+ *  \endcode
+ *
+ *  \see for_each
+ *  \see http://www.sgi.com/tech/stl/for_each.html
+ */
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename Size,
+         typename UnaryFunction>
+InputIterator for_each_n(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                         InputIterator first,
+                         Size n,
+                         UnaryFunction f);
+
+/*! \p for_each applies the function object \p f to each element
+ *  in the range <tt>[first, last)</tt>; \p f's return value, if any,
+ *  is ignored. Unlike the C++ Standard Template Library function
+ *  <tt>std::for_each</tt>, this version offers no guarantee on
+ *  order of execution. For this reason, this version of \p for_each
+ *  does not return a copy of the function object.
+ *
+ *  \param first The beginning of the sequence.
+ *  \param last The end of the sequence.
+ *  \param f The function object to apply to the range <tt>[first, last)</tt>.
+ *  \return last
+ *
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator">Input Iterator</a>,
+ *          and \p InputIterator's \c value_type is convertible to \p UnaryFunction's \c argument_type.
+ *  \tparam UnaryFunction is a model of <a href="http://www.sgi.com/tech/stl/UnaryFunction">Unary Function</a>,
+ *          and \p UnaryFunction does not apply any non-constant operation through its argument.
+ *
+ *  The following code snippet demonstrates how to use \p for_each to print the elements
+ *  of a \p device_vector.
+ *
+ *  \code
+ *  #include <thrust/for_each.h>
+ *  #include <thrust/device_vector.h>
+ *  #include <stdio.h>
+ *
+ *  struct printf_functor
+ *  {
+ *    __host__ __device__
+ *    void operator()(int x)
+ *    {
+ *      // note that using printf in a __device__ function requires
+ *      // code compiled for a GPU with compute capability 2.0 or
+ *      // higher (nvcc --arch=sm_20)
+ *      printf("%d\n");
+ *    }
+ *  };
+ *  ...
+ *  thrust::device_vector<int> d_vec(3);
+ *  d_vec[0] = 0; d_vec[1] = 1; d_vec[2] = 2;
+ *
+ *  thrust::for_each(d_vec.begin(), d_vec.end(), printf_functor());
+ *
+ *  // 0 1 2 is printed to standard output in some unspecified order
+ *  \endcode
+ *
+ *  \see for_each_n
+ *  \see http://www.sgi.com/tech/stl/for_each.html
+ */
+template<typename InputIterator,
+         typename UnaryFunction>
+InputIterator for_each(InputIterator first,
+                       InputIterator last,
+                       UnaryFunction f);
+
+
+/*! \p for_each_n applies the function object \p f to each element
+ *  in the range <tt>[first, first + n)</tt>; \p f's return value, if any,
+ *  is ignored. Unlike the C++ Standard Template Library function
+ *  <tt>std::for_each</tt>, this version offers no guarantee on
+ *  order of execution.
+ *
+ *  \param first The beginning of the sequence.
+ *  \param n The size of the input sequence.
+ *  \param f The function object to apply to the range <tt>[first, first + n)</tt>.
+ *  \return <tt>first + n</tt>
+ *
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator">Input Iterator</a>,
+ *          and \p InputIterator's \c value_type is convertible to \p UnaryFunction's \c argument_type.
+ *  \tparam Size is an integral type.
+ *  \tparam UnaryFunction is a model of <a href="http://www.sgi.com/tech/stl/UnaryFunction">Unary Function</a>,
+ *          and \p UnaryFunction does not apply any non-constant operation through its argument.
+ *
+ *  The following code snippet demonstrates how to use \p for_each_n to print the elements
+ *  of a \p device_vector.
+ *
+ *  \code
+ *  #include <thrust/for_each.h>
+ *  #include <thrust/device_vector.h>
+ *  #include <stdio.h>
+ *
+ *  struct printf_functor
+ *  {
+ *    __host__ __device__
+ *    void operator()(int x)
+ *    {
+ *      // note that using printf in a __device__ function requires
+ *      // code compiled for a GPU with compute capability 2.0 or
+ *      // higher (nvcc --arch=sm_20)
+ *      printf("%d\n");
+ *    }
+ *  };
+ *  ...
+ *  thrust::device_vector<int> d_vec(3);
+ *  d_vec[0] = 0; d_vec[1] = 1; d_vec[2] = 2;
+ *
+ *  thrust::for_each_n(d_vec.begin(), d_vec.size(), printf_functor());
+ *
+ *  // 0 1 2 is printed to standard output in some unspecified order
+ *  \endcode
+ *
+ *  \see for_each
+ *  \see http://www.sgi.com/tech/stl/for_each.html
+ */
+template<typename InputIterator,
+         typename Size,
+         typename UnaryFunction>
+InputIterator for_each_n(InputIterator first,
+                         Size n,
+                         UnaryFunction f);
+
+/*! \} // end modifying
+ */
+
+} // end namespace thrust
+
+#include <thrust/detail/for_each.inl>
+
diff --git a/compat/thrust/functional.h b/compat/thrust/functional.h
new file mode 100644
index 0000000000..b3d47f9179
--- /dev/null
+++ b/compat/thrust/functional.h
@@ -0,0 +1,1079 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file functional.h
+ *  \brief Function objects and tools for manipulating them
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <functional>
+#include <thrust/detail/functional/placeholder.h>
+
+namespace thrust
+{
+
+/*! \addtogroup function_objects Function Objects
+ */
+
+template<typename Operation> struct unary_traits;
+
+template<typename Operation> struct binary_traits;
+
+/*! \addtogroup function_object_adaptors Function Object Adaptors
+ *  \ingroup function_objects
+ *  \{
+ */
+
+/*! \p unary_function is an empty base class: it contains no member functions
+ *  or member variables, but only type information. The only reason it exists
+ *  is to make it more convenient to define types that are models of the
+ *  concept Adaptable Unary Function. Specifically, any model of Adaptable
+ *  Unary Function must define nested \c typedefs. Those \c typedefs are
+ *  provided by the base class \p unary_function.
+ *
+ *  The following code snippet demonstrates how to construct an 
+ *  Adaptable Unary Function using \p unary_function.
+ *
+ *  \code
+ *  struct sine : public thrust::unary_function<float,float>
+ *  {
+ *    __host__ __device__
+ *    float operator()(float x) { return sinf(x); }
+ *  };
+ *  \endcode
+ *
+ *  \note unary_function is currently redundant with the C++ STL type
+ *  \c std::unary_function. We reserve it here for potential additional
+ *  functionality at a later date.
+ *
+ *  \see http://www.sgi.com/tech/stl/unary_function.html
+ *  \see binary_function
+ */
+template<typename Argument,
+         typename Result>
+  struct unary_function
+    : public std::unary_function<Argument, Result>
+{
+}; // end unary_function
+
+/*! \p binary_function is an empty base class: it contains no member functions
+ *  or member variables, but only type information. The only reason it exists
+ *  is to make it more convenient to define types that are models of the
+ *  concept Adaptable Binary Function. Specifically, any model of Adaptable
+ *  Binary Function must define nested \c typedefs. Those \c typedefs are
+ *  provided by the base class \p binary_function.
+ *
+ *  The following code snippet demonstrates how to construct an 
+ *  Adaptable Binary Function using \p binary_function.
+ *
+ *  \code
+ *  struct exponentiate : public thrust::binary_function<float,float,float>
+ *  {
+ *    __host__ __device__
+ *    float operator()(float x, float y) { return powf(x,y); }
+ *  };
+ *  \endcode
+ *
+ *  \note binary_function is currently redundant with the C++ STL type
+ *  \c std::binary_function. We reserve it here for potential additional
+ *  functionality at a later date.
+ *
+ *  \see http://www.sgi.com/tech/stl/binary_function.html
+ *  \see unary_function
+ */
+template<typename Argument1,
+         typename Argument2,
+         typename Result>
+  struct binary_function
+    : public std::binary_function<Argument1, Argument2, Result>
+{
+}; // end binary_function
+
+/*! \}
+ */
+
+
+/*! \addtogroup predefined_function_objects Predefined Function Objects
+ *  \ingroup function_objects
+ */
+
+/*! \addtogroup arithmetic_operations Arithmetic Operations
+ *  \ingroup predefined_function_objects
+ *  \{
+ */
+
+/*! \p plus is a function object. Specifically, it is an Adaptable Binary Function.
+ *  If \c f is an object of class <tt>plus<T></tt>, and \c x and \c y are objects
+ *  of class \c T, then <tt>f(x,y)</tt> returns <tt>x+y</tt>.
+ *
+ *  \tparam T is a model of <a href="http://www.sgi.com/tech/stl/Assignable.html">Assignable</a>,
+ *          and if \c x and \c y are objects of type \p T, then <tt>x+y</tt> must be defined and must have a return type that is convertible to \c T.
+ *
+ *  The following code snippet demonstrates how to use <tt>plus</tt> to sum two
+ *  device_vectors of \c floats.
+ *
+ *  \code
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/functional.h>
+ *  #include <thrust/sequence.h>
+ *  #include <thrust/fill.h>
+ *  #include <thrust/transform.h>
+ *  ...
+ *  const int N = 1000;
+ *  thrust::device_vector<float> V1(N);
+ *  thrust::device_vector<float> V2(N);
+ *  thrust::device_vector<float> V3(N);
+ *
+ *  thrust::sequence(V1.begin(), V1.end(), 1);
+ *  thrust::fill(V2.begin(), V2.end(), 75);
+ *
+ *  thrust::transform(V1.begin(), V1.end(), V2.begin(), V3.begin(),
+ *                     thrust::plus<float>());
+ *  // V3 is now {76, 77, 78, ..., 1075}
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/plus.html
+ *  \see binary_function
+ */
+template<typename T>
+  struct plus : public binary_function<T,T,T>
+{
+  /*! Function call operator. The return value is <tt>lhs + rhs</tt>.
+   */
+  __host__ __device__ T operator()(const T &lhs, const T &rhs) const {return lhs + rhs;}
+}; // end plus
+
+/*! \p minus is a function object. Specifically, it is an Adaptable Binary Function.
+ *  If \c f is an object of class <tt>minus<T></tt>, and \c x and \c y are objects
+ *  of class \c T, then <tt>f(x,y)</tt> returns <tt>x-y</tt>.
+ *
+ *  \tparam T is a model of <a href="http://www.sgi.com/tech/stl/Assignable.html">Assignable</a>,
+ *          and if \c x and \c y are objects of type \p T, then <tt>x-y</tt> must be defined and must have a return type that is convertible to \c T.
+ *
+ *  The following code snippet demonstrates how to use <tt>minus</tt> to subtract
+ *  a device_vector of \c floats from another.
+ *
+ *  \code
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/functional.h>
+ *  #include <thrust/sequence.h>
+ *  #include <thrust/fill.h>
+ *  #include <thrust/transform.h>
+ *  ...
+ *  const int N = 1000;
+ *  thrust::device_vector<float> V1(N);
+ *  thrust::device_vector<float> V2(N);
+ *  thrust::device_vector<float> V3(N);
+ *
+ *  thrust::sequence(V1.begin(), V1.end(), 1);
+ *  thrust::fill(V2.begin(), V2.end(), 75);
+ *
+ *  thrust::transform(V1.begin(), V1.end(), V2.begin(), V3.begin(),
+ *                     thrust::minus<float>());
+ *  // V3 is now {-74, -75, -76, ..., -925}
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/minus.html
+ *  \see binary_function
+ */
+template<typename T>
+  struct minus : public binary_function<T,T,T>
+{
+  /*! Function call operator. The return value is <tt>lhs - rhs</tt>.
+   */
+  __host__ __device__ T operator()(const T &lhs, const T &rhs) const {return lhs - rhs;}
+}; // end minus
+
+/*! \p multiplies is a function object. Specifically, it is an Adaptable Binary Function.
+ *  If \c f is an object of class <tt>minus<T></tt>, and \c x and \c y are objects
+ *  of class \c T, then <tt>f(x,y)</tt> returns <tt>x*y</tt>.
+ *
+ *  \tparam T is a model of <a href="http://www.sgi.com/tech/stl/Assignable.html">Assignable</a>,
+ *          and if \c x and \c y are objects of type \p T, then <tt>x*y</tt> must be defined and must have a return type that is convertible to \c T.
+ *
+ *  The following code snippet demonstrates how to use <tt>multiplies</tt> to multiply
+ *  two device_vectors of \c floats.
+ *
+ *  \code
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/functional.h>
+ *  #include <thrust/sequence.h>
+ *  #include <thrust/fill.h>
+ *  #include <thrust/transform.h>
+ *  ...
+ *  const int N = 1000;
+ *  thrust::device_vector<float> V1(N);
+ *  thrust::device_vector<float> V2(N);
+ *  thrust::device_vector<float> V3(N);
+ *
+ *  thrust::sequence(V1.begin(), V1.end(), 1);
+ *  thrust::fill(V2.begin(), V2.end(), 75);
+ *
+ *  thrust::transform(V1.begin(), V1.end(), V2.begin(), V3.begin(),
+ *                     thrust::multiplies<float>());
+ *  // V3 is now {75, 150, 225, ..., 75000}
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/multiplies.html
+ *  \see binary_function
+ */
+template<typename T>
+  struct multiplies : public binary_function<T,T,T>
+{
+  /*! Function call operator. The return value is <tt>lhs * rhs</tt>.
+   */
+  __host__ __device__ T operator()(const T &lhs, const T &rhs) const {return lhs * rhs;}
+}; // end multiplies
+
+/*! \p divides is a function object. Specifically, it is an Adaptable Binary Function.
+ *  If \c f is an object of class <tt>divides<T></tt>, and \c x and \c y are objects
+ *  of class \c T, then <tt>f(x,y)</tt> returns <tt>x/y</tt>.
+ *
+ *  \tparam T is a model of <a href="http://www.sgi.com/tech/stl/Assignable.html">Assignable</a>,
+ *          and if \c x and \c y are objects of type \p T, then <tt>x/y</tt> must be defined and must have a return type that is convertible to \c T.
+ *
+ *  The following code snippet demonstrates how to use <tt>divides</tt> to divide
+ *  one device_vectors of \c floats by another.
+ *
+ *  \code
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/functional.h>
+ *  #include <thrust/sequence.h>
+ *  #include <thrust/fill.h>
+ *  #include <thrust/transform.h>
+ *  ...
+ *  const int N = 1000;
+ *  thrust::device_vector<float> V1(N);
+ *  thrust::device_vector<float> V2(N);
+ *  thrust::device_vector<float> V3(N);
+ *
+ *  thrust::sequence(V1.begin(), V1.end(), 1);
+ *  thrust::fill(V2.begin(), V2.end(), 75);
+ *
+ *  thrust::transform(V1.begin(), V1.end(), V2.begin(), V3.begin(),
+ *                     thrust::divides<float>());
+ *  // V3 is now {1/75, 2/75, 3/75, ..., 1000/75}
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/divides.html
+ *  \see binary_function
+ */
+template<typename T>
+  struct divides : public binary_function<T,T,T>
+{
+  /*! Function call operator. The return value is <tt>lhs / rhs</tt>.
+   */
+  __host__ __device__ T operator()(const T &lhs, const T &rhs) const {return lhs / rhs;}
+}; // end divides
+
+/*! \p modulus is a function object. Specifically, it is an Adaptable Binary Function.
+ *  If \c f is an object of class <tt>divides<T></tt>, and \c x and \c y are objects
+ *  of class \c T, then <tt>f(x,y)</tt> returns <tt>x%y</tt>.
+ *
+ *  \tparam T is a model of <a href="http://www.sgi.com/tech/stl/Assignable.html">Assignable</a>,
+ *          and if \c x and \c y are objects of type \p T, then <tt>x%y</tt> must be defined and must have a return type that is convertible to \c T.
+ *
+ *  The following code snippet demonstrates how to use <tt>modulus</tt> to take
+ *  the modulus of one device_vectors of \c floats by another.
+ *
+ *  \code
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/functional.h>
+ *  #include <thrust/sequence.h>
+ *  #include <thrust/fill.h>
+ *  #include <thrust/transform.h>
+ *  ...
+ *  const int N = 1000;
+ *  thrust::device_vector<float> V1(N);
+ *  thrust::device_vector<float> V2(N);
+ *  thrust::device_vector<float> V3(N);
+ *
+ *  thrust::sequence(V1.begin(), V1.end(), 1);
+ *  thrust::fill(V2.begin(), V2.end(), 75);
+ *
+ *  thrust::transform(V1.begin(), V1.end(), V2.begin(), V3.begin(),
+ *                     thrust::modulus<int>());
+ *  // V3 is now {1%75, 2%75, 3%75, ..., 1000%75}
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/modulus.html
+ *  \see binary_function
+ */
+template<typename T>
+  struct modulus : public binary_function<T,T,T>
+{
+  /*! Function call operator. The return value is <tt>lhs % rhs</tt>.
+   */
+  __host__ __device__ T operator()(const T &lhs, const T &rhs) const {return lhs % rhs;}
+}; // end modulus
+
+/*! \p negate is a function object. Specifically, it is an Adaptable Unary Function.
+ *  If \c f is an object of class <tt>negate<T></tt>, and \c x is an object
+ *  of class \c T, then <tt>f(x)</tt> returns <tt>-x</tt>.
+ *
+ *  \tparam T is a model of <a href="http://www.sgi.com/tech/stl/Assignable.html">Assignable</a>,
+ *          and if \c x is an object of type \p T, then <tt>-x</tt> must be defined and must have a return type that is convertible to \c T.
+ *
+ *  The following code snippet demonstrates how to use <tt>negate</tt> to negate
+ *  the element of a device_vector of \c floats.
+ *
+ *  \code
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/functional.h>
+ *  #include <thrust/sequence.h>
+ *  #include <thrust/transform.h>
+ *  ...
+ *  const int N = 1000;
+ *  thrust::device_vector<float> V1(N);
+ *  thrust::device_vector<float> V2(N);
+ *
+ *  thrust::sequence(V1.begin(), V1.end(), 1);
+ *
+ *  thrust::transform(V1.begin(), V1.end(), V2.begin(),
+ *                     thrust::negate<float>());
+ *  // V2 is now {-1, -2, -3, ..., -1000}
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/negate.html
+ *  \see unary_function
+ */
+template<typename T>
+  struct negate : public unary_function<T,T>
+{
+  /*! Function call operator. The return value is <tt>-x</tt>.
+   */
+  __host__ __device__ T operator()(const T &x) const {return -x;}
+}; // end negate
+
+/*! \}
+ */
+
+/*! \addtogroup comparison_operations Comparison Operations
+ *  \ingroup predefined_function_objects
+ *  \{
+ */
+
+/*! \p equal_to is a function object. Specifically, it is an Adaptable Binary
+ *  Predicate, which means it is a function object that tests the truth or falsehood
+ *  of some condition. If \c f is an object of class <tt>equal_to<T></tt> and \c x
+ *  and \c y are objects of class \c T, then <tt>f(x,y)</tt> returns \c true if
+ *  <tt>x == y</tt> and \c false otherwise.
+ *
+ *  \tparam T is a model of <a href="http://www.sgi.com/tech/stl/EqualityComparable.html">Equality Comparable</a>.
+ *
+ *  \see http://www.sgi.com/tech/stl/equal_to.html
+ *  \see binary_function
+ */
+template<typename T>
+  struct equal_to : public binary_function<T,T,bool>
+{
+  /*! Function call operator. The return value is <tt>lhs == rhs</tt>.
+   */
+  __host__ __device__ bool operator()(const T &lhs, const T &rhs) const {return lhs == rhs;}
+}; // end equal_to
+
+/*! \p not_equal_to is a function object. Specifically, it is an Adaptable Binary
+ *  Predicate, which means it is a function object that tests the truth or falsehood
+ *  of some condition. If \c f is an object of class <tt>not_equal_to<T></tt> and \c x
+ *  and \c y are objects of class \c T, then <tt>f(x,y)</tt> returns \c true if
+ *  <tt>x != y</tt> and \c false otherwise.
+ *
+ *  \tparam T is a model of <a href="http://www.sgi.com/tech/stl/EqualityComparable.html">Equality Comparable</a>.
+ *
+ *  \see http://www.sgi.com/tech/stl/not_equal_to.html
+ *  \see binary_function
+ */
+template<typename T>
+  struct not_equal_to : public binary_function<T,T,bool>
+{
+  /*! Function call operator. The return value is <tt>lhs != rhs</tt>.
+   */
+  __host__ __device__ bool operator()(const T &lhs, const T &rhs) const {return lhs != rhs;}
+}; // end not_equal_to
+
+/*! \p greater is a function object. Specifically, it is an Adaptable Binary
+ *  Predicate, which means it is a function object that tests the truth or falsehood
+ *  of some condition. If \c f is an object of class <tt>greater<T></tt> and \c x
+ *  and \c y are objects of class \c T, then <tt>f(x,y)</tt> returns \c true if
+ *  <tt>x > y</tt> and \c false otherwise.
+ *
+ *  \tparam T is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable.html">LessThan Comparable</a>.
+ *
+ *  \see http://www.sgi.com/tech/stl/greater.html
+ *  \see binary_function
+ */
+template<typename T>
+  struct greater : public binary_function<T,T,bool>
+{
+  /*! Function call operator. The return value is <tt>lhs > rhs</tt>.
+   */
+  __host__ __device__ bool operator()(const T &lhs, const T &rhs) const {return lhs > rhs;}
+}; // end greater
+
+/*! \p less is a function object. Specifically, it is an Adaptable Binary
+ *  Predicate, which means it is a function object that tests the truth or falsehood
+ *  of some condition. If \c f is an object of class <tt>less<T></tt> and \c x
+ *  and \c y are objects of class \c T, then <tt>f(x,y)</tt> returns \c true if
+ *  <tt>x < y</tt> and \c false otherwise.
+ *
+ *  \tparam T is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable.html">LessThan Comparable</a>.
+ *
+ *  \see http://www.sgi.com/tech/stl/less.html
+ *  \see binary_function
+ */
+template<typename T>
+  struct less : public binary_function<T,T,bool>
+{
+  /*! Function call operator. The return value is <tt>lhs < rhs</tt>.
+   */
+  __host__ __device__ bool operator()(const T &lhs, const T &rhs) const {return lhs < rhs;}
+}; // end less
+
+/*! \p greater_equal is a function object. Specifically, it is an Adaptable Binary
+ *  Predicate, which means it is a function object that tests the truth or falsehood
+ *  of some condition. If \c f is an object of class <tt>greater_equal<T></tt> and \c x
+ *  and \c y are objects of class \c T, then <tt>f(x,y)</tt> returns \c true if
+ *  <tt>x >= y</tt> and \c false otherwise.
+ *
+ *  \tparam T is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable.html">LessThan Comparable</a>.
+ *
+ *  \see http://www.sgi.com/tech/stl/greater_equal.html
+ *  \see binary_function
+ */
+template<typename T>
+  struct greater_equal : public binary_function<T,T,bool>
+{
+  /*! Function call operator. The return value is <tt>lhs >= rhs</tt>.
+   */
+  __host__ __device__ bool operator()(const T &lhs, const T &rhs) const {return lhs >= rhs;}
+}; // end greater_equal
+
+/*! \p less_equal is a function object. Specifically, it is an Adaptable Binary
+ *  Predicate, which means it is a function object that tests the truth or falsehood
+ *  of some condition. If \c f is an object of class <tt>less_equal<T></tt> and \c x
+ *  and \c y are objects of class \c T, then <tt>f(x,y)</tt> returns \c true if
+ *  <tt>x <= y</tt> and \c false otherwise.
+ *
+ *  \tparam T is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable.html">LessThan Comparable</a>.
+ *
+ *  \see http://www.sgi.com/tech/stl/less_equal.html
+ *  \see binary_function
+ */
+template<typename T>
+  struct less_equal : public binary_function<T,T,bool>
+{
+  /*! Function call operator. The return value is <tt>lhs <= rhs</tt>.
+   */
+  __host__ __device__ bool operator()(const T &lhs, const T &rhs) const {return lhs <= rhs;}
+}; // end less_equal
+
+/*! \}
+ */
+
+
+/*! \addtogroup logical_operations Logical Operations
+ *  \ingroup predefined_function_objects
+ *  \{
+ */
+
+/*! \p logical_and is a function object. Specifically, it is an Adaptable Binary Predicate,
+ *  which means it is a function object that tests the truth or falsehood of some condition.
+ *  If \c f is an object of class <tt>logical_and<T></tt> and \c x and \c y are objects of
+ *  class \c T (where \c T is convertible to \c bool) then <tt>f(x,y)</tt> returns \c true
+ *  if and only if both \c x and \c y are \c true.
+ *
+ *  \tparam T must be convertible to \c bool.
+ *
+ *  \see http://www.sgi.com/tech/stl/logical_and.html
+ *  \see binary_function
+ */
+template<typename T>
+  struct logical_and : public binary_function<T,T,bool>
+{
+  /*! Function call operator. The return value is <tt>lhs && rhs</tt>.
+   */
+  __host__ __device__ bool operator()(const T &lhs, const T &rhs) const {return lhs && rhs;}
+}; // end logical_and
+
+/*! \p logical_or is a function object. Specifically, it is an Adaptable Binary Predicate,
+ *  which means it is a function object that tests the truth or falsehood of some condition.
+ *  If \c f is an object of class <tt>logical_or<T></tt> and \c x and \c y are objects of
+ *  class \c T (where \c T is convertible to \c bool) then <tt>f(x,y)</tt> returns \c true
+ *  if and only if either \c x or \c y are \c true.
+ *
+ *  \tparam T must be convertible to \c bool.
+ *
+ *  \see http://www.sgi.com/tech/stl/logical_or.html
+ *  \see binary_function
+ */
+template<typename T>
+  struct logical_or : public binary_function<T,T,bool>
+{
+  /*! Function call operator. The return value is <tt>lhs || rhs</tt>.
+   */
+  __host__ __device__ bool operator()(const T &lhs, const T &rhs) const {return lhs || rhs;}
+}; // end logical_or
+
+/*! \p logical_not is a function object. Specifically, it is an Adaptable Predicate,
+ *  which means it is a function object that tests the truth or falsehood of some condition.
+ *  If \c f is an object of class <tt>logical_not<T></tt> and \c x is an object of
+ *  class \c T (where \c T is convertible to \c bool) then <tt>f(x)</tt> returns \c true
+ *  if and only if \c x is \c false.
+ *
+ *  \tparam T must be convertible to \c bool.
+ *
+ *  The following code snippet demonstrates how to use \p logical_not to transform
+ *  a device_vector of \c bools into its logical complement.
+ *
+ *  \code
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/transform.h>
+ *  #include <thrust/functional.h>
+ *  ...
+ *  thrust::device_vector<bool> V;
+ *  ...
+ *  thrust::transform(V.begin(), V.end(), V.begin(), thrust::logical_not<bool>());
+ *  // The elements of V are now the logical complement of what they were prior
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/logical_not.html
+ *  \see unary_function
+ */
+template<typename T>
+  struct logical_not : public unary_function<T,bool>
+{
+  /*! Function call operator. The return value is <tt>!x</tt>.
+   */
+  __host__ __device__ bool operator()(const T &x) const {return !x;}
+}; // end logical_not
+
+/*! \}
+ */
+
+/*! \addtogroup bitwise_operations Bitwise Operations
+ *  \ingroup predefined_function_objects
+ *  \{
+ */
+
+/*! \p bit_and is a function object. Specifically, it is an Adaptable Binary Function.
+ *  If \c f is an object of class <tt>bit_and<T></tt>, and \c x and \c y are objects
+ *  of class \c T, then <tt>f(x,y)</tt> returns <tt>x&y</tt>.
+ *
+ *  \tparam T is a model of <a href="http://www.sgi.com/tech/stl/Assignable.html">Assignable</a>,
+ *          and if \c x and \c y are objects of type \p T, then <tt>x&y</tt> must be defined and must have a return type that is convertible to \c T.
+ *
+ *  The following code snippet demonstrates how to use <tt>bit_and</tt> to take
+ *  the bitwise AND of one device_vector of \c ints by another.
+ *
+ *  \code
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/functional.h>
+ *  #include <thrust/sequence.h>
+ *  #include <thrust/fill.h>
+ *  #include <thrust/transform.h>
+ *  ...
+ *  const int N = 1000;
+ *  thrust::device_vector<int> V1(N);
+ *  thrust::device_vector<int> V2(N);
+ *  thrust::device_vector<int> V3(N);
+ *
+ *  thrust::sequence(V1.begin(), V1.end(), 1);
+ *  thrust::fill(V2.begin(), V2.end(), 13);
+ *
+ *  thrust::transform(V1.begin(), V1.end(), V2.begin(), V3.begin(),
+ *                    thrust::bit_and<int>());
+ *  // V3 is now {1&13, 2&13, 3&13, ..., 1000%13}
+ *  \endcode
+ *
+ *  \see binary_function
+ */
+template<typename T>
+  struct bit_and : public binary_function<T,T,T>
+{
+  /*! Function call operator. The return value is <tt>lhs & rhs</tt>.
+   */
+  __host__ __device__ T operator()(const T &lhs, const T &rhs) const {return lhs & rhs;}
+}; // end bit_and
+
+/*! \p bit_or is a function object. Specifically, it is an Adaptable Binary Function.
+ *  If \c f is an object of class <tt>bit_and<T></tt>, and \c x and \c y are objects
+ *  of class \c T, then <tt>f(x,y)</tt> returns <tt>x|y</tt>.
+ *
+ *  \tparam T is a model of <a href="http://www.sgi.com/tech/stl/Assignable.html">Assignable</a>,
+ *          and if \c x and \c y are objects of type \p T, then <tt>x|y</tt> must be defined and must have a return type that is convertible to \c T.
+ *
+ *  The following code snippet demonstrates how to use <tt>bit_or</tt> to take
+ *  the bitwise OR of one device_vector of \c ints by another.
+ *
+ *  \code
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/functional.h>
+ *  #include <thrust/sequence.h>
+ *  #include <thrust/fill.h>
+ *  #include <thrust/transform.h>
+ *  ...
+ *  const int N = 1000;
+ *  thrust::device_vector<int> V1(N);
+ *  thrust::device_vector<int> V2(N);
+ *  thrust::device_vector<int> V3(N);
+ *
+ *  thrust::sequence(V1.begin(), V1.end(), 1);
+ *  thrust::fill(V2.begin(), V2.end(), 13);
+ *
+ *  thrust::transform(V1.begin(), V1.end(), V2.begin(), V3.begin(),
+ *                    thrust::bit_or<int>());
+ *  // V3 is now {1|13, 2|13, 3|13, ..., 1000|13}
+ *  \endcode
+ *
+ *  \see binary_function
+ */
+template<typename T>
+  struct bit_or : public binary_function<T,T,T>
+{
+  /*! Function call operator. The return value is <tt>lhs | rhs</tt>.
+   */
+  __host__ __device__ T operator()(const T &lhs, const T &rhs) const {return lhs | rhs;}
+}; // end bit_or
+
+/*! \p bit_xor is a function object. Specifically, it is an Adaptable Binary Function.
+ *  If \c f is an object of class <tt>bit_and<T></tt>, and \c x and \c y are objects
+ *  of class \c T, then <tt>f(x,y)</tt> returns <tt>x^y</tt>.
+ *
+ *  \tparam T is a model of <a href="http://www.sgi.com/tech/stl/Assignable.html">Assignable</a>,
+ *          and if \c x and \c y are objects of type \p T, then <tt>x^y</tt> must be defined and must have a return type that is convertible to \c T.
+ *
+ *  The following code snippet demonstrates how to use <tt>bit_xor</tt> to take
+ *  the bitwise XOR of one device_vector of \c ints by another.
+ *
+ *  \code
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/functional.h>
+ *  #include <thrust/sequence.h>
+ *  #include <thrust/fill.h>
+ *  #include <thrust/transform.h>
+ *  ...
+ *  const int N = 1000;
+ *  thrust::device_vector<int> V1(N);
+ *  thrust::device_vector<int> V2(N);
+ *  thrust::device_vector<int> V3(N);
+ *
+ *  thrust::sequence(V1.begin(), V1.end(), 1);
+ *  thrust::fill(V2.begin(), V2.end(), 13);
+ *
+ *  thrust::transform(V1.begin(), V1.end(), V2.begin(), V3.begin(),
+ *                    thrust::bit_xor<int>());
+ *  // V3 is now {1^13, 2^13, 3^13, ..., 1000^13}
+ *  \endcode
+ *
+ *  \see binary_function
+ */
+template<typename T>
+  struct bit_xor : public binary_function<T,T,T>
+{
+  /*! Function call operator. The return value is <tt>lhs ^ rhs</tt>.
+   */
+  __host__ __device__ T operator()(const T &lhs, const T &rhs) const {return lhs ^ rhs;}
+}; // end bit_xor
+
+/*! \}
+ */
+
+/*! \addtogroup generalized_identity_operations Generalized Identity Operations
+ *  \ingroup predefined_function_objects
+ *  \{
+ */
+
+/*! \p identity is a Unary Function that represents the identity function: it takes
+ *  a single argument \c x, and returns \c x.
+ *
+ *  \tparam T No requirements on \p T.
+ *
+ *  The following code snippet demonstrates that \p identity returns its
+ *  argument.
+ *
+ *  \code
+ *  #include <thrust/functional.h>
+ *  #include <assert.h>
+ *  ...
+ *  int x = 137;
+ *  thrust::identity<int> id;
+ *  assert(x == id(x));
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/identity.html
+ *  \see unary_function
+ */
+template<typename T>
+  struct identity : public unary_function<T,T>
+{
+  /*! Function call operator. The return value is <tt>x</tt>.
+   */
+  __host__ __device__ const T &operator()(const T &x) const {return x;}
+}; // end identity
+
+/*! \p maximum is a function object that takes two arguments and returns the greater
+ *  of the two. Specifically, it is an Adaptable Binary Function. If \c f is an
+ *  object of class <tt>maximum<T></tt> and \c x and \c y are objects of class \c T
+ *  <tt>f(x,y)</tt> returns \c x if <tt>x > y</tt> and \c y, otherwise.
+ *
+ *  \tparam T is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable.html">LessThan Comparable</a>.
+ *
+ *  The following code snippet demonstrates that \p maximum returns its
+ *  greater argument.
+ *
+ *  \code
+ *  #include <thrust/functional.h>
+ *  #include <assert.h>
+ *  ...
+ *  int x =  137;
+ *  int y = -137;
+ *  thrust::maximum<int> mx;
+ *  assert(x == mx(x,y));
+ *  \endcode
+ *
+ *  \see minimum
+ *  \see min
+ *  \see binary_function
+ */
+template<typename T>
+  struct maximum : public binary_function<T,T,T>
+{
+  /*! Function call operator. The return value is <tt>rhs < lhs ? lhs : rhs</tt>.
+   */
+  __host__ __device__ T operator()(const T &lhs, const T &rhs) const {return lhs < rhs ? rhs : lhs;}
+}; // end maximum
+
+/*! \p minimum is a function object that takes two arguments and returns the lesser
+ *  of the two. Specifically, it is an Adaptable Binary Function. If \c f is an
+ *  object of class <tt>minimum<T></tt> and \c x and \c y are objects of class \c T
+ *  <tt>f(x,y)</tt> returns \c x if <tt>x < y</tt> and \c y, otherwise.
+ *
+ *  \tparam T is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable.html">LessThan Comparable</a>.
+ *
+ *  The following code snippet demonstrates that \p minimum returns its
+ *  lesser argument.
+ *
+ *  \code
+ *  #include <thrust/functional.h>
+ *  #include <assert.h>
+ *  ...
+ *  int x =  137;
+ *  int y = -137;
+ *  thrust::minimum<int> mn;
+ *  assert(y == mn(x,y));
+ *  \endcode
+ *
+ *  \see maximum
+ *  \see max
+ *  \see binary_function
+ */
+template<typename T>
+  struct minimum : public binary_function<T,T,T>
+{
+  /*! Function call operator. The return value is <tt>lhs < rhs ? lhs : rhs</tt>.
+   */
+  __host__ __device__ T operator()(const T &lhs, const T &rhs) const {return lhs < rhs ? lhs : rhs;}
+}; // end minimum
+
+/*! \p project1st is a function object that takes two arguments and returns 
+ *  its first argument; the second argument is unused. It is essentially a
+ *  generalization of identity to the case of a Binary Function.
+ *
+ *  \code
+ *  #include <thrust/functional.h>
+ *  #include <assert.h>
+ *  ...
+ *  int x =  137;
+ *  int y = -137;
+ *  thrust::project1st<int> pj1;
+ *  assert(x == pj1(x,y));
+ *  \endcode
+ *
+ *  \see identity
+ *  \see project2nd
+ *  \see binary_function
+ */
+template<typename T1, typename T2>
+  struct project1st : public binary_function<T1,T2,T1>
+{
+  /*! Function call operator. The return value is <tt>lhs</tt>.
+   */
+  __host__ __device__ const T1 &operator()(const T1 &lhs, const T2 &rhs) const {return lhs;}
+}; // end project1st
+
+/*! \p project2nd is a function object that takes two arguments and returns 
+ *  its second argument; the first argument is unused. It is essentially a
+ *  generalization of identity to the case of a Binary Function.
+ *
+ *  \code
+ *  #include <thrust/functional.h>
+ *  #include <assert.h>
+ *  ...
+ *  int x =  137;
+ *  int y = -137;
+ *  thrust::project2nd<int> pj2;
+ *  assert(y == pj2(x,y));
+ *  \endcode
+ *
+ *  \see identity
+ *  \see project1st
+ *  \see binary_function
+ */
+template<typename T1, typename T2>
+  struct project2nd : public binary_function<T1,T2,T2>
+{
+  /*! Function call operator. The return value is <tt>rhs</tt>.
+   */
+  __host__ __device__ const T2 &operator()(const T1 &lhs, const T2 &rhs) const {return rhs;}
+}; // end project2nd
+
+/*! \}
+ */
+
+
+// odds and ends
+
+/*! \addtogroup function_object_adaptors
+ *  \{
+ */
+
+/*! \p unary_negate is a function object adaptor: it is an Adaptable Predicate
+ *  that represents the logical negation of some other Adaptable Predicate.
+ *  That is: if \c f is an object of class <tt>unary_negate<AdaptablePredicate></tt>,
+ *  then there exists an object \c pred of class \c AdaptablePredicate such
+ *  that <tt>f(x)</tt> always returns the same value as <tt>!pred(x)</tt>.
+ *  There is rarely any reason to construct a <tt>unary_negate</tt> directly;
+ *  it is almost always easier to use the helper function not1.
+ *
+ *  \see http://www.sgi.com/tech/stl/unary_negate.html
+ *  \see not1
+ */
+template<typename Predicate>
+struct unary_negate 
+    : public thrust::unary_function<typename Predicate::argument_type, bool>
+{
+  /*! Constructor takes a \p Predicate object to negate.
+   *  \param p The \p Predicate object to negate.
+   */
+  __host__ __device__
+  explicit unary_negate(Predicate p) : pred(p){}
+
+  /*! Function call operator. The return value is <tt>!pred(x)</tt>.
+   */
+  __host__ __device__
+  bool operator()(const typename Predicate::argument_type& x) { return !pred(x); }
+
+  /*! \cond */
+  Predicate pred;
+  /*! \endcond */
+}; // end unary_negate
+
+/*! \p not1 is a helper function to simplify the creation of Adaptable Predicates:
+ *  it takes an Adaptable Predicate \p pred as an argument and returns a new Adaptable
+ *  Predicate that represents the negation of \p pred. That is: if \c pred is an object
+ *  of a type which models Adaptable Predicate, then the the type of the result
+ *  \c npred of <tt>not1(pred)</tt> is also a model of Adaptable Predicate and
+ *  <tt>npred(x)</tt> always returns the same value as <tt>!pred(x)</tt>.
+ *
+ *  \param pred The Adaptable Predicate to negate.
+ *  \return A new object, <tt>npred</tt> such that <tt>npred(x)</tt> always returns
+ *          the same value as <tt>!pred(x)</tt>.
+ *
+ *  \tparam Predicate is a model of <a href="http://www.sgi.com/tech/stl/AdaptablePredicate.html">Adaptable Predicate</a>.
+ *
+ *  \see unary_negate
+ *  \see not2
+ */
+template<typename Predicate>
+  __host__ __device__
+  unary_negate<Predicate> not1(const Predicate &pred);
+
+/*! \p binary_negate is a function object adaptor: it is an Adaptable Binary 
+ *  Predicate that represents the logical negation of some other Adaptable
+ *  Binary Predicate. That is: if \c f is an object of class <tt>binary_negate<AdaptablePredicate></tt>,
+ *  then there exists an object \c pred of class \c AdaptableBinaryPredicate
+ *  such that <tt>f(x,y)</tt> always returns the same value as <tt>!pred(x,y)</tt>.
+ *  There is rarely any reason to construct a <tt>binary_negate</tt> directly;
+ *  it is almost always easier to use the helper function not2.
+ *
+ *  \see http://www.sgi.com/tech/stl/binary_negate.html
+ */
+template<typename Predicate>
+struct binary_negate
+    : public thrust::binary_function<typename Predicate::first_argument_type,
+                                     typename Predicate::second_argument_type,
+                                     bool>
+{
+  /*! Constructor takes a \p Predicate object to negate.
+   *  \param p The \p Predicate object to negate.
+   */
+  __host__ __device__
+  explicit binary_negate(Predicate p) : pred(p){}
+
+  /*! Function call operator. The return value is <tt>!pred(x,y)</tt>.
+   */
+  __host__ __device__
+  bool operator()(const typename Predicate::first_argument_type& x, const typename Predicate::second_argument_type& y)
+  { 
+      return !pred(x,y); 
+  }
+
+  /*! \cond */
+  Predicate pred;
+  /*! \endcond */
+}; // end binary_negate
+
+/*! \p not2 is a helper function to simplify the creation of Adaptable Binary Predicates:
+ *  it takes an Adaptable Binary Predicate \p pred as an argument and returns a new Adaptable
+ *  Binary Predicate that represents the negation of \p pred. That is: if \c pred is an object
+ *  of a type which models Adaptable Binary Predicate, then the the type of the result
+ *  \c npred of <tt>not2(pred)</tt> is also a model of Adaptable Binary Predicate and
+ *  <tt>npred(x,y)</tt> always returns the same value as <tt>!pred(x,y)</tt>.
+ *
+ *  \param pred The Adaptable Binary Predicate to negate.
+ *  \return A new object, <tt>npred</tt> such that <tt>npred(x,y)</tt> always returns
+ *          the same value as <tt>!pred(x,y)</tt>.
+ *
+ *  \tparam Binary Predicate is a model of <a href="http://www.sgi.com/tech/stl/AdaptableBinaryPredicate.html">Adaptable Binary Predicate</a>.
+ *
+ *  \see binary_negate
+ *  \see not1
+ */
+template<typename BinaryPredicate>
+  __host__ __device__
+  binary_negate<BinaryPredicate> not2(const BinaryPredicate &pred);
+
+/*! \}
+ */
+
+
+/*! \addtogroup placeholder_objects Placeholder Objects
+ *  \ingroup function_objects
+ *  \{
+ */
+
+
+/*! \namespace placeholders
+ *  \brief Facilities for constructing simple functions inline.
+ *
+ *  Objects in the \p thrust::placeholders namespace may be used to create simple arithmetic functions inline
+ *  in an algorithm invocation. Combining placeholders such as \p _1 and \p _2 with arithmetic operations such as \c +
+ *  creates an unnamed function object which applies the operation to their arguments.
+ *
+ *  The type of placeholder objects is implementation-defined.
+ *
+ *  The following code snippet demonstrates how to use the placeholders \p _1 and \p _2 with \p thrust::transform
+ *  to implement the SAXPY computation:
+ *
+ *  \code
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/transform.h>
+ *  #include <thrust/functional.h>
+ *
+ *  int main()
+ *  {
+ *    thrust::device_vector<float> x(4), y(4);
+ *    x[0] = 1;
+ *    x[1] = 2;
+ *    x[2] = 3;
+ *    x[3] = 4;
+ *    
+ *    y[0] = 1;
+ *    y[1] = 1;
+ *    y[2] = 1;
+ *    y[3] = 1;
+ *
+ *    float a = 2.0f;
+ *
+ *    using namespace thrust::placeholders;
+ *
+ *    thrust::transform(x.begin(), x.end(), y.begin(), y.begin(),
+ *      a * _1 + 2
+ *    );
+ *
+ *    // y is now {3, 5, 7, 9}
+ *  }
+ *  \endcode
+ */
+namespace placeholders
+{
+
+
+/*! \p thrust::placeholders::_1 is the placeholder for the first function parameter.
+ */
+static const thrust::detail::functional::placeholder<0>::type _1;
+
+
+/*! \p thrust::placeholders::_2 is the placeholder for the second function parameter.
+ */
+static const thrust::detail::functional::placeholder<1>::type _2;
+
+
+/*! \p thrust::placeholders::_3 is the placeholder for the third function parameter.
+ */
+static const thrust::detail::functional::placeholder<2>::type _3;
+
+
+/*! \p thrust::placeholders::_4 is the placeholder for the fourth function parameter.
+ */
+static const thrust::detail::functional::placeholder<3>::type _4;
+
+
+/*! \p thrust::placeholders::_5 is the placeholder for the fifth function parameter.
+ */
+static const thrust::detail::functional::placeholder<4>::type _5;
+
+
+/*! \p thrust::placeholders::_6 is the placeholder for the sixth function parameter.
+ */
+static const thrust::detail::functional::placeholder<5>::type _6;
+
+
+/*! \p thrust::placeholders::_7 is the placeholder for the seventh function parameter.
+ */
+static const thrust::detail::functional::placeholder<6>::type _7;
+
+
+/*! \p thrust::placeholders::_8 is the placeholder for the eighth function parameter.
+ */
+static const thrust::detail::functional::placeholder<7>::type _8;
+
+
+/*! \p thrust::placeholders::_9 is the placeholder for the ninth function parameter.
+ */
+static const thrust::detail::functional::placeholder<8>::type _9;
+
+
+/*! \p thrust::placeholders::_10 is the placeholder for the tenth function parameter.
+ */
+static const thrust::detail::functional::placeholder<9>::type _10;
+
+
+} // end placeholders
+
+
+/*! \} // placeholder_objects
+ */
+
+
+} // end thrust
+
+#include <thrust/detail/functional.inl>
+#include <thrust/detail/functional/operators.h>
+
diff --git a/compat/thrust/gather.h b/compat/thrust/gather.h
new file mode 100644
index 0000000000..f2b8233657
--- /dev/null
+++ b/compat/thrust/gather.h
@@ -0,0 +1,438 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file gather.h
+ *  \brief Irregular copying from a source range
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/execution_policy.h>
+
+namespace thrust
+{
+
+
+/*! \addtogroup gathering
+ *  \ingroup copying
+ *  \{
+ */
+
+
+/*! \p gather copies elements from a source array into a destination range according 
+ *  to a map. For each input iterator \c i in the range <tt>[map_first, map_last)</tt>, the
+ *  value <tt>input_first[\*i]</tt> is assigned to <tt>*(result + (i - map_first))</tt>.
+ *  \p RandomAccessIterator must permit random access.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param map_first Beginning of the range of gather locations.
+ *  \param map_last End of the range of gather locations.
+ *  \param input_first Beginning of the source range.
+ *  \param result Beginning of the destination range.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator must be a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a> and \c InputIterator's \c value_type must be convertible to \c RandomAccessIterator's \c difference_type.
+ *  \tparam RandomAccessIterator must be a model of <a href="http://www.sgi.com/tech/stl/RandomAccessIterator.html">Random Access Iterator</a> and \c RandomAccessIterator's \c value_type must be convertible to \c OutputIterator's \c value_type.
+ *  \tparam OutputIterator must be a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *
+ *  \pre The range <tt>[map_first, map_last)</tt> shall not overlap the range <tt>[result, result + (map_last - map_first))</tt>.
+ *
+ *  \remark \p gather is the inverse of thrust::scatter.
+ *
+ *  The following code snippet demonstrates how to use \p gather to reorder
+ *  a range using the \p thrust::device execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/gather.h>
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  // mark even indices with a 1; odd indices with a 0
+ *  int values[10] = {1, 0, 1, 0, 1, 0, 1, 0, 1, 0};
+ *  thrust::device_vector<int> d_values(values, values + 10);
+ *
+ *  // gather all even indices into the first half of the range
+ *  // and odd indices to the last half of the range
+ *  int map[10]   = {0, 2, 4, 6, 8, 1, 3, 5, 7, 9};
+ *  thrust::device_vector<int> d_map(map, map + 10);
+ *
+ *  thrust::device_vector<int> d_output(10);
+ *  thrust::gather(thrust::device,
+ *                 d_map.begin(), d_map.end(),
+ *                 d_values.begin(),
+ *                 d_output.begin());
+ *  // d_output is now {1, 1, 1, 1, 1, 0, 0, 0, 0, 0}
+ *  \endcode
+ */
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename RandomAccessIterator,
+         typename OutputIterator>
+  OutputIterator gather(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                        InputIterator                                               map_first,
+                        InputIterator                                               map_last,
+                        RandomAccessIterator                                        input_first,
+                        OutputIterator                                              result);
+
+
+/*! \p gather copies elements from a source array into a destination range according 
+ *  to a map. For each input iterator \c i in the range <tt>[map_first, map_last)</tt>, the
+ *  value <tt>input_first[\*i]</tt> is assigned to <tt>*(result + (i - map_first))</tt>.
+ *  \p RandomAccessIterator must permit random access.
+ *
+ *  \param map_first Beginning of the range of gather locations.
+ *  \param map_last End of the range of gather locations.
+ *  \param input_first Beginning of the source range.
+ *  \param result Beginning of the destination range.
+ *
+ *  \tparam InputIterator must be a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a> and \c InputIterator's \c value_type must be convertible to \c RandomAccessIterator's \c difference_type.
+ *  \tparam RandomAccessIterator must be a model of <a href="http://www.sgi.com/tech/stl/RandomAccessIterator.html">Random Access Iterator</a> and \c RandomAccessIterator's \c value_type must be convertible to \c OutputIterator's \c value_type.
+ *  \tparam OutputIterator must be a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *
+ *  \pre The range <tt>[map_first, map_last)</tt> shall not overlap the range <tt>[result, result + (map_last - map_first))</tt>.
+ *
+ *  \remark \p gather is the inverse of thrust::scatter.
+ *
+ *  The following code snippet demonstrates how to use \p gather to reorder
+ *  a range.
+ *
+ *  \code
+ *  #include <thrust/gather.h>
+ *  #include <thrust/device_vector.h>
+ *  ...
+ *  // mark even indices with a 1; odd indices with a 0
+ *  int values[10] = {1, 0, 1, 0, 1, 0, 1, 0, 1, 0};
+ *  thrust::device_vector<int> d_values(values, values + 10);
+ *
+ *  // gather all even indices into the first half of the range
+ *  // and odd indices to the last half of the range
+ *  int map[10]   = {0, 2, 4, 6, 8, 1, 3, 5, 7, 9};
+ *  thrust::device_vector<int> d_map(map, map + 10);
+ *
+ *  thrust::device_vector<int> d_output(10);
+ *  thrust::gather(d_map.begin(), d_map.end(),
+ *                 d_values.begin(),
+ *                 d_output.begin());
+ *  // d_output is now {1, 1, 1, 1, 1, 0, 0, 0, 0, 0}
+ *  \endcode
+ */
+template<typename InputIterator,
+         typename RandomAccessIterator,
+         typename OutputIterator>
+  OutputIterator gather(InputIterator        map_first,
+                        InputIterator        map_last,
+                        RandomAccessIterator input_first,
+                        OutputIterator       result);
+
+
+/*! \p gather_if conditionally copies elements from a source array into a destination 
+ *  range according to a map. For each input iterator \c i in the range <tt>[map_first, map_last)</tt>,
+ *  such that the value of <tt>\*(stencil + (i - map_first))</tt> is \c true, the value
+ *  <tt>input_first[\*i]</tt> is assigned to <tt>*(result + (i - map_first))</tt>.
+ *  \p RandomAccessIterator must permit random access.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param map_first Beginning of the range of gather locations.
+ *  \param map_last End of the range of gather locations.
+ *  \param stencil Beginning of the range of predicate values.
+ *  \param input_first Beginning of the source range.
+ *  \param result Beginning of the destination range.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator1 must be a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a> and \c InputIterator1's \c value_type must be convertible to \c RandomAccessIterator's \c difference_type.
+ *  \tparam InputIterator2 must be a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a> and \c InputIterator2's \c value_type must be convertible to \c bool.
+ *  \tparam RandomAccessIterator must be a model of <a href="http://www.sgi.com/tech/stl/RandomAccessIterator.html">Random Access iterator</a> and \c RandomAccessIterator's \c value_type must be convertible to \c OutputIterator's \c value_type.
+ *  \tparam OutputIterator must be a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *
+ *  \pre The range <tt>[map_first, map_last)</tt> shall not overlap the range <tt>[result, result + (map_last - map_first))</tt>.
+ *  \pre The range <tt>[stencil, stencil + (map_last - map_first))</tt> shall not overlap the range <tt>[result, result + (map_last - map_first))</tt>.
+ *
+ *  \remark \p gather_if is the inverse of \p scatter_if.
+ *
+ *  The following code snippet demonstrates how to use \p gather_if to gather selected values from
+ *  an input range using the \p thrust::device execution policy:
+ *
+ *  \code
+ *  #include <thrust/gather.h>
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *
+ *  int values[10] = {0, 1, 2, 3, 4, 5, 6, 7, 8, 9};
+ *  thrust::device_vector<int> d_values(values, values + 10);
+ *
+ *  // select elements at even-indexed locations
+ *  int stencil[10] = {1, 0, 1, 0, 1, 0, 1, 0, 1, 0};
+ *  thrust::device_vector<int> d_stencil(stencil, stencil + 10);
+ *
+ *  // map all even indices into the first half of the range
+ *  // and odd indices to the last half of the range
+ *  int map[10]   = {0, 2, 4, 6, 8, 1, 3, 5, 7, 9};
+ *  thrust::device_vector<int> d_map(map, map + 10);
+ *
+ *  thrust::device_vector<int> d_output(10, 7);
+ *  thrust::gather_if(thrust::device,
+ *                    d_map.begin(), d_map.end(),
+ *                    d_stencil.begin(),
+ *                    d_values.begin(),
+ *                    d_output.begin());
+ *  // d_output is now {0, 7, 4, 7, 8, 7, 3, 7, 7, 7}
+ *  \endcode
+ */
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename RandomAccessIterator,
+         typename OutputIterator>
+  OutputIterator gather_if(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                           InputIterator1                                              map_first,
+                           InputIterator1                                              map_last,
+                           InputIterator2                                              stencil,
+                           RandomAccessIterator                                        input_first,
+                           OutputIterator                                              result);
+
+
+/*! \p gather_if conditionally copies elements from a source array into a destination 
+ *  range according to a map. For each input iterator \c i in the range <tt>[map_first, map_last)</tt>,
+ *  such that the value of <tt>\*(stencil + (i - map_first))</tt> is \c true, the value
+ *  <tt>input_first[\*i]</tt> is assigned to <tt>*(result + (i - map_first))</tt>.
+ *  \p RandomAccessIterator must permit random access.
+ *
+ *  \param map_first Beginning of the range of gather locations.
+ *  \param map_last End of the range of gather locations.
+ *  \param stencil Beginning of the range of predicate values.
+ *  \param input_first Beginning of the source range.
+ *  \param result Beginning of the destination range.
+ *
+ *  \tparam InputIterator1 must be a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a> and \c InputIterator1's \c value_type must be convertible to \c RandomAccessIterator's \c difference_type.
+ *  \tparam InputIterator2 must be a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a> and \c InputIterator2's \c value_type must be convertible to \c bool.
+ *  \tparam RandomAccessIterator must be a model of <a href="http://www.sgi.com/tech/stl/RandomAccessIterator.html">Random Access iterator</a> and \c RandomAccessIterator's \c value_type must be convertible to \c OutputIterator's \c value_type.
+ *  \tparam OutputIterator must be a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *
+ *  \pre The range <tt>[map_first, map_last)</tt> shall not overlap the range <tt>[result, result + (map_last - map_first))</tt>.
+ *  \pre The range <tt>[stencil, stencil + (map_last - map_first))</tt> shall not overlap the range <tt>[result, result + (map_last - map_first))</tt>.
+ *
+ *  \remark \p gather_if is the inverse of \p scatter_if.
+ *
+ *  The following code snippet demonstrates how to use \p gather_if to gather selected values from
+ *  an input range.
+ *
+ *  \code
+ *  #include <thrust/gather.h>
+ *  #include <thrust/device_vector.h>
+ *  ...
+ *
+ *  int values[10] = {0, 1, 2, 3, 4, 5, 6, 7, 8, 9};
+ *  thrust::device_vector<int> d_values(values, values + 10);
+ *
+ *  // select elements at even-indexed locations
+ *  int stencil[10] = {1, 0, 1, 0, 1, 0, 1, 0, 1, 0};
+ *  thrust::device_vector<int> d_stencil(stencil, stencil + 10);
+ *
+ *  // map all even indices into the first half of the range
+ *  // and odd indices to the last half of the range
+ *  int map[10]   = {0, 2, 4, 6, 8, 1, 3, 5, 7, 9};
+ *  thrust::device_vector<int> d_map(map, map + 10);
+ *
+ *  thrust::device_vector<int> d_output(10, 7);
+ *  thrust::gather_if(d_map.begin(), d_map.end(),
+ *                    d_stencil.begin(),
+ *                    d_values.begin(),
+ *                    d_output.begin());
+ *  // d_output is now {0, 7, 4, 7, 8, 7, 3, 7, 7, 7}
+ *  \endcode
+ */
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename RandomAccessIterator,
+         typename OutputIterator>
+  OutputIterator gather_if(InputIterator1       map_first,
+                           InputIterator1       map_last,
+                           InputIterator2       stencil,
+                           RandomAccessIterator input_first,
+                           OutputIterator       result);
+
+
+/*! \p gather_if conditionally copies elements from a source array into a destination 
+ *  range according to a map. For each input iterator \c i in the range <tt>[map_first, map_last)</tt>
+ *  such that the value of <tt>pred(\*(stencil + (i - map_first)))</tt> is \c true,
+ *  the value <tt>input_first[\*i]</tt> is assigned to <tt>*(result + (i - map_first))</tt>.
+ *  \p RandomAccessIterator must permit random access.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param map_first Beginning of the range of gather locations.
+ *  \param map_last End of the range of gather locations.
+ *  \param stencil Beginning of the range of predicate values.
+ *  \param input_first Beginning of the source range.
+ *  \param result Beginning of the destination range.
+ *  \param pred Predicate to apply to the stencil values.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator1 must be a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a> and \c InputIterator1's \c value_type must be convertible to \c RandomAccessIterator's \c difference_type.
+ *  \tparam InputIterator2 must be a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a> and \c InputIterator2's \c value_type must be convertible to \c Predicate's \c argument_type.
+ *  \tparam RandomAccessIterator must be a model of <a href="http://www.sgi.com/tech/stl/RandomAccessIterator.html">Random Access iterator</a> and \c RandomAccessIterator's \c value_type must be convertible to \c OutputIterator's \c value_type.
+ *  \tparam OutputIterator must be a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam Predicate must be a model of <a href="http://www.sgi.com/tech/stl/Predicate.html">Predicate</a>.
+ *
+ *  \pre The range <tt>[map_first, map_last)</tt> shall not overlap the range <tt>[result, result + (map_last - map_first))</tt>.
+ *  \pre The range <tt>[stencil, stencil + (map_last - map_first))</tt> shall not overlap the range <tt>[result, result + (map_last - map_first))</tt>.
+ *
+ *  \remark \p gather_if is the inverse of \p scatter_if.
+ *
+ *  The following code snippet demonstrates how to use \p gather_if to gather selected values from
+ *  an input range based on an arbitrary selection function using the \p thrust::device execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/gather.h>
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/execution_policy.h>
+ *  
+ *  struct is_even
+ *  {
+ *    __host__ __device__
+ *    bool operator()(const int x)
+ *    {
+ *      return (x % 2) == 0;
+ *    }
+ *  };
+ *  ...
+ *
+ *  int values[10] = {0, 1, 2, 3, 4, 5, 6, 7, 8, 9};
+ *  thrust::device_vector<int> d_values(values, values + 10);
+ *
+ *  // we will select an element when our stencil is even
+ *  int stencil[10] = {0, 3, 4, 1, 4, 1, 2, 7, 8, 9};
+ *  thrust::device_vector<int> d_stencil(stencil, stencil + 10);
+ *
+ *  // map all even indices into the first half of the range
+ *  // and odd indices to the last half of the range
+ *  int map[10]   = {0, 2, 4, 6, 8, 1, 3, 5, 7, 9};
+ *  thrust::device_vector<int> d_map(map, map + 10);
+ *
+ *  thrust::device_vector<int> d_output(10, 7);
+ *  thrust::gather_if(thrust::device,
+ *                    d_map.begin(), d_map.end(),
+ *                    d_stencil.begin(),
+ *                    d_values.begin(),
+ *                    d_output.begin(),
+ *                    is_even());
+ *  // d_output is now {0, 7, 4, 7, 8, 7, 3, 7, 7, 7}
+ *  \endcode
+ */
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename RandomAccessIterator,
+         typename OutputIterator,
+         typename Predicate>
+  OutputIterator gather_if(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                           InputIterator1                                              map_first,
+                           InputIterator1                                              map_last,
+                           InputIterator2                                              stencil,
+                           RandomAccessIterator                                        input_first,
+                           OutputIterator                                              result,
+                           Predicate                                                   pred);
+
+
+/*! \p gather_if conditionally copies elements from a source array into a destination 
+ *  range according to a map. For each input iterator \c i in the range <tt>[map_first, map_last)</tt>
+ *  such that the value of <tt>pred(\*(stencil + (i - map_first)))</tt> is \c true,
+ *  the value <tt>input_first[\*i]</tt> is assigned to <tt>*(result + (i - map_first))</tt>.
+ *  \p RandomAccessIterator must permit random access.
+ *
+ *  \param map_first Beginning of the range of gather locations.
+ *  \param map_last End of the range of gather locations.
+ *  \param stencil Beginning of the range of predicate values.
+ *  \param input_first Beginning of the source range.
+ *  \param result Beginning of the destination range.
+ *  \param pred Predicate to apply to the stencil values.
+ *
+ *  \tparam InputIterator1 must be a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a> and \c InputIterator1's \c value_type must be convertible to \c RandomAccessIterator's \c difference_type.
+ *  \tparam InputIterator2 must be a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a> and \c InputIterator2's \c value_type must be convertible to \c Predicate's \c argument_type.
+ *  \tparam RandomAccessIterator must be a model of <a href="http://www.sgi.com/tech/stl/RandomAccessIterator.html">Random Access iterator</a> and \c RandomAccessIterator's \c value_type must be convertible to \c OutputIterator's \c value_type.
+ *  \tparam OutputIterator must be a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam Predicate must be a model of <a href="http://www.sgi.com/tech/stl/Predicate.html">Predicate</a>.
+ *
+ *  \pre The range <tt>[map_first, map_last)</tt> shall not overlap the range <tt>[result, result + (map_last - map_first))</tt>.
+ *  \pre The range <tt>[stencil, stencil + (map_last - map_first))</tt> shall not overlap the range <tt>[result, result + (map_last - map_first))</tt>.
+ *
+ *  \remark \p gather_if is the inverse of \p scatter_if.
+ *
+ *  The following code snippet demonstrates how to use \p gather_if to gather selected values from
+ *  an input range based on an arbitrary selection function.
+ *
+ *  \code
+ *  #include <thrust/gather.h>
+ *  #include <thrust/device_vector.h>
+ *  
+ *  struct is_even
+ *  {
+ *    __host__ __device__
+ *    bool operator()(const int x)
+ *    {
+ *      return (x % 2) == 0;
+ *    }
+ *  };
+ *  ...
+ *
+ *  int values[10] = {0, 1, 2, 3, 4, 5, 6, 7, 8, 9};
+ *  thrust::device_vector<int> d_values(values, values + 10);
+ *
+ *  // we will select an element when our stencil is even
+ *  int stencil[10] = {0, 3, 4, 1, 4, 1, 2, 7, 8, 9};
+ *  thrust::device_vector<int> d_stencil(stencil, stencil + 10);
+ *
+ *  // map all even indices into the first half of the range
+ *  // and odd indices to the last half of the range
+ *  int map[10]   = {0, 2, 4, 6, 8, 1, 3, 5, 7, 9};
+ *  thrust::device_vector<int> d_map(map, map + 10);
+ *
+ *  thrust::device_vector<int> d_output(10, 7);
+ *  thrust::gather_if(d_map.begin(), d_map.end(),
+ *                    d_stencil.begin(),
+ *                    d_values.begin(),
+ *                    d_output.begin(),
+ *                    is_even());
+ *  // d_output is now {0, 7, 4, 7, 8, 7, 3, 7, 7, 7}
+ *  \endcode
+ */
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename RandomAccessIterator,
+         typename OutputIterator,
+         typename Predicate>
+  OutputIterator gather_if(InputIterator1       map_first,
+                           InputIterator1       map_last,
+                           InputIterator2       stencil,
+                           RandomAccessIterator input_first,
+                           OutputIterator       result,
+                           Predicate            pred);
+
+/*! \} // gathering
+ */
+
+} // end namespace thrust
+
+#include <thrust/detail/gather.inl>
+
diff --git a/compat/thrust/generate.h b/compat/thrust/generate.h
new file mode 100644
index 0000000000..1d52721a78
--- /dev/null
+++ b/compat/thrust/generate.h
@@ -0,0 +1,211 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file generate.h
+ *  \brief Fills a range with values "generated" from a function of no arguments
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/execution_policy.h>
+
+namespace thrust
+{
+
+
+/*! \addtogroup transformations
+ *  \{
+ */
+
+
+/*! \p generate assigns the result of invoking \p gen, a function object that takes no arguments,
+ *  to each element in the range <tt>[first,last)</tt>.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The first element in the range of interest.
+ *  \param last The last element in the range of interest.
+ *  \param gen A function argument, taking no parameters, used to generate values to assign to
+ *             elements in the range <tt>[first,last)</tt>.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>,
+ *          and \p ForwardIterator is mutable.
+ *  \tparam Generator is a model of <a href="http://www.sgi.com/tech/stl/Generator.html">Generator</a>,
+ *          and \p Generator's \c result_type is convertible to \p ForwardIterator's \c value_type.
+ *
+ *  The following code snippet demonstrates how to fill a \c host_vector with random numbers,
+ *  using the standard C library function \c rand using the \p thrust::host execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/generate.h>
+ *  #include <thrust/host_vector.h>
+ *  #include <thrust/execution_policy.h>
+ *  #include <cstdlib>
+ *  ...
+ *  thrust::host_vector<int> v(10);
+ *  srand(13);
+ *  thrust::generate(thrust::host, v.begin(), v.end(), rand);
+ *
+ *  // the elements of v are now pseudo-random numbers
+ *  \endcode
+ *
+ *  \see generate_n
+ *  \see http://www.sgi.com/tech/stl/generate.html
+ */
+template<typename DerivedPolicy,
+         typename ForwardIterator,
+         typename Generator>
+  void generate(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                ForwardIterator first,
+                ForwardIterator last,
+                Generator gen);
+
+
+/*! \p generate assigns the result of invoking \p gen, a function object that takes no arguments,
+ *  to each element in the range <tt>[first,last)</tt>.
+ *
+ *  \param first The first element in the range of interest.
+ *  \param last The last element in the range of interest.
+ *  \param gen A function argument, taking no parameters, used to generate values to assign to
+ *             elements in the range <tt>[first,last)</tt>.
+ *
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>,
+ *          and \p ForwardIterator is mutable.
+ *  \tparam Generator is a model of <a href="http://www.sgi.com/tech/stl/Generator.html">Generator</a>,
+ *          and \p Generator's \c result_type is convertible to \p ForwardIterator's \c value_type.
+ *
+ *  The following code snippet demonstrates how to fill a \c host_vector with random numbers,
+ *  using the standard C library function \c rand.
+ *
+ *  \code
+ *  #include <thrust/generate.h>
+ *  #include <thrust/host_vector.h>
+ *  #include <thrust/execution_policy.h>
+ *  #include <cstdlib>
+ *  ...
+ *  thrust::host_vector<int> v(10);
+ *  srand(13);
+ *  thrust::generate(v.begin(), v.end(), rand);
+ *
+ *  // the elements of v are now pseudo-random numbers
+ *  \endcode
+ *
+ *  \see generate_n
+ *  \see http://www.sgi.com/tech/stl/generate.html
+ */
+template<typename ForwardIterator,
+         typename Generator>
+  void generate(ForwardIterator first,
+                ForwardIterator last,
+                Generator gen);
+
+
+/*! \p generate_n assigns the result of invoking \p gen, a function object that takes no arguments,
+ *  to each element in the range <tt>[first,first + n)</tt>. The return value is <tt>first + n</tt>.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The first element in the range of interest.
+ *  \param n The size of the range of interest.
+ *  \param gen A function argument, taking no parameters, used to generate values to assign to
+ *             elements in the range <tt>[first,first + n)</tt>.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam Size is an integral type (either signed or unsigned).
+ *  \tparam Generator is a model of <a href="http://www.sgi.com/tech/stl/Generator.html">Generator</a>,
+ *          and \p Generator's \c result_type is convertible to a type in \p OutputIterator's set of \c value_types.
+ *
+ *  The following code snippet demonstrates how to fill a \c host_vector with random numbers,
+ *  using the standard C library function \c rand using the \p thrust::host execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/generate.h>
+ *  #include <thrust/host_vector.h>
+ *  #include <thrust/execution_policy.h>
+ *  #include <cstdlib>
+ *  ...
+ *  thrust::host_vector<int> v(10);
+ *  srand(13);
+ *  thrust::generate_n(thrust::host, v.begin(), 10, rand);
+ *
+ *  // the elements of v are now pseudo-random numbers
+ *  \endcode
+ *
+ *  \see generate
+ *  \see http://www.sgi.com/tech/stl/generate.html
+ */
+template<typename DerivedPolicy,
+         typename OutputIterator,
+         typename Size,
+         typename Generator>
+  OutputIterator generate_n(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                            OutputIterator first,
+                            Size n,
+                            Generator gen);
+
+
+/*! \p generate_n assigns the result of invoking \p gen, a function object that takes no arguments,
+ *  to each element in the range <tt>[first,first + n)</tt>. The return value is <tt>first + n</tt>.
+ *
+ *  \param first The first element in the range of interest.
+ *  \param n The size of the range of interest.
+ *  \param gen A function argument, taking no parameters, used to generate values to assign to
+ *             elements in the range <tt>[first,first + n)</tt>.
+ *
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam Size is an integral type (either signed or unsigned).
+ *  \tparam Generator is a model of <a href="http://www.sgi.com/tech/stl/Generator.html">Generator</a>,
+ *          and \p Generator's \c result_type is convertible to a type in \p OutputIterator's set of \c value_types.
+ *
+ *  The following code snippet demonstrates how to fill a \c host_vector with random numbers,
+ *  using the standard C library function \c rand.
+ *
+ *  \code
+ *  #include <thrust/generate.h>
+ *  #include <thrust/host_vector.h>
+ *  #include <stdlib.h>
+ *  ...
+ *  thrust::host_vector<int> v(10);
+ *  srand(13);
+ *  thrust::generate_n(v.begin(), 10, rand);
+ *
+ *  // the elements of v are now pseudo-random numbers
+ *  \endcode
+ *
+ *  \see generate
+ *  \see http://www.sgi.com/tech/stl/generate.html
+ */
+template<typename OutputIterator,
+         typename Size,
+         typename Generator>
+  OutputIterator generate_n(OutputIterator first,
+                            Size n,
+                            Generator gen);
+
+
+/*! \} // end transformations
+ */
+
+} // end namespace thrust
+
+#include <thrust/detail/generate.inl>
+
diff --git a/compat/thrust/host_vector.h b/compat/thrust/host_vector.h
new file mode 100644
index 0000000000..11b1ae0685
--- /dev/null
+++ b/compat/thrust/host_vector.h
@@ -0,0 +1,424 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file host_vector.h
+ *  \brief A dynamically-sizable array of elements which reside in the "host" memory space
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <memory>
+#include <thrust/detail/vector_base.h>
+#include <vector>
+
+namespace thrust
+{
+
+// forward declaration of device_vector
+template<typename T, typename Alloc> class device_vector;
+
+/*! \addtogroup container_classes Container Classes
+ *  \addtogroup host_containers Host Containers
+ *  \ingroup container_classes
+ *  \{
+ */
+
+/*! A \p host_vector is a container that supports random access to elements,
+ *  constant time removal of elements at the end, and linear time insertion
+ *  and removal of elements at the beginning or in the middle. The number of
+ *  elements in a \p host_vector may vary dynamically; memory management is
+ *  automatic. The memory associated with a \p host_vector resides in the memory
+ *  space of the host associated with a parallel device.
+ *
+ *  \see http://www.sgi.com/tech/stl/Vector.html
+ *  \see device_vector
+ */
+template<typename T, typename Alloc = std::allocator<T> >
+  class host_vector
+    : public detail::vector_base<T,Alloc>
+{
+  private:
+    typedef detail::vector_base<T,Alloc> Parent;
+
+  public:
+    /*! \cond */
+    typedef typename Parent::size_type  size_type;
+    typedef typename Parent::value_type value_type;
+    /*! \endcond */
+
+    /*! This constructor creates an empty \p host_vector.
+     */
+    __host__
+    host_vector(void)
+      :Parent() {}
+
+    /*! This constructor creates a \p host_vector with the given
+     *  size.
+     *  \param n The number of elements to initially craete.
+     */
+    __host__
+    explicit host_vector(size_type n)
+      :Parent(n) {}
+
+    /*! This constructor creates a \p host_vector with copies
+     *  of an exemplar element.
+     *  \param n The number of elements to initially create.
+     *  \param value An element to copy.
+     */
+    __host__
+    explicit host_vector(size_type n, const value_type &value)
+      :Parent(n,value) {}
+
+    /*! Copy constructor copies from an exemplar \p host_vector.
+     *  \param v The \p host_vector to copy.
+     */
+    __host__
+    host_vector(const host_vector &v)
+      :Parent(v) {}
+
+    /*! Assign operator copies from an exemplar \p host_vector.
+     *  \param v The \p host_vector to copy.
+     */
+    __host__
+    host_vector &operator=(const host_vector &v)
+    { Parent::operator=(v); return *this; }
+
+    /*! Copy constructor copies from an exemplar \p host_vector with different type.
+     *  \param v The \p host_vector to copy.
+     */
+    template<typename OtherT, typename OtherAlloc>
+    __host__
+    host_vector(const host_vector<OtherT,OtherAlloc> &v)
+      :Parent(v) {}
+
+    /*! Assign operator copies from an exemplar \p host_vector with different type.
+     *  \param v The \p host_vector to copy.
+     */
+    template<typename OtherT, typename OtherAlloc>
+    __host__
+    host_vector &operator=(const host_vector<OtherT,OtherAlloc> &v)
+    { Parent::operator=(v); return *this; }
+
+    /*! Copy constructor copies from an exemplar <tt>std::vector</tt>.
+     *  \param v The <tt>std::vector</tt> to copy.
+     */
+    template<typename OtherT, typename OtherAlloc>
+    __host__
+    host_vector(const std::vector<OtherT,OtherAlloc> &v)
+      :Parent(v) {}
+
+    /*! Assign operator copies from an exemplar <tt>std::vector</tt>.
+     *  \param v The <tt>std::vector</tt> to copy.
+     */
+    template<typename OtherT, typename OtherAlloc>
+    __host__
+    host_vector &operator=(const std::vector<OtherT,OtherAlloc> &v)
+    { Parent::operator=(v); return *this;}
+
+    /*! Copy constructor copies from an exemplar \p device_vector with possibly different type.
+     *  \param v The \p device_vector to copy.
+     */
+    template<typename OtherT, typename OtherAlloc>
+    __host__
+    host_vector(const device_vector<OtherT,OtherAlloc> &v);
+
+    /*! Assign operator copies from an exemplar \p device_vector.
+     *  \param v The \p device_vector to copy.
+     */
+    template<typename OtherT, typename OtherAlloc>
+    __host__
+    host_vector &operator=(const device_vector<OtherT,OtherAlloc> &v)
+    { Parent::operator=(v); return *this; }
+
+    /*! This constructor builds a \p host_vector from a range.
+     *  \param first The beginning of the range.
+     *  \param last The end of the range.
+     */
+    template<typename InputIterator>
+    __host__
+    host_vector(InputIterator first, InputIterator last)
+      :Parent(first, last) {}
+
+// declare these members for the purpose of Doxygenating them
+// they actually exist in a derived-from class
+#if 0
+    /*! \brief Resizes this vector to the specified number of elements.
+     *  \param new_size Number of elements this vector should contain.
+     *  \param x Data with which new elements should be populated.
+     *  \throw std::length_error If n exceeds max_size().
+     *
+     *  This method will resize this vector to the specified number of
+     *  elements.  If the number is smaller than this vector's current
+     *  size this vector is truncated, otherwise this vector is
+     *  extended and new elements are populated with given data.
+     */
+    void resize(size_type new_size, const value_type &x = value_type());
+
+    /*! Returns the number of elements in this vector.
+     */
+    size_type size(void) const;
+
+    /*! Returns the size() of the largest possible vector.
+     *  \return The largest possible return value of size().
+     */
+    size_type max_size(void) const;
+
+    /*! \brief If n is less than or equal to capacity(), this call has no effect.
+     *         Otherwise, this method is a request for allocation of additional memory. If
+     *         the request is successful, then capacity() is greater than or equal to
+     *         n; otherwise, capacity() is unchanged. In either case, size() is unchanged.
+     *  \throw std::length_error If n exceeds max_size().
+     */
+    void reserve(size_type n);
+
+    /*! Returns the number of elements which have been reserved in this
+     *  vector.
+     */
+    size_type capacity(void) const;
+
+    /*! This method shrinks the capacity of this vector to exactly
+     *  fit its elements.
+     */
+    void shrink_to_fit(void);
+
+    /*! \brief Subscript access to the data contained in this vector_dev.
+     *  \param n The index of the element for which data should be accessed.
+     *  \return Read/write reference to data.
+     *
+     *  This operator allows for easy, array-style, data access.
+     *  Note that data access with this operator is unchecked and
+     *  out_of_range lookups are not defined.
+     */
+    reference operator[](size_type n);
+
+    /*! \brief Subscript read access to the data contained in this vector_dev.
+     *  \param n The index of the element for which data should be accessed.
+     *  \return Read reference to data.
+     *
+     *  This operator allows for easy, array-style, data access.
+     *  Note that data access with this operator is unchecked and
+     *  out_of_range lookups are not defined.
+     */
+    const_reference operator[](size_type n) const;
+
+    /*! This method returns an iterator pointing to the beginning of
+     *  this vector.
+     *  \return mStart
+     */
+    iterator begin(void);
+
+    /*! This method returns a const_iterator pointing to the beginning
+     *  of this vector.
+     *  \return mStart
+     */
+    const_iterator begin(void) const;
+
+    /*! This method returns a const_iterator pointing to the beginning
+     *  of this vector.
+     *  \return mStart
+     */
+    const_iterator cbegin(void) const;
+
+    /*! This method returns a reverse_iterator pointing to the beginning of
+     *  this vector's reversed sequence.
+     *  \return A reverse_iterator pointing to the beginning of this
+     *          vector's reversed sequence.
+     */
+    reverse_iterator rbegin(void);
+
+    /*! This method returns a const_reverse_iterator pointing to the beginning of
+     *  this vector's reversed sequence.
+     *  \return A const_reverse_iterator pointing to the beginning of this
+     *          vector's reversed sequence.
+     */
+    const_reverse_iterator rbegin(void) const;
+
+    /*! This method returns a const_reverse_iterator pointing to the beginning of
+     *  this vector's reversed sequence.
+     *  \return A const_reverse_iterator pointing to the beginning of this
+     *          vector's reversed sequence.
+     */
+    const_reverse_iterator crbegin(void) const;
+
+    /*! This method returns an iterator pointing to one element past the
+     *  last of this vector.
+     *  \return begin() + size().
+     */
+    iterator end(void);
+
+    /*! This method returns a const_iterator pointing to one element past the
+     *  last of this vector.
+     *  \return begin() + size().
+     */
+    const_iterator end(void) const;
+
+    /*! This method returns a const_iterator pointing to one element past the
+     *  last of this vector.
+     *  \return begin() + size().
+     */
+    const_iterator cend(void) const;
+
+    /*! This method returns a reverse_iterator pointing to one element past the
+     *  last of this vector's reversed sequence.
+     *  \return rbegin() + size().
+     */
+    reverse_iterator rend(void);
+
+    /*! This method returns a const_reverse_iterator pointing to one element past the
+     *  last of this vector's reversed sequence.
+     *  \return rbegin() + size().
+     */
+    const_reverse_iterator rend(void) const;
+
+    /*! This method returns a const_reverse_iterator pointing to one element past the
+     *  last of this vector's reversed sequence.
+     *  \return rbegin() + size().
+     */
+    const_reverse_iterator crend(void) const;
+
+    /*! This method returns a const_reference referring to the first element of this
+     *  vector.
+     *  \return The first element of this vector.
+     */
+    const_reference front(void) const;
+
+    /*! This method returns a reference pointing to the first element of this
+     *  vector.
+     *  \return The first element of this vector.
+     */
+    reference front(void);
+
+    /*! This method returns a const reference pointing to the last element of
+     *  this vector.
+     *  \return The last element of this vector.
+     */
+    const_reference back(void) const;
+
+    /*! This method returns a reference referring to the last element of
+     *  this vector_dev.
+     *  \return The last element of this vector.
+     */
+    reference back(void);
+
+    /*! This method returns a pointer to this vector's first element.
+     *  \return A pointer to the first element of this vector.
+     */
+    pointer data(void);
+
+    /*! This method returns a const_pointer to this vector's first element.
+     *  \return a const_pointer to the first element of this vector.
+     */
+    const_pointer data(void) const;
+
+    /*! This method resizes this vector to 0.
+     */
+    void clear(void);
+
+    /*! This method returns true iff size() == 0.
+     *  \return true if size() == 0; false, otherwise.
+     */
+    bool empty(void) const;
+
+    /*! This method appends the given element to the end of this vector.
+     *  \param x The element to append.
+     */
+    void push_back(const value_type &x);
+
+    /*! This method erases the last element of this vector, invalidating
+     *  all iterators and references to it.
+     */
+    void pop_back(void);
+
+    /*! This method swaps the contents of this vector_base with another vector.
+     *  \param v The vector with which to swap.
+     */
+    void swap(host_vector &v);
+
+    /*! This method removes the element at position pos.
+     *  \param pos The position of the element of interest.
+     *  \return An iterator pointing to the new location of the element that followed the element
+     *          at position pos.
+     */
+    iterator erase(iterator pos);
+
+    /*! This method removes the range of elements [first,last) from this vector.
+     *  \param first The beginning of the range of elements to remove.
+     *  \param last The end of the range of elements to remove.
+     *  \return An iterator pointing to the new location of the element that followed the last
+     *          element in the sequence [first,last).
+     */
+    iterator erase(iterator first, iterator last);
+
+    /*! This method inserts a single copy of a given exemplar value at the
+     *  specified position in this vector.
+     *  \param position The insertion position.
+     *  \param x The exemplar element to copy & insert.
+     *  \return An iterator pointing to the newly inserted element.
+     */
+    iterator insert(iterator position, const T &x); 
+
+    /*! This method inserts a copy of an exemplar value to a range at the
+     *  specified position in this vector.
+     *  \param position The insertion position
+     *  \param n The number of insertions to perform.
+     *  \param x The value to replicate and insert.
+     */
+    void insert(iterator position, size_type n, const T &x);
+
+    /*! This method inserts a copy of an input range at the specified position
+     *  in this vector.
+     *  \param position The insertion position.
+     *  \param first The beginning of the range to copy.
+     *  \param last  The end of the range to copy.
+     *
+     *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html>Input Iterator</a>,
+     *                        and \p InputIterator's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/Assignable.html">Assignable</a>.
+     */
+    template<typename InputIterator>
+    void insert(iterator position, InputIterator first, InputIterator last);
+
+    /*! This version of \p assign replicates a given exemplar
+     *  \p n times into this vector.
+     *  \param n The number of times to copy \p x.
+     *  \param x The exemplar element to replicate.
+     */
+    void assign(size_type n, const T &x);
+
+    /*! This version of \p assign makes this vector a copy of a given input range.
+     *  \param first The beginning of the range to copy.
+     *  \param last  The end of the range to copy.
+     *
+     *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator">Input Iterator</a>.
+     */
+    template<typename InputIterator>
+    void assign(InputIterator first, InputIterator last);
+
+    /*! This method returns a copy of this vector's allocator.
+     *  \return A copy of the alloctor used by this vector.
+     */
+    allocator_type get_allocator(void) const;
+#endif // end doxygen-only members
+}; // end host_vector
+
+/*! \}
+ */
+
+} // end thrust
+
+#include <thrust/detail/host_vector.inl>
+
diff --git a/compat/thrust/inner_product.h b/compat/thrust/inner_product.h
new file mode 100644
index 0000000000..01f55414bd
--- /dev/null
+++ b/compat/thrust/inner_product.h
@@ -0,0 +1,262 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file inner_product.h
+ *  \brief Mathematical inner product between ranges
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/execution_policy.h>
+
+namespace thrust
+{
+
+
+/*! \addtogroup reductions
+ *  \{
+ *  \addtogroup transformed_reductions Transformed Reductions
+ *  \ingroup reductions
+ *  \{
+ */
+
+
+/*! \p inner_product calculates an inner product of the ranges
+ *  <tt>[first1, last1)</tt> and <tt>[first2, first2 + (last1 - first1))</tt>.
+ *
+ *  Specifically, this version of \p inner_product computes the sum
+ *  <tt>init + (*first1 * *first2) + (*(first1+1) * *(first2+1)) + ... </tt>
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first1 The beginning of the first sequence.
+ *  \param last1 The end of the first sequence.
+ *  \param first2 The beginning of the second sequence.
+ *  \param init Initial value of the result.
+ *  \return The inner product of sequences <tt>[first1, last1)</tt>
+ *          and <tt>[first2, last2)</tt> plus \p init.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *  \tparam OutputType is a model of <a href="http://www.sgi.com/tech/stl/Assignable.html">Assignable</a>,
+ *          and if \c x is an object of type \p OutputType, and \c y is an object of \p InputIterator1's \c value_type,
+ *          and \c z is an object of \p InputIterator2's \c value_type, then <tt>x + y * z</tt> is defined
+ *          and is convertible to \p OutputType.
+ *
+ *  The following code demonstrates how to use \p inner_product to
+ *  compute the dot product of two vectors using the \p thrust::host execution policy for parallelization.
+ *
+ *  \code
+ *  #include <thrust/inner_product.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  float vec1[3] = {1.0f, 2.0f, 5.0f};
+ *  float vec2[3] = {4.0f, 1.0f, 5.0f};
+ *
+ *  float result = thrust::inner_product(thrust::host, vec1, vec1 + 3, vec2, 0.0f);
+ *
+ *  // result == 31.0f
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/inner_product.html
+ */
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputType>
+OutputType inner_product(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                         InputIterator1 first1,
+                         InputIterator1 last1,
+                         InputIterator2 first2,
+                         OutputType init);
+
+
+/*! \p inner_product calculates an inner product of the ranges
+ *  <tt>[first1, last1)</tt> and <tt>[first2, first2 + (last1 - first1))</tt>.
+ *
+ *  Specifically, this version of \p inner_product computes the sum
+ *  <tt>init + (*first1 * *first2) + (*(first1+1) * *(first2+1)) + ... </tt>
+ *
+ *  Unlike the C++ Standard Template Library function <tt>std::inner_product</tt>,
+ *  this version offers no guarantee on order of execution.
+ *
+ *  \param first1 The beginning of the first sequence.
+ *  \param last1 The end of the first sequence.
+ *  \param first2 The beginning of the second sequence.
+ *  \param init Initial value of the result.
+ *  \return The inner product of sequences <tt>[first1, last1)</tt>
+ *          and <tt>[first2, last2)</tt> plus \p init.
+ *
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *  \tparam OutputType is a model of <a href="http://www.sgi.com/tech/stl/Assignable.html">Assignable</a>,
+ *          and if \c x is an object of type \p OutputType, and \c y is an object of \p InputIterator1's \c value_type,
+ *          and \c z is an object of \p InputIterator2's \c value_type, then <tt>x + y * z</tt> is defined
+ *          and is convertible to \p OutputType.
+ *
+ *  The following code demonstrates how to use \p inner_product to
+ *  compute the dot product of two vectors.
+ *
+ *  \code
+ *  #include <thrust/inner_product.h>
+ *  ...
+ *  float vec1[3] = {1.0f, 2.0f, 5.0f};
+ *  float vec2[3] = {4.0f, 1.0f, 5.0f};
+ *
+ *  float result = thrust::inner_product(vec1, vec1 + 3, vec2, 0.0f);
+ *
+ *  // result == 31.0f
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/inner_product.html
+ */
+template <typename InputIterator1, typename InputIterator2, typename OutputType>
+OutputType inner_product(InputIterator1 first1, InputIterator1 last1,
+                         InputIterator2 first2, OutputType init);
+
+
+/*! \p inner_product calculates an inner product of the ranges
+ *  <tt>[first1, last1)</tt> and <tt>[first2, first2 + (last1 - first1))</tt>.
+ *
+ *  This version of \p inner_product is identical to the first, except that is uses
+ *  two user-supplied function objects instead of \c operator+ and \c operator*.
+ *
+ *  Specifically, this version of \p inner_product computes the sum
+ *  <tt>binary_op1( init, binary_op2(*first1, *first2) ), ... </tt>
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first1 The beginning of the first sequence.
+ *  \param last1 The end of the first sequence.
+ *  \param first2 The beginning of the second sequence.
+ *  \param init Initial value of the result.
+ *  \param binary_op1 Generalized addition operation.
+ *  \param binary_op2 Generalized multiplication operation.
+ *  \return The inner product of sequences <tt>[first1, last1)</tt> and <tt>[first2, last2)</tt>.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator1's \c value_type is convertible to \p BinaryFunction2's \c first_argument_type.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>.
+ *          and \p InputIterator2's \c value_type is convertible to \p BinaryFunction2's \c second_argument_type.
+ *  \tparam OutputType is a model of <a href="http://www.sgi.com/tech/stl/Assignable.html">Assignable</a>,
+ *          and \p OutputType is convertible to \p BinaryFunction1's \c first_argument_type.
+ *  \tparam BinaryFunction1 is a model of <a href="http://www.sgi.com/tech/stl/BinaryFunction.html">Binary Function</a>,
+ *          and \p BinaryFunction1's \c return_type is convertible to \p OutputType.
+ *  \tparam BinaryFunction2 is a model of <a href="http://www.sgi.com/tech/stl/BinaryFunction.html">Binary Function</a>,
+ *          and \p BinaryFunction2's \c return_type is convertible to \p BinaryFunction1's \c second_argument_type.
+ * 
+ *  \code
+ *  #include <thrust/inner_product.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  float vec1[3] = {1.0f, 2.0f, 5.0f};
+ *  float vec2[3] = {4.0f, 1.0f, 5.0f};
+ *
+ *  float init = 0.0f;
+ *  thrust::plus<float>       binary_op1;
+ *  thrust::multiplies<float> binary_op2;
+ *
+ *  float result = thrust::inner_product(thrust::host, vec1, vec1 + 3, vec2, init, binary_op1, binary_op2);
+ *
+ *  // result == 31.0f
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/inner_product.html
+ */
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputType,
+         typename BinaryFunction1,
+         typename BinaryFunction2>
+OutputType inner_product(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                         InputIterator1 first1,
+                         InputIterator1 last1,
+                         InputIterator2 first2,
+                         OutputType init, 
+                         BinaryFunction1 binary_op1,
+                         BinaryFunction2 binary_op2);
+
+
+/*! \p inner_product calculates an inner product of the ranges
+ *  <tt>[first1, last1)</tt> and <tt>[first2, first2 + (last1 - first1))</tt>.
+ *
+ *  This version of \p inner_product is identical to the first, except that is uses
+ *  two user-supplied function objects instead of \c operator+ and \c operator*.
+ *
+ *  Specifically, this version of \p inner_product computes the sum
+ *  <tt>binary_op1( init, binary_op2(*first1, *first2) ), ... </tt>
+ *
+ *  Unlike the C++ Standard Template Library function <tt>std::inner_product</tt>,
+ *  this version offers no guarantee on order of execution.
+ *
+ *  \param first1 The beginning of the first sequence.
+ *  \param last1 The end of the first sequence.
+ *  \param first2 The beginning of the second sequence.
+ *  \param init Initial value of the result.
+ *  \param binary_op1 Generalized addition operation.
+ *  \param binary_op2 Generalized multiplication operation.
+ *  \return The inner product of sequences <tt>[first1, last1)</tt> and <tt>[first2, last2)</tt>.
+ *
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator1's \c value_type is convertible to \p BinaryFunction2's \c first_argument_type.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>.
+ *          and \p InputIterator2's \c value_type is convertible to \p BinaryFunction2's \c second_argument_type.
+ *  \tparam OutputType is a model of <a href="http://www.sgi.com/tech/stl/Assignable.html">Assignable</a>,
+ *          and \p OutputType is convertible to \p BinaryFunction1's \c first_argument_type.
+ *  \tparam BinaryFunction1 is a model of <a href="http://www.sgi.com/tech/stl/BinaryFunction.html">Binary Function</a>,
+ *          and \p BinaryFunction1's \c return_type is convertible to \p OutputType.
+ *  \tparam BinaryFunction2 is a model of <a href="http://www.sgi.com/tech/stl/BinaryFunction.html">Binary Function</a>,
+ *          and \p BinaryFunction2's \c return_type is convertible to \p BinaryFunction1's \c second_argument_type.
+ * 
+ *  \code
+ *  #include <thrust/inner_product.h>
+ *  ...
+ *  float vec1[3] = {1.0f, 2.0f, 5.0f};
+ *  float vec2[3] = {4.0f, 1.0f, 5.0f};
+ *
+ *  float init = 0.0f;
+ *  thrust::plus<float>       binary_op1;
+ *  thrust::multiplies<float> binary_op2;
+ *
+ *  float result = thrust::inner_product(vec1, vec1 + 3, vec2, init, binary_op1, binary_op2);
+ *
+ *  // result == 31.0f
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/inner_product.html
+ */
+template <typename InputIterator1, typename InputIterator2, typename OutputType,
+          typename BinaryFunction1, typename BinaryFunction2>
+OutputType inner_product(InputIterator1 first1, InputIterator1 last1,
+                         InputIterator2 first2, OutputType init, 
+                         BinaryFunction1 binary_op1, BinaryFunction2 binary_op2);
+
+
+/*! \} // end transformed_reductions
+ *  \} // end reductions
+ */
+
+} // end namespace thrust
+
+#include <thrust/detail/inner_product.inl>
+
diff --git a/compat/thrust/iterator/constant_iterator.h b/compat/thrust/iterator/constant_iterator.h
new file mode 100644
index 0000000000..e9e03c18c2
--- /dev/null
+++ b/compat/thrust/iterator/constant_iterator.h
@@ -0,0 +1,251 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file thrust/iterator/constant_iterator.h
+ *  \brief An iterator which returns a constant value when
+ *         dereferenced
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/iterator/detail/constant_iterator_base.h>
+#include <thrust/iterator/iterator_facade.h>
+
+namespace thrust
+{
+
+/*! \addtogroup iterators
+ *  \{
+ */
+
+/*! \addtogroup fancyiterator Fancy Iterators
+ *  \ingroup iterators
+ *  \{
+ */
+
+/*! \p constant_iterator is an iterator which represents a pointer into a range
+ *  of constant values. This iterator is useful for creating a range filled with the same
+ *  value without explicitly storing it in memory. Using \p constant_iterator saves both
+ *  memory capacity and bandwidth.
+ *
+ *  The following code snippet demonstrates how to create a \p constant_iterator whose
+ *  \c value_type is \c int and whose value is \c 10.
+ *
+ *  \code
+ *  #include <thrust/iterator/constant_iterator.h>
+ *
+ *  thrust::constant_iterator<int> iter(10);
+ *
+ *  *iter;    // returns 10
+ *  iter[0];  // returns 10
+ *  iter[1];  // returns 10
+ *  iter[13]; // returns 10
+ *
+ *  // and so on...
+ *  \endcode
+ *
+ *  This next example demonstrates how to use a \p constant_iterator with the
+ *  \p thrust::transform function to increment all elements of a sequence by the
+ *  same value. We will create a temporary \p constant_iterator with the function
+ *  \p make_constant_iterator function in order to avoid explicitly specifying
+ *  its type:
+ *
+ *  \code
+ *  #include <thrust/iterator/constant_iterator.h>
+ *  #include <thrust/transform.h>
+ *  #include <thrust/functional.h>
+ *  #include <thrust/device_vector.h>
+ *
+ *  int main(void)
+ *  {
+ *    thrust::device_vector<int> data(4);
+ *    data[0] = 3;
+ *    data[1] = 7;
+ *    data[2] = 2;
+ *    data[3] = 5;
+ *    
+ *    // add 10 to all values in data
+ *    thrust::transform(data.begin(), data.end(),
+ *                      thrust::make_constant_iterator(10),
+ *                      data.begin(),
+ *                      thrust::plus<int>());
+ *    
+ *    // data is now [13, 17, 12, 15]
+ *    
+ *    return 0;
+ *  }
+ *  \endcode
+ *
+ *  \see make_constant_iterator
+ */
+template<typename Value,
+         typename Incrementable = use_default,
+         typename System = use_default>
+  class constant_iterator
+    : public detail::constant_iterator_base<Value, Incrementable, System>::type
+{
+    /*! \cond
+     */
+    friend class thrust::iterator_core_access;
+    typedef typename detail::constant_iterator_base<Value, Incrementable, System>::type          super_t;
+    typedef typename detail::constant_iterator_base<Value, Incrementable, System>::incrementable incrementable;
+    typedef typename detail::constant_iterator_base<Value, Incrementable, System>::base_iterator base_iterator;
+
+  public:
+    typedef typename super_t::reference  reference;
+    typedef typename super_t::value_type value_type;
+
+    /*! \endcond
+     */
+
+    /*! Null constructor initializes this \p constant_iterator's constant using its
+     *  null constructor.
+     */
+    __host__ __device__
+    constant_iterator(void)
+      : super_t(), m_value(){};
+
+    /*! Copy constructor copies the value of another \p constant_iterator into this
+     *  \p constant_iterator.
+     *
+     *  \p rhs The constant_iterator to copy.
+     */
+    __host__ __device__
+    constant_iterator(constant_iterator const &rhs)
+      : super_t(rhs.base()), m_value(rhs.m_value) {}
+
+    /*! Copy constructor copies the value of another \p constant_iterator with related
+     *  System type.
+     *
+     *  \param rhs The \p constant_iterator to copy.
+     */
+    template<typename OtherSystem>
+    __host__ __device__
+    constant_iterator(constant_iterator<Value,Incrementable,OtherSystem> const &rhs,
+                      typename thrust::detail::enable_if_convertible<
+                        typename thrust::iterator_system<constant_iterator<Value,Incrementable,OtherSystem> >::type,
+                        typename thrust::iterator_system<super_t>::type
+                      >::type * = 0)
+      : super_t(rhs.base()), m_value(rhs.value()) {}
+
+    /*! This constructor receives a value to use as the constant value of this
+     *  \p constant_iterator and an index specifying the location of this
+     *  \p constant_iterator in a sequence.
+     *  
+     *  \p v The value of this \p constant_iterator's constant value.
+     *  \p i The index of this \p constant_iterator in a sequence. Defaults to the
+     *       value returned by \c Incrementable's null constructor. For example,
+     *       when <tt>Incrementable == int</tt>, \c 0.
+     */
+    __host__ __device__
+    constant_iterator(value_type const& v, incrementable const &i = incrementable())
+      : super_t(base_iterator(i)), m_value(v) {}
+
+    /*! This constructor is templated to allow construction from a value type and
+     *  incrementable type related this this \p constant_iterator's respective types.
+     *
+     *  \p v The value of this \p constant_iterator's constant value.
+     *  \p i The index of this \p constant_iterator in a sequence. Defaults to the
+     *       value returned by \c Incrementable's null constructor. For example,
+     *       when <tt>Incrementable == int</tt>, \c 0.
+     */
+    template<typename OtherValue, typename OtherIncrementable>
+    __host__ __device__
+    constant_iterator(OtherValue const& v, OtherIncrementable const& i = incrementable())
+      : super_t(base_iterator(i)), m_value(v) {}
+
+    /*! This method returns the value of this \p constant_iterator's constant value.
+     *  \return A \c const reference to this \p constant_iterator's constant value.
+     */
+    __host__ __device__
+    Value const& value(void) const
+    { return m_value; }
+
+    /*! \cond
+     */
+
+  protected:
+    __host__ __device__
+    Value const& value_reference(void) const
+    { return m_value; }
+
+    __host__ __device__
+    Value & value_reference(void)
+    { return m_value; }
+  
+  private: // Core iterator interface
+    __host__ __device__
+    reference dereference(void) const
+    {
+      return m_value;
+    }
+
+  private:
+    Value m_value;
+
+    /*! \endcond
+     */
+}; // end constant_iterator
+
+
+/*! This version of \p make_constant_iterator creates a \p constant_iterator
+ *  from values given for both value and index. The type of \p constant_iterator
+ *  may be inferred by the compiler from the types of its parameters.
+ *
+ *  \param x The value of the returned \p constant_iterator's constant value.
+ *  \param i The index of the returned \p constant_iterator within a sequence.
+ *           The type of this parameter defaults to \c int. In the default case,
+ *           the value of this parameter is \c 0.
+ *
+ *  \return A new \p constant_iterator with constant value & index as given
+ *          by \p x & \p i.
+ *
+ *  \see constant_iterator
+ */
+template<typename V, typename I>
+inline __host__ __device__
+constant_iterator<V,I> make_constant_iterator(V x, I i = int())
+{
+  return constant_iterator<V,I>(x, i);
+} // end make_constant_iterator()
+
+
+/*! This version of \p make_constant_iterator creates a \p constant_iterator
+ *  using only a parameter for the desired constant value. The value of the
+ *  returned \p constant_iterator's index is set to \c 0.
+ *
+ *  \param x The value of the returned \p constant_iterator's constant value.
+ *  \return A new \p constant_iterator with constant value equal to \p x and
+ *          index equal to \c 0.
+ *  \see constant_iterator
+ */
+template<typename V>
+inline __host__ __device__
+constant_iterator<V> make_constant_iterator(V x)
+{
+  return constant_iterator<V>(x, 0);
+} // end make_constant_iterator()
+
+/*! \} // end fancyiterators
+ */
+
+/*! \} // end iterators
+ */
+
+} // end namespace thrust
+
diff --git a/compat/thrust/iterator/counting_iterator.h b/compat/thrust/iterator/counting_iterator.h
new file mode 100644
index 0000000000..99812cae17
--- /dev/null
+++ b/compat/thrust/iterator/counting_iterator.h
@@ -0,0 +1,243 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file thrust/iterator/counting_iterator.h
+ *  \brief An iterator which returns an increasing incrementable value
+ *         when dereferenced
+ */
+
+/*
+ * Copyright David Abrahams 2003.
+ * 
+ * Distributed under the Boost Software License, Version 1.0.
+ * (See accompanying NOTICE file for the complete license)
+ *
+ * For more information, see http://www.boost.org
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/iterator/iterator_adaptor.h>
+#include <thrust/iterator/iterator_facade.h>
+#include <thrust/iterator/iterator_categories.h>
+
+// #include the details first
+#include <thrust/iterator/detail/counting_iterator.inl>
+
+namespace thrust
+{
+
+/*! \addtogroup iterators
+ *  \{
+ */
+
+/*! \addtogroup fancyiterator Fancy Iterators
+ *  \ingroup iterators
+ *  \{
+ */
+
+/*! \p counting_iterator is an iterator which represents a pointer into a range
+ *  of sequentially changing values. This iterator is useful for creating a range
+ *  filled with a sequence without explicitly storing it in memory. Using
+ *  \p counting_iterator saves memory capacity and bandwidth.
+ *
+ *  The following code snippet demonstrates how to create a \p counting_iterator whose
+ *  \c value_type is \c int and which sequentially increments by \c 1.
+ *
+ *  \code
+ *  #include <thrust/iterator/counting_iterator.h>
+ *  ...
+ *  // create iterators
+ *  thrust::counting_iterator<int> first(10);
+ *  thrust::counting_iterator<int> last = first + 3;
+ *   
+ *  first[0]   // returns 10
+ *  first[1]   // returns 11
+ *  first[100] // returns 110
+ *   
+ *  // sum of [first, last)
+ *  thrust::reduce(first, last);   // returns 33 (i.e. 10 + 11 + 12)
+ *   
+ *  // initialize vector to [0,1,2,..]
+ *  thrust::counting_iterator<int> iter(0);
+ *  thrust::device_vector<int> vec(500);
+ *  thrust::copy(iter, iter + vec.size(), vec.begin());
+ *  \endcode
+ *
+ *  This next example demonstrates how to use a \p counting_iterator with the
+ *  \p thrust::copy_if function to compute the indices of the non-zero elements
+ *  of a \p device_vector. In this example, we use the \p make_counting_iterator
+ *  function to avoid specifying the type of the \p counting_iterator.
+ *
+ *  \code
+ *  #include <thrust/iterator/counting_iterator.h>
+ *  #include <thrust/copy.h>
+ *  #include <thrust/functional.h>
+ *  #include <thrust/device_vector.h>
+ *   
+ *  int main(void)
+ *  {
+ *   // this example computes indices for all the nonzero values in a sequence
+ *   
+ *   // sequence of zero and nonzero values
+ *   thrust::device_vector<int> stencil(8);
+ *   stencil[0] = 0;
+ *   stencil[1] = 1;
+ *   stencil[2] = 1;
+ *   stencil[3] = 0;
+ *   stencil[4] = 0;
+ *   stencil[5] = 1;
+ *   stencil[6] = 0;
+ *   stencil[7] = 1;
+ *   
+ *   // storage for the nonzero indices
+ *   thrust::device_vector<int> indices(8);
+ *   
+ *   // compute indices of nonzero elements
+ *   typedef thrust::device_vector<int>::iterator IndexIterator;
+ *   
+ *   // use make_counting_iterator to define the sequence [0, 8)
+ *   IndexIterator indices_end = thrust::copy_if(thrust::make_counting_iterator(0),
+ *                                               thrust::make_counting_iterator(8),
+ *                                               stencil.begin(),
+ *                                               indices.begin(),
+ *                                               thrust::identity<int>());
+ *   // indices now contains [1,2,5,7]
+ *   
+ *   return 0;
+ *  }
+ *  \endcode
+ *
+ *  \see make_counting_iterator
+ */
+template<typename Incrementable,
+         typename System = use_default,
+         typename Traversal = use_default,
+         typename Difference = use_default>
+  class counting_iterator
+    : public detail::counting_iterator_base<Incrementable, System, Traversal, Difference>::type
+{
+    /*! \cond
+     */
+    typedef typename detail::counting_iterator_base<Incrementable, System, Traversal, Difference>::type super_t;
+
+    friend class thrust::iterator_core_access;
+
+  public:
+    typedef typename super_t::reference       reference;
+    typedef typename super_t::difference_type difference_type;
+
+    /*! \endcond
+     */
+
+    /*! Null constructor initializes this \p counting_iterator's \c Incrementable
+     *  counter using its null constructor.
+     */
+    __host__ __device__
+    counting_iterator(void){};
+
+    /*! Copy constructor copies the value of another \p counting_iterator into a
+     *  new \p counting_iterator.
+     *
+     *  \p rhs The \p counting_iterator to copy.
+     */
+    __host__ __device__
+    counting_iterator(counting_iterator const &rhs):super_t(rhs.base()){}
+
+    /*! Copy constructor copies the value of another counting_iterator 
+     *  with related System type.
+     *
+     *  \param rhs The \p counting_iterator to copy.
+     */
+    template<typename OtherSystem>
+    __host__ __device__
+    counting_iterator(counting_iterator<Incrementable, OtherSystem, Traversal, Difference> const &rhs,
+                      typename thrust::detail::enable_if_convertible<
+                        typename thrust::iterator_system<counting_iterator<Incrementable,OtherSystem,Traversal,Difference> >::type,
+                        typename thrust::iterator_system<super_t>::type
+                      >::type * = 0)
+      : super_t(rhs.base()){}
+
+    /*! This \c explicit constructor copies the value of an \c Incrementable
+     *  into a new \p counting_iterator's \c Incrementable counter.
+     *  
+     *  \param x The initial value of the new \p counting_iterator's \c Incrementable
+     *         counter.
+     */
+    __host__ __device__
+    explicit counting_iterator(Incrementable x):super_t(x){}
+
+    /*! \cond
+     */
+  private:
+    __host__ __device__
+    reference dereference(void) const
+    {
+      return this->base_reference();
+    }
+
+    // note that we implement equal specially for floating point counting_iterator
+    template <typename OtherIncrementable, typename OtherSystem, typename OtherTraversal, typename OtherDifference>
+    __host__ __device__
+    bool equal(counting_iterator<OtherIncrementable, OtherSystem, OtherTraversal, OtherDifference> const& y) const
+    {
+      typedef thrust::detail::counting_iterator_equal<difference_type,Incrementable,OtherIncrementable> e;
+      return e::equal(this->base(), y.base());
+    }
+
+    template <class OtherIncrementable>
+    __host__ __device__
+    difference_type
+    distance_to(counting_iterator<OtherIncrementable, System, Traversal, Difference> const& y) const
+    {
+      typedef typename
+      thrust::detail::eval_if<
+        thrust::detail::is_numeric<Incrementable>::value,
+        thrust::detail::identity_<thrust::detail::number_distance<difference_type, Incrementable, OtherIncrementable> >,
+        thrust::detail::identity_<thrust::detail::iterator_distance<difference_type, Incrementable, OtherIncrementable> >
+      >::type d;
+
+      return d::distance(this->base(), y.base());
+    }
+
+    /*! \endcond
+     */
+}; // end counting_iterator
+
+
+/*! \p make_counting_iterator creates a \p counting_iterator
+ *  using an initial value for its \c Incrementable counter.
+ *
+ *  \param x The initial value of the new \p counting_iterator's counter.
+ *  \return A new \p counting_iterator whose counter has been initialized to \p x.
+ */
+template <typename Incrementable>
+inline __host__ __device__
+counting_iterator<Incrementable> make_counting_iterator(Incrementable x)
+{
+  return counting_iterator<Incrementable>(x);
+}
+
+/*! \} // end fancyiterators
+ */
+
+/*! \} // end iterators
+ */
+
+} // end thrust
+
diff --git a/compat/thrust/iterator/detail/any_assign.h b/compat/thrust/iterator/detail/any_assign.h
new file mode 100644
index 0000000000..e08a829ec0
--- /dev/null
+++ b/compat/thrust/iterator/detail/any_assign.h
@@ -0,0 +1,55 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+namespace thrust
+{
+namespace detail
+{
+
+
+// a type which may be assigned any other type
+struct any_assign
+{
+  inline __host__ __device__ any_assign(void)
+  {}
+
+  template<typename T>
+  inline __host__ __device__ any_assign(T)
+  {}
+
+  template<typename T>
+  inline __host__ __device__
+  any_assign &operator=(T)
+  {
+    if(0)
+    {
+      // trick the compiler into silencing "warning: this expression has no effect"
+      int *x = 0;
+      *x = 13;
+    } // end if
+
+    return *this;
+  }
+};
+
+
+} // end detail
+} // end thrust
+
diff --git a/compat/thrust/iterator/detail/any_system_tag.h b/compat/thrust/iterator/detail/any_system_tag.h
new file mode 100644
index 0000000000..fc6417ad8a
--- /dev/null
+++ b/compat/thrust/iterator/detail/any_system_tag.h
@@ -0,0 +1,37 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/execution_policy.h>
+
+namespace thrust
+{
+
+struct any_system_tag
+  : thrust::execution_policy<any_system_tag>
+{
+  // allow any_system_tag to convert to any type at all
+  // XXX make this safer using enable_if<is_tag<T>> upon c++11
+  template<typename T> operator T () const {return T();}
+};
+
+// TODO remove this in 1.7.0
+typedef THRUST_DEPRECATED any_system_tag any_space_tag;
+
+} // end thrust
+
diff --git a/compat/thrust/iterator/detail/constant_iterator_base.h b/compat/thrust/iterator/detail/constant_iterator_base.h
new file mode 100644
index 0000000000..276e5ff0ef
--- /dev/null
+++ b/compat/thrust/iterator/detail/constant_iterator_base.h
@@ -0,0 +1,70 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/iterator/counting_iterator.h>
+#include <thrust/iterator/iterator_adaptor.h>
+
+namespace thrust
+{
+
+// forward declaration of constant_iterator
+template<typename,typename,typename> class constant_iterator;
+
+namespace detail
+{
+
+template<typename Value,
+         typename Incrementable,
+         typename System>
+  struct constant_iterator_base
+{
+  typedef Value              value_type;
+
+  // the reference type is the same as the value_type.
+  // we wish to avoid returning a reference to the internal state
+  // of the constant_iterator, which is prone to subtle bugs.
+  // consider the temporary iterator created in the expression
+  // *(iter + i)
+  typedef value_type         reference;
+
+  // the incrementable type is int unless otherwise specified
+  typedef typename thrust::detail::ia_dflt_help<
+    Incrementable,
+    thrust::detail::identity_<int>
+  >::type incrementable;
+
+  typedef typename thrust::counting_iterator<
+    incrementable,
+    System,
+    thrust::random_access_traversal_tag
+  > base_iterator;
+
+  typedef typename thrust::iterator_adaptor<
+    constant_iterator<Value, Incrementable, System>,
+    base_iterator,
+    value_type, // XXX we may need to pass const value_type here as boost counting_iterator does
+    typename thrust::iterator_system<base_iterator>::type,
+    typename thrust::iterator_traversal<base_iterator>::type,
+    reference
+  > type;
+}; // end constant_iterator_base
+
+} // end detail
+  
+} // end thrust
+
diff --git a/compat/thrust/iterator/detail/counting_iterator.inl b/compat/thrust/iterator/detail/counting_iterator.inl
new file mode 100644
index 0000000000..ad4fcffaa6
--- /dev/null
+++ b/compat/thrust/iterator/detail/counting_iterator.inl
@@ -0,0 +1,141 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/iterator/counting_iterator.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/detail/numeric_traits.h>
+#include <thrust/detail/type_traits.h>
+#include <cstddef>
+
+namespace thrust
+{
+
+// forward declaration of counting_iterator
+template <typename Incrementable, typename System, typename Traversal, typename Difference>
+  class counting_iterator;
+
+namespace detail
+{
+
+template <typename Incrementable, typename System, typename Traversal, typename Difference>
+  struct counting_iterator_base
+{
+  typedef typename thrust::detail::eval_if<
+    // use any_system_tag if we are given use_default
+    thrust::detail::is_same<System,use_default>::value,
+    thrust::detail::identity_<thrust::any_system_tag>,
+    thrust::detail::identity_<System>
+  >::type system;
+
+  typedef typename thrust::detail::ia_dflt_help<
+      Traversal,
+      thrust::detail::eval_if<
+          thrust::detail::is_numeric<Incrementable>::value,
+          thrust::detail::identity_<random_access_traversal_tag>,
+          thrust::iterator_traversal<Incrementable>
+      >
+  >::type traversal;
+
+  // unlike Boost, we explicitly use std::ptrdiff_t as the difference type
+  // for floating point counting_iterators
+  typedef typename thrust::detail::ia_dflt_help<
+    Difference,
+    thrust::detail::eval_if<
+      thrust::detail::is_numeric<Incrementable>::value,
+        thrust::detail::eval_if<
+          thrust::detail::is_integral<Incrementable>::value,
+          thrust::detail::numeric_difference<Incrementable>,
+          thrust::detail::identity_<std::ptrdiff_t>
+        >,
+      thrust::iterator_difference<Incrementable>
+    >
+  >::type difference;
+
+  // our implementation departs from Boost's in that counting_iterator::dereference
+  // returns a copy of its counter, rather than a reference to it. returning a reference
+  // to the internal state of an iterator causes subtle bugs (consider the temporary
+  // iterator created in the expression *(iter + i) ) and has no compelling use case
+  typedef thrust::iterator_adaptor<
+    counting_iterator<Incrementable, System, Traversal, Difference>, // self
+    Incrementable,                                                  // Base
+    Incrementable,                                                  // XXX we may need to pass const here as Boost does
+    system,
+    traversal,
+    Incrementable,
+    difference
+  > type;
+}; // end counting_iterator_base
+
+
+template<typename Difference, typename Incrementable1, typename Incrementable2>
+  struct iterator_distance
+{
+  __host__ __device__
+  static Difference distance(Incrementable1 x, Incrementable2 y)
+  {
+    return y - x;
+  }
+};
+
+
+template<typename Difference, typename Incrementable1, typename Incrementable2>
+  struct number_distance
+{
+  __host__ __device__
+  static Difference distance(Incrementable1 x, Incrementable2 y)
+  {
+      return static_cast<Difference>(numeric_distance(x,y));
+  }
+};
+
+
+template<typename Difference, typename Incrementable1, typename Incrementable2, typename Enable = void>
+  struct counting_iterator_equal
+{
+  __host__ __device__
+  static bool equal(Incrementable1 x, Incrementable2 y)
+  {
+    return x == y;
+  }
+};
+
+
+// specialization for floating point equality
+template<typename Difference, typename Incrementable1, typename Incrementable2>
+  struct counting_iterator_equal<
+    Difference,
+    Incrementable1,
+    Incrementable2,
+    typename thrust::detail::enable_if<
+      thrust::detail::is_floating_point<Incrementable1>::value ||
+      thrust::detail::is_floating_point<Incrementable2>::value
+    >::type
+  >
+{
+  __host__ __device__
+  static bool equal(Incrementable1 x, Incrementable2 y)
+  {
+    typedef number_distance<Difference,Incrementable1,Incrementable2> d;
+    return d::distance(x,y) == 0;
+  }
+};
+
+
+} // end detail
+} // end thrust
+
diff --git a/compat/thrust/iterator/detail/device_system_tag.h b/compat/thrust/iterator/detail/device_system_tag.h
new file mode 100644
index 0000000000..ab66fb48bf
--- /dev/null
+++ b/compat/thrust/iterator/detail/device_system_tag.h
@@ -0,0 +1,40 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// #include the device system's execution_policy header
+#define __THRUST_DEVICE_SYSTEM_TAG_HEADER <__THRUST_DEVICE_SYSTEM_ROOT/detail/execution_policy.h>
+#include __THRUST_DEVICE_SYSTEM_TAG_HEADER
+#undef __THRUST_DEVICE_SYSTEM_TAG_HEADER
+
+namespace thrust
+{
+
+typedef thrust::system::__THRUST_DEVICE_SYSTEM_NAMESPACE::tag device_system_tag;
+
+} // end thrust
+
+// TODO remove this in 1.8.0
+namespace thrust
+{
+
+typedef THRUST_DEPRECATED device_system_tag device_space_tag;
+
+} // end thrust
+
diff --git a/compat/thrust/iterator/detail/discard_iterator_base.h b/compat/thrust/iterator/detail/discard_iterator_base.h
new file mode 100644
index 0000000000..1909ca8239
--- /dev/null
+++ b/compat/thrust/iterator/detail/discard_iterator_base.h
@@ -0,0 +1,65 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/iterator/counting_iterator.h>
+#include <thrust/iterator/iterator_adaptor.h>
+#include <thrust/iterator/detail/any_assign.h>
+#include <cstddef> // for std::ptrdiff_t
+
+namespace thrust
+{
+
+// forward declaration of discard_iterator
+template<typename> class discard_iterator;
+
+namespace detail
+{
+
+
+template<typename System>
+  struct discard_iterator_base
+{
+  // XXX value_type should actually be void
+  //     but this interferes with zip_iterator<discard_iterator>
+  typedef any_assign         value_type;
+  typedef any_assign&        reference;
+  typedef std::ptrdiff_t     incrementable;
+
+  typedef typename thrust::counting_iterator<
+    incrementable,
+    System,
+    thrust::random_access_traversal_tag
+  > base_iterator;
+
+  typedef typename thrust::iterator_adaptor<
+    discard_iterator<System>,
+    base_iterator,
+    value_type,
+    typename thrust::iterator_system<base_iterator>::type,
+    typename thrust::iterator_traversal<base_iterator>::type,
+    reference
+  > type;
+}; // end discard_iterator_base
+
+
+} // end detail
+  
+} // end thrust
+
+
diff --git a/compat/thrust/iterator/detail/distance_from_result.h b/compat/thrust/iterator/detail/distance_from_result.h
new file mode 100644
index 0000000000..bf83e6ca44
--- /dev/null
+++ b/compat/thrust/iterator/detail/distance_from_result.h
@@ -0,0 +1,42 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/type_traits.h>
+
+namespace thrust
+{
+
+namespace detail
+{
+
+// since both arguments are known to be specializations of iterator_facade,
+// it's legal to access IteratorFacade2::difference_type
+template<typename IteratorFacade1, typename IteratorFacade2>
+  struct distance_from_result
+    : eval_if<
+        is_convertible<IteratorFacade2,IteratorFacade1>::value,
+        identity_<typename IteratorFacade1::difference_type>,
+        identity_<typename IteratorFacade2::difference_type>
+      >
+{};
+
+} // end detail
+
+} // end thrust
+
diff --git a/compat/thrust/iterator/detail/host_system_tag.h b/compat/thrust/iterator/detail/host_system_tag.h
new file mode 100644
index 0000000000..26d3f7d73f
--- /dev/null
+++ b/compat/thrust/iterator/detail/host_system_tag.h
@@ -0,0 +1,40 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// #include the host system's execution_policy header
+#define __THRUST_HOST_SYSTEM_TAG_HEADER <__THRUST_HOST_SYSTEM_ROOT/detail/execution_policy.h>
+#include __THRUST_HOST_SYSTEM_TAG_HEADER
+#undef __THRUST_HOST_SYSTEM_TAG_HEADER
+
+namespace thrust
+{
+
+typedef thrust::system::__THRUST_HOST_SYSTEM_NAMESPACE::tag host_system_tag;
+
+} // end thrust
+
+// TODO remove this in 1.8.0
+namespace thrust
+{
+
+typedef THRUST_DEPRECATED host_system_tag host_space_tag;
+
+} // end thrust
+
diff --git a/compat/thrust/iterator/detail/is_iterator_category.h b/compat/thrust/iterator/detail/is_iterator_category.h
new file mode 100644
index 0000000000..95f14d558c
--- /dev/null
+++ b/compat/thrust/iterator/detail/is_iterator_category.h
@@ -0,0 +1,60 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/iterator/iterator_categories.h>
+#include <thrust/detail/type_traits.h>
+
+namespace thrust
+{
+
+namespace detail
+{
+
+template <typename T>
+  struct is_host_iterator_category
+    : thrust::detail::or_<
+        thrust::detail::is_convertible<T, thrust::input_host_iterator_tag>,
+        thrust::detail::is_convertible<T, thrust::output_host_iterator_tag>
+      >
+{
+}; // end is_host_iterator_category
+
+template <typename T>
+  struct is_device_iterator_category
+    : thrust::detail::or_<
+        thrust::detail::is_convertible<T, thrust::input_device_iterator_tag>,
+        thrust::detail::is_convertible<T, thrust::output_device_iterator_tag>
+      >
+{
+}; // end is_device_iterator_category
+
+
+template <typename T>
+  struct is_iterator_category
+    : thrust::detail::or_<
+        is_host_iterator_category<T>,
+        is_device_iterator_category<T>
+      >
+{
+}; // end is_iterator_category
+
+} // end detail
+
+} // end thrust
+
diff --git a/compat/thrust/iterator/detail/is_trivial_iterator.h b/compat/thrust/iterator/detail/is_trivial_iterator.h
new file mode 100644
index 0000000000..ca37e74e64
--- /dev/null
+++ b/compat/thrust/iterator/detail/is_trivial_iterator.h
@@ -0,0 +1,96 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/type_traits/pointer_traits.h>
+
+#if __GNUC__
+// forward declaration of gnu's __normal_iterator
+namespace __gnu_cxx
+{
+
+template<typename Iterator, typename Container> class __normal_iterator;
+
+} // end __gnu_cxx
+#endif // __GNUC__
+
+#if _MSC_VER
+// forward declaration of MSVC's "normal iterators"
+namespace std
+{
+
+template<typename Value, typename Difference, typename Pointer, typename Reference> struct _Ranit;
+
+} // end std
+#endif // _MSC_VER
+
+namespace thrust
+{
+namespace detail
+{
+
+#ifdef __GNUC__
+template<typename T>
+  struct is_gnu_normal_iterator
+    : false_type
+{};
+
+
+// catch gnu __normal_iterators
+template<typename Iterator, typename Container>
+  struct is_gnu_normal_iterator< __gnu_cxx::__normal_iterator<Iterator, Container> >
+    : true_type
+{};
+#endif // __GNUC__
+
+
+#ifdef _MSC_VER
+// catch msvc _Ranit
+template<typename Iterator>
+  struct is_convertible_to_msvc_Ranit :
+    is_convertible<
+      Iterator,
+      std::_Ranit<
+        typename iterator_value<Iterator>::type,
+        typename iterator_difference<Iterator>::type,
+        typename iterator_pointer<Iterator>::type,
+        typename iterator_reference<Iterator>::type
+      >
+    >
+{};
+#endif // _MSC_VER
+
+
+template<typename T>
+  struct is_trivial_iterator :
+    integral_constant<
+      bool,
+        is_pointer<T>::value
+      | thrust::detail::is_thrust_pointer<T>::value
+#if __GNUC__
+      | is_gnu_normal_iterator<T>::value
+#endif // __GNUC__
+#ifdef _MSC_VER
+      | is_convertible_to_msvc_Ranit<T>::value
+#endif // _MSC_VER
+    >
+{};
+
+} // end detail
+} // end thrust
+
diff --git a/compat/thrust/iterator/detail/iterator_adaptor_base.h b/compat/thrust/iterator/detail/iterator_adaptor_base.h
new file mode 100644
index 0000000000..8b77f05d81
--- /dev/null
+++ b/compat/thrust/iterator/detail/iterator_adaptor_base.h
@@ -0,0 +1,111 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/type_traits.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/detail/use_default.h>
+#include <thrust/iterator/iterator_facade.h>
+
+namespace thrust
+{
+
+
+// forward declaration of iterator_adaptor for iterator_adaptor_base below
+template<typename Derived,
+         typename Base,
+         typename Value,
+         typename System,
+         typename Traversal,
+         typename Reference,
+         typename Difference
+>
+class iterator_adaptor;
+
+
+namespace detail
+{
+
+// If T is use_default, return the result of invoking
+// DefaultNullaryFn, otherwise return T.
+// XXX rename to dflt_help
+template <class T, class DefaultNullaryFn>
+struct ia_dflt_help
+  : thrust::detail::eval_if<
+        thrust::detail::is_same<T, thrust::use_default>::value
+      , DefaultNullaryFn
+      , thrust::detail::identity_<T>
+    >
+{
+}; // end ia_dflt_help
+
+
+// A metafunction which computes an iterator_adaptor's base class,
+// a specialization of iterator_facade.
+template<typename Derived,
+         typename Base,
+         typename Value,
+         typename System,
+         typename Traversal,
+         typename Reference,
+         typename Difference
+>
+  struct iterator_adaptor_base
+{
+  typedef typename ia_dflt_help<
+    Value,
+    iterator_value<Base>
+  >::type value;
+
+  typedef typename ia_dflt_help<
+    System,
+    thrust::iterator_system<Base>
+  >::type system;
+
+  typedef typename ia_dflt_help<
+    Traversal,
+    thrust::iterator_traversal<Base>
+  >::type traversal;
+
+  typedef typename ia_dflt_help<
+    Reference,
+    thrust::detail::eval_if<
+      thrust::detail::is_same<Value,use_default>::value,
+      thrust::iterator_reference<Base>,
+      thrust::detail::add_reference<Value>
+    >
+  >::type reference;
+
+  typedef typename ia_dflt_help<
+    Difference,
+    iterator_difference<Base>
+  >::type difference;
+
+  typedef thrust::iterator_facade<
+    Derived,
+    value,
+    system,
+    traversal,
+    reference,
+    difference
+  > type;
+}; // end iterator_adaptor_base
+
+
+} // end detail
+} // end thrust
+
diff --git a/compat/thrust/iterator/detail/iterator_category_to_system.h b/compat/thrust/iterator/detail/iterator_category_to_system.h
new file mode 100644
index 0000000000..17e7d78c8f
--- /dev/null
+++ b/compat/thrust/iterator/detail/iterator_category_to_system.h
@@ -0,0 +1,95 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/iterator/iterator_categories.h>
+#include <thrust/iterator/detail/iterator_traversal_tags.h>
+#include <thrust/iterator/detail/host_system_tag.h>
+#include <thrust/iterator/detail/device_system_tag.h>
+#include <thrust/iterator/detail/any_system_tag.h>
+#include <thrust/detail/type_traits.h>
+
+namespace thrust
+{
+
+// XXX WAR circular #inclusion with forward declarations
+struct random_access_universal_iterator_tag;
+struct input_universal_iterator_tag;
+struct output_universal_iterator_tag;
+
+namespace detail
+{
+
+// forward declaration
+template <typename> struct is_iterator_system;
+
+template <typename> struct device_iterator_category_to_backend_system;
+
+// XXX this should work entirely differently
+// we should just specialize this metafunction for iterator_category_with_system_and_traversal
+template<typename Category>
+  struct iterator_category_to_system
+    // convertible to any iterator?
+    : eval_if<
+        or_<
+          is_convertible<Category, thrust::input_universal_iterator_tag>,
+          is_convertible<Category, thrust::output_universal_iterator_tag>
+        >::value,
+
+        detail::identity_<thrust::any_system_tag>,
+
+        // convertible to host iterator?
+        eval_if<
+          or_<
+            is_convertible<Category, thrust::input_host_iterator_tag>,
+            is_convertible<Category, thrust::output_host_iterator_tag>
+          >::value,
+
+          detail::identity_<thrust::host_system_tag>,
+          
+          // convertible to device iterator?
+          eval_if<
+            or_<
+              is_convertible<Category, thrust::input_device_iterator_tag>,
+              is_convertible<Category, thrust::output_device_iterator_tag>
+            >::value,
+
+            detail::identity_<thrust::device_system_tag>,
+
+            // unknown system
+            detail::identity_<void>
+          > // if device
+        > // if host
+      > // if any
+{
+}; // end iterator_category_to_system
+
+
+template<typename CategoryOrTraversal>
+  struct iterator_category_or_traversal_to_system
+    : eval_if<
+        is_iterator_system<CategoryOrTraversal>::value,
+        detail::identity_<CategoryOrTraversal>,
+        iterator_category_to_system<CategoryOrTraversal>
+      >
+{
+}; // end iterator_category_or_traversal_to_system
+
+} // end detail
+} // end thrust
+
diff --git a/compat/thrust/iterator/detail/iterator_category_to_traversal.h b/compat/thrust/iterator/detail/iterator_category_to_traversal.h
new file mode 100644
index 0000000000..04ef60c0c2
--- /dev/null
+++ b/compat/thrust/iterator/detail/iterator_category_to_traversal.h
@@ -0,0 +1,178 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/iterator/iterator_categories.h>
+#include <thrust/iterator/detail/iterator_traversal_tags.h>
+#include <thrust/iterator/detail/iterator_category_to_system.h>
+#include <thrust/detail/type_traits.h>
+
+namespace thrust
+{
+
+// XXX WAR circular #inclusion with these forward declarations
+struct bidirectional_universal_iterator_tag;
+struct forward_universal_iterator_tag;
+
+namespace detail
+{
+
+// forward declarations
+template <typename> struct is_iterator_system;
+template <typename> struct is_iterator_traversal;
+
+// make type_traits easy to access
+using namespace thrust::detail;
+
+template <typename Category>
+  struct host_system_category_to_traversal
+    : eval_if<
+        is_convertible<Category, random_access_host_iterator_tag>::value,
+        detail::identity_<random_access_traversal_tag>,
+        eval_if<
+          is_convertible<Category, bidirectional_host_iterator_tag>::value,
+          detail::identity_<bidirectional_traversal_tag>,
+          eval_if<
+            is_convertible<Category, forward_host_iterator_tag>::value,
+            detail::identity_<forward_traversal_tag>,
+            eval_if<
+              is_convertible<Category, input_host_iterator_tag>::value,
+              detail::identity_<single_pass_traversal_tag>,
+              eval_if<
+                is_convertible<Category, output_host_iterator_tag>::value,
+                detail::identity_<incrementable_traversal_tag>,
+                void
+              >
+            >
+          >
+        >
+      >
+{
+}; // end host_system_category_to_traversal
+
+
+
+template <typename Category>
+  struct device_system_category_to_traversal
+    : eval_if<
+        is_convertible<Category, random_access_device_iterator_tag>::value,
+        detail::identity_<random_access_traversal_tag>,
+        eval_if<
+          is_convertible<Category, bidirectional_device_iterator_tag>::value,
+          detail::identity_<bidirectional_traversal_tag>,
+          eval_if<
+            is_convertible<Category, forward_device_iterator_tag>::value,
+            detail::identity_<forward_traversal_tag>,
+            eval_if<
+              is_convertible<Category, input_device_iterator_tag>::value,
+              detail::identity_<single_pass_traversal_tag>,
+              eval_if<
+                is_convertible<Category, output_device_iterator_tag>::value,
+                detail::identity_<incrementable_traversal_tag>,
+                void
+              >
+            >
+          >
+        >
+      >
+{
+}; // end device_system_category_to_traversal
+
+
+
+template <typename Category>
+  struct any_system_category_to_traversal
+    : eval_if<
+        is_convertible<Category, random_access_universal_iterator_tag>::value,
+        identity_<random_access_traversal_tag>,
+        eval_if<
+          is_convertible<Category, bidirectional_universal_iterator_tag>::value,
+          identity_<bidirectional_traversal_tag>,
+          eval_if<
+            is_convertible<Category, forward_universal_iterator_tag>::value,
+            identity_<forward_traversal_tag>,
+            eval_if<
+              is_convertible<Category, input_universal_iterator_tag>::value,
+              identity_<single_pass_traversal_tag>,
+              eval_if<
+                is_convertible<Category, output_universal_iterator_tag>::value,
+                identity_<incrementable_traversal_tag>,
+
+                // unknown traversal
+                void
+              >
+            >
+          >
+        >
+      >
+{
+}; // end any_system_category_to_traversal
+
+
+template<typename Category>
+  struct category_to_traversal
+      // check for any system
+    : eval_if<
+        or_<
+          is_convertible<Category, thrust::input_universal_iterator_tag>,
+          is_convertible<Category, thrust::output_universal_iterator_tag>
+        >::value,
+
+        any_system_category_to_traversal<Category>,
+
+        // check for host system
+        eval_if<
+          or_<
+            is_convertible<Category, thrust::input_host_iterator_tag>,
+            is_convertible<Category, thrust::output_host_iterator_tag>
+          >::value,
+
+          host_system_category_to_traversal<Category>,
+
+          // check for device system
+          eval_if<
+            or_<
+              is_convertible<Category, thrust::input_device_iterator_tag>,
+              is_convertible<Category, thrust::output_device_iterator_tag>
+            >::value,
+
+            device_system_category_to_traversal<Category>,
+
+            // unknown category
+            void
+          >
+        >
+      >
+{};
+
+
+template <typename CategoryOrTraversal>
+  struct iterator_category_to_traversal
+    : eval_if<
+        is_iterator_traversal<CategoryOrTraversal>::value,
+        detail::identity_<CategoryOrTraversal>,
+        category_to_traversal<CategoryOrTraversal>
+      >
+{
+}; // end iterator_category_to_traversal
+
+
+} // end detail
+
+} // end thrust
+
diff --git a/compat/thrust/iterator/detail/iterator_facade_category.h b/compat/thrust/iterator/detail/iterator_facade_category.h
new file mode 100644
index 0000000000..fbb8bd6451
--- /dev/null
+++ b/compat/thrust/iterator/detail/iterator_facade_category.h
@@ -0,0 +1,283 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/type_traits.h>
+#include <thrust/iterator/detail/host_system_tag.h>
+#include <thrust/iterator/detail/device_system_tag.h>
+#include <thrust/iterator/detail/any_system_tag.h>
+#include <thrust/iterator/iterator_categories.h>
+#include <thrust/iterator/detail/iterator_traversal_tags.h>
+#include <thrust/iterator/detail/is_iterator_category.h>
+#include <thrust/iterator/detail/iterator_category_to_traversal.h>
+
+namespace thrust
+{
+
+namespace detail
+{
+
+template<typename Category, typename System, typename Traversal>
+  struct iterator_category_with_system_and_traversal
+    : Category
+{
+}; // end iterator_category_with_system_and_traversal
+
+// specialize iterator_category_to_system for iterator_category_with_system_and_traversal
+template<typename Category> struct iterator_category_to_system;
+
+template<typename Category, typename System, typename Traversal>
+  struct iterator_category_to_system<iterator_category_with_system_and_traversal<Category,System,Traversal> >
+{
+  typedef System type;
+}; // end iterator_category_with_system_and_traversal
+
+
+// adapted from http://www.boost.org/doc/libs/1_37_0/libs/iterator/doc/iterator_facade.html#iterator-category
+//
+// in our implementation, R need not be a reference type to result in a category
+// derived from forward_XXX_iterator_tag
+//
+// iterator-category(T,V,R) :=
+//   if(T is convertible to input_host_iterator_tag
+//      || T is convertible to output_host_iterator_tag
+//      || T is convertible to input_device_iterator_tag
+//      || T is convertible to output_device_iterator_tag
+//   )
+//     return T
+//
+//   else if (T is not convertible to incrementable_traversal_tag)
+//     the program is ill-formed
+//
+//   else return a type X satisfying the following two constraints:
+//
+//     1. X is convertible to X1, and not to any more-derived
+//        type, where X1 is defined by:
+//
+//        if (T is convertible to forward_traversal_tag)
+//        {
+//          if (T is convertible to random_access_traversal_tag)
+//            X1 = random_access_host_iterator_tag
+//          else if (T is convertible to bidirectional_traversal_tag)
+//            X1 = bidirectional_host_iterator_tag
+//          else
+//            X1 = forward_host_iterator_tag
+//        }
+//        else
+//        {
+//          if (T is convertible to single_pass_traversal_tag
+//              && R is convertible to V)
+//            X1 = input_host_iterator_tag
+//          else
+//            X1 = T
+//        }
+//
+//     2. category-to-traversal(X) is convertible to the most
+//        derived traversal tag type to which X is also convertible,
+//        and not to any more-derived traversal tag type.
+
+
+template<typename System, typename Traversal, typename ValueParam, typename Reference>
+  struct iterator_facade_default_category;
+
+
+// Thrust's implementation of iterator_facade_default_category is slightly
+// different from Boost's equivalent.
+// Thrust does not check is_convertible<Reference, ValueParam> because Reference
+// may not be a complete type at this point, and implementations of is_convertible
+// typically require that both types be complete.
+// Instead, it simply assumes that if is_convertible<Traversal, single_pass_traversal_tag>,
+// then the category is input_iterator_tag
+
+
+// this is the function for standard system iterators
+template<typename Traversal, typename ValueParam, typename Reference>
+  struct iterator_facade_default_category_std :
+    thrust::detail::eval_if<
+      thrust::detail::is_convertible<Traversal, thrust::forward_traversal_tag>::value,
+      thrust::detail::eval_if<
+        thrust::detail::is_convertible<Traversal, thrust::random_access_traversal_tag>::value,
+        thrust::detail::identity_<std::random_access_iterator_tag>,
+        thrust::detail::eval_if<
+          thrust::detail::is_convertible<Traversal, thrust::bidirectional_traversal_tag>::value,
+          thrust::detail::identity_<std::bidirectional_iterator_tag>,
+          thrust::detail::identity_<std::forward_iterator_tag>
+        >
+      >,
+      thrust::detail::eval_if< // XXX note we differ from Boost here
+        thrust::detail::is_convertible<Traversal, thrust::single_pass_traversal_tag>::value,
+        thrust::detail::identity_<std::input_iterator_tag>,
+        thrust::detail::identity_<Traversal>
+      >
+    >
+{
+}; // end iterator_facade_default_category_std
+
+
+// this is the function for host system iterators
+template<typename Traversal, typename ValueParam, typename Reference>
+  struct iterator_facade_default_category_host :
+    thrust::detail::eval_if<
+      thrust::detail::is_convertible<Traversal, thrust::forward_traversal_tag>::value,
+      thrust::detail::eval_if<
+        thrust::detail::is_convertible<Traversal, thrust::random_access_traversal_tag>::value,
+        thrust::detail::identity_<thrust::random_access_host_iterator_tag>,
+        thrust::detail::eval_if<
+          thrust::detail::is_convertible<Traversal, thrust::bidirectional_traversal_tag>::value,
+          thrust::detail::identity_<thrust::bidirectional_host_iterator_tag>,
+          thrust::detail::identity_<thrust::forward_host_iterator_tag>
+        >
+      >,
+      thrust::detail::eval_if< // XXX note we differ from Boost here
+        thrust::detail::is_convertible<Traversal, thrust::single_pass_traversal_tag>::value,
+        thrust::detail::identity_<thrust::input_host_iterator_tag>,
+        thrust::detail::identity_<Traversal>
+      >
+    >
+{
+}; // end iterator_facade_default_category_host
+
+
+// this is the function for device system iterators
+template<typename Traversal, typename ValueParam, typename Reference>
+  struct iterator_facade_default_category_device :
+    thrust::detail::eval_if<
+      thrust::detail::is_convertible<Traversal, thrust::forward_traversal_tag>::value,
+      thrust::detail::eval_if<
+        thrust::detail::is_convertible<Traversal, thrust::random_access_traversal_tag>::value,
+        thrust::detail::identity_<thrust::random_access_device_iterator_tag>,
+        thrust::detail::eval_if<
+          thrust::detail::is_convertible<Traversal, thrust::bidirectional_traversal_tag>::value,
+          thrust::detail::identity_<thrust::bidirectional_device_iterator_tag>,
+          thrust::detail::identity_<thrust::forward_device_iterator_tag>
+        >
+      >,
+      thrust::detail::eval_if<
+        thrust::detail::is_convertible<Traversal, thrust::single_pass_traversal_tag>::value, // XXX note we differ from Boost here
+        thrust::detail::identity_<thrust::input_device_iterator_tag>,
+        thrust::detail::identity_<Traversal>
+      >
+    >
+{
+}; // end iterator_facade_default_category_device
+
+
+// this is the function for any system iterators
+template<typename Traversal, typename ValueParam, typename Reference>
+  struct iterator_facade_default_category_any :
+    thrust::detail::eval_if<
+
+      thrust::detail::is_convertible<Traversal, thrust::forward_traversal_tag>::value,
+
+      thrust::detail::eval_if<
+        thrust::detail::is_convertible<Traversal, thrust::random_access_traversal_tag>::value,
+        thrust::detail::identity_<thrust::random_access_universal_iterator_tag>,
+
+        thrust::detail::eval_if<
+          thrust::detail::is_convertible<Traversal, thrust::bidirectional_traversal_tag>::value,
+          thrust::detail::identity_<thrust::bidirectional_universal_iterator_tag>,
+          thrust::detail::identity_<thrust::forward_universal_iterator_tag>
+        >
+      >,
+
+      thrust::detail::eval_if<
+        thrust::detail::is_convertible<Traversal, thrust::single_pass_traversal_tag>::value, // XXX note we differ from Boost here
+        thrust::detail::identity_<thrust::input_universal_iterator_tag>,
+        thrust::detail::identity_<Traversal>
+      >
+    >
+{
+}; // end iterator_facade_default_category_any
+
+
+template<typename System, typename Traversal, typename ValueParam, typename Reference>
+  struct iterator_facade_default_category
+      // check for any system
+    : thrust::detail::eval_if<
+        thrust::detail::is_convertible<System, thrust::any_system_tag>::value,
+        iterator_facade_default_category_any<Traversal, ValueParam, Reference>,
+
+        // check for host system
+        thrust::detail::eval_if<
+          thrust::detail::is_convertible<System, thrust::host_system_tag>::value,
+          iterator_facade_default_category_host<Traversal, ValueParam, Reference>,
+
+          // check for device system
+          thrust::detail::eval_if<
+            thrust::detail::is_convertible<System, thrust::device_system_tag>::value,
+            iterator_facade_default_category_device<Traversal, ValueParam, Reference>,
+
+            // if we don't recognize the system, get a standard iterator category
+            // and combine it with System & Traversal
+            thrust::detail::identity_<
+              thrust::detail::iterator_category_with_system_and_traversal<
+                typename iterator_facade_default_category_std<Traversal, ValueParam, Reference>::type,
+                System,
+                Traversal
+              >
+            >
+          >
+        >
+      >
+{};
+
+
+template<typename System, typename Traversal, typename ValueParam, typename Reference>
+  struct iterator_facade_category_impl
+{
+  typedef typename iterator_facade_default_category<
+    System,Traversal,ValueParam,Reference
+  >::type category;
+
+  // we must be able to deduce both Traversal & System from category
+  // otherwise, munge them all together
+  typedef typename thrust::detail::eval_if<
+    thrust::detail::and_<
+      thrust::detail::is_same<
+        Traversal,
+        typename thrust::detail::iterator_category_to_traversal<category>::type
+      >,
+      thrust::detail::is_same<
+        System,
+        typename thrust::detail::iterator_category_to_system<category>::type
+      >
+    >::value,
+    thrust::detail::identity_<category>,
+    thrust::detail::identity_<thrust::detail::iterator_category_with_system_and_traversal<category,System,Traversal> >
+  >::type type;
+}; // end iterator_facade_category_impl
+
+
+template<typename CategoryOrSystem,
+         typename CategoryOrTraversal,
+         typename ValueParam,
+         typename Reference>
+  struct iterator_facade_category
+{
+  typedef typename
+  thrust::detail::eval_if<
+    thrust::detail::is_iterator_category<CategoryOrTraversal>::value,
+    thrust::detail::identity_<CategoryOrTraversal>, // categories are fine as-is
+    iterator_facade_category_impl<CategoryOrSystem, CategoryOrTraversal, ValueParam, Reference>
+  >::type type;
+}; // end iterator_facade_category
+
+
+} // end detail
+} // end thrust
+
diff --git a/compat/thrust/iterator/detail/iterator_traits.inl b/compat/thrust/iterator/detail/iterator_traits.inl
new file mode 100644
index 0000000000..924eabb187
--- /dev/null
+++ b/compat/thrust/iterator/detail/iterator_traits.inl
@@ -0,0 +1,112 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file iterator_traits.inl
+ *  \brief Inline file for iterator_traits.h.
+ */
+
+#include <thrust/iterator/iterator_categories.h>
+#include <thrust/iterator/detail/iterator_category_to_traversal.h>
+#include <thrust/detail/type_traits.h>
+
+namespace thrust
+{
+
+template<typename Iterator>
+  struct iterator_value
+{
+  typedef typename thrust::iterator_traits<Iterator>::value_type type;
+}; // end iterator_value
+
+
+template<typename Iterator>
+  struct iterator_pointer
+{
+  typedef typename thrust::iterator_traits<Iterator>::pointer type;
+}; // end iterator_pointer
+
+
+template<typename Iterator>
+  struct iterator_reference
+{
+  typedef typename iterator_traits<Iterator>::reference type;
+}; // end iterator_reference
+
+
+template<typename Iterator>
+  struct iterator_difference
+{
+  typedef typename thrust::iterator_traits<Iterator>::difference_type type;
+}; // end iterator_difference
+
+
+template<typename Iterator>
+  struct iterator_system
+    : detail::iterator_category_to_system<
+        typename thrust::iterator_traits<Iterator>::iterator_category
+      >
+{
+}; // end iterator_system
+
+// specialize iterator_system for void *, which has no category
+template<>
+  struct iterator_system<void *>
+{
+  typedef thrust::iterator_system<int*>::type type;
+}; // end iterator_system<void*>
+
+template<>
+  struct iterator_system<const void *>
+{
+  typedef thrust::iterator_system<const int*>::type type;
+}; // end iterator_system<void*>
+
+
+template <typename Iterator>
+  struct iterator_traversal
+    : detail::iterator_category_to_traversal<
+        typename thrust::iterator_traits<Iterator>::iterator_category
+      >
+{
+}; // end iterator_traversal
+
+namespace detail
+{
+
+template <typename T>
+  struct is_iterator_traversal
+    : thrust::detail::is_convertible<T, incrementable_traversal_tag>
+{
+}; // end is_iterator_traversal
+
+
+template<typename T>
+  struct is_iterator_system
+    : detail::or_<
+        detail::is_convertible<T, any_system_tag>,
+        detail::or_<
+          detail::is_convertible<T, host_system_tag>,
+          detail::is_convertible<T, device_system_tag>
+        >
+      >
+{
+}; // end is_iterator_system
+
+
+} // end namespace detail
+} // end namespace thrust
+
diff --git a/compat/thrust/iterator/detail/iterator_traversal_tags.h b/compat/thrust/iterator/detail/iterator_traversal_tags.h
new file mode 100644
index 0000000000..dcbebf3fda
--- /dev/null
+++ b/compat/thrust/iterator/detail/iterator_traversal_tags.h
@@ -0,0 +1,41 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+namespace thrust
+{
+
+// define Boost's traversal tags
+struct no_traversal_tag {};
+
+struct incrementable_traversal_tag
+  : no_traversal_tag {};
+
+struct single_pass_traversal_tag
+  : incrementable_traversal_tag {};
+
+struct forward_traversal_tag
+  : single_pass_traversal_tag {};
+
+struct bidirectional_traversal_tag
+  : forward_traversal_tag {};
+
+struct random_access_traversal_tag
+  : bidirectional_traversal_tag {};
+
+} // end thrust
+
diff --git a/compat/thrust/iterator/detail/minimum_category.h b/compat/thrust/iterator/detail/minimum_category.h
new file mode 100644
index 0000000000..e07e09636e
--- /dev/null
+++ b/compat/thrust/iterator/detail/minimum_category.h
@@ -0,0 +1,52 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/type_traits/minimum_type.h>
+
+namespace thrust
+{
+
+namespace detail
+{ 
+
+template<typename T1,
+         typename T2  = minimum_type_detail::any_conversion,
+         typename T3  = minimum_type_detail::any_conversion,
+         typename T4  = minimum_type_detail::any_conversion,
+         typename T5  = minimum_type_detail::any_conversion,
+         typename T6  = minimum_type_detail::any_conversion,
+         typename T7  = minimum_type_detail::any_conversion,
+         typename T8  = minimum_type_detail::any_conversion,
+         typename T9  = minimum_type_detail::any_conversion,
+         typename T10 = minimum_type_detail::any_conversion,
+         typename T11 = minimum_type_detail::any_conversion,
+         typename T12 = minimum_type_detail::any_conversion,
+         typename T13 = minimum_type_detail::any_conversion,
+         typename T14 = minimum_type_detail::any_conversion,
+         typename T15 = minimum_type_detail::any_conversion,
+         typename T16 = minimum_type_detail::any_conversion>
+  struct minimum_category
+    : minimum_type<T1,T2,T3,T4,T5,T6,T7,T8,T9,T10,T11,T12,T13,T14,T15,T16>
+{
+}; // end minimum_category
+
+} // end detail
+
+} // end thrust
+
+
diff --git a/compat/thrust/iterator/detail/minimum_system.h b/compat/thrust/iterator/detail/minimum_system.h
new file mode 100644
index 0000000000..5448a0d1f0
--- /dev/null
+++ b/compat/thrust/iterator/detail/minimum_system.h
@@ -0,0 +1,49 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/type_traits/minimum_type.h>
+
+namespace thrust
+{
+namespace detail
+{ 
+
+template<typename T1,
+         typename T2  = minimum_type_detail::any_conversion,
+         typename T3  = minimum_type_detail::any_conversion,
+         typename T4  = minimum_type_detail::any_conversion,
+         typename T5  = minimum_type_detail::any_conversion,
+         typename T6  = minimum_type_detail::any_conversion,
+         typename T7  = minimum_type_detail::any_conversion,
+         typename T8  = minimum_type_detail::any_conversion,
+         typename T9  = minimum_type_detail::any_conversion,
+         typename T10 = minimum_type_detail::any_conversion,
+         typename T11 = minimum_type_detail::any_conversion,
+         typename T12 = minimum_type_detail::any_conversion,
+         typename T13 = minimum_type_detail::any_conversion,
+         typename T14 = minimum_type_detail::any_conversion,
+         typename T15 = minimum_type_detail::any_conversion,
+         typename T16 = minimum_type_detail::any_conversion>
+  struct minimum_system
+    : minimum_type<T1,T2,T3,T4,T5,T6,T7,T8,T9,T10,T11,T12,T13,T14,T15,T16>
+{
+}; // end minimum_system
+
+} // end detail
+} // end thrust
+
diff --git a/compat/thrust/iterator/detail/normal_iterator.h b/compat/thrust/iterator/detail/normal_iterator.h
new file mode 100644
index 0000000000..7fe61bfed3
--- /dev/null
+++ b/compat/thrust/iterator/detail/normal_iterator.h
@@ -0,0 +1,76 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file normal_iterator.h
+ *  \brief Defines the interface to an iterator class
+ *         which adapts a pointer type.
+ */
+
+#pragma once
+
+#include <thrust/iterator/iterator_adaptor.h>
+#include <thrust/iterator/detail/is_trivial_iterator.h>
+#include <thrust/detail/type_traits.h>
+
+namespace thrust
+{
+namespace detail
+{
+
+
+template<typename Pointer>
+  class normal_iterator
+    : public iterator_adaptor<
+        normal_iterator<Pointer>,
+        Pointer
+      >
+{
+  typedef iterator_adaptor<normal_iterator<Pointer>, Pointer> super_t;
+
+  public:
+    __host__ __device__
+    normal_iterator() {}
+
+    __host__ __device__
+    normal_iterator(Pointer p)
+      : super_t(p) {}
+    
+    template<typename OtherPointer>
+    __host__ __device__
+    normal_iterator(const normal_iterator<OtherPointer> &other,
+                    typename thrust::detail::enable_if_convertible<
+                      OtherPointer,
+                      Pointer
+                    >::type * = 0)
+      : super_t(other.base()) {}
+
+}; // end normal_iterator
+
+
+template<typename Pointer>
+  inline __host__ __device__ normal_iterator<Pointer> make_normal_iterator(Pointer ptr)
+{
+  return normal_iterator<Pointer>(ptr);
+}
+
+
+template<typename T> struct is_trivial_iterator< normal_iterator<T> > : public true_type {};
+
+
+} // end detail
+} // end thrust
+
diff --git a/compat/thrust/iterator/detail/permutation_iterator_base.h b/compat/thrust/iterator/detail/permutation_iterator_base.h
new file mode 100644
index 0000000000..a145b88aec
--- /dev/null
+++ b/compat/thrust/iterator/detail/permutation_iterator_base.h
@@ -0,0 +1,53 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/iterator/iterator_adaptor.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/detail/type_traits.h>
+#include <thrust/iterator/detail/minimum_system.h>
+
+namespace thrust
+{
+
+template<typename,typename> class permutation_iterator;
+
+
+namespace detail
+{
+
+template<typename ElementIterator,
+         typename IndexIterator>
+  struct permutation_iterator_base
+{
+  typedef typename thrust::iterator_system<ElementIterator>::type System1;
+  typedef typename thrust::iterator_system<IndexIterator>::type System2;
+
+  typedef thrust::iterator_adaptor<
+    permutation_iterator<ElementIterator,IndexIterator>,
+    IndexIterator,
+    typename thrust::iterator_value<ElementIterator>::type,
+    typename detail::minimum_system<System1,System2>::type,
+    thrust::use_default,
+    typename thrust::iterator_reference<ElementIterator>::type
+  > type;
+}; // end permutation_iterator_base
+
+} // end detail
+
+} // end thrust
+
diff --git a/compat/thrust/iterator/detail/retag.h b/compat/thrust/iterator/detail/retag.h
new file mode 100644
index 0000000000..4417fa5604
--- /dev/null
+++ b/compat/thrust/iterator/detail/retag.h
@@ -0,0 +1,140 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/type_traits.h>
+#include <thrust/iterator/detail/tagged_iterator.h>
+#include <thrust/detail/pointer.h>
+
+namespace thrust
+{
+namespace detail
+{
+
+
+// we can retag an iterator if FromTag converts to ToTag
+// or vice versa
+template<typename FromTag, typename ToTag>
+  struct is_retaggable
+    : integral_constant<
+        bool,
+        (is_convertible<FromTag,ToTag>::value || is_convertible<ToTag,FromTag>::value)
+      >
+{};
+
+
+template<typename FromTag, typename ToTag, typename Result>
+  struct enable_if_retaggable
+    : enable_if<
+        is_retaggable<FromTag,ToTag>::value,
+        Result
+      >
+{}; // end enable_if_retaggable
+
+
+} // end detail
+
+
+template<typename Tag, typename Iterator>
+  thrust::detail::tagged_iterator<Iterator,Tag>
+    reinterpret_tag(Iterator iter)
+{
+  return thrust::detail::tagged_iterator<Iterator,Tag>(iter);
+} // end reinterpret_tag()
+
+
+// specialization for raw pointer
+template<typename Tag, typename T>
+  thrust::pointer<T,Tag>
+    reinterpret_tag(T *ptr)
+{
+  return thrust::pointer<T,Tag>(ptr);
+} // end reinterpret_tag()
+
+
+// specialization for thrust::pointer
+template<typename Tag, typename T, typename OtherTag, typename Reference, typename Derived>
+  thrust::pointer<T,Tag>
+    reinterpret_tag(thrust::pointer<T,OtherTag,Reference,Derived> ptr)
+{
+  return reinterpret_tag<Tag>(ptr.get());
+} // end reinterpret_tag()
+
+
+// avoid deeply-nested tagged_iterator
+template<typename Tag, typename BaseIterator, typename OtherTag>
+  thrust::detail::tagged_iterator<BaseIterator,Tag>
+    reinterpret_tag(thrust::detail::tagged_iterator<BaseIterator,OtherTag> iter)
+{
+  return reinterpret_tag<Tag>(iter.base());
+} // end reinterpret_tag()
+
+
+template<typename Tag, typename Iterator>
+  typename thrust::detail::enable_if_retaggable<
+    typename thrust::iterator_system<Iterator>::type,
+    Tag,
+    thrust::detail::tagged_iterator<Iterator,Tag>
+  >::type
+    retag(Iterator iter)
+{
+  return reinterpret_tag<Tag>(iter);
+} // end retag()
+
+
+// specialization for raw pointer
+template<typename Tag, typename T>
+  typename thrust::detail::enable_if_retaggable<
+    typename thrust::iterator_system<T*>::type,
+    Tag,
+    thrust::pointer<T,Tag>
+  >::type
+    retag(T *ptr)
+{
+  return reinterpret_tag<Tag>(ptr);
+} // end retag()
+
+
+// specialization for thrust::pointer
+template<typename Tag, typename T, typename OtherTag>
+  typename thrust::detail::enable_if_retaggable<
+    OtherTag,
+    Tag,
+    thrust::pointer<T,Tag>
+  >::type
+    retag(thrust::pointer<T,OtherTag> ptr)
+{
+  return reinterpret_tag<Tag>(ptr);
+} // end retag()
+
+
+// avoid deeply-nested tagged_iterator
+template<typename Tag, typename BaseIterator, typename OtherTag>
+  typename thrust::detail::enable_if_retaggable<
+    OtherTag,
+    Tag,
+    thrust::detail::tagged_iterator<BaseIterator,Tag>
+  >::type
+    retag(thrust::detail::tagged_iterator<BaseIterator,OtherTag> iter)
+{
+  return reinterpret_tag<Tag>(iter);
+} // end retag()
+
+
+} // end thrust
+
diff --git a/compat/thrust/iterator/detail/reverse_iterator.inl b/compat/thrust/iterator/detail/reverse_iterator.inl
new file mode 100644
index 0000000000..03e9032130
--- /dev/null
+++ b/compat/thrust/iterator/detail/reverse_iterator.inl
@@ -0,0 +1,108 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/iterator/reverse_iterator.h>
+#include <thrust/iterator/iterator_traits.h>
+
+namespace thrust
+{
+
+namespace detail
+{
+
+__thrust_hd_warning_disable__
+template<typename Iterator>
+__host__ __device__
+  Iterator prior(Iterator x)
+{
+  return --x;
+} // end prior()
+
+} // end detail
+
+template<typename BidirectionalIterator>
+  reverse_iterator<BidirectionalIterator>
+    ::reverse_iterator(BidirectionalIterator x)
+      :super_t(x)
+{
+} // end reverse_iterator::reverse_iterator()
+
+template<typename BidirectionalIterator>
+  template<typename OtherBidirectionalIterator>
+    reverse_iterator<BidirectionalIterator>
+      ::reverse_iterator(reverse_iterator<OtherBidirectionalIterator> const &r
+// XXX msvc screws this up
+#ifndef _MSC_VER
+                     , typename thrust::detail::enable_if<
+                         thrust::detail::is_convertible<
+                           OtherBidirectionalIterator,
+                           BidirectionalIterator
+                         >::value
+                       >::type *
+#endif // _MSC_VER
+                     )
+        :super_t(r.base())
+{
+} // end reverse_iterator::reverse_iterator()
+
+template<typename BidirectionalIterator>
+  typename reverse_iterator<BidirectionalIterator>::super_t::reference
+    reverse_iterator<BidirectionalIterator>
+      ::dereference(void) const
+{
+  return *thrust::detail::prior(this->base());
+} // end reverse_iterator::increment()
+
+template<typename BidirectionalIterator>
+  void reverse_iterator<BidirectionalIterator>
+    ::increment(void)
+{
+  --this->base_reference();
+} // end reverse_iterator::increment()
+
+template<typename BidirectionalIterator>
+  void reverse_iterator<BidirectionalIterator>
+    ::decrement(void)
+{
+  ++this->base_reference();
+} // end reverse_iterator::decrement()
+
+template<typename BidirectionalIterator>
+  void reverse_iterator<BidirectionalIterator>
+    ::advance(typename super_t::difference_type n)
+{
+  this->base_reference() += -n;
+} // end reverse_iterator::advance()
+
+template<typename BidirectionalIterator>
+  template<typename OtherBidirectionalIterator>
+    typename reverse_iterator<BidirectionalIterator>::super_t::difference_type
+      reverse_iterator<BidirectionalIterator>
+        ::distance_to(reverse_iterator<OtherBidirectionalIterator> const &y) const
+{
+  return this->base_reference() - y.base();
+} // end reverse_iterator::distance_to()
+
+template<typename BidirectionalIterator>
+__host__ __device__
+reverse_iterator<BidirectionalIterator> make_reverse_iterator(BidirectionalIterator x)
+{
+  return reverse_iterator<BidirectionalIterator>(x);
+} // end make_reverse_iterator()
+
+
+} // end thrust
+
diff --git a/compat/thrust/iterator/detail/reverse_iterator_base.h b/compat/thrust/iterator/detail/reverse_iterator_base.h
new file mode 100644
index 0000000000..c10c5b73ff
--- /dev/null
+++ b/compat/thrust/iterator/detail/reverse_iterator_base.h
@@ -0,0 +1,42 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/iterator/iterator_adaptor.h>
+#include <thrust/iterator/iterator_traits.h>
+
+namespace thrust
+{
+
+template <typename> class reverse_iterator;
+
+namespace detail
+{
+
+template<typename BidirectionalIterator>
+  struct reverse_iterator_base
+{
+  typedef thrust::iterator_adaptor<
+    thrust::reverse_iterator<BidirectionalIterator>,
+    BidirectionalIterator
+  > type;
+}; // end reverse_iterator_base
+
+} // end detail
+
+} // end thrust
+
diff --git a/compat/thrust/iterator/detail/tagged_iterator.h b/compat/thrust/iterator/detail/tagged_iterator.h
new file mode 100644
index 0000000000..69e6445183
--- /dev/null
+++ b/compat/thrust/iterator/detail/tagged_iterator.h
@@ -0,0 +1,74 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/iterator/iterator_adaptor.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/detail/use_default.h>
+
+namespace thrust
+{
+namespace detail
+{
+
+template <typename,typename> class tagged_iterator;
+
+template<typename Iterator, typename Tag>
+  struct tagged_iterator_base
+{
+  typedef thrust::iterator_adaptor<
+    tagged_iterator<Iterator,Tag>,
+    Iterator,
+    typename thrust::iterator_value<Iterator>::type,
+    Tag,
+    typename thrust::iterator_traversal<Iterator>::type,
+    typename thrust::iterator_reference<Iterator>::type,
+    typename thrust::iterator_difference<Iterator>::type
+  > type;
+}; // end tagged_iterator_base
+
+template<typename Iterator, typename Tag>
+  class tagged_iterator
+    : public tagged_iterator_base<Iterator,Tag>::type
+{
+  private:
+    typedef typename tagged_iterator_base<Iterator,Tag>::type super_t;
+
+  public:
+    __host__ __device__
+    tagged_iterator(void) {}
+
+    __host__ __device__
+    explicit tagged_iterator(Iterator x)
+      : super_t(x) {}
+}; // end tagged_iterator
+
+
+// specialize is_trivial_iterator for tagged_iterator
+template<typename> struct is_trivial_iterator;
+
+// tagged_iterator is trivial if its base iterator is
+template<typename BaseIterator, typename Tag>
+  struct is_trivial_iterator<tagged_iterator<BaseIterator,Tag> >
+    : is_trivial_iterator<BaseIterator>
+{};
+
+
+} // end detail
+} // end thrust
+
diff --git a/compat/thrust/iterator/detail/transform_iterator.inl b/compat/thrust/iterator/detail/transform_iterator.inl
new file mode 100644
index 0000000000..a5a36a78be
--- /dev/null
+++ b/compat/thrust/iterator/detail/transform_iterator.inl
@@ -0,0 +1,72 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/iterator/transform_iterator.h>
+#include <thrust/iterator/iterator_adaptor.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/detail/type_traits.h>
+#include <thrust/detail/type_traits/result_of.h>
+
+namespace thrust
+{
+
+template <class UnaryFunction, class Iterator, class Reference, class Value>
+  class transform_iterator;
+  
+namespace detail 
+{
+
+// Compute the iterator_adaptor instantiation to be used for transform_iterator
+template <class UnaryFunc, class Iterator, class Reference, class Value>
+struct transform_iterator_base
+{
+ private:
+    // By default, dereferencing the iterator yields the same as the function.
+    typedef typename thrust::detail::ia_dflt_help<
+      Reference,
+      thrust::detail::result_of<UnaryFunc(typename thrust::iterator_value<Iterator>::type)>
+    >::type reference;
+
+    // To get the default for Value: remove any reference on the
+    // result type, but retain any constness to signal
+    // non-writability.  Note that if we adopt Thomas' suggestion
+    // to key non-writability *only* on the Reference argument,
+    // we'd need to strip constness here as well.
+    typedef typename thrust::detail::ia_dflt_help<
+      Value,
+      thrust::detail::remove_reference<reference>
+    >::type cv_value_type;
+
+ public:
+    typedef thrust::iterator_adaptor
+    <
+        transform_iterator<UnaryFunc, Iterator, Reference, Value>
+      , Iterator
+      , cv_value_type
+      , thrust::use_default   // Leave the system alone
+        //, thrust::use_default   // Leave the traversal alone
+        // use the Iterator's category to let any system iterators remain random access even though
+        // transform_iterator's reference type may not be a reference
+        // XXX figure out why only iterators whose reference types are true references are random access
+        , typename thrust::iterator_traits<Iterator>::iterator_category
+      , reference
+    > type;
+};
+
+
+} // end detail
+} // end thrust
+
diff --git a/compat/thrust/iterator/detail/tuple_of_iterator_references.h b/compat/thrust/iterator/detail/tuple_of_iterator_references.h
new file mode 100644
index 0000000000..fdbf6b8f66
--- /dev/null
+++ b/compat/thrust/iterator/detail/tuple_of_iterator_references.h
@@ -0,0 +1,246 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/tuple.h>
+#include <thrust/pair.h>
+#include <thrust/detail/reference_forward_declaration.h>
+
+namespace thrust
+{
+namespace detail
+{
+
+  
+template<
+  typename T0, typename T1, typename T2,
+  typename T3, typename T4, typename T5,
+  typename T6, typename T7, typename T8,
+  typename T9
+>
+  class tuple_of_iterator_references
+    : public thrust::tuple<T0,T1,T2,T3,T4,T5,T6,T7,T8,T9>
+{
+  private:
+    typedef thrust::tuple<T0,T1,T2,T3,T4,T5,T6,T7,T8,T9> super_t;
+
+  public:
+    // allow implicit construction from tuple<refs>
+    inline __host__ __device__
+    tuple_of_iterator_references(const super_t &other)
+      : super_t(other)
+    {}
+
+    // allow assignment from tuples
+    // XXX might be worthwhile to guard this with an enable_if is_assignable
+    template<typename U1, typename U2>
+    inline __host__ __device__
+    tuple_of_iterator_references &operator=(const detail::cons<U1,U2> &other)
+    {
+      super_t::operator=(other);
+      return *this;
+    }
+
+    // allow assignment from pairs
+    // XXX might be worthwhile to guard this with an enable_if is_assignable
+    template<typename U1, typename U2>
+    inline __host__ __device__
+    tuple_of_iterator_references &operator=(const thrust::pair<U1,U2> &other)
+    {
+      super_t::operator=(other);
+      return *this;
+    }
+
+    // allow assignment from reference<tuple>
+    // XXX perhaps we should generalize to reference<T>
+    //     we could captures reference<pair> this way
+    template<typename U0, typename U1, typename U2,
+             typename U3, typename U4, typename U5,
+             typename U6, typename U7, typename U8,
+             typename U9,
+             typename Pointer, typename Derived>
+    inline __host__ __device__
+// XXX gcc-4.2 crashes on is_assignable
+//    typename thrust::detail::enable_if<
+//      thrust::detail::is_assignable<
+//        super_t,
+//        const thrust::tuple<U0,U1,U2,U3,U4,U5,U6,U7,U8,U9>
+//      >::value,
+//      tuple_of_iterator_references &
+//    >::type
+    tuple_of_iterator_references &
+    operator=(const thrust::reference<thrust::tuple<U0,U1,U2,U3,U4,U5,U6,U7,U8,U9>, Pointer, Derived> &other)
+    {
+      typedef thrust::tuple<U0,U1,U2,U3,U4,U5,U6,U7,U8,U9> tuple_type;
+
+      // XXX perhaps this could be accelerated
+      tuple_type other_tuple = other;
+      super_t::operator=(other_tuple);
+      return *this;
+    }
+
+
+    // duplicate thrust::tuple's constructors
+    inline __host__ __device__
+    tuple_of_iterator_references() {}
+
+    inline __host__ __device__ 
+    tuple_of_iterator_references(typename access_traits<T0>::parameter_type t0)
+      : super_t(t0,
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type()))
+    {}
+
+    inline __host__ __device__ 
+    tuple_of_iterator_references(typename access_traits<T0>::parameter_type t0,
+                                 typename access_traits<T1>::parameter_type t1)
+      : super_t(t0, t1,
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type()))
+    {}
+
+    inline __host__ __device__ 
+    tuple_of_iterator_references(typename access_traits<T0>::parameter_type t0,
+                                 typename access_traits<T1>::parameter_type t1,
+                                 typename access_traits<T2>::parameter_type t2)
+      : super_t(t0, t1, t2,
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type()))
+    {}
+
+    inline __host__ __device__ 
+    tuple_of_iterator_references(typename access_traits<T0>::parameter_type t0,
+                                 typename access_traits<T1>::parameter_type t1,
+                                 typename access_traits<T2>::parameter_type t2,
+                                 typename access_traits<T3>::parameter_type t3)
+      : super_t(t0, t1, t2, t3,
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type()))
+    {}
+
+    inline __host__ __device__ 
+    tuple_of_iterator_references(typename access_traits<T0>::parameter_type t0,
+                                 typename access_traits<T1>::parameter_type t1,
+                                 typename access_traits<T2>::parameter_type t2,
+                                 typename access_traits<T3>::parameter_type t3,
+                                 typename access_traits<T4>::parameter_type t4)
+      : super_t(t0, t1, t2, t3, t4,
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type()))
+    {}
+
+    inline __host__ __device__ 
+    tuple_of_iterator_references(typename access_traits<T0>::parameter_type t0,
+                                 typename access_traits<T1>::parameter_type t1,
+                                 typename access_traits<T2>::parameter_type t2,
+                                 typename access_traits<T3>::parameter_type t3,
+                                 typename access_traits<T4>::parameter_type t4,
+                                 typename access_traits<T5>::parameter_type t5)
+      : super_t(t0, t1, t2, t3, t4, t5,
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type()))
+    {}
+
+    inline __host__ __device__ 
+    tuple_of_iterator_references(typename access_traits<T0>::parameter_type t0,
+                                 typename access_traits<T1>::parameter_type t1,
+                                 typename access_traits<T2>::parameter_type t2,
+                                 typename access_traits<T3>::parameter_type t3,
+                                 typename access_traits<T4>::parameter_type t4,
+                                 typename access_traits<T5>::parameter_type t5,
+                                 typename access_traits<T6>::parameter_type t6)
+      : super_t(t0, t1, t2, t3, t4, t5, t6,
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type()))
+    {}
+
+    inline __host__ __device__ 
+    tuple_of_iterator_references(typename access_traits<T0>::parameter_type t0,
+                                 typename access_traits<T1>::parameter_type t1,
+                                 typename access_traits<T2>::parameter_type t2,
+                                 typename access_traits<T3>::parameter_type t3,
+                                 typename access_traits<T4>::parameter_type t4,
+                                 typename access_traits<T5>::parameter_type t5,
+                                 typename access_traits<T6>::parameter_type t6,
+                                 typename access_traits<T7>::parameter_type t7)
+      : super_t(t0, t1, t2, t3, t4, t5, t6, t7,
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type()))
+    {}
+
+    inline __host__ __device__ 
+    tuple_of_iterator_references(typename access_traits<T0>::parameter_type t0,
+                                 typename access_traits<T1>::parameter_type t1,
+                                 typename access_traits<T2>::parameter_type t2,
+                                 typename access_traits<T3>::parameter_type t3,
+                                 typename access_traits<T4>::parameter_type t4,
+                                 typename access_traits<T5>::parameter_type t5,
+                                 typename access_traits<T6>::parameter_type t6,
+                                 typename access_traits<T7>::parameter_type t7,
+                                 typename access_traits<T8>::parameter_type t8)
+      : super_t(t0, t1, t2, t3, t4, t5, t6, t7, t8,
+                static_cast<const null_type&>(null_type()))
+    {}
+
+    inline __host__ __device__ 
+    tuple_of_iterator_references(typename access_traits<T0>::parameter_type t0,
+                                 typename access_traits<T1>::parameter_type t1,
+                                 typename access_traits<T2>::parameter_type t2,
+                                 typename access_traits<T3>::parameter_type t3,
+                                 typename access_traits<T4>::parameter_type t4,
+                                 typename access_traits<T5>::parameter_type t5,
+                                 typename access_traits<T6>::parameter_type t6,
+                                 typename access_traits<T7>::parameter_type t7,
+                                 typename access_traits<T8>::parameter_type t8,
+                                 typename access_traits<T9>::parameter_type t9)
+      : super_t(t0, t1, t2, t3, t4, t5, t6, t7, t8, t9)
+    {}
+};
+
+
+} // end detail
+} // end thrust
+
diff --git a/compat/thrust/iterator/detail/universal_categories.h b/compat/thrust/iterator/detail/universal_categories.h
new file mode 100644
index 0000000000..7c3922210c
--- /dev/null
+++ b/compat/thrust/iterator/detail/universal_categories.h
@@ -0,0 +1,85 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/iterator/iterator_categories.h>
+
+namespace thrust
+{
+
+// define these types without inheritance to avoid ambiguous conversion to base classes
+
+struct input_universal_iterator_tag
+{
+  operator input_host_iterator_tag () {return input_host_iterator_tag();}
+
+  operator input_device_iterator_tag () {return input_device_iterator_tag();}
+};
+
+struct output_universal_iterator_tag
+{
+  operator output_host_iterator_tag () {return output_host_iterator_tag();}
+
+  operator output_device_iterator_tag () {return output_device_iterator_tag();}
+};
+
+struct forward_universal_iterator_tag
+  : input_universal_iterator_tag
+{
+  operator forward_host_iterator_tag () {return forward_host_iterator_tag();};
+
+  operator forward_device_iterator_tag () {return forward_device_iterator_tag();};
+};
+
+struct bidirectional_universal_iterator_tag
+  : forward_universal_iterator_tag
+{
+  operator bidirectional_host_iterator_tag () {return bidirectional_host_iterator_tag();};
+
+  operator bidirectional_device_iterator_tag () {return bidirectional_device_iterator_tag();};
+};
+
+
+namespace detail
+{
+
+// create this struct to control conversion precedence in random_access_universal_iterator_tag
+template<typename T>
+struct one_degree_of_separation
+  : T
+{
+};
+
+} // end detail
+
+
+struct random_access_universal_iterator_tag
+{
+  // these conversions are all P0
+  operator random_access_host_iterator_tag () {return random_access_host_iterator_tag();};
+
+  operator random_access_device_iterator_tag () {return random_access_device_iterator_tag();};
+
+  // bidirectional_universal_iterator_tag is P1
+  operator detail::one_degree_of_separation<bidirectional_universal_iterator_tag> () {return detail::one_degree_of_separation<bidirectional_universal_iterator_tag>();}
+
+};
+
+
+} // end thrust
+
diff --git a/compat/thrust/iterator/detail/zip_iterator.inl b/compat/thrust/iterator/detail/zip_iterator.inl
new file mode 100644
index 0000000000..fddd0ada11
--- /dev/null
+++ b/compat/thrust/iterator/detail/zip_iterator.inl
@@ -0,0 +1,151 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/iterator/zip_iterator.h>
+#include <thrust/detail/tuple_transform.h>
+
+namespace thrust
+{
+
+
+template <typename IteratorTuple>
+  zip_iterator<IteratorTuple>
+    ::zip_iterator(void)
+{
+} // end zip_iterator::zip_iterator()
+
+
+template <typename IteratorTuple>
+  zip_iterator<IteratorTuple>
+    ::zip_iterator(IteratorTuple iterator_tuple)
+      :m_iterator_tuple(iterator_tuple)
+{
+} // end zip_iterator::zip_iterator()
+
+
+template <typename IteratorTuple>
+  template <typename OtherIteratorTuple>
+    zip_iterator<IteratorTuple>
+      ::zip_iterator(const zip_iterator<OtherIteratorTuple> &other,
+                     typename thrust::detail::enable_if_convertible<
+                       OtherIteratorTuple,
+                       IteratorTuple
+                     >::type *)
+        :m_iterator_tuple(other.get_iterator_tuple())
+{
+} // end zip_iterator::zip_iterator()
+
+
+template <typename IteratorTuple>
+const IteratorTuple &zip_iterator<IteratorTuple>
+  ::get_iterator_tuple(void) const
+{
+  return m_iterator_tuple;
+} // end zip_iterator::get_iterator_tuple()
+
+
+template <typename IteratorTuple>
+  typename zip_iterator<IteratorTuple>::super_t::reference
+    zip_iterator<IteratorTuple>
+      ::dereference(void) const
+{
+  using namespace detail::tuple_impl_specific;
+
+  return thrust::detail::tuple_host_device_transform<detail::dereference_iterator::template apply>(get_iterator_tuple(), detail::dereference_iterator());
+} // end zip_iterator::dereference()
+
+
+__thrust_hd_warning_disable__
+template <typename IteratorTuple>
+  template <typename OtherIteratorTuple>
+    bool zip_iterator<IteratorTuple>
+      ::equal(const zip_iterator<OtherIteratorTuple> &other) const
+{
+  return get<0>(get_iterator_tuple()) == get<0>(other.get_iterator_tuple());
+} // end zip_iterator::equal()
+
+
+template <typename IteratorTuple>
+  void zip_iterator<IteratorTuple>
+    ::advance(typename super_t::difference_type n)
+{
+  using namespace detail::tuple_impl_specific;
+
+  // XXX note that we use a pointer to System to dispatch to avoid
+  //     default construction of a System
+  typename thrust::iterator_system<zip_iterator>::type *use_me_to_dispatch = 0;
+
+  // dispatch on system
+  tuple_for_each(m_iterator_tuple,
+                 detail::advance_iterator<typename super_t::difference_type>(n),
+                 use_me_to_dispatch);
+} // end zip_iterator::advance()
+
+
+template <typename IteratorTuple>
+  void zip_iterator<IteratorTuple>
+    ::increment(void)
+{
+  using namespace detail::tuple_impl_specific;
+
+  // XXX note that we use a pointer to System to dispatch to avoid
+  //     default construction of a System
+  typename thrust::iterator_system<zip_iterator>::type *use_me_to_dispatch = 0;
+
+  // dispatch on system
+  tuple_for_each(m_iterator_tuple, detail::increment_iterator(),
+                 use_me_to_dispatch);
+} // end zip_iterator::increment()
+
+
+template <typename IteratorTuple>
+  void zip_iterator<IteratorTuple>
+    ::decrement(void)
+{
+  using namespace detail::tuple_impl_specific;
+
+  // XXX note that we use a pointer to System to dispatch to avoid
+  //     default construction of a System
+  typename thrust::iterator_system<zip_iterator>::type *use_me_to_dispatch = 0;
+
+  // dispatch on system
+  tuple_for_each(m_iterator_tuple, detail::decrement_iterator(),
+                 use_me_to_dispatch);
+} // end zip_iterator::decrement()
+
+
+__thrust_hd_warning_disable__
+template <typename IteratorTuple>
+  template <typename OtherIteratorTuple>
+    typename zip_iterator<IteratorTuple>::super_t::difference_type
+      zip_iterator<IteratorTuple>
+        ::distance_to(const zip_iterator<OtherIteratorTuple> &other) const
+{
+  return get<0>(other.get_iterator_tuple()) - get<0>(get_iterator_tuple());
+} // end zip_iterator::distance_to()
+
+
+template <typename IteratorTuple>
+  zip_iterator<IteratorTuple> make_zip_iterator(IteratorTuple t)
+{
+  return zip_iterator<IteratorTuple>(t);
+} // end make_zip_iterator()
+
+
+} // end thrust
+
diff --git a/compat/thrust/iterator/detail/zip_iterator_base.h b/compat/thrust/iterator/detail/zip_iterator_base.h
new file mode 100644
index 0000000000..9dd7789e52
--- /dev/null
+++ b/compat/thrust/iterator/detail/zip_iterator_base.h
@@ -0,0 +1,418 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/iterator/iterator_facade.h>
+#include <thrust/iterator/iterator_categories.h>
+#include <thrust/iterator/detail/minimum_category.h>
+#include <thrust/iterator/detail/minimum_system.h>
+#include <thrust/tuple.h>
+#include <thrust/detail/tuple_meta_transform.h>
+#include <thrust/detail/tuple_transform.h>
+#include <thrust/detail/type_traits.h>
+#include <thrust/iterator/detail/tuple_of_iterator_references.h>
+
+namespace thrust
+{
+
+// forward declare zip_iterator for zip_iterator_base
+template<typename IteratorTuple> class zip_iterator;
+
+namespace detail
+{
+
+
+// Functors to be used with tuple algorithms
+//
+template<typename DiffType>
+class advance_iterator
+{
+public:
+  inline __host__ __device__
+  advance_iterator(DiffType step) : m_step(step) {}
+  
+  template<typename Iterator>
+  inline __host__ __device__
+  void operator()(Iterator& it) const
+  { it += m_step; }
+
+private:
+  DiffType m_step;
+}; // end advance_iterator
+
+
+struct increment_iterator
+{
+  template<typename Iterator>
+  inline __host__ __device__
+  void operator()(Iterator& it)
+  { ++it; }
+}; // end increment_iterator
+
+
+struct decrement_iterator
+{
+  template<typename Iterator>
+  inline __host__ __device__
+  void operator()(Iterator& it)
+  { --it; }
+}; // end decrement_iterator
+
+
+struct dereference_iterator
+{
+  template<typename Iterator>
+  struct apply
+  { 
+    typedef typename
+      iterator_traits<Iterator>::reference
+    type;
+  }; // end apply
+
+  // XXX silence warnings of the form "calling a __host__ function from a __host__ __device__ function is not allowed
+  __thrust_hd_warning_disable__
+  template<typename Iterator>
+  __host__ __device__
+    typename apply<Iterator>::type operator()(Iterator const& it)
+  {
+    return *it;
+  }
+}; // end dereference_iterator
+
+
+// The namespace tuple_impl_specific provides two meta-
+// algorithms and two algorithms for tuples.
+namespace tuple_impl_specific
+{
+
+// define apply1 for tuple_meta_transform_impl
+template<typename UnaryMetaFunctionClass, class Arg>
+  struct apply1
+    : UnaryMetaFunctionClass::template apply<Arg>
+{
+}; // end apply1
+
+
+// define apply2 for tuple_meta_accumulate_impl
+template<typename UnaryMetaFunctionClass, class Arg1, class Arg2>
+  struct apply2
+    : UnaryMetaFunctionClass::template apply<Arg1,Arg2>
+{
+}; // end apply2
+
+
+// Meta-accumulate algorithm for tuples. Note: The template 
+// parameter StartType corresponds to the initial value in 
+// ordinary accumulation.
+//
+template<class Tuple, class BinaryMetaFun, class StartType>
+  struct tuple_meta_accumulate;
+
+template<
+    typename Tuple
+  , class BinaryMetaFun
+  , typename StartType
+>
+  struct tuple_meta_accumulate_impl
+{
+   typedef typename apply2<
+       BinaryMetaFun
+     , typename Tuple::head_type
+     , typename tuple_meta_accumulate<
+           typename Tuple::tail_type
+         , BinaryMetaFun
+         , StartType 
+       >::type
+   >::type type;
+};
+
+
+template<
+    typename Tuple
+  , class BinaryMetaFun
+  , typename StartType
+>
+struct tuple_meta_accumulate
+  : thrust::detail::eval_if<
+        thrust::detail::is_same<Tuple, thrust::null_type>::value
+      , thrust::detail::identity_<StartType>
+      , tuple_meta_accumulate_impl<
+            Tuple
+          , BinaryMetaFun
+          , StartType
+        >
+    > // end eval_if
+{
+}; // end tuple_meta_accumulate
+
+
+// transform algorithm for tuples. The template parameter Fun
+// must be a unary functor which is also a unary metafunction
+// class that computes its return type based on its argument
+// type. For example:
+//
+// struct to_ptr
+// {
+//     template <class Arg>
+//     struct apply
+//     {
+//          typedef Arg* type;
+//     }
+//
+//     template <class Arg>
+//     Arg* operator()(Arg x);
+// };
+
+
+
+// for_each algorithm for tuples.
+//
+template<typename Fun, typename System>
+inline __host__ __device__
+Fun tuple_for_each(thrust::null_type, Fun f, System *)
+{
+  return f;
+} // end tuple_for_each()
+
+
+template<typename Tuple, typename Fun, typename System>
+inline __host__ __device__
+Fun tuple_for_each(Tuple& t, Fun f, System *dispatch_tag)
+{ 
+  f( t.get_head() );
+  return tuple_for_each(t.get_tail(), f, dispatch_tag);
+} // end tuple_for_each()
+
+
+template<typename Tuple, typename Fun>
+inline __host__ __device__
+Fun tuple_for_each(Tuple& t, Fun f, thrust::host_system_tag *dispatch_tag)
+{ 
+// XXX this path is required in order to accomodate pure host iterators
+//     (such as std::vector::iterator) in a zip_iterator
+#ifndef __CUDA_ARCH__
+  f( t.get_head() );
+  return tuple_for_each(t.get_tail(), f, dispatch_tag);
+#else
+  // this code will never be called
+  return f;
+#endif
+} // end tuple_for_each()
+
+
+// Equality of tuples. NOTE: "==" for tuples currently (7/2003)
+// has problems under some compilers, so I just do my own.
+// No point in bringing in a bunch of #ifdefs here. This is
+// going to go away with the next tuple implementation anyway.
+//
+__host__ __device__
+inline bool tuple_equal(thrust::null_type, thrust::null_type)
+{ return true; }
+
+
+template<typename Tuple1, typename Tuple2>
+__host__ __device__
+bool tuple_equal(Tuple1 const& t1, Tuple2 const& t2)
+{ 
+  return t1.get_head() == t2.get_head() && 
+  tuple_equal(t1.get_tail(), t2.get_tail());
+} // end tuple_equal()
+
+} // end end tuple_impl_specific
+
+
+// Metafunction to obtain the type of the tuple whose element types
+// are the value_types of an iterator tupel.
+//
+template<typename IteratorTuple>
+  struct tuple_of_value_types
+    : tuple_meta_transform<
+          IteratorTuple,
+          iterator_value
+        >
+{
+}; // end tuple_of_value_types
+
+
+struct minimum_category_lambda
+{
+  template<typename T1, typename T2>
+    struct apply : minimum_category<T1,T2>
+  {};
+};
+
+
+
+// Metafunction to obtain the minimal traversal tag in a tuple
+// of iterators.
+//
+template<typename IteratorTuple>
+struct minimum_traversal_category_in_iterator_tuple
+{
+  typedef typename tuple_meta_transform<
+      IteratorTuple
+    , thrust::iterator_traversal
+  >::type tuple_of_traversal_tags;
+      
+  typedef typename tuple_impl_specific::tuple_meta_accumulate<
+      tuple_of_traversal_tags
+    , minimum_category_lambda
+    , thrust::random_access_traversal_tag
+  >::type type;
+};
+
+
+struct minimum_system_lambda
+{
+  template<typename T1, typename T2>
+    struct apply : minimum_system<T1,T2>
+  {};
+};
+
+
+
+// Metafunction to obtain the minimal system tag in a tuple
+// of iterators.
+template<typename IteratorTuple>
+struct minimum_system_in_iterator_tuple
+{
+  typedef typename thrust::detail::tuple_meta_transform<
+    IteratorTuple,
+    thrust::iterator_system
+  >::type tuple_of_system_tags;
+
+  typedef typename tuple_impl_specific::tuple_meta_accumulate<
+    tuple_of_system_tags,
+    minimum_system_lambda,
+    thrust::any_system_tag
+  >::type type;
+};
+
+namespace zip_iterator_base_ns
+{
+
+
+template<int i, typename Tuple>
+  struct tuple_elements_helper
+    : eval_if<
+        (i < tuple_size<Tuple>::value),
+        tuple_element<i,Tuple>,
+        identity_<thrust::null_type>
+      >
+{};
+
+
+template<typename Tuple>
+  struct tuple_elements
+{
+  typedef typename tuple_elements_helper<0,Tuple>::type T0;
+  typedef typename tuple_elements_helper<1,Tuple>::type T1;
+  typedef typename tuple_elements_helper<2,Tuple>::type T2;
+  typedef typename tuple_elements_helper<3,Tuple>::type T3;
+  typedef typename tuple_elements_helper<4,Tuple>::type T4;
+  typedef typename tuple_elements_helper<5,Tuple>::type T5;
+  typedef typename tuple_elements_helper<6,Tuple>::type T6;
+  typedef typename tuple_elements_helper<7,Tuple>::type T7;
+  typedef typename tuple_elements_helper<8,Tuple>::type T8;
+  typedef typename tuple_elements_helper<9,Tuple>::type T9;
+};
+
+
+template<typename IteratorTuple>
+  struct tuple_of_iterator_references
+{
+  // get a thrust::tuple of the iterators' references
+  typedef typename tuple_meta_transform<
+    IteratorTuple,
+    iterator_reference
+  >::type tuple_of_references;
+
+  // get at the individual tuple element types by name
+  typedef tuple_elements<tuple_of_references> elements;
+
+  // map thrust::tuple<T...> to tuple_of_iterator_references<T...>
+  typedef thrust::detail::tuple_of_iterator_references<
+    typename elements::T0,
+    typename elements::T1,
+    typename elements::T2,
+    typename elements::T3,
+    typename elements::T4,
+    typename elements::T5,
+    typename elements::T6,
+    typename elements::T7,
+    typename elements::T8,
+    typename elements::T9
+  > type;
+};
+
+
+} // end zip_iterator_base_ns
+
+///////////////////////////////////////////////////////////////////
+//
+// Class zip_iterator_base
+//
+// Builds and exposes the iterator facade type from which the zip 
+// iterator will be derived.
+//
+template<typename IteratorTuple>
+  struct zip_iterator_base
+{
+ //private:
+    // reference type is the type of the tuple obtained from the
+    // iterators' reference types.
+    typedef typename zip_iterator_base_ns::tuple_of_iterator_references<IteratorTuple>::type reference;
+
+    // Boost's Value type is the same as reference type.
+    //typedef reference value_type;
+    typedef typename tuple_of_value_types<IteratorTuple>::type value_type;
+
+    // Difference type is the first iterator's difference type
+    typedef typename thrust::iterator_traits<
+      typename thrust::tuple_element<0, IteratorTuple>::type
+    >::difference_type difference_type;
+
+    // Iterator system is the minimum system tag in the
+    // iterator tuple
+    typedef typename
+    minimum_system_in_iterator_tuple<IteratorTuple>::type system;
+
+    // Traversal category is the minimum traversal category in the
+    // iterator tuple
+    typedef typename
+    minimum_traversal_category_in_iterator_tuple<IteratorTuple>::type traversal_category;
+  
+ public:
+  
+    // The iterator facade type from which the zip iterator will
+    // be derived.
+    typedef thrust::iterator_facade<
+        zip_iterator<IteratorTuple>,
+        value_type,  
+        system,
+        traversal_category,
+        reference,
+        difference_type
+    > type;
+}; // end zip_iterator_base
+
+} // end detail
+
+} // end thrust
+
+
diff --git a/compat/thrust/iterator/discard_iterator.h b/compat/thrust/iterator/discard_iterator.h
new file mode 100644
index 0000000000..6e089b567e
--- /dev/null
+++ b/compat/thrust/iterator/discard_iterator.h
@@ -0,0 +1,171 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file thrust/iterator/discard_iterator.h
+ *  \brief An iterator which "discards" (ignores) values assigned to it upon dereference
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/iterator/detail/discard_iterator_base.h>
+#include <thrust/iterator/iterator_facade.h>
+
+__THRUST_DISABLE_MSVC_POSSIBLE_LOSS_OF_DATA_WARNING_BEGIN
+
+namespace thrust
+{
+
+/*! \addtogroup iterators
+ *  \{
+ */
+
+/*! \addtogroup fancyiterator Fancy Iterators
+ *  \ingroup iterators
+ *  \{
+ */
+
+/*! \p discard_iterator is an iterator which represents a special kind of pointer that
+ *  ignores values written to it upon dereference. This iterator is useful for ignoring
+ *  the output of certain algorithms without wasting memory capacity or bandwidth.
+ *  \p discard_iterator may also be used to count the size of an algorithm's output which
+ *  may not be known a priori.
+ *
+ *  The following code snippet demonstrates how to use \p discard_iterator to ignore
+ *  ignore one of the output ranges of reduce_by_key
+ *
+ *  \code
+ *  #include <thrust/iterator/discard_iterator.h>
+ *  #include <thrust/reduce.h>
+ *  #include <thrust/device_vector.h>
+ *
+ *  int main(void)
+ *  {
+ *    thrust::device_vector<int> keys(7), values(7);
+ *
+ *    keys[0] = 1;
+ *    keys[1] = 3;
+ *    keys[2] = 3;
+ *    keys[3] = 3;
+ *    keys[4] = 2;
+ *    keys[5] = 2;
+ *    keys[6] = 1;
+ *
+ *    values[0] = 9;
+ *    values[1] = 8;
+ *    values[2] = 7;
+ *    values[3] = 6;
+ *    values[4] = 5;
+ *    values[5] = 4;
+ *    values[6] = 3;
+ *
+ *    thrust::device_vector<int> result(4);
+ *
+ *    // we are only interested in the reduced values
+ *    // use discard_iterator to ignore the output keys
+ *    thrust::reduce_by_key(keys.begin(), keys.end(),
+ *                          values.begin(), values.end(),
+ *                          thrust::make_discard_iterator(),
+ *                          result.begin());
+ *    
+ *    // result is now [9, 21, 9, 3]
+ *    
+ *    return 0;
+ *  }
+ *  \endcode
+ *
+ *  \see make_discard_iterator
+ */
+template<typename System = use_default>
+  class discard_iterator
+    : public detail::discard_iterator_base<System>::type
+{
+    /*! \cond
+     */
+    friend class thrust::iterator_core_access;
+    typedef typename detail::discard_iterator_base<System>::type          super_t;
+    typedef typename detail::discard_iterator_base<System>::incrementable incrementable;
+    typedef typename detail::discard_iterator_base<System>::base_iterator base_iterator;
+
+  public:
+    typedef typename super_t::reference  reference;
+    typedef typename super_t::value_type value_type;
+
+    /*! \endcond
+     */
+
+    /*! Copy constructor copies from a source discard_iterator.
+     *
+     *  \p rhs The discard_iterator to copy.
+     */
+    __host__ __device__
+    discard_iterator(discard_iterator const &rhs)
+      : super_t(rhs.base()) {}
+
+    /*! This constructor receives an optional index specifying the position of this
+     *  \p discard_iterator in a range.
+     *  
+     *  \p i The index of this \p discard_iterator in a range. Defaults to the
+     *       value returned by \c Incrementable's null constructor. For example,
+     *       when <tt>Incrementable == int</tt>, \c 0.
+     */
+    __host__ __device__
+    discard_iterator(incrementable const &i = incrementable())
+      : super_t(base_iterator(i)) {}
+
+    /*! \cond
+     */
+  
+  private: // Core iterator interface
+    __host__ __device__
+    reference dereference(void) const
+    {
+      return m_element;
+    }
+
+    mutable value_type m_element;
+
+    /*! \endcond
+     */
+}; // end constant_iterator
+
+
+/*! \p make_discard_iterator creates a \p discard_iterator from an optional index parameter.
+ *
+ *  \param i The index of the returned \p discard_iterator within a range.
+ *           In the default case, the value of this parameter is \c 0.
+ *
+ *  \return A new \p discard_iterator with index as given by \p i.
+ *
+ *  \see constant_iterator
+ */
+inline __host__ __device__
+discard_iterator<> make_discard_iterator(discard_iterator<>::difference_type i = discard_iterator<>::difference_type(0))
+{
+  return discard_iterator<>(i);
+} // end make_discard_iterator()
+
+/*! \} // end fancyiterators
+ */
+
+/*! \} // end iterators
+ */
+
+} // end namespace thrust
+  
+__THRUST_DISABLE_MSVC_POSSIBLE_LOSS_OF_DATA_WARNING_END
+
diff --git a/compat/thrust/iterator/iterator_adaptor.h b/compat/thrust/iterator/iterator_adaptor.h
new file mode 100644
index 0000000000..7b9cca308a
--- /dev/null
+++ b/compat/thrust/iterator/iterator_adaptor.h
@@ -0,0 +1,239 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file thrust/iterator/iterator_adaptor.h
+ *  \brief An iterator which adapts a base iterator
+ */
+
+/*
+ * (C) Copyright David Abrahams 2002.
+ * (C) Copyright Jeremy Siek    2002.
+ * (C) Copyright Thomas Witt    2002.
+ * 
+ * Distributed under the Boost Software License, Version 1.0.
+ * (See accompanying NOTICE file for the complete license)
+ *
+ * For more information, see http://www.boost.org
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/iterator/iterator_facade.h>
+#include <thrust/detail/use_default.h>
+#include <thrust/iterator/detail/iterator_adaptor_base.h>
+
+namespace thrust
+{
+
+/*! \addtogroup iterators
+ *  \{
+ */
+
+/*! \addtogroup fancyiterator Fancy Iterators
+ *  \ingroup iterators
+ *  \{
+ */
+
+/*! \p iterator_adaptor is an iterator which adapts an existing type of iterator to create a new type of
+ *  iterator. Most of Thrust's fancy iterators are defined via inheritance from \p iterator_adaptor.
+ *  While composition of these existing Thrust iterators is often sufficient for expressing the desired
+ *  functionality, it is occasionally more straightforward to derive from \p iterator_adaptor directly.
+ *
+ *  To see how to use \p iterator_adaptor to create a novel iterator type, let's examine how to use it to
+ *  define \p repeat_iterator, a fancy iterator which repeats elements from another range a given number of time:
+ *
+ *  \code
+ *  #include <thrust/iterator/iterator_adaptor.h>
+ *
+ *  // derive repeat_iterator from iterator_adaptor
+ *  template<typename Iterator>
+ *    class repeat_iterator
+ *      : public thrust::iterator_adaptor<
+ *          repeat_iterator<Iterator>, // the first template parameter is the name of the iterator we're creating
+ *          Iterator                   // the second template parameter is the name of the iterator we're adapting
+ *                                     // we can use the default for the additional template parameters
+ *        >
+ *  {
+ *    public:
+ *      // shorthand for the name of the iterator_adaptor we're deriving from
+ *      typedef thrust::iterator_adaptor<
+ *        repeat_iterator<Iterator>,
+ *        Iterator
+ *      > super_t;
+ *
+ *      __host__ __device__
+ *      repeat_iterator(const Iterator &x, int n) : super_t(x), begin(x), n(n) {}
+ *
+ *      // befriend thrust::iterator_core_access to allow it access to the private interface below
+ *      friend class thrust::iterator_core_access;
+ *
+ *    private:
+ *      // repeat each element of the adapted range n times
+ *      unsigned int n;
+ *
+ *      // used to keep track of where we began
+ *      const Iterator begin;
+ *
+ *      // it is private because only thrust::iterator_core_access needs access to it
+ *      __host__ __device__
+ *      typename super_t::reference dereference() const
+ *      {
+ *        return *(begin + (this->base() - begin) / n);
+ *      }
+ *  };
+ *  \endcode
+ *
+ *  Except for the first two, \p iterator_adaptor's template parameters are optional. When omitted, or when the
+ *  user specifies \p thrust::use_default in its place, \p iterator_adaptor will use a default type inferred from \p Base.
+ *
+ *  \p iterator_adaptor's functionality is derived from and generally equivalent to \p boost::iterator_adaptor.
+ *  The exception is Thrust's addition of the template parameter \p System, which is necessary to allow Thrust
+ *  to dispatch an algorithm to one of several parallel backend systems.
+ *
+ *  \p iterator_adaptor is a powerful tool for creating custom iterators directly. However, the large set of iterator semantics which must be satisfied
+ *  for algorithm compatibility can make \p iterator_adaptor difficult to use correctly. Unless you require the full expressivity of \p iterator_adaptor,
+ *  consider building a custom iterator through composition of existing higher-level fancy iterators instead. 
+ *
+ *  Interested users may refer to <tt>boost::iterator_adaptor</tt>'s documentation for further usage examples.
+ */
+template<typename Derived,
+         typename Base,
+         typename Value      = use_default,
+         typename System     = use_default,
+         typename Traversal  = use_default,
+         typename Reference  = use_default,
+         typename Difference = use_default>
+  class iterator_adaptor:
+    public detail::iterator_adaptor_base<
+      Derived, Base, Value, System, Traversal, Reference, Difference
+    >::type
+{
+  /*! \cond
+   */
+
+    friend class thrust::iterator_core_access;
+
+  protected:
+    typedef typename detail::iterator_adaptor_base<
+        Derived, Base, Value, System, Traversal, Reference, Difference
+    >::type super_t;
+
+  /*! \endcond
+   */
+  
+  public:
+    /*! \p iterator_adaptor's default constructor does nothing.
+     */
+    __host__ __device__
+    iterator_adaptor(){}
+
+    /*! This constructor copies from a given instance of the \p Base iterator.
+     */
+    __host__ __device__
+    explicit iterator_adaptor(Base const& iter)
+      : m_iterator(iter)
+    {}
+
+    /*! The type of iterator this \p iterator_adaptor's \p adapts.
+     */
+    typedef Base       base_type;
+                                                                                              
+    /*! \cond
+     */
+    typedef typename super_t::reference reference;
+                                                                                              
+    typedef typename super_t::difference_type difference_type;
+    /*! \endcond
+     */
+
+    /*! \return A \p const reference to the \p Base iterator this \p iterator_adaptor adapts.
+     */
+    __host__ __device__
+    Base const& base() const
+    { return m_iterator; }
+
+  protected:
+    /*! \return A \p const reference to the \p Base iterator this \p iterator_adaptor adapts.
+     */
+    __host__ __device__
+    Base const& base_reference() const
+    { return m_iterator; }
+
+    /*! \return A mutable reference to the \p Base iterator this \p iterator_adaptor adapts.
+     */
+    __host__ __device__
+    Base& base_reference()
+    { return m_iterator; }
+
+    /*! \cond
+     */
+  private: // Core iterator interface for iterator_facade
+
+    __thrust_hd_warning_disable__
+    __host__ __device__
+    typename iterator_adaptor::reference dereference() const
+    { return *m_iterator; }
+
+    __thrust_hd_warning_disable__
+    template<typename OtherDerived, typename OtherIterator, typename V, typename S, typename T, typename R, typename D>
+    __host__ __device__
+    bool equal(iterator_adaptor<OtherDerived, OtherIterator, V, S, T, R, D> const& x) const
+    { return m_iterator == x.base(); }
+
+    __thrust_hd_warning_disable__
+    __host__ __device__
+    void advance(typename iterator_adaptor::difference_type n)
+    {
+      // XXX statically assert on random_access_traversal_tag
+      m_iterator += n;
+    }
+
+    __thrust_hd_warning_disable__
+    __host__ __device__
+    void increment()
+    { ++m_iterator; }
+
+    __thrust_hd_warning_disable__
+    __host__ __device__
+    void decrement()
+    {
+      // XXX statically assert on bidirectional_traversal_tag
+      --m_iterator;
+    }
+
+    __thrust_hd_warning_disable__
+    template<typename OtherDerived, typename OtherIterator, typename V, typename S, typename T, typename R, typename D>
+    __host__ __device__
+    typename iterator_adaptor::difference_type distance_to(iterator_adaptor<OtherDerived, OtherIterator, V, S, T, R, D> const& y) const
+    { return y.base() - m_iterator; }
+
+  private:
+    Base m_iterator;
+
+    /*! \endcond
+     */
+}; // end iterator_adaptor
+
+/*! \} // end fancyiterators
+ */
+
+/*! \} // end iterators
+ */
+
+} // end thrust
+
diff --git a/compat/thrust/iterator/iterator_categories.h b/compat/thrust/iterator/iterator_categories.h
new file mode 100644
index 0000000000..81601b4a40
--- /dev/null
+++ b/compat/thrust/iterator/iterator_categories.h
@@ -0,0 +1,191 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file thrust/iterator/iterator_categories.h
+ *  \brief Types for reasoning about the categories of iterators
+ */
+
+/*
+ * (C) Copyright Jeremy Siek 2002.
+ * 
+ * Distributed under the Boost Software License, Version 1.0.
+ * (See accompanying NOTICE file for the complete license)
+ *
+ * For more information, see http://www.boost.org
+ */
+
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// #include this for stl's iterator tags
+#include <iterator>
+
+namespace thrust
+{
+
+/*! \addtogroup iterators
+ *  \addtogroup iterator_tags Iterator Tags
+ *  \ingroup iterators
+ *  \addtogroup iterator_tag_classes Iterator Tag Classes
+ *  \ingroup iterator_tags
+ *  \{
+ */
+
+/*! \p input_device_iterator_tag is an empty class: it has no member functions,
+ *  member variables, or nested types. It is used solely as a "tag": a
+ *  representation of the Input Device Iterator concept within the C++ type
+ *  system.
+ *
+ *  \see http://www.sgi.com/tech/sgi/input_iterator_tag.html, iterator_traits,
+ *  output_device_iterator_tag, forward_device_iterator_tag,
+ *  bidirectional_device_iterator_tag, random_access_device_iterator_tag,
+ *  input_host_iterator_tag, output_host_iterator_tag, forward_host_iterator_tag,
+ *  bidirectional_host_iterator_tag, random_access_host_iterator_tag
+ */
+struct input_device_iterator_tag {};
+
+/*! \p output_device_iterator_tag is an empty class: it has no member functions,
+ *  member variables, or nested types. It is used solely as a "tag": a
+ *  representation of the Output Device Iterator concept within the C++ type
+ *  system.
+ *
+ *  \see http://www.sgi.com/tech/sgi/output_iterator_tag.html, iterator_traits,
+ *  input_device_iterator_tag, forward_device_iterator_tag,
+ *  bidirectional_device_iterator_tag, random_access_device_iterator_tag,
+ *  input_host_iterator_tag, output_host_iterator_tag, forward_host_iterator_tag,
+ *  bidirectional_host_iterator_tag, random_access_host_iterator_tag
+ */
+struct output_device_iterator_tag {};
+
+/*! \p forward_device_iterator_tag is an empty class: it has no member functions,
+ *  member variables, or nested types. It is used solely as a "tag": a
+ *  representation of the Forward Device Iterator concept within the C++ type
+ *  system.
+ *
+ *  \see http://www.sgi.com/tech/sgi/forward_iterator_tag.html, iterator_traits,
+ *  input_device_iterator_tag, output_device_iterator_tag,
+ *  bidirectional_device_iterator_tag, random_access_device_iterator_tag,
+ *  input_host_iterator_tag, output_host_iterator_tag, forward_host_iterator_tag,
+ *  bidirectional_host_iterator_tag, random_access_host_iterator_tag
+ */
+struct forward_device_iterator_tag : public input_device_iterator_tag {};
+
+/*! \p bidirectional_device_iterator_tag is an empty class: it has no member
+ *  functions, member variables, or nested types. It is used solely as a "tag": a
+ *  representation of the Bidirectional Device Iterator concept within the C++
+ *  type system.
+ *
+ *  \see http://www.sgi.com/tech/sgi/bidirectional_iterator_tag.html,
+ *  iterator_traits, input_device_iterator_tag, output_device_iterator_tag,
+ *  forward_device_iterator_tag, random_access_device_iterator_tag,
+ *  input_host_iterator_tag, output_host_iterator_tag, forward_host_iterator_tag,
+ *  bidirectional_host_iterator_tag, random_access_host_iterator_tag
+ */
+struct bidirectional_device_iterator_tag : public forward_device_iterator_tag {};
+
+/*! \p random_access_device_iterator_tag is an empty class: it has no member
+ *  functions, member variables, or nested types. It is used solely as a "tag": a
+ *  representation of the Random Access Device Iterator concept within the C++
+ *  type system.
+ *
+ *  \see http://www.sgi.com/tech/sgi/random_access_iterator_tag.html,
+ *  iterator_traits, input_device_iterator_tag, output_device_iterator_tag,
+ *  forward_device_iterator_tag, bidirectional_device_iterator_tag,
+ *  input_host_iterator_tag, output_host_iterator_tag, forward_host_iterator_tag,
+ *  bidirectional_host_iterator_tag, random_access_host_iterator_tag
+ */
+struct random_access_device_iterator_tag : public bidirectional_device_iterator_tag {};
+
+/*! \p input_host_iterator_tag is an empty class: it has no member
+ *  functions, member variables, or nested types. It is used solely as a "tag": a
+ *  representation of the Input Host Iterator concept within the C++
+ *  type system.
+ *
+ *  \see http://www.sgi.com/tech/sgi/input_iterator_tag.html,
+ *  iterator_traits, input_device_iterator_tag, output_device_iterator_tag,
+ *  forward_device_iterator_tag, bidirectional_device_iterator_tag,
+ *  random_access_device_iterator_tag,
+ *  output_host_iterator_tag, forward_host_iterator_tag,
+ *  bidirectional_host_iterator_tag, random_access_host_iterator_tag
+ */
+typedef std::input_iterator_tag input_host_iterator_tag;
+
+/*! \p output_host_iterator_tag is an empty class: it has no member
+ *  functions, member variables, or nested types. It is used solely as a "tag": a
+ *  representation of the Output Host Iterator concept within the C++
+ *  type system.
+ *
+ *  \see http://www.sgi.com/tech/sgi/output_iterator_tag.html,
+ *  iterator_traits, input_device_iterator_tag, output_device_iterator_tag,
+ *  forward_device_iterator_tag, bidirectional_device_iterator_tag,
+ *  random_access_device_iterator_tag,
+ *  input_host_iterator_tag, forward_host_iterator_tag,
+ *  bidirectional_host_iterator_tag, random_access_host_iterator_tag
+ */
+typedef std::output_iterator_tag output_host_iterator_tag;
+
+/*! \p forward_host_iterator_tag is an empty class: it has no member
+ *  functions, member variables, or nested types. It is used solely as a "tag": a
+ *  representation of the Forward Host Iterator concept within the C++
+ *  type system.
+ *
+ *  \see http://www.sgi.com/tech/sgi/forward_iterator_tag.html,
+ *  iterator_traits, input_device_iterator_tag, output_device_iterator_tag,
+ *  forward_device_iterator_tag, bidirectional_device_iterator_tag,
+ *  random_access_device_iterator_tag,
+ *  input_host_iterator_tag, output_host_iterator_tag,
+ *  bidirectional_host_iterator_tag, random_access_host_iterator_tag
+ */
+typedef std::forward_iterator_tag forward_host_iterator_tag;
+
+/*! \p bidirectional_host_iterator_tag is an empty class: it has no member
+ *  functions, member variables, or nested types. It is used solely as a "tag": a
+ *  representation of the Forward Host Iterator concept within the C++
+ *  type system.
+ *
+ *  \see http://www.sgi.com/tech/sgi/bidirectional_iterator_tag.html,
+ *  iterator_traits, input_device_iterator_tag, output_device_iterator_tag,
+ *  forward_device_iterator_tag, bidirectional_device_iterator_tag,
+ *  random_access_device_iterator_tag,
+ *  input_host_iterator_tag, output_host_iterator_tag,
+ *  forward_host_iterator_tag, random_access_host_iterator_tag
+ */
+typedef std::bidirectional_iterator_tag bidirectional_host_iterator_tag;
+
+/*! \p random_access_host_iterator_tag is an empty class: it has no member
+ *  functions, member variables, or nested types. It is used solely as a "tag": a
+ *  representation of the Forward Host Iterator concept within the C++
+ *  type system.
+ *
+ *  \see http://www.sgi.com/tech/sgi/random_access_iterator_tag.html,
+ *  iterator_traits, input_device_iterator_tag, output_device_iterator_tag,
+ *  forward_device_iterator_tag, bidirectional_device_iterator_tag,
+ *  random_access_device_iterator_tag,
+ *  input_host_iterator_tag, output_host_iterator_tag,
+ *  forward_host_iterator_tag, bidirectional_host_iterator_tag
+ */
+typedef std::random_access_iterator_tag random_access_host_iterator_tag;
+
+/*! \} // end iterator_tag_classes
+ */
+
+} // end namespace thrust
+
+#include <thrust/iterator/detail/universal_categories.h>
+
diff --git a/compat/thrust/iterator/iterator_facade.h b/compat/thrust/iterator/iterator_facade.h
new file mode 100644
index 0000000000..232c150a6f
--- /dev/null
+++ b/compat/thrust/iterator/iterator_facade.h
@@ -0,0 +1,538 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+/*! \file thrust/iterator/iterator_facade.h
+ *  \brief A class which exposes a public interface for iterators
+ */
+
+/*
+ * (C) Copyright David Abrahams 2002.
+ * (C) Copyright Jeremy Siek    2002.
+ * (C) Copyright Thomas Witt    2002.
+ * 
+ * Distributed under the Boost Software License, Version 1.0.
+ * (See accompanying NOTICE file for the complete license)
+ *
+ * For more information, see http://www.boost.org
+ */
+
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/type_traits.h>
+#include <thrust/iterator/detail/iterator_facade_category.h>
+#include <thrust/iterator/detail/distance_from_result.h>
+
+namespace thrust
+{
+
+/*! \addtogroup iterators
+ *  \{
+ */
+
+/*! \addtogroup fancyiterator Fancy Iterators
+ *  \ingroup iterators
+ *  \{
+ */
+
+
+// This forward declaration is required for the friend declaration
+// in iterator_core_access
+template<typename Derived, typename Value, typename System, typename Traversal, typename Reference, typename Difference> class iterator_facade;
+
+
+/*! \p iterator_core_access is the class which user iterator types derived from \p thrust::iterator_adaptor
+ *  or \p thrust::iterator_facade must befriend to allow it to access their private interface.
+ */
+class iterator_core_access
+{
+    /*! \cond
+     */
+
+    // declare our friends
+    template<typename Derived, typename Value, typename System, typename Traversal, typename Reference, typename Difference> friend class iterator_facade;
+
+    // iterator comparisons are our friends
+    template <typename Derived1, typename Value1, typename System1, typename Traversal1, typename Reference1, typename Difference1,
+              typename Derived2, typename Value2, typename System2, typename Traversal2, typename Reference2, typename Difference2>
+    inline __host__ __device__
+    friend bool
+    operator ==(iterator_facade<Derived1,Value1,System1,Traversal1,Reference1,Difference1> const& lhs,
+                iterator_facade<Derived2,Value2,System2,Traversal2,Reference2,Difference2> const& rhs);
+
+    template <typename Derived1, typename Value1, typename System1, typename Traversal1, typename Reference1, typename Difference1,
+              typename Derived2, typename Value2, typename System2, typename Traversal2, typename Reference2, typename Difference2>
+    inline __host__ __device__
+    friend bool
+    operator !=(iterator_facade<Derived1,Value1,System1,Traversal1,Reference1,Difference1> const& lhs,
+                iterator_facade<Derived2,Value2,System2,Traversal2,Reference2,Difference2> const& rhs);
+
+    template <typename Derived1, typename Value1, typename System1, typename Traversal1, typename Reference1, typename Difference1,
+              typename Derived2, typename Value2, typename System2, typename Traversal2, typename Reference2, typename Difference2>
+    inline __host__ __device__
+    friend bool
+    operator <(iterator_facade<Derived1,Value1,System1,Traversal1,Reference1,Difference1> const& lhs,
+               iterator_facade<Derived2,Value2,System2,Traversal2,Reference2,Difference2> const& rhs);
+
+    template <typename Derived1, typename Value1, typename System1, typename Traversal1, typename Reference1, typename Difference1,
+              typename Derived2, typename Value2, typename System2, typename Traversal2, typename Reference2, typename Difference2>
+    inline __host__ __device__
+    friend bool
+    operator >(iterator_facade<Derived1,Value1,System1,Traversal1,Reference1,Difference1> const& lhs,
+               iterator_facade<Derived2,Value2,System2,Traversal2,Reference2,Difference2> const& rhs);
+
+    template <typename Derived1, typename Value1, typename System1, typename Traversal1, typename Reference1, typename Difference1,
+              typename Derived2, typename Value2, typename System2, typename Traversal2, typename Reference2, typename Difference2>
+    inline __host__ __device__
+    friend bool
+    operator <=(iterator_facade<Derived1,Value1,System1,Traversal1,Reference1,Difference1> const& lhs,
+                iterator_facade<Derived2,Value2,System2,Traversal2,Reference2,Difference2> const& rhs);
+
+    template <typename Derived1, typename Value1, typename System1, typename Traversal1, typename Reference1, typename Difference1,
+              typename Derived2, typename Value2, typename System2, typename Traversal2, typename Reference2, typename Difference2>
+    inline __host__ __device__
+    friend bool
+    operator >=(iterator_facade<Derived1,Value1,System1,Traversal1,Reference1,Difference1> const& lhs,
+                iterator_facade<Derived2,Value2,System2,Traversal2,Reference2,Difference2> const& rhs);
+
+    // iterator difference is our friend
+    template <typename Derived1, typename Value1, typename System1, typename Traversal1, typename Reference1, typename Difference1,
+              typename Derived2, typename Value2, typename System2, typename Traversal2, typename Reference2, typename Difference2>
+    inline __host__ __device__
+    friend
+      typename thrust::detail::distance_from_result<
+        iterator_facade<Derived1,Value1,System1,Traversal1,Reference1,Difference1>,
+        iterator_facade<Derived2,Value2,System2,Traversal2,Reference2,Difference2>
+      >::type
+    operator-(iterator_facade<Derived1,Value1,System1,Traversal1,Reference1,Difference1> const& lhs,
+              iterator_facade<Derived2,Value2,System2,Traversal2,Reference2,Difference2> const& rhs);
+
+    template<typename Facade>
+    __host__ __device__
+    static typename Facade::reference dereference(Facade const& f)
+    {
+      return f.dereference();
+    }
+
+    template<typename Facade>
+    __host__ __device__
+    static void increment(Facade& f)
+    {
+      f.increment();
+    }
+
+    template<typename Facade>
+    __host__ __device__
+    static void decrement(Facade& f)
+    {
+      f.decrement();
+    }
+
+    template <class Facade1, class Facade2>
+    __host__ __device__
+    static bool equal(Facade1 const& f1, Facade2 const& f2)
+    {
+      return f1.equal(f2);
+    }
+
+    // XXX TODO: Investigate whether we need both of these cases
+    //template <class Facade1, class Facade2>
+    //__host__ __device__
+    //static bool equal(Facade1 const& f1, Facade2 const& f2, mpl::true_)
+    //{
+    //  return f1.equal(f2);
+    //}
+
+    //template <class Facade1, class Facade2>
+    //__host__ __device__
+    //static bool equal(Facade1 const& f1, Facade2 const& f2, mpl::false_)
+    //{
+    //  return f2.equal(f1);
+    //}
+
+    template <class Facade>
+    __host__ __device__
+    static void advance(Facade& f, typename Facade::difference_type n)
+    {
+      f.advance(n);
+    }
+
+    // Facade2 is convertible to Facade1,
+    // so return Facade1's difference_type
+    template <class Facade1, class Facade2>
+    __host__ __device__
+    static typename Facade1::difference_type
+      distance_from(Facade1 const& f1, Facade2 const& f2, thrust::detail::true_type)
+    {
+      return -f1.distance_to(f2);
+    }
+
+    // Facade2 is not convertible to Facade1,
+    // so return Facade2's difference_type
+    template <class Facade1, class Facade2>
+    __host__ __device__
+    static typename Facade2::difference_type
+      distance_from(Facade1 const& f1, Facade2 const& f2, thrust::detail::false_type)
+    {
+      return f2.distance_to(f1);
+    }
+    
+    template <class Facade1, class Facade2>
+    __host__ __device__
+    static typename thrust::detail::distance_from_result<Facade1,Facade2>::type
+      distance_from(Facade1 const& f1, Facade2 const& f2)
+    {
+      // dispatch the implementation of this method upon whether or not
+      // Facade2 is convertible to Facade1
+      return distance_from(f1, f2,
+        typename thrust::detail::is_convertible<Facade2,Facade1>::type());
+    }
+
+    //
+    // Curiously Recurring Template interface.
+    //
+    template <typename Derived, typename Value, typename System, typename Traversal, typename Reference, typename Difference>
+    __host__ __device__
+    static Derived& derived(iterator_facade<Derived,Value,System,Traversal,Reference,Difference>& facade)
+    {
+      return *static_cast<Derived*>(&facade);
+    }
+
+    template <typename Derived, typename Value, typename System, typename Traversal, typename Reference, typename Difference>
+    __host__ __device__
+    static Derived const& derived(iterator_facade<Derived,Value,System,Traversal,Reference,Difference> const& facade)
+    {
+      return *static_cast<Derived const*>(&facade);
+    }
+
+    /*! \endcond
+     */
+}; // end iterator_core_access
+
+
+/*! \p iterator_facade is a template which allows the programmer to define a novel iterator with a standards-conforming interface
+ *  which Thrust can use to reason about algorithm acceleration opportunities.
+ *
+ *  Because most of a standard iterator's interface is defined in terms of a small set of core primitives, \p iterator_facade
+ *  defines the non-primitive portion mechanically. In principle a novel iterator could explicitly provide the entire interface in
+ *  an ad hoc fashion but doing so might be tedious and prone to subtle errors.
+ *
+ *  Often \p iterator_facade is too primitive a tool to use for defining novel iterators. In these cases, \p iterator_adaptor
+ *  or a specific fancy iterator should be used instead.
+ *
+ *  \p iterator_facade's functionality is derived from and generally equivalent to \p boost::iterator_facade.
+ *  The exception is Thrust's addition of the template parameter \p System, which is necessary to allow Thrust
+ *  to dispatch an algorithm to one of several parallel backend systems. An additional exception is Thrust's omission
+ *  of the \c operator-> member function.
+ *
+ *  Interested users may refer to <tt>boost::iterator_facade</tt>'s documentation for usage examples.
+ *
+ *  \note \p iterator_facade's arithmetic operator free functions exist with the usual meanings but are omitted here for brevity.
+ */
+template<typename Derived, typename Value, typename System, typename Traversal, typename Reference, typename Difference>
+  class iterator_facade
+{
+  private:
+    /*! \cond
+     */
+
+    //
+    // Curiously Recurring Template interface.
+    //
+    __host__ __device__
+    Derived& derived()
+    {
+      return *static_cast<Derived*>(this);
+    }
+
+    __host__ __device__
+    Derived const& derived() const
+    {
+      return *static_cast<Derived const*>(this);
+    }
+    /*! \endcond
+     */
+
+  public:
+    /*! The type of element pointed to by \p iterator_facade.
+     */
+    typedef typename thrust::detail::remove_const<Value>::type value_type;
+
+    /*! The return type of \p iterator_facade::operator*().
+     */
+    typedef Reference                                          reference;
+
+    /*! The return type of \p iterator_facade's non-existent \c operator->()
+     *  member function. Unlike \c boost::iterator_facade, \p iterator_facade
+     *  disallows access to the \p value_type's members through expressions of the
+     *  form <tt>iter->member</tt>. \p pointer is defined to \c void to indicate
+     *  that these expressions are not allowed. This limitation may be relaxed in a
+     *  future version of Thrust.
+     */
+    typedef void                                               pointer;
+
+    /*! The type of expressions of the form <tt>x - y</tt> where <tt>x</tt> and <tt>y</tt>
+     *  are of type \p iterator_facade.
+     */
+    typedef Difference                                         difference_type;
+
+    /*! The type of iterator category of \p iterator_facade.
+     */
+    typedef typename thrust::detail::iterator_facade_category<
+      System, Traversal, Value, Reference
+    >::type                                                    iterator_category;
+
+    /*! \p operator*() dereferences this \p iterator_facade.
+     *  \return A reference to the element pointed to by this \p iterator_facade.
+     */
+    __host__ __device__
+    reference operator*() const
+    {
+      return iterator_core_access::dereference(this->derived());
+    }
+
+    // XXX unimplemented for now, consider implementing it later
+    //pointer operator->() const
+    //{
+    //  return;
+    //}
+
+    // XXX investigate whether or not we need to go to the lengths
+    //     boost does to determine the return type
+
+    /*! \p operator[] performs indexed dereference.
+     *  \return A reference to the element \p n distance away from this \p iterator_facade.
+     */
+    __host__ __device__
+    reference operator[](difference_type n) const
+    {
+      return *(this->derived() + n);
+    }
+
+    /*! \p operator++ pre-increments this \p iterator_facade to refer to the element in the next position.
+     *  \return <tt>*this</tt>
+     */
+    __host__ __device__
+    Derived& operator++()
+    {
+      iterator_core_access::increment(this->derived());
+      return this->derived();
+    }
+
+    /*! \p operator++ post-increments this \p iterator_facade and returns a new \p iterator_facade referring to the element in the next position.
+     *  \return A copy of <tt>*this</tt> before increment.
+     */
+    __host__ __device__
+    Derived  operator++(int)
+    {
+      Derived tmp(this->derived());
+      ++*this;
+      return tmp;
+    }
+
+    /*! \p operator-- pre-decrements this \p iterator_facade to refer to the element in the previous position.
+     *  \return <tt>*this</tt>
+     */
+    __host__ __device__
+    Derived& operator--()
+    {
+      iterator_core_access::decrement(this->derived());
+      return this->derived();
+    }
+
+    /*! \p operator-- post-decrements this \p iterator_facade and returns a new \p iterator_facade referring to the element in the previous position.
+     *  \return A copy of <tt>*this</tt> before decrement.
+     */
+    __host__ __device__
+    Derived  operator--(int)
+    {
+      Derived tmp(this->derived());
+      --*this;
+      return tmp;
+    }
+
+    /*! \p operator+= increments this \p iterator_facade to refer to an element a given distance after its current position.
+     *  \param n The quantity to increment.
+     *  \return <tt>*this</tt>
+     */
+    __host__ __device__
+    Derived& operator+=(difference_type n)
+    {
+      iterator_core_access::advance(this->derived(), n);
+      return this->derived();
+    }
+
+    /*! \p operator-= decrements this \p iterator_facade to refer to an element a given distance before its current postition.
+     *  \param n The quantity to decrement.
+     *  \return <tt>*this</tt>
+     */
+    __host__ __device__
+    Derived& operator-=(difference_type n)
+    {
+      iterator_core_access::advance(this->derived(), -n);
+      return this->derived();
+    }
+
+    /*! \p operator- subtracts a given quantity from this \p iterator_facade and returns a new \p iterator_facade referring to the element at the given position before this \p iterator_facade.
+     *  \param n The quantity to decrement
+     *  \return An \p iterator_facade pointing \p n elements before this \p iterator_facade.
+     */
+    __host__ __device__
+    Derived  operator-(difference_type n) const
+    {
+      Derived result(this->derived());
+      return result -= n;
+    }
+}; // end iterator_facade
+
+/*! \cond
+ */
+
+// Comparison operators
+template <typename Derived1, typename Value1, typename System1, typename Traversal1, typename Reference1, typename Difference1,
+          typename Derived2, typename Value2, typename System2, typename Traversal2, typename Reference2, typename Difference2>
+inline __host__ __device__
+// XXX it might be nice to implement this at some point
+//typename enable_if_interoperable<Dr1,Dr2,bool>::type // exposition
+bool
+operator ==(iterator_facade<Derived1,Value1,System1,Traversal1,Reference1,Difference1> const& lhs,
+            iterator_facade<Derived2,Value2,System2,Traversal2,Reference2,Difference2> const& rhs)
+{
+  return iterator_core_access
+    ::equal(*static_cast<Derived1 const*>(&lhs),
+            *static_cast<Derived2 const*>(&rhs));
+}
+
+template <typename Derived1, typename Value1, typename System1, typename Traversal1, typename Reference1, typename Difference1,
+          typename Derived2, typename Value2, typename System2, typename Traversal2, typename Reference2, typename Difference2>
+inline __host__ __device__
+// XXX it might be nice to implement this at some point
+//typename enable_if_interoperable<Dr1,Dr2,bool>::type // exposition
+bool
+operator !=(iterator_facade<Derived1,Value1,System1,Traversal1,Reference1,Difference1> const& lhs,
+            iterator_facade<Derived2,Value2,System2,Traversal2,Reference2,Difference2> const& rhs)
+{
+  return !iterator_core_access
+    ::equal(*static_cast<Derived1 const*>(&lhs),
+            *static_cast<Derived2 const*>(&rhs));
+}
+
+template <typename Derived1, typename Value1, typename System1, typename Traversal1, typename Reference1, typename Difference1,
+          typename Derived2, typename Value2, typename System2, typename Traversal2, typename Reference2, typename Difference2>
+inline __host__ __device__
+// XXX it might be nice to implement this at some point
+//typename enable_if_interoperable<Dr1,Dr2,bool>::type // exposition
+bool
+operator <(iterator_facade<Derived1,Value1,System1,Traversal1,Reference1,Difference1> const& lhs,
+           iterator_facade<Derived2,Value2,System2,Traversal2,Reference2,Difference2> const& rhs)
+{
+  return 0 > iterator_core_access
+    ::distance_from(*static_cast<Derived1 const*>(&lhs),
+                    *static_cast<Derived2 const*>(&rhs));
+}
+
+template <typename Derived1, typename Value1, typename System1, typename Traversal1, typename Reference1, typename Difference1,
+          typename Derived2, typename Value2, typename System2, typename Traversal2, typename Reference2, typename Difference2>
+inline __host__ __device__
+// XXX it might be nice to implement this at some point
+//typename enable_if_interoperable<Dr1,Dr2,bool>::type // exposition
+bool
+operator >(iterator_facade<Derived1,Value1,System1,Traversal1,Reference1,Difference1> const& lhs,
+           iterator_facade<Derived2,Value2,System2,Traversal2,Reference2,Difference2> const& rhs)
+{
+  return 0 < iterator_core_access
+    ::distance_from(*static_cast<Derived1 const*>(&lhs),
+                    *static_cast<Derived2 const*>(&rhs));
+}
+
+template <typename Derived1, typename Value1, typename System1, typename Traversal1, typename Reference1, typename Difference1,
+          typename Derived2, typename Value2, typename System2, typename Traversal2, typename Reference2, typename Difference2>
+inline __host__ __device__
+// XXX it might be nice to implement this at some point
+//typename enable_if_interoperable<Dr1,Dr2,bool>::type // exposition
+bool
+operator <=(iterator_facade<Derived1,Value1,System1,Traversal1,Reference1,Difference1> const& lhs,
+            iterator_facade<Derived2,Value2,System2,Traversal2,Reference2,Difference2> const& rhs)
+{
+  return 0 >= iterator_core_access
+    ::distance_from(*static_cast<Derived1 const*>(&lhs),
+                    *static_cast<Derived2 const*>(&rhs));
+}
+
+template <typename Derived1, typename Value1, typename System1, typename Traversal1, typename Reference1, typename Difference1,
+          typename Derived2, typename Value2, typename System2, typename Traversal2, typename Reference2, typename Difference2>
+inline __host__ __device__
+// XXX it might be nice to implement this at some point
+//typename enable_if_interoperable<Dr1,Dr2,bool>::type // exposition
+bool
+operator >=(iterator_facade<Derived1,Value1,System1,Traversal1,Reference1,Difference1> const& lhs,
+            iterator_facade<Derived2,Value2,System2,Traversal2,Reference2,Difference2> const& rhs)
+{
+  return 0 <= iterator_core_access
+    ::distance_from(*static_cast<Derived1 const*>(&lhs),
+                    *static_cast<Derived2 const*>(&rhs));
+}
+
+// Iterator difference
+template <typename Derived1, typename Value1, typename System1, typename Traversal1, typename Reference1, typename Difference1,
+          typename Derived2, typename Value2, typename System2, typename Traversal2, typename Reference2, typename Difference2>
+inline __host__ __device__
+
+// divine the type this operator returns
+typename thrust::detail::distance_from_result<
+  iterator_facade<Derived1,Value1,System1,Traversal1,Reference1,Difference1>,
+  iterator_facade<Derived2,Value2,System2,Traversal2,Reference2,Difference2>
+>::type
+
+operator-(iterator_facade<Derived1,Value1,System1,Traversal1,Reference1,Difference1> const& lhs,
+          iterator_facade<Derived2,Value2,System2,Traversal2,Reference2,Difference2> const& rhs)
+{
+  return iterator_core_access
+    ::distance_from(*static_cast<Derived1 const*>(&lhs),
+                    *static_cast<Derived2 const*>(&rhs));
+}
+
+// Iterator addition
+template <typename Derived, typename Value, typename System, typename Traversal, typename Reference, typename Difference>
+inline __host__ __device__
+Derived operator+ (iterator_facade<Derived,Value,System,Traversal,Reference,Difference> const& i,
+                   typename Derived::difference_type n)
+{
+  Derived tmp(static_cast<Derived const&>(i));
+  return tmp += n;
+}
+
+template <typename Derived, typename Value, typename System, typename Traversal, typename Reference, typename Difference>
+inline __host__ __device__
+Derived operator+ (typename Derived::difference_type n,
+                   iterator_facade<Derived,Value,System,Traversal,Reference,Difference> const& i)
+{
+  Derived tmp(static_cast<Derived const&>(i));
+  return tmp += n;
+}
+
+/*! \endcond
+ */
+
+/*! \} // end fancyiterators
+ */
+
+/*! \} // end iterators
+ */
+
+} // end thrust
+
diff --git a/compat/thrust/iterator/iterator_traits.h b/compat/thrust/iterator/iterator_traits.h
new file mode 100644
index 0000000000..a16f219b07
--- /dev/null
+++ b/compat/thrust/iterator/iterator_traits.h
@@ -0,0 +1,76 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file thrust/iterator/iterator_traits.h
+ *  \brief Traits and metafunctions for reasoning about the traits of iterators
+ */
+
+/*
+ * (C) Copyright David Abrahams 2003.
+ * 
+ * Distributed under the Boost Software License, Version 1.0.
+ * (See accompanying NOTICE file for the complete license)
+ *
+ * For more information, see http://www.boost.org
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <iterator>
+
+namespace thrust
+{
+
+/*! \p iterator_traits is a type trait class that provides a uniform
+ *  interface for querying the properties of iterators at compile-time.
+ */
+template<typename T>
+  struct iterator_traits
+    : public std::iterator_traits<T>
+{
+}; // end iterator_traits
+
+
+template<typename Iterator> struct iterator_value;
+
+template<typename Iterator> struct iterator_pointer;
+
+template<typename Iterator> struct iterator_reference;
+
+template<typename Iterator> struct iterator_difference;
+
+template<typename Iterator> struct iterator_traversal;
+
+template<typename Iterator> struct iterator_system;
+
+// TODO remove this in Thrust v1.7.0
+template<typename Iterator>
+  struct THRUST_DEPRECATED iterator_space
+{
+  typedef THRUST_DEPRECATED typename iterator_system<Iterator>::type type;
+};
+
+
+} // end thrust
+
+#include <thrust/iterator/detail/iterator_traversal_tags.h>
+#include <thrust/iterator/detail/host_system_tag.h>
+#include <thrust/iterator/detail/device_system_tag.h>
+#include <thrust/iterator/detail/any_system_tag.h>
+#include <thrust/iterator/detail/iterator_traits.inl>
+
diff --git a/compat/thrust/iterator/permutation_iterator.h b/compat/thrust/iterator/permutation_iterator.h
new file mode 100644
index 0000000000..509097b347
--- /dev/null
+++ b/compat/thrust/iterator/permutation_iterator.h
@@ -0,0 +1,210 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+/*! \file thrust/iterator/permutation_iterator.h
+ *  \brief An iterator which performs a gather or scatter operation when dereferenced
+ */
+
+/*
+ * (C) Copyright Toon Knapen    2001.
+ * (C) Copyright David Abrahams 2003.
+ * (C) Copyright Roland Richter 2003.
+ * 
+ * Distributed under the Boost Software License, Version 1.0.
+ * (See accompanying NOTICE file for the complete license)
+ *
+ * For more information, see http://www.boost.org
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/type_traits.h>
+#include <thrust/iterator/detail/permutation_iterator_base.h>
+#include <thrust/iterator/iterator_facade.h>
+#include <thrust/iterator/iterator_traits.h>
+
+namespace thrust
+{
+
+
+/*! \addtogroup iterators
+ *  \{
+ */
+
+/*! \addtogroup fancyiterator Fancy Iterators
+ *  \ingroup iterators
+ *  \{
+ */
+
+/*! \p permutation_iterator is an iterator which represents a pointer into a
+ *  reordered view of a given range. \p permutation_iterator is an imprecise name;
+ *  the reordered view need not be a strict permutation. This iterator is useful
+ *  for fusing a scatter or gather operation with other algorithms.
+ *
+ *  This iterator takes two arguments:
+ *
+ *    - an iterator to the range \c V on which the "permutation" will be applied
+ *    - the reindexing scheme that defines how the elements of \c V will be permuted.
+ *
+ *  Note that \p permutation_iterator is not limited to strict permutations of the
+ *  given range \c V. The distance between begin and end of the reindexing iterators
+ *  is allowed to be smaller compared to the size of the range \c V, in which case
+ *  the \p permutation_iterator only provides a "permutation" of a subrange of \c V.
+ *  The indices neither need to be unique. In this same context, it must be noted
+ *  that the past-the-end \p permutation_iterator is completely defined by means of
+ *  the past-the-end iterator to the indices.
+ *
+ *  The following code snippet demonstrates how to create a \p permutation_iterator
+ *  which represents a reordering of the contents of a \p device_vector.
+ *
+ *  \code
+ *  #include <thrust/iterator/permutation_iterator.h>
+ *  #include <thrust/device_vector.h>
+ *  ...
+ *  thrust::device_vector<float> values(4);
+ *  values[0] = 10.0f;
+ *  values[1] = 20.0f;
+ *  values[2] = 30.0f;
+ *  values[3] = 40.0f;
+ *  values[4] = 50.0f;
+ *  values[5] = 60.0f;
+ *  values[6] = 70.0f;
+ *  values[7] = 80.0f;
+ *
+ *  thrust::device_vector<int> indices(4);
+ *  indices[0] = 2;
+ *  indices[1] = 6;
+ *  indices[2] = 1;
+ *  indices[3] = 3;
+ *
+ *  typedef thrust::device_vector<float>::iterator ElementIterator;
+ *  typedef thrust::device_vector<int>::iterator   IndexIterator;
+ *
+ *  thrust::permutation_iterator<ElementIterator,IndexIterator> iter(values.begin(), indices.begin());
+ *
+ *  *iter;   // returns 30.0f;
+ *  iter[0]; // returns 30.0f;
+ *  iter[1]; // returns 70.0f;
+ *  iter[2]; // returns 20.0f;
+ *  iter[3]; // returns 40.0f;
+ *
+ *  // iter[4] is an out-of-bounds error
+ *
+ *  *iter   = -1.0f; // sets values[2] to -1.0f;
+ *  iter[0] = -1.0f; // sets values[2] to -1.0f;
+ *  iter[1] = -1.0f; // sets values[6] to -1.0f;
+ *  iter[2] = -1.0f; // sets values[1] to -1.0f;
+ *  iter[3] = -1.0f; // sets values[3] to -1.0f;
+ *
+ *  // values is now {10, -1, -1, -1, 50, 60, -1, 80}
+ *  \endcode
+ *
+ *  \see make_permutation_iterator
+ */
+template <typename ElementIterator,
+          typename IndexIterator>
+  class permutation_iterator
+    : public thrust::detail::permutation_iterator_base<
+        ElementIterator,
+        IndexIterator
+      >::type
+{
+  /*! \cond
+   */
+  private:
+    typedef typename detail::permutation_iterator_base<ElementIterator,IndexIterator>::type super_t;
+
+    friend class thrust::iterator_core_access;
+  /*! \endcond
+   */
+
+  public:
+    /*! Null constructor calls the null constructor of this \p permutation_iterator's
+     *  element iterator.
+     */
+    __host__ __device__
+    permutation_iterator()
+      : m_element_iterator() {}
+
+    /*! Constructor accepts an \c ElementIterator into a range of values and an
+     *  \c IndexIterator into a range of indices defining the indexing scheme on the
+     *  values.
+     *
+     *  \param x An \c ElementIterator pointing this \p permutation_iterator's range of values.
+     *  \param y An \c IndexIterator pointing to an indexing scheme to use on \p x.
+     */
+    __host__ __device__
+    explicit permutation_iterator(ElementIterator x, IndexIterator y)
+      : super_t(y), m_element_iterator(x) {}
+
+    /*! Copy constructor accepts a related \p permutation_iterator.
+     *  \param r A compatible \p permutation_iterator to copy from.
+     */
+    template<typename OtherElementIterator, typename OtherIndexIterator>
+    __host__ __device__
+    permutation_iterator(permutation_iterator<OtherElementIterator,OtherIndexIterator> const &r
+    // XXX remove these guards when we have static_assert
+    , typename detail::enable_if_convertible<OtherElementIterator, ElementIterator>::type* = 0
+    , typename detail::enable_if_convertible<OtherIndexIterator, IndexIterator>::type* = 0
+    )
+      : super_t(r.base()), m_element_iterator(r.m_element_iterator)
+    {}
+
+  /*! \cond
+   */
+  private:
+    __thrust_hd_warning_disable__
+    __host__ __device__
+    typename super_t::reference dereference() const
+    {
+      return *(m_element_iterator + *this->base());
+    }
+
+    // make friends for the copy constructor
+    template<typename,typename> friend class permutation_iterator;
+
+    ElementIterator m_element_iterator;
+  /*! \endcond
+   */
+}; // end permutation_iterator
+
+
+/*! \p make_permutation_iterator creates a \p permutation_iterator
+ *  from an \c ElementIterator pointing to a range of elements to "permute"
+ *  and an \c IndexIterator pointing to a range of indices defining an indexing
+ *  scheme on the values.
+ *
+ *  \param e An \c ElementIterator pointing to a range of values.
+ *  \param i An \c IndexIterator pointing to an indexing scheme to use on \p e.
+ *  \return A new \p permutation_iterator which permutes the range \p e by \p i.
+ *  \see permutation_iterator
+ */
+template<typename ElementIterator, typename IndexIterator>
+__host__ __device__
+permutation_iterator<ElementIterator,IndexIterator> make_permutation_iterator(ElementIterator e, IndexIterator i)
+{
+  return permutation_iterator<ElementIterator,IndexIterator>(e,i);
+}
+
+/*! \} // end fancyiterators
+ */
+
+/*! \} // end iterators
+ */
+
+} // end thrust
+
diff --git a/compat/thrust/iterator/retag.h b/compat/thrust/iterator/retag.h
new file mode 100644
index 0000000000..660da8f2fd
--- /dev/null
+++ b/compat/thrust/iterator/retag.h
@@ -0,0 +1,68 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+/*! \file thrust/iterator/retag.h
+ *  \brief Functionality for altering an iterator's associated system.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/iterator/detail/retag.h>
+
+namespace thrust
+{
+
+
+/*! \ingroup iterator_tags
+ *  \{
+ */
+
+#if 0
+/*! \p reinterpret_tag returns a copy of an iterator and changes the type of the result's system tag.
+ *  \tparam Tag Any system tag.
+ *  \tparam Iterator Any iterator type.
+ *  \param iter The iterator of interest.
+ *  \return An iterator of unspecified type whose system tag is \p Tag and whose behavior is otherwise
+ *          equivalent to \p iter.
+ *  \note Unlike \p retag, \p reinterpret_tag does not enforce that the converted-to system tag be
+ *        related to the converted-from system tag.
+ *  \see retag
+ */
+template<typename Tag, typename Iterator>
+unspecified_iterator_type reinterpret_tag(Iterator iter);
+
+/*! \p retag returns a copy of an iterator and changes the type of the result's system tag.
+ *  \tparam Tag \p Tag shall be convertible to <tt>thrust::iterator_system<Iterator>::type</tt>,
+ *              or <tt>thrust::iterator_system<Iterator>::type</tt> is a base type of \p Tag.
+ *  \tparam Iterator Any iterator type.
+ *  \param iter The iterator of interest.
+ *  \return An iterator of unspecified type whose system tag is \p Tag and whose behavior is
+ *          otherwise equivalent to \p iter.
+ *  \note Unlike \p reinterpret_tag, \p retag enforces that the converted-to system tag be
+ *        related to the converted-from system tag.
+ *  \see reinterpret_tag
+ */
+template<typename Tag, typename Iterator>
+unspecified_iterator_type retag(Iterator iter);
+#endif
+
+/*! \} // iterator_tags
+ */
+
+
+} // end thrust
+
diff --git a/compat/thrust/iterator/reverse_iterator.h b/compat/thrust/iterator/reverse_iterator.h
new file mode 100644
index 0000000000..03f03396d3
--- /dev/null
+++ b/compat/thrust/iterator/reverse_iterator.h
@@ -0,0 +1,238 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file thrust/iterator/reverse_iterator.h
+ *  \brief An iterator adaptor which adapts another iterator to traverse backwards
+ */
+
+/*
+ * (C) Copyright David Abrahams 2002.
+ * (C) Copyright Jeremy Siek    2002.
+ * (C) Copyright Thomas Witt    2002.
+ * 
+ * Distributed under the Boost Software License, Version 1.0.
+ * (See accompanying NOTICE file for the complete license)
+ *
+ * For more information, see http://www.boost.org
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/type_traits.h>
+#include <thrust/iterator/detail/reverse_iterator_base.h>
+#include <thrust/iterator/iterator_facade.h>
+
+namespace thrust
+{
+
+/*! \addtogroup iterators
+ *  \{
+ */
+
+/*! \addtogroup fancyiterator Fancy Iterators
+ *  \ingroup iterators
+ *  \{
+ */
+
+/*! \p reverse_iterator is an iterator which represents a pointer into a
+ *  reversed view of a given range. In this way, \p reverse_iterator allows
+ *  backwards iteration through a bidirectional input range.
+ *
+ *  It is important to note that although \p reverse_iterator is constructed
+ *  from a given iterator, it points to the element preceding it. In this way,
+ *  the past-the-end \p reverse_iterator of a given range points to the element
+ *  preceding the first element of the input range. By the same token, the first
+ *  \p reverse_iterator of a given range is constructed from a past-the-end iterator
+ *  of the original range yet points to the last element of the input.
+ *
+ *  The following code snippet demonstrates how to create a \p reverse_iterator
+ *  which represents a reversed view of the contents of a \p device_vector.
+ *
+ *  \code
+ *  #include <thrust/iterator/reverse_iterator.h>
+ *  #include <thrust/device_vector.h>
+ *  ...
+ *  thrust::device_vector<float> v(4);
+ *  v[0] = 0.0f;
+ *  v[1] = 1.0f;
+ *  v[2] = 2.0f;
+ *  v[3] = 3.0f;
+ *
+ *  typedef thrust::device_vector<float>::iterator Iterator;
+ *
+ *  // note that we point the iterator to the *end* of the device_vector
+ *  thrust::reverse_iterator<Iterator> iter(values.end());
+ *
+ *  *iter;   // returns 3.0f;
+ *  iter[0]; // returns 3.0f;
+ *  iter[1]; // returns 2.0f;
+ *  iter[2]; // returns 1.0f;
+ *  iter[3]; // returns 0.0f;
+ *
+ *  // iter[4] is an out-of-bounds error
+ *  \endcode
+ *
+ *  Since reversing a range is a common operation, containers like \p device_vector
+ *  have nested typedefs for declaration shorthand and methods for constructing
+ *  reverse_iterators. The following code snippet is equivalent to the previous:
+ *
+ *  \code
+ *  #include <thrust/device_vector.h>
+ *  ...
+ *  thrust::device_vector<float> v(4);
+ *  v[0] = 0.0f;
+ *  v[1] = 1.0f;
+ *  v[2] = 2.0f;
+ *  v[3] = 3.0f;
+ *
+ *  // we use the nested type reverse_iterator to refer to a reversed view of
+ *  // a device_vector and the method rbegin() to create a reverse_iterator pointing
+ *  // to the beginning of the reversed device_vector
+ *  thrust::device_iterator<float>::reverse_iterator iter = values.rbegin();
+ *
+ *  *iter;   // returns 3.0f;
+ *  iter[0]; // returns 3.0f;
+ *  iter[1]; // returns 2.0f;
+ *  iter[2]; // returns 1.0f;
+ *  iter[3]; // returns 0.0f;
+ *
+ *  // iter[4] is an out-of-bounds error
+ *
+ *  // similarly, rend() points to the end of the reversed sequence:
+ *  assert(values.rend() == (iter + 4));
+ *  \endcode
+ *
+ *  Finally, the following code snippet demonstrates how to use reverse_iterator to
+ *  perform a reversed prefix sum operation on the contents of a device_vector:
+ *
+ *  \code
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/scan.h>
+ *  ...
+ *  thrust::device_vector<int> v(5);
+ *  v[0] = 0;
+ *  v[1] = 1;
+ *  v[2] = 2;
+ *  v[3] = 3;
+ *  v[4] = 4;
+ *
+ *  thrust::device_vector<int> result(5);
+ *
+ *  // exclusive scan v into result in reverse
+ *  thrust::exclusive_scan(v.rbegin(), v.rend(), result.begin());
+ *
+ *  // result is now {0, 4, 7, 9, 10}
+ *  \endcode
+ *
+ *  \see make_reverse_iterator
+ */
+template<typename BidirectionalIterator>
+  class reverse_iterator
+    : public detail::reverse_iterator_base<BidirectionalIterator>::type
+{
+  /*! \cond
+   */
+  private:
+    typedef typename thrust::detail::reverse_iterator_base<
+      BidirectionalIterator
+    >::type super_t;
+
+    friend class thrust::iterator_core_access;
+  /*! \endcond
+   */
+
+  public:
+    /*! Default constructor does nothing.
+     */
+    __host__ __device__
+    reverse_iterator(void) {}
+
+    /*! \p Constructor accepts a \c BidirectionalIterator pointing to a range
+     *  for this \p reverse_iterator to reverse.
+     *
+     *  \param x A \c BidirectionalIterator pointing to a range to reverse.
+     */
+    __host__ __device__
+    explicit reverse_iterator(BidirectionalIterator x);
+
+    /*! \p Copy constructor allows construction from a related compatible
+     *  \p reverse_iterator.
+     *
+     *  \param r A \p reverse_iterator to copy from.
+     */
+    template<typename OtherBidirectionalIterator>
+    __host__ __device__
+    reverse_iterator(reverse_iterator<OtherBidirectionalIterator> const &r
+// XXX msvc screws this up
+// XXX remove these guards when we have static_assert
+#ifndef _MSC_VER
+                     , typename thrust::detail::enable_if<
+                         thrust::detail::is_convertible<
+                           OtherBidirectionalIterator,
+                           BidirectionalIterator
+                         >::value
+                       >::type * = 0
+#endif // _MSC_VER
+                     );
+
+  /*! \cond
+   */
+  private:
+    __thrust_hd_warning_disable__
+    __host__ __device__
+    typename super_t::reference dereference(void) const;
+
+    __host__ __device__
+    void increment(void);
+
+    __host__ __device__
+    void decrement(void);
+
+    __host__ __device__
+    void advance(typename super_t::difference_type n);
+
+    template<typename OtherBidirectionalIterator>
+    __host__ __device__
+    typename super_t::difference_type
+    distance_to(reverse_iterator<OtherBidirectionalIterator> const &y) const;
+  /*! \endcond
+   */
+}; // end reverse_iterator
+
+
+/*! \p make_reverse_iterator creates a \p reverse_iterator
+ *  from a \c BidirectionalIterator pointing to a range of elements to reverse.
+ *  
+ *  \param x A \c BidirectionalIterator pointing to a range to reverse.
+ *  \return A new \p reverse_iterator which reverses the range \p x.
+ */
+template<typename BidirectionalIterator>
+__host__ __device__
+reverse_iterator<BidirectionalIterator> make_reverse_iterator(BidirectionalIterator x);
+
+
+/*! \} // end fancyiterators
+ */
+
+/*! \} // end iterators
+ */
+
+} // end thrust
+
+#include <thrust/iterator/detail/reverse_iterator.inl>
+
diff --git a/compat/thrust/iterator/transform_iterator.h b/compat/thrust/iterator/transform_iterator.h
new file mode 100644
index 0000000000..985b61b775
--- /dev/null
+++ b/compat/thrust/iterator/transform_iterator.h
@@ -0,0 +1,344 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file thrust/iterator/transform_iterator.h
+ *  \brief An iterator which adapts another iterator by applying a function to the result of its dereference 
+ */
+
+/*
+ * (C) Copyright David Abrahams 2002.
+ * (C) Copyright Jeremy Siek    2002.
+ * (C) Copyright Thomas Witt    2002.
+ * 
+ * Distributed under the Boost Software License, Version 1.0.
+ * (See accompanying NOTICE file for the complete license)
+ *
+ * For more information, see http://www.boost.org
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// #include the details first
+#include <thrust/iterator/detail/transform_iterator.inl>
+#include <thrust/iterator/iterator_facade.h>
+#include <thrust/detail/type_traits.h>
+#include <thrust/detail/function.h>
+
+namespace thrust
+{
+
+/*! \addtogroup iterators
+ *  \{
+ */
+
+/*! \addtogroup fancyiterator Fancy Iterators
+ *  \ingroup iterators
+ *  \{
+ */
+
+/*! \p transform_iterator is an iterator which represents a pointer into a range
+ *  of values after transformation by a function. This iterator is useful for 
+ *  creating a range filled with the result of applying an operation to another range
+ *  without either explicitly storing it in memory, or explicitly executing the transformation.
+ *  Using \p transform_iterator facilitates kernel fusion by deferring the execution
+ *  of a transformation until the value is needed while saving both memory capacity
+ *  and bandwidth.
+ *
+ *  The following code snippet demonstrates how to create a \p transform_iterator
+ *  which represents the result of \c sqrtf applied to the contents of a \p device_vector.
+ *
+ *  \code
+ *  #include <thrust/iterator/transform_iterator.h>
+ *  #include <thrust/device_vector.h>
+ *  
+ *  // note: functor inherits from unary_function
+ *  struct square_root : public thrust::unary_function<float,float>
+ *  {
+ *    __host__ __device__
+ *    float operator()(float x) const
+ *    {
+ *      return sqrtf(x);
+ *    }
+ *  };
+ *  
+ *  int main(void)
+ *  {
+ *    thrust::device_vector<float> v(4);
+ *    v[0] = 1.0f;
+ *    v[1] = 4.0f;
+ *    v[2] = 9.0f;
+ *    v[3] = 16.0f;
+ *                                                                                           
+ *    typedef thrust::device_vector<float>::iterator FloatIterator;
+ *                                                                                           
+ *    thrust::transform_iterator<square_root, FloatIterator> iter(v.begin(), square_root());
+ *                                                                                           
+ *    *iter;   // returns 1.0f
+ *    iter[0]; // returns 1.0f;
+ *    iter[1]; // returns 2.0f;
+ *    iter[2]; // returns 3.0f;
+ *    iter[3]; // returns 4.0f;
+ *                                                                                           
+ *    // iter[4] is an out-of-bounds error
+ *  }
+ *  \endcode
+ *
+ *  This next example demonstrates how to use a \p transform_iterator with the
+ *  \p thrust::reduce function to compute the sum of squares of a sequence.
+ *  We will create temporary \p transform_iterators with the
+ *  \p make_transform_iterator function in order to avoid explicitly specifying their type:
+ *
+ *  \code
+ *  #include <thrust/iterator/transform_iterator.h>
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/reduce.h>
+ *  #include <iostream>
+ *  
+ *  // note: functor inherits from unary_function
+ *  struct square : public thrust::unary_function<float,float>
+ *  {
+ *    __host__ __device__
+ *    float operator()(float x) const
+ *    {
+ *      return x * x;
+ *    }
+ *  };
+ *  
+ *  int main(void)
+ *  {
+ *    // initialize a device array
+ *    thrust::device_vector<float> v(4);
+ *    v[0] = 1.0f;
+ *    v[1] = 2.0f;
+ *    v[2] = 3.0f;
+ *    v[3] = 4.0f;
+ *  
+ *    float sum_of_squares =
+ *     thrust::reduce(thrust::make_transform_iterator(v.begin(), square()),
+ *                    thrust::make_transform_iterator(v.end(),   square()));
+ *  
+ *    std::cout << "sum of squares: " << sum_of_squares << std::endl;
+ *    return 0;
+ *  }
+ *  \endcode
+ *
+ *  Note that in the previous two examples the transform functor (namely \c square_root 
+ *  and \c square) inherits from \c thrust::unary_function.  Inheriting from 
+ *  \c thrust::unary_function ensures that a functor is a valid \c AdaptableUnaryFunction
+ *  and provides all the necessary \c typedef declarations.  The \p transform_iterator
+ *  can also be applied to a \c UnaryFunction that does not inherit from 
+ *  \c thrust::unary_function using an optional template argument.  The following example
+ *  illustrates how to use the third template argument to specify the \c result_type of
+ *  the function.   
+ *
+ *  \code
+ *  #include <thrust/iterator/transform_iterator.h>
+ *  #include <thrust/device_vector.h>
+ *  
+ *  // note: functor *does not* inherit from unary_function
+ *  struct square_root
+ *  {
+ *    __host__ __device__
+ *    float operator()(float x) const
+ *    {
+ *      return sqrtf(x);
+ *    }
+ *  };
+ *  
+ *  int main(void)
+ *  {
+ *    thrust::device_vector<float> v(4);
+ *    v[0] = 1.0f;
+ *    v[1] = 4.0f;
+ *    v[2] = 9.0f;
+ *    v[3] = 16.0f;
+ *                                                                                           
+ *    typedef thrust::device_vector<float>::iterator FloatIterator;
+ *    
+ *    // note: float result_type is specified explicitly
+ *    thrust::transform_iterator<square_root, FloatIterator, float> iter(v.begin(), square_root());
+ *                                                                                           
+ *    *iter;   // returns 1.0f
+ *    iter[0]; // returns 1.0f;
+ *    iter[1]; // returns 2.0f;
+ *    iter[2]; // returns 3.0f;
+ *    iter[3]; // returns 4.0f;
+ *                                                                                           
+ *    // iter[4] is an out-of-bounds error
+ *  }
+ *  \endcode
+ *
+ *  \see make_transform_iterator
+ */
+template <class AdaptableUnaryFunction, class Iterator, class Reference = use_default, class Value = use_default>
+  class transform_iterator
+    : public detail::transform_iterator_base<AdaptableUnaryFunction, Iterator, Reference, Value>::type
+{
+  /*! \cond
+   */
+  public:
+    typedef typename
+    detail::transform_iterator_base<AdaptableUnaryFunction, Iterator, Reference, Value>::type
+    super_t;
+
+    friend class thrust::iterator_core_access;
+  /*! \endcond
+   */
+
+  public:
+    /*! Null constructor does nothing.
+     */
+    __host__ __device__
+    transform_iterator() {}
+  
+    /*! This constructor takes as arguments an \c Iterator and an \c AdaptableUnaryFunction
+     *  and copies them to a new \p transform_iterator.
+     *
+     *  \param x An \c Iterator pointing to the input to this \p transform_iterator's \c AdaptableUnaryFunction.
+     *  \param f An \c AdaptableUnaryFunction used to transform the objects pointed to by \p x.
+     */
+    __host__ __device__
+    transform_iterator(Iterator const& x, AdaptableUnaryFunction f)
+      : super_t(x), m_f(f) {
+    }
+  
+    /*! This explicit constructor copies the value of a given \c Iterator and creates
+     *  this \p transform_iterator's \c AdaptableUnaryFunction using its null constructor.
+     *
+     *  \param x An \c Iterator to copy.
+     */
+    __host__ __device__
+    explicit transform_iterator(Iterator const& x)
+      : super_t(x) { }
+
+    /*! This copy constructor creates a new \p transform_iterator from another
+     *  \p transform_iterator.
+     *
+     *  \param other The \p transform_iterator to copy.
+     */
+    template<typename OtherAdaptableUnaryFunction,
+             typename OtherIterator,
+             typename OtherReference,
+             typename OtherValue>
+    __host__ __device__
+    transform_iterator(const transform_iterator<OtherAdaptableUnaryFunction, OtherIterator, OtherReference, OtherValue> &other,
+                       typename thrust::detail::enable_if_convertible<OtherIterator, Iterator>::type* = 0,
+                       typename thrust::detail::enable_if_convertible<OtherAdaptableUnaryFunction, AdaptableUnaryFunction>::type* = 0)
+      : super_t(other.base()), m_f(other.functor()) {}
+
+    /*! Copy assignment operator copies from another \p transform_iterator.
+     *  \p other The other \p transform_iterator to copy
+     *  \return <tt>*this</tt>
+     *
+     *  \note If the type of this \p transform_iterator's functor is not copy assignable
+     *        (for example, if it is a lambda) it is not an error to call this function.
+     *        In this case, however, the functor will not be modified.
+     *
+     *        In any case, this \p transform_iterator's underlying iterator will be copy assigned.
+     */
+    __host__ __device__
+    transform_iterator &operator=(const transform_iterator &other)
+    {
+      return do_assign(other,
+      // XXX gcc 4.2.1 crashes on is_copy_assignable; just assume the functor is assignable as a WAR
+#if (THRUST_HOST_COMPILER == THRUST_HOST_COMPILER_GCC) && (THRUST_GCC_VERSION <= 40201)
+          thrust::detail::true_type()
+#else
+          typename thrust::detail::is_copy_assignable<AdaptableUnaryFunction>::type()
+#endif // THRUST_HOST_COMPILER
+      );
+    }
+
+    /*! This method returns a copy of this \p transform_iterator's \c AdaptableUnaryFunction.
+     *  \return A copy of this \p transform_iterator's \c AdaptableUnaryFunction.
+     */
+    __host__ __device__
+    AdaptableUnaryFunction functor() const
+      { return m_f; }
+
+    /*! \cond
+     */
+  private:
+    __host__ __device__
+    transform_iterator &do_assign(const transform_iterator &other, thrust::detail::true_type)
+    {
+      super_t::operator=(other);
+
+      // do assign to m_f
+      m_f = other.functor();
+
+      return *this;
+    }
+
+    __host__ __device__
+    transform_iterator &do_assign(const transform_iterator &other, thrust::detail::false_type)
+    {
+      super_t::operator=(other);
+
+      // don't assign to m_f
+
+      return *this;
+    }
+
+    __thrust_hd_warning_disable__
+    __host__ __device__
+    typename super_t::reference dereference() const
+    { 
+      // XXX consider making this a member instead of a temporary created inside dereference
+      thrust::detail::host_device_function<AdaptableUnaryFunction, typename super_t::reference> wrapped_f(m_f);
+
+      return wrapped_f(*this->base());
+    }
+
+    // tag this as mutable per Dave Abrahams in this thread:
+    // http://lists.boost.org/Archives/boost/2004/05/65332.php
+    mutable AdaptableUnaryFunction m_f;
+
+    /*! \endcond
+     */
+}; // end transform_iterator
+
+
+/*! \p make_transform_iterator creates a \p transform_iterator
+ *  from an \c Iterator and \c AdaptableUnaryFunction.
+ *
+ *  \param it The \c Iterator pointing to the input range of the
+ *            newly created \p transform_iterator.
+ *  \param fun The \c AdaptableUnaryFunction used to transform the range pointed
+ *             to by \p it in the newly created \p transform_iterator.
+ *  \return A new \p transform_iterator which transforms the range at
+ *          \p it by \p fun.
+ *  \see transform_iterator
+ */
+template <class AdaptableUnaryFunction, class Iterator>
+inline __host__ __device__
+transform_iterator<AdaptableUnaryFunction, Iterator>
+make_transform_iterator(Iterator it, AdaptableUnaryFunction fun)
+{
+  return transform_iterator<AdaptableUnaryFunction, Iterator>(it, fun);
+} // end make_transform_iterator
+
+/*! \} // end fancyiterators
+ */
+
+/*! \} // end iterators
+ */
+
+} // end thrust
+
diff --git a/compat/thrust/iterator/zip_iterator.h b/compat/thrust/iterator/zip_iterator.h
new file mode 100644
index 0000000000..8e7299c407
--- /dev/null
+++ b/compat/thrust/iterator/zip_iterator.h
@@ -0,0 +1,245 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file thrust/iterator/zip_iterator.h
+ *  \brief An iterator which returns a tuple of the result of dereferencing
+ *         a tuple of iterators when dereferenced
+ */
+
+/*
+ * Copyright David Abrahams and Thomas Becker 2000-2006.
+ * 
+ * Distributed under the Boost Software License, Version 1.0.
+ * (See accompanying NOTICE file for the complete license)
+ *
+ * For more information, see http://www.boost.org
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/iterator/detail/zip_iterator_base.h>
+#include <thrust/iterator/iterator_facade.h>
+#include <thrust/detail/type_traits.h>
+
+namespace thrust
+{
+
+/*! \addtogroup iterators
+ *  \{
+ */
+
+/*! \addtogroup fancyiterator Fancy Iterators
+ *  \ingroup iterators
+ *  \{
+ */
+
+/*! \p zip_iterator is an iterator which represents a pointer into a range
+ *  of \p tuples whose elements are themselves taken from a \p tuple of input
+ *  iterators. This iterator is useful for creating a virtual array of structures
+ *  while achieving the same performance and bandwidth as the structure of arrays
+ *  idiom. \p zip_iterator also facilitates kernel fusion by providing a convenient
+ *  means of amortizing the execution of the same operation over multiple ranges.
+ *
+ *  The following code snippet demonstrates how to create a \p zip_iterator
+ *  which represents the result of "zipping" multiple ranges together.
+ *  
+ *  \code
+ *  #include <thrust/iterator/zip_iterator.h>
+ *  #include <thrust/tuple.h>
+ *  #include <thrust/device_vector.h>
+ *  ...
+ *  thrust::device_vector<int> int_v(3);
+ *  int_v[0] = 0; int_v[1] = 1; int_v[2] = 2;
+ *
+ *  thrust::device_vector<float> float_v(3);
+ *  float_v[0] = 0.0f; float_v[1] = 1.0;f float_v[2] = 2.0f;
+ *
+ *  thrust::device_vector<char> char_v(3);
+ *  char_v[0] = 'a'; char_v[1] = 'b'; char_v[2] = 'c';
+ *
+ *  // typedef these iterators for shorthand
+ *  typedef thrust::device_vector<int>::iterator   IntIterator;
+ *  typedef thrust::device_vector<float>::iterator FloatIterator;
+ *  typedef thrust::device_vector<char>::iterator  CharIterator;
+ *
+ *  // typedef a tuple of these iterators
+ *  typedef thrust::tuple<IntIterator, FloatIterator, CharIterator> IteratorTuple;
+ *
+ *  // typedef the zip_iterator of this tuple
+ *  typedef thrust::zip_iterator<IteratorTuple> ZipIterator;
+ *
+ *  // finally, create the zip_iterator
+ *  ZipIterator iter(thrust::make_tuple(int_v.begin(), float_v.begin(), char_v.begin()));
+ *
+ *  *iter;   // returns (0, 0.0f, 'a')
+ *  iter[0]; // returns (0, 0.0f, 'a')
+ *  iter[1]; // returns (1, 1.0f, 'b')
+ *  iter[2]; // returns (2, 2.0f, 'c')
+ *
+ *  thrust::get<0>(iter[2]); // returns 2
+ *  thrust::get<1>(iter[0]); // returns 0.0f
+ *  thrust::get<2>(iter[1]); // returns 'b'
+ *
+ *  // iter[3] is an out-of-bounds error
+ *  \endcode
+ *
+ *  Defining the type of a \p zip_iterator can be complex. The next code example demonstrates
+ *  how to use the \p make_zip_iterator function with the \p make_tuple function to avoid
+ *  explicitly specifying the type of the \p zip_iterator. This example shows how to use
+ *  \p zip_iterator to copy multiple ranges with a single call to \p thrust::copy.
+ *
+ *  \code
+ *  #include <thrust/zip_iterator.h>
+ *  #include <thrust/tuple.h>
+ *  #include <thrust/device_vector.h>
+ *
+ *  int main(void)
+ *  {
+ *    thrust::device_vector<int> int_in(3), int_out(3);
+ *    int_in[0] = 0;
+ *    int_in[1] = 1;
+ *    int_in[2] = 2;
+ *
+ *    thrust::device_vector<float> float_in(3), float_out(3);
+ *    float_in[0] =  0.0f;
+ *    float_in[1] = 10.0f;
+ *    float_in[2] = 20.0f;
+ *
+ *    thrust::copy(thrust::make_zip_iterator(thrust::make_tuple(int_in.begin(), float_in.begin())),
+ *                 thrust::make_zip_iterator(thrust::make_tuple(int_in.end(),   float_in.end())),
+ *                 thrust::make_zip_iterator(thrust::make_tuple(int_out.begin(),float_out.begin())));
+ *
+ *    // int_out is now [0, 1, 2]
+ *    // float_out is now [0.0f, 10.0f, 20.0f]
+ *
+ *    return 0;
+ *  }
+ *  \endcode
+ *
+ *  \see make_zip_iterator
+ *  \see make_tuple
+ *  \see tuple
+ *  \see get
+ */
+template <typename IteratorTuple>
+  class zip_iterator
+    : public detail::zip_iterator_base<IteratorTuple>::type
+{
+  public:
+    /*! Null constructor does nothing.
+     */
+    inline __host__ __device__
+    zip_iterator(void);
+
+    /*! This constructor creates a new \p zip_iterator from a
+     *  \p tuple of iterators.
+     *  
+     *  \param iterator_tuple The \p tuple of iterators to copy from.
+     */
+    inline __host__ __device__
+    zip_iterator(IteratorTuple iterator_tuple);
+
+    /*! This copy constructor creates a new \p zip_iterator from another
+     *  \p zip_iterator.
+     *
+     *  \param other The \p zip_iterator to copy.
+     */
+    template<typename OtherIteratorTuple>
+    inline __host__ __device__
+    zip_iterator(const zip_iterator<OtherIteratorTuple> &other,
+                 typename thrust::detail::enable_if_convertible<
+                   OtherIteratorTuple,
+                   IteratorTuple
+                 >::type * = 0);
+
+    /*! This method returns a \c const reference to this \p zip_iterator's
+     *  \p tuple of iterators.
+     *
+     *  \return A \c const reference to this \p zip_iterator's \p tuple
+     *          of iterators.
+     */
+    inline __host__ __device__
+    const IteratorTuple &get_iterator_tuple() const;
+
+    /*! \cond
+     */
+  private:
+    typedef typename
+    detail::zip_iterator_base<IteratorTuple>::type super_t;
+
+    friend class thrust::iterator_core_access;
+
+    // Dereferencing returns a tuple built from the dereferenced
+    // iterators in the iterator tuple.
+    __host__ __device__
+    typename super_t::reference dereference() const;
+
+    // Two zip_iterators are equal if the two first iterators of the
+    // tuple are equal. Note this differs from Boost's implementation, which
+    // considers the entire tuple.
+    template<typename OtherIteratorTuple>
+    inline __host__ __device__
+    bool equal(const zip_iterator<OtherIteratorTuple> &other) const;
+
+    // Advancing a zip_iterator means to advance all iterators in the tuple
+    inline __host__ __device__
+    void advance(typename super_t::difference_type n);
+
+    // Incrementing a zip iterator means to increment all iterators in the tuple
+    inline __host__ __device__
+    void increment();
+
+    // Decrementing a zip iterator means to decrement all iterators in the tuple
+    inline __host__ __device__
+    void decrement();
+
+    // Distance is calculated using the first iterator in the tuple.
+    template<typename OtherIteratorTuple>
+    inline __host__ __device__
+      typename super_t::difference_type
+        distance_to(const zip_iterator<OtherIteratorTuple> &other) const;
+
+    // The iterator tuple.
+    IteratorTuple m_iterator_tuple;
+
+    /*! \endcond
+     */
+}; // end zip_iterator
+
+/*! \p make_zip_iterator creates a \p zip_iterator from a \p tuple
+ *  of iterators.
+ *
+ *  \param t The \p tuple of iterators to copy.
+ *  \return A newly created \p zip_iterator which zips the iterators encapsulated in \p t.
+ *
+ *  \see zip_iterator
+ */
+template<typename IteratorTuple>
+inline __host__ __device__
+zip_iterator<IteratorTuple> make_zip_iterator(IteratorTuple t);
+
+/*! \} // end fancyiterators
+ */
+
+/*! \} // end iterators
+ */
+
+} // end thrust
+
+#include <thrust/iterator/detail/zip_iterator.inl>
+
diff --git a/compat/thrust/logical.h b/compat/thrust/logical.h
new file mode 100644
index 0000000000..21510f3f21
--- /dev/null
+++ b/compat/thrust/logical.h
@@ -0,0 +1,276 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file logical.h
+ *  \brief Logical operations on ranges
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/execution_policy.h>
+
+namespace thrust
+{
+
+
+/*! \addtogroup reductions
+ *  \{
+ *  \addtogroup logical
+ *  \ingroup reductions
+ *  \{
+ */
+
+
+/*! \p all_of determines whether all elements in a range satify a predicate.
+ *  Specifically, \p all_of returns \c true if <tt>pred(*i)</tt> is \c true
+ *  for every iterator \c i in the range <tt>[first, last)</tt> and 
+ *  \c false otherwise.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the sequence.
+ *  \param last  The end of the sequence.
+ *  \param pred A predicate used to test range elements.
+ *  \return \c true, if all elements satisfy the predicate; \c false, otherwise.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *  \tparam Predicate must be a model of <a href="http://www.sgi.com/tech/stl/Predicate.html">Predicate</a>.
+ *
+ *  \code
+ *  #include <thrust/logical.h>
+ *  #include <thrust/functional.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  bool A[3] = {true, true, false};
+ *
+ *  thrust::all_of(thrust::host, A, A + 2, thrust::identity<bool>()); // returns true
+ *  thrust::all_of(thrust::host, A, A + 3, thrust::identity<bool>()); // returns false
+ *
+ *  // empty range
+ *  thrust::all_of(thrust::host, A, A, thrust::identity<bool>()); // returns false
+ *  
+ *  \endcode
+ *
+ *  \see any_of
+ *  \see none_of
+ *  \see transform_reduce
+ */
+template <typename DerivedPolicy, typename InputIterator, typename Predicate>
+bool all_of(const thrust::detail::execution_policy_base<DerivedPolicy> &exec, InputIterator first, InputIterator last, Predicate pred);
+
+
+/*! \p all_of determines whether all elements in a range satify a predicate.
+ * Specifically, \p all_of returns \c true if <tt>pred(*i)</tt> is \c true
+ * for every iterator \c i in the range <tt>[first, last)</tt> and 
+ * \c false otherwise.
+ *
+ *  \param first The beginning of the sequence.
+ *  \param last  The end of the sequence.
+ *  \param pred A predicate used to test range elements.
+ *  \return \c true, if all elements satisfy the predicate; \c false, otherwise.
+ *
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *  \tparam Predicate must be a model of <a href="http://www.sgi.com/tech/stl/Predicate.html">Predicate</a>.
+ *
+ *  \code
+ *  #include <thrust/logical.h>
+ *  #include <thrust/functional.h>
+ *  ...
+ *  bool A[3] = {true, true, false};
+ *
+ *  thrust::all_of(A, A + 2, thrust::identity<bool>()); // returns true
+ *  thrust::all_of(A, A + 3, thrust::identity<bool>()); // returns false
+ *
+ *  // empty range
+ *  thrust::all_of(A, A, thrust::identity<bool>()); // returns false
+ *  
+ *  \endcode
+ *
+ *  \see any_of
+ *  \see none_of
+ *  \see transform_reduce
+ */
+template <typename InputIterator, typename Predicate>
+bool all_of(InputIterator first, InputIterator last, Predicate pred);
+
+
+/*! \p any_of determines whether any element in a range satifies a predicate.
+ *  Specifically, \p any_of returns \c true if <tt>pred(*i)</tt> is \c true
+ *  for any iterator \c i in the range <tt>[first, last)</tt> and 
+ *  \c false otherwise.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the sequence.
+ *  \param last  The end of the sequence.
+ *  \param pred A predicate used to test range elements.
+ *  \return \c true, if any element satisfies the predicate; \c false, otherwise.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *  \tparam Predicate must be a model of <a href="http://www.sgi.com/tech/stl/Predicate.html">Predicate</a>.
+ *
+ *  \code
+ *  #include <thrust/logical.h>
+ *  #include <thrust/functional.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  bool A[3] = {true, true, false};
+ *
+ *  thrust::any_of(thrust::host, A, A + 2, thrust::identity<bool>()); // returns true
+ *  thrust::any_of(thrust::host, A, A + 3, thrust::identity<bool>()); // returns true
+ *
+ *  thrust::any_of(thrust::host, A + 2, A + 3, thrust::identity<bool>()); // returns false
+ *
+ *  // empty range
+ *  thrust::any_of(thrust::host, A, A, thrust::identity<bool>()); // returns false
+ *  \endcode
+ *
+ *  \see all_of
+ *  \see none_of
+ *  \see transform_reduce
+ */
+template <typename DerivedPolicy, typename InputIterator, typename Predicate>
+bool any_of(const thrust::detail::execution_policy_base<DerivedPolicy> &exec, InputIterator first, InputIterator last, Predicate pred);
+   
+
+/*! \p any_of determines whether any element in a range satifies a predicate.
+ * Specifically, \p any_of returns \c true if <tt>pred(*i)</tt> is \c true
+ * for any iterator \c i in the range <tt>[first, last)</tt> and 
+ * \c false otherwise.
+ *
+ *  \param first The beginning of the sequence.
+ *  \param last  The end of the sequence.
+ *  \param pred A predicate used to test range elements.
+ *  \return \c true, if any element satisfies the predicate; \c false, otherwise.
+ *
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *  \tparam Predicate must be a model of <a href="http://www.sgi.com/tech/stl/Predicate.html">Predicate</a>.
+ *
+ *  \code
+ *  #include <thrust/logical.h>
+ *  #include <thrust/functional.h>
+ *  ...
+ *  bool A[3] = {true, true, false};
+ *
+ *  thrust::any_of(A, A + 2, thrust::identity<bool>()); // returns true
+ *  thrust::any_of(A, A + 3, thrust::identity<bool>()); // returns true
+ *
+ *  thrust::any_of(A + 2, A + 3, thrust::identity<bool>()); // returns false
+ *
+ *  // empty range
+ *  thrust::any_of(A, A, thrust::identity<bool>()); // returns false
+ *  \endcode
+ *
+ *  \see all_of
+ *  \see none_of
+ *  \see transform_reduce
+ */
+template <typename InputIterator, typename Predicate>
+bool any_of(InputIterator first, InputIterator last, Predicate pred);
+
+
+/*! \p none_of determines whether no element in a range satifies a predicate.
+ *  Specifically, \p none_of returns \c true if there is no iterator \c i in 
+ *  the range <tt>[first, last)</tt> such that <tt>pred(*i)</tt> is \c true,
+ *  and \c false otherwise.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the sequence.
+ *  \param last  The end of the sequence.
+ *  \param pred A predicate used to test range elements.
+ *  \return \c true, if no element satisfies the predicate; \c false, otherwise.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *  \tparam Predicate must be a model of <a href="http://www.sgi.com/tech/stl/Predicate.html">Predicate</a>.
+ *
+ *  \code
+ *  #include <thrust/logical.h>
+ *  #include <thrust/functional.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  bool A[3] = {true, true, false};
+ *
+ *  thrust::none_of(thrust::host, A, A + 2, thrust::identity<bool>()); // returns false
+ *  thrust::none_of(thrust::host, A, A + 3, thrust::identity<bool>()); // returns false
+ *
+ *  thrust::none_of(thrust::host, A + 2, A + 3, thrust::identity<bool>()); // returns true
+ *
+ *  // empty range
+ *  thrust::none_of(thrust::host, A, A, thrust::identity<bool>()); // returns true
+ *  \endcode
+ *
+ *  \see all_of
+ *  \see any_of
+ *  \see transform_reduce
+ */
+template <typename DerivedPolicy, typename InputIterator, typename Predicate>
+bool none_of(const thrust::detail::execution_policy_base<DerivedPolicy> &exec, InputIterator first, InputIterator last, Predicate pred);
+
+
+/*! \p none_of determines whether no element in a range satifies a predicate.
+ *  Specifically, \p none_of returns \c true if there is no iterator \c i in 
+ *  the range <tt>[first, last)</tt> such that <tt>pred(*i)</tt> is \c true,
+ *  and \c false otherwise.
+ *
+ *  \param first The beginning of the sequence.
+ *  \param last  The end of the sequence.
+ *  \param pred A predicate used to test range elements.
+ *  \return \c true, if no element satisfies the predicate; \c false, otherwise.
+ *
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *  \tparam Predicate must be a model of <a href="http://www.sgi.com/tech/stl/Predicate.html">Predicate</a>.
+ *
+ *  \code
+ *  #include <thrust/logical.h>
+ *  #include <thrust/functional.h>
+ *  ...
+ *  bool A[3] = {true, true, false};
+ *
+ *  thrust::none_of(A, A + 2, thrust::identity<bool>()); // returns false
+ *  thrust::none_of(A, A + 3, thrust::identity<bool>()); // returns false
+ *
+ *  thrust::none_of(A + 2, A + 3, thrust::identity<bool>()); // returns true
+ *
+ *  // empty range
+ *  thrust::none_of(A, A, thrust::identity<bool>()); // returns true
+ *  \endcode
+ *
+ *  \see all_of
+ *  \see any_of
+ *  \see transform_reduce
+ */
+template <typename InputIterator, typename Predicate>
+bool none_of(InputIterator first, InputIterator last, Predicate pred);
+
+
+/*! \} // end logical
+ *  \} // end reductions
+ */
+
+
+} // end namespace thrust
+
+#include <thrust/detail/logical.inl>
+
diff --git a/compat/thrust/memory.h b/compat/thrust/memory.h
new file mode 100644
index 0000000000..6362de4064
--- /dev/null
+++ b/compat/thrust/memory.h
@@ -0,0 +1,538 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+/*! \file thrust/memory.h
+ *  \brief Abstractions for Thrust's memory model.
+ */
+
+#include <thrust/detail/config.h>
+
+#include <thrust/detail/type_traits/pointer_traits.h>
+#include <thrust/detail/pointer.h>
+#include <thrust/detail/reference.h>
+#include <thrust/detail/raw_pointer_cast.h>
+#include <thrust/detail/raw_reference_cast.h>
+#include <thrust/detail/malloc_and_free.h>
+#include <thrust/detail/temporary_buffer.h>
+
+namespace thrust
+{
+
+
+/*! \addtogroup memory_management Memory Management
+ *  \addtogroup memory_management_classes Memory Management Classes
+ *  \ingroup memory_management
+ *  \{
+ */
+
+/*! \p pointer stores a pointer to an object allocated in memory. Like \p device_ptr, this
+ *  type ensures type safety when dispatching standard algorithms on ranges resident in memory.
+ *
+ *  \p pointer generalizes \p device_ptr by relaxing the backend system associated with the \p pointer.
+ *  Instead of the backend system specified by \p THRUST_DEFAULT_DEVICE_BACKEND, \p pointer's
+ *  system is given by its second template parameter, \p Tag. For the purpose of Thrust dispatch,
+ *  <tt>device_ptr<Element></tt> and <tt>pointer<Element,device_system_tag></tt> are considered equivalent.
+ *
+ *  The raw pointer encapsulated by a \p pointer may be obtained through its <tt>get</tt> member function
+ *  or the \p raw_pointer_cast free function.
+ *
+ *  \tparam Element specifies the type of the pointed-to object.
+ *
+ *  \tparam Tag specifies the system with which this \p pointer is associated. This may be any Thrust
+ *          backend system, or a user-defined tag.
+ *
+ *  \tparam Reference allows the client to specify the reference type returned upon derereference.
+ *          By default, this type is <tt>reference<Element,pointer></tt>.
+ *
+ *  \tparam Derived allows the client to specify the name of the derived type when \p pointer is used as
+ *          a base class. This is useful to ensure that arithmetic on values of the derived type return
+ *          values of the derived type as a result. By default, this type is <tt>pointer<Element,Tag,Reference></tt>.
+ *
+ *  \note \p pointer is not a smart pointer; it is the client's responsibility to deallocate memory
+ *        pointer to by \p pointer.
+ *
+ *  \see device_ptr
+ *  \see reference
+ *  \see raw_pointer_cast
+ */
+// define pointer for the purpose of Doxygenating it
+// it is actually defined elsewhere
+#if 0
+template<typename Element, typename Tag, typename Reference = thrust::use_default, typename Derived = thrust::use_default>
+  class pointer
+{
+  public:
+    /*! The type of the raw pointer
+     */
+    typedef typename super_t::base_type raw_pointer;
+    
+    /*! \p pointer's default constructor initializes its encapsulated pointer to \c 0
+     */
+    __host__ __device__
+    pointer();
+
+    /*! This constructor allows construction of a <tt>pointer<const T, ...></tt> from a <tt>T*</tt>.
+     *
+     *  \param ptr A raw pointer to copy from, presumed to point to a location in \p Tag's memory.
+     *  \tparam OtherElement \p OtherElement shall be convertible to \p Element.
+     */
+    template<typename OtherElement>
+    __host__ __device__
+    explicit pointer(OtherElement *ptr);
+
+    /*! This contructor allows initialization from another pointer-like object.
+     *
+     *  \param other The \p OtherPointer to copy.
+     *
+     *  \tparam OtherPointer The tag associated with \p OtherPointer shall be convertible to \p Tag,
+     *                       and its element type shall be convertible to \p Element.
+     */
+    template<typename OtherPointer>
+    __host__ __device__
+    pointer(const OtherPointer &other,
+            typename thrust::detail::enable_if_pointer_is_convertible<
+              OtherPointer,
+              pointer<Element,Tag,Reference,Derived>
+            >::type * = 0);
+
+    /*! Assignment operator allows assigning from another pointer-like object with related type.
+     *
+     *  \param other The other pointer-like object to assign from.
+     *  \return <tt>*this</tt>
+     *
+     *  \tparam OtherPointer The tag associated with \p OtherPointer shall be convertible to \p Tag,
+     *                       and its element type shall be convertible to \p Element.
+     */
+    template<typename OtherPointer>
+    __host__ __device__
+    typename thrust::detail::enable_if_pointer_is_convertible<
+      OtherPointer,
+      pointer,
+      derived_type &
+    >::type
+    operator=(const OtherPointer &other);
+
+    /*! \p get returns this \p pointer's encapsulated raw pointer.
+     *  \return This \p pointer's raw pointer.
+     */
+    __host__ __device__
+    Element *get() const;
+};
+#endif
+
+/*! \p reference is a wrapped reference to an object stored in memory. \p reference generalizes
+ *  \p device_reference by relaxing the type of pointer associated with the object. \p reference
+ *  is the type of the result of dereferencing a tagged pointer-like object such as \p pointer, and
+ *  intermediates operations on objects existing in a remote memory.
+ *
+ *  \tparam Element specifies the type of the referent object.
+ *  \tparam Pointer specifies the type of the result of taking the address of \p reference.
+ *  \tparam Derived allows the client to specify the name of the derived type when \p reference is used as
+ *          a base class. This is useful to ensure that assignment to objects of the derived type return
+ *          values of the derived type as a result. By default, this type is <tt>reference<Element,Pointer></tt>.
+ */
+// define pointer for the purpose of Doxygenating it
+// it is actually defined elsewhere
+#if 0
+template<typename Element, typename Pointer, typename Derived = thrust::use_default>
+  class reference
+{
+  public:
+    /*! The type of this \p reference's wrapped pointers.
+     */
+    typedef Pointer                                              pointer;
+
+    /*! The \p value_type of this \p reference.
+     */
+    typedef typename thrust::detail::remove_const<Element>::type value_type;
+
+    /*! This copy constructor initializes this \p reference
+     *  to refer to an object pointed to by the given \p pointer. After
+     *  this \p reference is constructed, it shall refer to the
+     *  object pointed to by \p ptr.
+     *
+     *  \param ptr A \p pointer to copy from.
+     */
+    __host__ __device__
+    explicit reference(const pointer &ptr);
+
+    /*! This copy constructor accepts a const reference to another
+     *  \p reference of related type. After this \p reference is constructed,
+     *  it shall refer to the same object as \p other.
+     *  
+     *  \param other A \p reference to copy from.
+     *  \tparam OtherElement the element type of the other \p reference.
+     *  \tparam OtherPointer the pointer type of the other \p reference.
+     *  \tparam OtherDerived the derived type of the other \p reference.
+     *
+     *  \note This constructor is templated primarily to allow initialization of 
+     *  <tt>reference<const T,...></tt> from <tt>reference<T,...></tt>.
+     */
+    template<typename OtherElement, typename OtherPointer, typename OtherDerived>
+    __host__ __device__
+    reference(const reference<OtherElement,OtherPointer,OtherDerived> &other,
+              typename thrust::detail::enable_if_convertible<
+                typename reference<OtherElement,OtherPointer,OtherDerived>::pointer,
+                pointer
+              >::type * = 0);
+
+    /*! Copy assignment operator copy assigns from another \p reference.
+     *
+     *  \param other The other \p reference to assign from.
+     *  \return <tt>static_cast<derived_type&>(*this)</tt>
+     */
+    __host__ __device__
+    derived_type &operator=(const reference &other);
+
+    /*! Assignment operator copy assigns from another \p reference of related type.
+     *
+     *  \param other The other \p reference to assign from.
+     *  \return <tt>static_cast<derived_type&>(*this)</tt>
+     *
+     *  \tparam OtherElement the element type of the other \p reference.
+     *  \tparam OtherPointer the pointer type of the other \p reference.
+     *  \tparam OtherDerived the derived type of the other \p reference.
+     */
+    template<typename OtherElement, typename OtherPointer, typename OtherDerived>
+    __host__ __device__
+    derived_type &operator=(const reference<OtherElement,OtherPointer,OtherDerived> &other);
+
+    /*! Assignment operator assigns from a \p value_type.
+     *
+     *  \param x The \p value_type to assign from.
+     *  \return <tt>static_cast<derived_type&>(*this)</tt>.
+     */
+    __host__ __device__
+    derived_type &operator=(const value_type &x);
+
+    /*! Address-of operator returns a \p pointer pointing to the object
+     *  referenced by this \p reference. It does not return the address of this
+     *  \p reference.
+     *
+     *  \return A \p pointer pointing to the referenct object.
+     */
+    __host__ __device__
+    pointer operator&() const;
+
+    /*! Conversion operator converts this \p reference to \p value_type by
+     *  returning a copy of the referent object.
+     *  
+     *  \return A copy of the referent object.
+     */
+    __host__ __device__
+    operator value_type () const;
+
+    /*! Swaps the value of the referent object with another.
+     *
+     *  \param other The other \p reference with which to swap.
+     *  \note The argument is of type \p derived_type rather than \p reference.
+     */
+    __host__ __device__
+    void swap(derived_type &other);
+
+    /*! Prefix increment operator increments the referent object.
+     *
+     *  \return <tt>static_Cast<derived_type&>(*this)</tt>.
+     *
+     *  \note Documentation for other arithmetic operators omitted for brevity.
+     */
+    derived_type &operator++();
+};
+#endif
+
+/*! \}
+ */
+
+/*!
+ *  \addtogroup memory_management_functions Memory Management Functions
+ *  \ingroup memory_management
+ *  \{
+ */
+
+
+/*! \addtogroup allocation_functions
+ *  \{
+ */
+
+
+/*! This version of \p malloc allocates untyped uninitialized storage associated with a given system.
+ *
+ *  \param system The Thrust system with which to associate the storage.
+ *  \param n The number of bytes of storage to allocate.
+ *  \return If allocation succeeds, a pointer to the allocated storage; a null pointer otherwise.
+ *          The pointer must be deallocated with \p thrust::free.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *
+ *  \pre \p DerivedPolicy must be publically derived from <code>thrust::execution_policy<DerivedPolicy></code>.
+ *
+ *  The following code snippet demonstrates how to use \p malloc to allocate a range of memory
+ *  associated with Thrust's device system.
+ *
+ *  \code
+ *  #include <thrust/memory.h>
+ *  ...
+ *  // allocate some memory with thrust::malloc
+ *  const int N = 100;
+ *  thrust::device_system_tag device_sys;
+ *  thrust::pointer<void,thrust::device_space_tag> void_ptr = thrust::malloc(device_sys, N);
+ *
+ *  // manipulate memory
+ *  ...
+ *
+ *  // deallocate void_ptr with thrust::free
+ *  thrust::free(device_sys, void_ptr);
+ *  \endcode
+ *
+ *  \see free
+ *  \see device_malloc
+ */
+template<typename DerivedPolicy>
+pointer<void,DerivedPolicy> malloc(const thrust::detail::execution_policy_base<DerivedPolicy> &system, std::size_t n);
+
+
+/*! This version of \p malloc allocates typed uninitialized storage associated with a given system.
+ *
+ *  \param system The Thrust system with which to associate the storage.
+ *  \param n The number of elements of type \c T which the storage should accomodate.
+ *  \return If allocation succeeds, a pointer to an allocation large enough to accomodate \c n
+ *          elements of type \c T; a null pointer otherwise.
+ *          The pointer must be deallocated with \p thrust::free.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *
+ *  \pre \p DerivedPolicy must be publically derived from <code>thrust::execution_policy<DerivedPolicy></code>.
+ *
+ *  The following code snippet demonstrates how to use \p malloc to allocate a range of memory
+ *  to accomodate integers associated with Thrust's device system.
+ *
+ *  \code
+ *  #include <thrust/memory.h>
+ *  ...
+ *  // allocate storage for 100 ints with thrust::malloc
+ *  const int N = 100;
+ *  thrust::device_system_tag device_sys;
+ *  thrust::pointer<int,thrust::device_system_tag> ptr = thrust::malloc<int>(device_sys, N);
+ *
+ *  // manipulate memory
+ *  ...
+ *
+ *  // deallocate ptr with thrust::free
+ *  thrust::free(device_sys, ptr);
+ *  \endcode
+ *
+ *  \see free
+ *  \see device_malloc
+ */
+template<typename T, typename DerivedPolicy>
+pointer<T,DerivedPolicy> malloc(const thrust::detail::execution_policy_base<DerivedPolicy> &system, std::size_t n);
+
+
+/*! \p get_temporary_buffer returns a pointer to storage associated with a given Thrust system sufficient to store up to
+ *  \p n objects of type \c T. If not enough storage is available to accomodate \p n objects, an implementation may return
+ *  a smaller buffer. The number of objects the returned buffer can accomodate is also returned.
+ *
+ *  Thrust uses \p get_temporary_buffer internally when allocating temporary storage required by algorithm implementations.
+ *
+ *  The storage allocated with \p get_temporary_buffer must be returned to the system with \p return_temporary_buffer.
+ *
+ *  \param system The Thrust system with which to associate the storage.
+ *  \param n The requested number of objects of type \c T the storage should accomodate.
+ *  \return A pair \c p such that <tt>p.first</tt> is a pointer to the allocated storage and <tt>p.second</tt> is the number of
+ *          contiguous objects of type \c T that the storage can accomodate. If no storage can be allocated, <tt>p.first</tt> if
+ *          no storage can be obtained. The storage must be returned to the system using \p return_temporary_buffer.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *
+ *  \pre \p DerivedPolicy must be publically derived from <code>thrust::execution_policy<DerivedPolicy></code>.
+ *
+ *  The following code snippet demonstrates how to use \p get_temporary_buffer to allocate a range of memory
+ *  to accomodate integers associated with Thrust's device system.
+ *
+ *  \code
+ *  #include <thrust/memory.h>
+ *  ...
+ *  // allocate storage for 100 ints with thrust::get_temporary_buffer
+ *  const int N = 100;
+ *
+ *  typedef thrust::pair<
+ *    thrust::pointer<int,thrust::device_system_tag>,
+ *    std::ptrdiff_t
+ *  > ptr_and_size_t;
+ *
+ *  thrust::device_system_tag device_sys;
+ *  ptr_and_size_t ptr_and_size = thrust::get_temporary_buffer<int>(device_sys, N);
+ *
+ *  // manipulate up to 100 ints
+ *  for(int i = 0; i < ptr_and_size.second; ++i)
+ *  {
+ *    *ptr_and_size.first = i;
+ *  }
+ *
+ *  // deallocate storage with thrust::return_temporary_buffer
+ *  thrust::return_temporary_buffer(device_sys, ptr_and_size.first);
+ *  \endcode
+ *
+ *  \see malloc
+ *  \see return_temporary_buffer
+ */
+template<typename T, typename DerivedPolicy>
+thrust::pair<thrust::pointer<T,DerivedPolicy>, typename thrust::pointer<T,DerivedPolicy>::difference_type>
+get_temporary_buffer(const thrust::detail::execution_policy_base<DerivedPolicy> &system, typename thrust::pointer<T,DerivedPolicy>::difference_type n);
+
+
+/*! \} allocation_functions
+ */
+
+
+/*! \addtogroup deallocation_functions
+ *  \{
+ */
+
+
+/*! \p free deallocates the storage previously allocated by \p thrust::malloc.
+ *
+ *  \param system The Thrust system with which the storage is associated.
+ *  \param ptr A pointer previously returned by \p thrust::malloc. If \p ptr is null, \p free
+ *         does nothing.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *
+ *  \pre \p ptr shall have been returned by a previous call to <tt>thrust::malloc(system, n)</tt> or <tt>thrust::malloc<T>(system, n)</tt> for some type \c T.
+ *
+ *  The following code snippet demonstrates how to use \p free to deallocate a range of memory
+ *  previously allocated with \p thrust::malloc.
+ *
+ *  \code
+ *  #include <thrust/memory.h>
+ *  ...
+ *  // allocate storage for 100 ints with thrust::malloc
+ *  const int N = 100;
+ *  thrust::device_system_tag device_sys;
+ *  thrust::pointer<int,thrust::device_system_tag> ptr = thrust::malloc<int>(device_sys, N);
+ *
+ *  // mainpulate memory
+ *  ...
+ *
+ *  // deallocate ptr with thrust::free
+ *  thrust::free(device_sys, ptr);
+ *  \endcode
+ */
+template<typename DerivedPolicy, typename Pointer>
+void free(const thrust::detail::execution_policy_base<DerivedPolicy> &system, Pointer ptr);
+
+
+/*! \p return_temporary_buffer deallocates storage associated with a given Thrust system previously allocated by \p get_temporary_buffer.
+ *
+ *  Thrust uses \p return_temporary_buffer internally when deallocating temporary storage required by algorithm implementations.
+ *
+ *  \param system The Thrust system with which the storage is associated.
+ *  \param p A pointer previously returned by \p thrust::get_temporary_buffer. If \p ptr is null, \p return_temporary_buffer does nothing.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *
+ *  \pre \p p shall have been previously allocated by \p thrust::get_temporary_buffer.
+ *
+ *  The following code snippet demonstrates how to use \p return_temporary_buffer to deallocate a range of memory
+ *  previously allocated by \p get_temporary_buffer.
+ *
+ *  \code
+ *  #include <thrust/memory.h>
+ *  ...
+ *  // allocate storage for 100 ints with thrust::get_temporary_buffer
+ *  const int N = 100;
+ *
+ *  typedef thrust::pair<
+ *    thrust::pointer<int,thrust::device_system_tag>,
+ *    std::ptrdiff_t
+ *  > ptr_and_size_t;
+ *
+ *  thrust::device_system_tag device_sys;
+ *  ptr_and_size_t ptr_and_size = thrust::get_temporary_buffer<int>(device_sys, N);
+ *
+ *  // manipulate up to 100 ints
+ *  for(int i = 0; i < ptr_and_size.second; ++i)
+ *  {
+ *    *ptr_and_size.first = i;
+ *  }
+ *
+ *  // deallocate storage with thrust::return_temporary_buffer
+ *  thrust::return_temporary_buffer(device_sys, ptr_and_size.first);
+ *  \endcode
+ *
+ *  \see free
+ *  \see get_temporary_buffer
+ */
+template<typename DerivedPolicy, typename Pointer>
+void return_temporary_buffer(const thrust::detail::execution_policy_base<DerivedPolicy> &system, Pointer p);
+
+
+/*! \} deallocation_functions
+ */
+
+
+/*! \p raw_pointer_cast creates a "raw" pointer from a pointer-like type,
+ *  simply returning the wrapped pointer, should it exist.
+ *
+ *  \param ptr The pointer of interest.
+ *  \return <tt>ptr.get()</tt>, if the expression is well formed; <tt>ptr</tt>, otherwise.
+ *  \see raw_reference_cast
+ */
+template<typename Pointer>
+__host__ __device__
+inline typename thrust::detail::pointer_traits<Pointer>::raw_pointer
+  raw_pointer_cast(const Pointer &ptr);
+
+
+/*! \p raw_reference_cast creates a "raw" reference from a wrapped reference type,
+ *  simply returning the underlying reference, should it exist.
+ *
+ *  If the argument is not a reference wrapper, the result is a reference to the argument.
+ *
+ *  \param ref The reference of interest.
+ *  \return <tt>*thrust::raw_pointer_cast(&ref)</tt>.
+ *  \note There are two versions of \p raw_reference_cast. One for <tt>const</tt> references,
+ *        and one for non-<tt>const</tt>.
+ *  \see raw_pointer_cast
+ */
+template<typename T>
+__host__ __device__
+inline typename detail::raw_reference<T>::type
+  raw_reference_cast(T &ref);
+
+
+/*! \p raw_reference_cast creates a "raw" reference from a wrapped reference type,
+ *  simply returning the underlying reference, should it exist.
+ *
+ *  If the argument is not a reference wrapper, the result is a reference to the argument.
+ *
+ *  \param ref The reference of interest.
+ *  \return <tt>*thrust::raw_pointer_cast(&ref)</tt>.
+ *  \note There are two versions of \p raw_reference_cast. One for <tt>const</tt> references,
+ *        and one for non-<tt>const</tt>.
+ *  \see raw_pointer_cast
+ */
+template<typename T>
+__host__ __device__
+inline typename detail::raw_reference<const T>::type
+  raw_reference_cast(const T &ref);
+
+
+/*! \}
+ */
+
+} // end thrust
+
diff --git a/compat/thrust/merge.h b/compat/thrust/merge.h
new file mode 100644
index 0000000000..e5fa7b47ac
--- /dev/null
+++ b/compat/thrust/merge.h
@@ -0,0 +1,676 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+/*! \file merge.h
+ *  \brief Merging sorted ranges
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/execution_policy.h>
+#include <thrust/pair.h>
+
+namespace thrust
+{
+
+
+/*! \addtogroup merging Merging
+ *  \ingroup algorithms
+ *  \{
+ */
+
+
+/*! \p merge combines two sorted ranges <tt>[first1, last1)</tt> and <tt>[first2, last2)</tt>
+ *  into a single sorted range. That is, it copies from <tt>[first1, last1)</tt> and
+ *  <tt>[first2, last2)</tt> into <tt>[result, result + (last1 - first1) + (last2 - first2))</tt>
+ *  such that the resulting range is in ascending order. \p merge is stable, meaning both that the
+ *  relative order of elements within each input range is preserved, and that for equivalent elements
+ *  in both input ranges the element from the first range precedes the element from the second. The
+ *  return value is <tt>result + (last1 - first1) + (last2 - first2)</tt>.
+ *
+ *  This version of \p merge compares elements using \c operator<.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first1 The beginning of the first input range.
+ *  \param last1 The end of the first input range.
+ *  \param first2 The beginning of the second input range.
+ *  \param last2 The end of the second input range.
+ *  \param result The beginning of the merged output.
+ *  \return The end of the output range.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator1 and \p InputIterator2 have the same \c value_type,
+ *          \p InputIterator1's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a>,
+ *          the ordering on \p InputIterator1's \c value_type is a strict weak ordering, as defined in the <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a> requirements,
+ *          and \p InputIterator1's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator2 and \p InputIterator1 have the same \c value_type,
+ *          \p InputIterator2's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a>,
+ *          the ordering on \p InputIterator2's \c value_type is a strict weak ordering, as defined in the <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a> requirements,
+ *          and \p InputIterator2's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *
+ *  \pre The ranges <tt>[first1, last1)</tt> and <tt>[first2, last2)</tt> shall be sorted with respect to <tt>operator<</tt>.
+ *  \pre The resulting range shall not overlap with either input range.
+ *
+ *  The following code snippet demonstrates how to use
+ *  \p merge to compute the merger of two sorted sets of integers using the \p thrust::host execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/merge.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  int A1[6] = {1, 3, 5, 7, 9, 11};
+ *  int A2[7] = {1, 1, 2, 3, 5,  8, 13};
+ *
+ *  int result[13];
+ *
+ *  int *result_end =
+ *    thrust::merge(thrust::host,
+ *                  A1, A1 + 6,
+ *                  A2, A2 + 7,
+ *                  result);
+ *  // result = {1, 1, 1, 2, 3, 3, 5, 5, 7, 8, 9, 11, 13}
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/merge.html
+ *  \see \p set_union
+ *  \see \p sort
+ *  \see \p is_sorted
+ */
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator>
+  OutputIterator merge(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                       InputIterator1 first1,
+                       InputIterator1 last1,
+                       InputIterator2 first2,
+                       InputIterator2 last2,
+                       OutputIterator result);
+
+
+/*! \p merge combines two sorted ranges <tt>[first1, last1)</tt> and <tt>[first2, last2)</tt>
+ *  into a single sorted range. That is, it copies from <tt>[first1, last1)</tt> and
+ *  <tt>[first2, last2)</tt> into <tt>[result, result + (last1 - first1) + (last2 - first2))</tt>
+ *  such that the resulting range is in ascending order. \p merge is stable, meaning both that the
+ *  relative order of elements within each input range is preserved, and that for equivalent elements
+ *  in both input ranges the element from the first range precedes the element from the second. The
+ *  return value is <tt>result + (last1 - first1) + (last2 - first2)</tt>.
+ *
+ *  This version of \p merge compares elements using \c operator<.
+ *
+ *  \param first1 The beginning of the first input range.
+ *  \param last1 The end of the first input range.
+ *  \param first2 The beginning of the second input range.
+ *  \param last2 The end of the second input range.
+ *  \param result The beginning of the merged output.
+ *  \return The end of the output range.
+ *
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator1 and \p InputIterator2 have the same \c value_type,
+ *          \p InputIterator1's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a>,
+ *          the ordering on \p InputIterator1's \c value_type is a strict weak ordering, as defined in the <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a> requirements,
+ *          and \p InputIterator1's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator2 and \p InputIterator1 have the same \c value_type,
+ *          \p InputIterator2's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a>,
+ *          the ordering on \p InputIterator2's \c value_type is a strict weak ordering, as defined in the <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a> requirements,
+ *          and \p InputIterator2's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *
+ *  \pre The ranges <tt>[first1, last1)</tt> and <tt>[first2, last2)</tt> shall be sorted with respect to <tt>operator<</tt>.
+ *  \pre The resulting range shall not overlap with either input range.
+ *
+ *  The following code snippet demonstrates how to use
+ *  \p merge to compute the merger of two sorted sets of integers.
+ *
+ *  \code
+ *  #include <thrust/merge.h>
+ *  ...
+ *  int A1[6] = {1, 3, 5, 7, 9, 11};
+ *  int A2[7] = {1, 1, 2, 3, 5,  8, 13};
+ *
+ *  int result[13];
+ *
+ *  int *result_end = thrust::merge(A1, A1 + 6, A2, A2 + 7, result);
+ *  // result = {1, 1, 1, 2, 3, 3, 5, 5, 7, 8, 9, 11, 13}
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/merge.html
+ *  \see \p set_union
+ *  \see \p sort
+ *  \see \p is_sorted
+ */
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator>
+  OutputIterator merge(InputIterator1 first1,
+                       InputIterator1 last1,
+                       InputIterator2 first2,
+                       InputIterator2 last2,
+                       OutputIterator result);
+
+
+/*! \p merge combines two sorted ranges <tt>[first1, last1)</tt> and <tt>[first2, last2)</tt>
+ *  into a single sorted range. That is, it copies from <tt>[first1, last1)</tt> and
+ *  <tt>[first2, last2)</tt> into <tt>[result, result + (last1 - first1) + (last2 - first2))</tt>
+ *  such that the resulting range is in ascending order. \p merge is stable, meaning both that the
+ *  relative order of elements within each input range is preserved, and that for equivalent elements
+ *  in both input ranges the element from the first range precedes the element from the second. The
+ *  return value is <tt>result + (last1 - first1) + (last2 - first2)</tt>.
+ *
+ *  This version of \p merge compares elements using a function object \p comp.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first1 The beginning of the first input range.
+ *  \param last1 The end of the first input range.
+ *  \param first2 The beginning of the second input range.
+ *  \param last2 The end of the second input range.
+ *  \param result The beginning of the merged output.
+ *  \param comp Comparison operator.
+ *  \return The end of the output range.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator1's \c value_type is convertable to \p StrictWeakCompare's \c first_argument_type.
+ *          and \p InputIterator1's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator2's \c value_type is convertable to \p StrictWeakCompare's \c second_argument_type.
+ *          and \p InputIterator2's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam StrictWeakCompare is a model of <a href="http://www.sgi.com/tech/stl/StrictWeakOrdering.html">Strict Weak Ordering</a>.
+ *
+ *  \pre The ranges <tt>[first1, last1)</tt> and <tt>[first2, last2)</tt> shall be sorted with respect to \p comp.
+ *  \pre The resulting range shall not overlap with either input range.
+ *
+ *  The following code snippet demonstrates how to use
+ *  \p merge to compute the merger of two sets of integers sorted in
+ *  descending order using the \p thrust::host execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/merge.h>
+ *  #include <thrust/functional.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  int A1[6] = {11, 9, 7, 5, 3, 1};
+ *  int A2[7] = {13, 8, 5, 3, 2, 1, 1};
+ *
+ *  int result[13];
+ *
+ *  int *result_end = thrust::merge(thrust::host,
+ *                                  A1, A1 + 6,
+ *                                  A2, A2 + 7,
+ *                                  result,
+ *                                  thrust::greater<int>());
+ *  // result = {13, 11, 9, 8, 7, 5, 5, 3, 3, 2, 1, 1, 1}
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/merge.html
+ *  \see \p sort
+ *  \see \p is_sorted
+ */
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename StrictWeakCompare>
+  OutputIterator merge(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                       InputIterator1 first1,
+                       InputIterator1 last1,
+                       InputIterator2 first2,
+                       InputIterator2 last2,
+                       OutputIterator result,
+                       StrictWeakCompare comp);
+
+
+/*! \p merge combines two sorted ranges <tt>[first1, last1)</tt> and <tt>[first2, last2)</tt>
+ *  into a single sorted range. That is, it copies from <tt>[first1, last1)</tt> and
+ *  <tt>[first2, last2)</tt> into <tt>[result, result + (last1 - first1) + (last2 - first2))</tt>
+ *  such that the resulting range is in ascending order. \p merge is stable, meaning both that the
+ *  relative order of elements within each input range is preserved, and that for equivalent elements
+ *  in both input ranges the element from the first range precedes the element from the second. The
+ *  return value is <tt>result + (last1 - first1) + (last2 - first2)</tt>.
+ *
+ *  This version of \p merge compares elements using a function object \p comp.
+ *
+ *  \param first1 The beginning of the first input range.
+ *  \param last1 The end of the first input range.
+ *  \param first2 The beginning of the second input range.
+ *  \param last2 The end of the second input range.
+ *  \param result The beginning of the merged output.
+ *  \param comp Comparison operator.
+ *  \return The end of the output range.
+ *
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator1's \c value_type is convertable to \p StrictWeakCompare's \c first_argument_type.
+ *          and \p InputIterator1's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator2's \c value_type is convertable to \p StrictWeakCompare's \c second_argument_type.
+ *          and \p InputIterator2's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam StrictWeakCompare is a model of <a href="http://www.sgi.com/tech/stl/StrictWeakOrdering.html">Strict Weak Ordering</a>.
+ *
+ *  \pre The ranges <tt>[first1, last1)</tt> and <tt>[first2, last2)</tt> shall be sorted with respect to \p comp.
+ *  \pre The resulting range shall not overlap with either input range.
+ *
+ *  The following code snippet demonstrates how to use
+ *  \p merge to compute the merger of two sets of integers sorted in
+ *  descending order.
+ *
+ *  \code
+ *  #include <thrust/merge.h>
+ *  #include <thrust/functional.h>
+ *  ...
+ *  int A1[6] = {11, 9, 7, 5, 3, 1};
+ *  int A2[7] = {13, 8, 5, 3, 2, 1, 1};
+ *
+ *  int result[13];
+ *
+ *  int *result_end = thrust::merge(A1, A1 + 6, A2, A2 + 7, result, thrust::greater<int>());
+ *  // result = {13, 11, 9, 8, 7, 5, 5, 3, 3, 2, 1, 1, 1}
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/merge.html
+ *  \see \p sort
+ *  \see \p is_sorted
+ */
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename StrictWeakCompare>
+  OutputIterator merge(InputIterator1 first1,
+                       InputIterator1 last1,
+                       InputIterator2 first2,
+                       InputIterator2 last2,
+                       OutputIterator result,
+                       StrictWeakCompare comp);
+
+
+/*! \p merge_by_key performs a key-value merge. That is, \p merge_by_key copies elements from
+ *  <tt>[keys_first1, keys_last1)</tt> and <tt>[keys_first2, keys_last2)</tt> into a single range,
+ *  <tt>[keys_result, keys_result + (keys_last1 - keys_first1) + (keys_last2 - keys_first2))</tt> such that
+ *  the resulting range is in ascending key order.
+ *
+ *  At the same time, \p merge_by_key copies elements from the two associated ranges <tt>[values_first1 + (keys_last1 - keys_first1))</tt>
+ *  and <tt>[values_first2 + (keys_last2 - keys_first2))</tt> into a single range,
+ *  <tt>[values_result, values_result + (keys_last1 - keys_first1) + (keys_last2 - keys_first2))</tt> such that
+ *  the resulting range is in ascending order implied by each input element's associated key.
+ *
+ *  \p merge_by_key is stable, meaning both that the relative order of elements within each input range is
+ *  preserved, and that for equivalent elements in all input key ranges the element from the first range
+ *  precedes the element from the second.
+ *
+ *  The return value is is <tt>(keys_result + (keys_last1 - keys_first1) + (keys_last2 - keys_first2))</tt>
+ *  and <tt>(values_result + (keys_last1 - keys_first1) + (keys_last2 - keys_first2))</tt>.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param keys_first1 The beginning of the first input range of keys.
+ *  \param keys_last1 The end of the first input range of keys.
+ *  \param keys_first2 The beginning of the second input range of keys.
+ *  \param keys_last2 The end of the second input range of keys.
+ *  \param values_first1 The beginning of the first input range of values.
+ *  \param values_first2 The beginning of the first input range of values.
+ *  \param keys_result The beginning of the merged output range of keys.
+ *  \param values_result The beginning of the merged output range of values.
+ *  \return A \p pair \c p such that <tt>p.first</tt> is the end of the output range of keys,
+ *          and such that <tt>p.second</tt> is the end of the output range of values.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator1 and \p InputIterator2 have the same \c value_type,
+ *          \p InputIterator1's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a>,
+ *          the ordering on \p InputIterator1's \c value_type is a strict weak ordering, as defined in the <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a> requirements,
+ *          and \p InputIterator1's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator2 and \p InputIterator1 have the same \c value_type,
+ *          \p InputIterator2's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a>,
+ *          the ordering on \p InputIterator2's \c value_type is a strict weak ordering, as defined in the <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a> requirements,
+ *          and \p InputIterator2's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam InputIterator3 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator3's \c value_type is convertible to a type in \p OutputIterator2's set of \c value_types.
+ *  \tparam InputIterator4 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator4's \c value_type is convertible to a type in \p OutputIterator2's set of \c value_types.
+ *  \tparam OutputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam OutputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *
+ *  \pre The ranges <tt>[keys_first1, keys_last1)</tt> and <tt>[keys_first2, keys_last2)</tt> shall be sorted with respect to <tt>operator<</tt>.
+ *  \pre The resulting ranges shall not overlap with any input range.
+ *
+ *  The following code snippet demonstrates how to use
+ *  \p merge_by_key to compute the merger of two sets of integers sorted in
+ *  ascending order using the \p thrust::host execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/merge.h>
+ *  #include <thrust/functional.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  int A_keys[6] = {1, 3, 5, 7, 9, 11};
+ *  int A_vals[6] = {0, 0, 0, 0, 0, 0};
+ *
+ *  int B_keys[7] = {1, 1, 2, 3, 5, 8, 13};
+ *  int B_vals[7] = {1, 1, 1, 1, 1, 1, 1};
+ *
+ *  int keys_result[13];
+ *  int vals_result[13];
+ *
+ *  thrust::pair<int*,int*> end =
+ *    thrust::merge_by_key(thrust::host,
+ *                         A_keys, A_keys + 6,
+ *                         B_keys, B_keys + 7,
+ *                         A_vals, B_vals,
+ *                         keys_result, vals_result);
+ *
+ *  // keys_result = {1, 1, 1, 2, 3, 3, 5, 5, 7, 8, 9, 11, 13}
+ *  // vals_result = {0, 1, 1, 1, 0, 1, 0, 1, 0, 1, 0,  0,  1}
+ *  \endcode
+ *
+ *  \see merge
+ *  \see \p sort_by_key
+ *  \see \p is_sorted
+ */
+template<typename DerivedPolicy, typename InputIterator1, typename InputIterator2, typename InputIterator3, typename InputIterator4, typename OutputIterator1, typename OutputIterator2>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    merge_by_key(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                 InputIterator1 keys_first1, InputIterator1 keys_last1,
+                 InputIterator2 keys_first2, InputIterator2 keys_last2,
+                 InputIterator3 values_first1, InputIterator4 values_first2,
+                 OutputIterator1 keys_result,
+                 OutputIterator2 values_result);
+
+
+/*! \p merge_by_key performs a key-value merge. That is, \p merge_by_key copies elements from
+ *  <tt>[keys_first1, keys_last1)</tt> and <tt>[keys_first2, keys_last2)</tt> into a single range,
+ *  <tt>[keys_result, keys_result + (keys_last1 - keys_first1) + (keys_last2 - keys_first2))</tt> such that
+ *  the resulting range is in ascending key order.
+ *
+ *  At the same time, \p merge_by_key copies elements from the two associated ranges <tt>[values_first1 + (keys_last1 - keys_first1))</tt>
+ *  and <tt>[values_first2 + (keys_last2 - keys_first2))</tt> into a single range,
+ *  <tt>[values_result, values_result + (keys_last1 - keys_first1) + (keys_last2 - keys_first2))</tt> such that
+ *  the resulting range is in ascending order implied by each input element's associated key.
+ *
+ *  \p merge_by_key is stable, meaning both that the relative order of elements within each input range is
+ *  preserved, and that for equivalent elements in all input key ranges the element from the first range
+ *  precedes the element from the second.
+ *
+ *  The return value is is <tt>(keys_result + (keys_last1 - keys_first1) + (keys_last2 - keys_first2))</tt>
+ *  and <tt>(values_result + (keys_last1 - keys_first1) + (keys_last2 - keys_first2))</tt>.
+ *
+ *  \param keys_first1 The beginning of the first input range of keys.
+ *  \param keys_last1 The end of the first input range of keys.
+ *  \param keys_first2 The beginning of the second input range of keys.
+ *  \param keys_last2 The end of the second input range of keys.
+ *  \param values_first1 The beginning of the first input range of values.
+ *  \param values_first2 The beginning of the first input range of values.
+ *  \param keys_result The beginning of the merged output range of keys.
+ *  \param values_result The beginning of the merged output range of values.
+ *  \return A \p pair \c p such that <tt>p.first</tt> is the end of the output range of keys,
+ *          and such that <tt>p.second</tt> is the end of the output range of values.
+ *
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator1 and \p InputIterator2 have the same \c value_type,
+ *          \p InputIterator1's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a>,
+ *          the ordering on \p InputIterator1's \c value_type is a strict weak ordering, as defined in the <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a> requirements,
+ *          and \p InputIterator1's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator2 and \p InputIterator1 have the same \c value_type,
+ *          \p InputIterator2's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a>,
+ *          the ordering on \p InputIterator2's \c value_type is a strict weak ordering, as defined in the <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a> requirements,
+ *          and \p InputIterator2's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam InputIterator3 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator3's \c value_type is convertible to a type in \p OutputIterator2's set of \c value_types.
+ *  \tparam InputIterator4 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator4's \c value_type is convertible to a type in \p OutputIterator2's set of \c value_types.
+ *  \tparam OutputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam OutputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *
+ *  \pre The ranges <tt>[keys_first1, keys_last1)</tt> and <tt>[keys_first2, keys_last2)</tt> shall be sorted with respect to <tt>operator<</tt>.
+ *  \pre The resulting ranges shall not overlap with any input range.
+ *
+ *  The following code snippet demonstrates how to use
+ *  \p merge_by_key to compute the merger of two sets of integers sorted in
+ *  ascending order.
+ *
+ *  \code
+ *  #include <thrust/merge.h>
+ *  #include <thrust/functional.h>
+ *  ...
+ *  int A_keys[6] = {1, 3, 5, 7, 9, 11};
+ *  int A_vals[6] = {0, 0, 0, 0, 0, 0};
+ *
+ *  int B_keys[7] = {1, 1, 2, 3, 5, 8, 13};
+ *  int B_vals[7] = {1, 1, 1, 1, 1, 1, 1};
+ *
+ *  int keys_result[13];
+ *  int vals_result[13];
+ *
+ *  thrust::pair<int*,int*> end = thrust::merge_by_key(A_keys, A_keys + 6, B_keys, B_keys + 7, A_vals, B_vals, keys_result, vals_result);
+ *
+ *  // keys_result = {1, 1, 1, 2, 3, 3, 5, 5, 7, 8, 9, 11, 13}
+ *  // vals_result = {0, 1, 1, 1, 0, 1, 0, 1, 0, 1, 0,  0,  1}
+ *  \endcode
+ *
+ *  \see merge
+ *  \see \p sort_by_key
+ *  \see \p is_sorted
+ */
+template<typename InputIterator1, typename InputIterator2, typename InputIterator3, typename InputIterator4, typename OutputIterator1, typename OutputIterator2>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    merge_by_key(InputIterator1 keys_first1, InputIterator1 keys_last1,
+                 InputIterator2 keys_first2, InputIterator2 keys_last2,
+                 InputIterator3 values_first1, InputIterator4 values_first2,
+                 OutputIterator1 keys_result,
+                 OutputIterator2 values_result);
+
+
+/*! \p merge_by_key performs a key-value merge. That is, \p merge_by_key copies elements from
+ *  <tt>[keys_first1, keys_last1)</tt> and <tt>[keys_first2, keys_last2)</tt> into a single range,
+ *  <tt>[keys_result, keys_result + (keys_last1 - keys_first1) + (keys_last2 - keys_first2))</tt> such that
+ *  the resulting range is in ascending key order.
+ *
+ *  At the same time, \p merge_by_key copies elements from the two associated ranges <tt>[values_first1 + (keys_last1 - keys_first1))</tt>
+ *  and <tt>[values_first2 + (keys_last2 - keys_first2))</tt> into a single range,
+ *  <tt>[values_result, values_result + (keys_last1 - keys_first1) + (keys_last2 - keys_first2))</tt> such that
+ *  the resulting range is in ascending order implied by each input element's associated key.
+ *
+ *  \p merge_by_key is stable, meaning both that the relative order of elements within each input range is
+ *  preserved, and that for equivalent elements in all input key ranges the element from the first range
+ *  precedes the element from the second.
+ *
+ *  The return value is is <tt>(keys_result + (keys_last1 - keys_first1) + (keys_last2 - keys_first2))</tt>
+ *  and <tt>(values_result + (keys_last1 - keys_first1) + (keys_last2 - keys_first2))</tt>.
+ *
+ *  This version of \p merge_by_key compares key elements using a function object \p comp.
+ *
+ *  The algorithm's execution is parallelized using \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param keys_first1 The beginning of the first input range of keys.
+ *  \param keys_last1 The end of the first input range of keys.
+ *  \param keys_first2 The beginning of the second input range of keys.
+ *  \param keys_last2 The end of the second input range of keys.
+ *  \param values_first1 The beginning of the first input range of values.
+ *  \param values_first2 The beginning of the first input range of values.
+ *  \param keys_result The beginning of the merged output range of keys.
+ *  \param values_result The beginning of the merged output range of values.
+ *  \param comp Comparison operator.
+ *  \return A \p pair \c p such that <tt>p.first</tt> is the end of the output range of keys,
+ *          and such that <tt>p.second</tt> is the end of the output range of values.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator1's \c value_type is convertable to \p StrictWeakCompare's \c first_argument_type.
+ *          and \p InputIterator1's \c value_type is convertable to a type in \p OutputIterator1's set of \c value_types.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator2's \c value_type is convertable to \p StrictWeakCompare's \c second_argument_type.
+ *          and \p InputIterator2's \c value_type is convertable to a type in \p OutputIterator1's set of \c value_types.
+ *  \tparam InputIterator3 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator3's \c value_type is convertible to a type in \p OutputIterator2's set of \c value_types.
+ *  \tparam InputIterator4 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator4's \c value_type is convertible to a type in \p OutputIterator2's set of \c value_types.
+ *  \tparam OutputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam OutputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam StrictWeakCompare is a model of <a href="http://www.sgi.com/tech/stl/StrictWeakOrdering.html">Strict Weak Ordering</a>.
+ *
+ *  \pre The ranges <tt>[keys_first1, keys_last1)</tt> and <tt>[keys_first2, keys_last2)</tt> shall be sorted with respect to \p comp.
+ *  \pre The resulting ranges shall not overlap with any input range.
+ *
+ *  The following code snippet demonstrates how to use
+ *  \p merge_by_key to compute the merger of two sets of integers sorted in
+ *  descending order using the \p thrust::host execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/merge.h>
+ *  #include <thrust/functional.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  int A_keys[6] = {11, 9, 7, 5, 3, 1};
+ *  int A_vals[6] = { 0, 0, 0, 0, 0, 0};
+ *
+ *  int B_keys[7] = {13, 8, 5, 3, 2, 1, 1};
+ *  int B_vals[7] = { 1, 1, 1, 1, 1, 1, 1};
+ *
+ *  int keys_result[13];
+ *  int vals_result[13];
+ *
+ *  thrust::pair<int*,int*> end =
+ *    thrust::merge_by_key(thrust::host,
+ *                         A_keys, A_keys + 6,
+ *                         B_keys, B_keys + 7,
+ *                         A_vals, B_vals,
+ *                         keys_result, vals_result,
+ *                         thrust::greater<int>());
+ *
+ *  // keys_result = {13, 11, 9, 8, 7, 5, 5, 3, 3, 2, 1, 1, 1}
+ *  // vals_result = { 1,  0, 0, 1, 0, 0, 1, 0, 1, 1, 0, 1, 1}
+ *  \endcode
+ *
+ *  \see merge
+ *  \see \p sort_by_key
+ *  \see \p is_sorted
+ */
+template<typename DerivedPolicy, typename InputIterator1, typename InputIterator2, typename InputIterator3, typename InputIterator4, typename OutputIterator1, typename OutputIterator2, typename Compare>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    merge_by_key(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                 InputIterator1 keys_first1, InputIterator1 keys_last1,
+                 InputIterator2 keys_first2, InputIterator2 keys_last2,
+                 InputIterator3 values_first1, InputIterator4 values_first2,
+                 OutputIterator1 keys_result,
+                 OutputIterator2 values_result,
+                 Compare comp);
+
+
+/*! \p merge_by_key performs a key-value merge. That is, \p merge_by_key copies elements from
+ *  <tt>[keys_first1, keys_last1)</tt> and <tt>[keys_first2, keys_last2)</tt> into a single range,
+ *  <tt>[keys_result, keys_result + (keys_last1 - keys_first1) + (keys_last2 - keys_first2))</tt> such that
+ *  the resulting range is in ascending key order.
+ *
+ *  At the same time, \p merge_by_key copies elements from the two associated ranges <tt>[values_first1 + (keys_last1 - keys_first1))</tt>
+ *  and <tt>[values_first2 + (keys_last2 - keys_first2))</tt> into a single range,
+ *  <tt>[values_result, values_result + (keys_last1 - keys_first1) + (keys_last2 - keys_first2))</tt> such that
+ *  the resulting range is in ascending order implied by each input element's associated key.
+ *
+ *  \p merge_by_key is stable, meaning both that the relative order of elements within each input range is
+ *  preserved, and that for equivalent elements in all input key ranges the element from the first range
+ *  precedes the element from the second.
+ *
+ *  The return value is is <tt>(keys_result + (keys_last1 - keys_first1) + (keys_last2 - keys_first2))</tt>
+ *  and <tt>(values_result + (keys_last1 - keys_first1) + (keys_last2 - keys_first2))</tt>.
+ *
+ *  This version of \p merge_by_key compares key elements using a function object \p comp.
+ *
+ *  \param keys_first1 The beginning of the first input range of keys.
+ *  \param keys_last1 The end of the first input range of keys.
+ *  \param keys_first2 The beginning of the second input range of keys.
+ *  \param keys_last2 The end of the second input range of keys.
+ *  \param values_first1 The beginning of the first input range of values.
+ *  \param values_first2 The beginning of the first input range of values.
+ *  \param keys_result The beginning of the merged output range of keys.
+ *  \param values_result The beginning of the merged output range of values.
+ *  \param comp Comparison operator.
+ *  \return A \p pair \c p such that <tt>p.first</tt> is the end of the output range of keys,
+ *          and such that <tt>p.second</tt> is the end of the output range of values.
+ *
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator1's \c value_type is convertable to \p StrictWeakCompare's \c first_argument_type.
+ *          and \p InputIterator1's \c value_type is convertable to a type in \p OutputIterator1's set of \c value_types.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator2's \c value_type is convertable to \p StrictWeakCompare's \c second_argument_type.
+ *          and \p InputIterator2's \c value_type is convertable to a type in \p OutputIterator1's set of \c value_types.
+ *  \tparam InputIterator3 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator3's \c value_type is convertible to a type in \p OutputIterator2's set of \c value_types.
+ *  \tparam InputIterator4 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator4's \c value_type is convertible to a type in \p OutputIterator2's set of \c value_types.
+ *  \tparam OutputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam OutputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam StrictWeakCompare is a model of <a href="http://www.sgi.com/tech/stl/StrictWeakOrdering.html">Strict Weak Ordering</a>.
+ *
+ *  \pre The ranges <tt>[keys_first1, keys_last1)</tt> and <tt>[keys_first2, keys_last2)</tt> shall be sorted with respect to \p comp.
+ *  \pre The resulting ranges shall not overlap with any input range.
+ *
+ *  The following code snippet demonstrates how to use
+ *  \p merge_by_key to compute the merger of two sets of integers sorted in
+ *  descending order.
+ *
+ *  \code
+ *  #include <thrust/merge.h>
+ *  #include <thrust/functional.h>
+ *  ...
+ *  int A_keys[6] = {11, 9, 7, 5, 3, 1};
+ *  int A_vals[6] = { 0, 0, 0, 0, 0, 0};
+ *
+ *  int B_keys[7] = {13, 8, 5, 3, 2, 1, 1};
+ *  int B_vals[7] = { 1, 1, 1, 1, 1, 1, 1};
+ *
+ *  int keys_result[13];
+ *  int vals_result[13];
+ *
+ *  thrust::pair<int*,int*> end = thrust::merge_by_key(A_keys, A_keys + 6, B_keys, B_keys + 7, A_vals, B_vals, keys_result, vals_result, thrust::greater<int>());
+ *
+ *  // keys_result = {13, 11, 9, 8, 7, 5, 5, 3, 3, 2, 1, 1, 1}
+ *  // vals_result = { 1,  0, 0, 1, 0, 0, 1, 0, 1, 1, 0, 1, 1}
+ *  \endcode
+ *
+ *  \see merge
+ *  \see \p sort_by_key
+ *  \see \p is_sorted
+ */
+template<typename InputIterator1, typename InputIterator2, typename InputIterator3, typename InputIterator4, typename OutputIterator1, typename OutputIterator2, typename StrictWeakCompare>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    merge_by_key(InputIterator1 keys_first1, InputIterator1 keys_last1,
+                 InputIterator2 keys_first2, InputIterator2 keys_last2,
+                 InputIterator3 values_first1, InputIterator4 values_first2,
+                 OutputIterator1 keys_result,
+                 OutputIterator2 values_result,
+                 StrictWeakCompare comp);
+
+
+/*! \} // merging
+ */
+
+} // end thrust
+
+#include <thrust/detail/merge.inl>
+
diff --git a/compat/thrust/mismatch.h b/compat/thrust/mismatch.h
new file mode 100644
index 0000000000..898157ac90
--- /dev/null
+++ b/compat/thrust/mismatch.h
@@ -0,0 +1,258 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file mismatch.h
+ *  \brief Search for differences between ranges
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/execution_policy.h>
+#include <thrust/pair.h>
+
+namespace thrust
+{
+
+
+/*! \addtogroup algorithms
+ */
+
+/*! \addtogroup searching
+ *  \ingroup algorithms
+ *  \{
+ */
+
+
+/*! \p mismatch finds the first position where the two ranges <tt>[first1, last1)</tt>
+ *  and <tt>[first2, first2 + (last1 - first1))</tt> differ. The two versions of 
+ *  \p mismatch use different tests for whether elements differ.
+ *
+ *  This version of \p mismatch finds the first iterator \c i in <tt>[first1, last1)</tt>
+ *  such that <tt>*i == *(first2 + (i - first1))</tt> is \c false. The return value is a
+ *  \c pair whose first element is \c i and whose second element is <tt>*(first2 + (i - first1))</tt>.
+ *  If no such iterator \c i exists, the return value is a \c pair whose first element
+ *  is \c last1 and whose second element is <tt>*(first2 + (last1 - first1))</tt>.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first1 The beginning of the first sequence.
+ *  \param last1  The end of the first sequence.
+ *  \param first2 The beginning of the second sequence.
+ *  \return The first position where the sequences differ.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>
+ *          and \p InputIterator1's \c value_type is equality comparable to \p InputIterator2's \c value_type.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>.
+ *
+ *  \code
+ *  #include <thrust/mismatch.h>
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  thrust::device_vector<int> vec1(4);
+ *  thrust::device_vector<int> vec2(4);
+ *
+ *  vec1[0] = 0;  vec2[0] = 0; 
+ *  vec1[1] = 5;  vec2[1] = 5;
+ *  vec1[2] = 3;  vec2[2] = 8;
+ *  vec1[3] = 7;  vec2[3] = 7;
+ *
+ *  typedef thrust::device_vector<int>::iterator Iterator;
+ *  thrust::pair<Iterator,Iterator> result;
+ *
+ *  result = thrust::mismatch(thrust::device, vec1.begin(), vec1.end(), vec2.begin());
+ *
+ *  // result.first  is vec1.begin() + 2
+ *  // result.second is vec2.begin() + 2
+ *  \endcode
+ *
+ *  \see find
+ *  \see find_if
+ */
+template<typename DerivedPolicy, typename InputIterator1, typename InputIterator2>
+thrust::pair<InputIterator1, InputIterator2> mismatch(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                                      InputIterator1 first1,
+                                                      InputIterator1 last1,
+                                                      InputIterator2 first2);
+
+
+/*! \p mismatch finds the first position where the two ranges <tt>[first1, last1)</tt>
+ * and <tt>[first2, first2 + (last1 - first1))</tt> differ. The two versions of 
+ * \p mismatch use different tests for whether elements differ.
+ *
+ * This version of \p mismatch finds the first iterator \c i in <tt>[first1, last1)</tt>
+ * such that <tt>*i == *(first2 + (i - first1))</tt> is \c false. The return value is a
+ * \c pair whose first element is \c i and whose second element is <tt>*(first2 + (i - first1))</tt>.
+ * If no such iterator \c i exists, the return value is a \c pair whose first element
+ * is \c last1 and whose second element is <tt>*(first2 + (last1 - first1))</tt>.
+ *
+ *  \param first1 The beginning of the first sequence.
+ *  \param last1  The end of the first sequence.
+ *  \param first2 The beginning of the second sequence.
+ *  \return The first position where the sequences differ.
+ *
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>
+ *          and \p InputIterator1's \c value_type is equality comparable to \p InputIterator2's \c value_type.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>.
+ *
+ *  \code
+ *  #include <thrust/mismatch.h>
+ *  #include <thrust/device_vector.h>
+ *  ...
+ *  thrust::device_vector<int> vec1(4);
+ *  thrust::device_vector<int> vec2(4);
+ *
+ *  vec1[0] = 0;  vec2[0] = 0; 
+ *  vec1[1] = 5;  vec2[1] = 5;
+ *  vec1[2] = 3;  vec2[2] = 8;
+ *  vec1[3] = 7;  vec2[3] = 7;
+ *
+ *  typedef thrust::device_vector<int>::iterator Iterator;
+ *  thrust::pair<Iterator,Iterator> result;
+ *
+ *  result = thrust::mismatch(vec1.begin(), vec1.end(), vec2.begin());
+ *
+ *  // result.first  is vec1.begin() + 2
+ *  // result.second is vec2.begin() + 2
+ *  \endcode
+ *
+ *  \see find
+ *  \see find_if
+ */
+template <typename InputIterator1, typename InputIterator2>
+thrust::pair<InputIterator1, InputIterator2> mismatch(InputIterator1 first1,
+                                                      InputIterator1 last1,
+                                                      InputIterator2 first2);
+
+
+/*! \p mismatch finds the first position where the two ranges <tt>[first1, last1)</tt>
+ *  and <tt>[first2, first2 + (last1 - first1))</tt> differ. The two versions of 
+ *  \p mismatch use different tests for whether elements differ.
+ *
+ *  This version of \p mismatch finds the first iterator \c i in <tt>[first1, last1)</tt>
+ *  such that <tt>pred(\*i, \*(first2 + (i - first1))</tt> is \c false. The return value is a
+ *  \c pair whose first element is \c i and whose second element is <tt>*(first2 + (i - first1))</tt>.
+ *  If no such iterator \c i exists, the return value is a \c pair whose first element is
+ *  \c last1 and whose second element is <tt>*(first2 + (last1 - first1))</tt>.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first1 The beginning of the first sequence.
+ *  \param last1  The end of the first sequence.
+ *  \param first2 The beginning of the second sequence.
+ *  \param pred   The binary predicate to compare elements.
+ *  \return The first position where the sequences differ.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>.
+ *  \tparam Predicate is a model of <a href="http://www.sgi.com/tech/stl/BinaryPredicate.html">Input Iterator</a>.
+ *
+ *  \code
+ *  #include <thrust/mismatch.h>
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  thrust::device_vector<int> vec1(4);
+ *  thrust::device_vector<int> vec2(4);
+ *
+ *  vec1[0] = 0;  vec2[0] = 0; 
+ *  vec1[1] = 5;  vec2[1] = 5;
+ *  vec1[2] = 3;  vec2[2] = 8;
+ *  vec1[3] = 7;  vec2[3] = 7;
+ *
+ *  typedef thrust::device_vector<int>::iterator Iterator;
+ *  thrust::pair<Iterator,Iterator> result;
+ *
+ *  result = thrust::mismatch(thrust::device, vec1.begin(), vec1.end(), vec2.begin(), thrust::equal_to<int>());
+ *
+ *  // result.first  is vec1.begin() + 2
+ *  // result.second is vec2.begin() + 2
+ *  \endcode
+ *
+ *  \see find
+ *  \see find_if
+ */
+template<typename DerivedPolicy, typename InputIterator1, typename InputIterator2, typename BinaryPredicate>
+thrust::pair<InputIterator1, InputIterator2> mismatch(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                                      InputIterator1 first1,
+                                                      InputIterator1 last1,
+                                                      InputIterator2 first2,
+                                                      BinaryPredicate pred);
+
+
+/*! \p mismatch finds the first position where the two ranges <tt>[first1, last1)</tt>
+ * and <tt>[first2, first2 + (last1 - first1))</tt> differ. The two versions of 
+ * \p mismatch use different tests for whether elements differ.
+ *
+ * This version of \p mismatch finds the first iterator \c i in <tt>[first1, last1)</tt>
+ * such that <tt>pred(\*i, \*(first2 + (i - first1))</tt> is \c false. The return value is a
+ * \c pair whose first element is \c i and whose second element is <tt>*(first2 + (i - first1))</tt>.
+ * If no such iterator \c i exists, the return value is a \c pair whose first element is
+ * \c last1 and whose second element is <tt>*(first2 + (last1 - first1))</tt>.
+ *
+ *  \param first1 The beginning of the first sequence.
+ *  \param last1  The end of the first sequence.
+ *  \param first2 The beginning of the second sequence.
+ *  \param pred   The binary predicate to compare elements.
+ *  \return The first position where the sequences differ.
+ *
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>.
+ *  \tparam Predicate is a model of <a href="http://www.sgi.com/tech/stl/BinaryPredicate.html">Input Iterator</a>.
+ *
+ *  \code
+ *  #include <thrust/mismatch.h>
+ *  #include <thrust/device_vector.h>
+ *  ...
+ *  thrust::device_vector<int> vec1(4);
+ *  thrust::device_vector<int> vec2(4);
+ *
+ *  vec1[0] = 0;  vec2[0] = 0; 
+ *  vec1[1] = 5;  vec2[1] = 5;
+ *  vec1[2] = 3;  vec2[2] = 8;
+ *  vec1[3] = 7;  vec2[3] = 7;
+ *
+ *  typedef thrust::device_vector<int>::iterator Iterator;
+ *  thrust::pair<Iterator,Iterator> result;
+ *
+ *  result = thrust::mismatch(vec1.begin(), vec1.end(), vec2.begin(), thrust::equal_to<int>());
+ *
+ *  // result.first  is vec1.begin() + 2
+ *  // result.second is vec2.begin() + 2
+ *  \endcode
+ *
+ *  \see find
+ *  \see find_if
+ */
+template <typename InputIterator1, typename InputIterator2, typename BinaryPredicate>
+thrust::pair<InputIterator1, InputIterator2> mismatch(InputIterator1 first1,
+                                                      InputIterator1 last1,
+                                                      InputIterator2 first2,
+                                                      BinaryPredicate pred);
+
+/*! \} // end searching
+ */
+
+} // end namespace thrust
+
+#include <thrust/detail/mismatch.inl>
+
diff --git a/compat/thrust/pair.h b/compat/thrust/pair.h
new file mode 100644
index 0000000000..897cc078e0
--- /dev/null
+++ b/compat/thrust/pair.h
@@ -0,0 +1,283 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+/*! \file pair.h
+ *  \brief A type encapsulating a heterogeneous pair of elements
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <utility>
+
+namespace thrust
+{
+
+/*! \addtogroup utility
+ *  \{
+ */
+
+/*! \addtogroup pair
+ *  \{
+ */
+
+/*! \p pair is a generic data structure encapsulating a heterogeneous
+ *  pair of values.
+ *
+ *  \tparam T1 The type of \p pair's first object type.  There are no
+ *          requirements on the type of \p T1. <tt>T1</tt>'s type is
+ *          provided by <tt>pair::first_type</tt>.
+ *
+ *  \tparam T2 The type of \p pair's second object type.  There are no
+ *          requirements on the type of \p T2. <tt>T2</tt>'s type is
+ *          provided by <tt>pair::second_type</tt>.
+ */
+template <typename T1, typename T2>
+  struct pair
+{
+  /*! \p first_type is the type of \p pair's first object type.
+   */
+  typedef T1 first_type;
+
+  /*! \p second_type is the type of \p pair's second object type.
+   */
+  typedef T2 second_type;
+
+  /*! The \p pair's first object.
+   */
+  first_type first;
+
+  /*! The \p pair's second object.
+   */
+  second_type second;
+
+  /*! \p pair's default constructor constructs \p first
+   *  and \p second using \c first_type & \c second_type's
+   *  default constructors, respectively.
+   */
+  __host__ __device__ pair(void);
+
+  /*! This constructor accepts two objects to copy into this \p pair.
+   *
+   *  \param x The object to copy into \p first.
+   *  \param y The object to copy into \p second.
+   */
+  inline __host__ __device__
+  pair(const T1 &x, const T2 &y);
+
+  /*! This copy constructor copies from a \p pair whose types are
+   *  convertible to this \p pair's \c first_type and \c second_type,
+   *  respectively.
+   *
+   *  \param p The \p pair to copy from.
+   *
+   *  \tparam U1 is convertible to \c first_type.
+   *  \tparam U2 is convertible to \c second_type.
+   */
+  template <typename U1, typename U2>
+  inline __host__ __device__
+  pair(const pair<U1,U2> &p);
+
+  /*! This copy constructor copies from a <tt>std::pair</tt> whose types are
+   *  convertible to this \p pair's \c first_type and \c second_type,
+   *  respectively.
+   *
+   *  \param p The <tt>std::pair</tt> to copy from.
+   *
+   *  \tparam U1 is convertible to \c first_type.
+   *  \tparam U2 is convertible to \c second_type.
+   */
+  template <typename U1, typename U2>
+  inline __host__ __device__
+  pair(const std::pair<U1,U2> &p);
+
+  /*! \p swap swaps the elements of two <tt>pair</tt>s.
+   *  
+   *  \param p The other <tt>pair</tt> with which to swap.
+   */
+  inline __host__ __device__
+  void swap(pair &p);
+}; // end pair
+
+
+/*! This operator tests two \p pairs for equality.
+ *
+ *  \param x The first \p pair to compare.
+ *  \param y The second \p pair to compare.
+ *  \return \c true if and only if <tt>x.first == y.first && x.second == y.second</tt>.
+ *  
+ *  \tparam T1 is a model of <a href="http://www.sgi.com/tech/stl/EqualityComparable.html">Equality Comparable</a>.
+ *  \tparam T2 is a model of <a href="http://www.sgi.com/tech/stl/EqualityComparable.html">Equality Comparable</a>.
+ */
+template <typename T1, typename T2>
+  inline __host__ __device__
+    bool operator==(const pair<T1,T2> &x, const pair<T1,T2> &y);
+
+
+/*! This operator tests two pairs for ascending ordering.
+ *
+ *  \param x The first \p pair to compare.
+ *  \param y The second \p pair to compare.
+ *  \return \c true if and only if <tt>x.first < y.first || (!(y.first < x.first) && x.second < y.second)</tt>.
+ *
+ *  \tparam T1 is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable.html">LessThan Comparable</a>.
+ *  \tparam T2 is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable.html">LessThan Comparable</a>.
+ */
+template <typename T1, typename T2>
+  inline __host__ __device__
+    bool operator<(const pair<T1,T2> &x, const pair<T1,T2> &y);
+
+
+/*! This operator tests two pairs for inequality.
+ *
+ *  \param x The first \p pair to compare.
+ *  \param y The second \p pair to compare.
+ *  \return \c true if and only if <tt>!(x == y)</tt>.
+ *
+ *  \tparam T1 is a model of <a href="http://www.sgi.com/tech/stl/EqualityComparable.html">Equality Comparable</a>.
+ *  \tparam T2 is a model of <a href="http://www.sgi.com/tech/stl/EqualityComparable.html">Equality Comparable</a>.
+ */
+template <typename T1, typename T2>
+  inline __host__ __device__
+    bool operator!=(const pair<T1,T2> &x, const pair<T1,T2> &y);
+
+
+/*! This operator tests two pairs for descending ordering.
+ *
+ *  \param x The first \p pair to compare.
+ *  \param y The second \p pair to compare.
+ *  \return \c true if and only if <tt>y < x</tt>.
+ *
+ *  \tparam T1 is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable.html">LessThan Comparable</a>.
+ *  \tparam T2 is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable.html">LessThan Comparable</a>.
+ */
+template <typename T1, typename T2>
+  inline __host__ __device__
+    bool operator>(const pair<T1,T2> &x, const pair<T1,T2> &y);
+
+
+/*! This operator tests two pairs for ascending ordering or equivalence.
+ *
+ *  \param x The first \p pair to compare.
+ *  \param y The second \p pair to compare.
+ *  \return \c true if and only if <tt>!(y < x)</tt>.
+ *
+ *  \tparam T1 is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable.html">LessThan Comparable</a>.
+ *  \tparam T2 is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable.html">LessThan Comparable</a>.
+ */
+template <typename T1, typename T2>
+  inline __host__ __device__
+    bool operator<=(const pair<T1,T2> &x, const pair<T1,T2> &y);
+
+
+/*! This operator tests two pairs for descending ordering or equivalence.
+ *
+ *  \param x The first \p pair to compare.
+ *  \param y The second \p pair to compare.
+ *  \return \c true if and only if <tt>!(x < y)</tt>.
+ *
+ *  \tparam T1 is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable.html">LessThan Comparable</a>.
+ *  \tparam T2 is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable.html">LessThan Comparable</a>.
+ */
+template <typename T1, typename T2>
+  inline __host__ __device__
+    bool operator>=(const pair<T1,T2> &x, const pair<T1,T2> &y);
+
+
+/*! \p swap swaps the contents of two <tt>pair</tt>s.
+ *
+ *  \param x The first \p pair to swap.
+ *  \param y The second \p pair to swap.
+ */
+template <typename T1, typename T2>
+  inline __host__ __device__
+    void swap(pair<T1,T2> &x, pair<T1,T2> &y);
+
+
+/*! This convenience function creates a \p pair from two objects.
+ *
+ *  \param x The first object to copy from.
+ *  \param y The second object to copy from.
+ *  \return A newly-constructed \p pair copied from \p a and \p b.
+ *
+ *  \tparam T1 There are no requirements on the type of \p T1.
+ *  \tparam T2 There are no requirements on the type of \p T2.
+ */
+template <typename T1, typename T2>
+  inline __host__ __device__
+    pair<T1,T2> make_pair(T1 x, T2 y);
+
+
+/*! This convenience metafunction is included for compatibility with
+ *  \p tuple. It returns either the type of a \p pair's
+ *  \c first_type or \c second_type in its nested type, \c type.
+ *
+ *  \tparam N This parameter selects the member of interest.
+ *  \tparam T A \c pair type of interest.
+ */
+template<int N, typename T> struct tuple_element;
+
+
+/*! This convenience metafunction is included for compatibility with
+ *  \p tuple. It returns \c 2, the number of elements of a \p pair,
+ *  in its nested data member, \c value.
+ *
+ *  \tparam Pair A \c pair type of interest.
+ */
+template<typename Pair> struct tuple_size;
+
+
+/*! This convenience function returns a reference to either the first or
+ *  second member of a \p pair.
+ *
+ *  \param p The \p pair of interest.
+ *  \return \c p.first or \c p.second, depending on the template
+ *          parameter.
+ *
+ *  \tparam N This parameter selects the member of interest.
+ */
+// XXX comment out these prototypes as a WAR to a problem on MSVC 2005
+//template<unsigned int N, typename T1, typename T2>
+//  inline __host__ __device__
+//    typename tuple_element<N, pair<T1,T2> >::type &
+//      get(pair<T1,T2> &p);
+
+
+/*! This convenience function returns a const reference to either the
+ *  first or second member of a \p pair.
+ *
+ *  \param p The \p pair of interest.
+ *  \return \c p.first or \c p.second, depending on the template
+ *          parameter.
+ *
+ *  \tparam i This parameter selects the member of interest.
+ */
+// XXX comment out these prototypes as a WAR to a problem on MSVC 2005
+//template<int N, typename T1, typename T2>
+//  inline __host__ __device__
+//    const typename tuple_element<N, pair<T1,T2> >::type &
+//      get(const pair<T1,T2> &p);
+
+/*! \} // pair
+ */
+
+/*! \} // utility
+ */
+
+} // end thrust
+
+#include <thrust/detail/pair.inl>
+
diff --git a/compat/thrust/partition.h b/compat/thrust/partition.h
new file mode 100644
index 0000000000..61a6278a8d
--- /dev/null
+++ b/compat/thrust/partition.h
@@ -0,0 +1,1429 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file partition.h
+ *  \brief Reorganizes a range based on a predicate
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/execution_policy.h>
+#include <thrust/pair.h>
+
+namespace thrust
+{
+
+
+/*! \addtogroup reordering
+ *  \ingroup algorithms
+ *
+ *  \addtogroup partitioning
+ *  \ingroup reordering
+ *  \{
+ */
+
+
+/*! \p partition reorders the elements <tt>[first, last)</tt> based on the function
+ *  object \p pred, such that all of the elements that satisfy \p pred precede the
+ *  elements that fail to satisfy it. The postcondition is that, for some iterator
+ *  \c middle in the range <tt>[first, last)</tt>, <tt>pred(*i)</tt> is \c true for every
+ *  iterator \c i in the range <tt>[first,middle)</tt> and \c false for every iterator
+ *  \c i in the range <tt>[middle, last)</tt>. The return value of \p partition is
+ *  \c middle.
+ *
+ *  Note that the relative order of elements in the two reordered sequences is not
+ *  necessarily the same as it was in the original sequence. A different algorithm,
+ *  \ref stable_partition, does guarantee to preserve the relative order.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the sequence to reorder.
+ *  \param last The end of the sequence to reorder.
+ *  \param pred A function object which decides to which partition each element of the
+ *              sequence <tt>[first, last)</tt> belongs.
+ *  \return An iterator referring to the first element of the second partition, that is,
+ *          the sequence of the elements which do not satisfy \p pred.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>,
+ *          and \p ForwardIterator's \c value_type is convertible to \p Predicate's \c argument_type,
+ *          and \p ForwardIterator is mutable.
+ *  \tparam Predicate is a model of <a href="http://www.sgi.com/tech/stl/Predicate.html">Predicate</a>.
+ *
+ *  The following code snippet demonstrates how to use \p partition to reorder a
+ *  sequence so that even numbers precede odd numbers using the \p thrust::host execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/partition.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  struct is_even
+ *  {
+ *    __host__ __device__
+ *    bool operator()(const int &x)
+ *    {
+ *      return (x % 2) == 0;
+ *    }
+ *  };
+ *  ...
+ *  int A[] = {1, 2, 3, 4, 5, 6, 7, 8, 9, 10};
+ *  const int N = sizeof(A)/sizeof(int);
+ *  thrust::partition(thrust::host,
+ *                    A, A + N,
+ *                    is_even());
+ *  // A is now {2, 4, 6, 8, 10, 1, 3, 5, 7, 9}
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/partition.html
+ *  \see \p stable_partition
+ *  \see \p partition_copy
+ */
+template<typename DerivedPolicy,
+         typename ForwardIterator,
+         typename Predicate>
+  ForwardIterator partition(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                            ForwardIterator first,
+                            ForwardIterator last,
+                            Predicate pred);
+
+
+/*! \p partition reorders the elements <tt>[first, last)</tt> based on the function
+ *  object \p pred, such that all of the elements that satisfy \p pred precede the
+ *  elements that fail to satisfy it. The postcondition is that, for some iterator
+ *  \c middle in the range <tt>[first, last)</tt>, <tt>pred(*i)</tt> is \c true for every
+ *  iterator \c i in the range <tt>[first,middle)</tt> and \c false for every iterator
+ *  \c i in the range <tt>[middle, last)</tt>. The return value of \p partition is
+ *  \c middle.
+ *
+ *  Note that the relative order of elements in the two reordered sequences is not
+ *  necessarily the same as it was in the original sequence. A different algorithm,
+ *  \ref stable_partition, does guarantee to preserve the relative order.
+ *
+ *  \param first The beginning of the sequence to reorder.
+ *  \param last The end of the sequence to reorder.
+ *  \param pred A function object which decides to which partition each element of the
+ *              sequence <tt>[first, last)</tt> belongs.
+ *  \return An iterator referring to the first element of the second partition, that is,
+ *          the sequence of the elements which do not satisfy \p pred.
+ *
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>,
+ *          and \p ForwardIterator's \c value_type is convertible to \p Predicate's \c argument_type,
+ *          and \p ForwardIterator is mutable.
+ *  \tparam Predicate is a model of <a href="http://www.sgi.com/tech/stl/Predicate.html">Predicate</a>.
+ *
+ *  The following code snippet demonstrates how to use \p partition to reorder a
+ *  sequence so that even numbers precede odd numbers.
+ *
+ *  \code
+ *  #include <thrust/partition.h>
+ *  ...
+ *  struct is_even
+ *  {
+ *    __host__ __device__
+ *    bool operator()(const int &x)
+ *    {
+ *      return (x % 2) == 0;
+ *    }
+ *  };
+ *  ...
+ *  int A[] = {1, 2, 3, 4, 5, 6, 7, 8, 9, 10};
+ *  const int N = sizeof(A)/sizeof(int);
+ *  thrust::partition(A, A + N,
+ *                     is_even());
+ *  // A is now {2, 4, 6, 8, 10, 1, 3, 5, 7, 9}
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/partition.html
+ *  \see \p stable_partition
+ *  \see \p partition_copy
+ */
+template<typename ForwardIterator,
+         typename Predicate>
+  ForwardIterator partition(ForwardIterator first,
+                            ForwardIterator last,
+                            Predicate pred);
+
+
+/*! \p partition reorders the elements <tt>[first, last)</tt> based on the function
+ *  object \p pred applied to a stencil range <tt>[stencil, stencil + (last - first))</tt>,
+ *  such that all of the elements whose corresponding stencil element satisfies \p pred precede all of the elements whose
+ *  corresponding stencil element fails to satisfy it. The postcondition is that, for some iterator
+ *  \c middle in the range <tt>[first, last)</tt>, <tt>pred(*stencil_i)</tt> is \c true for every iterator
+ *  \c stencil_i in the range <tt>[stencil,stencil + (middle - first))</tt> and \c false for every iterator \c stencil_i
+ *  in the range <tt>[stencil + (middle - first), stencil + (last - first))</tt>.
+ *  The return value of \p stable_partition is \c middle.
+ *
+ *  Note that the relative order of elements in the two reordered sequences is not
+ *  necessarily the same as it was in the original sequence. A different algorithm,
+ *  \ref stable_partition, does guarantee to preserve the relative order.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the sequence to reorder.
+ *  \param last The end of the sequence to reorder.
+ *  \param stencil The beginning of the stencil sequence.
+ *  \param pred A function object which decides to which partition each element of the
+ *              sequence <tt>[first, last)</tt> belongs.
+ *  \return An iterator referring to the first element of the second partition, that is,
+ *          the sequence of the elements whose stencil elements do not satisfy \p pred.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>,
+ *          and \p ForwardIterator is mutable.
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator's \c value_type is convertible to \p Predicate's \c argument_type.
+ *  \tparam Predicate is a model of <a href="http://www.sgi.com/tech/stl/Predicate.html">Predicate</a>.
+ *
+ *  \pre The ranges <tt>[first,last)</tt> and <tt>[stencil, stencil + (last - first))</tt> shall not overlap.
+ *
+ *  The following code snippet demonstrates how to use \p partition to reorder a
+ *  sequence so that even numbers precede odd numbers using the \p thrust::host execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/partition.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  struct is_even
+ *  {
+ *    __host__ __device__
+ *    bool operator()(const int &x)
+ *    {
+ *      return (x % 2) == 0;
+ *    }
+ *  };
+ *  ...
+ *  int A[] = {0, 1, 0, 1, 0, 1, 0, 1, 0,  1};
+ *  int S[] = {1, 2, 3, 4, 5, 6, 7, 8, 9, 10};
+ *  const int N = sizeof(A)/sizeof(int);
+ *  thrust::partition(thrust::host, A, A + N, S, is_even());
+ *  // A is now {1, 1, 1, 1, 1, 0, 0, 0, 0, 0}
+ *  // S is unmodified
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/partition.html
+ *  \see \p stable_partition
+ *  \see \p partition_copy
+ */
+template<typename DerivedPolicy,
+         typename ForwardIterator,
+         typename InputIterator,
+         typename Predicate>
+  ForwardIterator partition(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                            ForwardIterator first,
+                            ForwardIterator last,
+                            InputIterator stencil,
+                            Predicate pred);
+
+
+/*! \p partition reorders the elements <tt>[first, last)</tt> based on the function
+ *  object \p pred applied to a stencil range <tt>[stencil, stencil + (last - first))</tt>,
+ *  such that all of the elements whose corresponding stencil element satisfies \p pred precede all of the elements whose
+ *  corresponding stencil element fails to satisfy it. The postcondition is that, for some iterator
+ *  \c middle in the range <tt>[first, last)</tt>, <tt>pred(*stencil_i)</tt> is \c true for every iterator
+ *  \c stencil_i in the range <tt>[stencil,stencil + (middle - first))</tt> and \c false for every iterator \c stencil_i
+ *  in the range <tt>[stencil + (middle - first), stencil + (last - first))</tt>.
+ *  The return value of \p stable_partition is \c middle.
+ *
+ *  Note that the relative order of elements in the two reordered sequences is not
+ *  necessarily the same as it was in the original sequence. A different algorithm,
+ *  \ref stable_partition, does guarantee to preserve the relative order.
+ *
+ *  \param first The beginning of the sequence to reorder.
+ *  \param last The end of the sequence to reorder.
+ *  \param stencil The beginning of the stencil sequence.
+ *  \param pred A function object which decides to which partition each element of the
+ *              sequence <tt>[first, last)</tt> belongs.
+ *  \return An iterator referring to the first element of the second partition, that is,
+ *          the sequence of the elements whose stencil elements do not satisfy \p pred.
+ *
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>,
+ *          and \p ForwardIterator is mutable.
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator's \c value_type is convertible to \p Predicate's \c argument_type.
+ *  \tparam Predicate is a model of <a href="http://www.sgi.com/tech/stl/Predicate.html">Predicate</a>.
+ *
+ *  \pre The ranges <tt>[first,last)</tt> and <tt>[stencil, stencil + (last - first))</tt> shall not overlap.
+ *
+ *  The following code snippet demonstrates how to use \p partition to reorder a
+ *  sequence so that even numbers precede odd numbers.
+ *
+ *  \code
+ *  #include <thrust/partition.h>
+ *  ...
+ *  struct is_even
+ *  {
+ *    __host__ __device__
+ *    bool operator()(const int &x)
+ *    {
+ *      return (x % 2) == 0;
+ *    }
+ *  };
+ *  ...
+ *  int A[] = {0, 1, 0, 1, 0, 1, 0, 1, 0,  1};
+ *  int S[] = {1, 2, 3, 4, 5, 6, 7, 8, 9, 10};
+ *  const int N = sizeof(A)/sizeof(int);
+ *  thrust::partition(A, A + N, S, is_even());
+ *  // A is now {1, 1, 1, 1, 1, 0, 0, 0, 0, 0}
+ *  // S is unmodified
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/partition.html
+ *  \see \p stable_partition
+ *  \see \p partition_copy
+ */
+template<typename ForwardIterator,
+         typename InputIterator,
+         typename Predicate>
+  ForwardIterator partition(ForwardIterator first,
+                            ForwardIterator last,
+                            InputIterator stencil,
+                            Predicate pred);
+
+
+/*! \p partition_copy differs from \ref partition only in that the reordered
+ *  sequence is written to difference output sequences, rather than in place.
+ *
+ *  \p partition_copy copies the elements <tt>[first, last)</tt> based on the
+ *  function object \p pred. All of the elements that satisfy \p pred are copied
+ *  to the range beginning at \p out_true and all the elements that fail to satisfy it
+ *  are copied to the range beginning at \p out_false.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the sequence to reorder.
+ *  \param last The end of the sequence to reorder.
+ *  \param out_true The destination of the resulting sequence of elements which satisfy \p pred.
+ *  \param out_false The destination of the resulting sequence of elements which fail to satisfy \p pred.
+ *  \param pred A function object which decides to which partition each element of the
+ *              sequence <tt>[first, last)</tt> belongs.
+ *  \return A \p pair p such that <tt>p.first</tt> is the end of the output range beginning
+ *          at \p out_true and <tt>p.second</tt> is the end of the output range beginning at
+ *          \p out_false.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator's \c value_type is convertible to \p Predicate's \c argument_type and \p InputIterator's \c value_type
+ *          is convertible to \p OutputIterator1 and \p OutputIterator2's \c value_types.
+ *  \tparam OutputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam OutputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam Predicate is a model of <a href="http://www.sgi.com/tech/stl/Predicate.html">Predicate</a>.
+ *
+ *  \pre The input range shall not overlap with either output range.
+ *
+ *  The following code snippet demonstrates how to use \p partition_copy to separate a
+ *  sequence into two output sequences of even and odd numbers using the \p thrust::host execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/partition.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  struct is_even
+ *  {
+ *    __host__ __device__
+ *    bool operator()(const int &x)
+ *    {
+ *      return (x % 2) == 0;
+ *    }
+ *  };
+ *  ...
+ *  int A[] = {1, 2, 3, 4, 5, 6, 7, 8, 9, 10};
+ *  int result[10];
+ *  const int N = sizeof(A)/sizeof(int);
+ *  int *evens = result;
+ *  int *odds  = result + 5;
+ *  thrust::partition_copy(thrust::host, A, A + N, evens, odds, is_even());
+ *  // A remains {1, 2, 3, 4, 5, 6, 7, 8, 9, 10}
+ *  // result is now {2, 4, 6, 8, 10, 1, 3, 5, 7, 9}
+ *  // evens points to {2, 4, 6, 8, 10}
+ *  // odds points to {1, 3, 5, 7, 9}
+ *  \endcode
+ *
+ *  \note The relative order of elements in the two reordered sequences is not
+ *  necessarily the same as it was in the original sequence. A different algorithm,
+ *  \ref stable_partition_copy, does guarantee to preserve the relative order.
+ *
+ *  \see http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2569.pdf
+ *  \see \p stable_partition_copy
+ *  \see \p partition
+ */
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename Predicate>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    partition_copy(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                   InputIterator first,
+                   InputIterator last,
+                   OutputIterator1 out_true,
+                   OutputIterator2 out_false,
+                   Predicate pred);
+
+
+/*! \p partition_copy differs from \ref partition only in that the reordered
+ *  sequence is written to difference output sequences, rather than in place.
+ *
+ *  \p partition_copy copies the elements <tt>[first, last)</tt> based on the
+ *  function object \p pred. All of the elements that satisfy \p pred are copied
+ *  to the range beginning at \p out_true and all the elements that fail to satisfy it
+ *  are copied to the range beginning at \p out_false.
+ *
+ *  \param first The beginning of the sequence to reorder.
+ *  \param last The end of the sequence to reorder.
+ *  \param out_true The destination of the resulting sequence of elements which satisfy \p pred.
+ *  \param out_false The destination of the resulting sequence of elements which fail to satisfy \p pred.
+ *  \param pred A function object which decides to which partition each element of the
+ *              sequence <tt>[first, last)</tt> belongs.
+ *  \return A \p pair p such that <tt>p.first</tt> is the end of the output range beginning
+ *          at \p out_true and <tt>p.second</tt> is the end of the output range beginning at
+ *          \p out_false.
+ *
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator's \c value_type is convertible to \p Predicate's \c argument_type and \p InputIterator's \c value_type
+ *          is convertible to \p OutputIterator1 and \p OutputIterator2's \c value_types.
+ *  \tparam OutputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam OutputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam Predicate is a model of <a href="http://www.sgi.com/tech/stl/Predicate.html">Predicate</a>.
+ *
+ *  \pre The input range shall not overlap with either output range.
+ *
+ *  The following code snippet demonstrates how to use \p partition_copy to separate a
+ *  sequence into two output sequences of even and odd numbers.
+ *
+ *  \code
+ *  #include <thrust/partition.h>
+ *  ...
+ *  struct is_even
+ *  {
+ *    __host__ __device__
+ *    bool operator()(const int &x)
+ *    {
+ *      return (x % 2) == 0;
+ *    }
+ *  };
+ *  ...
+ *  int A[] = {1, 2, 3, 4, 5, 6, 7, 8, 9, 10};
+ *  int result[10];
+ *  const int N = sizeof(A)/sizeof(int);
+ *  int *evens = result;
+ *  int *odds  = result + 5;
+ *  thrust::partition_copy(A, A + N, evens, odds, is_even());
+ *  // A remains {1, 2, 3, 4, 5, 6, 7, 8, 9, 10}
+ *  // result is now {2, 4, 6, 8, 10, 1, 3, 5, 7, 9}
+ *  // evens points to {2, 4, 6, 8, 10}
+ *  // odds points to {1, 3, 5, 7, 9}
+ *  \endcode
+ *
+ *  \note The relative order of elements in the two reordered sequences is not
+ *  necessarily the same as it was in the original sequence. A different algorithm,
+ *  \ref stable_partition_copy, does guarantee to preserve the relative order.
+ *
+ *  \see http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2569.pdf
+ *  \see \p stable_partition_copy
+ *  \see \p partition
+ */
+template<typename InputIterator,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename Predicate>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    partition_copy(InputIterator first,
+                   InputIterator last,
+                   OutputIterator1 out_true,
+                   OutputIterator2 out_false,
+                   Predicate pred);
+
+
+/*! \p partition_copy differs from \ref partition only in that the reordered
+ *  sequence is written to difference output sequences, rather than in place.
+ *
+ *  \p partition_copy copies the elements <tt>[first, last)</tt> based on the
+ *  function object \p pred which is applied to a range of stencil elements. All of the elements
+ *  whose corresponding stencil element satisfies \p pred are copied to the range beginning at \p out_true
+ *  and all the elements whose stencil element fails to satisfy it are copied to the range beginning
+ *  at \p out_false.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the sequence to reorder.
+ *  \param last The end of the sequence to reorder.
+ *  \param stencil The beginning of the stencil sequence.
+ *  \param out_true The destination of the resulting sequence of elements which satisfy \p pred.
+ *  \param out_false The destination of the resulting sequence of elements which fail to satisfy \p pred.
+ *  \param pred A function object which decides to which partition each element of the
+ *              sequence <tt>[first, last)</tt> belongs.
+ *  \return A \p pair p such that <tt>p.first</tt> is the end of the output range beginning
+ *          at \p out_true and <tt>p.second</tt> is the end of the output range beginning at
+ *          \p out_false.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator's \c value_type is convertible to \p OutputIterator1 and \p OutputIterator2's \c value_types.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator2's \c value_type is convertible to \p Predicate's \c argument_type.
+ *  \tparam OutputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam OutputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam Predicate is a model of <a href="http://www.sgi.com/tech/stl/Predicate.html">Predicate</a>.
+ *
+ *  \pre The input ranges shall not overlap with either output range.
+ *
+ *  The following code snippet demonstrates how to use \p partition_copy to separate a
+ *  sequence into two output sequences of even and odd numbers using the \p thrust::host execution
+ *  policy for parallelization.
+ *
+ *  \code
+ *  #include <thrust/partition.h>
+ *  #include <thrust/functional.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  int A[] = {1, 2, 3, 4, 5, 6, 7, 8, 9, 10};
+ *  int S[] = {0, 1, 0, 1, 0, 1, 0, 1, 0,  1};
+ *  int result[10];
+ *  const int N = sizeof(A)/sizeof(int);
+ *  int *evens = result;
+ *  int *odds  = result + 5;
+ *  thrust::stable_partition_copy(thrust::host, A, A + N, S, evens, odds, thrust::identity<int>());
+ *  // A remains {1, 2, 3, 4, 5, 6, 7, 8, 9, 10}
+ *  // S remains {0, 1, 0, 1, 0, 1, 0, 1, 0,  1}
+ *  // result is now {2, 4, 6, 8, 10, 1, 3, 5, 7, 9}
+ *  // evens points to {2, 4, 6, 8, 10}
+ *  // odds points to {1, 3, 5, 7, 9}
+ *  \endcode
+ *
+ *  \note The relative order of elements in the two reordered sequences is not
+ *  necessarily the same as it was in the original sequence. A different algorithm,
+ *  \ref stable_partition_copy, does guarantee to preserve the relative order.
+ *
+ *  \see http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2569.pdf
+ *  \see \p stable_partition_copy
+ *  \see \p partition
+ */
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename Predicate>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    partition_copy(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                   InputIterator1 first,
+                   InputIterator1 last,
+                   InputIterator2 stencil,
+                   OutputIterator1 out_true,
+                   OutputIterator2 out_false,
+                   Predicate pred);
+
+
+/*! \p partition_copy differs from \ref partition only in that the reordered
+ *  sequence is written to difference output sequences, rather than in place.
+ *
+ *  \p partition_copy copies the elements <tt>[first, last)</tt> based on the
+ *  function object \p pred which is applied to a range of stencil elements. All of the elements
+ *  whose corresponding stencil element satisfies \p pred are copied to the range beginning at \p out_true
+ *  and all the elements whose stencil element fails to satisfy it are copied to the range beginning
+ *  at \p out_false.
+ *
+ *  \param first The beginning of the sequence to reorder.
+ *  \param last The end of the sequence to reorder.
+ *  \param stencil The beginning of the stencil sequence.
+ *  \param out_true The destination of the resulting sequence of elements which satisfy \p pred.
+ *  \param out_false The destination of the resulting sequence of elements which fail to satisfy \p pred.
+ *  \param pred A function object which decides to which partition each element of the
+ *              sequence <tt>[first, last)</tt> belongs.
+ *  \return A \p pair p such that <tt>p.first</tt> is the end of the output range beginning
+ *          at \p out_true and <tt>p.second</tt> is the end of the output range beginning at
+ *          \p out_false.
+ *
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator's \c value_type is convertible to \p OutputIterator1 and \p OutputIterator2's \c value_types.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator2's \c value_type is convertible to \p Predicate's \c argument_type.
+ *  \tparam OutputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam OutputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam Predicate is a model of <a href="http://www.sgi.com/tech/stl/Predicate.html">Predicate</a>.
+ *
+ *  \pre The input ranges shall not overlap with either output range.
+ *
+ *  The following code snippet demonstrates how to use \p partition_copy to separate a
+ *  sequence into two output sequences of even and odd numbers.
+ *
+ *  \code
+ *  #include <thrust/partition.h>
+ *  #include <thrust/functional.h>
+ *  ...
+ *  int A[] = {1, 2, 3, 4, 5, 6, 7, 8, 9, 10};
+ *  int S[] = {0, 1, 0, 1, 0, 1, 0, 1, 0,  1};
+ *  int result[10];
+ *  const int N = sizeof(A)/sizeof(int);
+ *  int *evens = result;
+ *  int *odds  = result + 5;
+ *  thrust::stable_partition_copy(A, A + N, S, evens, odds, thrust::identity<int>());
+ *  // A remains {1, 2, 3, 4, 5, 6, 7, 8, 9, 10}
+ *  // S remains {0, 1, 0, 1, 0, 1, 0, 1, 0,  1}
+ *  // result is now {2, 4, 6, 8, 10, 1, 3, 5, 7, 9}
+ *  // evens points to {2, 4, 6, 8, 10}
+ *  // odds points to {1, 3, 5, 7, 9}
+ *  \endcode
+ *
+ *  \note The relative order of elements in the two reordered sequences is not
+ *  necessarily the same as it was in the original sequence. A different algorithm,
+ *  \ref stable_partition_copy, does guarantee to preserve the relative order.
+ *
+ *  \see http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2569.pdf
+ *  \see \p stable_partition_copy
+ *  \see \p partition
+ */
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename Predicate>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    partition_copy(InputIterator1 first,
+                   InputIterator1 last,
+                   InputIterator2 stencil,
+                   OutputIterator1 out_true,
+                   OutputIterator2 out_false,
+                   Predicate pred);
+
+
+/*! \p stable_partition is much like \ref partition : it reorders the elements in the
+ *  range <tt>[first, last)</tt> based on the function object \p pred, such that all of
+ *  the elements that satisfy \p pred precede all of the elements that fail to satisfy
+ *  it. The postcondition is that, for some iterator \p middle in the range
+ *  <tt>[first, last)</tt>, <tt>pred(*i)</tt> is \c true for every iterator \c i in the
+ *  range <tt>[first,middle)</tt> and \c false for every iterator \c i in the range
+ *  <tt>[middle, last)</tt>. The return value of \p stable_partition is \c middle.
+ *
+ *  \p stable_partition differs from \ref partition in that \p stable_partition is
+ *  guaranteed to preserve relative order. That is, if \c x and \c y are elements in
+ *  <tt>[first, last)</tt>, and \c stencil_x and \c stencil_y are the stencil elements
+ *  in corresponding positions within <tt>[stencil, stencil + (last - first))</tt>,
+ *  and <tt>pred(stencil_x) == pred(stencil_y)</tt>, and if \c x precedes
+ *  \c y, then it will still be true after \p stable_partition that \c x precedes \c y.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The first element of the sequence to reorder.
+ *  \param last One position past the last element of the sequence to reorder.
+ *  \param pred A function object which decides to which partition each element of the
+ *              sequence <tt>[first, last)</tt> belongs.
+ *  \return An iterator referring to the first element of the second partition, that is,
+ *          the sequence of the elements which do not satisfy pred.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>,
+ *          and \p ForwardIterator's \c value_type is convertible to \p Predicate's \c argument_type,
+ *          and \p ForwardIterator is mutable.
+ *  \tparam Predicate is a model of <a href="http://www.sgi.com/tech/stl/Predicate.html">Predicate</a>.
+ *
+ *  The following code snippet demonstrates how to use \p stable_partition to reorder a
+ *  sequence so that even numbers precede odd numbers using the \p thrust::host execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/partition.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  struct is_even
+ *  {
+ *    __host__ __device__
+ *    bool operator()(const int &x)
+ *    {
+ *      return (x % 2) == 0;
+ *    }
+ *  };
+ *  ...
+ *  int A[] = {1, 2, 3, 4, 5, 6, 7, 8, 9, 10};
+ *  const int N = sizeof(A)/sizeof(int);
+ *  thrust::stable_partition(thrust::host,
+ *                           A, A + N,
+ *                           is_even());
+ *  // A is now {2, 4, 6, 8, 10, 1, 3, 5, 7, 9}
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/stable_partition.html
+ *  \see \p partition
+ *  \see \p stable_partition_copy
+ */
+template<typename DerivedPolicy,
+         typename ForwardIterator,
+         typename Predicate>
+  ForwardIterator stable_partition(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                   ForwardIterator first,
+                                   ForwardIterator last,
+                                   Predicate pred);
+
+
+/*! \p stable_partition is much like \ref partition : it reorders the elements in the
+ *  range <tt>[first, last)</tt> based on the function object \p pred, such that all of
+ *  the elements that satisfy \p pred precede all of the elements that fail to satisfy
+ *  it. The postcondition is that, for some iterator \p middle in the range
+ *  <tt>[first, last)</tt>, <tt>pred(*i)</tt> is \c true for every iterator \c i in the
+ *  range <tt>[first,middle)</tt> and \c false for every iterator \c i in the range
+ *  <tt>[middle, last)</tt>. The return value of \p stable_partition is \c middle.
+ *
+ *  \p stable_partition differs from \ref partition in that \p stable_partition is
+ *  guaranteed to preserve relative order. That is, if \c x and \c y are elements in
+ *  <tt>[first, last)</tt>, and \c stencil_x and \c stencil_y are the stencil elements
+ *  in corresponding positions within <tt>[stencil, stencil + (last - first))</tt>,
+ *  and <tt>pred(stencil_x) == pred(stencil_y)</tt>, and if \c x precedes
+ *  \c y, then it will still be true after \p stable_partition that \c x precedes \c y.
+ *
+ *  \param first The first element of the sequence to reorder.
+ *  \param last One position past the last element of the sequence to reorder.
+ *  \param pred A function object which decides to which partition each element of the
+ *              sequence <tt>[first, last)</tt> belongs.
+ *  \return An iterator referring to the first element of the second partition, that is,
+ *          the sequence of the elements which do not satisfy pred.
+ *
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>,
+ *          and \p ForwardIterator's \c value_type is convertible to \p Predicate's \c argument_type,
+ *          and \p ForwardIterator is mutable.
+ *  \tparam Predicate is a model of <a href="http://www.sgi.com/tech/stl/Predicate.html">Predicate</a>.
+ *
+ *  The following code snippet demonstrates how to use \p stable_partition to reorder a
+ *  sequence so that even numbers precede odd numbers.
+ *
+ *  \code
+ *  #include <thrust/partition.h>
+ *  ...
+ *  struct is_even
+ *  {
+ *    __host__ __device__
+ *    bool operator()(const int &x)
+ *    {
+ *      return (x % 2) == 0;
+ *    }
+ *  };
+ *  ...
+ *  int A[] = {1, 2, 3, 4, 5, 6, 7, 8, 9, 10};
+ *  const int N = sizeof(A)/sizeof(int);
+ *  thrust::stable_partition(A, A + N,
+ *                            is_even());
+ *  // A is now {2, 4, 6, 8, 10, 1, 3, 5, 7, 9}
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/stable_partition.html
+ *  \see \p partition
+ *  \see \p stable_partition_copy
+ */
+template<typename ForwardIterator,
+         typename Predicate>
+  ForwardIterator stable_partition(ForwardIterator first,
+                                   ForwardIterator last,
+                                   Predicate pred);
+
+
+/*! \p stable_partition is much like \p partition: it reorders the elements in the
+ *  range <tt>[first, last)</tt> based on the function object \p pred applied to a stencil
+ *  range <tt>[stencil, stencil + (last - first))</tt>, such that all of
+ *  the elements whose corresponding stencil element satisfies \p pred precede all of the elements whose
+ *  corresponding stencil element fails to satisfy it. The postcondition is that, for some iterator
+ *  \c middle in the range <tt>[first, last)</tt>, <tt>pred(*stencil_i)</tt> is \c true for every iterator
+ *  \c stencil_i in the range <tt>[stencil,stencil + (middle - first))</tt> and \c false for every iterator \c stencil_i
+ *  in the range <tt>[stencil + (middle - first), stencil + (last - first))</tt>.
+ *  The return value of \p stable_partition is \c middle.
+ *
+ *  \p stable_partition differs from \ref partition in that \p stable_partition is
+ *  guaranteed to preserve relative order. That is, if \c x and \c y are elements in
+ *  <tt>[first, last)</tt>, such that <tt>pred(x) == pred(y)</tt>, and if \c x precedes
+ *  \c y, then it will still be true after \p stable_partition that \c x precedes \c y.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The first element of the sequence to reorder.
+ *  \param last One position past the last element of the sequence to reorder.
+ *  \param stencil The beginning of the stencil sequence.
+ *  \param pred A function object which decides to which partition each element of the
+ *              sequence <tt>[first, last)</tt> belongs.
+ *  \return An iterator referring to the first element of the second partition, that is,
+ *          the sequence of the elements whose stencil elements do not satisfy \p pred.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>,
+ *          and \p ForwardIterator is mutable.
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator's \c value_type is convertible to \p Predicate's \c argument_type.
+ *  \tparam Predicate is a model of <a href="http://www.sgi.com/tech/stl/Predicate.html">Predicate</a>.
+ *
+ *  \pre The range <tt>[first, last)</tt> shall not overlap with the range <tt>[stencil, stencil + (last - first))</tt>.
+ *
+ *  The following code snippet demonstrates how to use \p stable_partition to reorder a
+ *  sequence so that even numbers precede odd numbers using the \p thrust::host execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/partition.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  struct is_even
+ *  {
+ *    __host__ __device__
+ *    bool operator()(const int &x)
+ *    {
+ *      return (x % 2) == 0;
+ *    }
+ *  };
+ *  ...
+ *  int A[] = {0, 1, 0, 1, 0, 1, 0, 1, 0,  1};
+ *  int S[] = {1, 2, 3, 4, 5, 6, 7, 8, 9, 10};
+ *  const int N = sizeof(A)/sizeof(int);
+ *  thrust::stable_partition(thrust::host, A, A + N, S, is_even());
+ *  // A is now {1, 1, 1, 1, 1, 0, 0, 0, 0, 0}
+ *  // S is unmodified
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/stable_partition.html
+ *  \see \p partition
+ *  \see \p stable_partition_copy
+ */
+template<typename DerivedPolicy,
+         typename ForwardIterator,
+         typename InputIterator,
+         typename Predicate>
+  ForwardIterator stable_partition(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                   ForwardIterator first,
+                                   ForwardIterator last,
+                                   InputIterator stencil,
+                                   Predicate pred);
+
+
+/*! \p stable_partition is much like \p partition: it reorders the elements in the
+ *  range <tt>[first, last)</tt> based on the function object \p pred applied to a stencil
+ *  range <tt>[stencil, stencil + (last - first))</tt>, such that all of
+ *  the elements whose corresponding stencil element satisfies \p pred precede all of the elements whose
+ *  corresponding stencil element fails to satisfy it. The postcondition is that, for some iterator
+ *  \c middle in the range <tt>[first, last)</tt>, <tt>pred(*stencil_i)</tt> is \c true for every iterator
+ *  \c stencil_i in the range <tt>[stencil,stencil + (middle - first))</tt> and \c false for every iterator \c stencil_i
+ *  in the range <tt>[stencil + (middle - first), stencil + (last - first))</tt>.
+ *  The return value of \p stable_partition is \c middle.
+ *
+ *  \p stable_partition differs from \ref partition in that \p stable_partition is
+ *  guaranteed to preserve relative order. That is, if \c x and \c y are elements in
+ *  <tt>[first, last)</tt>, such that <tt>pred(x) == pred(y)</tt>, and if \c x precedes
+ *  \c y, then it will still be true after \p stable_partition that \c x precedes \c y.
+ *
+ *  \param first The first element of the sequence to reorder.
+ *  \param last One position past the last element of the sequence to reorder.
+ *  \param stencil The beginning of the stencil sequence.
+ *  \param pred A function object which decides to which partition each element of the
+ *              sequence <tt>[first, last)</tt> belongs.
+ *  \return An iterator referring to the first element of the second partition, that is,
+ *          the sequence of the elements whose stencil elements do not satisfy \p pred.
+ *
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>,
+ *          and \p ForwardIterator is mutable.
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator's \c value_type is convertible to \p Predicate's \c argument_type.
+ *  \tparam Predicate is a model of <a href="http://www.sgi.com/tech/stl/Predicate.html">Predicate</a>.
+ *
+ *  \pre The range <tt>[first, last)</tt> shall not overlap with the range <tt>[stencil, stencil + (last - first))</tt>.
+ *
+ *  The following code snippet demonstrates how to use \p stable_partition to reorder a
+ *  sequence so that even numbers precede odd numbers.
+ *
+ *  \code
+ *  #include <thrust/partition.h>
+ *  ...
+ *  struct is_even
+ *  {
+ *    __host__ __device__
+ *    bool operator()(const int &x)
+ *    {
+ *      return (x % 2) == 0;
+ *    }
+ *  };
+ *  ...
+ *  int A[] = {0, 1, 0, 1, 0, 1, 0, 1, 0,  1};
+ *  int S[] = {1, 2, 3, 4, 5, 6, 7, 8, 9, 10};
+ *  const int N = sizeof(A)/sizeof(int);
+ *  thrust::stable_partition(A, A + N, S, is_even());
+ *  // A is now {1, 1, 1, 1, 1, 0, 0, 0, 0, 0}
+ *  // S is unmodified
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/stable_partition.html
+ *  \see \p partition
+ *  \see \p stable_partition_copy
+ */
+template<typename ForwardIterator,
+         typename InputIterator,
+         typename Predicate>
+  ForwardIterator stable_partition(ForwardIterator first,
+                                   ForwardIterator last,
+                                   InputIterator stencil,
+                                   Predicate pred);
+
+
+/*! \p stable_partition_copy differs from \ref stable_partition only in that the reordered
+ *  sequence is written to different output sequences, rather than in place.
+ *
+ *  \p stable_partition_copy copies the elements <tt>[first, last)</tt> based on the
+ *  function object \p pred. All of the elements that satisfy \p pred are copied
+ *  to the range beginning at \p out_true and all the elements that fail to satisfy it
+ *  are copied to the range beginning at \p out_false.
+ *
+ *  \p stable_partition_copy differs from \ref partition_copy in that
+ *  \p stable_partition_copy is guaranteed to preserve relative order. That is, if
+ *  \c x and \c y are elements in <tt>[first, last)</tt>, such that
+ *  <tt>pred(x) == pred(y)</tt>, and if \c x precedes \c y, then it will still be true
+ *  after \p stable_partition_copy that \c x precedes \c y in the output.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The first element of the sequence to reorder.
+ *  \param last One position past the last element of the sequence to reorder.
+ *  \param out_true The destination of the resulting sequence of elements which satisfy \p pred.
+ *  \param out_false The destination of the resulting sequence of elements which fail to satisfy \p pred.
+ *  \param pred A function object which decides to which partition each element of the
+ *              sequence <tt>[first, last)</tt> belongs.
+ *  \return A \p pair p such that <tt>p.first</tt> is the end of the output range beginning
+ *          at \p out_true and <tt>p.second</tt> is the end of the output range beginning at
+ *          \p out_false.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator's \c value_type is convertible to \p Predicate's \c argument_type and \p InputIterator's \c value_type
+ *          is convertible to \p OutputIterator1 and \p OutputIterator2's \c value_types.
+ *  \tparam OutputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam OutputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam Predicate is a model of <a href="http://www.sgi.com/tech/stl/Predicate.html">Predicate</a>.
+ *
+ *  \pre The input ranges shall not overlap with either output range.
+ *
+ *  The following code snippet demonstrates how to use \p stable_partition_copy to
+ *  reorder a sequence so that even numbers precede odd numbers using the \p thrust::host execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/partition.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  struct is_even
+ *  {
+ *    __host__ __device__
+ *    bool operator()(const int &x)
+ *    {
+ *      return (x % 2) == 0;
+ *    }
+ *  };
+ *  ...
+ *  int A[] = {1, 2, 3, 4, 5, 6, 7, 8, 9, 10};
+ *  int result[10];
+ *  const int N = sizeof(A)/sizeof(int);
+ *  int *evens = result;
+ *  int *odds  = result + 5;
+ *  thrust::stable_partition_copy(thrust::host, A, A + N, evens, odds, is_even());
+ *  // A remains {1, 2, 3, 4, 5, 6, 7, 8, 9, 10}
+ *  // result is now {2, 4, 6, 8, 10, 1, 3, 5, 7, 9}
+ *  // evens points to {2, 4, 6, 8, 10}
+ *  // odds points to {1, 3, 5, 7, 9}
+ *  \endcode
+ *
+ *  \see http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2569.pdf
+ *  \see \p partition_copy
+ *  \see \p stable_partition
+ */
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename Predicate>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    stable_partition_copy(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                          InputIterator first,
+                          InputIterator last,
+                          OutputIterator1 out_true,
+                          OutputIterator2 out_false,
+                          Predicate pred);
+
+
+/*! \p stable_partition_copy differs from \ref stable_partition only in that the reordered
+ *  sequence is written to different output sequences, rather than in place.
+ *
+ *  \p stable_partition_copy copies the elements <tt>[first, last)</tt> based on the
+ *  function object \p pred. All of the elements that satisfy \p pred are copied
+ *  to the range beginning at \p out_true and all the elements that fail to satisfy it
+ *  are copied to the range beginning at \p out_false.
+ *
+ *  \p stable_partition_copy differs from \ref partition_copy in that
+ *  \p stable_partition_copy is guaranteed to preserve relative order. That is, if
+ *  \c x and \c y are elements in <tt>[first, last)</tt>, such that
+ *  <tt>pred(x) == pred(y)</tt>, and if \c x precedes \c y, then it will still be true
+ *  after \p stable_partition_copy that \c x precedes \c y in the output.
+ *
+ *  \param first The first element of the sequence to reorder.
+ *  \param last One position past the last element of the sequence to reorder.
+ *  \param out_true The destination of the resulting sequence of elements which satisfy \p pred.
+ *  \param out_false The destination of the resulting sequence of elements which fail to satisfy \p pred.
+ *  \param pred A function object which decides to which partition each element of the
+ *              sequence <tt>[first, last)</tt> belongs.
+ *  \return A \p pair p such that <tt>p.first</tt> is the end of the output range beginning
+ *          at \p out_true and <tt>p.second</tt> is the end of the output range beginning at
+ *          \p out_false.
+ *
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator's \c value_type is convertible to \p Predicate's \c argument_type and \p InputIterator's \c value_type
+ *          is convertible to \p OutputIterator1 and \p OutputIterator2's \c value_types.
+ *  \tparam OutputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam OutputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam Predicate is a model of <a href="http://www.sgi.com/tech/stl/Predicate.html">Predicate</a>.
+ *
+ *  \pre The input ranges shall not overlap with either output range.
+ *
+ *  The following code snippet demonstrates how to use \p stable_partition_copy to
+ *  reorder a sequence so that even numbers precede odd numbers.
+ *
+ *  \code
+ *  #include <thrust/partition.h>
+ *  ...
+ *  struct is_even
+ *  {
+ *    __host__ __device__
+ *    bool operator()(const int &x)
+ *    {
+ *      return (x % 2) == 0;
+ *    }
+ *  };
+ *  ...
+ *  int A[] = {1, 2, 3, 4, 5, 6, 7, 8, 9, 10};
+ *  int result[10];
+ *  const int N = sizeof(A)/sizeof(int);
+ *  int *evens = result;
+ *  int *odds  = result + 5;
+ *  thrust::stable_partition_copy(A, A + N, evens, odds, is_even());
+ *  // A remains {1, 2, 3, 4, 5, 6, 7, 8, 9, 10}
+ *  // result is now {2, 4, 6, 8, 10, 1, 3, 5, 7, 9}
+ *  // evens points to {2, 4, 6, 8, 10}
+ *  // odds points to {1, 3, 5, 7, 9}
+ *  \endcode
+ *
+ *  \see http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2569.pdf
+ *  \see \p partition_copy
+ *  \see \p stable_partition
+ */
+template<typename InputIterator,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename Predicate>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    stable_partition_copy(InputIterator first,
+                          InputIterator last,
+                          OutputIterator1 out_true,
+                          OutputIterator2 out_false,
+                          Predicate pred);
+
+
+/*! \p stable_partition_copy differs from \ref stable_partition only in that the reordered
+ *  sequence is written to different output sequences, rather than in place.
+ *
+ *  \p stable_partition_copy copies the elements <tt>[first, last)</tt> based on the
+ *  function object \p pred which is applied to a range of stencil elements. All of the elements
+ *  whose corresponding stencil element satisfies \p pred are copied to the range beginning at \p out_true
+ *  and all the elements whose stencil element fails to satisfy it are copied to the range beginning
+ *  at \p out_false.
+ *
+ *  \p stable_partition_copy differs from \ref partition_copy in that
+ *  \p stable_partition_copy is guaranteed to preserve relative order. That is, if
+ *  \c x and \c y are elements in <tt>[first, last)</tt>, such that
+ *  <tt>pred(x) == pred(y)</tt>, and if \c x precedes \c y, then it will still be true
+ *  after \p stable_partition_copy that \c x precedes \c y in the output.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The first element of the sequence to reorder.
+ *  \param last One position past the last element of the sequence to reorder.
+ *  \param stencil The beginning of the stencil sequence.
+ *  \param out_true The destination of the resulting sequence of elements which satisfy \p pred.
+ *  \param out_false The destination of the resulting sequence of elements which fail to satisfy \p pred.
+ *  \param pred A function object which decides to which partition each element of the
+ *              sequence <tt>[first, last)</tt> belongs.
+ *  \return A \p pair p such that <tt>p.first</tt> is the end of the output range beginning
+ *          at \p out_true and <tt>p.second</tt> is the end of the output range beginning at
+ *          \p out_false.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator's \c value_type is convertible to \p OutputIterator1 and \p OutputIterator2's \c value_types.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator2's \c value_type is convertible to \p Predicate's \c argument_type.
+ *  \tparam OutputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam OutputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam Predicate is a model of <a href="http://www.sgi.com/tech/stl/Predicate.html">Predicate</a>.
+ *
+ *  \pre The input ranges shall not overlap with either output range.
+ *
+ *  The following code snippet demonstrates how to use \p stable_partition_copy to
+ *  reorder a sequence so that even numbers precede odd numbers using the \p thrust::host execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/partition.h>
+ *  #include <thrust/functional.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  int A[] = {1, 2, 3, 4, 5, 6, 7, 8, 9, 10};
+ *  int S[] = {0, 1, 0, 1, 0, 1, 0, 1, 0,  1};
+ *  int result[10];
+ *  const int N = sizeof(A)/sizeof(int);
+ *  int *evens = result;
+ *  int *odds  = result + 5;
+ *  thrust::stable_partition_copy(thrust::host, A, A + N, S, evens, odds, thrust::identity<int>());
+ *  // A remains {1, 2, 3, 4, 5, 6, 7, 8, 9, 10}
+ *  // S remains {0, 1, 0, 1, 0, 1, 0, 1, 0,  1}
+ *  // result is now {2, 4, 6, 8, 10, 1, 3, 5, 7, 9}
+ *  // evens points to {2, 4, 6, 8, 10}
+ *  // odds points to {1, 3, 5, 7, 9}
+ *  \endcode
+ *
+ *  \see http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2569.pdf
+ *  \see \p partition_copy
+ *  \see \p stable_partition
+ */
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename Predicate>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    stable_partition_copy(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                          InputIterator1 first,
+                          InputIterator1 last,
+                          InputIterator2 stencil,
+                          OutputIterator1 out_true,
+                          OutputIterator2 out_false,
+                          Predicate pred);
+
+
+/*! \p stable_partition_copy differs from \ref stable_partition only in that the reordered
+ *  sequence is written to different output sequences, rather than in place.
+ *
+ *  \p stable_partition_copy copies the elements <tt>[first, last)</tt> based on the
+ *  function object \p pred which is applied to a range of stencil elements. All of the elements
+ *  whose corresponding stencil element satisfies \p pred are copied to the range beginning at \p out_true
+ *  and all the elements whose stencil element fails to satisfy it are copied to the range beginning
+ *  at \p out_false.
+ *
+ *  \p stable_partition_copy differs from \ref partition_copy in that
+ *  \p stable_partition_copy is guaranteed to preserve relative order. That is, if
+ *  \c x and \c y are elements in <tt>[first, last)</tt>, such that
+ *  <tt>pred(x) == pred(y)</tt>, and if \c x precedes \c y, then it will still be true
+ *  after \p stable_partition_copy that \c x precedes \c y in the output.
+ *
+ *  \param first The first element of the sequence to reorder.
+ *  \param last One position past the last element of the sequence to reorder.
+ *  \param stencil The beginning of the stencil sequence.
+ *  \param out_true The destination of the resulting sequence of elements which satisfy \p pred.
+ *  \param out_false The destination of the resulting sequence of elements which fail to satisfy \p pred.
+ *  \param pred A function object which decides to which partition each element of the
+ *              sequence <tt>[first, last)</tt> belongs.
+ *  \return A \p pair p such that <tt>p.first</tt> is the end of the output range beginning
+ *          at \p out_true and <tt>p.second</tt> is the end of the output range beginning at
+ *          \p out_false.
+ *
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator's \c value_type is convertible to \p OutputIterator1 and \p OutputIterator2's \c value_types.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator2's \c value_type is convertible to \p Predicate's \c argument_type.
+ *  \tparam OutputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam OutputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam Predicate is a model of <a href="http://www.sgi.com/tech/stl/Predicate.html">Predicate</a>.
+ *
+ *  \pre The input ranges shall not overlap with either output range.
+ *
+ *  The following code snippet demonstrates how to use \p stable_partition_copy to
+ *  reorder a sequence so that even numbers precede odd numbers.
+ *
+ *  \code
+ *  #include <thrust/partition.h>
+ *  #include <thrust/functional.h>
+ *  ...
+ *  int A[] = {1, 2, 3, 4, 5, 6, 7, 8, 9, 10};
+ *  int S[] = {0, 1, 0, 1, 0, 1, 0, 1, 0,  1};
+ *  int result[10];
+ *  const int N = sizeof(A)/sizeof(int);
+ *  int *evens = result;
+ *  int *odds  = result + 5;
+ *  thrust::stable_partition_copy(A, A + N, S, evens, odds, thrust::identity<int>());
+ *  // A remains {1, 2, 3, 4, 5, 6, 7, 8, 9, 10}
+ *  // S remains {0, 1, 0, 1, 0, 1, 0, 1, 0,  1}
+ *  // result is now {2, 4, 6, 8, 10, 1, 3, 5, 7, 9}
+ *  // evens points to {2, 4, 6, 8, 10}
+ *  // odds points to {1, 3, 5, 7, 9}
+ *  \endcode
+ *
+ *  \see http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2569.pdf
+ *  \see \p partition_copy
+ *  \see \p stable_partition
+ */
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename Predicate>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    stable_partition_copy(InputIterator1 first,
+                          InputIterator1 last,
+                          InputIterator2 stencil,
+                          OutputIterator1 out_true,
+                          OutputIterator2 out_false,
+                          Predicate pred);
+
+
+/*! \} // end stream_compaction
+ */
+
+/*! \} // end reordering
+ */
+
+/*! \addtogroup searching
+ *  \{
+ */
+
+
+/*! \p partition_point returns an iterator pointing to the end of the true
+ *  partition of a partitioned range. \p partition_point requires the input range
+ *  <tt>[first,last)</tt> to be a partition; that is, all elements which satisfy
+ *  <tt>pred</tt> shall appear before those that do not.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the range to consider.
+ *  \param last The end of the range to consider.
+ *  \param pred A function object which decides to which partition each element of the
+ *              range <tt>[first, last)</tt> belongs.
+ *  \return An iterator \c mid such that <tt>all_of(first, mid, pred)</tt>
+ *          and <tt>none_of(mid, last, pred)</tt> are both true.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>,
+ *          and \p ForwardIterator's \c value_type is convertible to \p Predicate's \c argument_type.
+ *  \tparam Predicate is a model of <a href="http://www.sgi.com/tech/stl/Predicate.html">Predicate</a>.
+ *
+ *  \pre The range <tt>[first, last)</tt> shall be partitioned by \p pred.
+ *
+ *  \note Though similar, \p partition_point is not redundant with \p find_if_not.
+ *        \p partition_point's precondition provides an opportunity for a
+ *        faster implemention.
+ *
+ *  \code
+ *  #include <thrust/partition.h>
+ *  #include <thrust/execution_policy.h>
+ *
+ *  struct is_even
+ *  {
+ *    __host__ __device__
+ *    bool operator()(const int &x)
+ *    {
+ *      return (x % 2) == 0;
+ *    }
+ *  };
+ *  
+ *  ...
+ *
+ *  int A[] = {2, 4, 6, 8, 10, 1, 3, 5, 7, 9};
+ *  int * B = thrust::partition_point(thrust::host, A, A + 10, is_even());
+ *  // B - A is 5
+ *  // [A, B) contains only even values
+ *  \endcode
+ *
+ *  \see \p partition
+ *  \see \p find_if_not
+ */
+template<typename DerivedPolicy, typename ForwardIterator, typename Predicate>
+  ForwardIterator partition_point(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                  ForwardIterator first,
+                                  ForwardIterator last,
+                                  Predicate pred);
+
+
+/*! \p partition_point returns an iterator pointing to the end of the true
+ *  partition of a partitioned range. \p partition_point requires the input range
+ *  <tt>[first,last)</tt> to be a partition; that is, all elements which satisfy
+ *  <tt>pred</tt> shall appear before those that do not.
+ *  \param first The beginning of the range to consider.
+ *  \param last The end of the range to consider.
+ *  \param pred A function object which decides to which partition each element of the
+ *              range <tt>[first, last)</tt> belongs.
+ *  \return An iterator \c mid such that <tt>all_of(first, mid, pred)</tt>
+ *          and <tt>none_of(mid, last, pred)</tt> are both true.
+ *
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>,
+ *          and \p ForwardIterator's \c value_type is convertible to \p Predicate's \c argument_type.
+ *  \tparam Predicate is a model of <a href="http://www.sgi.com/tech/stl/Predicate.html">Predicate</a>.
+ *
+ *  \pre The range <tt>[first, last)</tt> shall be partitioned by \p pred.
+ *
+ *  \note Though similar, \p partition_point is not redundant with \p find_if_not.
+ *        \p partition_point's precondition provides an opportunity for a
+ *        faster implemention.
+ *
+ *  \code
+ *  #include <thrust/partition.h>
+ *
+ *  struct is_even
+ *  {
+ *    __host__ __device__
+ *    bool operator()(const int &x)
+ *    {
+ *      return (x % 2) == 0;
+ *    }
+ *  };
+ *  
+ *  ...
+ *
+ *  int A[] = {2, 4, 6, 8, 10, 1, 3, 5, 7, 9};
+ *  int * B = thrust::partition_point(A, A + 10, is_even());
+ *  // B - A is 5
+ *  // [A, B) contains only even values
+ *  \endcode
+ *
+ *  \see \p partition
+ *  \see \p find_if_not
+ */
+template<typename ForwardIterator, typename Predicate>
+  ForwardIterator partition_point(ForwardIterator first,
+                                  ForwardIterator last,
+                                  Predicate pred);
+
+/*! \} // searching
+ */
+
+/*! \addtogroup reductions
+ *  \{
+ *  \addtogroup predicates
+ *  \{
+ */
+
+
+/*! \p is_partitioned returns \c true if the given range 
+ *  is partitioned with respect to a predicate, and \c false otherwise.
+ *
+ *  Specifically, \p is_partitioned returns \c true if <tt>[first, last)</tt>
+ *  is empty of if <tt>[first, last)</tt> is partitioned by \p pred, i.e. if
+ *  all elements that satisfy \p pred appear before those that do not.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the range to consider.
+ *  \param last The end of the range to consider.
+ *  \param pred A function object which decides to which partition each element of the
+ *         range <tt>[first, last)</tt> belongs.
+ *  \return \c true if the range <tt>[first, last)</tt> is partitioned with respect
+ *          to \p pred, or if <tt>[first, last)</tt> is empty. \c false, otherwise.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Input Iterator</a>,
+ *          and \p InputIterator's \c value_type is convertible to \p Predicate's \c argument_type.
+ *  \tparam Predicate is a model of <a href="http://www.sgi.com/tech/stl/Predicate.html">Predicate</a>.
+ *  
+ *  \code
+ *  #include <thrust/partition.h>
+ *  #include <thrust/execution_policy.h>
+ *
+ *  struct is_even
+ *  {
+ *    __host__ __device__
+ *    bool operator()(const int &x)
+ *    {
+ *      return (x % 2) == 0;
+ *    }
+ *  };
+ *  
+ *  ...
+ *
+ *  int A[] = {2, 4, 6, 8, 10, 1, 3, 5, 7, 9};
+ *  int B[] = {1, 2, 3, 4, 5, 6, 7, 8, 9, 10};
+ *
+ *  thrust::is_partitioned(thrust::host, A, A + 10); // returns true
+ *  thrust::is_partitioned(thrust::host, B, B + 10); // returns false
+ *  \endcode
+ *
+ *  \see \p partition
+ */
+template<typename DerivedPolicy, typename InputIterator, typename Predicate>
+  bool is_partitioned(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                      InputIterator first,
+                      InputIterator last,
+                      Predicate pred);
+
+
+/*! \p is_partitioned returns \c true if the given range 
+ *  is partitioned with respect to a predicate, and \c false otherwise.
+ *
+ *  Specifically, \p is_partitioned returns \c true if <tt>[first, last)</tt>
+ *  is empty of if <tt>[first, last)</tt> is partitioned by \p pred, i.e. if
+ *  all elements that satisfy \p pred appear before those that do not.
+ *
+ *  \param first The beginning of the range to consider.
+ *  \param last The end of the range to consider.
+ *  \param pred A function object which decides to which partition each element of the
+ *         range <tt>[first, last)</tt> belongs.
+ *  \return \c true if the range <tt>[first, last)</tt> is partitioned with respect
+ *          to \p pred, or if <tt>[first, last)</tt> is empty. \c false, otherwise.
+ *
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Input Iterator</a>,
+ *          and \p InputIterator's \c value_type is convertible to \p Predicate's \c argument_type.
+ *  \tparam Predicate is a model of <a href="http://www.sgi.com/tech/stl/Predicate.html">Predicate</a>.
+ *  
+ *  \code
+ *  #include <thrust/partition.h>
+ *
+ *  struct is_even
+ *  {
+ *    __host__ __device__
+ *    bool operator()(const int &x)
+ *    {
+ *      return (x % 2) == 0;
+ *    }
+ *  };
+ *  
+ *  ...
+ *
+ *  int A[] = {2, 4, 6, 8, 10, 1, 3, 5, 7, 9};
+ *  int B[] = {1, 2, 3, 4, 5, 6, 7, 8, 9, 10};
+ *
+ *  thrust::is_partitioned(A, A + 10); // returns true
+ *  thrust::is_partitioned(B, B + 10); // returns false
+ *  \endcode
+ *
+ *  \see \p partition
+ */
+template<typename InputIterator, typename Predicate>
+  bool is_partitioned(InputIterator first,
+                      InputIterator last,
+                      Predicate pred);
+
+
+/*! \} // end predicates
+ *  \} // end reductions
+ */
+
+
+} // end thrust
+
+#include <thrust/detail/partition.inl>
+
diff --git a/compat/thrust/random.h b/compat/thrust/random.h
new file mode 100644
index 0000000000..5a2c00da86
--- /dev/null
+++ b/compat/thrust/random.h
@@ -0,0 +1,120 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+/*! \file random.h
+ *  \brief Pseudo-random number generators.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/cstdint.h>
+
+// RNGs
+#include <thrust/random/discard_block_engine.h>
+#include <thrust/random/linear_congruential_engine.h>
+#include <thrust/random/linear_feedback_shift_engine.h>
+#include <thrust/random/subtract_with_carry_engine.h>
+#include <thrust/random/xor_combine_engine.h>
+
+// distributions
+#include <thrust/random/uniform_int_distribution.h>
+#include <thrust/random/uniform_real_distribution.h>
+#include <thrust/random/normal_distribution.h>
+
+namespace thrust
+{
+
+
+/*! \addtogroup random Random Number Generation
+ *  \{
+ */
+
+
+/*! \namespace thrust::random
+ *  \brief \p thrust::random is the namespace which contains random number engine class templates,
+ *  random number engine adaptor class templates, engines with predefined parameters,
+ *  and random number distribution class templates. They are provided in a separate namespace
+ *  for import convenience but are also aliased in the top-level \p thrust namespace for
+ *  easy access.
+ */
+namespace random
+{
+
+/*! \addtogroup predefined_random Random Number Engines with Predefined Parameters
+ *  \ingroup random
+ *  \{
+ */
+
+/*! \typedef ranlux24
+ *  \brief A random number engine with predefined parameters which implements the
+ *         RANLUX level-3 random number generation algorithm.
+ *  \note The 10000th consecutive invocation of a default-constructed object of type \p ranlux24
+ *        shall produce the value \c 9901578 .
+ */
+typedef discard_block_engine<ranlux24_base, 223, 23> ranlux24;
+
+
+/*! \typedef ranlux48
+ *  \brief A random number engine with predefined parameters which implements the
+ *         RANLUX level-4 random number generation algorithm.
+ *  \note The 10000th consecutive invocation of a default-constructed object of type \p ranlux48
+ *        shall produce the value \c 88229545517833 .
+ */
+typedef discard_block_engine<ranlux48_base, 389, 11> ranlux48;
+
+
+/*! \typedef taus88
+ *  \brief A random number engine with predefined parameters which implements
+ *         L'Ecuyer's 1996 three-component Tausworthe random number generator.
+ *
+ *  \note The 10000th consecutive invocation of a default-constructed object of type \p taus88
+ *        shall produce the value \c 3535848941 .
+ */
+typedef xor_combine_engine<
+  linear_feedback_shift_engine<thrust::detail::uint32_t, 32u, 31u, 13u, 12u>,
+  0,
+  xor_combine_engine<
+    linear_feedback_shift_engine<thrust::detail::uint32_t, 32u, 29u,  2u,  4u>, 0,
+    linear_feedback_shift_engine<thrust::detail::uint32_t, 32u, 28u,  3u, 17u>, 0
+  >,
+  0
+> taus88;
+
+/*! \typedef default_random_engine
+ *  \brief An implementation-defined "default" random number engine.
+ *  \note \p default_random_engine is currently an alias for \p minstd_rand, and may change
+ *        in a future version.
+ */
+typedef minstd_rand default_random_engine;
+
+/*! \} // end predefined_random
+ */
+
+} // end random
+
+
+/*! \} // end random
+ */
+
+// import names into thrust::
+using random::ranlux24;
+using random::ranlux48;
+using random::taus88;
+using random::default_random_engine;
+
+} // end thrust
+
diff --git a/compat/thrust/random/detail/discard_block_engine.inl b/compat/thrust/random/detail/discard_block_engine.inl
new file mode 100644
index 0000000000..5f01bd1816
--- /dev/null
+++ b/compat/thrust/random/detail/discard_block_engine.inl
@@ -0,0 +1,201 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/random/discard_block_engine.h>
+
+namespace thrust
+{
+
+namespace random
+{
+
+
+template<typename Engine, size_t p, size_t r>
+  discard_block_engine<Engine,p,r>
+    ::discard_block_engine()
+      : m_e(), m_n(0)
+{}
+
+
+template<typename Engine, size_t p, size_t r>
+  discard_block_engine<Engine,p,r>
+    ::discard_block_engine(result_type s)
+      : m_e(s), m_n(0)
+{}
+
+
+template<typename Engine, size_t p, size_t r>
+  discard_block_engine<Engine,p,r>
+    ::discard_block_engine(const base_type &urng)
+      : m_e(urng), m_n(0)
+{}
+
+
+template<typename Engine, size_t p, size_t r>
+  void discard_block_engine<Engine,p,r>
+    ::seed(void)
+{
+  m_e.seed();
+  m_n = 0;
+}
+
+
+template<typename Engine, size_t p, size_t r>
+  void discard_block_engine<Engine,p,r>
+    ::seed(result_type s)
+{
+  m_e.seed(s);
+  m_n = 0;
+}
+
+
+template<typename Engine, size_t p, size_t r>
+  typename discard_block_engine<Engine,p,r>::result_type
+    discard_block_engine<Engine,p,r>
+      ::operator()(void)
+{
+  if(m_n >= used_block)
+  {
+    m_e.discard(block_size - m_n);
+//    for(; m_n < block_size; ++m_n)
+//      m_e();
+    m_n = 0;
+  }
+
+  ++m_n;
+
+  return m_e();
+}
+
+
+template<typename Engine, size_t p, size_t r>
+  void discard_block_engine<Engine,p,r>
+    ::discard(unsigned long long z)
+{
+  // XXX this should be accelerated
+  for(; z > 0; --z)
+  {
+    this->operator()();
+  } // end for
+}
+
+
+template<typename Engine, size_t p, size_t r>
+  const typename discard_block_engine<Engine,p,r>::base_type &
+    discard_block_engine<Engine,p,r>
+      ::base(void) const
+{
+  return m_e;
+}
+
+
+template<typename Engine, size_t p, size_t r>
+  template<typename CharT, typename Traits>
+    std::basic_ostream<CharT,Traits>& discard_block_engine<Engine,p,r>
+      ::stream_out(std::basic_ostream<CharT,Traits> &os) const
+{
+  typedef std::basic_ostream<CharT,Traits> ostream_type;
+  typedef typename ostream_type::ios_base  ios_base;
+
+  // save old flags & fill character
+  const typename ios_base::fmtflags flags = os.flags();
+  const CharT fill = os.fill();
+
+  const CharT space = os.widen(' ');
+  os.flags(ios_base::dec | ios_base::fixed | ios_base::left);
+  os.fill(space);
+
+  // output the base engine followed by n
+  os << m_e << space << m_n;
+
+  // restore flags & fill character
+  os.flags(flags);
+  os.fill(fill);
+
+  return os;
+}
+
+
+template<typename Engine, size_t p, size_t r>
+  template<typename CharT, typename Traits>
+    std::basic_istream<CharT,Traits>& discard_block_engine<Engine,p,r>
+      ::stream_in(std::basic_istream<CharT,Traits> &is)
+{
+  typedef std::basic_istream<CharT,Traits> istream_type;
+  typedef typename istream_type::ios_base  ios_base;
+
+  // save old flags
+  const typename ios_base::fmtflags flags = is.flags();
+
+  is.flags(ios_base::skipws);
+
+  // input the base engine and then n
+  is >> m_e >> m_n;
+
+  // restore old flags
+  is.flags(flags);
+  return is;
+}
+
+
+template<typename Engine, size_t p, size_t r>
+  bool discard_block_engine<Engine,p,r>
+    ::equal(const discard_block_engine<Engine,p,r> &rhs) const
+{
+  return (m_e == rhs.m_e) && (m_n == rhs.m_n);
+}
+
+
+template<typename Engine, size_t p, size_t r,
+         typename CharT, typename Traits>
+std::basic_ostream<CharT,Traits>&
+operator<<(std::basic_ostream<CharT,Traits> &os,
+           const discard_block_engine<Engine,p,r> &e)
+{
+  return thrust::random::detail::random_core_access::stream_out(os,e);
+}
+
+
+template<typename Engine, size_t p, size_t r,
+         typename CharT, typename Traits>
+std::basic_istream<CharT,Traits>&
+operator>>(std::basic_istream<CharT,Traits> &is,
+           discard_block_engine<Engine,p,r> &e)
+{
+  return thrust::random::detail::random_core_access::stream_in(is,e);
+}
+
+
+template<typename Engine, size_t p, size_t r>
+bool operator==(const discard_block_engine<Engine,p,r> &lhs,
+                const discard_block_engine<Engine,p,r> &rhs)
+{
+  return thrust::random::detail::random_core_access::equal(lhs,rhs);
+}
+
+
+template<typename Engine, size_t p, size_t r>
+bool operator!=(const discard_block_engine<Engine,p,r> &lhs,
+                const discard_block_engine<Engine,p,r> &rhs)
+{
+  return !(lhs == rhs);
+}
+
+
+} // end random
+
+} // end thrust
+
diff --git a/compat/thrust/random/detail/linear_congruential_engine.inl b/compat/thrust/random/detail/linear_congruential_engine.inl
new file mode 100644
index 0000000000..f040563cc2
--- /dev/null
+++ b/compat/thrust/random/detail/linear_congruential_engine.inl
@@ -0,0 +1,163 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/random/linear_congruential_engine.h>
+#include <thrust/random/detail/mod.h>
+#include <thrust/random/detail/random_core_access.h>
+
+namespace thrust
+{
+
+namespace random
+{
+
+
+template<typename UIntType, UIntType a, UIntType c, UIntType m>
+  linear_congruential_engine<UIntType,a,c,m>
+    ::linear_congruential_engine(result_type s)
+{
+  seed(s);
+} // end linear_congruential_engine::linear_congruential_engine()
+
+
+template<typename UIntType, UIntType a, UIntType c, UIntType m>
+  void linear_congruential_engine<UIntType,a,c,m>
+    ::seed(result_type s)
+{
+  if((detail::mod<UIntType, 1, 0, m>(c) == 0) &&
+     (detail::mod<UIntType, 1, 0, m>(s) == 0))
+    m_x = detail::mod<UIntType, 1, 0, m>(1);
+  else
+    m_x = detail::mod<UIntType, 1, 0, m>(s);
+} // end linear_congruential_engine::seed()
+
+
+template<typename UIntType, UIntType a, UIntType c, UIntType m>
+  typename linear_congruential_engine<UIntType,a,c,m>::result_type
+    linear_congruential_engine<UIntType,a,c,m>
+      ::operator()(void)
+{
+  m_x = detail::mod<UIntType,a,c,m>(m_x);
+  return m_x;
+} // end linear_congruential_engine::operator()()
+
+
+template<typename UIntType, UIntType a, UIntType c, UIntType m>
+  void linear_congruential_engine<UIntType,a,c,m>
+    ::discard(unsigned long long z)
+{
+  thrust::random::detail::linear_congruential_engine_discard::discard(*this,z);
+} // end linear_congruential_engine::discard()
+
+
+template<typename UIntType, UIntType a, UIntType c, UIntType m>
+  template<typename CharT, typename Traits>
+    std::basic_ostream<CharT,Traits>& linear_congruential_engine<UIntType,a,c,m>
+      ::stream_out(std::basic_ostream<CharT,Traits> &os) const
+{
+  typedef std::basic_ostream<CharT,Traits> ostream_type;
+  typedef typename ostream_type::ios_base  ios_base;
+
+  // save old flags & fill character
+  const typename ios_base::fmtflags flags = os.flags();
+  const CharT fill = os.fill();
+
+  os.flags(ios_base::dec | ios_base::fixed | ios_base::left);
+  os.fill(os.widen(' '));
+
+  // output one word of state
+  os << m_x;
+
+  // restore flags & fill character
+  os.flags(flags);
+  os.fill(fill);
+
+  return os;
+}
+
+
+template<typename UIntType, UIntType a, UIntType c, UIntType m>
+  template<typename CharT, typename Traits>
+    std::basic_istream<CharT,Traits>& linear_congruential_engine<UIntType,a,c,m>
+      ::stream_in(std::basic_istream<CharT,Traits> &is)
+{
+  typedef std::basic_istream<CharT,Traits> istream_type;
+  typedef typename istream_type::ios_base     ios_base;
+
+  // save old flags
+  const typename ios_base::fmtflags flags = is.flags();
+
+  is.flags(ios_base::dec);
+
+  // input one word of state
+  is >> m_x;
+
+  // restore flags
+  is.flags(flags);
+
+  return is;
+}
+
+
+template<typename UIntType, UIntType a, UIntType c, UIntType m>
+bool linear_congruential_engine<UIntType,a,c,m>
+  ::equal(const linear_congruential_engine<UIntType,a,c,m> &rhs) const
+{
+  return m_x == rhs.m_x;
+}
+
+
+template<typename UIntType_, UIntType_ a_, UIntType_ c_, UIntType_ m_>
+__host__ __device__
+bool operator==(const linear_congruential_engine<UIntType_,a_,c_,m_> &lhs,
+                const linear_congruential_engine<UIntType_,a_,c_,m_> &rhs)
+{
+  return detail::random_core_access::equal(lhs,rhs);
+}
+
+
+template<typename UIntType, UIntType a, UIntType c, UIntType m>
+bool operator!=(const linear_congruential_engine<UIntType,a,c,m> &lhs,
+                const linear_congruential_engine<UIntType,a,c,m> &rhs)
+{
+  return !(lhs == rhs);
+}
+
+
+template<typename UIntType_, UIntType_ a_, UIntType_ c_, UIntType_ m_,
+         typename CharT, typename Traits>
+std::basic_ostream<CharT,Traits>&
+operator<<(std::basic_ostream<CharT,Traits> &os,
+           const linear_congruential_engine<UIntType_,a_,c_,m_> &e)
+{
+  return detail::random_core_access::stream_out(os,e);
+}
+
+
+template<typename UIntType_, UIntType_ a_, UIntType_ c_, UIntType_ m_,
+         typename CharT, typename Traits>
+std::basic_istream<CharT,Traits>&
+operator>>(std::basic_istream<CharT,Traits> &is,
+           linear_congruential_engine<UIntType_,a_,c_,m_> &e)
+{
+  return detail::random_core_access::stream_in(is,e);
+}
+
+
+} // end random
+
+} // end thrust
+
diff --git a/compat/thrust/random/detail/linear_congruential_engine_discard.h b/compat/thrust/random/detail/linear_congruential_engine_discard.h
new file mode 100644
index 0000000000..f4ec23364e
--- /dev/null
+++ b/compat/thrust/random/detail/linear_congruential_engine_discard.h
@@ -0,0 +1,107 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/cstdint.h>
+#include <thrust/random/detail/mod.h>
+
+namespace thrust
+{
+
+namespace random
+{
+
+namespace detail
+{
+
+
+template<typename UIntType, UIntType a, unsigned long long c, UIntType m>
+  struct linear_congruential_engine_discard_implementation
+{
+  __host__ __device__
+  static void discard(UIntType &state, unsigned long long z)
+  {
+    for(; z > 0; --z)
+    {
+      state = detail::mod<UIntType,a,c,m>(state);
+    }
+  }
+}; // end linear_congruential_engine_discard
+
+
+// specialize for small integers and c == 0
+// XXX figure out a robust implemenation of this for any unsigned integer type later
+template<thrust::detail::uint32_t a, thrust::detail::uint32_t m>
+  struct linear_congruential_engine_discard_implementation<thrust::detail::uint32_t,a,0,m>
+{
+  __host__ __device__
+  static void discard(thrust::detail::uint32_t &state, unsigned long long z)
+  {
+    const thrust::detail::uint32_t modulus = m;
+
+    // XXX we need to use unsigned long long here or we will encounter overflow in the
+    //     multiplies below
+    //     figure out a robust implementation of this later
+    unsigned long long multiplier = a;
+    unsigned long long multiplier_to_z = 1;
+    
+    // see http://en.wikipedia.org/wiki/Modular_exponentiation
+    while(z > 0)
+    {
+      if(z & 1)
+      {
+        // multiply in this bit's contribution while using modulus to keep result small
+        multiplier_to_z = (multiplier_to_z * multiplier) % modulus;
+      }
+
+      // move to the next bit of the exponent, square (and mod) the base accordingly
+      z >>= 1;
+      multiplier = (multiplier * multiplier) % modulus;
+    }
+
+    state = static_cast<thrust::detail::uint32_t>((multiplier_to_z * state) % modulus);
+  }
+}; // end linear_congruential_engine_discard
+
+
+struct linear_congruential_engine_discard
+{
+  template<typename LinearCongruentialEngine>
+  __host__ __device__
+  static void discard(LinearCongruentialEngine &lcg, unsigned long long z)
+  {
+    typedef typename LinearCongruentialEngine::result_type result_type;
+    const result_type c = LinearCongruentialEngine::increment;
+    const result_type a = LinearCongruentialEngine::multiplier;
+    const result_type m = LinearCongruentialEngine::modulus;
+    
+    // XXX WAR unused variable warnings
+    (void) c;
+    (void) a;
+    (void) m;
+
+    linear_congruential_engine_discard_implementation<result_type,a,c,m>::discard(lcg.m_x, z);
+  }
+}; // end linear_congruential_engine_discard
+
+
+} // end detail
+
+} // end random
+
+} // end thrust
+
diff --git a/compat/thrust/random/detail/linear_feedback_shift_engine.inl b/compat/thrust/random/detail/linear_feedback_shift_engine.inl
new file mode 100644
index 0000000000..4e8dad5140
--- /dev/null
+++ b/compat/thrust/random/detail/linear_feedback_shift_engine.inl
@@ -0,0 +1,158 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/random/linear_feedback_shift_engine.h>
+
+namespace thrust
+{
+
+namespace random
+{
+
+template<typename UIntType, size_t w, size_t k, size_t q, size_t s>
+  linear_feedback_shift_engine<UIntType,w,k,q,s>
+    ::linear_feedback_shift_engine(result_type value)
+{
+  seed(value);
+} // end linear_feedback_shift_engine::linear_feedback_shift_engine()
+
+template<typename UIntType, size_t w, size_t k, size_t q, size_t s>
+  void linear_feedback_shift_engine<UIntType,w,k,q,s>
+    ::seed(result_type value)
+{
+  m_value = value;
+} // end linear_feedback_shift_engine::seed()
+
+template<typename UIntType, size_t w, size_t k, size_t q, size_t s>
+  typename linear_feedback_shift_engine<UIntType,w,k,q,s>::result_type
+    linear_feedback_shift_engine<UIntType,w,k,q,s>
+      ::operator()(void)
+{
+  const UIntType b = (((m_value << q) ^ m_value) & wordmask) >> (k-s);
+  const UIntType mask = ( (~static_cast<UIntType>(0)) << (w-k) ) & wordmask;
+  m_value = ((m_value & mask) << s) ^ b;
+  return m_value;
+} // end linear_feedback_shift_engine::operator()()
+
+
+template<typename UIntType, size_t w, size_t k, size_t q, size_t s>
+  void linear_feedback_shift_engine<UIntType,w,k,q,s>
+    ::discard(unsigned long long z)
+{
+  for(; z > 0; --z)
+  {
+    this->operator()();
+  } // end for
+} // end linear_feedback_shift_engine::discard()
+
+
+template<typename UIntType, size_t w, size_t k, size_t q, size_t s>
+  template<typename CharT, typename Traits>
+    std::basic_ostream<CharT,Traits>& linear_feedback_shift_engine<UIntType,w,k,q,s>
+      ::stream_out(std::basic_ostream<CharT,Traits> &os) const
+{
+  typedef std::basic_ostream<CharT,Traits> ostream_type;
+  typedef typename ostream_type::ios_base  ios_base;
+
+  // save old flags & fill character
+  const typename ios_base::fmtflags flags = os.flags();
+  const CharT fill = os.fill();
+
+  os.flags(ios_base::dec | ios_base::fixed | ios_base::left);
+  os.fill(os.widen(' '));
+
+  // output one word of state
+  os << m_value;
+
+  // restore flags & fill character
+  os.flags(flags);
+  os.fill(fill);
+
+  return os;
+}
+
+
+template<typename UIntType, size_t w, size_t k, size_t q, size_t s>
+  template<typename CharT, typename Traits>
+    std::basic_istream<CharT,Traits>& linear_feedback_shift_engine<UIntType,w,k,q,s>
+      ::stream_in(std::basic_istream<CharT,Traits> &is)
+{
+  typedef std::basic_istream<CharT,Traits> istream_type;
+  typedef typename istream_type::ios_base     ios_base;
+
+  // save old flags
+  const typename ios_base::fmtflags flags = is.flags();
+
+  is.flags(ios_base::skipws);
+
+  // input one word of state
+  is >> m_value;
+
+  // restore flags
+  is.flags(flags);
+
+  return is;
+}
+
+
+template<typename UIntType, size_t w, size_t k, size_t q, size_t s>
+  bool linear_feedback_shift_engine<UIntType,w,k,q,s>
+    ::equal(const linear_feedback_shift_engine<UIntType,w,k,q,s> &rhs) const
+{
+  return m_value == rhs.m_value;
+}
+
+
+template<typename UIntType, size_t w, size_t k, size_t q, size_t s>
+bool operator==(const linear_feedback_shift_engine<UIntType,w,k,q,s> &lhs,
+                const linear_feedback_shift_engine<UIntType,w,k,q,s> &rhs)
+{
+  return thrust::random::detail::random_core_access::equal(lhs,rhs);
+}
+
+
+template<typename UIntType, size_t w, size_t k, size_t q, size_t s>
+bool operator!=(const linear_feedback_shift_engine<UIntType,w,k,q,s> &lhs,
+                const linear_feedback_shift_engine<UIntType,w,k,q,s> &rhs)
+{
+  return !(lhs == rhs);
+}
+
+
+template<typename UIntType_, size_t w_, size_t k_, size_t q_, size_t s_,
+         typename CharT, typename Traits>
+std::basic_ostream<CharT,Traits>&
+operator<<(std::basic_ostream<CharT,Traits> &os,
+           const linear_feedback_shift_engine<UIntType_,w_,k_,q_,s_> &e)
+{
+  return thrust::random::detail::random_core_access::stream_out(os,e);
+}
+
+
+template<typename UIntType_, size_t w_, size_t k_, size_t q_, size_t s_,
+         typename CharT, typename Traits>
+std::basic_istream<CharT,Traits>&
+operator>>(std::basic_istream<CharT,Traits> &is,
+           linear_feedback_shift_engine<UIntType_,w_,k_,q_,s_> &e)
+{
+  return thrust::random::detail::random_core_access::stream_in(is,e);
+}
+
+
+} // end random
+
+} // end thrust
+
diff --git a/compat/thrust/random/detail/linear_feedback_shift_engine_wordmask.h b/compat/thrust/random/detail/linear_feedback_shift_engine_wordmask.h
new file mode 100644
index 0000000000..ed9e51e925
--- /dev/null
+++ b/compat/thrust/random/detail/linear_feedback_shift_engine_wordmask.h
@@ -0,0 +1,47 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+namespace thrust
+{
+
+namespace random
+{
+
+namespace detail
+{
+
+template<typename T, int w, int i = w-1>
+  struct linear_feedback_shift_engine_wordmask
+{
+  static const T value =
+    (T(1u) << i) |
+    linear_feedback_shift_engine_wordmask<T, w, i-1>::value;
+}; // end linear_feedback_shift_engine_wordmask
+
+template<typename T, int w>
+  struct linear_feedback_shift_engine_wordmask<T, w, 0>
+{
+  static const T value = 0;
+}; // end linear_feedback_shift_engine_wordmask
+
+} // end detail
+
+} // end random
+
+} // end thrust
+
diff --git a/compat/thrust/random/detail/mod.h b/compat/thrust/random/detail/mod.h
new file mode 100644
index 0000000000..ceb2191552
--- /dev/null
+++ b/compat/thrust/random/detail/mod.h
@@ -0,0 +1,97 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+namespace thrust
+{
+
+namespace random
+{
+
+namespace detail
+{
+
+template<typename T, T a, T c, T m, bool = (m == 0)>
+  struct static_mod
+{
+  static const T q = m / a;
+  static const T r = m % a;
+
+  __host__ __device__
+  T operator()(T x) const
+  {
+    if(a == 1)
+    {
+      x %= m;
+    }
+    else
+    {
+      T t1 = a * (x % q);
+      T t2 = r * (x / q);
+      if(t1 >= t2)
+      {
+        x = t1 - t2;
+      }
+      else
+      {
+        x = m - t2 + t1;
+      }
+    }
+
+    if(c != 0)
+    {
+      const T d = m - x;
+      if(d > c)
+      {
+        x += c;
+      }
+      else
+      {
+        x = c - d;
+      }
+    }
+
+    return x;
+  }
+}; // end static_mod
+
+
+// Rely on machine overflow handling
+template<typename T, T a, T c, T m>
+  struct static_mod<T,a,c,m,true>
+{
+  __host__ __device__
+  T operator()(T x) const
+  {
+    return a * x + c;
+  }
+}; // end static_mod
+
+template<typename T, T a, T c, T m>
+__host__ __device__
+  T mod(T x)
+{
+  static_mod<T,a,c,m> f;
+  return f(x);
+} // end static_mod
+
+} // end detail
+
+} // end random
+
+} // end thrust
+
diff --git a/compat/thrust/random/detail/normal_distribution.inl b/compat/thrust/random/detail/normal_distribution.inl
new file mode 100644
index 0000000000..1bb55d75b2
--- /dev/null
+++ b/compat/thrust/random/detail/normal_distribution.inl
@@ -0,0 +1,241 @@
+/*
+ *
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/random/normal_distribution.h>
+#include <thrust/random/uniform_real_distribution.h>
+#include <thrust/detail/cstdint.h>
+#include <thrust/detail/integer_traits.h>
+
+// for floating point infinity
+#if THRUST_DEVICE_COMPILER == THRUST_DEVICE_COMPILER_NVCC
+#include <math_constants.h>
+#else
+#include <limits>
+#endif
+
+namespace thrust
+{
+
+namespace random
+{
+
+
+template<typename RealType>
+  normal_distribution<RealType>
+    ::normal_distribution(RealType a, RealType b)
+      :super_t(),m_param(a,b)
+{
+} // end normal_distribution::normal_distribution()
+
+
+template<typename RealType>
+  normal_distribution<RealType>
+    ::normal_distribution(const param_type &parm)
+      :super_t(),m_param(parm)
+{
+} // end normal_distribution::normal_distribution()
+
+
+template<typename RealType>
+  void normal_distribution<RealType>
+    ::reset(void)
+{
+  super_t::reset();
+} // end normal_distribution::reset()
+
+
+template<typename RealType>
+  template<typename UniformRandomNumberGenerator>
+    typename normal_distribution<RealType>::result_type
+      normal_distribution<RealType>
+        ::operator()(UniformRandomNumberGenerator &urng)
+{
+  return operator()(urng, m_param);
+} // end normal_distribution::operator()()
+
+
+template<typename RealType>
+  template<typename UniformRandomNumberGenerator>
+    typename normal_distribution<RealType>::result_type
+      normal_distribution<RealType>
+        ::operator()(UniformRandomNumberGenerator &urng,
+                     const param_type &parm)
+{
+  return super_t::sample(urng, parm.first, parm.second);
+} // end normal_distribution::operator()()
+
+
+template<typename RealType>
+  typename normal_distribution<RealType>::param_type
+    normal_distribution<RealType>
+      ::param(void) const
+{
+  return m_param;
+} // end normal_distribution::param()
+
+
+template<typename RealType>
+  void normal_distribution<RealType>
+    ::param(const param_type &parm)
+{
+  m_param = parm;
+} // end normal_distribution::param()
+
+
+template<typename RealType>
+  typename normal_distribution<RealType>::result_type
+    normal_distribution<RealType>
+      ::min THRUST_PREVENT_MACRO_SUBSTITUTION (void) const
+{
+  return -this->max();
+} // end normal_distribution::min()
+
+
+template<typename RealType>
+  typename normal_distribution<RealType>::result_type
+    normal_distribution<RealType>
+      ::max THRUST_PREVENT_MACRO_SUBSTITUTION (void) const
+{
+  // XXX this solution is pretty terrible
+  // we can't use numeric_traits<RealType>::max because nvcc will
+  // complain that it is a __host__ function
+  union
+  {
+    thrust::detail::uint32_t inf_as_int;
+    float result;
+  } hack;
+
+  hack.inf_as_int = 0x7f800000u;
+
+  return hack.result;
+} // end normal_distribution::max()
+
+
+template<typename RealType>
+  typename normal_distribution<RealType>::result_type
+    normal_distribution<RealType>
+      ::mean(void) const
+{
+  return m_param.first;
+} // end normal_distribution::mean()
+
+
+template<typename RealType>
+  typename normal_distribution<RealType>::result_type
+    normal_distribution<RealType>
+      ::stddev(void) const
+{
+  return m_param.second;
+} // end normal_distribution::stddev()
+
+
+template<typename RealType>
+  bool normal_distribution<RealType>
+    ::equal(const normal_distribution &rhs) const
+{
+  return m_param == rhs.param();
+}
+
+
+template<typename RealType>
+  template<typename CharT, typename Traits>
+    std::basic_ostream<CharT,Traits>&
+      normal_distribution<RealType>
+        ::stream_out(std::basic_ostream<CharT,Traits> &os) const
+{
+  typedef std::basic_ostream<CharT,Traits> ostream_type;
+  typedef typename ostream_type::ios_base  ios_base;
+
+  // save old flags and fill character
+  const typename ios_base::fmtflags flags = os.flags();
+  const CharT fill = os.fill();
+
+  const CharT space = os.widen(' ');
+  os.flags(ios_base::dec | ios_base::fixed | ios_base::left);
+  os.fill(space);
+
+  os << mean() << space << stddev();
+
+  // restore old flags and fill character
+  os.flags(flags);
+  os.fill(fill);
+  return os;
+}
+
+
+template<typename RealType>
+  template<typename CharT, typename Traits>
+    std::basic_istream<CharT,Traits>&
+      normal_distribution<RealType>
+        ::stream_in(std::basic_istream<CharT,Traits> &is)
+{
+  typedef std::basic_istream<CharT,Traits> istream_type;
+  typedef typename istream_type::ios_base  ios_base;
+
+  // save old flags
+  const typename ios_base::fmtflags flags = is.flags();
+
+  is.flags(ios_base::skipws);
+
+  is >> m_param.first >> m_param.second;
+
+  // restore old flags
+  is.flags(flags);
+  return is;
+}
+
+
+template<typename RealType>
+bool operator==(const normal_distribution<RealType> &lhs,
+                const normal_distribution<RealType> &rhs)
+{
+  return thrust::random::detail::random_core_access::equal(lhs,rhs);
+}
+
+
+template<typename RealType>
+bool operator!=(const normal_distribution<RealType> &lhs,
+                const normal_distribution<RealType> &rhs)
+{
+  return !(lhs == rhs);
+}
+
+
+template<typename RealType,
+         typename CharT, typename Traits>
+std::basic_ostream<CharT,Traits>&
+operator<<(std::basic_ostream<CharT,Traits> &os,
+           const normal_distribution<RealType> &d)
+{
+  return thrust::random::detail::random_core_access::stream_out(os,d);
+}
+
+
+template<typename RealType,
+         typename CharT, typename Traits>
+std::basic_istream<CharT,Traits>&
+operator>>(std::basic_istream<CharT,Traits> &is,
+           normal_distribution<RealType> &d)
+{
+  return thrust::random::detail::random_core_access::stream_in(is,d);
+}
+
+
+} // end random
+
+} // end thrust
+
diff --git a/compat/thrust/random/detail/normal_distribution_base.h b/compat/thrust/random/detail/normal_distribution_base.h
new file mode 100644
index 0000000000..d9166112ad
--- /dev/null
+++ b/compat/thrust/random/detail/normal_distribution_base.h
@@ -0,0 +1,149 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+/*
+ * Copyright Jens Maurer 2000-2001
+ * Distributed under the Boost Software License, Version 1.0. (See
+ * accompanying file LICENSE_1_0.txt or copy at
+ * http://www.boost.org/LICENSE_1_0.txt)
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/pair.h>
+#include <thrust/random/uniform_real_distribution.h>
+#include <limits>
+#include <cmath>
+
+namespace thrust
+{
+namespace random
+{
+namespace detail
+{
+
+// this version samples the normal distribution directly 
+// and uses the non-standard math function erfcinv
+template<typename RealType>
+  class normal_distribution_nvcc
+{
+  protected:
+    template<typename UniformRandomNumberGenerator>
+    __host__ __device__
+    RealType sample(UniformRandomNumberGenerator &urng, const RealType mean, const RealType stddev)
+    {
+      typedef typename UniformRandomNumberGenerator::result_type uint_type;
+      const uint_type urng_range = UniformRandomNumberGenerator::max - UniformRandomNumberGenerator::min;
+
+      // Constants for conversion
+      const RealType S1 = static_cast<RealType>(1) / urng_range;
+      const RealType S2 = S1 / 2;
+
+      RealType S3 = static_cast<RealType>(-1.4142135623730950488016887242097); // -sqrt(2)
+      
+      // Get the integer value
+      uint_type u = urng() - UniformRandomNumberGenerator::min;
+
+      // Ensure the conversion to float will give a value in the range [0,0.5)
+      if(u > (urng_range / 2))
+      {
+        u = urng_range - u;
+        S3 = -S3;
+      }
+
+      // Convert to floating point in [0,0.5)
+      RealType p = u*S1 + S2;
+
+      // Apply inverse error function
+      return mean + stddev * S3 * erfcinv(2 * p);
+    }
+
+    // no-op
+    __host__ __device__
+    void reset() {}
+};
+
+// this version samples the normal distribution using 
+// Marsaglia's "polar method"
+template<typename RealType>
+  class normal_distribution_portable
+{
+  protected:
+    normal_distribution_portable()
+      : m_valid(false)
+    {}
+
+    normal_distribution_portable(const normal_distribution_portable &other)
+      : m_valid(other.m_valid)
+    {}
+
+    void reset()
+    {
+      m_valid = false;
+    }
+
+    // note that we promise to call this member function with the same mean and stddev
+    template<typename UniformRandomNumberGenerator>
+    __host__ __device__
+    RealType sample(UniformRandomNumberGenerator &urng, const RealType mean, const RealType stddev)
+    {
+      // implementation from Boost
+      // allow for Koenig lookup
+      using std::sqrt; using std::log; using std::sin; using std::cos;
+
+      if(!m_valid)
+      {
+        uniform_real_distribution<RealType> u01;
+        m_r1 = u01(urng);
+        m_r2 = u01(urng);
+        m_cached_rho = sqrt(-RealType(2) * log(RealType(1)-m_r2));
+
+        m_valid = true;
+      }
+      else
+      {
+        m_valid = false;
+      }
+
+      const RealType pi = RealType(3.14159265358979323846);
+
+      RealType result = m_cached_rho * (m_valid ?
+                          cos(RealType(2)*pi*m_r1) :
+                          sin(RealType(2)*pi*m_r1));
+
+      return result;
+    }
+
+  private:
+    RealType m_r1, m_r2, m_cached_rho;
+    bool m_valid;
+};
+
+template<typename RealType>
+  struct normal_distribution_base
+{
+#if THRUST_DEVICE_COMPILER == THRUST_DEVICE_COMPILER_NVCC
+  typedef normal_distribution_nvcc<RealType> type;
+#else
+  typedef normal_distribution_portable<RealType> type;
+#endif
+};
+
+} // end detail
+} // end random
+} // end thrust
+
diff --git a/compat/thrust/random/detail/random_core_access.h b/compat/thrust/random/detail/random_core_access.h
new file mode 100644
index 0000000000..81f58e2207
--- /dev/null
+++ b/compat/thrust/random/detail/random_core_access.h
@@ -0,0 +1,57 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+namespace thrust
+{
+
+namespace random
+{
+
+namespace detail
+{
+
+struct random_core_access
+{
+
+template<typename OStream, typename EngineOrDistribution>
+static OStream &stream_out(OStream &os, const EngineOrDistribution &x)
+{
+  return x.stream_out(os);
+}
+
+template<typename IStream, typename EngineOrDistribution>
+static IStream &stream_in(IStream &is, EngineOrDistribution &x)
+{
+  return x.stream_in(is);
+}
+
+template<typename EngineOrDistribution>
+__host__ __device__
+static bool equal(const EngineOrDistribution &lhs, const EngineOrDistribution &rhs)
+{
+  return lhs.equal(rhs);
+}
+
+}; // end random_core_access
+
+} // end detail
+
+} // end random
+
+} // end thrust
+
diff --git a/compat/thrust/random/detail/subtract_with_carry_engine.inl b/compat/thrust/random/detail/subtract_with_carry_engine.inl
new file mode 100644
index 0000000000..a58b2665b2
--- /dev/null
+++ b/compat/thrust/random/detail/subtract_with_carry_engine.inl
@@ -0,0 +1,203 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/random/linear_congruential_engine.h>
+#include <thrust/random/subtract_with_carry_engine.h>
+#include <thrust/random/detail/mod.h>
+#include <thrust/random/detail/random_core_access.h>
+
+namespace thrust
+{
+
+namespace random
+{
+
+
+template<typename UIntType, size_t w, size_t s, size_t r>
+  subtract_with_carry_engine<UIntType,w,s,r>
+    ::subtract_with_carry_engine(result_type value)
+{
+  seed(value);
+} // end subtract_with_carry_engine::subtract_with_carry_engine()
+
+
+template<typename UIntType, size_t w, size_t s, size_t r>
+  void subtract_with_carry_engine<UIntType,w,s,r>
+    ::seed(result_type value)
+{
+  thrust::random::linear_congruential_engine<result_type,
+    40014u, 0u, 2147483563u> e(value == 0u ? default_seed : value);
+
+  // initialize state
+  for(size_t i = 0; i < long_lag; ++i)
+  {
+    m_x[i] = detail::mod<UIntType, 1, 0, modulus>(e());
+  } // end for i
+
+  m_carry = (m_x[long_lag-1] == 0);
+  m_k = 0;
+} // end subtract_with_carry_engine::seed()
+
+
+template<typename UIntType, size_t w, size_t s, size_t r>
+  typename subtract_with_carry_engine<UIntType,w,s,r>::result_type
+    subtract_with_carry_engine<UIntType,w,s,r>
+      ::operator()(void)
+{
+  // XXX we probably need to cache these m_x[m_k] in a register
+  //     maybe we need to cache the use of all member variables
+  int short_index = m_k - short_lag;
+  if(short_index < 0)
+    short_index += long_lag;
+  result_type xi;
+  if (m_x[short_index] >= m_x[m_k] + m_carry)
+  {
+    // x(n) >= 0
+    xi =  m_x[short_index] - m_x[m_k] - m_carry;
+    m_carry = 0;
+  }
+  else
+  {
+    // x(n) < 0
+    xi = modulus - m_x[m_k] - m_carry + m_x[short_index];
+    m_carry = 1;
+  }
+  m_x[m_k] = xi;
+  ++m_k;
+  if(m_k >= long_lag)
+    m_k = 0;
+  return xi;
+} // end subtract_with_carry_engine::operator()()
+
+
+template<typename UIntType, size_t w, size_t s, size_t r>
+  void subtract_with_carry_engine<UIntType,w,s,r>
+    ::discard(unsigned long long z)
+{
+  for(; z > 0; --z)
+  {
+    this->operator()();
+  } // end for
+} // end subtract_with_carry_engine::discard()
+
+
+template<typename UIntType, size_t w, size_t s, size_t r>
+  template<typename CharT, typename Traits>
+    std::basic_ostream<CharT,Traits>& subtract_with_carry_engine<UIntType,w,s,r>
+      ::stream_out(std::basic_ostream<CharT,Traits> &os) const
+{
+  typedef std::basic_ostream<CharT,Traits> ostream_type;
+  typedef typename ostream_type::ios_base     ios_base;
+                  
+  const typename ios_base::fmtflags flags = os.flags();
+  const CharT fill  = os.fill();
+  const CharT space = os.widen(' ');
+  os.flags(ios_base::dec | ios_base::fixed | ios_base::left);
+  os.fill(space);
+
+  const UIntType long_lag = r;
+                                                          
+  for(size_t i = 0; i < r; ++i)
+    os << m_x[(i + m_k) % long_lag] << space;
+  os << m_carry;
+                                                                          
+  os.flags(flags);
+  os.fill(fill);
+  return os;
+}
+
+
+template<typename UIntType, size_t w, size_t s, size_t r>
+  template<typename CharType, typename Traits>
+    std::basic_istream<CharType,Traits>& subtract_with_carry_engine<UIntType,w,s,r>
+      ::stream_in(std::basic_istream<CharType,Traits> &is)
+{
+  typedef std::basic_istream<CharType,Traits> istream_type;
+  typedef typename istream_type::ios_base     ios_base;
+
+  const typename ios_base::fmtflags flags = is.flags();
+  is.flags(ios_base::dec | ios_base::skipws);
+
+  for(size_t i = 0; i < r; ++i)
+    is >> m_x[i];
+  is >> m_carry;
+
+  m_k = 0;
+
+  is.flags(flags);
+  return is;
+}
+
+
+template<typename UIntType, size_t w, size_t s, size_t r>
+  bool subtract_with_carry_engine<UIntType,w,s,r>
+    ::equal(const subtract_with_carry_engine<UIntType,w,s,r> &rhs) const
+{
+  const UIntType long_lag = r;
+
+  bool result = true;
+  for(size_t i = 0; i < r; ++i)
+  {
+    result &= (m_x[(i + m_k) % long_lag] == rhs.m_x[(i + rhs.m_k) % long_lag]);
+  }
+
+  // XXX not sure if this last check is necessary
+  result &= (m_carry == rhs.m_carry);
+
+  return result;
+}
+
+
+template<typename UIntType, size_t w, size_t s, size_t r,
+         typename CharT, typename Traits>
+  std::basic_ostream<CharT,Traits>&
+    operator<<(std::basic_ostream<CharT,Traits> &os,
+               const subtract_with_carry_engine<UIntType,w,s,r> &e)
+{
+  return thrust::random::detail::random_core_access::stream_out(os,e);
+}
+
+
+template<typename UIntType, size_t w, size_t s, size_t r,
+         typename CharType, typename Traits>
+  std::basic_istream<CharType,Traits>&
+    operator>>(std::basic_istream<CharType,Traits> &is,
+               subtract_with_carry_engine<UIntType,w,s,r> &e)
+{
+  return thrust::random::detail::random_core_access::stream_in(is,e);
+}
+
+
+template<typename UIntType, size_t w, size_t s, size_t r>
+  bool operator==(const subtract_with_carry_engine<UIntType,w,s,r> &lhs,
+                  const subtract_with_carry_engine<UIntType,w,s,r> &rhs)
+{
+  return thrust::random::detail::random_core_access::equal(lhs,rhs);
+}
+
+
+template<typename UIntType, size_t w, size_t s, size_t r>
+  bool operator!=(const subtract_with_carry_engine<UIntType,w,s,r> &lhs,
+                  const subtract_with_carry_engine<UIntType,w,s,r> &rhs)
+{
+  return !(lhs == rhs);
+}
+
+
+} // end random
+
+} // end thrust
+
diff --git a/compat/thrust/random/detail/uniform_int_distribution.inl b/compat/thrust/random/detail/uniform_int_distribution.inl
new file mode 100644
index 0000000000..e92754c5e7
--- /dev/null
+++ b/compat/thrust/random/detail/uniform_int_distribution.inl
@@ -0,0 +1,232 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/random/uniform_int_distribution.h>
+#include <thrust/random/uniform_real_distribution.h>
+#include <thrust/detail/type_traits.h>
+
+namespace thrust
+{
+
+namespace random
+{
+
+
+template<typename IntType>
+  uniform_int_distribution<IntType>
+    ::uniform_int_distribution(IntType a, IntType b)
+      :m_param(a,b)
+{
+} // end uniform_int_distribution::uniform_int_distribution()
+
+
+template<typename IntType>
+  uniform_int_distribution<IntType>
+    ::uniform_int_distribution(const param_type &parm)
+      :m_param(parm)
+{
+} // end uniform_int_distribution::uniform_int_distribution()
+
+
+template<typename IntType>
+  void uniform_int_distribution<IntType>
+    ::reset(void)
+{
+} // end uniform_int_distribution::reset()
+
+
+template<typename IntType>
+  template<typename UniformRandomNumberGenerator>
+    typename uniform_int_distribution<IntType>::result_type
+      uniform_int_distribution<IntType>
+        ::operator()(UniformRandomNumberGenerator &urng)
+{
+  return operator()(urng, m_param);
+} // end uniform_int_distribution::operator()()
+
+
+template<typename IntType>
+  template<typename UniformRandomNumberGenerator>
+    typename uniform_int_distribution<IntType>::result_type
+      uniform_int_distribution<IntType>
+        ::operator()(UniformRandomNumberGenerator &urng, const param_type &parm)
+{
+  // XXX this implementation is somewhat hacky and will skip
+  //     values if the range of the RNG is smaller than the range of the distribution
+  //     we should improve this implementation in a later version
+
+  typedef typename thrust::detail::largest_available_float::type float_type;
+
+  const float_type real_min(parm.first);
+  const float_type real_max(parm.second);
+
+  // add one to the right end of the interval because it is half-open
+  // XXX adding 1.0 to a potentially large floating point number seems like a bad idea
+  uniform_real_distribution<float_type> real_dist(real_min, real_max + float_type(1));
+
+  return static_cast<result_type>(real_dist(urng));
+} // end uniform_int_distribution::operator()()
+
+
+template<typename IntType>
+  typename uniform_int_distribution<IntType>::result_type
+    uniform_int_distribution<IntType>
+      ::a(void) const
+{
+  return m_param.first;
+} // end uniform_int_distribution<IntType>::a()
+
+
+template<typename IntType>
+  typename uniform_int_distribution<IntType>::result_type
+    uniform_int_distribution<IntType>
+      ::b(void) const
+{
+  return m_param.second;
+} // end uniform_int_distribution::b()
+
+
+template<typename IntType>
+  typename uniform_int_distribution<IntType>::param_type
+    uniform_int_distribution<IntType>
+      ::param(void) const
+{
+  return m_param;
+} // end uniform_int_distribution::param()
+
+
+template<typename IntType>
+  void uniform_int_distribution<IntType>
+    ::param(const param_type &parm)
+{
+  m_param = parm;
+} // end uniform_int_distribution::param()
+
+
+template<typename IntType>
+  typename uniform_int_distribution<IntType>::result_type
+    uniform_int_distribution<IntType>
+      ::min THRUST_PREVENT_MACRO_SUBSTITUTION (void) const
+{
+  return a();
+} // end uniform_int_distribution::min()
+
+
+template<typename IntType>
+  typename uniform_int_distribution<IntType>::result_type
+    uniform_int_distribution<IntType>
+      ::max THRUST_PREVENT_MACRO_SUBSTITUTION (void) const
+{
+  return b();
+} // end uniform_int_distribution::max()
+
+
+template<typename IntType>
+  bool uniform_int_distribution<IntType>
+    ::equal(const uniform_int_distribution &rhs) const
+{
+  return param() == rhs.param();
+}
+
+
+template<typename IntType>
+  template<typename CharT, typename Traits>
+    std::basic_ostream<CharT,Traits>&
+      uniform_int_distribution<IntType>
+        ::stream_out(std::basic_ostream<CharT,Traits> &os) const
+{
+  typedef std::basic_ostream<CharT,Traits> ostream_type;
+  typedef typename ostream_type::ios_base  ios_base;
+
+  // save old flags and fill character
+  const typename ios_base::fmtflags flags = os.flags();
+  const CharT fill = os.fill();
+
+  const CharT space = os.widen(' ');
+  os.flags(ios_base::dec | ios_base::fixed | ios_base::left);
+  os.fill(space);
+
+  os << a() << space << b();
+
+  // restore old flags and fill character
+  os.flags(flags);
+  os.fill(fill);
+  return os;
+}
+
+
+template<typename IntType>
+  template<typename CharT, typename Traits>
+    std::basic_istream<CharT,Traits>&
+      uniform_int_distribution<IntType>
+        ::stream_in(std::basic_istream<CharT,Traits> &is)
+{
+  typedef std::basic_istream<CharT,Traits> istream_type;
+  typedef typename istream_type::ios_base  ios_base;
+
+  // save old flags
+  const typename ios_base::fmtflags flags = is.flags();
+
+  is.flags(ios_base::skipws);
+
+  is >> m_param.first >> m_param.second;
+
+  // restore old flags
+  is.flags(flags);
+  return is;
+}
+
+
+template<typename IntType>
+bool operator==(const uniform_int_distribution<IntType> &lhs,
+                const uniform_int_distribution<IntType> &rhs)
+{
+  return thrust::random::detail::random_core_access::equal(lhs,rhs);
+}
+
+
+template<typename IntType>
+bool operator!=(const uniform_int_distribution<IntType> &lhs,
+                const uniform_int_distribution<IntType> &rhs)
+{
+  return !(lhs == rhs);
+}
+
+
+template<typename IntType,
+         typename CharT, typename Traits>
+std::basic_ostream<CharT,Traits>&
+operator<<(std::basic_ostream<CharT,Traits> &os,
+           const uniform_int_distribution<IntType> &d)
+{
+  return thrust::random::detail::random_core_access::stream_out(os,d);
+}
+
+
+template<typename IntType,
+         typename CharT, typename Traits>
+std::basic_istream<CharT,Traits>&
+operator>>(std::basic_istream<CharT,Traits> &is,
+           uniform_int_distribution<IntType> &d)
+{
+  return thrust::random::detail::random_core_access::stream_in(is,d);
+}
+
+
+} // end random
+
+} // end thrust
+
diff --git a/compat/thrust/random/detail/uniform_real_distribution.inl b/compat/thrust/random/detail/uniform_real_distribution.inl
new file mode 100644
index 0000000000..6f6d6b57b5
--- /dev/null
+++ b/compat/thrust/random/detail/uniform_real_distribution.inl
@@ -0,0 +1,217 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/random/uniform_real_distribution.h>
+
+namespace thrust
+{
+
+namespace random
+{
+
+
+template<typename RealType>
+  uniform_real_distribution<RealType>
+    ::uniform_real_distribution(RealType a, RealType b)
+      :m_param(a,b)
+{
+} // end uniform_real_distribution::uniform_real_distribution()
+
+template<typename RealType>
+  uniform_real_distribution<RealType>
+    ::uniform_real_distribution(const param_type &parm)
+      :m_param(parm)
+{
+} // end uniform_real_distribution::uniform_real_distribution()
+
+template<typename RealType>
+  void uniform_real_distribution<RealType>
+    ::reset(void)
+{
+} // end uniform_real_distribution::reset()
+
+template<typename RealType>
+  template<typename UniformRandomNumberGenerator>
+    typename uniform_real_distribution<RealType>::result_type
+      uniform_real_distribution<RealType>
+        ::operator()(UniformRandomNumberGenerator &urng)
+{
+  return operator()(urng, m_param);
+} // end uniform_real::operator()()
+
+template<typename RealType>
+  template<typename UniformRandomNumberGenerator>
+    typename uniform_real_distribution<RealType>::result_type
+      uniform_real_distribution<RealType>
+        ::operator()(UniformRandomNumberGenerator &urng,
+                     const param_type &parm)
+{
+  // call the urng & map its result to [0,1)
+  result_type result = static_cast<result_type>(urng() - UniformRandomNumberGenerator::min);
+
+  // adding one to the denominator ensures that the interval is half-open at 1.0
+  // XXX adding 1.0 to a potentially large floating point number seems like a bad idea
+  // XXX OTOH adding 1 to what is potentially UINT_MAX also seems like a bad idea
+  // XXX we could statically check if 1u + (max - min) is representable and do that, otherwise use the current implementation
+  result /= (result_type(1) + static_cast<result_type>(UniformRandomNumberGenerator::max - UniformRandomNumberGenerator::min));
+
+  return (result * (parm.second - parm.first)) + parm.first;
+} // end uniform_real::operator()()
+
+template<typename RealType>
+  typename uniform_real_distribution<RealType>::result_type
+    uniform_real_distribution<RealType>
+      ::a(void) const
+{
+  return m_param.first;
+} // end uniform_real::a()
+
+template<typename RealType>
+  typename uniform_real_distribution<RealType>::result_type
+    uniform_real_distribution<RealType>
+      ::b(void) const
+{
+  return m_param.second;
+} // end uniform_real_distribution::b()
+
+template<typename RealType>
+  typename uniform_real_distribution<RealType>::param_type
+    uniform_real_distribution<RealType>
+      ::param(void) const
+{
+  return m_param;;
+} // end uniform_real_distribution::param()
+
+template<typename RealType>
+  void uniform_real_distribution<RealType>
+    ::param(const param_type &parm)
+{
+  m_param = parm;
+} // end uniform_real_distribution::param()
+
+template<typename RealType>
+  typename uniform_real_distribution<RealType>::result_type
+    uniform_real_distribution<RealType>
+      ::min THRUST_PREVENT_MACRO_SUBSTITUTION (void) const
+{
+  return a();
+} // end uniform_real_distribution::min()
+
+template<typename RealType>
+  typename uniform_real_distribution<RealType>::result_type
+    uniform_real_distribution<RealType>
+      ::max THRUST_PREVENT_MACRO_SUBSTITUTION (void) const
+{
+  return b();
+} // end uniform_real_distribution::max()
+
+
+template<typename RealType>
+  bool uniform_real_distribution<RealType>
+    ::equal(const uniform_real_distribution &rhs) const
+{
+  return m_param == rhs.param();
+}
+
+
+template<typename RealType>
+  template<typename CharT, typename Traits>
+    std::basic_ostream<CharT,Traits>&
+      uniform_real_distribution<RealType>
+        ::stream_out(std::basic_ostream<CharT,Traits> &os) const
+{
+  typedef std::basic_ostream<CharT,Traits> ostream_type;
+  typedef typename ostream_type::ios_base  ios_base;
+
+  // save old flags and fill character
+  const typename ios_base::fmtflags flags = os.flags();
+  const CharT fill = os.fill();
+
+  const CharT space = os.widen(' ');
+  os.flags(ios_base::dec | ios_base::fixed | ios_base::left);
+  os.fill(space);
+
+  os << a() << space << b();
+
+  // restore old flags and fill character
+  os.flags(flags);
+  os.fill(fill);
+  return os;
+}
+
+
+template<typename RealType>
+  template<typename CharT, typename Traits>
+    std::basic_istream<CharT,Traits>&
+      uniform_real_distribution<RealType>
+        ::stream_in(std::basic_istream<CharT,Traits> &is)
+{
+  typedef std::basic_istream<CharT,Traits> istream_type;
+  typedef typename istream_type::ios_base  ios_base;
+
+  // save old flags
+  const typename ios_base::fmtflags flags = is.flags();
+
+  is.flags(ios_base::skipws);
+
+  is >> m_param.first >> m_param.second;
+
+  // restore old flags
+  is.flags(flags);
+  return is;
+}
+
+
+template<typename RealType>
+bool operator==(const uniform_real_distribution<RealType> &lhs,
+                const uniform_real_distribution<RealType> &rhs)
+{
+  return thrust::random::detail::random_core_access::equal(lhs,rhs);
+}
+
+
+template<typename RealType>
+bool operator!=(const uniform_real_distribution<RealType> &lhs,
+                const uniform_real_distribution<RealType> &rhs)
+{
+  return !(lhs == rhs);
+}
+
+
+template<typename RealType,
+         typename CharT, typename Traits>
+std::basic_ostream<CharT,Traits>&
+operator<<(std::basic_ostream<CharT,Traits> &os,
+           const uniform_real_distribution<RealType> &d)
+{
+  return thrust::random::detail::random_core_access::stream_out(os,d);
+}
+
+
+template<typename RealType,
+         typename CharT, typename Traits>
+std::basic_istream<CharT,Traits>&
+operator>>(std::basic_istream<CharT,Traits> &is,
+           uniform_real_distribution<RealType> &d)
+{
+  return thrust::random::detail::random_core_access::stream_in(is,d);
+}
+
+
+} // end random
+
+} // end thrust
+
diff --git a/compat/thrust/random/detail/xor_combine_engine.inl b/compat/thrust/random/detail/xor_combine_engine.inl
new file mode 100644
index 0000000000..b138722f8b
--- /dev/null
+++ b/compat/thrust/random/detail/xor_combine_engine.inl
@@ -0,0 +1,203 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/random/xor_combine_engine.h>
+#include <thrust/random/detail/random_core_access.h>
+
+namespace thrust
+{
+
+namespace random
+{
+
+template<typename Engine1, size_t s1,
+         typename Engine2, size_t s2>
+  xor_combine_engine<Engine1,s1,Engine2,s2>
+    ::xor_combine_engine(void)
+      :m_b1(),m_b2()
+{
+} // end xor_combine_engine::xor_combine_engine()
+
+template<typename Engine1, size_t s1,
+         typename Engine2, size_t s2>
+  xor_combine_engine<Engine1,s1,Engine2,s2>
+    ::xor_combine_engine(const base1_type &urng1, const base2_type &urng2)
+      :m_b1(urng1),m_b2(urng2)
+{
+} // end xor_combine_engine::xor_combine_engine()
+
+template<typename Engine1, size_t s1,
+         typename Engine2, size_t s2>
+  xor_combine_engine<Engine1,s1,Engine2,s2>
+    ::xor_combine_engine(result_type s)
+      :m_b1(s),m_b2(s)
+{
+} // end xor_combine_engine::xor_combine_engine()
+
+template<typename Engine1, size_t s1,
+         typename Engine2, size_t s2>
+  void xor_combine_engine<Engine1,s1,Engine2,s2>
+    ::seed(void)
+{
+  m_b1.seed();
+  m_b2.seed();
+} // end xor_combine_engine::seed()
+
+template<typename Engine1, size_t s1,
+         typename Engine2, size_t s2>
+  void xor_combine_engine<Engine1,s1,Engine2,s2>
+    ::seed(result_type s)
+{
+  m_b1.seed(s);
+  m_b2.seed(s);
+} // end xor_combine_engine::seed()
+
+template<typename Engine1, size_t s1,
+         typename Engine2, size_t s2>
+  const typename xor_combine_engine<Engine1,s1,Engine2,s2>::base1_type &
+    xor_combine_engine<Engine1,s1,Engine2,s2>
+      ::base1(void) const
+{
+  return m_b1;
+} // end xor_combine_engine::base1()
+
+template<typename Engine1, size_t s1,
+         typename Engine2, size_t s2>
+  const typename xor_combine_engine<Engine1,s1,Engine2,s2>::base2_type &
+    xor_combine_engine<Engine1,s1,Engine2,s2>
+      ::base2(void) const
+{
+  return m_b2;
+} // end xor_combine_engine::base2()
+
+template<typename Engine1, size_t s1,
+         typename Engine2, size_t s2>
+  typename xor_combine_engine<Engine1,s1,Engine2,s2>::result_type
+    xor_combine_engine<Engine1,s1,Engine2,s2>
+      ::operator()(void)
+{
+  return (result_type(m_b1() - base1_type::min) << shift1) ^
+         (result_type(m_b2() - base2_type::min) << shift2);
+} // end xor_combine_engine::operator()()
+
+template<typename Engine1, size_t s1,
+         typename Engine2, size_t s2>
+  void xor_combine_engine<Engine1, s1, Engine2, s2>
+    ::discard(unsigned long long z)
+{
+  for(; z > 0; --z)
+  {
+    this->operator()();
+  } // end for
+} // end xor_combine_engine::discard()
+
+
+template<typename Engine1, size_t s1, typename Engine2, size_t s2>
+  template<typename CharT, typename Traits>
+    std::basic_ostream<CharT,Traits>& xor_combine_engine<Engine1,s1,Engine2,s2>
+      ::stream_out(std::basic_ostream<CharT,Traits> &os) const
+{
+  typedef std::basic_ostream<CharT,Traits> ostream_type;
+  typedef typename ostream_type::ios_base  ios_base;
+
+  // save old flags and fill character
+  const typename ios_base::fmtflags flags = os.flags();
+  const CharT fill = os.fill();
+
+  const CharT space = os.widen(' ');
+  os.flags(ios_base::dec | ios_base::fixed | ios_base::left);
+  os.fill(space);
+
+  // output each base engine in turn
+  os << base1() << space << base2();
+
+  // restore old flags and fill character
+  os.flags(flags);
+  os.fill(fill);
+  return os;
+}
+
+
+template<typename Engine1, size_t s1, typename Engine2, size_t s2>
+  template<typename CharT, typename Traits>
+    std::basic_istream<CharT,Traits>& xor_combine_engine<Engine1,s1,Engine2,s2>
+      ::stream_in(std::basic_istream<CharT,Traits> &is)
+{
+  typedef std::basic_istream<CharT,Traits> istream_type;
+  typedef typename istream_type::ios_base  ios_base;
+
+  // save old flags
+  const typename ios_base::fmtflags flags = is.flags();
+
+  is.flags(ios_base::skipws);
+
+  // input each base engine in turn
+  is >> m_b1 >> m_b2;
+
+  // restore old flags
+  is.flags(flags);
+  return is;
+}
+
+
+template<typename Engine1, size_t s1, typename Engine2, size_t s2>
+  bool xor_combine_engine<Engine1,s1,Engine2,s2>
+    ::equal(const xor_combine_engine<Engine1,s1,Engine2,s2> &rhs) const
+{
+  return (m_b1 == rhs.m_b1) && (m_b2 == rhs.m_b2);
+}
+
+
+template<typename Engine1, size_t s1, typename Engine2, size_t s2,
+         typename CharT, typename Traits>
+std::basic_ostream<CharT,Traits>&
+operator<<(std::basic_ostream<CharT,Traits> &os,
+           const xor_combine_engine<Engine1,s1,Engine2,s2> &e)
+{
+  return thrust::random::detail::random_core_access::stream_out(os,e);
+}
+
+
+template<typename Engine1, size_t s1, typename Engine2, size_t s2,
+         typename CharT, typename Traits>
+std::basic_istream<CharT,Traits>&
+operator>>(std::basic_istream<CharT,Traits> &is,
+           xor_combine_engine<Engine1,s1,Engine2,s2> &e)
+{
+  return thrust::random::detail::random_core_access::stream_in(is,e);
+}
+
+
+template<typename Engine1, size_t s1, typename Engine2, size_t s2>
+bool operator==(const xor_combine_engine<Engine1,s1,Engine2,s2> &lhs,
+                const xor_combine_engine<Engine1,s1,Engine2,s2> &rhs)
+{
+  return thrust::random::detail::random_core_access::equal(lhs,rhs);
+}
+
+
+template<typename Engine1, size_t s1, typename Engine2, size_t s2>
+bool operator!=(const xor_combine_engine<Engine1,s1,Engine2,s2> &lhs,
+                const xor_combine_engine<Engine1,s1,Engine2,s2> &rhs)
+{
+  return !(lhs == rhs);
+}
+
+
+} // end random
+
+} // end thrust
+
diff --git a/compat/thrust/random/detail/xor_combine_engine_max.h b/compat/thrust/random/detail/xor_combine_engine_max.h
new file mode 100644
index 0000000000..8bad9a462c
--- /dev/null
+++ b/compat/thrust/random/detail/xor_combine_engine_max.h
@@ -0,0 +1,324 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/type_traits.h>
+#include <thrust/detail/mpl/math.h>
+#include <limits>
+#include <cstddef>
+
+namespace thrust
+{
+
+namespace random
+{
+
+namespace detail
+{
+
+
+namespace math = thrust::detail::mpl::math;
+
+
+namespace detail
+{
+
+// two cases for this function avoids compile-time warnings of overflow
+template<typename UIntType, UIntType w,
+         UIntType lhs, UIntType rhs,
+         bool shift_will_overflow>
+  struct lshift_w
+{
+  static const UIntType value = 0;
+};
+
+
+template<typename UIntType, UIntType w,
+         UIntType lhs, UIntType rhs>
+  struct lshift_w<UIntType,w,lhs,rhs,false>
+{
+  static const UIntType value = lhs << rhs;
+};
+
+} // end detail
+
+
+template<typename UIntType, UIntType w,
+         UIntType lhs, UIntType rhs>
+  struct lshift_w
+{
+  static const bool shift_will_overflow = rhs >= w;
+
+  static const UIntType value = detail::lshift_w<UIntType, w, lhs, rhs, shift_will_overflow>::value;
+};
+
+
+template<typename UIntType, UIntType lhs, UIntType rhs>
+  struct lshift
+    : lshift_w<UIntType, std::numeric_limits<UIntType>::digits, lhs, rhs>
+{};
+
+
+template<typename UIntType, int p>
+  struct two_to_the_power
+    : lshift<UIntType, 1, p>
+{};
+
+
+template<typename result_type, result_type a, result_type b, int d>
+  class xor_combine_engine_max_aux_constants
+{
+  public:
+    static const result_type two_to_the_d = two_to_the_power<result_type, d>::value;
+    static const result_type c = lshift<result_type, a, d>::value;
+
+    static const result_type t =
+      math::max<
+        result_type,
+        c,
+        b
+      >::value;
+
+    static const result_type u =
+      math::min<
+        result_type,
+        c,
+        b
+      >::value;
+
+    static const result_type p            = math::log2<u>::value;
+    static const result_type two_to_the_p = two_to_the_power<result_type, p>::value;
+
+    static const result_type k = math::div<result_type, t, two_to_the_p>::value;
+};
+
+
+template<typename result_type, result_type, result_type, int> struct xor_combine_engine_max_aux;
+
+
+template<typename result_type, result_type a, result_type b, int d>
+  struct xor_combine_engine_max_aux_case4
+{
+  typedef xor_combine_engine_max_aux_constants<result_type,a,b,d> constants;
+
+  static const result_type k_plus_1_times_two_to_the_p =
+    lshift<
+      result_type,
+      math::plus<result_type,constants::k,1>::value,
+      constants::p
+    >::value;
+
+  static const result_type M =
+    xor_combine_engine_max_aux<
+      result_type,
+      math::div<
+        result_type,
+        math::mod<
+          result_type,
+          constants::u,
+          constants::two_to_the_p
+        >::value,
+        constants::two_to_the_p
+      >::value,
+      math::mod<
+        result_type,
+        constants::t,
+        constants::two_to_the_p
+      >::value,
+      d
+    >::value;
+
+  static const result_type value = math::plus<result_type, k_plus_1_times_two_to_the_p, M>::value;
+};
+
+
+template<typename result_type, result_type a, result_type b, int d>
+  struct xor_combine_engine_max_aux_case3
+{
+  typedef xor_combine_engine_max_aux_constants<result_type,a,b,d> constants;
+
+  static const result_type k_plus_1_times_two_to_the_p =
+    lshift<
+      result_type,
+      math::plus<result_type,constants::k,1>::value,
+      constants::p
+    >::value;
+
+  static const result_type M =
+    xor_combine_engine_max_aux<
+      result_type,
+      math::div<
+        result_type,
+        math::mod<
+          result_type,
+          constants::t,
+          constants::two_to_the_p
+        >::value,
+        constants::two_to_the_p
+      >::value,
+      math::mod<
+        result_type,
+        constants::u,
+        constants::two_to_the_p
+      >::value,
+      d
+    >::value;
+
+  static const result_type value = math::plus<result_type, k_plus_1_times_two_to_the_p, M>::value;
+};
+
+
+
+template<typename result_type, result_type a, result_type b, int d>
+  struct xor_combine_engine_max_aux_case2
+{
+  typedef xor_combine_engine_max_aux_constants<result_type,a,b,d> constants;
+
+  static const result_type k_plus_1_times_two_to_the_p =
+    lshift<
+      result_type,
+      math::plus<result_type,constants::k,1>::value,
+      constants::p
+    >::value;
+
+  static const result_type value =
+    math::minus<
+      result_type,
+      k_plus_1_times_two_to_the_p,
+      1
+    >::value;
+};
+
+
+template<typename result_type, result_type a, result_type b, int d>
+  struct xor_combine_engine_max_aux_case1
+{
+  static const result_type c     = lshift<result_type, a, d>::value;
+
+  static const result_type value = math::plus<result_type,c,b>::value;
+};
+
+
+template<typename result_type, result_type a, result_type b, int d>
+  struct xor_combine_engine_max_aux_2
+{
+  typedef xor_combine_engine_max_aux_constants<result_type,a,b,d> constants;
+
+  static const result_type value = 
+    thrust::detail::eval_if<
+      // if k is odd...
+      math::is_odd<result_type, constants::k>::value,
+      thrust::detail::identity_<
+        thrust::detail::integral_constant<
+          result_type,
+          xor_combine_engine_max_aux_case2<result_type,a,b,d>::value
+        >
+      >,
+      thrust::detail::eval_if<
+        // otherwise if a * 2^3 >= b, then case 3
+        a * constants::two_to_the_d >= b,
+        thrust::detail::identity_<
+          thrust::detail::integral_constant<
+            result_type,
+            xor_combine_engine_max_aux_case3<result_type,a,b,d>::value
+          >
+        >,
+        // otherwise, case 4
+        thrust::detail::identity_<
+          thrust::detail::integral_constant<
+            result_type,
+            xor_combine_engine_max_aux_case4<result_type,a,b,d>::value
+          >
+        >
+      >
+    >::type::value;
+};
+
+
+template<typename result_type,
+         result_type a,
+         result_type b,
+         int d,
+         bool use_case1 = (a == 0) || (b < two_to_the_power<result_type,d>::value)>
+  struct xor_combine_engine_max_aux_1
+    : xor_combine_engine_max_aux_case1<result_type,a,b,d>
+{};
+
+
+template<typename result_type,
+         result_type a,
+         result_type b,
+         int d>
+  struct xor_combine_engine_max_aux_1<result_type,a,b,d,false>
+    : xor_combine_engine_max_aux_2<result_type,a,b,d>
+{};
+
+
+template<typename result_type,
+         result_type a,
+         result_type b,
+         int d>
+  struct xor_combine_engine_max_aux
+    : xor_combine_engine_max_aux_1<result_type,a,b,d>
+{};
+
+
+template<typename Engine1, size_t s1, typename Engine2, size_t s2, typename result_type>
+  struct xor_combine_engine_max
+{
+  static const size_t w = std::numeric_limits<result_type>::digits;
+
+  static const result_type m1 =
+    math::min<
+      result_type,
+      result_type(Engine1::max - Engine1::min),
+      two_to_the_power<result_type, w-s1>::value - 1 
+    >::value;
+
+  static const result_type m2 =
+    math::min<
+      result_type,
+      result_type(Engine2::max - Engine2::min),
+      two_to_the_power<result_type, w-s2>::value - 1
+    >::value;
+
+  static const result_type s = s1 - s2;
+
+  static const result_type M =
+    xor_combine_engine_max_aux<
+      result_type,
+      m1,
+      m2,
+      s
+    >::value;
+
+  // the value is M(m1,m2,s) lshift_w s2
+  static const result_type value =
+    lshift_w<
+      result_type,
+      w,
+      M,
+      s2
+    >::value;
+}; // end xor_combine_engine_max
+
+} // end detail
+
+} // end random
+
+} // end thrust
+
diff --git a/compat/thrust/random/discard_block_engine.h b/compat/thrust/random/discard_block_engine.h
new file mode 100644
index 0000000000..c902c5863b
--- /dev/null
+++ b/compat/thrust/random/discard_block_engine.h
@@ -0,0 +1,252 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file discard_block_engine.h
+ *  \brief A random number engine which adapts a base engine and produces
+ *         numbers by discarding all but a contiguous blocks of its values.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+#include <thrust/detail/config.h>
+#include <iostream>
+#include <thrust/detail/cstdint.h>
+#include <thrust/random/detail/random_core_access.h>
+
+namespace thrust
+{
+
+namespace random
+{
+
+/*! \addtogroup random_number_engine_adaptors Random Number Engine Adaptor Class Templates
+ *  \ingroup random
+ *  \{
+ */
+
+/*! \class discard_block_engine
+ *  \brief A \p discard_block_engine adapts an existing base random number engine and produces
+ *         random values by discarding some of the values returned by its base engine.
+ *         Each cycle of the compound engine begins by returning \c r values successively produced
+ *         by the base engine and ends by discarding <tt>p-r</tt> such values. The engine's state
+ *         is the state of its base engine followed by the number of calls to <tt>operator()</tt>
+ *         that have occurred since the beginning of the current cycle.
+ *
+ *  \tparam Engine The type of the base random number engine to adapt.
+ *  \tparam p The discard cycle length.
+ *  \tparam r The number of values to return of the base engine. Because <tt>p-r</tt> will be
+ *            discarded, <tt>r <= p</tt>.
+ *
+ *  The following code snippet shows an example of using a \p discard_block_engine instance:
+ *
+ *  \code
+ *  #include <thrust/random/linear_congruential_engine.h>
+ *  #include <thrust/random/discard_block_engine.h>
+ *  #include <iostream>
+ *
+ *  int main(void)
+ *  {
+ *    // create a discard_block_engine from minstd_rand, with a cycle length of 13
+ *    // keep every first 10 values, and discard the next 3
+ *    thrust::discard_block_engine<thrust::minstd_rand, 13, 10> rng;
+ *
+ *    // print a random number to standard output
+ *    std::cout << rng() << std::endl;
+ *
+ *    return 0;
+ *  }
+ *  \endcode
+ */         
+template<typename Engine, size_t p, size_t r>
+  class discard_block_engine
+{
+  public:
+    // types
+
+    /*! \typedef base_type
+     *  \brief The type of the adapted base random number engine.
+     */
+    typedef Engine base_type;
+
+    /*! \typedef result_type
+     *  \brief The type of the unsigned integer produced by this \p linear_congruential_engine.
+     */
+    typedef typename base_type::result_type result_type;
+
+    // engine characteristics
+
+    /*! The length of the production cycle.
+     */
+    static const size_t block_size = p;
+
+    /*! The number of used numbers per production cycle.
+     */
+    static const size_t used_block = r;
+
+    /*! The smallest value this \p discard_block_engine may potentially produce.
+     */
+    static const result_type min = base_type::min;
+
+    /*! The largest value this \p discard_block_engine may potentially produce.
+     */
+    static const result_type max = base_type::max;
+
+    // constructors and seeding functions
+
+    /*! This constructor constructs a new \p discard_block_engine and constructs
+     *  its \p base_type engine using its null constructor.
+     */
+    __host__ __device__
+    discard_block_engine();
+
+    /*! This constructor constructs a new \p discard_block_engine using
+     *  a given \p base_type engine to initialize its adapted base engine.
+     *
+     *  \param urng A \p base_type to use to initialize this \p discard_block_engine's
+     *         adapted base engine.
+     */
+    __host__ __device__
+    explicit discard_block_engine(const base_type &urng);
+
+    /*! This constructor initializes a new \p discard_block_engine with a given seed.
+     *  
+     *  \param s The seed used to intialize this \p discard_block_engine's adapted base engine.
+     */
+    __host__ __device__
+    explicit discard_block_engine(result_type s);
+
+    /*! This method initializes the state of this \p discard_block_engine's adapted base engine
+     *  by using its \p default_seed value.
+     */
+    __host__ __device__
+    void seed(void);
+
+    /*! This method initializes the state of this \p discard_block_engine's adapted base engine
+     *  by using the given seed.
+     *
+     *  \param s The seed with which to intialize this \p discard_block_engine's adapted base engine.
+     */
+    __host__ __device__
+    void seed(result_type s);
+
+    // generating functions
+    
+    /*! This member function produces a new random value and updates this \p discard_block_engine's state.
+     *  \return A new random number.
+     */
+    __host__ __device__
+    result_type operator()(void);
+
+    /*! This member function advances this \p discard_block_engine's state a given number of times
+     *  and discards the results.
+     *
+     *  \param z The number of random values to discard.
+     *  \note This function is provided because an implementation may be able to accelerate it.
+     */
+    __host__ __device__
+    void discard(unsigned long long z);
+
+    // property functions
+
+    /*! This member function returns a const reference to this \p discard_block_engine's
+     *  adapted base engine.
+     *
+     *  \return A const reference to the base engine this \p discard_block_engine adapts.
+     */
+    __host__ __device__
+    const base_type &base(void) const;
+
+    /*! \cond
+     */
+  private:
+    base_type m_e;
+    unsigned int m_n;
+
+    friend struct thrust::random::detail::random_core_access;
+
+    __host__ __device__
+    bool equal(const discard_block_engine &rhs) const;
+
+    template<typename CharT, typename Traits>
+    std::basic_ostream<CharT,Traits>& stream_out(std::basic_ostream<CharT,Traits> &os) const;
+
+    template<typename CharT, typename Traits>
+    std::basic_istream<CharT,Traits>& stream_in(std::basic_istream<CharT,Traits> &is);
+    /*! \endcond
+     */
+}; // end discard_block_engine
+
+
+/*! This function checks two \p discard_block_engines for equality.
+ *  \param lhs The first \p discard_block_engine to test.
+ *  \param rhs The second \p discard_block_engine to test.
+ *  \return \c true if \p lhs is equal to \p rhs; \c false, otherwise.
+ */
+template<typename Engine, size_t p, size_t r>
+__host__ __device__
+bool operator==(const discard_block_engine<Engine,p,r> &lhs,
+                const discard_block_engine<Engine,p,r> &rhs);
+
+
+/*! This function checks two \p discard_block_engines for inequality.
+ *  \param lhs The first \p discard_block_engine to test.
+ *  \param rhs The second \p discard_block_engine to test.
+ *  \return \c true if \p lhs is not equal to \p rhs; \c false, otherwise.
+ */
+template<typename Engine, size_t p, size_t r>
+__host__ __device__
+bool operator!=(const discard_block_engine<Engine,p,r> &lhs,
+                const discard_block_engine<Engine,p,r> &rhs);
+
+
+/*! This function streams a discard_block_engine to a \p std::basic_ostream.
+ *  \param os The \p basic_ostream to stream out to.
+ *  \param e The \p discard_block_engine to stream out.
+ *  \return \p os
+ */
+template<typename Engine, size_t p, size_t r,
+         typename CharT, typename Traits>
+std::basic_ostream<CharT,Traits>&
+operator<<(std::basic_ostream<CharT,Traits> &os,
+           const discard_block_engine<Engine,p,r> &e);
+
+
+/*! This function streams a discard_block_engine in from a std::basic_istream.
+ *  \param is The \p basic_istream to stream from.
+ *  \param e The \p discard_block_engine to stream in.
+ *  \return \p is
+ */
+template<typename Engine, size_t p, size_t r,
+         typename CharT, typename Traits>
+std::basic_istream<CharT,Traits>&
+operator>>(std::basic_istream<CharT,Traits> &is,
+           discard_block_engine<Engine,p,r> &e);
+
+/*! \} // end random_number_engine_adaptors
+ */
+
+} // end random
+
+// import names into thrust::
+using random::discard_block_engine;
+
+} // end thrust
+
+#include <thrust/random/detail/discard_block_engine.inl>
+
diff --git a/compat/thrust/random/linear_congruential_engine.h b/compat/thrust/random/linear_congruential_engine.h
new file mode 100644
index 0000000000..0added0069
--- /dev/null
+++ b/compat/thrust/random/linear_congruential_engine.h
@@ -0,0 +1,295 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file linear_congruential_engine.h
+ *  \brief A linear congruential pseudorandom number engine.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <iostream>
+#include <thrust/detail/cstdint.h>
+#include <thrust/random/detail/random_core_access.h>
+#include <thrust/random/detail/linear_congruential_engine_discard.h>
+
+namespace thrust
+{
+
+namespace random
+{
+
+/*! \addtogroup random_number_engine_templates Random Number Engine Class Templates
+ *  \ingroup random
+ *  \{
+ */
+
+/*! \class linear_congruential_engine
+ *  \brief A \p linear_congruential_engine random number engine produces unsigned integer
+ *         random numbers using a linear congruential random number generation algorithm.
+ *
+ *         The generation algorithm has the form <tt>x_i = (a * x_{i-1} + c) mod m</tt>.
+ *
+ *  \tparam UIntType The type of unsigned integer to produce.
+ *  \tparam a The multiplier used in the generation algorithm.
+ *  \tparam c The increment used in the generation algorithm.
+ *  \tparam m The modulus used in the generation algorithm.
+ *
+ *  \note Inexperienced users should not use this class template directly.  Instead, use
+ *  \p minstd_rand or \p minstd_rand0.
+ *
+ *  The following code snippet shows examples of use of a \p linear_congruential_engine instance:
+ *
+ *  \code
+ *  #include <thrust/random/linear_congruential_engine.h>
+ *  #include <iostream>
+ *
+ *  int main(void)
+ *  {
+ *    // create a minstd_rand object, which is an instance of linear_congruential_engine
+ *    thrust::minstd_rand rng1;
+ *
+ *    // output some random values to cout
+ *    std::cout << rng1() << std::endl;
+ *
+ *    // a random value is printed
+ *
+ *    // create a new minstd_rand from a seed
+ *    thrust::minstd_rand rng2(13);
+ *
+ *    // discard some random values
+ *    rng2.discard(13);
+ *
+ *    // stream the object to an iostream
+ *    std::cout << rng2 << std::endl;
+ *
+ *    // rng2's current state is printed
+ *
+ *    // print the minimum and maximum values that minstd_rand can produce
+ *    std::cout << thrust::minstd_rand::min << std::endl;
+ *    std::cout << thrust::minstd_rand::max << std::endl;
+ *
+ *    // the range of minstd_rand is printed
+ *
+ *    // save the state of rng2 to a different object
+ *    thrust::minstd_rand rng3 = rng2;
+ *
+ *    // compare rng2 and rng3
+ *    std::cout << (rng2 == rng3) << std::endl;
+ *
+ *    // 1 is printed
+ *
+ *    // re-seed rng2 with a different seed
+ *    rng2.seed(7);
+ *
+ *    // compare rng2 and rng3
+ *    std::cout << (rng2 == rng3) << std::endl;
+ *
+ *    // 0 is printed
+ *
+ *    return 0;
+ *  }
+ *
+ *  \endcode
+ *
+ *  \see thrust::random::minstd_rand
+ *  \see thrust::random::minstd_rand0
+ */
+template<typename UIntType, UIntType a, UIntType c, UIntType m>
+  class linear_congruential_engine
+{
+  public:
+    // types
+    
+    /*! \typedef result_type
+     *  \brief The type of the unsigned integer produced by this \p linear_congruential_engine.
+     */
+    typedef UIntType result_type;
+
+    // engine characteristics
+
+    /*! The multiplier used in the generation algorithm.
+     */
+    static const result_type multiplier = a;
+
+    /*! The increment used in the generation algorithm.
+     */
+    static const result_type increment = c;
+
+    /*! The modulus used in the generation algorithm.
+     */
+    static const result_type modulus = m;
+
+    /*! The smallest value this \p linear_congruential_engine may potentially produce.
+     */
+    static const result_type min = c == 0u ? 1u : 0u;
+
+    /*! The largest value this \p linear_congruential_engine may potentially produce.
+     */
+    static const result_type max = m - 1u;
+
+    /*! The default seed of this \p linear_congruential_engine.
+     */
+    static const result_type default_seed = 1u;
+
+    // constructors and seeding functions
+
+    /*! This constructor, which optionally accepts a seed, initializes a new
+     *  \p linear_congruential_engine.
+     *  
+     *  \param s The seed used to intialize this \p linear_congruential_engine's state.
+     */
+    __host__ __device__
+    explicit linear_congruential_engine(result_type s = default_seed);
+
+    /*! This method initializes this \p linear_congruential_engine's state, and optionally accepts
+     *  a seed value.
+     *
+     *  \param s The seed used to initializes this \p linear_congruential_engine's state.
+     */
+    __host__ __device__
+    void seed(result_type s = default_seed);
+
+    // generating functions
+
+    /*! This member function produces a new random value and updates this \p linear_congruential_engine's state.
+     *  \return A new random number.
+     */
+    __host__ __device__
+    result_type operator()(void);
+
+    /*! This member function advances this \p linear_congruential_engine's state a given number of times
+     *  and discards the results.
+     *
+     *  \param z The number of random values to discard.
+     *  \note This function is provided because an implementation may be able to accelerate it.
+     */
+    __host__ __device__
+    void discard(unsigned long long z);
+
+    /*! \cond
+     */
+  private:
+    result_type m_x;
+
+    static void transition(result_type &state);
+
+    friend struct thrust::random::detail::random_core_access;
+
+    friend struct thrust::random::detail::linear_congruential_engine_discard;
+
+    __host__ __device__
+    bool equal(const linear_congruential_engine &rhs) const;
+
+    template<typename CharT, typename Traits>
+    std::basic_ostream<CharT,Traits>& stream_out(std::basic_ostream<CharT,Traits> &os) const;
+
+    template<typename CharT, typename Traits>
+    std::basic_istream<CharT,Traits>& stream_in(std::basic_istream<CharT,Traits> &is);
+
+    /*! \endcond
+     */
+}; // end linear_congruential_engine
+
+
+/*! This function checks two \p linear_congruential_engines for equality.
+ *  \param lhs The first \p linear_congruential_engine to test.
+ *  \param rhs The second \p linear_congruential_engine to test.
+ *  \return \c true if \p lhs is equal to \p rhs; \c false, otherwise.
+ */
+template<typename UIntType_, UIntType_ a_, UIntType_ c_, UIntType_ m_>
+__host__ __device__
+bool operator==(const linear_congruential_engine<UIntType_,a_,c_,m_> &lhs,
+                const linear_congruential_engine<UIntType_,a_,c_,m_> &rhs);
+
+
+/*! This function checks two \p linear_congruential_engines for inequality.
+ *  \param lhs The first \p linear_congruential_engine to test.
+ *  \param rhs The second \p linear_congruential_engine to test.
+ *  \return \c true if \p lhs is not equal to \p rhs; \c false, otherwise.
+ */
+template<typename UIntType_, UIntType_ a_, UIntType_ c_, UIntType_ m_>
+__host__ __device__
+bool operator!=(const linear_congruential_engine<UIntType_,a_,c_,m_> &lhs,
+                const linear_congruential_engine<UIntType_,a_,c_,m_> &rhs);
+
+
+/*! This function streams a linear_congruential_engine to a \p std::basic_ostream.
+ *  \param os The \p basic_ostream to stream out to.
+ *  \param e The \p linear_congruential_engine to stream out.
+ *  \return \p os
+ */
+template<typename UIntType_, UIntType_ a_, UIntType_ c_, UIntType_ m_,
+         typename CharT, typename Traits>
+std::basic_ostream<CharT,Traits>&
+operator<<(std::basic_ostream<CharT,Traits> &os,
+           const linear_congruential_engine<UIntType_,a_,c_,m_> &e);
+
+
+/*! This function streams a linear_congruential_engine in from a std::basic_istream.
+ *  \param is The \p basic_istream to stream from.
+ *  \param e The \p linear_congruential_engine to stream in.
+ *  \return \p is
+ */
+template<typename UIntType_, UIntType_ a_, UIntType_ c_, UIntType_ m_,
+         typename CharT, typename Traits>
+std::basic_istream<CharT,Traits>&
+operator>>(std::basic_istream<CharT,Traits> &is,
+           linear_congruential_engine<UIntType_,a_,c_,m_> &e);
+
+
+/*! \} // random_number_engine_templates
+ */
+
+
+/*! \addtogroup predefined_random
+ *  \{
+ */
+
+// XXX the type N2111 used here was uint_fast32_t
+
+/*! \typedef minstd_rand0
+ *  \brief A random number engine with predefined parameters which implements a version of
+ *         the Minimal Standard random number generation algorithm.
+ *  \note The 10000th consecutive invocation of a default-constructed object of type \p minstd_rand0
+ *        shall produce the value \c 1043618065 .
+ */
+typedef linear_congruential_engine<thrust::detail::uint32_t, 16807, 0, 2147483647> minstd_rand0;
+
+
+/*! \typedef minstd_rand
+ *  \brief A random number engine with predefined parameters which implements a version of
+ *         the Minimal Standard random number generation algorithm.
+ *  \note The 10000th consecutive invocation of a default-constructed object of type \p minstd_rand
+ *        shall produce the value \c 399268537 .
+ */
+typedef linear_congruential_engine<thrust::detail::uint32_t, 48271, 0, 2147483647> minstd_rand;
+
+/*! \} // predefined_random
+ */
+  
+} // end random
+
+// import names into thrust::
+using random::linear_congruential_engine;
+using random::minstd_rand;
+using random::minstd_rand0;
+
+} // end thrust
+
+#include <thrust/random/detail/linear_congruential_engine.inl>
+
diff --git a/compat/thrust/random/linear_feedback_shift_engine.h b/compat/thrust/random/linear_feedback_shift_engine.h
new file mode 100644
index 0000000000..f5646c9483
--- /dev/null
+++ b/compat/thrust/random/linear_feedback_shift_engine.h
@@ -0,0 +1,230 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+/*! \file linear_feedback_shift_engine.h
+ *  \brief A linear feedback shift pseudorandom number generator.
+ */
+
+/*
+ * Copyright Jens Maurer 2002
+ *
+ * Distributed under the Boost Software License, Version 1.0.
+ * (See accompanying NOTICE file for the complete license)
+ *
+ * For more information, see http://www.boost.org
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/random/detail/linear_feedback_shift_engine_wordmask.h>
+#include <iostream>
+#include <cstddef> // for size_t
+#include <thrust/random/detail/random_core_access.h>
+
+namespace thrust
+{
+
+
+namespace random
+{
+
+/*! \addtogroup random_number_engine_templates
+ *  \{
+ */
+
+/*! \class linear_feedback_shift_engine
+ *  \brief A \p linear_feedback_shift_engine random number engine produces
+ *         unsigned integer random values using a linear feedback shift random number
+ *         generation algorithm.
+ *
+ *  \tparam UIntType The type of unsigned integer to produce.
+ *  \tparam w The word size of the produced values (<tt>w <= sizeof(UIntType)</tt>).
+ *  \tparam k The k parameter of Tausworthe's 1965 algorithm.
+ *  \tparam q The q exponent of Tausworthe's 1965 algorithm.
+ *  \tparam s The step size of Tausworthe's 1965 algorithm.
+ *
+ *  \note linear_feedback_shift_engine is based on the Boost Template Library's linear_feedback_shift.
+ */
+template<typename UIntType, size_t w, size_t k, size_t q, size_t s>
+  class linear_feedback_shift_engine
+{
+  public:
+    // types
+
+    /*! \typedef result_type
+     *  \brief The type of the unsigned integer produced by this \p linear_feedback_shift_engine.
+     */
+    typedef UIntType result_type;
+
+    // engine characteristics
+
+    /*! The word size of the produced values.
+     */
+    static const size_t word_size = w;
+
+    /*! A constant used in the generation algorithm.
+     */
+    static const size_t exponent1 = k;
+
+    /*! A constant used in the generation algorithm.
+     */
+    static const size_t exponent2 = q;
+
+    /*! The step size used in the generation algorithm.
+     */
+    static const size_t step_size = s;
+
+    /*! \cond
+     */
+  private:
+    static const result_type wordmask =
+      detail::linear_feedback_shift_engine_wordmask<
+        result_type,
+        w
+      >::value;
+    /*! \endcond
+     */
+
+  public:
+
+    /*! The smallest value this \p linear_feedback_shift_engine may potentially produce.
+     */
+    static const result_type min = 0;
+
+    /*! The largest value this \p linear_feedback_shift_engine may potentially produce.
+     */
+    static const result_type max = wordmask;
+
+    /*! The default seed of this \p linear_feedback_shift_engine.
+     */
+    static const result_type default_seed = 341u;
+
+    // constructors and seeding functions
+
+    /*! This constructor, which optionally accepts a seed, initializes a new
+     *  \p linear_feedback_shift_engine.
+     *  
+     *  \param value The seed used to intialize this \p linear_feedback_shift_engine's state.
+     */
+    __host__ __device__
+    explicit linear_feedback_shift_engine(result_type value = default_seed);
+
+    /*! This method initializes this \p linear_feedback_shift_engine's state, and optionally accepts
+     *  a seed value.
+     *
+     *  \param value The seed used to initializes this \p linear_feedback_shift_engine's state.
+     */
+    __host__ __device__
+    void seed(result_type value = default_seed);
+
+    // generating functions
+    
+    /*! This member function produces a new random value and updates this \p linear_feedback_shift_engine's state.
+     *  \return A new random number.
+     */
+    __host__ __device__
+    result_type operator()(void);
+
+    /*! This member function advances this \p linear_feedback_shift_engine's state a given number of times
+     *  and discards the results.
+     *
+     *  \param z The number of random values to discard.
+     *  \note This function is provided because an implementation may be able to accelerate it.
+     */
+    __host__ __device__
+    void discard(unsigned long long z);
+
+    /*! \cond
+     */
+  private:
+    result_type m_value;
+
+    friend struct thrust::random::detail::random_core_access;
+
+    __host__ __device__
+    bool equal(const linear_feedback_shift_engine &rhs) const;
+
+    template<typename CharT, typename Traits>
+    std::basic_ostream<CharT,Traits>& stream_out(std::basic_ostream<CharT,Traits> &os) const;
+
+    template<typename CharT, typename Traits>
+    std::basic_istream<CharT,Traits>& stream_in(std::basic_istream<CharT,Traits> &is);
+
+    /*! \endcond
+     */
+}; // end linear_feedback_shift_engine
+
+
+/*! This function checks two \p linear_feedback_shift_engines for equality.
+ *  \param lhs The first \p linear_feedback_shift_engine to test.
+ *  \param rhs The second \p linear_feedback_shift_engine to test.
+ *  \return \c true if \p lhs is equal to \p rhs; \c false, otherwise.
+ */
+template<typename UIntType_, size_t w_, size_t k_, size_t q_, size_t s_>
+__host__ __device__
+bool operator==(const linear_feedback_shift_engine<UIntType_,w_,k_,q_,s_> &lhs,
+                const linear_feedback_shift_engine<UIntType_,w_,k_,q_,s_> &rhs);
+
+
+/*! This function checks two \p linear_feedback_shift_engines for inequality.
+ *  \param lhs The first \p linear_feedback_shift_engine to test.
+ *  \param rhs The second \p linear_feedback_shift_engine to test.
+ *  \return \c true if \p lhs is not equal to \p rhs; \c false, otherwise.
+ */
+template<typename UIntType_, size_t w_, size_t k_, size_t q_, size_t s_>
+__host__ __device__
+bool operator!=(const linear_feedback_shift_engine<UIntType_,w_,k_,q_,s_> &lhs,
+                const linear_feedback_shift_engine<UIntType_,w_,k_,q_,s_> &rhs);
+
+
+/*! This function streams a linear_feedback_shift_engine to a \p std::basic_ostream.
+ *  \param os The \p basic_ostream to stream out to.
+ *  \param e The \p linear_feedback_shift_engine to stream out.
+ *  \return \p os
+ */
+template<typename UIntType_, size_t w_, size_t k_, size_t q_, size_t s_,
+         typename CharT, typename Traits>
+std::basic_ostream<CharT,Traits>&
+operator<<(std::basic_ostream<CharT,Traits> &os,
+           const linear_feedback_shift_engine<UIntType_,w_,k_,q_,s_> &e);
+
+
+/*! This function streams a linear_feedback_shift_engine in from a std::basic_istream.
+ *  \param is The \p basic_istream to stream from.
+ *  \param e The \p linear_feedback_shift_engine to stream in.
+ *  \return \p is
+ */
+template<typename UIntType_, size_t w_, size_t k_, size_t q_, size_t s_,
+         typename CharT, typename Traits>
+std::basic_istream<CharT,Traits>&
+operator>>(std::basic_istream<CharT,Traits> &is,
+           linear_feedback_shift_engine<UIntType_,w_,k_,q_,s_> &e);
+
+
+/*! \} // end random_number_engine_templates
+ */
+
+
+} // end random
+
+// import names into thrust::
+using random::linear_feedback_shift_engine;
+
+} // end thrust
+
+#include <thrust/random/detail/linear_feedback_shift_engine.inl>
+
diff --git a/compat/thrust/random/normal_distribution.h b/compat/thrust/random/normal_distribution.h
new file mode 100644
index 0000000000..5543f30a5f
--- /dev/null
+++ b/compat/thrust/random/normal_distribution.h
@@ -0,0 +1,275 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file normal_distribution.h
+ *  \brief A normal (Gaussian) distribution of real-valued numbers.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/pair.h>
+#include <thrust/random/detail/random_core_access.h>
+#include <thrust/random/detail/normal_distribution_base.h>
+#include <iostream>
+
+namespace thrust
+{
+
+namespace random
+{
+
+
+/*! \addtogroup random_number_distributions
+ *  \{
+ */
+
+/*! \class normal_distribution
+ *  \brief A \p normal_distribution random number distribution produces floating point
+ *         Normally distributed random numbers.
+ *
+ *  \tparam RealType The type of floating point number to produce.
+ *
+ *  The following code snippet demonstrates examples of using a \p normal_distribution with a 
+ *  random number engine to produce random values drawn from the Normal distribution with a given
+ *  mean and variance:
+ *
+ *  \code
+ *  #include <thrust/random/linear_congruential_engine.h>
+ *  #include <thrust/random/normal_distribution.h>
+ *
+ *  int main(void)
+ *  {
+ *    // create a minstd_rand object to act as our source of randomness
+ *    thrust::minstd_rand rng;
+ *
+ *    // create a normal_distribution to produce floats from the Normal distribution
+ *    // with mean 2.0 and standard deviation 3.5
+ *    thrust::random::normal_distribution<float> dist(2.0f, 3.5f);
+ *
+ *    // write a random number to standard output
+ *    std::cout << dist(rng) << std::endl;
+ *
+ *    // write the mean of the distribution, just in case we forgot
+ *    std::cout << dist.mean() << std::endl;
+ *
+ *    // 2.0 is printed
+ *
+ *    // and the standard deviation
+ *    std::cout << dist.stddev() << std::endl;
+ *
+ *    // 3.5 is printed
+ *
+ *    return 0;
+ *  }
+ *  \endcode
+ */
+template<typename RealType = double>
+  class normal_distribution
+    : public detail::normal_distribution_base<RealType>::type
+{
+  private:
+    typedef typename detail::normal_distribution_base<RealType>::type super_t;
+
+  public:
+    // types
+    
+    /*! \typedef result_type
+     *  \brief The type of the floating point number produced by this \p normal_distribution.
+     */
+    typedef RealType result_type;
+
+    /*! \typedef param_type
+     *  \brief The type of the object encapsulating this \p normal_distribution's parameters.
+     */
+    typedef thrust::pair<RealType,RealType> param_type;
+
+    // constructors and reset functions
+    
+    /*! This constructor creates a new \p normal_distribution from two values defining the
+     *  half-open interval of the distribution.
+     *  
+     *  \param mean The mean (expected value) of the distribution. Defaults to \c 0.0.
+     *  \param stddev The standard deviation of the distribution. Defaults to \c 1.0.
+     */
+    __host__ __device__
+    explicit normal_distribution(RealType mean = 0.0, RealType stddev = 1.0);
+
+    /*! This constructor creates a new \p normal_distribution from a \p param_type object
+     *  encapsulating the range of the distribution.
+     *  
+     *  \param parm A \p param_type object encapsulating the parameters (i.e., the mean and standard deviation) of the distribution.
+     */
+    __host__ __device__
+    explicit normal_distribution(const param_type &parm);
+
+    /*! Calling this member function guarantees that subsequent uses of this
+     *  \p normal_distribution do not depend on values produced by any random
+     *  number generator prior to invoking this function.
+     */
+    __host__ __device__
+    void reset(void);
+
+    // generating functions
+
+    /*! This method produces a new Normal random integer drawn from this \p normal_distribution's
+     *  range using a \p UniformRandomNumberGenerator as a source of randomness.
+     *
+     *  \param urng The \p UniformRandomNumberGenerator to use as a source of randomness.
+     */
+    template<typename UniformRandomNumberGenerator>
+    __host__ __device__
+    result_type operator()(UniformRandomNumberGenerator &urng);
+
+    /*! This method produces a new Normal random integer as if by creating a new \p normal_distribution 
+     *  from the given \p param_type object, and calling its <tt>operator()</tt> method with the given
+     *  \p UniformRandomNumberGenerator as a source of randomness.
+     *
+     *  \param urng The \p UniformRandomNumberGenerator to use as a source of randomness.
+     *  \param parm A \p param_type object encapsulating the parameters of the \p normal_distribution
+     *              to draw from.
+     */
+    template<typename UniformRandomNumberGenerator>
+    __host__ __device__
+    result_type operator()(UniformRandomNumberGenerator &urng, const param_type &parm);
+
+    // property functions
+
+    /*! This method returns the value of the parameter with which this \p normal_distribution
+     *  was constructed.
+     *
+     *  \return The mean (expected value) of this \p normal_distribution's output.
+     */
+    __host__ __device__
+    result_type mean(void) const;
+
+    /*! This method returns the value of the parameter with which this \p normal_distribution
+     *  was constructed.
+     *
+     *  \return The standard deviation of this \p uniform_real_distribution's output.
+     */
+    __host__ __device__
+    result_type stddev(void) const;
+
+    /*! This method returns a \p param_type object encapsulating the parameters with which this
+     *  \p normal_distribution was constructed.
+     *
+     *  \return A \p param_type object encapsulating the parameters (i.e., the mean and standard deviation) of this \p normal_distribution.
+     */
+    __host__ __device__
+    param_type param(void) const;
+
+    /*! This method changes the parameters of this \p normal_distribution using the values encapsulated
+     *  in a given \p param_type object.
+     *
+     *  \param parm A \p param_type object encapsulating the new parameters (i.e., the mean and variance) of this \p normal_distribution.
+     */
+    __host__ __device__
+    void param(const param_type &parm);
+
+    /*! This method returns the smallest floating point number this \p normal_distribution can potentially produce.
+     *
+     *  \return The lower bound of this \p normal_distribution's half-open interval.
+     */
+    __host__ __device__
+    result_type min THRUST_PREVENT_MACRO_SUBSTITUTION (void) const;
+
+    /*! This method returns the smallest number larger than largest floating point number this \p uniform_real_distribution can potentially produce.
+     *
+     *  \return The upper bound of this \p normal_distribution's half-open interval.
+     */
+    __host__ __device__
+    result_type max THRUST_PREVENT_MACRO_SUBSTITUTION (void) const;
+
+    /*! \cond
+     */
+  private:
+    param_type m_param;
+
+    friend struct thrust::random::detail::random_core_access;
+
+    __host__ __device__
+    bool equal(const normal_distribution &rhs) const;
+
+    template<typename CharT, typename Traits>
+    std::basic_ostream<CharT,Traits>& stream_out(std::basic_ostream<CharT,Traits> &os) const;
+
+    template<typename CharT, typename Traits>
+    std::basic_istream<CharT,Traits>& stream_in(std::basic_istream<CharT,Traits> &is);
+    /*! \endcond
+     */
+}; // end normal_distribution
+
+
+/*! This function checks two \p normal_distributions for equality.
+ *  \param lhs The first \p normal_distribution to test.
+ *  \param rhs The second \p normal_distribution to test.
+ *  \return \c true if \p lhs is equal to \p rhs; \c false, otherwise.
+ */
+template<typename RealType>
+__host__ __device__
+bool operator==(const normal_distribution<RealType> &lhs,
+                const normal_distribution<RealType> &rhs);
+
+
+/*! This function checks two \p normal_distributions for inequality.
+ *  \param lhs The first \p normal_distribution to test.
+ *  \param rhs The second \p normal_distribution to test.
+ *  \return \c true if \p lhs is not equal to \p rhs; \c false, otherwise.
+ */
+template<typename RealType>
+__host__ __device__
+bool operator!=(const normal_distribution<RealType> &lhs,
+                const normal_distribution<RealType> &rhs);
+
+
+/*! This function streams a normal_distribution to a \p std::basic_ostream.
+ *  \param os The \p basic_ostream to stream out to.
+ *  \param d The \p normal_distribution to stream out.
+ *  \return \p os
+ */
+template<typename RealType,
+         typename CharT, typename Traits>
+std::basic_ostream<CharT,Traits>&
+operator<<(std::basic_ostream<CharT,Traits> &os,
+           const normal_distribution<RealType> &d);
+
+
+/*! This function streams a normal_distribution in from a std::basic_istream.
+ *  \param is The \p basic_istream to stream from.
+ *  \param d The \p normal_distribution to stream in.
+ *  \return \p is
+ */
+template<typename RealType,
+         typename CharT, typename Traits>
+std::basic_istream<CharT,Traits>&
+operator>>(std::basic_istream<CharT,Traits> &is,
+           normal_distribution<RealType> &d);
+
+
+/*! \} // end random_number_distributions
+ */
+
+
+} // end random
+
+using random::normal_distribution;
+
+} // end thrust
+
+#include <thrust/random/detail/normal_distribution.inl>
+
diff --git a/compat/thrust/random/subtract_with_carry_engine.h b/compat/thrust/random/subtract_with_carry_engine.h
new file mode 100644
index 0000000000..b88810097b
--- /dev/null
+++ b/compat/thrust/random/subtract_with_carry_engine.h
@@ -0,0 +1,256 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+/*! \file subtract_with_carry_engine.h
+ *  \brief A subtract-with-carry pseudorandom number generator
+ *         based on Marsaglia & Zaman.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/random/detail/random_core_access.h>
+
+#include <thrust/detail/cstdint.h>
+#include <cstddef> // for size_t
+#include <iostream>
+
+namespace thrust
+{
+
+namespace random
+{
+
+
+/*! \addtogroup random_number_engine_templates
+ *  \{
+ */
+
+/*! \class subtract_with_carry_engine
+ *  \brief A \p subtract_with_carry_engine random number engine produces unsigned
+ *         integer random numbers using the subtract with carry algorithm of Marsaglia & Zaman.
+ *
+ *         The generation algorithm is performed as follows:
+ *         -# Let <tt>Y = X_{i-s}- X_{i-r} - c</tt>.
+ *         -# Set <tt>X_i</tt> to <tt>y = T mod m</tt>. Set \c c to \c 1 if <tt>Y < 0</tt>, otherwise set \c c to \c 0.
+ *
+ *         This algorithm corresponds to a modular linear function of the form
+ *
+ *         <tt>TA(x_i) = (a * x_i) mod b</tt>, where \c b is of the form <tt>m^r - m^s + 1</tt> and
+ *         <tt>a = b - (b-1)/m</tt>.
+ *
+ *  \tparam UIntType The type of unsigned integer to produce.
+ *  \tparam w The word size of the produced values (<tt> w <= sizeof(UIntType)</tt>).
+ *  \tparam s The short lag of the generation algorithm.
+ *  \tparam r The long lag of the generation algorithm.
+ *
+ *  \note Inexperienced users should not use this class template directly.  Instead, use
+ *  \p ranlux24_base or \p ranlux48_base, which are instances of \p subtract_with_carry_engine.
+ *
+ *  \see thrust::random::ranlux24_base
+ *  \see thrust::random::ranlux48_base
+ */
+template<typename UIntType, size_t w, size_t s, size_t r>
+  class subtract_with_carry_engine
+{
+    /*! \cond
+     */
+  private:
+    static const UIntType modulus = UIntType(1) << w;
+    /*! \endcond
+     */
+
+  public:
+    // types
+    
+    /*! \typedef result_type
+     *  \brief The type of the unsigned integer produced by this \p subtract_with_carry_engine.
+     */
+    typedef UIntType result_type;
+
+    // engine characteristics
+
+    /*! The word size of the produced values.
+     */
+    static const size_t word_size = w;
+
+    /*! The size of the short lag used in the generation algorithm.
+     */
+    static const size_t short_lag = s;
+
+    /*! The size of the long lag used in the generation algorithm.
+     */
+    static const size_t long_lag = r;
+
+    /*! The smallest value this \p subtract_with_carry_engine may potentially produce.
+     */
+    static const result_type min = 0;
+
+    /*! The largest value this \p subtract_with_carry_engine may potentially produce.
+     */
+    static const result_type max = modulus - 1;
+
+    /*! The default seed of this \p subtract_with_carry_engine.
+     */
+    static const result_type default_seed = 19780503u;
+
+    // constructors and seeding functions
+
+    /*! This constructor, which optionally accepts a seed, initializes a new
+     *  \p subtract_with_carry_engine.
+     *  
+     *  \param value The seed used to intialize this \p subtract_with_carry_engine's state.
+     */
+    __host__ __device__
+    explicit subtract_with_carry_engine(result_type value = default_seed);
+
+    /*! This method initializes this \p subtract_with_carry_engine's state, and optionally accepts
+     *  a seed value.
+     *
+     *  \param value The seed used to initializes this \p subtract_with_carry_engine's state.
+     */
+    __host__ __device__
+    void seed(result_type value = default_seed);
+
+    // generating functions
+    
+    /*! This member function produces a new random value and updates this \p subtract_with_carry_engine's state.
+     *  \return A new random number.
+     */
+    __host__ __device__
+    result_type operator()(void);
+
+    /*! This member function advances this \p subtract_with_carry_engine's state a given number of times
+     *  and discards the results.
+     *
+     *  \param z The number of random values to discard.
+     *  \note This function is provided because an implementation may be able to accelerate it.
+     */
+    __host__ __device__
+    void discard(unsigned long long z);
+
+    /*! \cond
+     */
+  private:
+    result_type m_x[long_lag];
+    unsigned int m_k;
+    int m_carry;
+
+    friend struct thrust::random::detail::random_core_access;
+
+    __host__ __device__
+    bool equal(const subtract_with_carry_engine &rhs) const;
+
+    template<typename CharT, typename Traits>
+    std::basic_ostream<CharT,Traits>& stream_out(std::basic_ostream<CharT,Traits> &os) const;
+
+    template<typename CharT, typename Traits>
+    std::basic_istream<CharT,Traits>& stream_in(std::basic_istream<CharT,Traits> &is);
+
+    /*! \endcond
+     */
+}; // end subtract_with_carry_engine
+
+
+/*! This function checks two \p subtract_with_carry_engines for equality.
+ *  \param lhs The first \p subtract_with_carry_engine to test.
+ *  \param rhs The second \p subtract_with_carry_engine to test.
+ *  \return \c true if \p lhs is equal to \p rhs; \c false, otherwise.
+ */
+template<typename UIntType_, size_t w_, size_t s_, size_t r_>
+__host__ __device__
+bool operator==(const subtract_with_carry_engine<UIntType_,w_,s_,r_> &lhs,
+                const subtract_with_carry_engine<UIntType_,w_,s_,r_> &rhs);
+
+
+/*! This function checks two \p subtract_with_carry_engines for inequality.
+ *  \param lhs The first \p subtract_with_carry_engine to test.
+ *  \param rhs The second \p subtract_with_carry_engine to test.
+ *  \return \c true if \p lhs is not equal to \p rhs; \c false, otherwise.
+ */
+template<typename UIntType_, size_t w_, size_t s_, size_t r_>
+__host__ __device__
+bool operator!=(const subtract_with_carry_engine<UIntType_,w_,s_,r_>&lhs,
+                const subtract_with_carry_engine<UIntType_,w_,s_,r_>&rhs);
+
+
+/*! This function streams a subtract_with_carry_engine to a \p std::basic_ostream.
+ *  \param os The \p basic_ostream to stream out to.
+ *  \param e The \p subtract_with_carry_engine to stream out.
+ *  \return \p os
+ */
+template<typename UIntType_, size_t w_, size_t s_, size_t r_,
+         typename CharT, typename Traits>
+std::basic_ostream<CharT,Traits>&
+operator<<(std::basic_ostream<CharT,Traits> &os,
+           const subtract_with_carry_engine<UIntType_,w_,s_,r_> &e);
+
+
+/*! This function streams a subtract_with_carry_engine in from a std::basic_istream.
+ *  \param is The \p basic_istream to stream from.
+ *  \param e The \p subtract_with_carry_engine to stream in.
+ *  \return \p is
+ */
+template<typename UIntType_, size_t w_, size_t s_, size_t r_,
+         typename CharT, typename Traits>
+std::basic_istream<CharT,Traits>&
+operator>>(std::basic_istream<CharT,Traits> &is,
+           subtract_with_carry_engine<UIntType_,w_,s_,r_> &e);
+
+
+/*! \} // end random_number_engine_templates
+ */
+
+
+/*! \addtogroup predefined_random
+ *  \{
+ */
+
+// XXX N2111 uses uint_fast32_t here
+
+/*! \typedef ranlux24_base
+ *  \brief A random number engine with predefined parameters which implements the
+ *         base engine of the \p ranlux24 random number engine.
+ *  \note The 10000th consecutive invocation of a default-constructed object of type \p ranlux24_base
+ *        shall produce the value \c 7937952 .
+ */
+typedef subtract_with_carry_engine<thrust::detail::uint32_t, 24, 10, 24> ranlux24_base;
+
+
+// XXX N2111 uses uint_fast64_t here
+
+/*! \typedef ranlux48_base
+ *  \brief A random number engine with predefined parameters which implements the
+ *         base engine of the \p ranlux48 random number engine.
+ *  \note The 10000th consecutive invocation of a default-constructed object of type \p ranlux48_base
+ *        shall produce the value \c 192113843633948 .
+ */
+typedef subtract_with_carry_engine<thrust::detail::uint64_t, 48,  5, 12> ranlux48_base;
+
+/*! \} // end predefined_random
+ */
+
+} // end random
+
+// import names into thrust::
+using random::subtract_with_carry_engine;
+using random::ranlux24_base;
+using random::ranlux48_base;
+
+} // end thrust
+
+#include <thrust/random/detail/subtract_with_carry_engine.inl>
+
diff --git a/compat/thrust/random/uniform_int_distribution.h b/compat/thrust/random/uniform_int_distribution.h
new file mode 100644
index 0000000000..d05f7faaf3
--- /dev/null
+++ b/compat/thrust/random/uniform_int_distribution.h
@@ -0,0 +1,276 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file uniform_int_distribution.h
+ *  \brief A uniform distribution of integer-valued numbers
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/pair.h>
+#include <thrust/detail/integer_traits.h>
+#include <thrust/random/detail/random_core_access.h>
+#include <iostream>
+
+namespace thrust
+{
+
+namespace random
+{
+
+/*! \addtogroup random_number_distributions Random Number Distributions Class Templates
+ *  \ingroup random
+ *  \{
+ */
+
+/*! \class uniform_int_distribution
+ *  \brief A \p uniform_int_distribution random number distribution produces signed or unsigned integer
+ *         uniform random numbers from a given range.
+ *
+ *  \tparam IntType The type of integer to produce.
+ *
+ *  The following code snippet demonstrates examples of using a \p uniform_int_distribution with a 
+ *  random number engine to produce random integers drawn from a given range:
+ *
+ *  \code
+ *  #include <thrust/random/linear_congruential_engine.h>
+ *  #include <thrust/random/uniform_int_distribution.h>
+ *
+ *  int main(void)
+ *  {
+ *    // create a minstd_rand object to act as our source of randomness
+ *    thrust::minstd_rand rng;
+ *
+ *    // create a uniform_int_distribution to produce ints from [-7,13]
+ *    thrust::uniform_int_distribution<int> dist(-7,13);
+ *
+ *    // write a random number from the range [-7,13] to standard output
+ *    std::cout << dist(rng) << std::endl;
+ *
+ *    // write the range of the distribution, just in case we forgot
+ *    std::cout << dist.min() << std::endl;
+ *
+ *    // -7 is printed
+ *
+ *    std::cout << dist.max() << std::endl;
+ *
+ *    // 13 is printed
+ *
+ *    // write the parameters of the distribution (which happen to be the bounds) to standard output
+ *    std::cout << dist.a() << std::endl;
+ *
+ *    // -7 is printed
+ *
+ *    std::cout << dist.b() << std::endl;
+ *
+ *    // 13 is printed
+ *
+ *    return 0;
+ *  }
+ *  \endcode
+ */
+template<typename IntType = int>
+  class uniform_int_distribution
+{
+  public:
+    // types
+
+    /*! \typedef result_type
+     *  \brief The type of the integer produced by this \p uniform_int_distribution.
+     */
+    typedef IntType result_type;
+
+    /*! \typedef param_type
+     *  \brief The type of the object encapsulating this \p uniform_int_distribution's parameters.
+     */
+    typedef thrust::pair<IntType,IntType> param_type;
+
+    // constructors and reset functions
+
+    /*! This constructor creates a new \p uniform_int_distribution from two values defining the
+     *  range of the distribution.
+     *  
+     *  \param a The smallest integer to potentially produce. Defaults to \c 0.
+     *  \param b The largest integer to potentially produce. Defaults to the largest representable integer in
+     *           the platform.
+     */
+    __host__ __device__
+    explicit uniform_int_distribution(IntType a = 0, IntType b = thrust::detail::integer_traits<IntType>::const_max);
+
+    /*! This constructor creates a new \p uniform_int_distribution from a \p param_type object
+     *  encapsulating the range of the distribution.
+     *  
+     *  \param parm A \p param_type object encapsulating the parameters (i.e., the range) of the distribution.
+     */
+    __host__ __device__
+    explicit uniform_int_distribution(const param_type &parm);
+
+    /*! This does nothing.  It is included to conform to the requirements of the RandomDistribution concept.
+     */
+    __host__ __device__
+    void reset(void);
+
+    // generating functions
+
+    /*! This method produces a new uniform random integer drawn from this \p uniform_int_distribution's
+     *  range using a \p UniformRandomNumberGenerator as a source of randomness.
+     *
+     *  \param urng The \p UniformRandomNumberGenerator to use as a source of randomness.
+     */
+    template<typename UniformRandomNumberGenerator>
+    __host__ __device__
+    result_type operator()(UniformRandomNumberGenerator &urng);
+
+    /*! This method produces a new uniform random integer as if by creating a new \p uniform_int_distribution 
+     *  from the given \p param_type object, and calling its <tt>operator()</tt> method with the given
+     *  \p UniformRandomNumberGenerator as a source of randomness.
+     *
+     *  \param urng The \p UniformRandomNumberGenerator to use as a source of randomness.
+     *  \param parm A \p param_type object encapsulating the parameters of the \p uniform_int_distribution
+     *              to draw from.
+     */
+    template<typename UniformRandomNumberGenerator>
+    __host__ __device__
+    result_type operator()(UniformRandomNumberGenerator &urng, const param_type &parm);
+
+    // property functions
+    
+    /*! This method returns the value of the parameter with which this \p uniform_int_distribution
+     *  was constructed.
+     *
+     *  \return The lower bound of this \p uniform_int_distribution's range.
+     */
+    __host__ __device__
+    result_type a(void) const;
+
+    /*! This method returns the value of the parameter with which this \p uniform_int_distribution
+     *  was constructed.
+     *
+     *  \return The upper bound of this \p uniform_int_distribution's range.
+     */
+    __host__ __device__
+    result_type b(void) const;
+
+    /*! This method returns a \p param_type object encapsulating the parameters with which this
+     *  \p uniform_int_distribution was constructed.
+     *
+     *  \return A \p param_type object enapsulating the range of this \p uniform_int_distribution.
+     */
+    __host__ __device__
+    param_type param(void) const;
+
+    /*! This method changes the parameters of this \p uniform_int_distribution using the values encapsulated
+     *  in a given \p param_type object.
+     *
+     *  \param parm A \p param_type object encapsulating the new range of this \p uniform_int_distribution.
+     */
+    __host__ __device__
+    void param(const param_type &parm);
+
+    /*! This method returns the smallest integer this \p uniform_int_distribution can potentially produce.
+     *
+     *  \return The lower bound of this \p uniform_int_distribution's range.
+     */
+    __host__ __device__
+    result_type min THRUST_PREVENT_MACRO_SUBSTITUTION (void) const;
+
+    /*! This method returns the largest integer this \p uniform_int_distribution can potentially produce.
+     *
+     *  \return The upper bound of this \p uniform_int_distribution's range.
+     */
+    __host__ __device__
+    result_type max THRUST_PREVENT_MACRO_SUBSTITUTION (void) const;
+
+    /*! \cond
+     */
+  private:
+    param_type m_param;
+
+    friend struct thrust::random::detail::random_core_access;
+
+    __host__ __device__
+    bool equal(const uniform_int_distribution &rhs) const;
+
+    template<typename CharT, typename Traits>
+    std::basic_ostream<CharT,Traits>& stream_out(std::basic_ostream<CharT,Traits> &os) const;
+
+    template<typename CharT, typename Traits>
+    std::basic_istream<CharT,Traits>& stream_in(std::basic_istream<CharT,Traits> &is);
+    /*! \endcond
+     */
+}; // end uniform_int_distribution
+
+
+/*! This function checks two \p uniform_int_distributions for equality.
+ *  \param lhs The first \p uniform_int_distribution to test.
+ *  \param rhs The second \p uniform_int_distribution to test.
+ *  \return \c true if \p lhs is equal to \p rhs; \c false, otherwise.
+ */
+template<typename IntType>
+__host__ __device__
+bool operator==(const uniform_int_distribution<IntType> &lhs,
+                const uniform_int_distribution<IntType> &rhs);
+
+
+/*! This function checks two \p uniform_int_distributions for inequality.
+ *  \param lhs The first \p uniform_int_distribution to test.
+ *  \param rhs The second \p uniform_int_distribution to test.
+ *  \return \c true if \p lhs is not equal to \p rhs; \c false, otherwise.
+ */
+template<typename IntType>
+__host__ __device__
+bool operator!=(const uniform_int_distribution<IntType> &lhs,
+                const uniform_int_distribution<IntType> &rhs);
+
+
+/*! This function streams a uniform_int_distribution to a \p std::basic_ostream.
+ *  \param os The \p basic_ostream to stream out to.
+ *  \param d The \p uniform_int_distribution to stream out.
+ *  \return \p os
+ */
+template<typename IntType,
+         typename CharT, typename Traits>
+std::basic_ostream<CharT,Traits>&
+operator<<(std::basic_ostream<CharT,Traits> &os,
+           const uniform_int_distribution<IntType> &d);
+
+
+/*! This function streams a uniform_int_distribution in from a std::basic_istream.
+ *  \param is The \p basic_istream to stream from.
+ *  \param d The \p uniform_int_distribution to stream in.
+ *  \return \p is
+ */
+template<typename IntType,
+         typename CharT, typename Traits>
+std::basic_istream<CharT,Traits>&
+operator>>(std::basic_istream<CharT,Traits> &is,
+           uniform_int_distribution<IntType> &d);
+
+
+/*! \} // end random_number_distributions
+ */
+
+
+} // end random
+
+using random::uniform_int_distribution;
+
+} // end thrust
+
+#include <thrust/random/detail/uniform_int_distribution.inl>
+
diff --git a/compat/thrust/random/uniform_real_distribution.h b/compat/thrust/random/uniform_real_distribution.h
new file mode 100644
index 0000000000..ab85ab33dc
--- /dev/null
+++ b/compat/thrust/random/uniform_real_distribution.h
@@ -0,0 +1,274 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file uniform_real_distribution.h
+ *  \brief A uniform distribution of real-valued numbers
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/pair.h>
+#include <thrust/random/detail/random_core_access.h>
+#include <iostream>
+
+namespace thrust
+{
+
+namespace random
+{
+
+
+/*! \addtogroup random_number_distributions
+ *  \{
+ */
+
+/*! \class uniform_real_distribution
+ *  \brief A \p uniform_real_distribution random number distribution produces floating point
+ *         uniform random numbers from a half-open interval.
+ *
+ *  \tparam RealType The type of floating point number to produce.
+ *
+ *  The following code snippet demonstrates examples of using a \p uniform_real_distribution with a 
+ *  random number engine to produce random integers drawn from a given range:
+ *
+ *  \code
+ *  #include <thrust/random/linear_congruential_engine.h>
+ *  #include <thrust/random/uniform_real_distribution.h>
+ *
+ *  int main(void)
+ *  {
+ *    // create a minstd_rand object to act as our source of randomness
+ *    thrust::minstd_rand rng;
+ *
+ *    // create a uniform_real_distribution to produce floats from [-7,13)
+ *    thrust::uniform_real_distribution<float> dist(-7,13);
+ *
+ *    // write a random number from the range [-7,13) to standard output
+ *    std::cout << dist(rng) << std::endl;
+ *
+ *    // write the range of the distribution, just in case we forgot
+ *    std::cout << dist.min() << std::endl;
+ *
+ *    // -7.0 is printed
+ *
+ *    std::cout << dist.max() << std::endl;
+ *
+ *    // 13.0 is printed
+ *
+ *    // write the parameters of the distribution (which happen to be the bounds) to standard output
+ *    std::cout << dist.a() << std::endl;
+ *
+ *    // -7.0 is printed
+ *
+ *    std::cout << dist.b() << std::endl;
+ *
+ *    // 13.0 is printed
+ *
+ *    return 0;
+ *  }
+ *  \endcode
+ */
+template<typename RealType = double>
+  class uniform_real_distribution
+{
+  public:
+    // types
+    
+    /*! \typedef result_type
+     *  \brief The type of the floating point number produced by this \p uniform_real_distribution.
+     */
+    typedef RealType result_type;
+
+    /*! \typedef param_type
+     *  \brief The type of the object encapsulating this \p uniform_real_distribution's parameters.
+     */
+    typedef thrust::pair<RealType,RealType> param_type;
+
+    // constructors and reset functions
+    
+    /*! This constructor creates a new \p uniform_real_distribution from two values defining the
+     *  half-open interval of the distribution.
+     *  
+     *  \param a The smallest floating point number to potentially produce. Defaults to \c 0.0.
+     *  \param b The smallest number larger than the largest floating point number to potentially produce. Defaults to \c 1.0.
+     */
+    __host__ __device__
+    explicit uniform_real_distribution(RealType a = 0.0, RealType b = 1.0);
+
+    /*! This constructor creates a new \p uniform_real_distribution from a \p param_type object
+     *  encapsulating the range of the distribution.
+     *  
+     *  \param parm A \p param_type object encapsulating the parameters (i.e., the range) of the distribution.
+     */
+    __host__ __device__
+    explicit uniform_real_distribution(const param_type &parm);
+
+    /*! This does nothing.  It is included to conform to the requirements of the RandomDistribution concept.
+     */
+    __host__ __device__
+    void reset(void);
+
+    // generating functions
+
+    /*! This method produces a new uniform random integer drawn from this \p uniform_real_distribution's
+     *  range using a \p UniformRandomNumberGenerator as a source of randomness.
+     *
+     *  \param urng The \p UniformRandomNumberGenerator to use as a source of randomness.
+     */
+    template<typename UniformRandomNumberGenerator>
+    __host__ __device__
+    result_type operator()(UniformRandomNumberGenerator &urng);
+
+    /*! This method produces a new uniform random integer as if by creating a new \p uniform_real_distribution 
+     *  from the given \p param_type object, and calling its <tt>operator()</tt> method with the given
+     *  \p UniformRandomNumberGenerator as a source of randomness.
+     *
+     *  \param urng The \p UniformRandomNumberGenerator to use as a source of randomness.
+     *  \param parm A \p param_type object encapsulating the parameters of the \p uniform_real_distribution
+     *              to draw from.
+     */
+    template<typename UniformRandomNumberGenerator>
+    __host__ __device__
+    result_type operator()(UniformRandomNumberGenerator &urng, const param_type &parm);
+
+    // property functions
+
+    /*! This method returns the value of the parameter with which this \p uniform_real_distribution
+     *  was constructed.
+     *
+     *  \return The lower bound of this \p uniform_real_distribution's half-open interval.
+     */
+    __host__ __device__
+    result_type a(void) const;
+
+    /*! This method returns the value of the parameter with which this \p uniform_real_distribution
+     *  was constructed.
+     *
+     *  \return The upper bound of this \p uniform_real_distribution's half-open interval.
+     */
+    __host__ __device__
+    result_type b(void) const;
+
+    /*! This method returns a \p param_type object encapsulating the parameters with which this
+     *  \p uniform_real_distribution was constructed.
+     *
+     *  \return A \p param_type object enapsulating the half-open interval of this \p uniform_real_distribution.
+     */
+    __host__ __device__
+    param_type param(void) const;
+
+    /*! This method changes the parameters of this \p uniform_real_distribution using the values encapsulated
+     *  in a given \p param_type object.
+     *
+     *  \param parm A \p param_type object encapsulating the new half-open interval of this \p uniform_real_distribution.
+     */
+    __host__ __device__
+    void param(const param_type &parm);
+
+    /*! This method returns the smallest floating point number this \p uniform_real_distribution can potentially produce.
+     *
+     *  \return The lower bound of this \p uniform_real_distribution's half-open interval.
+     */
+    __host__ __device__
+    result_type min THRUST_PREVENT_MACRO_SUBSTITUTION (void) const;
+
+    /*! This method returns the smallest number larger than largest floating point number this \p uniform_real_distribution can potentially produce.
+     *
+     *  \return The upper bound of this \p uniform_real_distribution's half-open interval.
+     */
+    __host__ __device__
+    result_type max THRUST_PREVENT_MACRO_SUBSTITUTION (void) const;
+
+    /*! \cond
+     */
+  private:
+    param_type m_param;
+
+    friend struct thrust::random::detail::random_core_access;
+
+    __host__ __device__
+    bool equal(const uniform_real_distribution &rhs) const;
+
+    template<typename CharT, typename Traits>
+    std::basic_ostream<CharT,Traits>& stream_out(std::basic_ostream<CharT,Traits> &os) const;
+
+    template<typename CharT, typename Traits>
+    std::basic_istream<CharT,Traits>& stream_in(std::basic_istream<CharT,Traits> &is);
+    /*! \endcond
+     */
+}; // end uniform_real_distribution
+
+
+/*! This function checks two \p uniform_real_distributions for equality.
+ *  \param lhs The first \p uniform_real_distribution to test.
+ *  \param rhs The second \p uniform_real_distribution to test.
+ *  \return \c true if \p lhs is equal to \p rhs; \c false, otherwise.
+ */
+template<typename RealType>
+__host__ __device__
+bool operator==(const uniform_real_distribution<RealType> &lhs,
+                const uniform_real_distribution<RealType> &rhs);
+
+
+/*! This function checks two \p uniform_real_distributions for inequality.
+ *  \param lhs The first \p uniform_real_distribution to test.
+ *  \param rhs The second \p uniform_real_distribution to test.
+ *  \return \c true if \p lhs is not equal to \p rhs; \c false, otherwise.
+ */
+template<typename RealType>
+__host__ __device__
+bool operator!=(const uniform_real_distribution<RealType> &lhs,
+                const uniform_real_distribution<RealType> &rhs);
+
+
+/*! This function streams a uniform_real_distribution to a \p std::basic_ostream.
+ *  \param os The \p basic_ostream to stream out to.
+ *  \param d The \p uniform_real_distribution to stream out.
+ *  \return \p os
+ */
+template<typename RealType,
+         typename CharT, typename Traits>
+std::basic_ostream<CharT,Traits>&
+operator<<(std::basic_ostream<CharT,Traits> &os,
+           const uniform_real_distribution<RealType> &d);
+
+
+/*! This function streams a uniform_real_distribution in from a std::basic_istream.
+ *  \param is The \p basic_istream to stream from.
+ *  \param d The \p uniform_real_distribution to stream in.
+ *  \return \p is
+ */
+template<typename RealType,
+         typename CharT, typename Traits>
+std::basic_istream<CharT,Traits>&
+operator>>(std::basic_istream<CharT,Traits> &is,
+           uniform_real_distribution<RealType> &d);
+
+
+/*! \} // end random_number_distributions
+ */
+
+
+} // end random
+
+using random::uniform_real_distribution;
+
+} // end thrust
+
+#include <thrust/random/detail/uniform_real_distribution.inl>
+
diff --git a/compat/thrust/random/xor_combine_engine.h b/compat/thrust/random/xor_combine_engine.h
new file mode 100644
index 0000000000..61eb5a50c2
--- /dev/null
+++ b/compat/thrust/random/xor_combine_engine.h
@@ -0,0 +1,271 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+/*! \file xor_combine_engine.h
+ *  \brief A pseudorandom number generator which produces pseudorandom
+ *         numbers from two integer base engines by merging their
+ *         pseudorandom numbers with bitwise exclusive-or.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/type_traits.h>
+#include <thrust/random/detail/xor_combine_engine_max.h>
+#include <thrust/random/detail/random_core_access.h>
+#include <iostream>
+#include <cstddef> // for size_t
+
+namespace thrust
+{
+
+namespace random
+{
+
+/*! \addtogroup random_number_engine_adaptors
+ *  \{
+ */
+
+/*! \class xor_combine_engine
+ *  \brief An \p xor_combine_engine adapts two existing base random number engines and
+ *         produces random values by combining the values produced by each.
+ *
+ *  \tparam Engine1 The type of the first base random number engine to adapt.
+ *  \tparam s1 The size of the first shift to use in the generation algorithm.
+ *  \tparam Engine2 The type of the second base random number engine to adapt.
+ *  \tparam s2 The second of the second shift to use in the generation algorithm. Defaults to \c 0.
+ *
+ *  The following code snippet shows an example of using an \p xor_combine_engine instance:
+ *
+ *  \code
+ *  #include <thrust/random/linear_congruential_engine.h>
+ *  #include <thrust/random/xor_combine_engine.h>
+ *  #include <iostream>
+ *
+ *  int main(void)
+ *  {
+ *    // create an xor_combine_engine from minstd_rand and minstd_rand0
+ *    // use a shift of 0 for each
+ *    thrust::xor_combine_engine<thrust::minstd_rand,0,thrust::minstd_rand0,0> rng;
+ *
+ *    // print a random number to standard output
+ *    std::cout << rng() << std::endl;
+ *
+ *    return 0;
+ *  }
+ *  \endcode
+ */
+template<typename Engine1, size_t s1,
+         typename Engine2, size_t s2=0u>
+  class xor_combine_engine
+{
+  public:
+    // types
+
+    /*! \typedef base1_type
+     *  \brief The type of the first adapted base random number engine.
+     */
+    typedef Engine1 base1_type;
+
+    /*! \typedef base2_type
+     *  \brief The type of the second adapted base random number engine.
+     */
+    typedef Engine2 base2_type;
+
+    /*! \typedef result_type
+     *  \brief The type of the unsigned integer produced by this \p xor_combine_engine.
+     */
+    typedef typename thrust::detail::eval_if<
+      (sizeof(typename base2_type::result_type) > sizeof(typename base1_type::result_type)),
+      thrust::detail::identity_<typename base2_type::result_type>,
+      thrust::detail::identity_<typename base1_type::result_type>
+    >::type result_type;
+    
+    /*! The size of the first shift used in the generation algorithm.
+     */
+    static const size_t shift1 = s1;
+
+    /*! The size of the second shift used in the generation algorithm.
+     */
+    static const size_t shift2 = s2;
+
+    /*! The smallest value this \p xor_combine_engine may potentially produce.
+     */
+    static const result_type min = 0;
+
+    /*! The largest value this \p xor_combine_engine may potentially produce.
+     */
+    static const result_type max =
+      detail::xor_combine_engine_max<
+        Engine1, s1, Engine2, s2, result_type
+      >::value;
+
+    // constructors and seeding functions
+
+    /*! This constructor constructs a new \p xor_combine_engine and constructs
+     *  its adapted engines using their null constructors.
+     */
+    __host__ __device__
+    xor_combine_engine(void);
+
+    /*! This constructor constructs a new \p xor_combine_engine using
+     *  given \p base1_type and \p base2_type engines to initialize its adapted base engines.
+     *
+     *  \param urng1 A \p base1_type to use to initialize this \p xor_combine_engine's
+     *         first adapted base engine.
+     *  \param urng2 A \p base2_type to use to initialize this \p xor_combine_engine's
+     *         first adapted base engine.
+     */
+    __host__ __device__
+    xor_combine_engine(const base1_type &urng1, const base2_type &urng2);
+
+    /*! This constructor initializes a new \p xor_combine_engine with a given seed.
+     *  
+     *  \param s The seed used to intialize this \p xor_combine_engine's adapted base engines.
+     */
+    __host__ __device__
+    xor_combine_engine(result_type s);
+
+    /*! This method initializes the state of this \p xor_combine_engine's adapted base engines
+     *  by using their \p default_seed values.
+     */
+    __host__ __device__
+    void seed(void);
+
+    /*! This method initializes the state of this \p xor_combine_engine's adapted base engines
+     *  by using the given seed.
+     *
+     *  \param s The seed with which to intialize this \p xor_combine_engine's adapted base engines.
+     */
+    __host__ __device__
+    void seed(result_type s);
+
+    // generating functions
+
+    /*! This member function produces a new random value and updates this \p xor_combine_engine's state.
+     *  \return A new random number.
+     */
+    __host__ __device__
+    result_type operator()(void);
+
+    /*! This member function advances this \p xor_combine_engine's state a given number of times
+     *  and discards the results.
+     *
+     *  \param z The number of random values to discard.
+     *  \note This function is provided because an implementation may be able to accelerate it.
+     */
+    __host__ __device__
+    void discard(unsigned long long z);
+
+    // property functions
+
+    /*! This member function returns a const reference to this \p xor_combine_engine's
+     *  first adapted base engine.
+     *
+     *  \return A const reference to the first base engine this \p xor_combine_engine adapts.
+     */
+    __host__ __device__
+    const base1_type &base1(void) const;
+
+    /*! This member function returns a const reference to this \p xor_combine_engine's
+     *  second adapted base engine.
+     *
+     *  \return A const reference to the second base engine this \p xor_combine_engine adapts.
+     */
+    __host__ __device__
+    const base2_type &base2(void) const;
+
+    /*! \cond
+     */
+  private:
+    base1_type m_b1;
+    base2_type m_b2;
+
+    friend struct thrust::random::detail::random_core_access;
+
+    __host__ __device__
+    bool equal(const xor_combine_engine &rhs) const;
+
+    template<typename CharT, typename Traits>
+    std::basic_istream<CharT,Traits>& stream_in(std::basic_istream<CharT,Traits> &is);
+
+    template<typename CharT, typename Traits>
+    std::basic_ostream<CharT,Traits>& stream_out(std::basic_ostream<CharT,Traits> &os) const;
+
+    /*! \endcond
+     */
+}; // end xor_combine_engine
+
+
+/*! This function checks two \p xor_combine_engines for equality.
+ *  \param lhs The first \p xor_combine_engine to test.
+ *  \param rhs The second \p xor_combine_engine to test.
+ *  \return \c true if \p lhs is equal to \p rhs; \c false, otherwise.
+ */
+template<typename Engine1_, size_t s1_, typename Engine2_, size_t s2_>
+__host__ __device__
+bool operator==(const xor_combine_engine<Engine1_,s1_,Engine2_,s2_> &lhs,
+                const xor_combine_engine<Engine1_,s1_,Engine2_,s2_> &rhs);
+
+
+/*! This function checks two \p xor_combine_engines for inequality.
+ *  \param lhs The first \p xor_combine_engine to test.
+ *  \param rhs The second \p xor_combine_engine to test.
+ *  \return \c true if \p lhs is not equal to \p rhs; \c false, otherwise.
+ */
+template<typename Engine1_, size_t s1_, typename Engine2_, size_t s2_>
+__host__ __device__
+bool operator!=(const xor_combine_engine<Engine1_,s1_,Engine2_,s2_> &lhs,
+                const xor_combine_engine<Engine1_,s1_,Engine2_,s2_> &rhs);
+
+
+/*! This function streams a xor_combine_engine to a \p std::basic_ostream.
+ *  \param os The \p basic_ostream to stream out to.
+ *  \param e The \p xor_combine_engine to stream out.
+ *  \return \p os
+ */
+template<typename Engine1_, size_t s1_, typename Engine2_, size_t s2_,
+         typename CharT, typename Traits>
+std::basic_ostream<CharT,Traits>&
+operator<<(std::basic_ostream<CharT,Traits> &os,
+           const xor_combine_engine<Engine1_,s1_,Engine2_,s2_> &e);
+
+
+/*! This function streams a xor_combine_engine in from a std::basic_istream.
+ *  \param is The \p basic_istream to stream from.
+ *  \param e The \p xor_combine_engine to stream in.
+ *  \return \p is
+ */
+template<typename Engine1_, size_t s1_, typename Engine2_, size_t s2_,
+         typename CharT, typename Traits>
+std::basic_istream<CharT,Traits>&
+operator>>(std::basic_istream<CharT,Traits> &is,
+           xor_combine_engine<Engine1_,s1_,Engine2_,s2_> &e);
+
+
+/*! \} // end random_number_engine_adaptors
+ */
+
+
+} // end random
+
+// import names into thrust::
+using random::xor_combine_engine;
+
+} // end thrust
+
+#include <thrust/random/detail/xor_combine_engine.inl>
+
diff --git a/compat/thrust/reduce.h b/compat/thrust/reduce.h
new file mode 100644
index 0000000000..1dc931f9a1
--- /dev/null
+++ b/compat/thrust/reduce.h
@@ -0,0 +1,779 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file reduce.h
+ *  \brief Functions for reducing a range to a single value
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/execution_policy.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/pair.h>
+
+namespace thrust
+{
+
+
+/*! \addtogroup reductions
+ *  \{
+ */
+
+
+/*! \p reduce is a generalization of summation: it computes the sum (or some
+ *  other binary operation) of all the elements in the range <tt>[first,
+ *  last)</tt>. This version of \p reduce uses \c 0 as the initial value of the
+ *  reduction. \p reduce is similar to the C++ Standard Template Library's
+ *  <tt>std::accumulate</tt>. The primary difference between the two functions
+ *  is that <tt>std::accumulate</tt> guarantees the order of summation, while
+ *  \p reduce requires associativity of the binary operation to parallelize
+ *  the reduction.
+ *
+ *  Note that \p reduce also assumes that the binary reduction operator (in this
+ *  case operator+) is commutative.  If the reduction operator is not commutative
+ *  then \p thrust::reduce should not be used.  Instead, one could use 
+ *  \p inclusive_scan (which does not require commutativity) and select the
+ *  last element of the output array.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the sequence.
+ *  \param last The end of the sequence.
+ *  \return The result of the reduction.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>
+ *          and if \c x and \c y are objects of \p InputIterator's \c value_type,
+ *          then <tt>x + y</tt> is defined and is convertible to \p InputIterator's
+ *          \c value_type. If \c T is \c InputIterator's \c value_type, then
+ *          <tt>T(0)</tt> is defined.
+ *
+ *  The following code snippet demonstrates how to use \p reduce to compute
+ *  the sum of a sequence of integers using the \p thrust::host execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/reduce.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  int data[6] = {1, 0, 2, 2, 1, 3};
+ *  int result = thrust::reduce(thrust::host, data, data + 6);
+ *
+ *  // result == 9
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/accumulate.html
+ */
+template<typename DerivedPolicy, typename InputIterator>
+  typename thrust::iterator_traits<InputIterator>::value_type
+    reduce(const thrust::detail::execution_policy_base<DerivedPolicy> &exec, InputIterator first, InputIterator last);
+
+
+/*! \p reduce is a generalization of summation: it computes the sum (or some
+ *  other binary operation) of all the elements in the range <tt>[first,
+ *  last)</tt>. This version of \p reduce uses \c 0 as the initial value of the
+ *  reduction. \p reduce is similar to the C++ Standard Template Library's
+ *  <tt>std::accumulate</tt>. The primary difference between the two functions
+ *  is that <tt>std::accumulate</tt> guarantees the order of summation, while
+ *  \p reduce requires associativity of the binary operation to parallelize
+ *  the reduction.
+ *
+ *  Note that \p reduce also assumes that the binary reduction operator (in this
+ *  case operator+) is commutative.  If the reduction operator is not commutative
+ *  then \p thrust::reduce should not be used.  Instead, one could use 
+ *  \p inclusive_scan (which does not require commutativity) and select the
+ *  last element of the output array.
+ *
+ *  \param first The beginning of the sequence.
+ *  \param last The end of the sequence.
+ *  \return The result of the reduction.
+ *
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>
+ *          and if \c x and \c y are objects of \p InputIterator's \c value_type,
+ *          then <tt>x + y</tt> is defined and is convertible to \p InputIterator's
+ *          \c value_type. If \c T is \c InputIterator's \c value_type, then
+ *          <tt>T(0)</tt> is defined.
+ *
+ *  The following code snippet demonstrates how to use \p reduce to compute
+ *  the sum of a sequence of integers.
+ *
+ *  \code
+ *  #include <thrust/reduce.h>
+ *  ...
+ *  int data[6] = {1, 0, 2, 2, 1, 3};
+ *  int result = thrust::reduce(data, data + 6);
+ *
+ *  // result == 9
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/accumulate.html
+ */
+template<typename InputIterator> typename
+  thrust::iterator_traits<InputIterator>::value_type reduce(InputIterator first, InputIterator last);
+
+
+/*! \p reduce is a generalization of summation: it computes the sum (or some
+ *  other binary operation) of all the elements in the range <tt>[first,
+ *  last)</tt>. This version of \p reduce uses \p init as the initial value of the
+ *  reduction. \p reduce is similar to the C++ Standard Template Library's
+ *  <tt>std::accumulate</tt>. The primary difference between the two functions
+ *  is that <tt>std::accumulate</tt> guarantees the order of summation, while
+ *  \p reduce requires associativity of the binary operation to parallelize
+ *  the reduction.
+ *
+ *  Note that \p reduce also assumes that the binary reduction operator (in this
+ *  case operator+) is commutative.  If the reduction operator is not commutative
+ *  then \p thrust::reduce should not be used.  Instead, one could use 
+ *  \p inclusive_scan (which does not require commutativity) and select the
+ *  last element of the output array.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the input sequence.
+ *  \param last The end of the input sequence.
+ *  \param init The initial value.
+ *  \return The result of the reduction.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>
+ *          and if \c x and \c y are objects of \p InputIterator's \c value_type,
+ *          then <tt>x + y</tt> is defined and is convertible to \p T.
+ *  \tparam T is convertible to \p InputIterator's \c value_type.
+ *
+ *  The following code snippet demonstrates how to use \p reduce to compute
+ *  the sum of a sequence of integers including an intialization value using the \p thrust::host
+ *  execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/reduce.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  int data[6] = {1, 0, 2, 2, 1, 3};
+ *  int result = thrust::reduce(thrust::host, data, data + 6, 1);
+ *
+ *  // result == 10
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/accumulate.html
+ */
+template<typename DerivedPolicy, typename InputIterator, typename T>
+  T reduce(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+           InputIterator first,
+           InputIterator last,
+           T init);
+
+
+/*! \p reduce is a generalization of summation: it computes the sum (or some
+ *  other binary operation) of all the elements in the range <tt>[first,
+ *  last)</tt>. This version of \p reduce uses \p init as the initial value of the
+ *  reduction. \p reduce is similar to the C++ Standard Template Library's
+ *  <tt>std::accumulate</tt>. The primary difference between the two functions
+ *  is that <tt>std::accumulate</tt> guarantees the order of summation, while
+ *  \p reduce requires associativity of the binary operation to parallelize
+ *  the reduction.
+ *
+ *  Note that \p reduce also assumes that the binary reduction operator (in this
+ *  case operator+) is commutative.  If the reduction operator is not commutative
+ *  then \p thrust::reduce should not be used.  Instead, one could use 
+ *  \p inclusive_scan (which does not require commutativity) and select the
+ *  last element of the output array.
+ *
+ *  \param first The beginning of the input sequence.
+ *  \param last The end of the input sequence.
+ *  \param init The initial value.
+ *  \return The result of the reduction.
+ *
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>
+ *          and if \c x and \c y are objects of \p InputIterator's \c value_type,
+ *          then <tt>x + y</tt> is defined and is convertible to \p T.
+ *  \tparam T is convertible to \p InputIterator's \c value_type.
+ *
+ *  The following code snippet demonstrates how to use \p reduce to compute
+ *  the sum of a sequence of integers including an intialization value.
+ *
+ *  \code
+ *  #include <thrust/reduce.h>
+ *  ...
+ *  int data[6] = {1, 0, 2, 2, 1, 3};
+ *  int result = thrust::reduce(data, data + 6, 1);
+ *
+ *  // result == 10
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/accumulate.html
+ */
+template<typename InputIterator, typename T>
+  T reduce(InputIterator first,
+           InputIterator last,
+           T init);
+
+
+/*! \p reduce is a generalization of summation: it computes the sum (or some
+ *  other binary operation) of all the elements in the range <tt>[first,
+ *  last)</tt>. This version of \p reduce uses \p init as the initial value of the
+ *  reduction and \p binary_op as the binary function used for summation. \p reduce
+ *  is similar to the C++ Standard Template Library's <tt>std::accumulate</tt>.
+ *  The primary difference between the two functions is that <tt>std::accumulate</tt>
+ *  guarantees the order of summation, while \p reduce requires associativity of
+ *  \p binary_op to parallelize the reduction.
+ *
+ *  Note that \p reduce also assumes that the binary reduction operator (in this
+ *  case \p binary_op) is commutative.  If the reduction operator is not commutative
+ *  then \p thrust::reduce should not be used.  Instead, one could use 
+ *  \p inclusive_scan (which does not require commutativity) and select the
+ *  last element of the output array.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the input sequence.
+ *  \param last The end of the input sequence.
+ *  \param init The initial value.
+ *  \param binary_op The binary function used to 'sum' values.
+ *  \return The result of the reduction.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>
+ *          and \c InputIterator's \c value_type is convertible to \c T.
+ *  \tparam T is a model of <a href="http://www.sgi.com/tech/stl/Assignable.html">Assignable</a>,
+ *          and is convertible to \p BinaryFunction's \c first_argument_type and \c second_argument_type.
+ *  \tparam BinaryFunction is a model of <a href="http://www.sgi.com/tech/stl/BinaryFunction.html">Binary Function</a>,
+ *          and \p BinaryFunction's \c result_type is convertible to \p OutputType.
+ *
+ *  The following code snippet demonstrates how to use \p reduce to
+ *  compute the maximum value of a sequence of integers using the \p thrust::host execution policy
+ *  for parallelization:
+ *
+ *  \code
+ *  #include <thrust/reduce.h>
+ *  #include <thrust/functional.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  int data[6] = {1, 0, 2, 2, 1, 3};
+ *  int result = thrust::reduce(thrust::host,
+ *                              data, data + 6,
+ *                              -1,
+ *                              thrust::maximum<int>());
+ *  // result == 3
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/accumulate.html
+ *  \see transform_reduce
+ */
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename T,
+         typename BinaryFunction>
+  T reduce(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+           InputIterator first,
+           InputIterator last,
+           T init,
+           BinaryFunction binary_op);
+
+
+/*! \p reduce is a generalization of summation: it computes the sum (or some
+ *  other binary operation) of all the elements in the range <tt>[first,
+ *  last)</tt>. This version of \p reduce uses \p init as the initial value of the
+ *  reduction and \p binary_op as the binary function used for summation. \p reduce
+ *  is similar to the C++ Standard Template Library's <tt>std::accumulate</tt>.
+ *  The primary difference between the two functions is that <tt>std::accumulate</tt>
+ *  guarantees the order of summation, while \p reduce requires associativity of
+ *  \p binary_op to parallelize the reduction.
+ *
+ *  Note that \p reduce also assumes that the binary reduction operator (in this
+ *  case \p binary_op) is commutative.  If the reduction operator is not commutative
+ *  then \p thrust::reduce should not be used.  Instead, one could use 
+ *  \p inclusive_scan (which does not require commutativity) and select the
+ *  last element of the output array.
+ *
+ *  \param first The beginning of the input sequence.
+ *  \param last The end of the input sequence.
+ *  \param init The initial value.
+ *  \param binary_op The binary function used to 'sum' values.
+ *  \return The result of the reduction.
+ *
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>
+ *          and \c InputIterator's \c value_type is convertible to \c T.
+ *  \tparam T is a model of <a href="http://www.sgi.com/tech/stl/Assignable.html">Assignable</a>,
+ *          and is convertible to \p BinaryFunction's \c first_argument_type and \c second_argument_type.
+ *  \tparam BinaryFunction is a model of <a href="http://www.sgi.com/tech/stl/BinaryFunction.html">Binary Function</a>,
+ *          and \p BinaryFunction's \c result_type is convertible to \p OutputType.
+ *
+ *  The following code snippet demonstrates how to use \p reduce to
+ *  compute the maximum value of a sequence of integers.
+ *
+ *  \code
+ *  #include <thrust/reduce.h>
+ *  #include <thrust/functional.h>
+ *  ...
+ *  int data[6] = {1, 0, 2, 2, 1, 3};
+ *  int result = thrust::reduce(data, data + 6,
+ *                              -1,
+ *                              thrust::maximum<int>());
+ *  // result == 3
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/accumulate.html
+ *  \see transform_reduce
+ */
+template<typename InputIterator,
+         typename T,
+         typename BinaryFunction>
+  T reduce(InputIterator first,
+           InputIterator last,
+           T init,
+           BinaryFunction binary_op);
+
+
+/*! \p reduce_by_key is a generalization of \p reduce to key-value pairs.
+ *  For each group of consecutive keys in the range <tt>[keys_first, keys_last)</tt>
+ *  that are equal, \p reduce_by_key copies the first element of the group to the
+ *  \c keys_output. The corresponding values in the range are reduced using the
+ *  \c plus and the result copied to \c values_output. 
+ *
+ *  This version of \p reduce_by_key uses the function object \c equal_to
+ *  to test for equality and \c plus to reduce values with equal keys.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param keys_first The beginning of the input key range.
+ *  \param keys_last  The end of the input key range.
+ *  \param values_first The beginning of the input value range.
+ *  \param keys_output The beginning of the output key range.
+ *  \param values_output The beginning of the output value range.
+ *  \return A pair of iterators at end of the ranges <tt>[keys_output, keys_output_last)</tt> and <tt>[values_output, values_output_last)</tt>.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *  \tparam OutputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a> and
+ *          and \p InputIterator1's \c value_type is convertible to \c OutputIterator1's \c value_type.
+ *  \tparam OutputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a> and
+ *          and \p InputIterator2's \c value_type is convertible to \c OutputIterator2's \c value_type.
+ *
+ *  \pre The input ranges shall not overlap either output range.
+ *
+ *  The following code snippet demonstrates how to use \p reduce_by_key to
+ *  compact a sequence of key/value pairs and sum values with equal keys using the \p thrust::host
+ *  execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/reduce.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  const int N = 7;
+ *  int A[N] = {1, 3, 3, 3, 2, 2, 1}; // input keys
+ *  int B[N] = {9, 8, 7, 6, 5, 4, 3}; // input values
+ *  int C[N];                         // output keys
+ *  int D[N];                         // output values
+ *
+ *  thrust::pair<int*,int*> new_end;
+ *  new_end = thrust::reduce_by_key(thrust::host, A, A + N, B, C, D);
+ *
+ *  // The first four keys in C are now {1, 3, 2, 1} and new_end.first - C is 4.
+ *  // The first four values in D are now {9, 21, 9, 3} and new_end.second - D is 4.
+ *  \endcode
+ *  
+ *  \see reduce
+ *  \see unique_copy
+ *  \see unique_by_key
+ *  \see unique_by_key_copy
+ */
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator1,
+         typename OutputIterator2>
+  thrust::pair<OutputIterator1,OutputIterator2>
+  reduce_by_key(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                InputIterator1 keys_first, 
+                InputIterator1 keys_last,
+                InputIterator2 values_first,
+                OutputIterator1 keys_output,
+                OutputIterator2 values_output);
+
+
+/*! \p reduce_by_key is a generalization of \p reduce to key-value pairs.
+ *  For each group of consecutive keys in the range <tt>[keys_first, keys_last)</tt>
+ *  that are equal, \p reduce_by_key copies the first element of the group to the
+ *  \c keys_output. The corresponding values in the range are reduced using the
+ *  \c plus and the result copied to \c values_output. 
+ *
+ *  This version of \p reduce_by_key uses the function object \c equal_to
+ *  to test for equality and \c plus to reduce values with equal keys.
+ *
+ *  \param keys_first The beginning of the input key range.
+ *  \param keys_last  The end of the input key range.
+ *  \param values_first The beginning of the input value range.
+ *  \param keys_output The beginning of the output key range.
+ *  \param values_output The beginning of the output value range.
+ *  \return A pair of iterators at end of the ranges <tt>[keys_output, keys_output_last)</tt> and <tt>[values_output, values_output_last)</tt>.
+ *
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *  \tparam OutputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a> and
+ *          and \p InputIterator1's \c value_type is convertible to \c OutputIterator1's \c value_type.
+ *  \tparam OutputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a> and
+ *          and \p InputIterator2's \c value_type is convertible to \c OutputIterator2's \c value_type.
+ *
+ *  \pre The input ranges shall not overlap either output range.
+ *
+ *  The following code snippet demonstrates how to use \p reduce_by_key to
+ *  compact a sequence of key/value pairs and sum values with equal keys.
+ *
+ *  \code
+ *  #include <thrust/reduce.h>
+ *  ...
+ *  const int N = 7;
+ *  int A[N] = {1, 3, 3, 3, 2, 2, 1}; // input keys
+ *  int B[N] = {9, 8, 7, 6, 5, 4, 3}; // input values
+ *  int C[N];                         // output keys
+ *  int D[N];                         // output values
+ *
+ *  thrust::pair<int*,int*> new_end;
+ *  new_end = thrust::reduce_by_key(A, A + N, B, C, D);
+ *
+ *  // The first four keys in C are now {1, 3, 2, 1} and new_end.first - C is 4.
+ *  // The first four values in D are now {9, 21, 9, 3} and new_end.second - D is 4.
+ *  \endcode
+ *  
+ *  \see reduce
+ *  \see unique_copy
+ *  \see unique_by_key
+ *  \see unique_by_key_copy
+ */
+template <typename InputIterator1,
+          typename InputIterator2,
+          typename OutputIterator1,
+          typename OutputIterator2>
+  thrust::pair<OutputIterator1,OutputIterator2>
+  reduce_by_key(InputIterator1 keys_first, 
+                InputIterator1 keys_last,
+                InputIterator2 values_first,
+                OutputIterator1 keys_output,
+                OutputIterator2 values_output);
+
+
+/*! \p reduce_by_key is a generalization of \p reduce to key-value pairs.
+ *  For each group of consecutive keys in the range <tt>[keys_first, keys_last)</tt>
+ *  that are equal, \p reduce_by_key copies the first element of the group to the
+ *  \c keys_output. The corresponding values in the range are reduced using the
+ *  \c plus and the result copied to \c values_output. 
+ *
+ *  This version of \p reduce_by_key uses the function object \c binary_pred
+ *  to test for equality and \c plus to reduce values with equal keys.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param keys_first The beginning of the input key range.
+ *  \param keys_last  The end of the input key range.
+ *  \param values_first The beginning of the input value range.
+ *  \param keys_output The beginning of the output key range.
+ *  \param values_output The beginning of the output value range.
+ *  \param binary_pred  The binary predicate used to determine equality.
+ *  \return A pair of iterators at end of the ranges <tt>[keys_output, keys_output_last)</tt> and <tt>[values_output, values_output_last)</tt>.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *  \tparam OutputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a> and
+ *          and \p InputIterator1's \c value_type is convertible to \c OutputIterator1's \c value_type.
+ *  \tparam OutputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a> and
+ *          and \p InputIterator2's \c value_type is convertible to \c OutputIterator2's \c value_type.
+ *  \tparam BinaryPredicate is a model of <a href="http://www.sgi.com/tech/stl/BinaryPredicate.html">Binary Predicate</a>.
+ *
+ *  \pre The input ranges shall not overlap either output range.
+ *
+ *  The following code snippet demonstrates how to use \p reduce_by_key to
+ *  compact a sequence of key/value pairs and sum values with equal keys using the \p thrust::host
+ *  execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/reduce.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  const int N = 7;
+ *  int A[N] = {1, 3, 3, 3, 2, 2, 1}; // input keys
+ *  int B[N] = {9, 8, 7, 6, 5, 4, 3}; // input values
+ *  int C[N];                         // output keys
+ *  int D[N];                         // output values
+ *
+ *  thrust::pair<int*,int*> new_end;
+ *  thrust::equal_to<int> binary_pred;
+ *  new_end = thrust::reduce_by_key(thrust::host, A, A + N, B, C, D, binary_pred);
+ *
+ *  // The first four keys in C are now {1, 3, 2, 1} and new_end.first - C is 4.
+ *  // The first four values in D are now {9, 21, 9, 3} and new_end.second - D is 4.
+ *  \endcode
+ *  
+ *  \see reduce
+ *  \see unique_copy
+ *  \see unique_by_key
+ *  \see unique_by_key_copy
+ */
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename BinaryPredicate>
+  thrust::pair<OutputIterator1,OutputIterator2>
+  reduce_by_key(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                InputIterator1 keys_first, 
+                InputIterator1 keys_last,
+                InputIterator2 values_first,
+                OutputIterator1 keys_output,
+                OutputIterator2 values_output,
+                BinaryPredicate binary_pred);
+
+
+/*! \p reduce_by_key is a generalization of \p reduce to key-value pairs.
+ *  For each group of consecutive keys in the range <tt>[keys_first, keys_last)</tt>
+ *  that are equal, \p reduce_by_key copies the first element of the group to the
+ *  \c keys_output. The corresponding values in the range are reduced using the
+ *  \c plus and the result copied to \c values_output. 
+ *
+ *  This version of \p reduce_by_key uses the function object \c binary_pred
+ *  to test for equality and \c plus to reduce values with equal keys.
+ *
+ *  \param keys_first The beginning of the input key range.
+ *  \param keys_last  The end of the input key range.
+ *  \param values_first The beginning of the input value range.
+ *  \param keys_output The beginning of the output key range.
+ *  \param values_output The beginning of the output value range.
+ *  \param binary_pred  The binary predicate used to determine equality.
+ *  \return A pair of iterators at end of the ranges <tt>[keys_output, keys_output_last)</tt> and <tt>[values_output, values_output_last)</tt>.
+ *
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *  \tparam OutputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a> and
+ *          and \p InputIterator1's \c value_type is convertible to \c OutputIterator1's \c value_type.
+ *  \tparam OutputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a> and
+ *          and \p InputIterator2's \c value_type is convertible to \c OutputIterator2's \c value_type.
+ *  \tparam BinaryPredicate is a model of <a href="http://www.sgi.com/tech/stl/BinaryPredicate.html">Binary Predicate</a>.
+ *
+ *  \pre The input ranges shall not overlap either output range.
+ *
+ *  The following code snippet demonstrates how to use \p reduce_by_key to
+ *  compact a sequence of key/value pairs and sum values with equal keys.
+ *
+ *  \code
+ *  #include <thrust/reduce.h>
+ *  ...
+ *  const int N = 7;
+ *  int A[N] = {1, 3, 3, 3, 2, 2, 1}; // input keys
+ *  int B[N] = {9, 8, 7, 6, 5, 4, 3}; // input values
+ *  int C[N];                         // output keys
+ *  int D[N];                         // output values
+ *
+ *  thrust::pair<int*,int*> new_end;
+ *  thrust::equal_to<int> binary_pred;
+ *  new_end = thrust::reduce_by_key(A, A + N, B, C, D, binary_pred);
+ *
+ *  // The first four keys in C are now {1, 3, 2, 1} and new_end.first - C is 4.
+ *  // The first four values in D are now {9, 21, 9, 3} and new_end.second - D is 4.
+ *  \endcode
+ *  
+ *  \see reduce
+ *  \see unique_copy
+ *  \see unique_by_key
+ *  \see unique_by_key_copy
+ */
+template <typename InputIterator1,
+          typename InputIterator2,
+          typename OutputIterator1,
+          typename OutputIterator2,
+          typename BinaryPredicate>
+  thrust::pair<OutputIterator1,OutputIterator2>
+  reduce_by_key(InputIterator1 keys_first, 
+                InputIterator1 keys_last,
+                InputIterator2 values_first,
+                OutputIterator1 keys_output,
+                OutputIterator2 values_output,
+                BinaryPredicate binary_pred);
+
+
+/*! \p reduce_by_key is a generalization of \p reduce to key-value pairs.
+ *  For each group of consecutive keys in the range <tt>[keys_first, keys_last)</tt>
+ *  that are equal, \p reduce_by_key copies the first element of the group to the
+ *  \c keys_output. The corresponding values in the range are reduced using the
+ *  \c BinaryFunction \c binary_op and the result copied to \c values_output. 
+ *  Specifically, if consecutive key iterators \c i and \c (i + 1) are 
+ *  such that <tt>binary_pred(*i, *(i+1))</tt> is \c true, then the corresponding
+ *  values are reduced to a single value with \c binary_op.
+ *
+ *  This version of \p reduce_by_key uses the function object \c binary_pred
+ *  to test for equality and \c binary_op to reduce values with equal keys.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param keys_first The beginning of the input key range.
+ *  \param keys_last  The end of the input key range.
+ *  \param values_first The beginning of the input value range.
+ *  \param keys_output The beginning of the output key range.
+ *  \param values_output The beginning of the output value range.
+ *  \param binary_pred  The binary predicate used to determine equality.
+ *  \param binary_op The binary function used to accumulate values.
+ *  \return A pair of iterators at end of the ranges <tt>[keys_output, keys_output_last)</tt> and <tt>[values_output, values_output_last)</tt>.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *  \tparam OutputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a> and
+ *          and \p InputIterator1's \c value_type is convertible to \c OutputIterator1's \c value_type.
+ *  \tparam OutputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a> and
+ *          and \p InputIterator2's \c value_type is convertible to \c OutputIterator2's \c value_type.
+ *  \tparam BinaryPredicate is a model of <a href="http://www.sgi.com/tech/stl/BinaryPredicate.html">Binary Predicate</a>.
+ *  \tparam BinaryFunction is a model of <a href="http://www.sgi.com/tech/stl/BinaryFunction.html">Binary Function</a>
+ *          and \c BinaryFunction's \c result_type is convertible to \c OutputIterator2's \c value_type.
+ *
+ *  \pre The input ranges shall not overlap either output range.
+ *
+ *  The following code snippet demonstrates how to use \p reduce_by_key to
+ *  compact a sequence of key/value pairs and sum values with equal keys using the \p thrust::host
+ *  execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/reduce.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  const int N = 7;
+ *  int A[N] = {1, 3, 3, 3, 2, 2, 1}; // input keys
+ *  int B[N] = {9, 8, 7, 6, 5, 4, 3}; // input values
+ *  int C[N];                         // output keys
+ *  int D[N];                         // output values
+ *
+ *  thrust::pair<int*,int*> new_end;
+ *  thrust::equal_to<int> binary_pred;
+ *  thrust::plus<int> binary_op;
+ *  new_end = thrust::reduce_by_key(thrust::host, A, A + N, B, C, D, binary_pred, binary_op);
+ *
+ *  // The first four keys in C are now {1, 3, 2, 1} and new_end.first - C is 4.
+ *  // The first four values in D are now {9, 21, 9, 3} and new_end.second - D is 4.
+ *  \endcode
+ *  
+ *  \see reduce
+ *  \see unique_copy
+ *  \see unique_by_key
+ *  \see unique_by_key_copy
+ */
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename BinaryPredicate,
+         typename BinaryFunction>
+  thrust::pair<OutputIterator1,OutputIterator2>
+  reduce_by_key(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                InputIterator1 keys_first, 
+                InputIterator1 keys_last,
+                InputIterator2 values_first,
+                OutputIterator1 keys_output,
+                OutputIterator2 values_output,
+                BinaryPredicate binary_pred,
+                BinaryFunction binary_op);
+
+
+/*! \p reduce_by_key is a generalization of \p reduce to key-value pairs.
+ *  For each group of consecutive keys in the range <tt>[keys_first, keys_last)</tt>
+ *  that are equal, \p reduce_by_key copies the first element of the group to the
+ *  \c keys_output. The corresponding values in the range are reduced using the
+ *  \c BinaryFunction \c binary_op and the result copied to \c values_output. 
+ *  Specifically, if consecutive key iterators \c i and \c (i + 1) are 
+ *  such that <tt>binary_pred(*i, *(i+1))</tt> is \c true, then the corresponding
+ *  values are reduced to a single value with \c binary_op.
+ *
+ *  This version of \p reduce_by_key uses the function object \c binary_pred
+ *  to test for equality and \c binary_op to reduce values with equal keys.
+ *
+ *  \param keys_first The beginning of the input key range.
+ *  \param keys_last  The end of the input key range.
+ *  \param values_first The beginning of the input value range.
+ *  \param keys_output The beginning of the output key range.
+ *  \param values_output The beginning of the output value range.
+ *  \param binary_pred  The binary predicate used to determine equality.
+ *  \param binary_op The binary function used to accumulate values.
+ *  \return A pair of iterators at end of the ranges <tt>[keys_output, keys_output_last)</tt> and <tt>[values_output, values_output_last)</tt>.
+ *
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *  \tparam OutputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a> and
+ *          and \p InputIterator1's \c value_type is convertible to \c OutputIterator1's \c value_type.
+ *  \tparam OutputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a> and
+ *          and \p InputIterator2's \c value_type is convertible to \c OutputIterator2's \c value_type.
+ *  \tparam BinaryPredicate is a model of <a href="http://www.sgi.com/tech/stl/BinaryPredicate.html">Binary Predicate</a>.
+ *  \tparam BinaryFunction is a model of <a href="http://www.sgi.com/tech/stl/BinaryFunction.html">Binary Function</a>
+ *          and \c BinaryFunction's \c result_type is convertible to \c OutputIterator2's \c value_type.
+ *
+ *  \pre The input ranges shall not overlap either output range.
+ *
+ *  The following code snippet demonstrates how to use \p reduce_by_key to
+ *  compact a sequence of key/value pairs and sum values with equal keys.
+ *
+ *  \code
+ *  #include <thrust/reduce.h>
+ *  ...
+ *  const int N = 7;
+ *  int A[N] = {1, 3, 3, 3, 2, 2, 1}; // input keys
+ *  int B[N] = {9, 8, 7, 6, 5, 4, 3}; // input values
+ *  int C[N];                         // output keys
+ *  int D[N];                         // output values
+ *
+ *  thrust::pair<int*,int*> new_end;
+ *  thrust::equal_to<int> binary_pred;
+ *  thrust::plus<int> binary_op;
+ *  new_end = thrust::reduce_by_key(A, A + N, B, C, D, binary_pred, binary_op);
+ *
+ *  // The first four keys in C are now {1, 3, 2, 1} and new_end.first - C is 4.
+ *  // The first four values in D are now {9, 21, 9, 3} and new_end.second - D is 4.
+ *  \endcode
+ *  
+ *  \see reduce
+ *  \see unique_copy
+ *  \see unique_by_key
+ *  \see unique_by_key_copy
+ */
+template <typename InputIterator1,
+          typename InputIterator2,
+          typename OutputIterator1,
+          typename OutputIterator2,
+          typename BinaryPredicate,
+          typename BinaryFunction>
+  thrust::pair<OutputIterator1,OutputIterator2>
+  reduce_by_key(InputIterator1 keys_first, 
+                InputIterator1 keys_last,
+                InputIterator2 values_first,
+                OutputIterator1 keys_output,
+                OutputIterator2 values_output,
+                BinaryPredicate binary_pred,
+                BinaryFunction binary_op);
+
+
+/*! \} // end reductions
+ */
+
+
+} // end namespace thrust
+
+#include <thrust/detail/reduce.inl>
+
diff --git a/compat/thrust/remove.h b/compat/thrust/remove.h
new file mode 100644
index 0000000000..c538776258
--- /dev/null
+++ b/compat/thrust/remove.h
@@ -0,0 +1,800 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file remove.h
+ *  \brief Functions for removing elements from a range
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/execution_policy.h>
+
+namespace thrust
+{
+
+
+/*! \addtogroup stream_compaction Stream Compaction
+ *  \ingroup reordering
+ *  \{
+ *
+ */
+
+
+/*! \p remove removes from the range <tt>[first, last)</tt> all elements that are
+ *  equal to \p value. That is, \p remove returns an iterator \p new_last such
+ *  that the range <tt>[first, new_last)</tt> contains no elements equal to
+ *  \p value. The iterators in the range <tt>[new_first,last)</tt> are all still
+ *  dereferenceable, but the elements that they point to are unspecified. \p remove
+ *  is stable, meaning that the relative order of elements that are not equal to
+ *  \p value is unchanged.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the range of interest.
+ *  \param last The end of the range of interest.
+ *  \param value The value to remove from the range <tt>[first, last)</tt>.
+ *         Elements which are equal to value are removed from the sequence.
+ *  \return A \p ForwardIterator pointing to the end of the resulting range of
+ *          elements which are not equal to \p value.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>,
+ *          and \p ForwardIterator is mutable.
+ *  \tparam T is a model of <a href="http://www.sgi.com/tech/stl/EqualityComparable.html">Equality Comparable</a>,
+ *          and objects of type \p T can be compared for equality with objects of \p ForwardIterator's \c value_type.
+ *
+ *  The following code snippet demonstrates how to use \p remove to remove a number
+ *  of interest from a range using the \p thrust::host execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/remove.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  const int N = 6;
+ *  int A[N] = {3, 1, 4, 1, 5, 9};
+ *  int *new_end = thrust::remove(A, A + N, 1);
+ *  // The first four values of A are now {3, 4, 5, 9}
+ *  // Values beyond new_end are unspecified
+ *  \endcode
+ *
+ *  \note The meaning of "removal" is somewhat subtle. \p remove does not destroy any
+ *  iterators, and does not change the distance between \p first and \p last.
+ *  (There's no way that it could do anything of the sort.) So, for example, if
+ *  \c V is a device_vector, <tt>remove(V.begin(), V.end(), 0)</tt> does not
+ *  change <tt>V.size()</tt>: \c V will contain just as many elements as it did
+ *  before. \p remove returns an iterator that points to the end of the resulting
+ *  range after elements have been removed from it; it follows that the elements
+ *  after that iterator are of no interest, and may be discarded. If you are
+ *  removing elements from a
+ *  <a href="http://www.sgi.com/tech/stl/Sequence.html">Sequence</a>, you may
+ *  simply erase them. That is, a reasonable way of removing elements from a
+ *  <a href="http://www.sgi.com/tech/stl/Sequence.html">Sequence</a> is
+ *  <tt>S.erase(remove(S.begin(), S.end(), x), S.end())</tt>.
+ *
+ *  \see http://www.sgi.com/tech/stl/remove.html
+ *  \see remove_if
+ *  \see remove_copy
+ *  \see remove_copy_if
+ */
+template<typename DerivedPolicy,
+         typename ForwardIterator,
+         typename T>
+  ForwardIterator remove(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                         ForwardIterator first,
+                         ForwardIterator last,
+                         const T &value);
+
+
+/*! \p remove removes from the range <tt>[first, last)</tt> all elements that are
+ *  equal to \p value. That is, \p remove returns an iterator \p new_last such
+ *  that the range <tt>[first, new_last)</tt> contains no elements equal to
+ *  \p value. The iterators in the range <tt>[new_first,last)</tt> are all still
+ *  dereferenceable, but the elements that they point to are unspecified. \p remove
+ *  is stable, meaning that the relative order of elements that are not equal to
+ *  \p value is unchanged.
+ *
+ *  \param first The beginning of the range of interest.
+ *  \param last The end of the range of interest.
+ *  \param value The value to remove from the range <tt>[first, last)</tt>.
+ *         Elements which are equal to value are removed from the sequence.
+ *  \return A \p ForwardIterator pointing to the end of the resulting range of
+ *          elements which are not equal to \p value.
+ *
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>,
+ *          and \p ForwardIterator is mutable.
+ *  \tparam T is a model of <a href="http://www.sgi.com/tech/stl/EqualityComparable.html">Equality Comparable</a>,
+ *          and objects of type \p T can be compared for equality with objects of \p ForwardIterator's \c value_type.
+ *
+ *  The following code snippet demonstrates how to use \p remove to remove a number
+ *  of interest from a range.
+ *
+ *  \code
+ *  #include <thrust/remove.h>
+ *  ...
+ *  const int N = 6;
+ *  int A[N] = {3, 1, 4, 1, 5, 9};
+ *  int *new_end = thrust::remove(A, A + N, 1);
+ *  // The first four values of A are now {3, 4, 5, 9}
+ *  // Values beyond new_end are unspecified
+ *  \endcode
+ *
+ *  \note The meaning of "removal" is somewhat subtle. \p remove does not destroy any
+ *  iterators, and does not change the distance between \p first and \p last.
+ *  (There's no way that it could do anything of the sort.) So, for example, if
+ *  \c V is a device_vector, <tt>remove(V.begin(), V.end(), 0)</tt> does not
+ *  change <tt>V.size()</tt>: \c V will contain just as many elements as it did
+ *  before. \p remove returns an iterator that points to the end of the resulting
+ *  range after elements have been removed from it; it follows that the elements
+ *  after that iterator are of no interest, and may be discarded. If you are
+ *  removing elements from a
+ *  <a href="http://www.sgi.com/tech/stl/Sequence.html">Sequence</a>, you may
+ *  simply erase them. That is, a reasonable way of removing elements from a
+ *  <a href="http://www.sgi.com/tech/stl/Sequence.html">Sequence</a> is
+ *  <tt>S.erase(remove(S.begin(), S.end(), x), S.end())</tt>.
+ *
+ *  \see http://www.sgi.com/tech/stl/remove.html
+ *  \see remove_if
+ *  \see remove_copy
+ *  \see remove_copy_if
+ */
+template<typename ForwardIterator,
+         typename T>
+  ForwardIterator remove(ForwardIterator first,
+                         ForwardIterator last,
+                         const T &value);
+
+
+/*! \p remove_copy copies elements that are not equal to \p value from the range
+ *  <tt>[first, last)</tt> to a range beginning at \p result. The return value is
+ *  the end of the resulting range. This operation is stable, meaning that the
+ *  relative order of the elements that are copied is the same as in
+ *  the range <tt>[first, last)</tt>.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the range of interest.
+ *  \param last The end of the range of interest.
+ *  \param result The resulting range is copied to the sequence beginning at this
+ *                location.
+ *  \param value The value to omit from the copied range.
+ *  \return An OutputIterator pointing to the end of the resulting range of elements
+ *          which are not equal to \p value.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator's \c value_type is convertible to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam T is a model of <a href="http://www.sgi.com/tech/stl/EqualityComparable">Equality Comparable</a>,
+ *          and objects of type \p T can be compared for equality with objects of \p InputIterator's \c value_type.
+ *
+ *  \pre The range <tt>[first, last)</tt> shall not overlap the range <tt>[result, result + (last - first))</tt>.
+ *
+ *  The following code snippet demonstrates how to use \p remove_copy to copy
+ *  a sequence of numbers to an output range while omitting a value of interest using the \p thrust::host
+ *  execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/remove.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  const int N = 6;
+ *  int V[N] = {-2, 0, -1, 0, 1, 2};
+ *  int result[N-2];
+ *  thrust::remove_copy(thrust::host, V, V + N, result, 0);
+ *  // V remains {-2, 0, -1, 0, 1, 2}
+ *  // result is now {-2, -1, 1, 2}
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/remove_copy.html
+ *  \see remove
+ *  \see remove_if
+ *  \see remove_copy_if
+ */
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator,
+         typename T>
+  OutputIterator remove_copy(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                             InputIterator first,
+                             InputIterator last,
+                             OutputIterator result,
+                             const T &value);
+
+
+/*! \p remove_copy copies elements that are not equal to \p value from the range
+ *  <tt>[first, last)</tt> to a range beginning at \p result. The return value is
+ *  the end of the resulting range. This operation is stable, meaning that the
+ *  relative order of the elements that are copied is the same as in
+ *  the range <tt>[first, last)</tt>.
+ *
+ *  \param first The beginning of the range of interest.
+ *  \param last The end of the range of interest.
+ *  \param result The resulting range is copied to the sequence beginning at this
+ *                location.
+ *  \param value The value to omit from the copied range.
+ *  \return An OutputIterator pointing to the end of the resulting range of elements
+ *          which are not equal to \p value.
+ *
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator's \c value_type is convertible to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam T is a model of <a href="http://www.sgi.com/tech/stl/EqualityComparable">Equality Comparable</a>,
+ *          and objects of type \p T can be compared for equality with objects of \p InputIterator's \c value_type.
+ *
+ *  \pre The range <tt>[first, last)</tt> shall not overlap the range <tt>[result, result + (last - first))</tt>.
+ *
+ *  The following code snippet demonstrates how to use \p remove_copy to copy
+ *  a sequence of numbers to an output range while omitting a value of interest.
+ *
+ *  \code
+ *  #include <thrust/remove.h>
+ *  ...
+ *  const int N = 6;
+ *  int V[N] = {-2, 0, -1, 0, 1, 2};
+ *  int result[N-2];
+ *  thrust::remove_copy(V, V + N, result, 0);
+ *  // V remains {-2, 0, -1, 0, 1, 2}
+ *  // result is now {-2, -1, 1, 2}
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/remove_copy.html
+ *  \see remove
+ *  \see remove_if
+ *  \see remove_copy_if
+ */
+template<typename InputIterator,
+         typename OutputIterator,
+         typename T>
+  OutputIterator remove_copy(InputIterator first,
+                             InputIterator last,
+                             OutputIterator result,
+                             const T &value);
+
+
+/*! \p remove_if removes from the range <tt>[first, last)</tt> every element \p x
+ *  such that <tt>pred(x)</tt> is \c true. That is, \p remove_if returns an
+ *  iterator \c new_last such that the range <tt>[first,new_last)</tt> contains
+ *  no elements for which \p pred is \c true. The iterators in the range
+ *  <tt>[new_last,last)</tt> are all still dereferenceable, but the elements that
+ *  they point to are unspecified. \p remove_if is stable, meaning that the
+ *  relative order of elements that are not removed is unchanged.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the range of interest.
+ *  \param last The end of the range of interest.
+ *  \param pred A predicate to evaluate for each element of the range
+ *              <tt>[first,last)</tt>. Elements for which \p pred evaluates to
+ *              \c false are removed from the sequence.
+ *  \return A ForwardIterator pointing to the end of the resulting range of
+ *          elements for which \p pred evaluated to \c true.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/ForwardIterator.html">Forward Iterator</a>,
+ *          \p ForwardIterator is mutable,
+ *          and \p ForwardIterator's \c value_type is convertible to \p Predicate's \c argument_type.
+ *  \tparam Predicate is a model of <a href="http://www.sgi.com/tech/Predicate.html">Predicate</a>.
+ *
+ *  The following code snippet demonstrates how to use \p remove_if to remove
+ *  all even numbers from an array of integers using the \p thrust::host execution policy for
+ *  parallelization:
+ *
+ *  \code
+ *  #include <thrust/remove.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  struct is_even
+ *  {
+ *    __host__ __device__
+ *    bool operator()(const int x)
+ *    {
+ *      return (x % 2) == 0;
+ *    }
+ *  };
+ *  ...
+ *  const int N = 6;
+ *  int A[N] = {1, 4, 2, 8, 5, 7};
+ *  int *new_end = thrust::remove_if(thrust::host, A, A + N, is_even());
+ *  // The first three values of A are now {1, 5, 7}
+ *  // Values beyond new_end are unspecified
+ *  \endcode
+ *
+ *  \note The meaning of "removal" is somewhat subtle. \p remove_if does not
+ *  destroy any iterators, and does not change the distance between \p first and
+ *  \p last. (There's no way that it could do anything of the sort.) So, for
+ *  example, if \c V is a device_vector,
+ *  <tt>remove_if(V.begin(), V.end(), pred)</tt> does not change
+ *  <tt>V.size()</tt>: \c V will contain just as many elements as it did before.
+ *  \p remove_if returns an iterator that points to the end of the resulting
+ *  range after elements have been removed from it; it follows that the elements
+ *  after that iterator are of no interest, and may be discarded. If you are
+ *  removing elements from a
+ *  <a href="http://www.sgi.com/tech/stl/Sequence.html">Sequence</a>, you may
+ *  simply erase them. That is, a reasonable way of removing elements from a
+ *  <a href="http://www.sgi.com/tech/stl/Sequence.html">Sequence</a> is
+ *  <tt>S.erase(remove_if(S.begin(), S.end(), pred), S.end())</tt>.
+ *
+ *  \see http://www.sgi.com/tech/stl/remove_if.html
+ *  \see remove
+ *  \see remove_copy
+ *  \see remove_copy_if
+ */
+template<typename DerivedPolicy,
+         typename ForwardIterator,
+         typename Predicate>
+  ForwardIterator remove_if(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                            ForwardIterator first,
+                            ForwardIterator last,
+                            Predicate pred);
+
+
+/*! \p remove_if removes from the range <tt>[first, last)</tt> every element \p x
+ *  such that <tt>pred(x)</tt> is \c true. That is, \p remove_if returns an
+ *  iterator \c new_last such that the range <tt>[first,new_last)</tt> contains
+ *  no elements for which \p pred is \c true. The iterators in the range
+ *  <tt>[new_last,last)</tt> are all still dereferenceable, but the elements that
+ *  they point to are unspecified. \p remove_if is stable, meaning that the
+ *  relative order of elements that are not removed is unchanged.
+ *
+ *  \param first The beginning of the range of interest.
+ *  \param last The end of the range of interest.
+ *  \param pred A predicate to evaluate for each element of the range
+ *              <tt>[first,last)</tt>. Elements for which \p pred evaluates to
+ *              \c false are removed from the sequence.
+ *  \return A ForwardIterator pointing to the end of the resulting range of
+ *          elements for which \p pred evaluated to \c true.
+ *
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/ForwardIterator.html">Forward Iterator</a>,
+ *          \p ForwardIterator is mutable,
+ *          and \p ForwardIterator's \c value_type is convertible to \p Predicate's \c argument_type.
+ *  \tparam Predicate is a model of <a href="http://www.sgi.com/tech/Predicate.html">Predicate</a>.
+ *
+ *  The following code snippet demonstrates how to use \p remove_if to remove
+ *  all even numbers from an array of integers.
+ *
+ *  \code
+ *  #include <thrust/remove.h>
+ *  ...
+ *  struct is_even
+ *  {
+ *    __host__ __device__
+ *    bool operator()(const int x)
+ *    {
+ *      return (x % 2) == 0;
+ *    }
+ *  };
+ *  ...
+ *  const int N = 6;
+ *  int A[N] = {1, 4, 2, 8, 5, 7};
+ *  int *new_end = thrust::remove_if(A, A + N, is_even());
+ *  // The first three values of A are now {1, 5, 7}
+ *  // Values beyond new_end are unspecified
+ *  \endcode
+ *
+ *  \note The meaning of "removal" is somewhat subtle. \p remove_if does not
+ *  destroy any iterators, and does not change the distance between \p first and
+ *  \p last. (There's no way that it could do anything of the sort.) So, for
+ *  example, if \c V is a device_vector,
+ *  <tt>remove_if(V.begin(), V.end(), pred)</tt> does not change
+ *  <tt>V.size()</tt>: \c V will contain just as many elements as it did before.
+ *  \p remove_if returns an iterator that points to the end of the resulting
+ *  range after elements have been removed from it; it follows that the elements
+ *  after that iterator are of no interest, and may be discarded. If you are
+ *  removing elements from a
+ *  <a href="http://www.sgi.com/tech/stl/Sequence.html">Sequence</a>, you may
+ *  simply erase them. That is, a reasonable way of removing elements from a
+ *  <a href="http://www.sgi.com/tech/stl/Sequence.html">Sequence</a> is
+ *  <tt>S.erase(remove_if(S.begin(), S.end(), pred), S.end())</tt>.
+ *
+ *  \see http://www.sgi.com/tech/stl/remove_if.html
+ *  \see remove
+ *  \see remove_copy
+ *  \see remove_copy_if
+ */
+template<typename ForwardIterator,
+         typename Predicate>
+  ForwardIterator remove_if(ForwardIterator first,
+                            ForwardIterator last,
+                            Predicate pred);
+
+
+/*! \p remove_copy_if copies elements from the range <tt>[first,last)</tt> to a
+ *  range beginning at \p result, except that elements for which \p pred is
+ *  \c true are not copied. The return value is the end of the resulting range.
+ *  This operation is stable, meaning that the relative order of the elements that
+ *  are copied is the same as the range <tt>[first,last)</tt>.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the range of interest.
+ *  \param last The end of the range of interest.
+ *  \param result The resulting range is copied to the sequence beginning at this
+ *                location.
+ *  \param pred A predicate to evaluate for each element of the range <tt>[first,last)</tt>.
+ *              Elements for which \p pred evaluates to \c false are not copied
+ *              to the resulting sequence.
+ *  \return An OutputIterator pointing to the end of the resulting range.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator's \c value_type is convertible to a type in \p OutputIterator's set of \c value_types,
+ *          and \p InputIterator's \c value_type is convertible to \p Predicate's \c argument_type.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam Predicate is a model of <a href="http://www.sgi.com/tech/stl/Predicate.html">Predicate</a>.
+ *
+ *  \pre The range <tt>[first, last)</tt> shall not overlap the range <tt>[result, result + (last - first))</tt>.
+ *
+ *  The following code snippet demonstrates how to use \p remove_copy_if to copy
+ *  a sequence of numbers to an output range while omitting even numbers using the \p thrust::host
+ *  execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/remove.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  struct is_even
+ *  {
+ *    __host__ __device__
+ *    bool operator()(const int x)
+ *    {
+ *      return (x % 2) == 0;
+ *    }
+ *  };
+ *  ...
+ *  const int N = 6;
+ *  int V[N] = {-2, 0, -1, 0, 1, 2};
+ *  int result[2];
+ *  thrust::remove_copy_if(thrust::host, V, V + N, result, is_even());
+ *  // V remains {-2, 0, -1, 0, 1, 2}
+ *  // result is now {-1, 1}
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/remove_copy_if.html
+ *  \see remove
+ *  \see remove_copy
+ *  \see remove_if
+ */
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator,
+         typename Predicate>
+  OutputIterator remove_copy_if(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                InputIterator first,
+                                InputIterator last,
+                                OutputIterator result,
+                                Predicate pred);
+
+
+/*! \p remove_copy_if copies elements from the range <tt>[first,last)</tt> to a
+ *  range beginning at \p result, except that elements for which \p pred is
+ *  \c true are not copied. The return value is the end of the resulting range.
+ *  This operation is stable, meaning that the relative order of the elements that
+ *  are copied is the same as the range <tt>[first,last)</tt>.
+ *
+ *  \param first The beginning of the range of interest.
+ *  \param last The end of the range of interest.
+ *  \param result The resulting range is copied to the sequence beginning at this
+ *                location.
+ *  \param pred A predicate to evaluate for each element of the range <tt>[first,last)</tt>.
+ *              Elements for which \p pred evaluates to \c false are not copied
+ *              to the resulting sequence.
+ *  \return An OutputIterator pointing to the end of the resulting range.
+ *
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator's \c value_type is convertible to a type in \p OutputIterator's set of \c value_types,
+ *          and \p InputIterator's \c value_type is convertible to \p Predicate's \c argument_type.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam Predicate is a model of <a href="http://www.sgi.com/tech/stl/Predicate.html">Predicate</a>.
+ *
+ *  \pre The range <tt>[first, last)</tt> shall not overlap the range <tt>[result, result + (last - first))</tt>.
+ *
+ *  The following code snippet demonstrates how to use \p remove_copy_if to copy
+ *  a sequence of numbers to an output range while omitting even numbers.
+ *
+ *  \code
+ *  #include <thrust/remove.h>
+ *  ...
+ *  struct is_even
+ *  {
+ *    __host__ __device__
+ *    bool operator()(const int x)
+ *    {
+ *      return (x % 2) == 0;
+ *    }
+ *  };
+ *  ...
+ *  const int N = 6;
+ *  int V[N] = {-2, 0, -1, 0, 1, 2};
+ *  int result[2];
+ *  thrust::remove_copy_if(V, V + N, result, is_even());
+ *  // V remains {-2, 0, -1, 0, 1, 2}
+ *  // result is now {-1, 1}
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/remove_copy_if.html
+ *  \see remove
+ *  \see remove_copy
+ *  \see remove_if
+ */
+template<typename InputIterator,
+         typename OutputIterator,
+         typename Predicate>
+  OutputIterator remove_copy_if(InputIterator first,
+                                InputIterator last,
+                                OutputIterator result,
+                                Predicate pred);
+
+
+/*! \p remove_if removes from the range <tt>[first, last)</tt> every element \p x
+ *  such that <tt>pred(x)</tt> is \c true. That is, \p remove_if returns an
+ *  iterator \c new_last such that the range <tt>[first, new_last)</tt> contains
+ *  no elements for which \p pred of the corresponding stencil value is \c true. 
+ *  The iterators in the range <tt>[new_last,last)</tt> are all still dereferenceable,
+ *  but the elements that they point to are unspecified. \p remove_if is stable,
+ *  meaning that the relative order of elements that are not removed is unchanged.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the range of interest.
+ *  \param last The end of the range of interest.
+ *  \param stencil The beginning of the stencil sequence.
+ *  \param pred A predicate to evaluate for each element of the range
+ *              <tt>[stencil, stencil + (last - first))</tt>. Elements for which \p pred evaluates to
+ *              \c false are removed from the sequence <tt>[first, last)</tt>
+ *  \return A ForwardIterator pointing to the end of the resulting range of
+ *          elements for which \p pred evaluated to \c true.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/ForwardIterator.html">Forward Iterator</a>
+ *          and \p ForwardIterator is mutable.
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator's \c value_type is convertible to \p Predicate's \c argument_type.
+ *  \tparam Predicate is a model of <a href="http://www.sgi.com/tech/Predicate.html">Predicate</a>.
+ *
+ *  \pre The range <tt>[first, last)</tt> shall not overlap the range <tt>[result, result + (last - first))</tt>.
+ *  \pre The range <tt>[stencil, stencil + (last - first))</tt> shall not overlap the range <tt>[result, result + (last - first))</tt>.
+ *
+ *  The following code snippet demonstrates how to use \p remove_if to remove
+ *  specific elements from an array of integers using the \p thrust::host execution policy for
+ *  parallelization:
+ *
+ *  \code
+ *  #include <thrust/remove.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  const int N = 6;
+ *  int A[N] = {1, 4, 2, 8, 5, 7};
+ *  int S[N] = {0, 1, 1, 1, 0, 0};
+ *
+ *  int *new_end = thrust::remove(thrust::host, A, A + N, S, thrust::identity<int>());
+ *  // The first three values of A are now {1, 5, 7}
+ *  // Values beyond new_end are unspecified
+ *  \endcode
+ *
+ *  \note The range <tt>[first, last)</tt> is not permitted to overlap with the range <tt>[stencil, stencil + (last - first))</tt>.
+ *
+ *  \see http://www.sgi.com/tech/stl/remove_if.html
+ *  \see remove
+ *  \see remove_copy
+ *  \see remove_copy_if
+ */
+template<typename DerivedPolicy,
+         typename ForwardIterator,
+         typename InputIterator,
+         typename Predicate>
+  ForwardIterator remove_if(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                            ForwardIterator first,
+                            ForwardIterator last,
+                            InputIterator stencil,
+                            Predicate pred);
+
+
+/*! \p remove_if removes from the range <tt>[first, last)</tt> every element \p x
+ *  such that <tt>pred(x)</tt> is \c true. That is, \p remove_if returns an
+ *  iterator \c new_last such that the range <tt>[first, new_last)</tt> contains
+ *  no elements for which \p pred of the corresponding stencil value is \c true. 
+ *  The iterators in the range <tt>[new_last,last)</tt> are all still dereferenceable,
+ *  but the elements that they point to are unspecified. \p remove_if is stable,
+ *  meaning that the relative order of elements that are not removed is unchanged.
+ *
+ *  \param first The beginning of the range of interest.
+ *  \param last The end of the range of interest.
+ *  \param stencil The beginning of the stencil sequence.
+ *  \param pred A predicate to evaluate for each element of the range
+ *              <tt>[stencil, stencil + (last - first))</tt>. Elements for which \p pred evaluates to
+ *              \c false are removed from the sequence <tt>[first, last)</tt>
+ *  \return A ForwardIterator pointing to the end of the resulting range of
+ *          elements for which \p pred evaluated to \c true.
+ *
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/ForwardIterator.html">Forward Iterator</a>
+ *          and \p ForwardIterator is mutable.
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator's \c value_type is convertible to \p Predicate's \c argument_type.
+ *  \tparam Predicate is a model of <a href="http://www.sgi.com/tech/Predicate.html">Predicate</a>.
+ *
+ *  \pre The range <tt>[first, last)</tt> shall not overlap the range <tt>[result, result + (last - first))</tt>.
+ *  \pre The range <tt>[stencil, stencil + (last - first))</tt> shall not overlap the range <tt>[result, result + (last - first))</tt>.
+ *
+ *  The following code snippet demonstrates how to use \p remove_if to remove
+ *  specific elements from an array of integers.
+ *
+ *  \code
+ *  #include <thrust/remove.h>
+ *  ...
+ *  const int N = 6;
+ *  int A[N] = {1, 4, 2, 8, 5, 7};
+ *  int S[N] = {0, 1, 1, 1, 0, 0};
+ *
+ *  int *new_end = thrust::remove(A, A + N, S, thrust::identity<int>());
+ *  // The first three values of A are now {1, 5, 7}
+ *  // Values beyond new_end are unspecified
+ *  \endcode
+ *
+ *  \note The range <tt>[first, last)</tt> is not permitted to overlap with the range <tt>[stencil, stencil + (last - first))</tt>.
+ *
+ *  \see http://www.sgi.com/tech/stl/remove_if.html
+ *  \see remove
+ *  \see remove_copy
+ *  \see remove_copy_if
+ */
+template<typename ForwardIterator,
+         typename InputIterator,
+         typename Predicate>
+  ForwardIterator remove_if(ForwardIterator first,
+                            ForwardIterator last,
+                            InputIterator stencil,
+                            Predicate pred);
+
+
+/*! \p remove_copy_if copies elements from the range <tt>[first,last)</tt> to a
+ *  range beginning at \p result, except that elements for which \p pred of the 
+ *  corresponding stencil value is \c true are not copied. The return value is 
+ *  the end of the resulting range.  This operation is stable, meaning that the
+ *  relative order of the elements that are copied is the same as the 
+ *  range <tt>[first,last)</tt>.
+ *
+ *  The algorithm's execution policy is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the range of interest.
+ *  \param last The end of the range of interest.
+ *  \param stencil The beginning of the stencil sequence.
+ *  \param result The resulting range is copied to the sequence beginning at this
+ *                location.
+ *  \param pred A predicate to evaluate for each element of the range <tt>[first,last)</tt>.
+ *              Elements for which \p pred evaluates to \c false are not copied
+ *              to the resulting sequence.
+ *  \return An OutputIterator pointing to the end of the resulting range.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator1's \c value_type is convertible to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator2's \c value_type is convertible to \p Predicate's \c argument_type.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam Predicate is a model of <a href="http://www.sgi.com/tech/stl/Predicate.html">Predicate</a>.
+ *
+ *  \pre The range <tt>[stencil, stencil + (last - first))</tt> shall not overlap the range <tt>[result, result + (last - first))</tt>.
+ *
+ *  The following code snippet demonstrates how to use \p remove_copy_if to copy
+ *  a sequence of numbers to an output range while omitting specific elements using the \p thrust::host
+ *  execution policy for parallelization.
+ *
+ *  \code
+ *  #include <thrust/remove.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  const int N = 6;
+ *  int V[N] = {-2, 0, -1, 0, 1, 2};
+ *  int S[N] = { 1, 1,  0, 1, 0, 1};
+ *  int result[2];
+ *  thrust::remove_copy_if(thrust::host, V, V + N, S, result, thrust::identity<int>());
+ *  // V remains {-2, 0, -1, 0, 1, 2}
+ *  // result is now {-1, 1}
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/remove_copy_if.html
+ *  \see remove
+ *  \see remove_copy
+ *  \see remove_if
+ *  \see copy_if
+ */
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename Predicate>
+  OutputIterator remove_copy_if(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                InputIterator1 first,
+                                InputIterator1 last,
+                                InputIterator2 stencil,
+                                OutputIterator result,
+                                Predicate pred);
+
+
+/*! \p remove_copy_if copies elements from the range <tt>[first,last)</tt> to a
+ *  range beginning at \p result, except that elements for which \p pred of the 
+ *  corresponding stencil value is \c true are not copied. The return value is 
+ *  the end of the resulting range.  This operation is stable, meaning that the
+ *  relative order of the elements that are copied is the same as the 
+ *  range <tt>[first,last)</tt>.
+ *
+ *  \param first The beginning of the range of interest.
+ *  \param last The end of the range of interest.
+ *  \param stencil The beginning of the stencil sequence.
+ *  \param result The resulting range is copied to the sequence beginning at this
+ *                location.
+ *  \param pred A predicate to evaluate for each element of the range <tt>[first,last)</tt>.
+ *              Elements for which \p pred evaluates to \c false are not copied
+ *              to the resulting sequence.
+ *  \return An OutputIterator pointing to the end of the resulting range.
+ *
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator1's \c value_type is convertible to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator2's \c value_type is convertible to \p Predicate's \c argument_type.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam Predicate is a model of <a href="http://www.sgi.com/tech/stl/Predicate.html">Predicate</a>.
+ *
+ *  \pre The range <tt>[stencil, stencil + (last - first))</tt> shall not overlap the range <tt>[result, result + (last - first))</tt>.
+ *
+ *  The following code snippet demonstrates how to use \p remove_copy_if to copy
+ *  a sequence of numbers to an output range while omitting specific elements.
+ *
+ *  \code
+ *  #include <thrust/remove.h>
+ *  ...
+ *  const int N = 6;
+ *  int V[N] = {-2, 0, -1, 0, 1, 2};
+ *  int S[N] = { 1, 1,  0, 1, 0, 1};
+ *  int result[2];
+ *  thrust::remove_copy_if(V, V + N, S, result, thrust::identity<int>());
+ *  // V remains {-2, 0, -1, 0, 1, 2}
+ *  // result is now {-1, 1}
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/remove_copy_if.html
+ *  \see remove
+ *  \see remove_copy
+ *  \see remove_if
+ *  \see copy_if
+ */
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename Predicate>
+  OutputIterator remove_copy_if(InputIterator1 first,
+                                InputIterator1 last,
+                                InputIterator2 stencil,
+                                OutputIterator result,
+                                Predicate pred);
+
+
+/*! \} // end stream_compaction
+ */
+
+
+} // end thrust
+
+#include <thrust/detail/remove.inl>
+
diff --git a/compat/thrust/replace.h b/compat/thrust/replace.h
new file mode 100644
index 0000000000..48e3e49e6c
--- /dev/null
+++ b/compat/thrust/replace.h
@@ -0,0 +1,817 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file replace.h
+ *  \brief Functions for replacing elements in a range with a particular value
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/execution_policy.h>
+
+namespace thrust
+{
+
+
+/*! \addtogroup transformations
+ *  \addtogroup replacing
+ *  \ingroup transformations
+ *  \{
+ */
+
+
+/*! \p replace replaces every element in the range [first, last) equal to \p old_value
+ *  with \p new_value. That is: for every iterator \c i, if <tt>*i == old_value</tt>
+ *  then it performs the <tt>assignment *i = new_value</tt>.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the sequence of interest.
+ *  \param last The end of the sequence of interest.
+ *  \param old_value The value to replace.
+ *  \param new_value The new value to replace \p old_value.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>,
+ *          and \p ForwardIterator is mutable.
+ *  \tparam T is a model of <a href="http://www.sgi.com/tech/stl/Assignable.html>Assignable.html">Assignable</a>,
+ *          \p T is a model of <a href="http://www.sgi.com/tech/stl/EqualityComparable.html">EqualityComparable</a>,
+ *          objects of \p T may be compared for equality with objects of
+ *          \p ForwardIterator's \c value_type,
+ *          and \p T is convertible to \p ForwardIterator's \c value_type.
+ *
+ *  The following code snippet demonstrates how to use \p replace to replace
+ *  a value of interest in a \c device_vector with another using the \p thrust::device
+ *  execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/replace.h>
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/execution_policy.h>
+ *
+ *  ...
+ *  
+ *  thrust::device_vector<int> A(4);
+ *  A[0] = 1;
+ *  A[1] = 2;
+ *  A[2] = 3;
+ *  A[3] = 1;
+ *
+ *  thrust::replace(thrust::device, A.begin(), A.end(), 1, 99);
+ *
+ *  // A contains [99, 2, 3, 99]
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/replace.html
+ *  \see \c replace_if
+ *  \see \c replace_copy
+ *  \see \c replace_copy_if
+ */
+template<typename DerivedPolicy, typename ForwardIterator, typename T>
+  void replace(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+               ForwardIterator first, ForwardIterator last,
+               const T &old_value,
+               const T &new_value);
+
+
+/*! \p replace replaces every element in the range [first, last) equal to \p old_value
+ *  with \p new_value. That is: for every iterator \c i, if <tt>*i == old_value</tt>
+ *  then it performs the <tt>assignment *i = new_value</tt>.
+ *
+ *  \param first The beginning of the sequence of interest.
+ *  \param last The end of the sequence of interest.
+ *  \param old_value The value to replace.
+ *  \param new_value The new value to replace \p old_value.
+ *
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>,
+ *          and \p ForwardIterator is mutable.
+ *  \tparam T is a model of <a href="http://www.sgi.com/tech/stl/Assignable.html>Assignable.html">Assignable</a>,
+ *          \p T is a model of <a href="http://www.sgi.com/tech/stl/EqualityComparable.html">EqualityComparable</a>,
+ *          objects of \p T may be compared for equality with objects of
+ *          \p ForwardIterator's \c value_type,
+ *          and \p T is convertible to \p ForwardIterator's \c value_type.
+ *
+ *  The following code snippet demonstrates how to use \p replace to replace
+ *  a value of interest in a \c device_vector with another.
+ *
+ *  \code
+ *  #include <thrust/replace.h>
+ *  #include <thrust/device_vector.h>
+ *
+ *  ...
+ *  
+ *  thrust::device_vector<int> A(4);
+ *  A[0] = 1;
+ *  A[1] = 2;
+ *  A[2] = 3;
+ *  A[3] = 1;
+ *
+ *  thrust::replace(A.begin(), A.end(), 1, 99);
+ *
+ *  // A contains [99, 2, 3, 99]
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/replace.html
+ *  \see \c replace_if
+ *  \see \c replace_copy
+ *  \see \c replace_copy_if
+ */
+template<typename ForwardIterator, typename T>
+  void replace(ForwardIterator first, ForwardIterator last, const T &old_value,
+               const T &new_value);
+
+
+/*! \p replace_if replaces every element in the range <tt>[first, last)</tt> for which
+ *  \p pred returns \c true with \p new_value. That is: for every iterator \c i, if
+ *  <tt>pred(*i)</tt> is \c true then it performs the assignment <tt>*i = new_value</tt>.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the sequence of interest.
+ *  \param last The end of the sequence of interest.
+ *  \param pred The predicate to test on every value of the range <tt>[first,last)</tt>.
+ *  \param new_value The new value to replace elements which <tt>pred(*i)</tt> evaluates
+ *         to \c true.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>,
+ *          \p ForwardIterator is mutable,
+ *          and \p ForwardIterator's \c value_type is convertible to \p Predicate's \c argument_type.
+ *  \tparam Predicate is a model of <a href="http://www.sgi.com/tech/stl/Predicate.html">Predicate</a>.
+ *  \tparam T is a model of <a href="http://www.sgi.com/tech/stl/Assignable.html">Assignable</a>,
+ *          and \p T is convertible to \p ForwardIterator's \c value_type.
+ *
+ *  The following code snippet demonstrates how to use \p replace_if to replace
+ *  a \c device_vector's negative elements with \c 0 using the \p thrust::device execution policy
+ *  for parallelization:
+ *
+ *  \code
+ *  #include <thrust/replace.h>
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  struct is_less_than_zero
+ *  {
+ *    __host__ __device__
+ *    bool operator()(int x)
+ *    {
+ *      return x < 0;
+ *    }
+ *  };
+ *
+ *  ...
+ *  
+ *  thrust::device_vector<int> A(4);
+ *  A[0] =  1;
+ *  A[1] = -3;
+ *  A[2] =  2;
+ *  A[3] = -1;
+ *
+ *  is_less_than_zero pred;
+ *
+ *  thrust::replace_if(thrust::device, A.begin(), A.end(), pred, 0);
+ *
+ *  // A contains [1, 0, 2, 0]
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/replace_if.html
+ *  \see \c replace
+ *  \see \c replace_copy
+ *  \see \c replace_copy_if
+ */
+template<typename DerivedPolicy, typename ForwardIterator, typename Predicate, typename T>
+  void replace_if(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                  ForwardIterator first, ForwardIterator last,
+                  Predicate pred,
+                  const T &new_value);
+
+
+/*! \p replace_if replaces every element in the range <tt>[first, last)</tt> for which
+ *  \p pred returns \c true with \p new_value. That is: for every iterator \c i, if
+ *  <tt>pred(*i)</tt> is \c true then it performs the assignment <tt>*i = new_value</tt>.
+ *
+ *  \param first The beginning of the sequence of interest.
+ *  \param last The end of the sequence of interest.
+ *  \param pred The predicate to test on every value of the range <tt>[first,last)</tt>.
+ *  \param new_value The new value to replace elements which <tt>pred(*i)</tt> evaluates
+ *         to \c true.
+ *
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>,
+ *          \p ForwardIterator is mutable,
+ *          and \p ForwardIterator's \c value_type is convertible to \p Predicate's \c argument_type.
+ *  \tparam Predicate is a model of <a href="http://www.sgi.com/tech/stl/Predicate.html">Predicate</a>.
+ *  \tparam T is a model of <a href="http://www.sgi.com/tech/stl/Assignable.html">Assignable</a>,
+ *          and \p T is convertible to \p ForwardIterator's \c value_type.
+ *
+ *  The following code snippet demonstrates how to use \p replace_if to replace
+ *  a \c device_vector's negative elements with \c 0.
+ *
+ *  \code
+ *  #include <thrust/replace.h>
+ *  #include <thrust/device_vector.h>
+ *  ...
+ *  struct is_less_than_zero
+ *  {
+ *    __host__ __device__
+ *    bool operator()(int x)
+ *    {
+ *      return x < 0;
+ *    }
+ *  };
+ *
+ *  ...
+ *  
+ *  thrust::device_vector<int> A(4);
+ *  A[0] =  1;
+ *  A[1] = -3;
+ *  A[2] =  2;
+ *  A[3] = -1;
+ *
+ *  is_less_than_zero pred;
+ *
+ *  thrust::replace_if(A.begin(), A.end(), pred, 0);
+ *
+ *  // A contains [1, 0, 2, 0]
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/replace_if.html
+ *  \see \c replace
+ *  \see \c replace_copy
+ *  \see \c replace_copy_if
+ */
+template<typename ForwardIterator, typename Predicate, typename T>
+  void replace_if(ForwardIterator first, ForwardIterator last,
+                  Predicate pred,
+                  const T &new_value);
+
+
+/*! \p replace_if replaces every element in the range <tt>[first, last)</tt> for which
+ *  <tt>pred(*s)</tt> returns \c true with \p new_value. That is: for every iterator
+ *  \c i in the range <tt>[first, last)</tt>, and \c s in the range <tt>[stencil, stencil + (last - first))</tt>,
+ *  if <tt>pred(*s)</tt> is \c true then it performs the assignment <tt>*i = new_value</tt>.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the sequence of interest.
+ *  \param last The end of the sequence of interest.
+ *  \param stencil The beginning of the stencil sequence.
+ *  \param pred The predicate to test on every value of the range <tt>[first,last)</tt>.
+ *  \param new_value The new value to replace elements which <tt>pred(*i)</tt> evaluates
+ *         to \c true.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>,
+ *          and \p ForwardIterator is mutable.
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator's \c value_type is convertible to \p Predicate's \c argument_type.
+ *  \tparam Predicate is a model of <a href="http://www.sgi.com/tech/stl/Predicate.html">Predicate</a>.
+ *  \tparam T is a model of <a href="http://www.sgi.com/tech/stl/Assignable.html">Assignable</a>,
+ *          and \p T is convertible to \p ForwardIterator's \c value_type.
+ *
+ *  The following code snippet demonstrates how to use \p replace_if to replace
+ *  a \c device_vector's element with \c 0 when its corresponding stencil element is less than zero
+ *  using the \p thrust::device execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/replace.h>
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/execution_policy.h>
+ *
+ *  struct is_less_than_zero
+ *  {
+ *    __host__ __device__
+ *    bool operator()(int x)
+ *    {
+ *      return x < 0;
+ *    }
+ *  };
+ *  
+ *  ...
+ *  
+ *  thrust::device_vector<int> A(4);
+ *  A[0] =  10;
+ *  A[1] =  20;
+ *  A[2] =  30;
+ *  A[3] =  40;
+ *
+ *  thrust::device_vector<int> S(4);
+ *  S[0] = -1;
+ *  S[1] =  0;
+ *  S[2] = -1;
+ *  S[3] =  0;
+ *
+ *  is_less_than_zero pred;
+ *  thrust::replace_if(thrust::device, A.begin(), A.end(), S.begin(), pred, 0);
+ *
+ *  // A contains [0, 20, 0, 40]
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/replace_if.html
+ *  \see \c replace
+ *  \see \c replace_copy
+ *  \see \c replace_copy_if
+ */
+template<typename DerivedPolicy, typename ForwardIterator, typename InputIterator, typename Predicate, typename T>
+  void replace_if(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                  ForwardIterator first, ForwardIterator last,
+                  InputIterator stencil,
+                  Predicate pred,
+                  const T &new_value);
+
+
+/*! \p replace_if replaces every element in the range <tt>[first, last)</tt> for which
+ *  <tt>pred(*s)</tt> returns \c true with \p new_value. That is: for every iterator
+ *  \c i in the range <tt>[first, last)</tt>, and \c s in the range <tt>[stencil, stencil + (last - first))</tt>,
+ *  if <tt>pred(*s)</tt> is \c true then it performs the assignment <tt>*i = new_value</tt>.
+ *
+ *  \param first The beginning of the sequence of interest.
+ *  \param last The end of the sequence of interest.
+ *  \param stencil The beginning of the stencil sequence.
+ *  \param pred The predicate to test on every value of the range <tt>[first,last)</tt>.
+ *  \param new_value The new value to replace elements which <tt>pred(*i)</tt> evaluates
+ *         to \c true.
+ *
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>,
+ *          and \p ForwardIterator is mutable.
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator's \c value_type is convertible to \p Predicate's \c argument_type.
+ *  \tparam Predicate is a model of <a href="http://www.sgi.com/tech/stl/Predicate.html">Predicate</a>.
+ *  \tparam T is a model of <a href="http://www.sgi.com/tech/stl/Assignable.html">Assignable</a>,
+ *          and \p T is convertible to \p ForwardIterator's \c value_type.
+ *
+ *  The following code snippet demonstrates how to use \p replace_if to replace
+ *  a \c device_vector's element with \c 0 when its corresponding stencil element is less than zero.
+ *
+ *  \code
+ *  #include <thrust/replace.h>
+ *  #include <thrust/device_vector.h>
+ *
+ *  struct is_less_than_zero
+ *  {
+ *    __host__ __device__
+ *    bool operator()(int x)
+ *    {
+ *      return x < 0;
+ *    }
+ *  };
+ *  
+ *  ...
+ *  
+ *  thrust::device_vector<int> A(4);
+ *  A[0] =  10;
+ *  A[1] =  20;
+ *  A[2] =  30;
+ *  A[3] =  40;
+ *
+ *  thrust::device_vector<int> S(4);
+ *  S[0] = -1;
+ *  S[1] =  0;
+ *  S[2] = -1;
+ *  S[3] =  0;
+ *
+ *  is_less_than_zero pred;
+ *  thrust::replace_if(A.begin(), A.end(), S.begin(), pred, 0);
+ *
+ *  // A contains [0, 20, 0, 40]
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/replace_if.html
+ *  \see \c replace
+ *  \see \c replace_copy
+ *  \see \c replace_copy_if
+ */
+template<typename ForwardIterator, typename InputIterator, typename Predicate, typename T>
+  void replace_if(ForwardIterator first, ForwardIterator last,
+                  InputIterator stencil,
+                  Predicate pred,
+                  const T &new_value);
+
+
+/*! \p replace_copy copies elements from the range <tt>[first, last)</tt> to the range
+ *  <tt>[result, result + (last-first))</tt>, except that any element equal to \p old_value
+ *  is not copied; \p new_value is copied instead.
+ *
+ *  More precisely, for every integer \c n such that <tt>0 <= n < last-first</tt>, \p replace_copy
+ *  performs the assignment <tt>*(result+n) = new_value</tt> if <tt>*(first+n) == old_value</tt>,
+ *  and <tt>*(result+n) = *(first+n)</tt> otherwise.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the sequence to copy from.
+ *  \param last The end of the sequence to copy from.
+ *  \param result The beginning of the sequence to copy to.
+ *  \param old_value The value to replace.
+ *  \param new_value The replacement value for which <tt>*i == old_value</tt> evaluates to \c true.
+ *  \return <tt>result + (last-first)</tt>
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam T is a model of <a href="http://www.sgi.com/tech/stl/Assignable.html">Assignable</a>,
+ *          \p T is a model of <a href="http://www.sgi.com/tech/stl/EqualityComparable.html">Equality Comparable</a>,
+ *          \p T may be compared for equality with \p InputIterator's \c value_type,
+ *          and \p T is convertible to \p OutputIterator's \c value_type.
+ *
+ *  \pre \p first may equal \p result, but the ranges <tt>[first, last)</tt> and <tt>[result, result + (last - first))</tt> shall not overlap otherwise.
+ *
+ *  \code
+ *  #include <thrust/replace.h>
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  thrust::device_vector<int> A(4);
+ *  A[0] = 1;
+ *  A[1] = 2;
+ *  A[2] = 3;
+ *  A[3] = 1;
+ *
+ *  thrust::device_vector<int> B(4);
+ *
+ *  thrust::replace_copy(thrust::device, A.begin(), A.end(), B.begin(), 1, 99);
+ *
+ *  // B contains [99, 2, 3, 99]
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/replace_copy.html
+ *  \see \c copy
+ *  \see \c replace
+ *  \see \c replace_if
+ *  \see \c replace_copy_if
+ */
+template<typename DerivedPolicy, typename InputIterator, typename OutputIterator, typename T>
+  OutputIterator replace_copy(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                              InputIterator first, InputIterator last,
+                              OutputIterator result,
+                              const T &old_value,
+                              const T &new_value);
+
+
+/*! \p replace_copy copies elements from the range <tt>[first, last)</tt> to the range
+ *  <tt>[result, result + (last-first))</tt>, except that any element equal to \p old_value
+ *  is not copied; \p new_value is copied instead.
+ *
+ *  More precisely, for every integer \c n such that <tt>0 <= n < last-first</tt>, \p replace_copy
+ *  performs the assignment <tt>*(result+n) = new_value</tt> if <tt>*(first+n) == old_value</tt>,
+ *  and <tt>*(result+n) = *(first+n)</tt> otherwise.
+ *
+ *  \param first The beginning of the sequence to copy from.
+ *  \param last The end of the sequence to copy from.
+ *  \param result The beginning of the sequence to copy to.
+ *  \param old_value The value to replace.
+ *  \param new_value The replacement value for which <tt>*i == old_value</tt> evaluates to \c true.
+ *  \return <tt>result + (last-first)</tt>
+ *
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam T is a model of <a href="http://www.sgi.com/tech/stl/Assignable.html">Assignable</a>,
+ *          \p T is a model of <a href="http://www.sgi.com/tech/stl/EqualityComparable.html">Equality Comparable</a>,
+ *          \p T may be compared for equality with \p InputIterator's \c value_type,
+ *          and \p T is convertible to \p OutputIterator's \c value_type.
+ *
+ *  \pre \p first may equal \p result, but the ranges <tt>[first, last)</tt> and <tt>[result, result + (last - first))</tt> shall not overlap otherwise.
+ *
+ *  \code
+ *  #include <thrust/replace.h>
+ *  #include <thrust/device_vector.h>
+ *  ...
+ *  thrust::device_vector<int> A(4);
+ *  A[0] = 1;
+ *  A[1] = 2;
+ *  A[2] = 3;
+ *  A[3] = 1;
+ *
+ *  thrust::device_vector<int> B(4);
+ *
+ *  thrust::replace_copy(A.begin(), A.end(), B.begin(), 1, 99);
+ *
+ *  // B contains [99, 2, 3, 99]
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/replace_copy.html
+ *  \see \c copy
+ *  \see \c replace
+ *  \see \c replace_if
+ *  \see \c replace_copy_if
+ */
+template<typename InputIterator, typename OutputIterator, typename T>
+  OutputIterator replace_copy(InputIterator first, InputIterator last,
+                              OutputIterator result, const T &old_value,
+                              const T &new_value);
+
+
+/*! \p replace_copy_if copies elements from the range <tt>[first, last)</tt> to the range
+ *  <tt>[result, result + (last-first))</tt>, except that any element for which \p pred
+ *  is \c true is not copied; \p new_value is copied instead.
+ *
+ *  More precisely, for every integer \c n such that 0 <= n < last-first,
+ *  \p replace_copy_if performs the assignment <tt>*(result+n) = new_value</tt> if
+ *  <tt>pred(*(first+n))</tt>, and <tt>*(result+n) = *(first+n)</tt> otherwise.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the sequence to copy from.
+ *  \param last The end of the sequence to copy from.
+ *  \param result The beginning of the sequence to copy to.
+ *  \param pred The predicate to test on every value of the range <tt>[first,last)</tt>.
+ *  \param new_value The replacement value to assign <tt>pred(*i)</tt> evaluates to \c true.
+ *  \return <tt>result + (last-first)</tt>
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator's \c value_type is convertible to \p Predicate's \c argument_type.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam Predicate is a model of <a href="http://www.sgi.com/tech/stl/Predicate.html">Predicate</a>.
+ *  \tparam T is a model of <a href="http://www.sgi.com/tech/stl/Assignable.html">Assignable</a>,
+ *          and \p T is convertible to \p OutputIterator's \c value_type.
+ *
+ *  \pre \p first may equal \p result, but the ranges <tt>[first, last)</tt> and <tt>[result, result + (last - first))</tt> shall not overlap otherwise.
+ *
+ *  \code
+ *  #include <thrust/replace.h>
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/execution_policy.h>
+ *
+ *  struct is_less_than_zero
+ *  {
+ *    __host__ __device__
+ *    bool operator()(int x)
+ *    {
+ *      return x < 0;
+ *    }
+ *  };
+ *
+ *  ...
+ *  
+ *  thrust::device_vector<int> A(4);
+ *  A[0] =  1;
+ *  A[1] = -3;
+ *  A[2] =  2;
+ *  A[3] = -1;
+ 
+ *  thrust::device_vector<int> B(4);
+ *  is_less_than_zero pred;
+ *
+ *  thrust::replace_copy_if(thrust::device, A.begin(), A.end(), B.begin(), pred, 0);
+ *
+ *  // B contains [1, 0, 2, 0]
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/replace_copy_if.html
+ *  \see \c replace
+ *  \see \c replace_if
+ *  \see \c replace_copy
+ */
+template<typename DerivedPolicy, typename InputIterator, typename OutputIterator, typename Predicate, typename T>
+  OutputIterator replace_copy_if(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                 InputIterator first, InputIterator last,
+                                 OutputIterator result,
+                                 Predicate pred,
+                                 const T &new_value);
+
+
+/*! \p replace_copy_if copies elements from the range <tt>[first, last)</tt> to the range
+ *  <tt>[result, result + (last-first))</tt>, except that any element for which \p pred
+ *  is \c true is not copied; \p new_value is copied instead.
+ *
+ *  More precisely, for every integer \c n such that 0 <= n < last-first,
+ *  \p replace_copy_if performs the assignment <tt>*(result+n) = new_value</tt> if
+ *  <tt>pred(*(first+n))</tt>, and <tt>*(result+n) = *(first+n)</tt> otherwise.
+ *
+ *  \param first The beginning of the sequence to copy from.
+ *  \param last The end of the sequence to copy from.
+ *  \param result The beginning of the sequence to copy to.
+ *  \param pred The predicate to test on every value of the range <tt>[first,last)</tt>.
+ *  \param new_value The replacement value to assign <tt>pred(*i)</tt> evaluates to \c true.
+ *  \return <tt>result + (last-first)</tt>
+ *
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator's \c value_type is convertible to \p Predicate's \c argument_type.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam Predicate is a model of <a href="http://www.sgi.com/tech/stl/Predicate.html">Predicate</a>.
+ *  \tparam T is a model of <a href="http://www.sgi.com/tech/stl/Assignable.html">Assignable</a>,
+ *          and \p T is convertible to \p OutputIterator's \c value_type.
+ *
+ *  \pre \p first may equal \p result, but the ranges <tt>[first, last)</tt> and <tt>[result, result + (last - first))</tt> shall not overlap otherwise.
+ *
+ *  \code
+ *  #include <thrust/replace.h>
+ *  #include <thrust/device_vector.h>
+ *
+ *  struct is_less_than_zero
+ *  {
+ *    __host__ __device__
+ *    bool operator()(int x)
+ *    {
+ *      return x < 0;
+ *    }
+ *  };
+ *
+ *  ...
+ *  
+ *  thrust::device_vector<int> A(4);
+ *  A[0] =  1;
+ *  A[1] = -3;
+ *  A[2] =  2;
+ *  A[3] = -1;
+ 
+ *  thrust::device_vector<int> B(4);
+ *  is_less_than_zero pred;
+ *
+ *  thrust::replace_copy_if(A.begin(), A.end(), B.begin(), pred, 0);
+ *
+ *  // B contains [1, 0, 2, 0]
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/replace_copy_if.html
+ *  \see \c replace
+ *  \see \c replace_if
+ *  \see \c replace_copy
+ */
+template<typename InputIterator, typename OutputIterator, typename Predicate, typename T>
+  OutputIterator replace_copy_if(InputIterator first, InputIterator last,
+                                 OutputIterator result,
+                                 Predicate pred,
+                                 const T &new_value);
+
+
+/*! This version of \p replace_copy_if copies elements from the range <tt>[first, last)</tt> to the range
+ *  <tt>[result, result + (last-first))</tt>, except that any element whose corresponding stencil
+ *  element causes \p pred to be \c true is not copied; \p new_value is copied instead.
+ *
+ *  More precisely, for every integer \c n such that <tt>0 <= n < last-first</tt>,
+ *  \p replace_copy_if performs the assignment <tt>*(result+n) = new_value</tt> if
+ *  <tt>pred(*(stencil+n))</tt>, and <tt>*(result+n) = *(first+n)</tt> otherwise.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the sequence to copy from.
+ *  \param last The end of the sequence to copy from.
+ *  \param stencil The beginning of the stencil sequence.
+ *  \param result The beginning of the sequence to copy to.
+ *  \param pred The predicate to test on every value of the range <tt>[stencil, stencil + (last - first))</tt>.
+ *  \param new_value The replacement value to assign when <tt>pred(*s)</tt> evaluates to \c true. 
+ *  \return <tt>result + (last-first)</tt>
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>
+ *                         and \p InputIterator2's \c value_type is convertible to \p Predicate's \c argument_type.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam Predicate is a model of <a href="http://www.sgi.com/tech/stl/Predicate.html">Predicate</a>.
+ *  \tparam T is a model of <a href="http://www.sgi.com/tech/stl/Assignable.html">Assignable</a>,
+ *          and \p T is convertible to \p OutputIterator's \c value_type.
+ *
+ *  \pre \p first may equal \p result, but the ranges <tt>[first, last)</tt> and <tt>[result, result + (last - first))</tt> shall not overlap otherwise.
+ *  \pre \p stencil may equal \p result, but the ranges <tt>[stencil, stencil + (last - first))</tt> and <tt>[result, result + (last - first))</tt> shall not overlap otherwise.
+ *
+ *  \code
+ *  #include <thrust/replace.h>
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/execution_policy.h>
+ *
+ *  struct is_less_than_zero
+ *  {
+ *    __host__ __device__
+ *    bool operator()(int x)
+ *    {
+ *      return x < 0;
+ *    }
+ *  };
+ *  
+ *  ...
+ *  
+ *  thrust::device_vector<int> A(4);
+ *  A[0] =  10;
+ *  A[1] =  20;
+ *  A[2] =  30;
+ *  A[3] =  40;
+ *
+ *  thrust::device_vector<int> S(4);
+ *  S[0] = -1;
+ *  S[1] =  0;
+ *  S[2] = -1;
+ *  S[3] =  0;
+ *
+ *  thrust::device_vector<int> B(4);
+ *  is_less_than_zero pred;
+ *
+ *  thrust::replace_if(thrust::device, A.begin(), A.end(), S.begin(), B.begin(), pred, 0);
+ *
+ *  // B contains [0, 20, 0, 40]
+ *  \endcode
+ *
+ *  \see \c replace_copy
+ *  \see \c replace_if
+ */
+template<typename DerivedPolicy, typename InputIterator1, typename InputIterator2, typename OutputIterator, typename Predicate, typename T>
+  OutputIterator replace_copy_if(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                 InputIterator1 first, InputIterator1 last,
+                                 InputIterator2 stencil,
+                                 OutputIterator result,
+                                 Predicate pred,
+                                 const T &new_value);
+
+
+/*! This version of \p replace_copy_if copies elements from the range <tt>[first, last)</tt> to the range
+ *  <tt>[result, result + (last-first))</tt>, except that any element whose corresponding stencil
+ *  element causes \p pred to be \c true is not copied; \p new_value is copied instead.
+ *
+ *  More precisely, for every integer \c n such that <tt>0 <= n < last-first</tt>,
+ *  \p replace_copy_if performs the assignment <tt>*(result+n) = new_value</tt> if
+ *  <tt>pred(*(stencil+n))</tt>, and <tt>*(result+n) = *(first+n)</tt> otherwise.
+ *
+ *  \param first The beginning of the sequence to copy from.
+ *  \param last The end of the sequence to copy from.
+ *  \param stencil The beginning of the stencil sequence.
+ *  \param result The beginning of the sequence to copy to.
+ *  \param pred The predicate to test on every value of the range <tt>[stencil, stencil + (last - first))</tt>.
+ *  \param new_value The replacement value to assign when <tt>pred(*s)</tt> evaluates to \c true. 
+ *  \return <tt>result + (last-first)</tt>
+ *
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>
+ *                         and \p InputIterator2's \c value_type is convertible to \p Predicate's \c argument_type.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam Predicate is a model of <a href="http://www.sgi.com/tech/stl/Predicate.html">Predicate</a>.
+ *  \tparam T is a model of <a href="http://www.sgi.com/tech/stl/Assignable.html">Assignable</a>,
+ *          and \p T is convertible to \p OutputIterator's \c value_type.
+ *
+ *  \pre \p first may equal \p result, but the ranges <tt>[first, last)</tt> and <tt>[result, result + (last - first))</tt> shall not overlap otherwise.
+ *  \pre \p stencil may equal \p result, but the ranges <tt>[stencil, stencil + (last - first))</tt> and <tt>[result, result + (last - first))</tt> shall not overlap otherwise.
+ *
+ *  \code
+ *  #include <thrust/replace.h>
+ *  #include <thrust/device_vector.h>
+ *
+ *  struct is_less_than_zero
+ *  {
+ *    __host__ __device__
+ *    bool operator()(int x)
+ *    {
+ *      return x < 0;
+ *    }
+ *  };
+ *  
+ *  ...
+ *  
+ *  thrust::device_vector<int> A(4);
+ *  A[0] =  10;
+ *  A[1] =  20;
+ *  A[2] =  30;
+ *  A[3] =  40;
+ *
+ *  thrust::device_vector<int> S(4);
+ *  S[0] = -1;
+ *  S[1] =  0;
+ *  S[2] = -1;
+ *  S[3] =  0;
+ *
+ *  thrust::device_vector<int> B(4);
+ *  is_less_than_zero pred;
+ *
+ *  thrust::replace_if(A.begin(), A.end(), S.begin(), B.begin(), pred, 0);
+ *
+ *  // B contains [0, 20, 0, 40]
+ *  \endcode
+ *
+ *  \see \c replace_copy
+ *  \see \c replace_if
+ */
+template<typename InputIterator1, typename InputIterator2, typename OutputIterator, typename Predicate, typename T>
+  OutputIterator replace_copy_if(InputIterator1 first, InputIterator1 last,
+                                 InputIterator2 stencil,
+                                 OutputIterator result,
+                                 Predicate pred,
+                                 const T &new_value);
+
+
+/*! \} // end replacing
+ *  \} // transformations
+ */
+
+
+} // end thrust
+
+#include <thrust/detail/replace.inl>
+
diff --git a/compat/thrust/reverse.h b/compat/thrust/reverse.h
new file mode 100644
index 0000000000..ba50c5d05a
--- /dev/null
+++ b/compat/thrust/reverse.h
@@ -0,0 +1,213 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file reverse.h
+ *  \brief Reverses the order of a range
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/execution_policy.h>
+
+namespace thrust
+{
+
+
+/*! \addtogroup reordering
+ *  \ingroup algorithms
+ */
+
+
+/*! \p reverse reverses a range. That is: for every <tt>i</tt> such that
+ *  <tt>0 <= i <= (last - first) / 2</tt>, it exchanges <tt>*(first + i)</tt>
+ *  and <tt>*(last - (i + 1))</tt>.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the range to reverse.
+ *  \param last The end of the range to reverse.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam BidirectionalIterator is a model of <a href="http://www.sgi.com/tech/stl/BidirectionalIterator.html">Bidirectional Iterator</a> and
+ *          \p BidirectionalIterator is mutable.
+ *
+ *  The following code snippet demonstrates how to use \p reverse to reverse a
+ *  \p device_vector of integers using the \p thrust::device execution policy for
+ *  parallelization:
+ *
+ *  \code
+ *  #include <thrust/reverse.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  const int N = 6;
+ *  int data[N] = {0, 1, 2, 3, 4, 5};
+ *  thrust::device_vector<int> v(data, data + N);
+ *  thrust::reverse(thrust::device, v.begin(), v.end());
+ *  // v is now {5, 4, 3, 2, 1, 0}
+ *  \endcode
+ *  
+ *  \see http://www.sgi.com/tech/stl/reverse.html
+ *  \see \p reverse_copy
+ *  \see \p reverse_iterator
+ */
+template<typename DerivedPolicy, typename BidirectionalIterator>
+  void reverse(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+               BidirectionalIterator first,
+               BidirectionalIterator last);
+
+
+/*! \p reverse reverses a range. That is: for every <tt>i</tt> such that
+ *  <tt>0 <= i <= (last - first) / 2</tt>, it exchanges <tt>*(first + i)</tt>
+ *  and <tt>*(last - (i + 1))</tt>.
+ *
+ *  \param first The beginning of the range to reverse.
+ *  \param last The end of the range to reverse.
+ *
+ *  \tparam BidirectionalIterator is a model of <a href="http://www.sgi.com/tech/stl/BidirectionalIterator.html">Bidirectional Iterator</a> and
+ *          \p BidirectionalIterator is mutable.
+ *
+ *  The following code snippet demonstrates how to use \p reverse to reverse a
+ *  \p device_vector of integers.
+ *
+ *  \code
+ *  #include <thrust/reverse.h>
+ *  ...
+ *  const int N = 6;
+ *  int data[N] = {0, 1, 2, 3, 4, 5};
+ *  thrust::device_vector<int> v(data, data + N);
+ *  thrust::reverse(v.begin(), v.end());
+ *  // v is now {5, 4, 3, 2, 1, 0}
+ *  \endcode
+ *  
+ *  \see http://www.sgi.com/tech/stl/reverse.html
+ *  \see \p reverse_copy
+ *  \see \p reverse_iterator
+ */
+template<typename BidirectionalIterator>
+  void reverse(BidirectionalIterator first,
+               BidirectionalIterator last);
+
+
+/*! \p reverse_copy differs from \ref reverse only in that the reversed range
+ *  is written to a different output range, rather than inplace.
+ *
+ *  \p reverse_copy copies elements from the range <tt>[first, last)</tt> to the
+ *  range <tt>[result, result + (last - first))</tt> such that the copy is a 
+ *  reverse of the original range. Specifically: for every <tt>i</tt> such that
+ *  <tt>0 <= i < (last - first)</tt>, \p reverse_copy performs the assignment
+ *  <tt>*(result + (last - first) - i) = *(first + i)</tt>.
+ *
+ *  The return value is <tt>result + (last - first))</tt>.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the range to reverse.
+ *  \param last The end of the range to reverse.
+ *  \param result The beginning of the output range.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam BidirectionalIterator is a model of <a href="http://www.sgi.com/tech/stl/BidirectionalIterator.html">Bidirectional Iterator</a>,
+ *          and \p BidirectionalIterator's \p value_type is convertible to \p OutputIterator's \p value_type.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *
+ *  \pre The range <tt>[first, last)</tt> and the range <tt>[result, result + (last - first))</tt> shall not overlap.
+ *
+ *  The following code snippet demonstrates how to use \p reverse_copy to reverse
+ *  an input \p device_vector of integers to an output \p device_vector using the \p thrust::device
+ *  execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/reverse.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  const int N = 6;
+ *  int data[N] = {0, 1, 2, 3, 4, 5};
+ *  thrust::device_vector<int> input(data, data + N);
+ *  thrust::device_vector<int> output(N);
+ *  thrust::reverse_copy(thrust::device, v.begin(), v.end(), output.begin());
+ *  // input is still {0, 1, 2, 3, 4, 5}
+ *  // output is now  {5, 4, 3, 2, 1, 0}
+ *  \endcode
+ *  
+ *  \see http://www.sgi.com/tech/stl/reverse_copy.html
+ *  \see \p reverse
+ *  \see \p reverse_iterator
+ */
+template<typename DerivedPolicy, typename BidirectionalIterator, typename OutputIterator>
+  OutputIterator reverse_copy(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                              BidirectionalIterator first,
+                              BidirectionalIterator last,
+                              OutputIterator result);
+
+
+/*! \p reverse_copy differs from \ref reverse only in that the reversed range
+ *  is written to a different output range, rather than inplace.
+ *
+ *  \p reverse_copy copies elements from the range <tt>[first, last)</tt> to the
+ *  range <tt>[result, result + (last - first))</tt> such that the copy is a 
+ *  reverse of the original range. Specifically: for every <tt>i</tt> such that
+ *  <tt>0 <= i < (last - first)</tt>, \p reverse_copy performs the assignment
+ *  <tt>*(result + (last - first) - i) = *(first + i)</tt>.
+ *
+ *  The return value is <tt>result + (last - first))</tt>.
+ *
+ *  \param first The beginning of the range to reverse.
+ *  \param last The end of the range to reverse.
+ *  \param result The beginning of the output range.
+ *
+ *  \tparam BidirectionalIterator is a model of <a href="http://www.sgi.com/tech/stl/BidirectionalIterator.html">Bidirectional Iterator</a>,
+ *          and \p BidirectionalIterator's \p value_type is convertible to \p OutputIterator's \p value_type.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *
+ *  \pre The range <tt>[first, last)</tt> and the range <tt>[result, result + (last - first))</tt> shall not overlap.
+ *
+ *  The following code snippet demonstrates how to use \p reverse_copy to reverse
+ *  an input \p device_vector of integers to an output \p device_vector.
+ *
+ *  \code
+ *  #include <thrust/reverse.h>
+ *  ...
+ *  const int N = 6;
+ *  int data[N] = {0, 1, 2, 3, 4, 5};
+ *  thrust::device_vector<int> input(data, data + N);
+ *  thrust::device_vector<int> output(N);
+ *  thrust::reverse_copy(v.begin(), v.end(), output.begin());
+ *  // input is still {0, 1, 2, 3, 4, 5}
+ *  // output is now  {5, 4, 3, 2, 1, 0}
+ *  \endcode
+ *  
+ *  \see http://www.sgi.com/tech/stl/reverse_copy.html
+ *  \see \p reverse
+ *  \see \p reverse_iterator
+ */
+template<typename BidirectionalIterator, typename OutputIterator>
+  OutputIterator reverse_copy(BidirectionalIterator first,
+                              BidirectionalIterator last,
+                              OutputIterator result);
+
+
+/*! \} // end reordering
+ */
+
+
+} // end thrust
+
+#include <thrust/detail/reverse.inl>
+
diff --git a/compat/thrust/scan.h b/compat/thrust/scan.h
new file mode 100644
index 0000000000..95074e6b90
--- /dev/null
+++ b/compat/thrust/scan.h
@@ -0,0 +1,1552 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file scan.h
+ *  \brief Functions for computing prefix sums
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/execution_policy.h>
+
+namespace thrust
+{
+
+
+/*! \addtogroup algorithms
+ */
+
+
+/*! \addtogroup prefixsums Prefix Sums
+ *  \ingroup algorithms
+ *  \{
+ */
+
+
+/*! \p inclusive_scan computes an inclusive prefix sum operation. The
+ *  term 'inclusive' means that each result includes the corresponding
+ *  input operand in the partial sum. More precisely, <tt>*first</tt> is 
+ *  assigned to <tt>*result</tt> and the sum of <tt>*first</tt> and 
+ *  <tt>*(first + 1)</tt> is assigned to <tt>*(result + 1)</tt>, and so on. 
+ *  This version of \p inclusive_scan assumes plus as the associative operator.  
+ *  When the input and output sequences are the same, the scan is performed 
+ *  in-place.
+ 
+ *  \p inclusive_scan is similar to \c std::partial_sum in the STL.  The primary
+ *  difference between the two functions is that \c std::partial_sum guarantees
+ *  a serial summation order, while \p inclusive_scan requires associativity of 
+ *  the binary operation to parallelize the prefix sum.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *    
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the input sequence.
+ *  \param last The end of the input sequence.
+ *  \param result The beginning of the output sequence.
+ *  \return The end of the output sequence.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>
+ *                        and \c InputIterator's \c value_type is convertible to
+ *                        \c OutputIterator's \c value_type.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>,
+ *                         and if \c x and \c y are objects of \c OutputIterator's
+ *                         \c value_type, then <tt>x + y</tt> is defined. If \c T is
+ *                         \c OutputIterator's \c value_type, then <tt>T(0)</tt> is
+ *                         defined.
+ *
+ *  \pre \p first may equal \p result but the range <tt>[first, last)</tt> and the range <tt>[result, result + (last - first))</tt> shall not overlap otherwise.
+ *
+ *  The following code snippet demonstrates how to use \p inclusive_scan to compute an in-place
+ *  prefix sum using the \p thrust::host execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/scan.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  
+ *  int data[6] = {1, 0, 2, 2, 1, 3};
+ *
+ *  thrust::inclusive_scan(thrust::host, data, data + 6, data); // in-place scan
+ *
+ *  // data is now {1, 1, 3, 5, 6, 9}
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/partial_sum.html
+ *
+ */
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator>
+  OutputIterator inclusive_scan(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                InputIterator first,
+                                InputIterator last,
+                                OutputIterator result);
+
+
+/*! \p inclusive_scan computes an inclusive prefix sum operation. The
+ *  term 'inclusive' means that each result includes the corresponding
+ *  input operand in the partial sum. More precisely, <tt>*first</tt> is 
+ *  assigned to <tt>*result</tt> and the sum of <tt>*first</tt> and 
+ *  <tt>*(first + 1)</tt> is assigned to <tt>*(result + 1)</tt>, and so on. 
+ *  This version of \p inclusive_scan assumes plus as the associative operator.  
+ *  When the input and output sequences are the same, the scan is performed 
+ *  in-place.
+ 
+ *  \p inclusive_scan is similar to \c std::partial_sum in the STL.  The primary
+ *  difference between the two functions is that \c std::partial_sum guarantees
+ *  a serial summation order, while \p inclusive_scan requires associativity of 
+ *  the binary operation to parallelize the prefix sum.
+ *    
+ *  \param first The beginning of the input sequence.
+ *  \param last The end of the input sequence.
+ *  \param result The beginning of the output sequence.
+ *  \return The end of the output sequence.
+ *
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>
+ *                        and \c InputIterator's \c value_type is convertible to
+ *                        \c OutputIterator's \c value_type.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>,
+ *                         and if \c x and \c y are objects of \c OutputIterator's
+ *                         \c value_type, then <tt>x + y</tt> is defined. If \c T is
+ *                         \c OutputIterator's \c value_type, then <tt>T(0)</tt> is
+ *                         defined.
+ *
+ *  \pre \p first may equal \p result but the range <tt>[first, last)</tt> and the range <tt>[result, result + (last - first))</tt> shall not overlap otherwise.
+ *
+ *  The following code snippet demonstrates how to use \p inclusive_scan
+ *
+ *  \code
+ *  #include <thrust/scan.h>
+ *  
+ *  int data[6] = {1, 0, 2, 2, 1, 3};
+ *
+ *  thrust::inclusive_scan(data, data + 6, data); // in-place scan
+ *
+ *  // data is now {1, 1, 3, 5, 6, 9}
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/partial_sum.html
+ *
+ */
+template<typename InputIterator,
+         typename OutputIterator>
+  OutputIterator inclusive_scan(InputIterator first,
+                                InputIterator last,
+                                OutputIterator result);
+
+
+/*! \p inclusive_scan computes an inclusive prefix sum operation. The
+ *  term 'inclusive' means that each result includes the corresponding
+ *  input operand in the partial sum.  When the input and output sequences 
+ *  are the same, the scan is performed in-place.
+ *    
+ *  \p inclusive_scan is similar to \c std::partial_sum in the STL.  The primary
+ *  difference between the two functions is that \c std::partial_sum guarantees
+ *  a serial summation order, while \p inclusive_scan requires associativity of 
+ *  the binary operation to parallelize the prefix sum.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the input sequence.
+ *  \param last The end of the input sequence.
+ *  \param result The beginning of the output sequence.
+ *  \param binary_op The associatve operator used to 'sum' values.
+ *  \return The end of the output sequence.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>
+ *                        and \c InputIterator's \c value_type is convertible to
+ *                        \c OutputIterator's \c value_type.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>
+ *                         and \c OutputIterator's \c value_type is convertible to
+ *                         both \c AssociativeOperator's \c first_argument_type and
+ *                         \c second_argument_type.
+ *  \tparam AssociativeOperator is a model of <a href="http://www.sgi.com/tech/stl/BinaryFunction.html">Binary Function</a>
+ *                              and \c AssociativeOperator's \c result_type is
+ *                              convertible to \c OutputIterator's \c value_type.
+ *
+ *  \pre \p first may equal \p result but the range <tt>[first, last)</tt> and the range <tt>[result, result + (last - first))</tt> shall not overlap otherwise.
+ *
+ *  The following code snippet demonstrates how to use \p inclusive_scan to compute an in-place
+ *  prefix sum using the \p thrust::host execution policy for parallelization:
+ *
+ *  \code
+ *  int data[10] = {-5, 0, 2, -3, 2, 4, 0, -1, 2, 8};
+ * 
+ *  thrust::maximum<int> binary_op;
+ *
+ *  thrust::inclusive_scan(thrust::host, data, data + 10, data, binary_op); // in-place scan
+ *
+ *  // data is now {-5, 0, 2, 2, 2, 4, 4, 4, 4, 8}
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/partial_sum.html
+ */
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator,
+         typename AssociativeOperator>
+  OutputIterator inclusive_scan(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                InputIterator first,
+                                InputIterator last,
+                                OutputIterator result,
+                                AssociativeOperator binary_op);
+
+
+/*! \p inclusive_scan computes an inclusive prefix sum operation. The
+ *  term 'inclusive' means that each result includes the corresponding
+ *  input operand in the partial sum.  When the input and output sequences 
+ *  are the same, the scan is performed in-place.
+ *    
+ *  \p inclusive_scan is similar to \c std::partial_sum in the STL.  The primary
+ *  difference between the two functions is that \c std::partial_sum guarantees
+ *  a serial summation order, while \p inclusive_scan requires associativity of 
+ *  the binary operation to parallelize the prefix sum.
+ *
+ *  \param first The beginning of the input sequence.
+ *  \param last The end of the input sequence.
+ *  \param result The beginning of the output sequence.
+ *  \param binary_op The associatve operator used to 'sum' values.
+ *  \return The end of the output sequence.
+ *
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>
+ *                        and \c InputIterator's \c value_type is convertible to
+ *                        \c OutputIterator's \c value_type.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>
+ *                         and \c OutputIterator's \c value_type is convertible to
+ *                         both \c AssociativeOperator's \c first_argument_type and
+ *                         \c second_argument_type.
+ *  \tparam AssociativeOperator is a model of <a href="http://www.sgi.com/tech/stl/BinaryFunction.html">Binary Function</a>
+ *                              and \c AssociativeOperator's \c result_type is
+ *                              convertible to \c OutputIterator's \c value_type.
+ *
+ *  \pre \p first may equal \p result but the range <tt>[first, last)</tt> and the range <tt>[result, result + (last - first))</tt> shall not overlap otherwise.
+ *
+ *  The following code snippet demonstrates how to use \p inclusive_scan
+ *
+ *  \code
+ *  int data[10] = {-5, 0, 2, -3, 2, 4, 0, -1, 2, 8};
+ * 
+ *  thrust::maximum<int> binary_op;
+ *
+ *  thrust::inclusive_scan(data, data + 10, data, binary_op); // in-place scan
+ *
+ *  // data is now {-5, 0, 2, 2, 2, 4, 4, 4, 4, 8}
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/partial_sum.html
+ */
+template<typename InputIterator,
+         typename OutputIterator,
+         typename AssociativeOperator>
+  OutputIterator inclusive_scan(InputIterator first,
+                                InputIterator last,
+                                OutputIterator result,
+                                AssociativeOperator binary_op);
+
+
+/*! \p exclusive_scan computes an exclusive prefix sum operation. The
+ *  term 'exclusive' means that each result does not include the 
+ *  corresponding input operand in the partial sum.  More precisely,
+ *  <tt>0</tt> is assigned to <tt>*result</tt> and the sum of 
+ *  <tt>0</tt> and <tt>*first</tt> is assigned to <tt>*(result + 1)</tt>,
+ *  and so on. This version of \p exclusive_scan assumes plus as the 
+ *  associative operator and \c 0 as the initial value.  When the input and 
+ *  output sequences are the same, the scan is performed in-place.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *    
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the input sequence.
+ *  \param last The end of the input sequence.
+ *  \param result The beginning of the output sequence.
+ *  \return The end of the output sequence.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>
+ *                        and \c InputIterator's \c value_type is convertible to
+ *                        \c OutputIterator's \c value_type.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>,
+ *                         and if \c x and \c y are objects of \c OutputIterator's
+ *                         \c value_type, then <tt>x + y</tt> is defined. If \c T is
+ *                         \c OutputIterator's \c value_type, then <tt>T(0)</tt> is
+ *                         defined.
+ *
+ *  \pre \p first may equal \p result but the range <tt>[first, last)</tt> and the range <tt>[result, result + (last - first))</tt> shall not overlap otherwise.
+ *
+ *  The following code snippet demonstrates how to use \p exclusive_scan to compute an in-place
+ *  prefix sum using the \p thrust::host execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/scan.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  
+ *  int data[6] = {1, 0, 2, 2, 1, 3};
+ *
+ *  thrust::exclusive_scan(thrust::host, data, data + 6, data); // in-place scan
+ *
+ *  // data is now {0, 1, 1, 3, 5, 6}
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/partial_sum.html
+ */
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator>
+  OutputIterator exclusive_scan(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                InputIterator first,
+                                InputIterator last,
+                                OutputIterator result);
+
+
+/*! \p exclusive_scan computes an exclusive prefix sum operation. The
+ *  term 'exclusive' means that each result does not include the 
+ *  corresponding input operand in the partial sum.  More precisely,
+ *  <tt>0</tt> is assigned to <tt>*result</tt> and the sum of 
+ *  <tt>0</tt> and <tt>*first</tt> is assigned to <tt>*(result + 1)</tt>,
+ *  and so on. This version of \p exclusive_scan assumes plus as the 
+ *  associative operator and \c 0 as the initial value.  When the input and 
+ *  output sequences are the same, the scan is performed in-place.
+ *    
+ *  \param first The beginning of the input sequence.
+ *  \param last The end of the input sequence.
+ *  \param result The beginning of the output sequence.
+ *  \return The end of the output sequence.
+ *
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>
+ *                        and \c InputIterator's \c value_type is convertible to
+ *                        \c OutputIterator's \c value_type.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>,
+ *                         and if \c x and \c y are objects of \c OutputIterator's
+ *                         \c value_type, then <tt>x + y</tt> is defined. If \c T is
+ *                         \c OutputIterator's \c value_type, then <tt>T(0)</tt> is
+ *                         defined.
+ *
+ *  \pre \p first may equal \p result but the range <tt>[first, last)</tt> and the range <tt>[result, result + (last - first))</tt> shall not overlap otherwise.
+ *
+ *  The following code snippet demonstrates how to use \p exclusive_scan
+ *
+ *  \code
+ *  #include <thrust/scan.h>
+ *  
+ *  int data[6] = {1, 0, 2, 2, 1, 3};
+ *
+ *  thrust::exclusive_scan(data, data + 6, data); // in-place scan
+ *
+ *  // data is now {0, 1, 1, 3, 5, 6}
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/partial_sum.html
+ */
+template<typename InputIterator,
+         typename OutputIterator>
+  OutputIterator exclusive_scan(InputIterator first,
+                                InputIterator last,
+                                OutputIterator result);
+
+
+/*! \p exclusive_scan computes an exclusive prefix sum operation. The
+ *  term 'exclusive' means that each result does not include the 
+ *  corresponding input operand in the partial sum.  More precisely,
+ *  \p init is assigned to <tt>*result</tt> and the sum of \p init and 
+ *  <tt>*first</tt> is assigned to <tt>*(result + 1)</tt>, and so on. 
+ *  This version of \p exclusive_scan assumes plus as the associative 
+ *  operator but requires an initial value \p init.  When the input and 
+ *  output sequences are the same, the scan is performed in-place.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the input sequence.
+ *  \param last The end of the input sequence.
+ *  \param result The beginning of the output sequence.
+ *  \param init The initial value.
+ *  \return The end of the output sequence.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>
+ *                        and \c InputIterator's \c value_type is convertible to
+ *                        \c OutputIterator's \c value_type.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>,
+ *                         and if \c x and \c y are objects of \c OutputIterator's
+ *                         \c value_type, then <tt>x + y</tt> is defined.
+ *  \tparam T is convertible to \c OutputIterator's \c value_type.
+ *
+ *  \pre \p first may equal \p result but the range <tt>[first, last)</tt> and the range <tt>[result, result + (last - first))</tt> shall not overlap otherwise.
+ *
+ *  The following code snippet demonstrates how to use \p exclusive_scan to compute an in-place
+ *  prefix sum using the \p thrust::host execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/scan.h>
+ *  #include <thrust/execution_policy.h>
+ *  
+ *  int data[6] = {1, 0, 2, 2, 1, 3};
+ *
+ *  thrust::exclusive_scan(thrust::host, data, data + 6, data, 4); // in-place scan
+ *
+ *  // data is now {4, 5, 5, 7, 9, 10}
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/partial_sum.html
+ */
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator,
+         typename T>
+  OutputIterator exclusive_scan(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                InputIterator first,
+                                InputIterator last,
+                                OutputIterator result,
+                                T init);
+
+
+/*! \p exclusive_scan computes an exclusive prefix sum operation. The
+ *  term 'exclusive' means that each result does not include the 
+ *  corresponding input operand in the partial sum.  More precisely,
+ *  \p init is assigned to <tt>*result</tt> and the sum of \p init and 
+ *  <tt>*first</tt> is assigned to <tt>*(result + 1)</tt>, and so on. 
+ *  This version of \p exclusive_scan assumes plus as the associative 
+ *  operator but requires an initial value \p init.  When the input and 
+ *  output sequences are the same, the scan is performed in-place.
+ *
+ *  \param first The beginning of the input sequence.
+ *  \param last The end of the input sequence.
+ *  \param result The beginning of the output sequence.
+ *  \param init The initial value.
+ *  \return The end of the output sequence.
+ *
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>
+ *                        and \c InputIterator's \c value_type is convertible to
+ *                        \c OutputIterator's \c value_type.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>,
+ *                         and if \c x and \c y are objects of \c OutputIterator's
+ *                         \c value_type, then <tt>x + y</tt> is defined.
+ *  \tparam T is convertible to \c OutputIterator's \c value_type.
+ *
+ *  \pre \p first may equal \p result but the range <tt>[first, last)</tt> and the range <tt>[result, result + (last - first))</tt> shall not overlap otherwise.
+ *
+ *  The following code snippet demonstrates how to use \p exclusive_scan
+ *
+ *  \code
+ *  #include <thrust/scan.h>
+ *  
+ *  int data[6] = {1, 0, 2, 2, 1, 3};
+ *
+ *  thrust::exclusive_scan(data, data + 6, data, 4); // in-place scan
+ *
+ *  // data is now {4, 5, 5, 7, 9, 10}
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/partial_sum.html
+ */
+template<typename InputIterator,
+         typename OutputIterator,
+         typename T>
+  OutputIterator exclusive_scan(InputIterator first,
+                                InputIterator last,
+                                OutputIterator result,
+                                T init);
+
+
+/*! \p exclusive_scan computes an exclusive prefix sum operation. The
+ *  term 'exclusive' means that each result does not include the 
+ *  corresponding input operand in the partial sum.  More precisely,
+ *  \p init is assigned to <tt>\*result</tt> and the value
+ *  <tt>binary_op(init, \*first)</tt> is assigned to <tt>\*(result + 1)</tt>,
+ *  and so on. This version of the function requires both and associative 
+ *  operator and an initial value \p init.  When the input and output
+ *  sequences are the same, the scan is performed in-place.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *    
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the input sequence.
+ *  \param last The end of the input sequence.
+ *  \param result The beginning of the output sequence.
+ *  \param init The initial value.
+ *  \param binary_op The associatve operator used to 'sum' values.
+ *  \return The end of the output sequence.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>
+ *                        and \c InputIterator's \c value_type is convertible to
+ *                        \c OutputIterator's \c value_type.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>
+ *                         and \c OutputIterator's \c value_type is convertible to
+ *                         both \c AssociativeOperator's \c first_argument_type and
+ *                         \c second_argument_type.
+ *  \tparam T is convertible to \c OutputIterator's \c value_type.
+ *  \tparam AssociativeOperator is a model of <a href="http://www.sgi.com/tech/stl/BinaryFunction.html">Binary Function</a>
+ *                              and \c AssociativeOperator's \c result_type is
+ *                              convertible to \c OutputIterator's \c value_type.
+ *
+ *  \pre \p first may equal \p result but the range <tt>[first, last)</tt> and the range <tt>[result, result + (last - first))</tt> shall not overlap otherwise.
+ *
+ *  The following code snippet demonstrates how to use \p exclusive_scan to compute an in-place
+ *  prefix sum using the \p thrust::host execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/scan.h>
+ *  #include <thrust/functional.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  
+ *  int data[10] = {-5, 0, 2, -3, 2, 4, 0, -1, 2, 8};
+ * 
+ *  thrust::maximum<int> binary_op;
+ *
+ *  thrust::exclusive_scan(thrust::host, data, data + 10, data, 1, binary_op); // in-place scan
+ *
+ *  // data is now {1, 1, 1, 2, 2, 2, 4, 4, 4, 4 }
+ *  \endcode
+ *  
+ *  \see http://www.sgi.com/tech/stl/partial_sum.html
+ */
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator,
+         typename T,
+         typename AssociativeOperator>
+  OutputIterator exclusive_scan(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                InputIterator first,
+                                InputIterator last,
+                                OutputIterator result,
+                                T init,
+                                AssociativeOperator binary_op);
+
+
+/*! \p exclusive_scan computes an exclusive prefix sum operation. The
+ *  term 'exclusive' means that each result does not include the 
+ *  corresponding input operand in the partial sum.  More precisely,
+ *  \p init is assigned to <tt>\*result</tt> and the value
+ *  <tt>binary_op(init, \*first)</tt> is assigned to <tt>\*(result + 1)</tt>,
+ *  and so on. This version of the function requires both and associative 
+ *  operator and an initial value \p init.  When the input and output
+ *  sequences are the same, the scan is performed in-place.
+ *    
+ *  \param first The beginning of the input sequence.
+ *  \param last The end of the input sequence.
+ *  \param result The beginning of the output sequence.
+ *  \param init The initial value.
+ *  \param binary_op The associatve operator used to 'sum' values.
+ *  \return The end of the output sequence.
+ *
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>
+ *                        and \c InputIterator's \c value_type is convertible to
+ *                        \c OutputIterator's \c value_type.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>
+ *                         and \c OutputIterator's \c value_type is convertible to
+ *                         both \c AssociativeOperator's \c first_argument_type and
+ *                         \c second_argument_type.
+ *  \tparam T is convertible to \c OutputIterator's \c value_type.
+ *  \tparam AssociativeOperator is a model of <a href="http://www.sgi.com/tech/stl/BinaryFunction.html">Binary Function</a>
+ *                              and \c AssociativeOperator's \c result_type is
+ *                              convertible to \c OutputIterator's \c value_type.
+ *
+ *  \pre \p first may equal \p result but the range <tt>[first, last)</tt> and the range <tt>[result, result + (last - first))</tt> shall not overlap otherwise.
+ *
+ *  The following code snippet demonstrates how to use \p exclusive_scan
+ *
+ *  \code
+ *  #include <thrust/scan.h>
+ *  #include <thrust/functional.h>
+ *  
+ *  int data[10] = {-5, 0, 2, -3, 2, 4, 0, -1, 2, 8};
+ * 
+ *  thrust::maximum<int> binary_op;
+ *
+ *  thrust::exclusive_scan(data, data + 10, data, 1, binary_op); // in-place scan
+ *
+ *  // data is now {1, 1, 1, 2, 2, 2, 4, 4, 4, 4 }
+ *  \endcode
+ *  
+ *  \see http://www.sgi.com/tech/stl/partial_sum.html
+ */
+template<typename InputIterator,
+         typename OutputIterator,
+         typename T,
+         typename AssociativeOperator>
+  OutputIterator exclusive_scan(InputIterator first,
+                                InputIterator last,
+                                OutputIterator result,
+                                T init,
+                                AssociativeOperator binary_op);
+
+
+/*! \addtogroup segmentedprefixsums Segmented Prefix Sums
+ *  \ingroup prefixsums
+ *  \{
+ */
+
+
+/*! \p inclusive_scan_by_key computes an inclusive key-value or 'segmented' prefix 
+ *  sum operation. The term 'inclusive' means that each result includes 
+ *  the corresponding input operand in the partial sum. The term 'segmented'
+ *  means that the partial sums are broken into distinct segments.  In other
+ *  words, within each segment a separate inclusive scan operation is computed.
+ *  Refer to the code sample below for example usage.
+ *
+ *  This version of \p inclusive_scan_by_key assumes \c equal_to as the binary
+ *  predicate used to compare adjacent keys.  Specifically, consecutive iterators
+ *  <tt>i</tt> and <tt>i+1</tt> in the range <tt>[first1, last1)</tt>
+ *  belong to the same segment if <tt>*i == *(i+1)</tt>, and belong to
+ *  different segments otherwise.
+ *
+ *  This version of \p inclusive_scan_by_key assumes \c plus as the associative
+ *  operator used to perform the prefix sum. When the input and output sequences
+ *  are the same, the scan is performed in-place.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first1 The beginning of the key sequence.
+ *  \param last1 The end of the key sequence.
+ *  \param first2 The beginning of the input value sequence.
+ *  \param result The beginning of the output value sequence.
+ *  \return The end of the output sequence.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>
+ *                         and \c InputIterator2's \c value_type is convertible to \c OutputIterator's \c value_type.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>,
+ *                         and if \c x and \c y are objects of \c OutputIterator's \c value_type, then 
+ *                         <tt>binary_op(x,y)</tt> is defined.
+ *
+ *  \pre \p first1 may equal \p result but the range <tt>[first1, last1)</tt> and the range <tt>[result, result + (last1 - first1))</tt> shall not overlap otherwise.
+ *  \pre \p first2 may equal \p result but the range <tt>[first2, first2 + (last1 - first1)</tt> and range <tt>[result, result + (last1 - first1))</tt> shall not overlap otherwise.
+ *
+ *  The following code snippet demonstrates how to use \p inclusive_scan_by_key using the \p thrust::host
+ *  execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/scan.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  
+ *  int data[10] = {1, 1, 1, 1, 1, 1, 1, 1, 1, 1};
+ *  int keys[10] = {0, 0, 0, 1, 1, 2, 3, 3, 3, 3};
+ *
+ *  thrust::inclusive_scan_by_key(thrust::host, keys, keys + 10, vals, vals); // in-place scan
+ *
+ *  // data is now {1, 2, 3, 1, 2, 1, 1, 2, 3, 4};
+ *  \endcode
+ *
+ *  \see inclusive_scan
+ *  \see exclusive_scan_by_key
+ *
+ */
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator>
+  OutputIterator inclusive_scan_by_key(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                       InputIterator1 first1,
+                                       InputIterator1 last1,
+                                       InputIterator2 first2,
+                                       OutputIterator result);
+ 
+
+/*! \p inclusive_scan_by_key computes an inclusive key-value or 'segmented' prefix 
+ *  sum operation. The term 'inclusive' means that each result includes 
+ *  the corresponding input operand in the partial sum. The term 'segmented'
+ *  means that the partial sums are broken into distinct segments.  In other
+ *  words, within each segment a separate inclusive scan operation is computed.
+ *  Refer to the code sample below for example usage.
+ *
+ *  This version of \p inclusive_scan_by_key assumes \c equal_to as the binary
+ *  predicate used to compare adjacent keys.  Specifically, consecutive iterators
+ *  <tt>i</tt> and <tt>i+1</tt> in the range <tt>[first1, last1)</tt>
+ *  belong to the same segment if <tt>*i == *(i+1)</tt>, and belong to
+ *  different segments otherwise.
+ *
+ *  This version of \p inclusive_scan_by_key assumes \c plus as the associative
+ *  operator used to perform the prefix sum. When the input and output sequences
+ *  are the same, the scan is performed in-place.
+ *
+ *  \param first1 The beginning of the key sequence.
+ *  \param last1 The end of the key sequence.
+ *  \param first2 The beginning of the input value sequence.
+ *  \param result The beginning of the output value sequence.
+ *  \return The end of the output sequence.
+ *
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>
+ *                         and \c InputIterator2's \c value_type is convertible to \c OutputIterator's \c value_type.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>,
+ *                         and if \c x and \c y are objects of \c OutputIterator's \c value_type, then 
+ *                         <tt>binary_op(x,y)</tt> is defined.
+ *
+ *  \pre \p first1 may equal \p result but the range <tt>[first1, last1)</tt> and the range <tt>[result, result + (last1 - first1))</tt> shall not overlap otherwise.
+ *  \pre \p first2 may equal \p result but the range <tt>[first2, first2 + (last1 - first1)</tt> and range <tt>[result, result + (last1 - first1))</tt> shall not overlap otherwise.
+ *
+ *  The following code snippet demonstrates how to use \p inclusive_scan_by_key
+ *
+ *  \code
+ *  #include <thrust/scan.h>
+ *  
+ *  int data[10] = {1, 1, 1, 1, 1, 1, 1, 1, 1, 1};
+ *  int keys[10] = {0, 0, 0, 1, 1, 2, 3, 3, 3, 3};
+ *
+ *  thrust::inclusive_scan_by_key(keys, keys + 10, vals, vals); // in-place scan
+ *
+ *  // data is now {1, 2, 3, 1, 2, 1, 1, 2, 3, 4};
+ *  \endcode
+ *
+ *  \see inclusive_scan
+ *  \see exclusive_scan_by_key
+ *
+ */
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator>
+  OutputIterator inclusive_scan_by_key(InputIterator1 first1,
+                                       InputIterator1 last1,
+                                       InputIterator2 first2,
+                                       OutputIterator result);
+
+
+/*! \p inclusive_scan_by_key computes an inclusive key-value or 'segmented' prefix 
+ *  sum operation. The term 'inclusive' means that each result includes 
+ *  the corresponding input operand in the partial sum. The term 'segmented'
+ *  means that the partial sums are broken into distinct segments.  In other
+ *  words, within each segment a separate inclusive scan operation is computed.
+ *  Refer to the code sample below for example usage.
+ *
+ *  This version of \p inclusive_scan_by_key uses the binary predicate 
+ *  \c pred to compare adjacent keys.  Specifically, consecutive iterators
+ *  <tt>i</tt> and <tt>i+1</tt> in the range <tt>[first1, last1)</tt>
+ *  belong to the same segment if <tt>binary_pred(*i, *(i+1))</tt> is true, and belong to 
+ *  different segments otherwise.
+ *
+ *  This version of \p inclusive_scan_by_key assumes \c plus as the associative
+ *  operator used to perform the prefix sum. When the input and output sequences
+ *  are the same, the scan is performed in-place.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec. 
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first1 The beginning of the key sequence.
+ *  \param last1 The end of the key sequence.
+ *  \param first2 The beginning of the input value sequence.
+ *  \param result The beginning of the output value sequence.
+ *  \param binary_pred  The binary predicate used to determine equality of keys.
+ *  \return The end of the output sequence.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>
+ *                         and \c InputIterator2's \c value_type is convertible to \c OutputIterator's \c value_type.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>,
+ *                         and if \c x and \c y are objects of \c OutputIterator's \c value_type, then 
+ *                         <tt>binary_op(x,y)</tt> is defined.
+ *  \tparam BinaryPredicate is a model of <a href="http://www.sgi.com/tech/stl/BinaryPredicate.html">Binary Predicate</a>.
+ *
+ *  \pre \p first1 may equal \p result but the range <tt>[first1, last1)</tt> and the range <tt>[result, result + (last1 - first1))</tt> shall not overlap otherwise.
+ *  \pre \p first2 may equal \p result but the range <tt>[first2, first2 + (last1 - first1)</tt> and range <tt>[result, result + (last1 - first1))</tt> shall not overlap otherwise.
+ *
+ *  The following code snippet demonstrates how to use \p inclusive_scan_by_key using the \p thrust::host
+ *  execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/scan.h>
+ *  #include <thrust/functional.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  
+ *  int data[10] = {1, 1, 1, 1, 1, 1, 1, 1, 1, 1};
+ *  int keys[10] = {0, 0, 0, 1, 1, 2, 3, 3, 3, 3};
+ *
+ *  thrust::equal_to<int> binary_pred;
+ *
+ *  thrust::inclusive_scan_by_key(thrust::host, keys, keys + 10, vals, vals, binary_pred); // in-place scan
+ *
+ *  // data is now {1, 2, 3, 1, 2, 1, 1, 2, 3, 4};
+ *  \endcode
+ *
+ *  \see inclusive_scan
+ *  \see exclusive_scan_by_key
+ *
+ */
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename BinaryPredicate>
+  OutputIterator inclusive_scan_by_key(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                       InputIterator1 first1,
+                                       InputIterator1 last1,
+                                       InputIterator2 first2,
+                                       OutputIterator result,
+                                       BinaryPredicate binary_pred);
+
+
+/*! \p inclusive_scan_by_key computes an inclusive key-value or 'segmented' prefix 
+ *  sum operation. The term 'inclusive' means that each result includes 
+ *  the corresponding input operand in the partial sum. The term 'segmented'
+ *  means that the partial sums are broken into distinct segments.  In other
+ *  words, within each segment a separate inclusive scan operation is computed.
+ *  Refer to the code sample below for example usage.
+ *
+ *  This version of \p inclusive_scan_by_key uses the binary predicate 
+ *  \c pred to compare adjacent keys.  Specifically, consecutive iterators
+ *  <tt>i</tt> and <tt>i+1</tt> in the range <tt>[first1, last1)</tt>
+ *  belong to the same segment if <tt>binary_pred(*i, *(i+1))</tt> is true, and belong to 
+ *  different segments otherwise.
+ *
+ *  This version of \p inclusive_scan_by_key assumes \c plus as the associative
+ *  operator used to perform the prefix sum. When the input and output sequences
+ *  are the same, the scan is performed in-place.
+ *
+ *  \param first1 The beginning of the key sequence.
+ *  \param last1 The end of the key sequence.
+ *  \param first2 The beginning of the input value sequence.
+ *  \param result The beginning of the output value sequence.
+ *  \param binary_pred  The binary predicate used to determine equality of keys.
+ *  \return The end of the output sequence.
+ *
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>
+ *                         and \c InputIterator2's \c value_type is convertible to \c OutputIterator's \c value_type.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>,
+ *                         and if \c x and \c y are objects of \c OutputIterator's \c value_type, then 
+ *                         <tt>binary_op(x,y)</tt> is defined.
+ *  \tparam BinaryPredicate is a model of <a href="http://www.sgi.com/tech/stl/BinaryPredicate.html">Binary Predicate</a>.
+ *
+ *  \pre \p first1 may equal \p result but the range <tt>[first1, last1)</tt> and the range <tt>[result, result + (last1 - first1))</tt> shall not overlap otherwise.
+ *  \pre \p first2 may equal \p result but the range <tt>[first2, first2 + (last1 - first1)</tt> and range <tt>[result, result + (last1 - first1))</tt> shall not overlap otherwise.
+ *
+ *  The following code snippet demonstrates how to use \p inclusive_scan_by_key
+ *
+ *  \code
+ *  #include <thrust/scan.h>
+ *  #include <thrust/functional.h>
+ *  
+ *  int data[10] = {1, 1, 1, 1, 1, 1, 1, 1, 1, 1};
+ *  int keys[10] = {0, 0, 0, 1, 1, 2, 3, 3, 3, 3};
+ *
+ *  thrust::equal_to<int> binary_pred;
+ *
+ *  thrust::inclusive_scan_by_key(keys, keys + 10, vals, vals, binary_pred); // in-place scan
+ *
+ *  // data is now {1, 2, 3, 1, 2, 1, 1, 2, 3, 4};
+ *  \endcode
+ *
+ *  \see inclusive_scan
+ *  \see exclusive_scan_by_key
+ *
+ */
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename BinaryPredicate>
+  OutputIterator inclusive_scan_by_key(InputIterator1 first1,
+                                       InputIterator1 last1,
+                                       InputIterator2 first2,
+                                       OutputIterator result,
+                                       BinaryPredicate binary_pred);
+
+
+/*! \p inclusive_scan_by_key computes an inclusive key-value or 'segmented' prefix 
+ *  sum operation. The term 'inclusive' means that each result includes 
+ *  the corresponding input operand in the partial sum. The term 'segmented'
+ *  means that the partial sums are broken into distinct segments.  In other
+ *  words, within each segment a separate inclusive scan operation is computed.
+ *  Refer to the code sample below for example usage.
+ *
+ *  This version of \p inclusive_scan_by_key uses the binary predicate 
+ *  \c pred to compare adjacent keys.  Specifically, consecutive iterators
+ *  <tt>i</tt> and <tt>i+1</tt> in the range <tt>[first1, last1)</tt>
+ *  belong to the same segment if <tt>binary_pred(*i, *(i+1))</tt> is true, and belong to 
+ *  different segments otherwise.
+ *
+ *  This version of \p inclusive_scan_by_key uses the associative operator 
+ *  \c binary_op to perform the prefix sum. When the input and output sequences
+ *  are the same, the scan is performed in-place.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first1 The beginning of the key sequence.
+ *  \param last1 The end of the key sequence.
+ *  \param first2 The beginning of the input value sequence.
+ *  \param result The beginning of the output value sequence.
+ *  \param binary_pred  The binary predicate used to determine equality of keys.
+ *  \param binary_op The associatve operator used to 'sum' values.
+ *  \return The end of the output sequence.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>
+ *                         and \c InputIterator2's \c value_type is convertible to \c OutputIterator's \c value_type.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>,
+ *                         and if \c x and \c y are objects of \c OutputIterator's \c value_type, then 
+ *                         <tt>binary_op(x,y)</tt> is defined.
+ *  \tparam BinaryPredicate is a model of <a href="http://www.sgi.com/tech/stl/BinaryPredicate.html">Binary Predicate</a>.
+ *  \tparam AssociativeOperator is a model of <a href="http://www.sgi.com/tech/stl/BinaryFunction.html">Binary Function</a>
+ *                              and \c AssociativeOperator's \c result_type is
+ *                              convertible to \c OutputIterator's \c value_type.
+ *
+ *  \pre \p first1 may equal \p result but the range <tt>[first1, last1)</tt> and the range <tt>[result, result + (last1 - first1))</tt> shall not overlap otherwise.
+ *  \pre \p first2 may equal \p result but the range <tt>[first2, first2 + (last1 - first1)</tt> and range <tt>[result, result + (last1 - first1))</tt> shall not overlap otherwise.
+ *
+ *  The following code snippet demonstrates how to use \p inclusive_scan_by_key using the \p thrust::host
+ *  execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/scan.h>
+ *  #include <thrust/functional.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  
+ *  int data[10] = {1, 1, 1, 1, 1, 1, 1, 1, 1, 1};
+ *  int keys[10] = {0, 0, 0, 1, 1, 2, 3, 3, 3, 3};
+ *
+ *  thrust::equal_to<int> binary_pred;
+ *  thrust::plus<int>     binary_op;
+ *
+ *  thrust::inclusive_scan_by_key(thrust::host, keys, keys + 10, vals, vals, binary_pred, binary_op); // in-place scan
+ *
+ *  // data is now {1, 2, 3, 1, 2, 1, 1, 2, 3, 4};
+ *  \endcode
+ *
+ *  \see inclusive_scan
+ *  \see exclusive_scan_by_key
+ *
+ */
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename BinaryPredicate,
+         typename AssociativeOperator>
+  OutputIterator inclusive_scan_by_key(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                       InputIterator1 first1,
+                                       InputIterator1 last1,
+                                       InputIterator2 first2,
+                                       OutputIterator result,
+                                       BinaryPredicate binary_pred,
+                                       AssociativeOperator binary_op);
+
+
+/*! \p inclusive_scan_by_key computes an inclusive key-value or 'segmented' prefix 
+ *  sum operation. The term 'inclusive' means that each result includes 
+ *  the corresponding input operand in the partial sum. The term 'segmented'
+ *  means that the partial sums are broken into distinct segments.  In other
+ *  words, within each segment a separate inclusive scan operation is computed.
+ *  Refer to the code sample below for example usage.
+ *
+ *  This version of \p inclusive_scan_by_key uses the binary predicate 
+ *  \c pred to compare adjacent keys.  Specifically, consecutive iterators
+ *  <tt>i</tt> and <tt>i+1</tt> in the range <tt>[first1, last1)</tt>
+ *  belong to the same segment if <tt>binary_pred(*i, *(i+1))</tt> is true, and belong to 
+ *  different segments otherwise.
+ *
+ *  This version of \p inclusive_scan_by_key uses the associative operator 
+ *  \c binary_op to perform the prefix sum. When the input and output sequences
+ *  are the same, the scan is performed in-place.
+ *
+ *  \param first1 The beginning of the key sequence.
+ *  \param last1 The end of the key sequence.
+ *  \param first2 The beginning of the input value sequence.
+ *  \param result The beginning of the output value sequence.
+ *  \param binary_pred  The binary predicate used to determine equality of keys.
+ *  \param binary_op The associatve operator used to 'sum' values.
+ *  \return The end of the output sequence.
+ *
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>
+ *                         and \c InputIterator2's \c value_type is convertible to \c OutputIterator's \c value_type.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>,
+ *                         and if \c x and \c y are objects of \c OutputIterator's \c value_type, then 
+ *                         <tt>binary_op(x,y)</tt> is defined.
+ *  \tparam BinaryPredicate is a model of <a href="http://www.sgi.com/tech/stl/BinaryPredicate.html">Binary Predicate</a>.
+ *  \tparam AssociativeOperator is a model of <a href="http://www.sgi.com/tech/stl/BinaryFunction.html">Binary Function</a>
+ *                              and \c AssociativeOperator's \c result_type is
+ *                              convertible to \c OutputIterator's \c value_type.
+ *
+ *  \pre \p first1 may equal \p result but the range <tt>[first1, last1)</tt> and the range <tt>[result, result + (last1 - first1))</tt> shall not overlap otherwise.
+ *  \pre \p first2 may equal \p result but the range <tt>[first2, first2 + (last1 - first1)</tt> and range <tt>[result, result + (last1 - first1))</tt> shall not overlap otherwise.
+ *
+ *  The following code snippet demonstrates how to use \p inclusive_scan_by_key
+ *
+ *  \code
+ *  #include <thrust/scan.h>
+ *  #include <thrust/functional.h>
+ *  
+ *  int data[10] = {1, 1, 1, 1, 1, 1, 1, 1, 1, 1};
+ *  int keys[10] = {0, 0, 0, 1, 1, 2, 3, 3, 3, 3};
+ *
+ *  thrust::equal_to<int> binary_pred;
+ *  thrust::plus<int>     binary_op;
+ *
+ *  thrust::inclusive_scan_by_key(keys, keys + 10, vals, vals, binary_pred, binary_op); // in-place scan
+ *
+ *  // data is now {1, 2, 3, 1, 2, 1, 1, 2, 3, 4};
+ *  \endcode
+ *
+ *  \see inclusive_scan
+ *  \see exclusive_scan_by_key
+ *
+ */
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename BinaryPredicate,
+         typename AssociativeOperator>
+  OutputIterator inclusive_scan_by_key(InputIterator1 first1,
+                                       InputIterator1 last1,
+                                       InputIterator2 first2,
+                                       OutputIterator result,
+                                       BinaryPredicate binary_pred,
+                                       AssociativeOperator binary_op);
+
+
+/*! \p exclusive_scan_by_key computes an exclusive segmented prefix 
+ *
+ *  This version of \p exclusive_scan_by_key uses the value \c 0 to
+ *  initialize the exclusive scan operation.
+ *
+ *  This version of \p exclusive_scan_by_key assumes \c plus as the associative
+ *  operator used to perform the prefix sum. When the input and output sequences
+ *  are the same, the scan is performed in-place.
+ * 
+ *  This version of \p exclusive_scan_by_key assumes \c equal_to as the binary
+ *  predicate used to compare adjacent keys.  Specifically, consecutive iterators
+ *  <tt>i</tt> and <tt>i+1</tt> in the range <tt>[first1, last1</tt>
+ *  belong to the same segment if <tt>*i == *(i+1)</tt>, and belong to 
+ *  different segments otherwise.
+ *
+ *  Refer to the most general form of \p exclusive_scan_by_key for additional details.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first1 The beginning of the key sequence.
+ *  \param last1 The end of the key sequence.
+ *  \param first2 The beginning of the input value sequence.
+ *  \param result The beginning of the output value sequence.
+ *
+ *  \pre \p first1 may equal \p result but the range <tt>[first1, last1)</tt> and the range <tt>[result, result + (last1 - first1))</tt> shall not overlap otherwise.
+ *  \pre \p first2 may equal \p result but the range <tt>[first2, first2 + (last1 - first1)</tt> and range <tt>[result, result + (last1 - first1))</tt> shall not overlap otherwise.
+ *
+ *  The following code snippet demonstrates how to use \p exclusive_scan_by_key using the
+ *  \p thrust::host execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/scan.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  
+ *  int keys[10] = {0, 0, 0, 1, 1, 2, 3, 3, 3, 3};
+ *  int vals[10] = {1, 1, 1, 1, 1, 1, 1, 1, 1, 1};
+ *
+ *  thrust::exclusive_scan_by_key(thrust::host, key, key + 10, vals, vals); // in-place scan
+ *
+ *  // vals is now {0, 1, 2, 0, 1, 0, 0, 1, 2, 3};
+ *  \endcode
+ *
+ *  \see exclusive_scan
+ *
+ */
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator>
+  OutputIterator exclusive_scan_by_key(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                       InputIterator1 first1,
+                                       InputIterator1 last1,
+                                       InputIterator2 first2,
+                                       OutputIterator result);
+
+
+/*! \p exclusive_scan_by_key computes an exclusive segmented prefix 
+ *
+ *  This version of \p exclusive_scan_by_key uses the value \c 0 to
+ *  initialize the exclusive scan operation.
+ *
+ *  This version of \p exclusive_scan_by_key assumes \c plus as the associative
+ *  operator used to perform the prefix sum. When the input and output sequences
+ *  are the same, the scan is performed in-place.
+ * 
+ *  This version of \p exclusive_scan_by_key assumes \c equal_to as the binary
+ *  predicate used to compare adjacent keys.  Specifically, consecutive iterators
+ *  <tt>i</tt> and <tt>i+1</tt> in the range <tt>[first1, last1</tt>
+ *  belong to the same segment if <tt>*i == *(i+1)</tt>, and belong to 
+ *  different segments otherwise.
+ *
+ *  Refer to the most general form of \p exclusive_scan_by_key for additional details.
+ *
+ *  \param first1 The beginning of the key sequence.
+ *  \param last1 The end of the key sequence.
+ *  \param first2 The beginning of the input value sequence.
+ *  \param result The beginning of the output value sequence.
+ *
+ *  \pre \p first1 may equal \p result but the range <tt>[first1, last1)</tt> and the range <tt>[result, result + (last1 - first1))</tt> shall not overlap otherwise.
+ *  \pre \p first2 may equal \p result but the range <tt>[first2, first2 + (last1 - first1)</tt> and range <tt>[result, result + (last1 - first1))</tt> shall not overlap otherwise.
+ *
+ *  The following code snippet demonstrates how to use \p exclusive_scan_by_key.
+ *
+ *  \code
+ *  #include <thrust/scan.h>
+ *  
+ *  int keys[10] = {0, 0, 0, 1, 1, 2, 3, 3, 3, 3};
+ *  int vals[10] = {1, 1, 1, 1, 1, 1, 1, 1, 1, 1};
+ *
+ *  thrust::exclusive_scan_by_key(key, key + 10, vals, vals); // in-place scan
+ *
+ *  // vals is now {0, 1, 2, 0, 1, 0, 0, 1, 2, 3};
+ *  \endcode
+ *
+ *  \see exclusive_scan
+ *
+ */
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator>
+  OutputIterator exclusive_scan_by_key(InputIterator1 first1,
+                                       InputIterator1 last1,
+                                       InputIterator2 first2,
+                                       OutputIterator result);
+
+
+/*! \p exclusive_scan_by_key computes an exclusive key-value or 'segmented' prefix 
+ *  sum operation. The term 'exclusive' means that each result does not include
+ *  the corresponding input operand in the partial sum. The term 'segmented'
+ *  means that the partial sums are broken into distinct segments.  In other
+ *  words, within each segment a separate exclusive scan operation is computed.
+ *  Refer to the code sample below for example usage.
+ *
+ *  This version of \p exclusive_scan_by_key uses the value \c init to
+ *  initialize the exclusive scan operation.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first1 The beginning of the key sequence.
+ *  \param last1 The end of the key sequence.
+ *  \param first2 The beginning of the input value sequence.
+ *  \param result The beginning of the output value sequence.
+ *  \param init The initial of the exclusive sum value.
+ *  \return The end of the output sequence.
+ *
+ *  \pre \p first1 may equal \p result but the range <tt>[first1, last1)</tt> and the range <tt>[result, result + (last1 - first1))</tt> shall not overlap otherwise.
+ *  \pre \p first2 may equal \p result but the range <tt>[first2, first2 + (last1 - first1)</tt> and range <tt>[result, result + (last1 - first1))</tt> shall not overlap otherwise.
+ *
+ *  The following code snippet demonstrates how to use \p exclusive_scan_by_key using the \p
+ *  thrust::host execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/scan.h>
+ *  #include <thrust/functional.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  
+ *  int keys[10] = {0, 0, 0, 1, 1, 2, 3, 3, 3, 3};
+ *  int vals[10] = {1, 1, 1, 1, 1, 1, 1, 1, 1, 1};
+ *
+ *  int init = 5;
+ *
+ *  thrust::exclusive_scan_by_key(thrust::host, key, key + 10, vals, vals, init); // in-place scan
+ *
+ *  // vals is now {5, 6, 7, 5, 6, 5, 5, 6, 7, 8};
+ *  \endcode
+ *
+ *  \see exclusive_scan
+ *  \see inclusive_scan_by_key
+ *
+ */
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename T>
+  OutputIterator exclusive_scan_by_key(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                       InputIterator1 first1,
+                                       InputIterator1 last1,
+                                       InputIterator2 first2,
+                                       OutputIterator result,
+                                       T init);
+
+
+/*! \p exclusive_scan_by_key computes an exclusive key-value or 'segmented' prefix 
+ *  sum operation. The term 'exclusive' means that each result does not include
+ *  the corresponding input operand in the partial sum. The term 'segmented'
+ *  means that the partial sums are broken into distinct segments.  In other
+ *  words, within each segment a separate exclusive scan operation is computed.
+ *  Refer to the code sample below for example usage.
+ *
+ *  This version of \p exclusive_scan_by_key uses the value \c init to
+ *  initialize the exclusive scan operation.
+ *
+ *  \param first1 The beginning of the key sequence.
+ *  \param last1 The end of the key sequence.
+ *  \param first2 The beginning of the input value sequence.
+ *  \param result The beginning of the output value sequence.
+ *  \param init The initial of the exclusive sum value.
+ *  \return The end of the output sequence.
+ *
+ *  \pre \p first1 may equal \p result but the range <tt>[first1, last1)</tt> and the range <tt>[result, result + (last1 - first1))</tt> shall not overlap otherwise.
+ *  \pre \p first2 may equal \p result but the range <tt>[first2, first2 + (last1 - first1)</tt> and range <tt>[result, result + (last1 - first1))</tt> shall not overlap otherwise.
+ *
+ *  The following code snippet demonstrates how to use \p exclusive_scan_by_key
+ *
+ *  \code
+ *  #include <thrust/scan.h>
+ *  #include <thrust/functional.h>
+ *  
+ *  int keys[10] = {0, 0, 0, 1, 1, 2, 3, 3, 3, 3};
+ *  int vals[10] = {1, 1, 1, 1, 1, 1, 1, 1, 1, 1};
+ *
+ *  int init = 5;
+ *
+ *  thrust::exclusive_scan_by_key(key, key + 10, vals, vals, init); // in-place scan
+ *
+ *  // vals is now {5, 6, 7, 5, 6, 5, 5, 6, 7, 8};
+ *  \endcode
+ *
+ *  \see exclusive_scan
+ *  \see inclusive_scan_by_key
+ *
+ */
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename T>
+  OutputIterator exclusive_scan_by_key(InputIterator1 first1,
+                                       InputIterator1 last1,
+                                       InputIterator2 first2,
+                                       OutputIterator result,
+                                       T init);
+
+
+/*! \p exclusive_scan_by_key computes an exclusive key-value or 'segmented' prefix 
+ *  sum operation. The term 'exclusive' means that each result does not include
+ *  the corresponding input operand in the partial sum. The term 'segmented'
+ *  means that the partial sums are broken into distinct segments.  In other
+ *  words, within each segment a separate exclusive scan operation is computed.
+ *  Refer to the code sample below for example usage.
+ *
+ *  This version of \p exclusive_scan_by_key uses the value \c init to
+ *  initialize the exclusive scan operation.
+ *
+ *  This version of \p exclusive_scan_by_key uses the binary predicate \c binary_pred
+ *  to compare adjacent keys.  Specifically, consecutive iterators <tt>i</tt> and
+ *  <tt>i+1</tt> in the range <tt>[first1, last1)</tt> belong to the same segment if
+ *  <tt>binary_pred(*i, *(i+1))</tt> is true, and belong to different segments otherwise.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first1 The beginning of the key sequence.
+ *  \param last1 The end of the key sequence.
+ *  \param first2 The beginning of the input value sequence.
+ *  \param result The beginning of the output value sequence.
+ *  \param init The initial of the exclusive sum value.
+ *  \param binary_pred The binary predicate used to determine equality of keys.
+ *  \return The end of the output sequence.
+ *
+ *  \pre \p first1 may equal \p result but the range <tt>[first1, last1)</tt> and the range <tt>[result, result + (last1 - first1))</tt> shall not overlap otherwise.
+ *  \pre \p first2 may equal \p result but the range <tt>[first2, first2 + (last1 - first1)</tt> and range <tt>[result, result + (last1 - first1))</tt> shall not overlap otherwise.
+ *
+ *  The following code snippet demonstrates how to use \p exclusive_scan_by_key using the
+ *  \p thrust::host execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/scan.h>
+ *  #include <thrust/functional.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  
+ *  int keys[10] = {0, 0, 0, 1, 1, 2, 3, 3, 3, 3};
+ *  int vals[10] = {1, 1, 1, 1, 1, 1, 1, 1, 1, 1};
+ *
+ *  int init = 5;
+ *
+ *  thrust::equal_to<int> binary_pred;
+ *
+ *  thrust::exclusive_scan_by_key(thrust::host, key, key + 10, vals, vals, init, binary_pred); // in-place scan
+ *
+ *  // vals is now {5, 6, 7, 5, 6, 5, 5, 6, 7, 8};
+ *  \endcode
+ *
+ *  \see exclusive_scan
+ *  \see inclusive_scan_by_key
+ *
+ */
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename T,
+         typename BinaryPredicate>
+  OutputIterator exclusive_scan_by_key(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                       InputIterator1 first1,
+                                       InputIterator1 last1,
+                                       InputIterator2 first2,
+                                       OutputIterator result,
+                                       T init,
+                                       BinaryPredicate binary_pred);
+
+
+/*! \p exclusive_scan_by_key computes an exclusive key-value or 'segmented' prefix 
+ *  sum operation. The term 'exclusive' means that each result does not include
+ *  the corresponding input operand in the partial sum. The term 'segmented'
+ *  means that the partial sums are broken into distinct segments.  In other
+ *  words, within each segment a separate exclusive scan operation is computed.
+ *  Refer to the code sample below for example usage.
+ *
+ *  This version of \p exclusive_scan_by_key uses the value \c init to
+ *  initialize the exclusive scan operation.
+ *
+ *  This version of \p exclusive_scan_by_key uses the binary predicate \c binary_pred
+ *  to compare adjacent keys.  Specifically, consecutive iterators <tt>i</tt> and
+ *  <tt>i+1</tt> in the range <tt>[first1, last1)</tt> belong to the same segment if
+ *  <tt>binary_pred(*i, *(i+1))</tt> is true, and belong to different segments otherwise.
+ *
+ *  \param first1 The beginning of the key sequence.
+ *  \param last1 The end of the key sequence.
+ *  \param first2 The beginning of the input value sequence.
+ *  \param result The beginning of the output value sequence.
+ *  \param init The initial of the exclusive sum value.
+ *  \param binary_pred The binary predicate used to determine equality of keys.
+ *  \return The end of the output sequence.
+ *
+ *  \pre \p first1 may equal \p result but the range <tt>[first1, last1)</tt> and the range <tt>[result, result + (last1 - first1))</tt> shall not overlap otherwise.
+ *  \pre \p first2 may equal \p result but the range <tt>[first2, first2 + (last1 - first1)</tt> and range <tt>[result, result + (last1 - first1))</tt> shall not overlap otherwise.
+ *
+ *  The following code snippet demonstrates how to use \p exclusive_scan_by_key
+ *
+ *  \code
+ *  #include <thrust/scan.h>
+ *  #include <thrust/functional.h>
+ *  
+ *  int keys[10] = {0, 0, 0, 1, 1, 2, 3, 3, 3, 3};
+ *  int vals[10] = {1, 1, 1, 1, 1, 1, 1, 1, 1, 1};
+ *
+ *  int init = 5;
+ *
+ *  thrust::equal_to<int> binary_pred;
+ *
+ *  thrust::exclusive_scan_by_key(key, key + 10, vals, vals, init, binary_pred); // in-place scan
+ *
+ *  // vals is now {5, 6, 7, 5, 6, 5, 5, 6, 7, 8};
+ *  \endcode
+ *
+ *  \see exclusive_scan
+ *  \see inclusive_scan_by_key
+ *
+ */
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename T,
+         typename BinaryPredicate>
+  OutputIterator exclusive_scan_by_key(InputIterator1 first1,
+                                       InputIterator1 last1,
+                                       InputIterator2 first2,
+                                       OutputIterator result,
+                                       T init,
+                                       BinaryPredicate binary_pred);
+
+
+/*! \p exclusive_scan_by_key computes an exclusive key-value or 'segmented' prefix 
+ *  sum operation. The term 'exclusive' means that each result does not include
+ *  the corresponding input operand in the partial sum. The term 'segmented'
+ *  means that the partial sums are broken into distinct segments.  In other
+ *  words, within each segment a separate exclusive scan operation is computed.
+ *  Refer to the code sample below for example usage.
+ *
+ *  This version of \p exclusive_scan_by_key uses the value \c init to
+ *  initialize the exclusive scan operation.
+ *
+ *  This version of \p exclusive_scan_by_key uses the binary predicate \c binary_pred
+ *  to compare adjacent keys.  Specifically, consecutive iterators <tt>i</tt> and
+ *  <tt>i+1</tt> in the range <tt>[first1, last1)</tt> belong to the same segment if 
+ *  <tt>binary_pred(*i, *(i+1))</tt> is true, and belong to different segments otherwise.
+ *
+ *  This version of \p exclusive_scan_by_key uses the associative operator 
+ *  \c binary_op to perform the prefix sum. When the input and output sequences
+ *  are the same, the scan is performed in-place.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first1 The beginning of the key sequence.
+ *  \param last1 The end of the key sequence.
+ *  \param first2 The beginning of the input value sequence.
+ *  \param result The beginning of the output value sequence.
+ *  \param init The initial of the exclusive sum value.
+ *  \param binary_pred The binary predicate used to determine equality of keys.
+ *  \param binary_op The associatve operator used to 'sum' values.
+ *  \return The end of the output sequence.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>
+ *                         and \c InputIterator2's \c value_type is convertible to \c OutputIterator's \c value_type.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>,
+ *                         and if \c x and \c y are objects of \c OutputIterator's \c value_type, then 
+ *                         <tt>binary_op(x,y)</tt> is defined.
+ *  \tparam T is convertible to \c OutputIterator's \c value_type.
+ *  \tparam BinaryPredicate is a model of <a href="http://www.sgi.com/tech/stl/BinaryPredicate.html">Binary Predicate</a>.
+ *  \tparam AssociativeOperator is a model of <a href="http://www.sgi.com/tech/stl/BinaryFunction.html">Binary Function</a>
+ *                         and \c AssociativeOperator's \c result_type is convertible to \c OutputIterator's \c value_type.
+ *
+ *  \pre \p first1 may equal \p result but the range <tt>[first1, last1)</tt> and the range <tt>[result, result + (last1 - first1))</tt> shall not overlap otherwise.
+ *  \pre \p first2 may equal \p result but the range <tt>[first2, first2 + (last1 - first1)</tt> and range <tt>[result, result + (last1 - first1))</tt> shall not overlap otherwise.
+ *
+ *  The following code snippet demonstrates how to use \p exclusive_scan_by_key using the
+ *  \p thrust::host execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/scan.h>
+ *  #include <thrust/functional.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  
+ *  int keys[10] = {0, 0, 0, 1, 1, 2, 3, 3, 3, 3};
+ *  int vals[10] = {1, 1, 1, 1, 1, 1, 1, 1, 1, 1};
+ *
+ *  int init = 5;
+ *
+ *  thrust::equal_to<int> binary_pred;
+ *  thrust::plus<int>     binary_op;
+ *
+ *  thrust::exclusive_scan_by_key(thrust::host, key, key + 10, vals, vals, init, binary_pred, binary_op); // in-place scan
+ *
+ *  // vals is now {5, 6, 7, 5, 6, 5, 5, 6, 7, 8};
+ *  \endcode
+ *
+ *  \see exclusive_scan
+ *  \see inclusive_scan_by_key
+ *
+ */
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename T,
+         typename BinaryPredicate,
+         typename AssociativeOperator>
+  OutputIterator exclusive_scan_by_key(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                       InputIterator1 first1,
+                                       InputIterator1 last1,
+                                       InputIterator2 first2,
+                                       OutputIterator result,
+                                       T init,
+                                       BinaryPredicate binary_pred,
+                                       AssociativeOperator binary_op);
+
+
+/*! \p exclusive_scan_by_key computes an exclusive key-value or 'segmented' prefix 
+ *  sum operation. The term 'exclusive' means that each result does not include
+ *  the corresponding input operand in the partial sum. The term 'segmented'
+ *  means that the partial sums are broken into distinct segments.  In other
+ *  words, within each segment a separate exclusive scan operation is computed.
+ *  Refer to the code sample below for example usage.
+ *
+ *  This version of \p exclusive_scan_by_key uses the value \c init to
+ *  initialize the exclusive scan operation.
+ *
+ *  This version of \p exclusive_scan_by_key uses the binary predicate \c binary_pred
+ *  to compare adjacent keys.  Specifically, consecutive iterators <tt>i</tt> and
+ *  <tt>i+1</tt> in the range <tt>[first1, last1)</tt> belong to the same segment if 
+ *  <tt>binary_pred(*i, *(i+1))</tt> is true, and belong to different segments otherwise.
+ *
+ *  This version of \p exclusive_scan_by_key uses the associative operator 
+ *  \c binary_op to perform the prefix sum. When the input and output sequences
+ *  are the same, the scan is performed in-place.
+ *
+ *  \param first1 The beginning of the key sequence.
+ *  \param last1 The end of the key sequence.
+ *  \param first2 The beginning of the input value sequence.
+ *  \param result The beginning of the output value sequence.
+ *  \param init The initial of the exclusive sum value.
+ *  \param binary_pred The binary predicate used to determine equality of keys.
+ *  \param binary_op The associatve operator used to 'sum' values.
+ *  \return The end of the output sequence.
+ *
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>
+ *                         and \c InputIterator2's \c value_type is convertible to \c OutputIterator's \c value_type.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>,
+ *                         and if \c x and \c y are objects of \c OutputIterator's \c value_type, then 
+ *                         <tt>binary_op(x,y)</tt> is defined.
+ *  \tparam T is convertible to \c OutputIterator's \c value_type.
+ *  \tparam BinaryPredicate is a model of <a href="http://www.sgi.com/tech/stl/BinaryPredicate.html">Binary Predicate</a>.
+ *  \tparam AssociativeOperator is a model of <a href="http://www.sgi.com/tech/stl/BinaryFunction.html">Binary Function</a>
+ *                         and \c AssociativeOperator's \c result_type is convertible to \c OutputIterator's \c value_type.
+ *
+ *  \pre \p first1 may equal \p result but the range <tt>[first1, last1)</tt> and the range <tt>[result, result + (last1 - first1))</tt> shall not overlap otherwise.
+ *  \pre \p first2 may equal \p result but the range <tt>[first2, first2 + (last1 - first1)</tt> and range <tt>[result, result + (last1 - first1))</tt> shall not overlap otherwise.
+ *
+ *  The following code snippet demonstrates how to use \p exclusive_scan_by_key
+ *
+ *  \code
+ *  #include <thrust/scan.h>
+ *  #include <thrust/functional.h>
+ *  
+ *  int keys[10] = {0, 0, 0, 1, 1, 2, 3, 3, 3, 3};
+ *  int vals[10] = {1, 1, 1, 1, 1, 1, 1, 1, 1, 1};
+ *
+ *  int init = 5;
+ *
+ *  thrust::equal_to<int> binary_pred;
+ *  thrust::plus<int>     binary_op;
+ *
+ *  thrust::exclusive_scan_by_key(key, key + 10, vals, vals, init, binary_pred, binary_op); // in-place scan
+ *
+ *  // vals is now {5, 6, 7, 5, 6, 5, 5, 6, 7, 8};
+ *  \endcode
+ *
+ *  \see exclusive_scan
+ *  \see inclusive_scan_by_key
+ *
+ */
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename T,
+         typename BinaryPredicate,
+         typename AssociativeOperator>
+  OutputIterator exclusive_scan_by_key(InputIterator1 first1,
+                                       InputIterator1 last1,
+                                       InputIterator2 first2,
+                                       OutputIterator result,
+                                       T init,
+                                       BinaryPredicate binary_pred,
+                                       AssociativeOperator binary_op);
+
+
+/*! \} // end segmentedprefixsums
+ */
+
+
+/*! \} // end prefix sums
+ */
+
+	
+} // end namespace thrust
+
+#include <thrust/detail/scan.inl>
+
diff --git a/compat/thrust/scatter.h b/compat/thrust/scatter.h
new file mode 100644
index 0000000000..59604ca170
--- /dev/null
+++ b/compat/thrust/scatter.h
@@ -0,0 +1,420 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file scatter.h
+ *  \brief Irregular copying to a destination range
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/execution_policy.h>
+
+namespace thrust
+{
+
+
+/*! \addtogroup scattering
+ *  \ingroup copying
+ *  \{
+ */
+
+
+/*! \p scatter copies elements from a source range into an output array
+ *  according to a map. For each iterator \c i in the range [\p first, \p last),
+ *  the value \c *i is assigned to <tt>output[*(map + (i - first))]</tt>. The 
+ *  output iterator must permit random access. If the same index 
+ *  appears more than once in the range <tt>[map, map + (last - first))</tt>,
+ *  the result is undefined.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first Beginning of the sequence of values to scatter.
+ *  \param last End of the sequence of values to scatter.
+ *  \param map  Beginning of the sequence of output indices.
+ *  \param result Destination of the source elements.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator1 must be a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a> and \c InputIterator1's \c value_type must be convertible to \c RandomAccessIterator's \c value_type.
+ *  \tparam InputIterator2 must be a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a> and \c InputIterator2's \c value_type must be convertible to \c RandomAccessIterator's \c difference_type.
+ *  \tparam RandomAccessIterator must be a model of <a href="http://www.sgi.com/tech/stl/RandomAccessIterator.html">Random Access iterator</a>.
+ *
+ *  \pre The iterator `result + i` shall not refer to any element referenced by any iterator `j` in the range `[first,last)` for all iterators `i` in the range `[map,map + (last - first))`.
+ *
+ *  \pre The iterator `result + i` shall not refer to any element referenced by any iterator `j` in the range `[map,map + (last - first))` for all iterators `i` in the range `[map,map + (last - first))`.
+ *
+ *  \pre The expression `result[*i]` shall be valid for all iterators in the range `[map,map + (last - first))`.
+ *
+ *  The following code snippet demonstrates how to use \p scatter to
+ *  reorder a range using the \p thrust::device execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/scatter.h>
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  // mark even indices with a 1; odd indices with a 0
+ *  int values[10] = {1, 0, 1, 0, 1, 0, 1, 0, 1, 0};
+ *  thrust::device_vector<int> d_values(values, values + 10);
+ *
+ *  // scatter all even indices into the first half of the
+ *  // range, and odd indices vice versa
+ *  int map[10]   = {0, 5, 1, 6, 2, 7, 3, 8, 4, 9};
+ *  thrust::device_vector<int> d_map(map, map + 10);
+ *
+ *  thrust::device_vector<int> d_output(10);
+ *  thrust::scatter(thrust::device,
+ *                  d_values.begin(), d_values.end(),
+ *                  d_map.begin(), d_output.begin());
+ *  // d_output is now {1, 1, 1, 1, 1, 0, 0, 0, 0, 0}
+ *  \endcode
+ *
+ *  \note \p scatter is the inverse of thrust::gather.
+ */
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename RandomAccessIterator>
+  void scatter(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+               InputIterator1 first,
+               InputIterator1 last,
+               InputIterator2 map,
+               RandomAccessIterator result);
+
+
+/*! \p scatter copies elements from a source range into an output array
+ *  according to a map. For each iterator \c i in the range [\p first, \p last),
+ *  the value \c *i is assigned to <tt>output[*(map + (i - first))]</tt>. The 
+ *  output iterator must permit random access. If the same index 
+ *  appears more than once in the range <tt>[map, map + (last - first))</tt>,
+ *  the result is undefined.
+ *
+ *  \param first Beginning of the sequence of values to scatter.
+ *  \param last End of the sequence of values to scatter.
+ *  \param map  Beginning of the sequence of output indices.
+ *  \param result Destination of the source elements.
+ *
+ *  \tparam InputIterator1 must be a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a> and \c InputIterator1's \c value_type must be convertible to \c RandomAccessIterator's \c value_type.
+ *  \tparam InputIterator2 must be a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a> and \c InputIterator2's \c value_type must be convertible to \c RandomAccessIterator's \c difference_type.
+ *  \tparam RandomAccessIterator must be a model of <a href="http://www.sgi.com/tech/stl/RandomAccessIterator.html">Random Access iterator</a>.
+ *
+ *  \pre The iterator `result + i` shall not refer to any element referenced by any iterator `j` in the range `[first,last)` for all iterators `i` in the range `[map,map + (last - first))`.
+ *
+ *  \pre The iterator `result + i` shall not refer to any element referenced by any iterator `j` in the range `[map,map + (last - first))` for all iterators `i` in the range `[map,map + (last - first))`.
+ *
+ *  \pre The expression `result[*i]` shall be valid for all iterators in the range `[map,map + (last - first))`.
+ *
+ *  The following code snippet demonstrates how to use \p scatter to
+ *  reorder a range.
+ *
+ *  \code
+ *  #include <thrust/scatter.h>
+ *  #include <thrust/device_vector.h>
+ *  ...
+ *  // mark even indices with a 1; odd indices with a 0
+ *  int values[10] = {1, 0, 1, 0, 1, 0, 1, 0, 1, 0};
+ *  thrust::device_vector<int> d_values(values, values + 10);
+ *
+ *  // scatter all even indices into the first half of the
+ *  // range, and odd indices vice versa
+ *  int map[10]   = {0, 5, 1, 6, 2, 7, 3, 8, 4, 9};
+ *  thrust::device_vector<int> d_map(map, map + 10);
+ *
+ *  thrust::device_vector<int> d_output(10);
+ *  thrust::scatter(d_values.begin(), d_values.end(),
+ *                  d_map.begin(), d_output.begin());
+ *  // d_output is now {1, 1, 1, 1, 1, 0, 0, 0, 0, 0}
+ *  \endcode
+ *
+ *  \note \p scatter is the inverse of thrust::gather.
+ */
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename RandomAccessIterator>
+  void scatter(InputIterator1 first,
+               InputIterator1 last,
+               InputIterator2 map,
+               RandomAccessIterator result);
+
+
+/*! \p scatter_if conditionally copies elements from a source range into an 
+ *  output array according to a map. For each iterator \c i in the 
+ *  range <tt>[first, last)</tt> such that <tt>*(stencil + (i - first))</tt> is
+ *  true, the value \c *i is assigned to <tt>output[*(map + (i - first))]</tt>.
+ *  The output iterator must permit random access. If the same index 
+ *  appears more than once in the range <tt>[map, map + (last - first))</tt>
+ *  the result is undefined.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first Beginning of the sequence of values to scatter.
+ *  \param last End of the sequence of values to scatter.
+ *  \param map Beginning of the sequence of output indices.
+ *  \param stencil Beginning of the sequence of predicate values.
+ *  \param output Beginning of the destination range.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator1 must be a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a> and \c InputIterator1's \c value_type must be convertible to \c RandomAccessIterator's \c value_type.
+ *  \tparam InputIterator2 must be a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a> and \c InputIterator2's \c value_type must be convertible to \c RandomAccessIterator's \c difference_type.
+ *  \tparam InputIterator3 must be a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a> and \c InputIterator3's \c value_type must be convertible to \c bool.
+ *  \tparam RandomAccessIterator must be a model of <a href="http://www.sgi.com/tech/stl/RandomAccessIterator.html">Random Access iterator</a>.
+ *
+ *  \pre The iterator `result + i` shall not refer to any element referenced by any iterator `j` in the range `[first,last)` for all iterators `i` in the range `[map,map + (last - first))`.
+ *
+ *  \pre The iterator `result + i` shall not refer to any element referenced by any iterator `j` in the range `[map,map + (last - first))` for all iterators `i` in the range `[map,map + (last - first))`.
+ *
+ *  \pre The iterator `result + i` shall not refer to any element referenced by any iterator `j` in the range `[stencil,stencil + (last - first))` for all iterators `i` in the range `[map,map + (last - first))`.
+ *
+ *  \pre The expression `result[*i]` shall be valid for all iterators `i` in the range `[map,map + (last - first))` for which the following condition holds: `*(stencil + i) != false`.
+ *
+ *  \code
+ *  #include <thrust/scatter.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  int V[8] = {10, 20, 30, 40, 50, 60, 70, 80};
+ *  int M[8] = {0, 5, 1, 6, 2, 7, 3, 4};
+ *  int S[8] = {1, 0, 1, 0, 1, 0, 1, 0};
+ *  int D[8] = {0, 0, 0, 0, 0, 0, 0, 0};
+ * 
+ *  thrust::scatter_if(thrust::host, V, V + 8, M, S, D);
+ * 
+ *  // D contains [10, 30, 50, 70, 0, 0, 0, 0];
+ *  \endcode
+ *
+ *  \note \p scatter_if is the inverse of thrust::gather_if.
+ */
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename RandomAccessIterator>
+  void scatter_if(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                  InputIterator1 first,
+                  InputIterator1 last,
+                  InputIterator2 map,
+                  InputIterator3 stencil,
+                  RandomAccessIterator output);
+
+
+/*! \p scatter_if conditionally copies elements from a source range into an 
+ *  output array according to a map. For each iterator \c i in the 
+ *  range <tt>[first, last)</tt> such that <tt>*(stencil + (i - first))</tt> is
+ *  true, the value \c *i is assigned to <tt>output[*(map + (i - first))]</tt>.
+ *  The output iterator must permit random access. If the same index 
+ *  appears more than once in the range <tt>[map, map + (last - first))</tt>
+ *  the result is undefined.
+ *
+ *  \param first Beginning of the sequence of values to scatter.
+ *  \param last End of the sequence of values to scatter.
+ *  \param map Beginning of the sequence of output indices.
+ *  \param stencil Beginning of the sequence of predicate values.
+ *  \param output Beginning of the destination range.
+ *
+ *  \tparam InputIterator1 must be a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a> and \c InputIterator1's \c value_type must be convertible to \c RandomAccessIterator's \c value_type.
+ *  \tparam InputIterator2 must be a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a> and \c InputIterator2's \c value_type must be convertible to \c RandomAccessIterator's \c difference_type.
+ *  \tparam InputIterator3 must be a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a> and \c InputIterator3's \c value_type must be convertible to \c bool.
+ *  \tparam RandomAccessIterator must be a model of <a href="http://www.sgi.com/tech/stl/RandomAccessIterator.html">Random Access iterator</a>.
+ *
+ *  \pre The iterator `result + i` shall not refer to any element referenced by any iterator `j` in the range `[first,last)` for all iterators `i` in the range `[map,map + (last - first))`.
+ *
+ *  \pre The iterator `result + i` shall not refer to any element referenced by any iterator `j` in the range `[map,map + (last - first))` for all iterators `i` in the range `[map,map + (last - first))`.
+ *
+ *  \pre The iterator `result + i` shall not refer to any element referenced by any iterator `j` in the range `[stencil,stencil + (last - first))` for all iterators `i` in the range `[map,map + (last - first))`.
+ *
+ *  \pre The expression `result[*i]` shall be valid for all iterators `i` in the range `[map,map + (last - first))` for which the following condition holds: `*(stencil + i) != false`.
+ *
+ *  \code
+ *  #include <thrust/scatter.h>
+ *  ...
+ *  int V[8] = {10, 20, 30, 40, 50, 60, 70, 80};
+ *  int M[8] = {0, 5, 1, 6, 2, 7, 3, 4};
+ *  int S[8] = {1, 0, 1, 0, 1, 0, 1, 0};
+ *  int D[8] = {0, 0, 0, 0, 0, 0, 0, 0};
+ * 
+ *  thrust::scatter_if(V, V + 8, M, S, D);
+ * 
+ *  // D contains [10, 30, 50, 70, 0, 0, 0, 0];
+ *  \endcode
+ *
+ *  \note \p scatter_if is the inverse of thrust::gather_if.
+ */
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename RandomAccessIterator>
+  void scatter_if(InputIterator1 first,
+                  InputIterator1 last,
+                  InputIterator2 map,
+                  InputIterator3 stencil,
+                  RandomAccessIterator output);
+
+
+/*! \p scatter_if conditionally copies elements from a source range into an 
+ *  output array according to a map. For each iterator \c i in the 
+ *  range <tt>[first, last)</tt> such that <tt>pred(*(stencil + (i - first)))</tt> is
+ *  \c true, the value \c *i is assigned to <tt>output[*(map + (i - first))]</tt>.
+ *  The output iterator must permit random access. If the same index 
+ *  appears more than once in the range <tt>[map, map + (last - first))</tt>
+ *  the result is undefined.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first Beginning of the sequence of values to scatter.
+ *  \param last End of the sequence of values to scatter.
+ *  \param map Beginning of the sequence of output indices.
+ *  \param stencil Beginning of the sequence of predicate values.
+ *  \param output Beginning of the destination range.
+ *  \param pred Predicate to apply to the stencil values.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator1 must be a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a> and \c InputIterator1's \c value_type must be convertible to \c RandomAccessIterator's \c value_type.
+ *  \tparam InputIterator2 must be a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a> and \c InputIterator2's \c value_type must be convertible to \c RandomAccessIterator's \c difference_type.
+ *  \tparam InputIterator3 must be a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a> and \c InputIterator3's \c value_type must be convertible to \c Predicate's \c argument_type.
+ *  \tparam RandomAccessIterator must be a model of <a href="http://www.sgi.com/tech/stl/RandomAccessIterator.html">Random Access iterator</a>.
+ *  \tparam Predicate must be a model of <a href="http://www.sgi.com/tech/stl/Predicate.html">Predicate</a>.
+ *
+ *  \pre The iterator `result + i` shall not refer to any element referenced by any iterator `j` in the range `[first,last)` for all iterators `i` in the range `[map,map + (last - first))`.
+ *
+ *  \pre The iterator `result + i` shall not refer to any element referenced by any iterator `j` in the range `[map,map + (last - first))` for all iterators `i` in the range `[map,map + (last - first))`.
+ *
+ *  \pre The iterator `result + i` shall not refer to any element referenced by any iterator `j` in the range `[stencil,stencil + (last - first))` for all iterators `i` in the range `[map,map + (last - first))`.
+ *
+ *  \pre The expression `result[*i]` shall be valid for all iterators `i` in the range `[map,map + (last - first))` for which the following condition holds: `pred(*(stencil + i)) != false`.
+ *
+ *  \code
+ *  #include <thrust/scatter.h>
+ *  #include <thrust/execution_policy.h>
+ *
+ *  struct is_even
+ *  {
+ *    __host__ __device__
+ *    bool operator()(int x)
+ *    {
+ *      return (x % 2) == 0;
+ *    }
+ *  };
+ *
+ *  ...
+ *
+ *  int V[8] = {10, 20, 30, 40, 50, 60, 70, 80};
+ *  int M[8] = {0, 5, 1, 6, 2, 7, 3, 4};
+ *  int S[8] = {2, 1, 2, 1, 2, 1, 2, 1};
+ *  int D[8] = {0, 0, 0, 0, 0, 0, 0, 0};
+ * 
+ *  is_even pred;
+ *  thrust::scatter_if(thrust::host, V, V + 8, M, S, D, pred);
+ * 
+ *  // D contains [10, 30, 50, 70, 0, 0, 0, 0];
+ *  \endcode
+ *  
+ *  \note \p scatter_if is the inverse of thrust::gather_if.
+ */
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename RandomAccessIterator,
+         typename Predicate>
+  void scatter_if(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                  InputIterator1 first,
+                  InputIterator1 last,
+                  InputIterator2 map,
+                  InputIterator3 stencil,
+                  RandomAccessIterator output,
+                  Predicate pred);
+                  
+
+/*! \p scatter_if conditionally copies elements from a source range into an 
+ *  output array according to a map. For each iterator \c i in the 
+ *  range <tt>[first, last)</tt> such that <tt>pred(*(stencil + (i - first)))</tt> is
+ *  \c true, the value \c *i is assigned to <tt>output[*(map + (i - first))]</tt>.
+ *  The output iterator must permit random access. If the same index 
+ *  appears more than once in the range <tt>[map, map + (last - first))</tt>
+ *  the result is undefined.
+ *
+ *  \param first Beginning of the sequence of values to scatter.
+ *  \param last End of the sequence of values to scatter.
+ *  \param map Beginning of the sequence of output indices.
+ *  \param stencil Beginning of the sequence of predicate values.
+ *  \param output Beginning of the destination range.
+ *  \param pred Predicate to apply to the stencil values.
+ *
+ *  \tparam InputIterator1 must be a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a> and \c InputIterator1's \c value_type must be convertible to \c RandomAccessIterator's \c value_type.
+ *  \tparam InputIterator2 must be a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a> and \c InputIterator2's \c value_type must be convertible to \c RandomAccessIterator's \c difference_type.
+ *  \tparam InputIterator3 must be a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a> and \c InputIterator3's \c value_type must be convertible to \c Predicate's \c argument_type.
+ *  \tparam RandomAccessIterator must be a model of <a href="http://www.sgi.com/tech/stl/RandomAccessIterator.html">Random Access iterator</a>.
+ *  \tparam Predicate must be a model of <a href="http://www.sgi.com/tech/stl/Predicate.html">Predicate</a>.
+ *
+ *  \pre The iterator `result + i` shall not refer to any element referenced by any iterator `j` in the range `[first,last)` for all iterators `i` in the range `[map,map + (last - first))`.
+ *
+ *  \pre The iterator `result + i` shall not refer to any element referenced by any iterator `j` in the range `[map,map + (last - first))` for all iterators `i` in the range `[map,map + (last - first))`.
+ *
+ *  \pre The iterator `result + i` shall not refer to any element referenced by any iterator `j` in the range `[stencil,stencil + (last - first))` for all iterators `i` in the range `[map,map + (last - first))`.
+ *
+ *  \pre The expression `result[*i]` shall be valid for all iterators `i` in the range `[map,map + (last - first))` for which the following condition holds: `pred(*(stencil + i)) != false`.
+ *
+ *  \code
+ *  #include <thrust/scatter.h>
+ *
+ *  struct is_even
+ *  {
+ *    __host__ __device__
+ *    bool operator()(int x)
+ *    {
+ *      return (x % 2) == 0;
+ *    }
+ *  };
+ *
+ *  ...
+ *
+ *  int V[8] = {10, 20, 30, 40, 50, 60, 70, 80};
+ *  int M[8] = {0, 5, 1, 6, 2, 7, 3, 4};
+ *  int S[8] = {2, 1, 2, 1, 2, 1, 2, 1};
+ *  int D[8] = {0, 0, 0, 0, 0, 0, 0, 0};
+ * 
+ *  is_even pred;
+ *  thrust::scatter_if(V, V + 8, M, S, D, pred);
+ * 
+ *  // D contains [10, 30, 50, 70, 0, 0, 0, 0];
+ *  \endcode
+ *  
+ *  \note \p scatter_if is the inverse of thrust::gather_if.
+ */
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename RandomAccessIterator,
+         typename Predicate>
+  void scatter_if(InputIterator1 first,
+                  InputIterator1 last,
+                  InputIterator2 map,
+                  InputIterator3 stencil,
+                  RandomAccessIterator output,
+                  Predicate pred);
+
+
+/*! \} // end scattering
+ */
+
+
+} // end namespace thrust
+
+#include <thrust/detail/scatter.inl>
+
diff --git a/compat/thrust/sequence.h b/compat/thrust/sequence.h
new file mode 100644
index 0000000000..6c54a5bbf4
--- /dev/null
+++ b/compat/thrust/sequence.h
@@ -0,0 +1,293 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file sequence.h
+ *  \brief Fills a range with a sequence of numbers
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/execution_policy.h>
+
+namespace thrust
+{
+
+
+/*! \addtogroup transformations
+ *  \{
+ */
+
+
+/*! \p sequence fills the range <tt>[first, last)</tt> with a sequence of numbers.
+ *
+ *  For each iterator \c i in the range <tt>[first, last)</tt>, this version of 
+ *  \p sequence performs the assignment <tt>*i =  (i - first)</tt>.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the sequence.
+ *  \param last The end of the sequence.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>,
+ *          and \p ForwardIterator is mutable,
+ *          and if \c x and \c y are objects of \c ForwardIterator's \c value_type, then <tt>x + y</tt> is defined,
+ *          and if \c T is \p ForwardIterator's \c value_type, then <tt>T(0)</tt> is defined.
+ *
+ *  The following code snippet demonstrates how to use \p sequence to fill a range
+ *  with a sequence of numbers using the \p thrust::host execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/sequence.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  const int N = 10;
+ *  int A[N];
+ *  thrust::sequence(thrust::host, A, A + 10);
+ *  // A is now {0, 1, 2, 3, 4, 5, 6, 7, 8, 9}
+ *  \endcode
+ *
+ *  \note Unlike the similar C++ STL function \c std::iota, \p sequence offers no
+ *        guarantee on order of execution.
+ *
+ *  \see http://www.sgi.com/tech/stl/iota.html
+ */
+template<typename DerivedPolicy, typename ForwardIterator>
+  void sequence(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                ForwardIterator first,
+                ForwardIterator last);
+
+
+/*! \p sequence fills the range <tt>[first, last)</tt> with a sequence of numbers.
+ *
+ *  For each iterator \c i in the range <tt>[first, last)</tt>, this version of 
+ *  \p sequence performs the assignment <tt>*i =  (i - first)</tt>.
+ *
+ *  \param first The beginning of the sequence.
+ *  \param last The end of the sequence.
+ *
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>,
+ *          and \p ForwardIterator is mutable,
+ *          and if \c x and \c y are objects of \c ForwardIterator's \c value_type, then <tt>x + y</tt> is defined,
+ *          and if \c T is \p ForwardIterator's \c value_type, then <tt>T(0)</tt> is defined.
+ *
+ *  The following code snippet demonstrates how to use \p sequence to fill a range
+ *  with a sequence of numbers.
+ *
+ *  \code
+ *  #include <thrust/sequence.h>
+ *  ...
+ *  const int N = 10;
+ *  int A[N];
+ *  thrust::sequence(A, A + 10);
+ *  // A is now {0, 1, 2, 3, 4, 5, 6, 7, 8, 9}
+ *  \endcode
+ *
+ *  \note Unlike the similar C++ STL function \c std::iota, \p sequence offers no
+ *        guarantee on order of execution.
+ *
+ *  \see http://www.sgi.com/tech/stl/iota.html
+ */
+template<typename ForwardIterator>
+  void sequence(ForwardIterator first,
+                ForwardIterator last);
+
+
+/*! \p sequence fills the range <tt>[first, last)</tt> with a sequence of numbers.
+ *
+ *  For each iterator \c i in the range <tt>[first, last)</tt>, this version of 
+ *  \p sequence performs the assignment <tt>*i =  init + (i - first)</tt>.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the sequence.
+ *  \param last The end of the sequence.
+ *  \param init The first value of the sequence of numbers.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>,
+ *          and \p ForwardIterator is mutable,
+ *          and if \c x and \c y are objects of \c ForwardIterator's \c value_type, then <tt>x + y</tt> is defined,
+ *          and if \c T is \p ForwardIterator's \c value_type, then <tt>T(0)</tt> is defined.
+ *  \tparam T is a model of <a href="http://www.sgi.com/tech/stl/Assignable.html">Assignable</a>,
+ *          and \p T is convertible to \p ForwardIterator's \c value_type.
+ *
+ *  The following code snippet demonstrates how to use \p sequence to fill a range
+ *  with a sequence of numbers starting from the value 1 using the \p thrust::host execution
+ *  policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/sequence.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  const int N = 10;
+ *  int A[N];
+ *  thrust::sequence(thrust::host, A, A + 10, 1);
+ *  // A is now {1, 2, 3, 4, 5, 6, 7, 8, 9, 10}
+ *  \endcode
+ *
+ *  \note Unlike the similar C++ STL function \c std::iota, \p sequence offers no
+ *        guarantee on order of execution.
+ *
+ *  \see http://www.sgi.com/tech/stl/iota.html
+ */
+template<typename DerivedPolicy, typename ForwardIterator, typename T>
+  void sequence(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                ForwardIterator first,
+                ForwardIterator last,
+                T init);
+
+
+/*! \p sequence fills the range <tt>[first, last)</tt> with a sequence of numbers.
+ *
+ *  For each iterator \c i in the range <tt>[first, last)</tt>, this version of 
+ *  \p sequence performs the assignment <tt>*i =  init + (i - first)</tt>.
+ *
+ *  \param first The beginning of the sequence.
+ *  \param last The end of the sequence.
+ *  \param init The first value of the sequence of numbers.
+ *
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>,
+ *          and \p ForwardIterator is mutable,
+ *          and if \c x and \c y are objects of \c ForwardIterator's \c value_type, then <tt>x + y</tt> is defined,
+ *          and if \c T is \p ForwardIterator's \c value_type, then <tt>T(0)</tt> is defined.
+ *  \tparam T is a model of <a href="http://www.sgi.com/tech/stl/Assignable.html">Assignable</a>,
+ *          and \p T is convertible to \p ForwardIterator's \c value_type.
+ *
+ *  The following code snippet demonstrates how to use \p sequence to fill a range
+ *  with a sequence of numbers starting from the value 1.
+ *
+ *  \code
+ *  #include <thrust/sequence.h>
+ *  ...
+ *  const int N = 10;
+ *  int A[N];
+ *  thrust::sequence(A, A + 10, 1);
+ *  // A is now {1, 2, 3, 4, 5, 6, 7, 8, 9, 10}
+ *  \endcode
+ *
+ *  \note Unlike the similar C++ STL function \c std::iota, \p sequence offers no
+ *        guarantee on order of execution.
+ *
+ *  \see http://www.sgi.com/tech/stl/iota.html
+ */
+template<typename ForwardIterator, typename T>
+  void sequence(ForwardIterator first,
+                ForwardIterator last,
+                T init);
+
+
+/*! \p sequence fills the range <tt>[first, last)</tt> with a sequence of numbers.
+ *
+ *  For each iterator \c i in the range <tt>[first, last)</tt>, this version of 
+ *  \p sequence performs the assignment <tt>*i =  init + step * (i - first)</tt>.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the sequence.
+ *  \param last The end of the sequence.
+ *  \param init The first value of the sequence of numbers
+ *  \param step The difference between consecutive elements.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>,
+ *          and \p ForwardIterator is mutable,
+ *          and if \c x and \c y are objects of \c ForwardIterator's \c value_type, then <tt>x + y</tt> is defined,
+ *          and if \c T is \p ForwardIterator's \c value_type, then <tt>T(0)</tt> is defined.
+ *  \tparam T is a model of <a href="http://www.sgi.com/tech/stl/Assignable.html">Assignable</a>,
+ *          and \p T is convertible to \p ForwardIterator's \c value_type.
+ *
+ *  The following code snippet demonstrates how to use \p sequence to fill a range
+ *  with a sequence of numbers starting from the value 1 with a step size of 3 using the \p thrust::host
+ *  execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/sequence.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  const int N = 10;
+ *  int A[N];
+ *  thrust::sequence(thrust::host, A, A + 10, 1, 3);
+ *  // A is now {1, 4, 7, 10, 13, 16, 19, 22, 25, 28}
+ *  \endcode
+ *
+ *  \note Unlike the similar C++ STL function \c std::iota, \p sequence offers no
+ *        guarantee on order of execution.
+ *
+ *  \see http://www.sgi.com/tech/stl/iota.html
+ */
+template<typename DerivedPolicy, typename ForwardIterator, typename T>
+  void sequence(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                ForwardIterator first,
+                ForwardIterator last,
+                T init,
+                T step);
+
+
+/*! \p sequence fills the range <tt>[first, last)</tt> with a sequence of numbers.
+ *
+ *  For each iterator \c i in the range <tt>[first, last)</tt>, this version of 
+ *  \p sequence performs the assignment <tt>*i =  init + step * (i - first)</tt>.
+ *
+ *  \param first The beginning of the sequence.
+ *  \param last The end of the sequence.
+ *  \param init The first value of the sequence of numbers
+ *  \param step The difference between consecutive elements.
+ *
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>,
+ *          and \p ForwardIterator is mutable,
+ *          and if \c x and \c y are objects of \c ForwardIterator's \c value_type, then <tt>x + y</tt> is defined,
+ *          and if \c T is \p ForwardIterator's \c value_type, then <tt>T(0)</tt> is defined.
+ *  \tparam T is a model of <a href="http://www.sgi.com/tech/stl/Assignable.html">Assignable</a>,
+ *          and \p T is convertible to \p ForwardIterator's \c value_type.
+ *
+ *  The following code snippet demonstrates how to use \p sequence to fill a range
+ *  with a sequence of numbers starting from the value 1 with a step size of 3.
+ *
+ *  \code
+ *  #include <thrust/sequence.h>
+ *  ...
+ *  const int N = 10;
+ *  int A[N];
+ *  thrust::sequence(A, A + 10, 1, 3);
+ *  // A is now {1, 4, 7, 10, 13, 16, 19, 22, 25, 28}
+ *  \endcode
+ *
+ *  \note Unlike the similar C++ STL function \c std::iota, \p sequence offers no
+ *        guarantee on order of execution.
+ *
+ *  \see http://www.sgi.com/tech/stl/iota.html
+ */
+template<typename ForwardIterator, typename T>
+  void sequence(ForwardIterator first,
+                ForwardIterator last,
+                T init,
+                T step);
+
+
+/*! \} // end transformations
+ */
+
+
+} // end namespace thrust
+
+#include <thrust/detail/sequence.inl>
+
diff --git a/compat/thrust/set_operations.h b/compat/thrust/set_operations.h
new file mode 100644
index 0000000000..a7ee624f92
--- /dev/null
+++ b/compat/thrust/set_operations.h
@@ -0,0 +1,2947 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file set_operations.h
+ *  \brief Set theoretic operations for sorted ranges
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/execution_policy.h>
+#include <thrust/pair.h>
+
+namespace thrust
+{
+
+
+/*! \addtogroup set_operations Set Operations
+ *  \ingroup algorithms
+ *  \{
+ */
+
+
+/*! \p set_difference constructs a sorted range that is the set difference of the sorted
+ *  ranges <tt>[first1, last1)</tt> and <tt>[first2, last2)</tt>. The return value is the
+ *  end of the output range.
+ *
+ *  In the simplest case, \p set_difference performs the "difference" operation from set
+ *  theory: the output range contains a copy of every element that is contained in
+ *  <tt>[first1, last1)</tt> and not contained in <tt>[first2, last1)</tt>. The general case
+ *  is more complicated, because the input ranges may contain duplicate elements.
+ *  The generalization is that if <tt>[first1, last1)</tt> contains \c m elements
+ *  that are equivalent to each other and if <tt>[first2, last2)</tt> contains \c n
+ *  elements that are equivalent to them, the last <tt>max(m-n,0)</tt> elements from
+ *  <tt>[first1, last1)</tt> range shall be copied to the output range.
+ *
+ *  This version of \p set_difference compares elements using \c operator<.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first1 The beginning of the first input range.
+ *  \param last1 The end of the first input range.
+ *  \param first2 The beginning of the second input range.
+ *  \param last2 The end of the second input range.
+ *  \param result The beginning of the output range.
+ *  \return The end of the output range.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator1 and \p InputIterator2 have the same \c value_type,
+ *          \p InputIterator1's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a>,
+ *          the ordering on \p InputIterator1's \c value_type is a strict weak ordering, as defined in the <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a> requirements,
+ *          and \p InputIterator1's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator2 and \p InputIterator1 have the same \c value_type,
+ *          \p InputIterator2's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a>,
+ *          the ordering on \p InputIterator2's \c value_type is a strict weak ordering, as defined in the <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a> requirements,
+ *          and \p InputIterator2's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *
+ *  \pre The ranges <tt>[first1, last1)</tt> and <tt>[first2, last2)</tt> shall be sorted with respect to <tt>operator<</tt>.
+ *  \pre The resulting range shall not overlap with either input range.
+ *
+ *  The following code snippet demonstrates how to use \p set_difference to compute the
+ *  set difference of two sets of integers sorted in ascending order using the \p thrust::host execution
+ *  policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/set_operations.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  int A1[6] = {0, 1, 3, 4, 5, 6, 9};
+ *  int A2[5] = {1, 3, 5, 7, 9};
+ *
+ *  int result[3];
+ *
+ *  int *result_end = thrust::set_difference(thrust::host, A1, A1 + 6, A2, A2 + 5, result);
+ *  // result is now {0, 4, 6}
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/set_difference.html
+ *  \see \p includes
+ *  \see \p set_union
+ *  \see \p set_intersection
+ *  \see \p set_symmetric_difference
+ *  \see \p sort
+ *  \see \p is_sorted
+ */
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator>
+  OutputIterator set_difference(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                InputIterator1                                              first1,
+                                InputIterator1                                              last1,
+                                InputIterator2                                              first2,
+                                InputIterator2                                              last2,
+                                OutputIterator                                              result);
+
+
+/*! \p set_difference constructs a sorted range that is the set difference of the sorted
+ *  ranges <tt>[first1, last1)</tt> and <tt>[first2, last2)</tt>. The return value is the
+ *  end of the output range.
+ *
+ *  In the simplest case, \p set_difference performs the "difference" operation from set
+ *  theory: the output range contains a copy of every element that is contained in
+ *  <tt>[first1, last1)</tt> and not contained in <tt>[first2, last1)</tt>. The general case
+ *  is more complicated, because the input ranges may contain duplicate elements.
+ *  The generalization is that if <tt>[first1, last1)</tt> contains \c m elements
+ *  that are equivalent to each other and if <tt>[first2, last2)</tt> contains \c n
+ *  elements that are equivalent to them, the last <tt>max(m-n,0)</tt> elements from
+ *  <tt>[first1, last1)</tt> range shall be copied to the output range.
+ *
+ *  This version of \p set_difference compares elements using \c operator<.
+ *
+ *  \param first1 The beginning of the first input range.
+ *  \param last1 The end of the first input range.
+ *  \param first2 The beginning of the second input range.
+ *  \param last2 The end of the second input range.
+ *  \param result The beginning of the output range.
+ *  \return The end of the output range.
+ *
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator1 and \p InputIterator2 have the same \c value_type,
+ *          \p InputIterator1's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a>,
+ *          the ordering on \p InputIterator1's \c value_type is a strict weak ordering, as defined in the <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a> requirements,
+ *          and \p InputIterator1's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator2 and \p InputIterator1 have the same \c value_type,
+ *          \p InputIterator2's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a>,
+ *          the ordering on \p InputIterator2's \c value_type is a strict weak ordering, as defined in the <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a> requirements,
+ *          and \p InputIterator2's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *
+ *  \pre The ranges <tt>[first1, last1)</tt> and <tt>[first2, last2)</tt> shall be sorted with respect to <tt>operator<</tt>.
+ *  \pre The resulting range shall not overlap with either input range.
+ *
+ *  The following code snippet demonstrates how to use \p set_difference to compute the
+ *  set difference of two sets of integers sorted in ascending order.
+ *
+ *  \code
+ *  #include <thrust/set_operations.h>
+ *  ...
+ *  int A1[6] = {0, 1, 3, 4, 5, 6, 9};
+ *  int A2[5] = {1, 3, 5, 7, 9};
+ *
+ *  int result[3];
+ *
+ *  int *result_end = thrust::set_difference(A1, A1 + 6, A2, A2 + 5, result);
+ *  // result is now {0, 4, 6}
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/set_difference.html
+ *  \see \p includes
+ *  \see \p set_union
+ *  \see \p set_intersection
+ *  \see \p set_symmetric_difference
+ *  \see \p sort
+ *  \see \p is_sorted
+ */
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator>
+  OutputIterator set_difference(InputIterator1 first1,
+                                InputIterator1 last1,
+                                InputIterator2 first2,
+                                InputIterator2 last2,
+                                OutputIterator result);
+
+
+/*! \p set_difference constructs a sorted range that is the set difference of the sorted
+ *  ranges <tt>[first1, last1)</tt> and <tt>[first2, last2)</tt>. The return value is the
+ *  end of the output range.
+ *
+ *  In the simplest case, \p set_difference performs the "difference" operation from set
+ *  theory: the output range contains a copy of every element that is contained in
+ *  <tt>[first1, last1)</tt> and not contained in <tt>[first2, last1)</tt>. The general case
+ *  is more complicated, because the input ranges may contain duplicate elements.
+ *  The generalization is that if <tt>[first1, last1)</tt> contains \c m elements
+ *  that are equivalent to each other and if <tt>[first2, last2)</tt> contains \c n
+ *  elements that are equivalent to them, the last <tt>max(m-n,0)</tt> elements from
+ *  <tt>[first1, last1)</tt> range shall be copied to the output range.
+ *
+ *  This version of \p set_difference compares elements using a function object \p comp.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first1 The beginning of the first input range.
+ *  \param last1 The end of the first input range.
+ *  \param first2 The beginning of the second input range.
+ *  \param last2 The end of the second input range.
+ *  \param result The beginning of the output range.
+ *  \param comp Comparison operator.
+ *  \return The end of the output range.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator1's \c value_type is convertable to \p StrictWeakCompare's \c first_argument_type.
+ *          and \p InputIterator1's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator2's \c value_type is convertable to \p StrictWeakCompare's \c second_argument_type.
+ *          and \p InputIterator2's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam StrictWeakCompare is a model of <a href="http://www.sgi.com/tech/stl/StrictWeakOrdering.html">Strict Weak Ordering</a>.
+ *
+ *  \pre The ranges <tt>[first1, last1)</tt> and <tt>[first2, last2)</tt> shall be sorted with respect to \p comp.
+ *  \pre The resulting range shall not overlap with either input range.
+ *
+ *  The following code snippet demonstrates how to use \p set_difference to compute the
+ *  set difference of two sets of integers sorted in descending order using the \p thrust::host execution
+ *  policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/set_operations.h>
+ *  #include <thrust/functional.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  int A1[6] = {9, 6, 5, 4, 3, 1, 0};
+ *  int A2[5] = {9, 7, 5, 3, 1};
+ *
+ *  int result[3];
+ *
+ *  int *result_end = thrust::set_difference(thrust::host, A1, A1 + 6, A2, A2 + 5, result, thrust::greater<int>());
+ *  // result is now {6, 4, 0}
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/set_difference.html
+ *  \see \p includes
+ *  \see \p set_union
+ *  \see \p set_intersection
+ *  \see \p set_symmetric_difference
+ *  \see \p sort
+ *  \see \p is_sorted
+ */
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename StrictWeakCompare>
+  OutputIterator set_difference(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                InputIterator1                                              first1,
+                                InputIterator1                                              last1,
+                                InputIterator2                                              first2,
+                                InputIterator2                                              last2,
+                                OutputIterator                                              result,
+                                StrictWeakCompare                                           comp);
+
+
+/*! \p set_difference constructs a sorted range that is the set difference of the sorted
+ *  ranges <tt>[first1, last1)</tt> and <tt>[first2, last2)</tt>. The return value is the
+ *  end of the output range.
+ *
+ *  In the simplest case, \p set_difference performs the "difference" operation from set
+ *  theory: the output range contains a copy of every element that is contained in
+ *  <tt>[first1, last1)</tt> and not contained in <tt>[first2, last1)</tt>. The general case
+ *  is more complicated, because the input ranges may contain duplicate elements.
+ *  The generalization is that if <tt>[first1, last1)</tt> contains \c m elements
+ *  that are equivalent to each other and if <tt>[first2, last2)</tt> contains \c n
+ *  elements that are equivalent to them, the last <tt>max(m-n,0)</tt> elements from
+ *  <tt>[first1, last1)</tt> range shall be copied to the output range.
+ *
+ *  This version of \p set_difference compares elements using a function object \p comp.
+ *
+ *  \param first1 The beginning of the first input range.
+ *  \param last1 The end of the first input range.
+ *  \param first2 The beginning of the second input range.
+ *  \param last2 The end of the second input range.
+ *  \param result The beginning of the output range.
+ *  \param comp Comparison operator.
+ *  \return The end of the output range.
+ *
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator1's \c value_type is convertable to \p StrictWeakCompare's \c first_argument_type.
+ *          and \p InputIterator1's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator2's \c value_type is convertable to \p StrictWeakCompare's \c second_argument_type.
+ *          and \p InputIterator2's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam StrictWeakCompare is a model of <a href="http://www.sgi.com/tech/stl/StrictWeakOrdering.html">Strict Weak Ordering</a>.
+ *
+ *  \pre The ranges <tt>[first1, last1)</tt> and <tt>[first2, last2)</tt> shall be sorted with respect to \p comp.
+ *  \pre The resulting range shall not overlap with either input range.
+ *
+ *  The following code snippet demonstrates how to use \p set_difference to compute the
+ *  set difference of two sets of integers sorted in descending order.
+ *
+ *  \code
+ *  #include <thrust/set_operations.h>
+ *  #include <thrust/functional.h>
+ *  ...
+ *  int A1[6] = {9, 6, 5, 4, 3, 1, 0};
+ *  int A2[5] = {9, 7, 5, 3, 1};
+ *
+ *  int result[3];
+ *
+ *  int *result_end = thrust::set_difference(A1, A1 + 6, A2, A2 + 5, result, thrust::greater<int>());
+ *  // result is now {6, 4, 0}
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/set_difference.html
+ *  \see \p includes
+ *  \see \p set_union
+ *  \see \p set_intersection
+ *  \see \p set_symmetric_difference
+ *  \see \p sort
+ *  \see \p is_sorted
+ */
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename StrictWeakCompare>
+  OutputIterator set_difference(InputIterator1 first1,
+                                InputIterator1 last1,
+                                InputIterator2 first2,
+                                InputIterator2 last2,
+                                OutputIterator result,
+                                StrictWeakCompare comp);
+
+
+/*! \p set_intersection constructs a sorted range that is the
+ *  intersection of sorted ranges <tt>[first1, last1)</tt> and
+ *  <tt>[first2, last2)</tt>. The return value is the end of the
+ *  output range.
+ *
+ *  In the simplest case, \p set_intersection performs the
+ *  "intersection" operation from set theory: the output range
+ *  contains a copy of every element that is contained in both
+ *  <tt>[first1, last1)</tt> and <tt>[first2, last2)</tt>. The
+ *  general case is more complicated, because the input ranges may
+ *  contain duplicate elements. The generalization is that if a value
+ *  appears \c m times in <tt>[first1, last1)</tt> and \c n times in
+ *  <tt>[first2, last2)</tt> (where \c m may be zero), then it
+ *  appears <tt>min(m,n)</tt> times in the output range.
+ *  \p set_intersection is stable, meaning that both elements are
+ *  copied from the first range rather than the second, and that the
+ *  relative order of elements in the output range is the same as in
+ *  the first input range.
+ *
+ *  This version of \p set_intersection compares objects using
+ *  \c operator<.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first1 The beginning of the first input range.
+ *  \param last1 The end of the first input range.
+ *  \param first2 The beginning of the second input range.
+ *  \param last2 The end of the second input range.
+ *  \param result The beginning of the output range.
+ *  \return The end of the output range.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator1 and \p InputIterator2 have the same \c value_type,
+ *          \p InputIterator1's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a>,
+ *          the ordering on \p InputIterator1's \c value_type is a strict weak ordering, as defined in the <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a> requirements,
+ *          and \p InputIterator1's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator2 and \p InputIterator1 have the same \c value_type,
+ *          \p InputIterator2's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a>,
+ *          the ordering on \p InputIterator2's \c value_type is a strict weak ordering, as defined in the <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a> requirements,
+ *          and \p InputIterator2's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *
+ *  \pre The ranges <tt>[first1, last1)</tt> and <tt>[first2, last2)</tt> shall be sorted with respect to <tt>operator<</tt>.
+ *  \pre The resulting range shall not overlap with either input range.
+ *
+ *  The following code snippet demonstrates how to use \p set_intersection to compute the
+ *  set intersection of two sets of integers sorted in ascending order using the \p thrust::host execution
+ *  policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/set_operations.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  int A1[6] = {1, 3, 5, 7, 9, 11};
+ *  int A2[7] = {1, 1, 2, 3, 5,  8, 13};
+ *
+ *  int result[7];
+ *
+ *  int *result_end = thrust::set_intersection(thrust::host, A1, A1 + 6, A2, A2 + 7, result);
+ *  // result is now {1, 3, 5}
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/set_intersection.html
+ *  \see \p includes
+ *  \see \p set_union
+ *  \see \p set_intersection
+ *  \see \p set_symmetric_difference
+ *  \see \p sort
+ *  \see \p is_sorted
+ */
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator>
+  OutputIterator set_intersection(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                  InputIterator1                                              first1,
+                                  InputIterator1                                              last1,
+                                  InputIterator2                                              first2,
+                                  InputIterator2                                              last2,
+                                  OutputIterator                                              result);
+
+
+/*! \p set_intersection constructs a sorted range that is the
+ *  intersection of sorted ranges <tt>[first1, last1)</tt> and
+ *  <tt>[first2, last2)</tt>. The return value is the end of the
+ *  output range.
+ *
+ *  In the simplest case, \p set_intersection performs the
+ *  "intersection" operation from set theory: the output range
+ *  contains a copy of every element that is contained in both
+ *  <tt>[first1, last1)</tt> and <tt>[first2, last2)</tt>. The
+ *  general case is more complicated, because the input ranges may
+ *  contain duplicate elements. The generalization is that if a value
+ *  appears \c m times in <tt>[first1, last1)</tt> and \c n times in
+ *  <tt>[first2, last2)</tt> (where \c m may be zero), then it
+ *  appears <tt>min(m,n)</tt> times in the output range.
+ *  \p set_intersection is stable, meaning that both elements are
+ *  copied from the first range rather than the second, and that the
+ *  relative order of elements in the output range is the same as in
+ *  the first input range.
+ *
+ *  This version of \p set_intersection compares objects using
+ *  \c operator<.
+ *
+ *  \param first1 The beginning of the first input range.
+ *  \param last1 The end of the first input range.
+ *  \param first2 The beginning of the second input range.
+ *  \param last2 The end of the second input range.
+ *  \param result The beginning of the output range.
+ *  \return The end of the output range.
+ *
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator1 and \p InputIterator2 have the same \c value_type,
+ *          \p InputIterator1's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a>,
+ *          the ordering on \p InputIterator1's \c value_type is a strict weak ordering, as defined in the <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a> requirements,
+ *          and \p InputIterator1's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator2 and \p InputIterator1 have the same \c value_type,
+ *          \p InputIterator2's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a>,
+ *          the ordering on \p InputIterator2's \c value_type is a strict weak ordering, as defined in the <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a> requirements,
+ *          and \p InputIterator2's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *
+ *  \pre The ranges <tt>[first1, last1)</tt> and <tt>[first2, last2)</tt> shall be sorted with respect to <tt>operator<</tt>.
+ *  \pre The resulting range shall not overlap with either input range.
+ *
+ *  The following code snippet demonstrates how to use \p set_intersection to compute the
+ *  set intersection of two sets of integers sorted in ascending order.
+ *
+ *  \code
+ *  #include <thrust/set_operations.h>
+ *  ...
+ *  int A1[6] = {1, 3, 5, 7, 9, 11};
+ *  int A2[7] = {1, 1, 2, 3, 5,  8, 13};
+ *
+ *  int result[7];
+ *
+ *  int *result_end = thrust::set_intersection(A1, A1 + 6, A2, A2 + 7, result);
+ *  // result is now {1, 3, 5}
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/set_intersection.html
+ *  \see \p includes
+ *  \see \p set_union
+ *  \see \p set_intersection
+ *  \see \p set_symmetric_difference
+ *  \see \p sort
+ *  \see \p is_sorted
+ */
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator>
+  OutputIterator set_intersection(InputIterator1 first1,
+                                  InputIterator1 last1,
+                                  InputIterator2 first2,
+                                  InputIterator2 last2,
+                                  OutputIterator result);
+
+
+/*! \p set_intersection constructs a sorted range that is the
+ *  intersection of sorted ranges <tt>[first1, last1)</tt> and
+ *  <tt>[first2, last2)</tt>. The return value is the end of the
+ *  output range.
+ *
+ *  In the simplest case, \p set_intersection performs the
+ *  "intersection" operation from set theory: the output range
+ *  contains a copy of every element that is contained in both
+ *  <tt>[first1, last1)</tt> and <tt>[first2, last2)</tt>. The
+ *  general case is more complicated, because the input ranges may
+ *  contain duplicate elements. The generalization is that if a value
+ *  appears \c m times in <tt>[first1, last1)</tt> and \c n times in
+ *  <tt>[first2, last2)</tt> (where \c m may be zero), then it
+ *  appears <tt>min(m,n)</tt> times in the output range.
+ *  \p set_intersection is stable, meaning that both elements are
+ *  copied from the first range rather than the second, and that the
+ *  relative order of elements in the output range is the same as in
+ *  the first input range.
+ *
+ *  This version of \p set_intersection compares elements using a function object \p comp.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first1 The beginning of the first input range.
+ *  \param last1 The end of the first input range.
+ *  \param first2 The beginning of the second input range.
+ *  \param last2 The end of the second input range.
+ *  \param result The beginning of the output range.
+ *  \param comp Comparison operator.
+ *  \return The end of the output range.
+ *
+ *  \pre The ranges <tt>[first1, last1)</tt> and <tt>[first2, last2)</tt> shall be sorted with respect to \p comp.
+ *  \pre The resulting range shall not overlap with either input range.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator1 and \p InputIterator2 have the same \c value_type,
+ *          \p InputIterator1's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a>,
+ *          the ordering on \p InputIterator1's \c value_type is a strict weak ordering, as defined in the <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a> requirements,
+ *          and \p InputIterator1's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator2 and \p InputIterator1 have the same \c value_type,
+ *          \p InputIterator2's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a>,
+ *          the ordering on \p InputIterator2's \c value_type is a strict weak ordering, as defined in the <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a> requirements,
+ *          and \p InputIterator2's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *
+ *  The following code snippet demonstrates how to use \p set_intersection to compute
+ *  the set intersection of sets of integers sorted in descending order using the \p thrust::host execution
+ *  policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/set_operations.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  int A1[6] = {11, 9, 7, 5, 3, 1};
+ *  int A2[7] = {13, 8, 5, 3, 2,  1, 1};
+ *
+ *  int result[3];
+ *
+ *  int *result_end = thrust::set_intersection(thrust::host, A1, A1 + 6, A2, A2 + 7, result, thrust::greater<int>());
+ *  // result is now {5, 3, 1}
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/set_intersection.html
+ *  \see \p includes
+ *  \see \p set_union
+ *  \see \p set_intersection
+ *  \see \p set_symmetric_difference
+ *  \see \p sort
+ *  \see \p is_sorted
+ */
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename StrictWeakCompare>
+  OutputIterator set_intersection(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                  InputIterator1                                              first1,
+                                  InputIterator1                                              last1,
+                                  InputIterator2                                              first2,
+                                  InputIterator2                                              last2,
+                                  OutputIterator                                              result,
+                                  StrictWeakCompare                                           comp);
+
+
+/*! \p set_intersection constructs a sorted range that is the
+ *  intersection of sorted ranges <tt>[first1, last1)</tt> and
+ *  <tt>[first2, last2)</tt>. The return value is the end of the
+ *  output range.
+ *
+ *  In the simplest case, \p set_intersection performs the
+ *  "intersection" operation from set theory: the output range
+ *  contains a copy of every element that is contained in both
+ *  <tt>[first1, last1)</tt> and <tt>[first2, last2)</tt>. The
+ *  general case is more complicated, because the input ranges may
+ *  contain duplicate elements. The generalization is that if a value
+ *  appears \c m times in <tt>[first1, last1)</tt> and \c n times in
+ *  <tt>[first2, last2)</tt> (where \c m may be zero), then it
+ *  appears <tt>min(m,n)</tt> times in the output range.
+ *  \p set_intersection is stable, meaning that both elements are
+ *  copied from the first range rather than the second, and that the
+ *  relative order of elements in the output range is the same as in
+ *  the first input range.
+ *
+ *  This version of \p set_intersection compares elements using a function object \p comp.
+ *
+ *  \param first1 The beginning of the first input range.
+ *  \param last1 The end of the first input range.
+ *  \param first2 The beginning of the second input range.
+ *  \param last2 The end of the second input range.
+ *  \param result The beginning of the output range.
+ *  \param comp Comparison operator.
+ *  \return The end of the output range.
+ *
+ *  \pre The ranges <tt>[first1, last1)</tt> and <tt>[first2, last2)</tt> shall be sorted with respect to \p comp.
+ *  \pre The resulting range shall not overlap with either input range.
+ *
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator1 and \p InputIterator2 have the same \c value_type,
+ *          \p InputIterator1's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a>,
+ *          the ordering on \p InputIterator1's \c value_type is a strict weak ordering, as defined in the <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a> requirements,
+ *          and \p InputIterator1's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator2 and \p InputIterator1 have the same \c value_type,
+ *          \p InputIterator2's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a>,
+ *          the ordering on \p InputIterator2's \c value_type is a strict weak ordering, as defined in the <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a> requirements,
+ *          and \p InputIterator2's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *
+ *  The following code snippet demonstrates how to use \p set_intersection to compute
+ *  the set intersection of sets of integers sorted in descending order.
+ *
+ *  \code
+ *  #include <thrust/set_operations.h>
+ *  ...
+ *  int A1[6] = {11, 9, 7, 5, 3, 1};
+ *  int A2[7] = {13, 8, 5, 3, 2,  1, 1};
+ *
+ *  int result[3];
+ *
+ *  int *result_end = thrust::set_intersection(A1, A1 + 6, A2, A2 + 7, result, thrust::greater<int>());
+ *  // result is now {5, 3, 1}
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/set_intersection.html
+ *  \see \p includes
+ *  \see \p set_union
+ *  \see \p set_intersection
+ *  \see \p set_symmetric_difference
+ *  \see \p sort
+ *  \see \p is_sorted
+ */
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename StrictWeakCompare>
+  OutputIterator set_intersection(InputIterator1 first1,
+                                  InputIterator1 last1,
+                                  InputIterator2 first2,
+                                  InputIterator2 last2,
+                                  OutputIterator result,
+                                  StrictWeakCompare comp);
+
+
+/*! \p set_symmetric_difference constructs a sorted range that is the set symmetric
+ *  difference of the sorted ranges <tt>[first1, last1)</tt> and <tt>[first2, last2)</tt>.
+ *  The return value is the end of the output range.
+ *
+ *  In the simplest case, \p set_symmetric_difference performs a set theoretic calculation:
+ *  it constructs the union of the two sets A - B and B - A, where A and B are the two
+ *  input ranges. That is, the output range contains a copy of every element that is
+ *  contained in <tt>[first1, last1)</tt> but not <tt>[first2, last1)</tt>, and a copy of
+ *  every element that is contained in <tt>[first2, last2)</tt> but not <tt>[first1, last1)</tt>.
+ *  The general case is more complicated, because the input ranges may contain duplicate elements.
+ *  The generalization is that if <tt>[first1, last1)</tt> contains \c m elements that are
+ *  equivalent to each other and <tt>[first2, last1)</tt> contains \c n elements that are
+ *  equivalent to them, then <tt>|m - n|</tt> of those elements shall be copied to the output
+ *  range: the last <tt>m - n</tt> elements from <tt>[first1, last1)</tt> if <tt>m > n</tt>, and
+ *  the last <tt>n - m</tt> of these elements from <tt>[first2, last2)</tt> if <tt>m < n</tt>.
+ *
+ *  This version of \p set_union compares elements using \c operator<.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first1 The beginning of the first input range.
+ *  \param last1 The end of the first input range.
+ *  \param first2 The beginning of the second input range.
+ *  \param last2 The end of the second input range.
+ *  \param result The beginning of the output range.
+ *  \return The end of the output range.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator1 and \p InputIterator2 have the same \c value_type,
+ *          \p InputIterator1's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a>,
+ *          the ordering on \p InputIterator1's \c value_type is a strict weak ordering, as defined in the <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a> requirements,
+ *          and \p InputIterator1's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator2 and \p InputIterator1 have the same \c value_type,
+ *          \p InputIterator2's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a>,
+ *          the ordering on \p InputIterator2's \c value_type is a strict weak ordering, as defined in the <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a> requirements,
+ *          and \p InputIterator2's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *
+ *  \pre The ranges <tt>[first1, last1)</tt> and <tt>[first2, last2)</tt> shall be sorted with respect to <tt>operator<</tt>.
+ *  \pre The resulting range shall not overlap with either input range.
+ *
+ *  The following code snippet demonstrates how to use \p set_symmetric_difference to compute
+ *  the symmetric difference of two sets of integers sorted in ascending order using the \p thrust::host
+ *  execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/set_operations.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  int A1[6] = {0, 1, 2, 2, 4, 6, 7};
+ *  int A2[5] = {1, 1, 2, 5, 8};
+ *
+ *  int result[6];
+ *
+ *  int *result_end = thrust::set_symmetric_difference(thrust::host, A1, A1 + 6, A2, A2 + 5, result);
+ *  // result = {0, 4, 5, 6, 7, 8}
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/set_symmetric_difference.html
+ *  \see \p merge
+ *  \see \p includes
+ *  \see \p set_difference
+ *  \see \p set_union
+ *  \see \p set_intersection
+ *  \see \p sort
+ *  \see \p is_sorted
+ */
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator>
+  OutputIterator set_symmetric_difference(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                          InputIterator1                                              first1,
+                                          InputIterator1                                              last1,
+                                          InputIterator2                                              first2,
+                                          InputIterator2                                              last2,
+                                          OutputIterator                                              result);
+
+
+/*! \p set_symmetric_difference constructs a sorted range that is the set symmetric
+ *  difference of the sorted ranges <tt>[first1, last1)</tt> and <tt>[first2, last2)</tt>.
+ *  The return value is the end of the output range.
+ *
+ *  In the simplest case, \p set_symmetric_difference performs a set theoretic calculation:
+ *  it constructs the union of the two sets A - B and B - A, where A and B are the two
+ *  input ranges. That is, the output range contains a copy of every element that is
+ *  contained in <tt>[first1, last1)</tt> but not <tt>[first2, last1)</tt>, and a copy of
+ *  every element that is contained in <tt>[first2, last2)</tt> but not <tt>[first1, last1)</tt>.
+ *  The general case is more complicated, because the input ranges may contain duplicate elements.
+ *  The generalization is that if <tt>[first1, last1)</tt> contains \c m elements that are
+ *  equivalent to each other and <tt>[first2, last1)</tt> contains \c n elements that are
+ *  equivalent to them, then <tt>|m - n|</tt> of those elements shall be copied to the output
+ *  range: the last <tt>m - n</tt> elements from <tt>[first1, last1)</tt> if <tt>m > n</tt>, and
+ *  the last <tt>n - m</tt> of these elements from <tt>[first2, last2)</tt> if <tt>m < n</tt>.
+ *
+ *  This version of \p set_union compares elements using \c operator<.
+ *
+ *  \param first1 The beginning of the first input range.
+ *  \param last1 The end of the first input range.
+ *  \param first2 The beginning of the second input range.
+ *  \param last2 The end of the second input range.
+ *  \param result The beginning of the output range.
+ *  \return The end of the output range.
+ *
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator1 and \p InputIterator2 have the same \c value_type,
+ *          \p InputIterator1's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a>,
+ *          the ordering on \p InputIterator1's \c value_type is a strict weak ordering, as defined in the <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a> requirements,
+ *          and \p InputIterator1's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator2 and \p InputIterator1 have the same \c value_type,
+ *          \p InputIterator2's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a>,
+ *          the ordering on \p InputIterator2's \c value_type is a strict weak ordering, as defined in the <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a> requirements,
+ *          and \p InputIterator2's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *
+ *  \pre The ranges <tt>[first1, last1)</tt> and <tt>[first2, last2)</tt> shall be sorted with respect to <tt>operator<</tt>.
+ *  \pre The resulting range shall not overlap with either input range.
+ *
+ *  The following code snippet demonstrates how to use \p set_symmetric_difference to compute
+ *  the symmetric difference of two sets of integers sorted in ascending order.
+ *
+ *  \code
+ *  #include <thrust/set_operations.h>
+ *  ...
+ *  int A1[6] = {0, 1, 2, 2, 4, 6, 7};
+ *  int A2[5] = {1, 1, 2, 5, 8};
+ *
+ *  int result[6];
+ *
+ *  int *result_end = thrust::set_symmetric_difference(A1, A1 + 6, A2, A2 + 5, result);
+ *  // result = {0, 4, 5, 6, 7, 8}
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/set_symmetric_difference.html
+ *  \see \p merge
+ *  \see \p includes
+ *  \see \p set_difference
+ *  \see \p set_union
+ *  \see \p set_intersection
+ *  \see \p sort
+ *  \see \p is_sorted
+ */
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator>
+  OutputIterator set_symmetric_difference(InputIterator1 first1,
+                                          InputIterator1 last1,
+                                          InputIterator2 first2,
+                                          InputIterator2 last2,
+                                          OutputIterator result);
+
+
+/*! \p set_symmetric_difference constructs a sorted range that is the set symmetric
+ *  difference of the sorted ranges <tt>[first1, last1)</tt> and <tt>[first2, last2)</tt>.
+ *  The return value is the end of the output range.
+ *
+ *  In the simplest case, \p set_symmetric_difference performs a set theoretic calculation:
+ *  it constructs the union of the two sets A - B and B - A, where A and B are the two
+ *  input ranges. That is, the output range contains a copy of every element that is
+ *  contained in <tt>[first1, last1)</tt> but not <tt>[first2, last1)</tt>, and a copy of
+ *  every element that is contained in <tt>[first2, last2)</tt> but not <tt>[first1, last1)</tt>.
+ *  The general case is more complicated, because the input ranges may contain duplicate elements.
+ *  The generalization is that if <tt>[first1, last1)</tt> contains \c m elements that are
+ *  equivalent to each other and <tt>[first2, last1)</tt> contains \c n elements that are
+ *  equivalent to them, then <tt>|m - n|</tt> of those elements shall be copied to the output
+ *  range: the last <tt>m - n</tt> elements from <tt>[first1, last1)</tt> if <tt>m > n</tt>, and
+ *  the last <tt>n - m</tt> of these elements from <tt>[first2, last2)</tt> if <tt>m < n</tt>.
+ *
+ *  This version of \p set_union compares elements using a function object \p comp.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first1 The beginning of the first input range.
+ *  \param last1 The end of the first input range.
+ *  \param first2 The beginning of the second input range.
+ *  \param last2 The end of the second input range.
+ *  \param result The beginning of the output range.
+ *  \param comp Comparison operator.
+ *  \return The end of the output range.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator1 and \p InputIterator2 have the same \c value_type,
+ *          \p InputIterator1's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a>,
+ *          the ordering on \p InputIterator1's \c value_type is a strict weak ordering, as defined in the <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a> requirements,
+ *          and \p InputIterator1's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator2 and \p InputIterator1 have the same \c value_type,
+ *          \p InputIterator2's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a>,
+ *          the ordering on \p InputIterator2's \c value_type is a strict weak ordering, as defined in the <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a> requirements,
+ *          and \p InputIterator2's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *
+ *  \pre The ranges <tt>[first1, last1)</tt> and <tt>[first2, last2)</tt> shall be sorted with respect to \p comp.
+ *  \pre The resulting range shall not overlap with either input range.
+ *
+ *  The following code snippet demonstrates how to use \p set_symmetric_difference to compute
+ *  the symmetric difference of two sets of integers sorted in descending order using the \p thrust::host
+ *  execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/set_operations.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  int A1[6] = {7, 6, 4, 2, 2, 1, 0};
+ *  int A2[5] = {8, 5, 2, 1, 1};
+ *
+ *  int result[6];
+ *
+ *  int *result_end = thrust::set_symmetric_difference(thrust::host, A1, A1 + 6, A2, A2 + 5, result);
+ *  // result = {8, 7, 6, 5, 4, 0}
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/set_symmetric_difference.html
+ *  \see \p merge
+ *  \see \p includes
+ *  \see \p set_difference
+ *  \see \p set_union
+ *  \see \p set_intersection
+ *  \see \p sort
+ *  \see \p is_sorted
+ */
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename StrictWeakCompare>
+  OutputIterator set_symmetric_difference(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                          InputIterator1                                              first1,
+                                          InputIterator1                                              last1,
+                                          InputIterator2                                              first2,
+                                          InputIterator2                                              last2,
+                                          OutputIterator                                              result,
+                                          StrictWeakCompare                                           comp);
+
+
+/*! \p set_symmetric_difference constructs a sorted range that is the set symmetric
+ *  difference of the sorted ranges <tt>[first1, last1)</tt> and <tt>[first2, last2)</tt>.
+ *  The return value is the end of the output range.
+ *
+ *  In the simplest case, \p set_symmetric_difference performs a set theoretic calculation:
+ *  it constructs the union of the two sets A - B and B - A, where A and B are the two
+ *  input ranges. That is, the output range contains a copy of every element that is
+ *  contained in <tt>[first1, last1)</tt> but not <tt>[first2, last1)</tt>, and a copy of
+ *  every element that is contained in <tt>[first2, last2)</tt> but not <tt>[first1, last1)</tt>.
+ *  The general case is more complicated, because the input ranges may contain duplicate elements.
+ *  The generalization is that if <tt>[first1, last1)</tt> contains \c m elements that are
+ *  equivalent to each other and <tt>[first2, last1)</tt> contains \c n elements that are
+ *  equivalent to them, then <tt>|m - n|</tt> of those elements shall be copied to the output
+ *  range: the last <tt>m - n</tt> elements from <tt>[first1, last1)</tt> if <tt>m > n</tt>, and
+ *  the last <tt>n - m</tt> of these elements from <tt>[first2, last2)</tt> if <tt>m < n</tt>.
+ *
+ *  This version of \p set_union compares elements using a function object \p comp.
+ *
+ *  \param first1 The beginning of the first input range.
+ *  \param last1 The end of the first input range.
+ *  \param first2 The beginning of the second input range.
+ *  \param last2 The end of the second input range.
+ *  \param result The beginning of the output range.
+ *  \param comp Comparison operator.
+ *  \return The end of the output range.
+ *
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator1 and \p InputIterator2 have the same \c value_type,
+ *          \p InputIterator1's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a>,
+ *          the ordering on \p InputIterator1's \c value_type is a strict weak ordering, as defined in the <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a> requirements,
+ *          and \p InputIterator1's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator2 and \p InputIterator1 have the same \c value_type,
+ *          \p InputIterator2's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a>,
+ *          the ordering on \p InputIterator2's \c value_type is a strict weak ordering, as defined in the <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a> requirements,
+ *          and \p InputIterator2's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *
+ *  \pre The ranges <tt>[first1, last1)</tt> and <tt>[first2, last2)</tt> shall be sorted with respect to \p comp.
+ *  \pre The resulting range shall not overlap with either input range.
+ *
+ *  The following code snippet demonstrates how to use \p set_symmetric_difference to compute
+ *  the symmetric difference of two sets of integers sorted in descending order.
+ *
+ *  \code
+ *  #include <thrust/set_operations.h>
+ *  ...
+ *  int A1[6] = {7, 6, 4, 2, 2, 1, 0};
+ *  int A2[5] = {8, 5, 2, 1, 1};
+ *
+ *  int result[6];
+ *
+ *  int *result_end = thrust::set_symmetric_difference(A1, A1 + 6, A2, A2 + 5, result);
+ *  // result = {8, 7, 6, 5, 4, 0}
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/set_symmetric_difference.html
+ *  \see \p merge
+ *  \see \p includes
+ *  \see \p set_difference
+ *  \see \p set_union
+ *  \see \p set_intersection
+ *  \see \p sort
+ *  \see \p is_sorted
+ */
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename StrictWeakCompare>
+  OutputIterator set_symmetric_difference(InputIterator1 first1,
+                                          InputIterator1 last1,
+                                          InputIterator2 first2,
+                                          InputIterator2 last2,
+                                          OutputIterator result,
+                                          StrictWeakCompare comp);
+
+
+/*! \p set_union constructs a sorted range that is the union of the sorted ranges
+ *  <tt>[first1, last1)</tt> and <tt>[first2, last2)</tt>. The return value is the
+ *  end of the output range.
+ *
+ *  In the simplest case, \p set_union performs the "union" operation from set
+ *  theory: the output range contains a copy of every element that is contained in
+ *  <tt>[first1, last1)</tt>, <tt>[first2, last1)</tt>, or both. The general case
+ *  is more complicated, because the input ranges may contain duplicate elements.
+ *  The generalization is that if <tt>[first1, last1)</tt> contains \c m elements
+ *  that are equivalent to each other and if <tt>[first2, last2)</tt> contains \c n
+ *  elements that are equivalent to them, then all \c m elements from the first
+ *  range shall be copied to the output range, in order, and then <tt>max(n - m, 0)</tt>
+ *  elements from the second range shall be copied to the output, in order.
+ *
+ *  This version of \p set_union compares elements using \c operator<.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first1 The beginning of the first input range.
+ *  \param last1 The end of the first input range.
+ *  \param first2 The beginning of the second input range.
+ *  \param last2 The end of the second input range.
+ *  \param result The beginning of the output range.
+ *  \return The end of the output range.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator1 and \p InputIterator2 have the same \c value_type,
+ *          \p InputIterator1's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a>,
+ *          the ordering on \p InputIterator1's \c value_type is a strict weak ordering, as defined in the <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a> requirements,
+ *          and \p InputIterator1's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator2 and \p InputIterator1 have the same \c value_type,
+ *          \p InputIterator2's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a>,
+ *          the ordering on \p InputIterator2's \c value_type is a strict weak ordering, as defined in the <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a> requirements,
+ *          and \p InputIterator2's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *
+ *  \pre The ranges <tt>[first1, last1)</tt> and <tt>[first2, last2)</tt> shall be sorted with respect to <tt>operator<</tt>.
+ *  \pre The resulting range shall not overlap with either input range.
+ *
+ *  The following code snippet demonstrates how to use \p set_union to compute the union of
+ *  two sets of integers sorted in ascending order using the \p thrust::host execution policy for
+ *  parallelization:
+ *
+ *  \code
+ *  #include <thrust/set_operations.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  int A1[6] = {0, 2, 4, 6, 8, 10, 12};
+ *  int A2[5] = {1, 3, 5, 7, 9};
+ *
+ *  int result[11];
+ *
+ *  int *result_end = thrust::set_union(thrust::host, A1, A1 + 6, A2, A2 + 5, result);
+ *  // result = {0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 12}
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/set_union.html
+ *  \see \p merge
+ *  \see \p includes
+ *  \see \p set_union
+ *  \see \p set_intersection
+ *  \see \p set_symmetric_difference
+ *  \see \p sort
+ *  \see \p is_sorted
+ */
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator>
+  OutputIterator set_union(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                           InputIterator1                                              first1,
+                           InputIterator1                                              last1,
+                           InputIterator2                                              first2,
+                           InputIterator2                                              last2,
+                           OutputIterator                                              result);
+
+
+/*! \p set_union constructs a sorted range that is the union of the sorted ranges
+ *  <tt>[first1, last1)</tt> and <tt>[first2, last2)</tt>. The return value is the
+ *  end of the output range.
+ *
+ *  In the simplest case, \p set_union performs the "union" operation from set
+ *  theory: the output range contains a copy of every element that is contained in
+ *  <tt>[first1, last1)</tt>, <tt>[first2, last1)</tt>, or both. The general case
+ *  is more complicated, because the input ranges may contain duplicate elements.
+ *  The generalization is that if <tt>[first1, last1)</tt> contains \c m elements
+ *  that are equivalent to each other and if <tt>[first2, last2)</tt> contains \c n
+ *  elements that are equivalent to them, then all \c m elements from the first
+ *  range shall be copied to the output range, in order, and then <tt>max(n - m, 0)</tt>
+ *  elements from the second range shall be copied to the output, in order.
+ *
+ *  This version of \p set_union compares elements using \c operator<.
+ *
+ *  \param first1 The beginning of the first input range.
+ *  \param last1 The end of the first input range.
+ *  \param first2 The beginning of the second input range.
+ *  \param last2 The end of the second input range.
+ *  \param result The beginning of the output range.
+ *  \return The end of the output range.
+ *
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator1 and \p InputIterator2 have the same \c value_type,
+ *          \p InputIterator1's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a>,
+ *          the ordering on \p InputIterator1's \c value_type is a strict weak ordering, as defined in the <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a> requirements,
+ *          and \p InputIterator1's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator2 and \p InputIterator1 have the same \c value_type,
+ *          \p InputIterator2's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a>,
+ *          the ordering on \p InputIterator2's \c value_type is a strict weak ordering, as defined in the <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a> requirements,
+ *          and \p InputIterator2's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *
+ *  \pre The ranges <tt>[first1, last1)</tt> and <tt>[first2, last2)</tt> shall be sorted with respect to <tt>operator<</tt>.
+ *  \pre The resulting range shall not overlap with either input range.
+ *
+ *  The following code snippet demonstrates how to use \p set_union to compute the union of
+ *  two sets of integers sorted in ascending order.
+ *
+ *  \code
+ *  #include <thrust/set_operations.h>
+ *  ...
+ *  int A1[6] = {0, 2, 4, 6, 8, 10, 12};
+ *  int A2[5] = {1, 3, 5, 7, 9};
+ *
+ *  int result[11];
+ *
+ *  int *result_end = thrust::set_union(A1, A1 + 6, A2, A2 + 5, result);
+ *  // result = {0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 12}
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/set_union.html
+ *  \see \p merge
+ *  \see \p includes
+ *  \see \p set_union
+ *  \see \p set_intersection
+ *  \see \p set_symmetric_difference
+ *  \see \p sort
+ *  \see \p is_sorted
+ */
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator>
+  OutputIterator set_union(InputIterator1 first1,
+                           InputIterator1 last1,
+                           InputIterator2 first2,
+                           InputIterator2 last2,
+                           OutputIterator result);
+
+
+/*! \p set_union constructs a sorted range that is the union of the sorted ranges
+ *  <tt>[first1, last1)</tt> and <tt>[first2, last2)</tt>. The return value is the
+ *  end of the output range.
+ *
+ *  In the simplest case, \p set_union performs the "union" operation from set
+ *  theory: the output range contains a copy of every element that is contained in
+ *  <tt>[first1, last1)</tt>, <tt>[first2, last1)</tt>, or both. The general case
+ *  is more complicated, because the input ranges may contain duplicate elements.
+ *  The generalization is that if <tt>[first1, last1)</tt> contains \c m elements
+ *  that are equivalent to each other and if <tt>[first2, last2)</tt> contains \c n
+ *  elements that are equivalent to them, then all \c m elements from the first
+ *  range shall be copied to the output range, in order, and then <tt>max(n - m, 0)</tt>
+ *  elements from the second range shall be copied to the output, in order.
+ *
+ *  This version of \p set_union compares elements using a function object \p comp.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first1 The beginning of the first input range.
+ *  \param last1 The end of the first input range.
+ *  \param first2 The beginning of the second input range.
+ *  \param last2 The end of the second input range.
+ *  \param result The beginning of the output range.
+ *  \param comp Comparison operator.
+ *  \return The end of the output range.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator1's \c value_type is convertable to \p StrictWeakCompare's \c first_argument_type.
+ *          and \p InputIterator1's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator2's \c value_type is convertable to \p StrictWeakCompare's \c second_argument_type.
+ *          and \p InputIterator2's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam StrictWeakCompare is a model of <a href="http://www.sgi.com/tech/stl/StrictWeakOrdering.html">Strict Weak Ordering</a>.
+ *
+ *  \pre The ranges <tt>[first1, last1)</tt> and <tt>[first2, last2)</tt> shall be sorted with respect to \p comp.
+ *  \pre The resulting range shall not overlap with either input range.
+ *
+ *  The following code snippet demonstrates how to use \p set_union to compute the union of
+ *  two sets of integers sorted in ascending order using the \p thrust::host execution policy for
+ *  parallelization:
+ *
+ *  \code
+ *  #include <thrust/set_operations.h>
+ *  #include <thrust/functional.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  int A1[6] = {12, 10, 8, 6, 4, 2, 0};
+ *  int A2[5] = {9, 7, 5, 3, 1};
+ *
+ *  int result[11];
+ *
+ *  int *result_end = thrust::set_union(thrust::host, A1, A1 + 6, A2, A2 + 5, result, thrust::greater<int>());
+ *  // result = {12, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0}
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/set_union.html
+ *  \see \p merge
+ *  \see \p includes
+ *  \see \p set_union
+ *  \see \p set_intersection
+ *  \see \p set_symmetric_difference
+ *  \see \p sort
+ *  \see \p is_sorted
+ */
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename StrictWeakCompare>
+  OutputIterator set_union(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                           InputIterator1                                              first1,
+                           InputIterator1                                              last1,
+                           InputIterator2                                              first2,
+                           InputIterator2                                              last2,
+                           OutputIterator                                              result,
+                           StrictWeakCompare                                           comp);
+
+
+/*! \p set_union constructs a sorted range that is the union of the sorted ranges
+ *  <tt>[first1, last1)</tt> and <tt>[first2, last2)</tt>. The return value is the
+ *  end of the output range.
+ *
+ *  In the simplest case, \p set_union performs the "union" operation from set
+ *  theory: the output range contains a copy of every element that is contained in
+ *  <tt>[first1, last1)</tt>, <tt>[first2, last1)</tt>, or both. The general case
+ *  is more complicated, because the input ranges may contain duplicate elements.
+ *  The generalization is that if <tt>[first1, last1)</tt> contains \c m elements
+ *  that are equivalent to each other and if <tt>[first2, last2)</tt> contains \c n
+ *  elements that are equivalent to them, then all \c m elements from the first
+ *  range shall be copied to the output range, in order, and then <tt>max(n - m, 0)</tt>
+ *  elements from the second range shall be copied to the output, in order.
+ *
+ *  This version of \p set_union compares elements using a function object \p comp.
+ *
+ *  \param first1 The beginning of the first input range.
+ *  \param last1 The end of the first input range.
+ *  \param first2 The beginning of the second input range.
+ *  \param last2 The end of the second input range.
+ *  \param result The beginning of the output range.
+ *  \param comp Comparison operator.
+ *  \return The end of the output range.
+ *
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator1's \c value_type is convertable to \p StrictWeakCompare's \c first_argument_type.
+ *          and \p InputIterator1's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator2's \c value_type is convertable to \p StrictWeakCompare's \c second_argument_type.
+ *          and \p InputIterator2's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam StrictWeakCompare is a model of <a href="http://www.sgi.com/tech/stl/StrictWeakOrdering.html">Strict Weak Ordering</a>.
+ *
+ *  \pre The ranges <tt>[first1, last1)</tt> and <tt>[first2, last2)</tt> shall be sorted with respect to \p comp.
+ *  \pre The resulting range shall not overlap with either input range.
+ *
+ *  The following code snippet demonstrates how to use \p set_union to compute the union of
+ *  two sets of integers sorted in ascending order.
+ *
+ *  \code
+ *  #include <thrust/set_operations.h>
+ *  #include <thrust/functional.h>
+ *  ...
+ *  int A1[6] = {12, 10, 8, 6, 4, 2, 0};
+ *  int A2[5] = {9, 7, 5, 3, 1};
+ *
+ *  int result[11];
+ *
+ *  int *result_end = thrust::set_union(A1, A1 + 6, A2, A2 + 5, result, thrust::greater<int>());
+ *  // result = {12, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0}
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/set_union.html
+ *  \see \p merge
+ *  \see \p includes
+ *  \see \p set_union
+ *  \see \p set_intersection
+ *  \see \p set_symmetric_difference
+ *  \see \p sort
+ *  \see \p is_sorted
+ */
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename StrictWeakCompare>
+  OutputIterator set_union(InputIterator1 first1,
+                           InputIterator1 last1,
+                           InputIterator2 first2,
+                           InputIterator2 last2,
+                           OutputIterator result,
+                           StrictWeakCompare comp);
+
+
+/*! \p set_difference_by_key performs a key-value difference operation from set theory.
+ *  \p set_difference_by_key constructs a sorted range that is the difference of the sorted
+ *  ranges <tt>[keys_first1, keys_last1)</tt> and <tt>[keys_first2, keys_last2)</tt>. Associated
+ *  with each element from the input and output key ranges is a value element. The associated input
+ *  value ranges need not be sorted.
+ *
+ *  In the simplest case, \p set_difference_by_key performs the "difference" operation from set
+ *  theory: the keys output range contains a copy of every element that is contained in
+ *  <tt>[keys_first1, keys_last1)</tt> and not contained in <tt>[keys_first2, keys_last2)</tt>.
+ *  The general case is more complicated, because the input ranges may contain duplicate elements.
+ *  The generalization is that if <tt>[keys_first1, keys_last1)</tt> contains \c m elements
+ *  that are equivalent to each other and if <tt>[keys_first2, keys_last2)</tt> contains \c n
+ *  elements that are equivalent to them, the last <tt>max(m-n,0)</tt> elements from
+ *  <tt>[keys_first1, keys_last1)</tt> range shall be copied to the output range.
+ *
+ *  Each time a key element is copied from <tt>[keys_first1, keys_last1)</tt> or
+ *  <tt>[keys_first2, keys_last2)</tt> is copied to the keys output range, the
+ *  corresponding value element is copied from the corresponding values input range (beginning at
+ *  \p values_first1 or \p values_first2) to the values output range.
+ *
+ *  This version of \p set_difference_by_key compares key elements using \c operator<.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param keys_first1 The beginning of the first input range of keys.
+ *  \param keys_last1 The end of the first input range of keys.
+ *  \param keys_first2 The beginning of the second input range of keys.
+ *  \param keys_last2 The end of the second input range of keys.
+ *  \param values_first1 The beginning of the first input range of values.
+ *  \param values_first2 The beginning of the first input range of values.
+ *  \param keys_result The beginning of the output range of keys.
+ *  \param values_result The beginning of the output range of values.
+ *  \return A \p pair \c p such that <tt>p.first</tt> is the end of the output range of keys,
+ *          and such that <tt>p.second</tt> is the end of the output range of values.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator1 and \p InputIterator2 have the same \c value_type,
+ *          \p InputIterator1's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a>,
+ *          the ordering on \p InputIterator1's \c value_type is a strict weak ordering, as defined in the <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a> requirements,
+ *          and \p InputIterator1's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator2 and \p InputIterator1 have the same \c value_type,
+ *          \p InputIterator2's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a>,
+ *          the ordering on \p InputIterator2's \c value_type is a strict weak ordering, as defined in the <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a> requirements,
+ *          and \p InputIterator2's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam InputIterator3 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator3's \c value_type is convertible to a type in \p OutputIterator2's set of \c value_types.
+ *  \tparam InputIterator4 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator4's \c value_type is convertible to a type in \p OutputIterator2's set of \c value_types.
+ *  \tparam OutputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam OutputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *
+ *  \pre The ranges <tt>[keys_first1, keys_last1)</tt> and <tt>[keys_first2, keys_last2)</tt> shall be sorted with respect to <tt>operator<</tt>.
+ *  \pre The resulting ranges shall not overlap with any input range.
+ *
+ *  The following code snippet demonstrates how to use \p set_difference_by_key to compute the
+ *  set difference of two sets of integers sorted in ascending order with their values using the \p thrust::host
+ *  execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/set_operations.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  int A_keys[6] = {0, 1, 3, 4, 5, 6, 9};
+ *  int A_vals[6] = {0, 0, 0, 0, 0, 0, 0};
+ *
+ *  int B_keys[5] = {1, 3, 5, 7, 9};
+ *  int B_vals[5] = {1, 1, 1, 1, 1};
+ *
+ *  int keys_result[3];
+ *  int vals_result[3];
+ *
+ *  thrust::pair<int*,int*> end = thrust::set_difference_by_key(thrust::host, A_keys, A_keys + 6, B_keys, B_keys + 5, A_vals, B_vals, keys_result, vals_result);
+ *  // keys_result is now {0, 4, 6}
+ *  // vals_result is now {0, 0, 0}
+ *  \endcode
+ *
+ *  \see \p set_union_by_key
+ *  \see \p set_intersection_by_key
+ *  \see \p set_symmetric_difference_by_key
+ *  \see \p sort_by_key
+ *  \see \p is_sorted
+ */
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename InputIterator4,
+         typename OutputIterator1,
+         typename OutputIterator2>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    set_difference_by_key(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                          InputIterator1                                              keys_first1,
+                          InputIterator1                                              keys_last1,
+                          InputIterator2                                              keys_first2,
+                          InputIterator2                                              keys_last2,
+                          InputIterator3                                              values_first1,
+                          InputIterator4                                              values_first2,
+                          OutputIterator1                                             keys_result,
+                          OutputIterator2                                             values_result);
+
+
+/*! \p set_difference_by_key performs a key-value difference operation from set theory.
+ *  \p set_difference_by_key constructs a sorted range that is the difference of the sorted
+ *  ranges <tt>[keys_first1, keys_last1)</tt> and <tt>[keys_first2, keys_last2)</tt>. Associated
+ *  with each element from the input and output key ranges is a value element. The associated input
+ *  value ranges need not be sorted.
+ *
+ *  In the simplest case, \p set_difference_by_key performs the "difference" operation from set
+ *  theory: the keys output range contains a copy of every element that is contained in
+ *  <tt>[keys_first1, keys_last1)</tt> and not contained in <tt>[keys_first2, keys_last2)</tt>.
+ *  The general case is more complicated, because the input ranges may contain duplicate elements.
+ *  The generalization is that if <tt>[keys_first1, keys_last1)</tt> contains \c m elements
+ *  that are equivalent to each other and if <tt>[keys_first2, keys_last2)</tt> contains \c n
+ *  elements that are equivalent to them, the last <tt>max(m-n,0)</tt> elements from
+ *  <tt>[keys_first1, keys_last1)</tt> range shall be copied to the output range.
+ *
+ *  Each time a key element is copied from <tt>[keys_first1, keys_last1)</tt> or
+ *  <tt>[keys_first2, keys_last2)</tt> is copied to the keys output range, the
+ *  corresponding value element is copied from the corresponding values input range (beginning at
+ *  \p values_first1 or \p values_first2) to the values output range.
+ *
+ *  This version of \p set_difference_by_key compares key elements using \c operator<.
+ *
+ *  \param keys_first1 The beginning of the first input range of keys.
+ *  \param keys_last1 The end of the first input range of keys.
+ *  \param keys_first2 The beginning of the second input range of keys.
+ *  \param keys_last2 The end of the second input range of keys.
+ *  \param values_first1 The beginning of the first input range of values.
+ *  \param values_first2 The beginning of the first input range of values.
+ *  \param keys_result The beginning of the output range of keys.
+ *  \param values_result The beginning of the output range of values.
+ *  \return A \p pair \c p such that <tt>p.first</tt> is the end of the output range of keys,
+ *          and such that <tt>p.second</tt> is the end of the output range of values.
+ *
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator1 and \p InputIterator2 have the same \c value_type,
+ *          \p InputIterator1's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a>,
+ *          the ordering on \p InputIterator1's \c value_type is a strict weak ordering, as defined in the <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a> requirements,
+ *          and \p InputIterator1's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator2 and \p InputIterator1 have the same \c value_type,
+ *          \p InputIterator2's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a>,
+ *          the ordering on \p InputIterator2's \c value_type is a strict weak ordering, as defined in the <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a> requirements,
+ *          and \p InputIterator2's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam InputIterator3 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator3's \c value_type is convertible to a type in \p OutputIterator2's set of \c value_types.
+ *  \tparam InputIterator4 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator4's \c value_type is convertible to a type in \p OutputIterator2's set of \c value_types.
+ *  \tparam OutputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam OutputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *
+ *  \pre The ranges <tt>[keys_first1, keys_last1)</tt> and <tt>[keys_first2, keys_last2)</tt> shall be sorted with respect to <tt>operator<</tt>.
+ *  \pre The resulting ranges shall not overlap with any input range.
+ *
+ *  The following code snippet demonstrates how to use \p set_difference_by_key to compute the
+ *  set difference of two sets of integers sorted in ascending order with their values.
+ *
+ *  \code
+ *  #include <thrust/set_operations.h>
+ *  ...
+ *  int A_keys[6] = {0, 1, 3, 4, 5, 6, 9};
+ *  int A_vals[6] = {0, 0, 0, 0, 0, 0, 0};
+ *
+ *  int B_keys[5] = {1, 3, 5, 7, 9};
+ *  int B_vals[5] = {1, 1, 1, 1, 1};
+ *
+ *  int keys_result[3];
+ *  int vals_result[3];
+ *
+ *  thrust::pair<int*,int*> end = thrust::set_difference_by_key(A_keys, A_keys + 6, B_keys, B_keys + 5, A_vals, B_vals, keys_result, vals_result);
+ *  // keys_result is now {0, 4, 6}
+ *  // vals_result is now {0, 0, 0}
+ *  \endcode
+ *
+ *  \see \p set_union_by_key
+ *  \see \p set_intersection_by_key
+ *  \see \p set_symmetric_difference_by_key
+ *  \see \p sort_by_key
+ *  \see \p is_sorted
+ */
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename InputIterator4,
+         typename OutputIterator1,
+         typename OutputIterator2>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    set_difference_by_key(InputIterator1                             keys_first1,
+                          InputIterator1                             keys_last1,
+                          InputIterator2                             keys_first2,
+                          InputIterator2                             keys_last2,
+                          InputIterator3                             values_first1,
+                          InputIterator4                             values_first2,
+                          OutputIterator1                            keys_result,
+                          OutputIterator2                            values_result);
+
+
+/*! \p set_difference_by_key performs a key-value difference operation from set theory.
+ *  \p set_difference_by_key constructs a sorted range that is the difference of the sorted
+ *  ranges <tt>[keys_first1, keys_last1)</tt> and <tt>[keys_first2, keys_last2)</tt>. Associated
+ *  with each element from the input and output key ranges is a value element. The associated input
+ *  value ranges need not be sorted.
+ *
+ *  In the simplest case, \p set_difference_by_key performs the "difference" operation from set
+ *  theory: the keys output range contains a copy of every element that is contained in
+ *  <tt>[keys_first1, keys_last1)</tt> and not contained in <tt>[keys_first2, keys_last2)</tt>.
+ *  The general case is more complicated, because the input ranges may contain duplicate elements.
+ *  The generalization is that if <tt>[keys_first1, keys_last1)</tt> contains \c m elements
+ *  that are equivalent to each other and if <tt>[keys_first2, keys_last2)</tt> contains \c n
+ *  elements that are equivalent to them, the last <tt>max(m-n,0)</tt> elements from
+ *  <tt>[keys_first1, keys_last1)</tt> range shall be copied to the output range.
+ *
+ *  Each time a key element is copied from <tt>[keys_first1, keys_last1)</tt> or
+ *  <tt>[keys_first2, keys_last2)</tt> is copied to the keys output range, the
+ *  corresponding value element is copied from the corresponding values input range (beginning at
+ *  \p values_first1 or \p values_first2) to the values output range.
+ *
+ *  This version of \p set_difference_by_key compares key elements using a function object \p comp.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param keys_first1 The beginning of the first input range of keys.
+ *  \param keys_last1 The end of the first input range of keys.
+ *  \param keys_first2 The beginning of the second input range of keys.
+ *  \param keys_last2 The end of the second input range of keys.
+ *  \param values_first1 The beginning of the first input range of values.
+ *  \param values_first2 The beginning of the first input range of values.
+ *  \param keys_result The beginning of the output range of keys.
+ *  \param values_result The beginning of the output range of values.
+ *  \param comp Comparison operator.
+ *  \return A \p pair \c p such that <tt>p.first</tt> is the end of the output range of keys,
+ *          and such that <tt>p.second</tt> is the end of the output range of values.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator1 and \p InputIterator2 have the same \c value_type,
+ *          \p InputIterator1's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a>,
+ *          the ordering on \p InputIterator1's \c value_type is a strict weak ordering, as defined in the <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a> requirements,
+ *          and \p InputIterator1's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator2 and \p InputIterator1 have the same \c value_type,
+ *          \p InputIterator2's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a>,
+ *          the ordering on \p InputIterator2's \c value_type is a strict weak ordering, as defined in the <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a> requirements,
+ *          and \p InputIterator2's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam InputIterator3 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator3's \c value_type is convertible to a type in \p OutputIterator2's set of \c value_types.
+ *  \tparam InputIterator4 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator4's \c value_type is convertible to a type in \p OutputIterator2's set of \c value_types.
+ *  \tparam OutputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam OutputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam StrictWeakCompare is a model of <a href="http://www.sgi.com/tech/stl/StrictWeakOrdering.html">Strict Weak Ordering</a>.
+ *
+ *  \pre The ranges <tt>[keys_first1, keys_last1)</tt> and <tt>[keys_first2, keys_last2)</tt> shall be sorted with respect to \p comp.
+ *  \pre The resulting ranges shall not overlap with any input range.
+ *
+ *  The following code snippet demonstrates how to use \p set_difference_by_key to compute the
+ *  set difference of two sets of integers sorted in descending order with their values using the \p thrust::host
+ *  execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/set_operations.h>
+ *  #include <thrust/functional.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  int A_keys[6] = {9, 6, 5, 4, 3, 1, 0};
+ *  int A_vals[6] = {0, 0, 0, 0, 0, 0, 0};
+ *
+ *  int B_keys[5] = {9, 7, 5, 3, 1};
+ *  int B_vals[5] = {1, 1, 1, 1, 1};
+ *
+ *  int keys_result[3];
+ *  int vals_result[3];
+ *
+ *  thrust::pair<int*,int*> end = thrust::set_difference_by_key(thrust::host, A_keys, A_keys + 6, B_keys, B_keys + 5, A_vals, B_vals, keys_result, vals_result, thrust::greater<int>());
+ *  // keys_result is now {0, 4, 6}
+ *  // vals_result is now {0, 0, 0}
+ *  \endcode
+ *
+ *  \see \p set_union_by_key
+ *  \see \p set_intersection_by_key
+ *  \see \p set_symmetric_difference_by_key
+ *  \see \p sort_by_key
+ *  \see \p is_sorted
+ */
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename InputIterator4,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename StrictWeakCompare>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    set_difference_by_key(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                          InputIterator1                                              keys_first1,
+                          InputIterator1                                              keys_last1,
+                          InputIterator2                                              keys_first2,
+                          InputIterator2                                              keys_last2,
+                          InputIterator3                                              values_first1,
+                          InputIterator4                                              values_first2,
+                          OutputIterator1                                             keys_result,
+                          OutputIterator2                                             values_result,
+                          StrictWeakCompare                                           comp);
+
+
+/*! \p set_difference_by_key performs a key-value difference operation from set theory.
+ *  \p set_difference_by_key constructs a sorted range that is the difference of the sorted
+ *  ranges <tt>[keys_first1, keys_last1)</tt> and <tt>[keys_first2, keys_last2)</tt>. Associated
+ *  with each element from the input and output key ranges is a value element. The associated input
+ *  value ranges need not be sorted.
+ *
+ *  In the simplest case, \p set_difference_by_key performs the "difference" operation from set
+ *  theory: the keys output range contains a copy of every element that is contained in
+ *  <tt>[keys_first1, keys_last1)</tt> and not contained in <tt>[keys_first2, keys_last2)</tt>.
+ *  The general case is more complicated, because the input ranges may contain duplicate elements.
+ *  The generalization is that if <tt>[keys_first1, keys_last1)</tt> contains \c m elements
+ *  that are equivalent to each other and if <tt>[keys_first2, keys_last2)</tt> contains \c n
+ *  elements that are equivalent to them, the last <tt>max(m-n,0)</tt> elements from
+ *  <tt>[keys_first1, keys_last1)</tt> range shall be copied to the output range.
+ *
+ *  Each time a key element is copied from <tt>[keys_first1, keys_last1)</tt> or
+ *  <tt>[keys_first2, keys_last2)</tt> is copied to the keys output range, the
+ *  corresponding value element is copied from the corresponding values input range (beginning at
+ *  \p values_first1 or \p values_first2) to the values output range.
+ *
+ *  This version of \p set_difference_by_key compares key elements using a function object \p comp.
+ *
+ *  \param keys_first1 The beginning of the first input range of keys.
+ *  \param keys_last1 The end of the first input range of keys.
+ *  \param keys_first2 The beginning of the second input range of keys.
+ *  \param keys_last2 The end of the second input range of keys.
+ *  \param values_first1 The beginning of the first input range of values.
+ *  \param values_first2 The beginning of the first input range of values.
+ *  \param keys_result The beginning of the output range of keys.
+ *  \param values_result The beginning of the output range of values.
+ *  \param comp Comparison operator.
+ *  \return A \p pair \c p such that <tt>p.first</tt> is the end of the output range of keys,
+ *          and such that <tt>p.second</tt> is the end of the output range of values.
+ *
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator1 and \p InputIterator2 have the same \c value_type,
+ *          \p InputIterator1's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a>,
+ *          the ordering on \p InputIterator1's \c value_type is a strict weak ordering, as defined in the <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a> requirements,
+ *          and \p InputIterator1's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator2 and \p InputIterator1 have the same \c value_type,
+ *          \p InputIterator2's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a>,
+ *          the ordering on \p InputIterator2's \c value_type is a strict weak ordering, as defined in the <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a> requirements,
+ *          and \p InputIterator2's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam InputIterator3 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator3's \c value_type is convertible to a type in \p OutputIterator2's set of \c value_types.
+ *  \tparam InputIterator4 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator4's \c value_type is convertible to a type in \p OutputIterator2's set of \c value_types.
+ *  \tparam OutputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam OutputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam StrictWeakCompare is a model of <a href="http://www.sgi.com/tech/stl/StrictWeakOrdering.html">Strict Weak Ordering</a>.
+ *
+ *  \pre The ranges <tt>[keys_first1, keys_last1)</tt> and <tt>[keys_first2, keys_last2)</tt> shall be sorted with respect to \p comp.
+ *  \pre The resulting ranges shall not overlap with any input range.
+ *
+ *  The following code snippet demonstrates how to use \p set_difference_by_key to compute the
+ *  set difference of two sets of integers sorted in descending order with their values.
+ *
+ *  \code
+ *  #include <thrust/set_operations.h>
+ *  #include <thrust/functional.h>
+ *  ...
+ *  int A_keys[6] = {9, 6, 5, 4, 3, 1, 0};
+ *  int A_vals[6] = {0, 0, 0, 0, 0, 0, 0};
+ *
+ *  int B_keys[5] = {9, 7, 5, 3, 1};
+ *  int B_vals[5] = {1, 1, 1, 1, 1};
+ *
+ *  int keys_result[3];
+ *  int vals_result[3];
+ *
+ *  thrust::pair<int*,int*> end = thrust::set_difference_by_key(A_keys, A_keys + 6, B_keys, B_keys + 5, A_vals, B_vals, keys_result, vals_result, thrust::greater<int>());
+ *  // keys_result is now {0, 4, 6}
+ *  // vals_result is now {0, 0, 0}
+ *  \endcode
+ *
+ *  \see \p set_union_by_key
+ *  \see \p set_intersection_by_key
+ *  \see \p set_symmetric_difference_by_key
+ *  \see \p sort_by_key
+ *  \see \p is_sorted
+ */
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename InputIterator4,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename StrictWeakCompare>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    set_difference_by_key(InputIterator1                             keys_first1,
+                          InputIterator1                             keys_last1,
+                          InputIterator2                             keys_first2,
+                          InputIterator2                             keys_last2,
+                          InputIterator3                             values_first1,
+                          InputIterator4                             values_first2,
+                          OutputIterator1                            keys_result,
+                          OutputIterator2                            values_result,
+                          StrictWeakCompare                          comp);
+
+
+/*! \p set_intersection_by_key performs a key-value intersection operation from set theory.
+ *  \p set_intersection_by_key constructs a sorted range that is the intersection of the sorted
+ *  ranges <tt>[keys_first1, keys_last1)</tt> and <tt>[keys_first2, keys_last2)</tt>. Associated
+ *  with each element from the input and output key ranges is a value element. The associated input
+ *  value ranges need not be sorted.
+ *
+ *  In the simplest case, \p set_intersection_by_key performs the "intersection" operation from set
+ *  theory: the keys output range contains a copy of every element that is contained in both
+ *  <tt>[keys_first1, keys_last1)</tt> <tt>[keys_first2, keys_last2)</tt>.
+ *  The general case is more complicated, because the input ranges may contain duplicate elements.
+ *  The generalization is that if an element appears \c m times in <tt>[keys_first1, keys_last1)</tt>
+ *  and \c n times in <tt>[keys_first2, keys_last2)</tt> (where \c m may be zero), then it
+ *  appears <tt>min(m,n)</tt> times in the keys output range.
+ *  \p set_intersection_by_key is stable, meaning both that elements are copied from the first
+ *  input range rather than the second, and that the relative order of elements in the output range
+ *  is the same as the first input range.
+ *
+ *  Each time a key element is copied from <tt>[keys_first1, keys_last1)</tt> to the keys output range,
+ *  the corresponding value element is copied from <tt>[values_first1, values_last1)</tt> to the values
+ *  output range.
+ *
+ *  This version of \p set_intersection_by_key compares objects using \c operator<.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param keys_first1 The beginning of the first input range of keys.
+ *  \param keys_last1 The end of the first input range of keys.
+ *  \param keys_first2 The beginning of the second input range of keys.
+ *  \param keys_last2 The end of the second input range of keys.
+ *  \param values_first1 The beginning of the first input range of values.
+ *  \param keys_result The beginning of the output range of keys.
+ *  \param values_result The beginning of the output range of values.
+ *  \return A \p pair \c p such that <tt>p.first</tt> is the end of the output range of keys,
+ *          and such that <tt>p.second</tt> is the end of the output range of values.
+ *
+ *  \note Unlike the other key-value set operations, \p set_intersection_by_key is unique in that it has no
+ *        \c values_first2 parameter because elements from the second input range are never copied to the output range.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator1 and \p InputIterator2 have the same \c value_type,
+ *          \p InputIterator1's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a>,
+ *          the ordering on \p InputIterator1's \c value_type is a strict weak ordering, as defined in the <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a> requirements,
+ *          and \p InputIterator1's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator2 and \p InputIterator1 have the same \c value_type,
+ *          \p InputIterator2's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a>,
+ *          the ordering on \p InputIterator2's \c value_type is a strict weak ordering, as defined in the <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a> requirements,
+ *          and \p InputIterator2's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam InputIterator3 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator3's \c value_type is convertible to a type in \p OutputIterator2's set of \c value_types.
+ *  \tparam OutputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam OutputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *
+ *  \pre The ranges <tt>[keys_first1, keys_last1)</tt> and <tt>[keys_first2, keys_last2)</tt> shall be sorted with respect to <tt>operator<</tt>.
+ *  \pre The resulting ranges shall not overlap with any input range.
+ *
+ *  The following code snippet demonstrates how to use \p set_intersection_by_key to compute the
+ *  set intersection of two sets of integers sorted in ascending order with their values using the \p thrust::host
+ *  execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/set_operations.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  int A_keys[6] = {1, 3, 5, 7, 9, 11};
+ *  int A_vals[6] = {0, 0, 0, 0, 0,  0};
+ *  
+ *  int B_keys[7] = {1, 1, 2, 3, 5,  8, 13};
+ *
+ *  int keys_result[7];
+ *  int vals_result[7];
+ *
+ *  thrust::pair<int*,int*> end = thrust::set_intersection_by_key(thrust::host, A_keys, A_keys + 6, B_keys, B_keys + 7, A_vals, keys_result, vals_result);
+ *
+ *  // keys_result is now {1, 3, 5}
+ *  // vals_result is now {0, 0, 0}
+ *  \endcode
+ *
+ *  \see \p set_union_by_key
+ *  \see \p set_difference_by_key
+ *  \see \p set_symmetric_difference_by_key
+ *  \see \p sort_by_key
+ *  \see \p is_sorted
+ */
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename OutputIterator1,
+         typename OutputIterator2>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    set_intersection_by_key(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                            InputIterator1                                              keys_first1,
+                            InputIterator1                                              keys_last1,
+                            InputIterator2                                              keys_first2,
+                            InputIterator2                                              keys_last2,
+                            InputIterator3                                              values_first1,
+                            OutputIterator1                                             keys_result,
+                            OutputIterator2                                             values_result);
+
+
+/*! \p set_intersection_by_key performs a key-value intersection operation from set theory.
+ *  \p set_intersection_by_key constructs a sorted range that is the intersection of the sorted
+ *  ranges <tt>[keys_first1, keys_last1)</tt> and <tt>[keys_first2, keys_last2)</tt>. Associated
+ *  with each element from the input and output key ranges is a value element. The associated input
+ *  value ranges need not be sorted.
+ *
+ *  In the simplest case, \p set_intersection_by_key performs the "intersection" operation from set
+ *  theory: the keys output range contains a copy of every element that is contained in both
+ *  <tt>[keys_first1, keys_last1)</tt> <tt>[keys_first2, keys_last2)</tt>.
+ *  The general case is more complicated, because the input ranges may contain duplicate elements.
+ *  The generalization is that if an element appears \c m times in <tt>[keys_first1, keys_last1)</tt>
+ *  and \c n times in <tt>[keys_first2, keys_last2)</tt> (where \c m may be zero), then it
+ *  appears <tt>min(m,n)</tt> times in the keys output range.
+ *  \p set_intersection_by_key is stable, meaning both that elements are copied from the first
+ *  input range rather than the second, and that the relative order of elements in the output range
+ *  is the same as the first input range.
+ *
+ *  Each time a key element is copied from <tt>[keys_first1, keys_last1)</tt> to the keys output range,
+ *  the corresponding value element is copied from <tt>[values_first1, values_last1)</tt> to the values
+ *  output range.
+ *
+ *  This version of \p set_intersection_by_key compares objects using \c operator<.
+ *
+ *  \param keys_first1 The beginning of the first input range of keys.
+ *  \param keys_last1 The end of the first input range of keys.
+ *  \param keys_first2 The beginning of the second input range of keys.
+ *  \param keys_last2 The end of the second input range of keys.
+ *  \param values_first1 The beginning of the first input range of values.
+ *  \param keys_result The beginning of the output range of keys.
+ *  \param values_result The beginning of the output range of values.
+ *  \return A \p pair \c p such that <tt>p.first</tt> is the end of the output range of keys,
+ *          and such that <tt>p.second</tt> is the end of the output range of values.
+ *
+ *  \note Unlike the other key-value set operations, \p set_intersection_by_key is unique in that it has no
+ *        \c values_first2 parameter because elements from the second input range are never copied to the output range.
+ *
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator1 and \p InputIterator2 have the same \c value_type,
+ *          \p InputIterator1's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a>,
+ *          the ordering on \p InputIterator1's \c value_type is a strict weak ordering, as defined in the <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a> requirements,
+ *          and \p InputIterator1's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator2 and \p InputIterator1 have the same \c value_type,
+ *          \p InputIterator2's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a>,
+ *          the ordering on \p InputIterator2's \c value_type is a strict weak ordering, as defined in the <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a> requirements,
+ *          and \p InputIterator2's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam InputIterator3 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator3's \c value_type is convertible to a type in \p OutputIterator2's set of \c value_types.
+ *  \tparam OutputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam OutputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *
+ *  \pre The ranges <tt>[keys_first1, keys_last1)</tt> and <tt>[keys_first2, keys_last2)</tt> shall be sorted with respect to <tt>operator<</tt>.
+ *  \pre The resulting ranges shall not overlap with any input range.
+ *
+ *  The following code snippet demonstrates how to use \p set_intersection_by_key to compute the
+ *  set intersection of two sets of integers sorted in ascending order with their values.
+ *
+ *  \code
+ *  #include <thrust/set_operations.h>
+ *  ...
+ *  int A_keys[6] = {1, 3, 5, 7, 9, 11};
+ *  int A_vals[6] = {0, 0, 0, 0, 0,  0};
+ *  
+ *  int B_keys[7] = {1, 1, 2, 3, 5,  8, 13};
+ *
+ *  int keys_result[7];
+ *  int vals_result[7];
+ *
+ *  thrust::pair<int*,int*> end = thrust::set_intersection_by_key(A_keys, A_keys + 6, B_keys, B_keys + 7, A_vals, keys_result, vals_result);
+ *
+ *  // keys_result is now {1, 3, 5}
+ *  // vals_result is now {0, 0, 0}
+ *  \endcode
+ *
+ *  \see \p set_union_by_key
+ *  \see \p set_difference_by_key
+ *  \see \p set_symmetric_difference_by_key
+ *  \see \p sort_by_key
+ *  \see \p is_sorted
+ */
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename OutputIterator1,
+         typename OutputIterator2>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    set_intersection_by_key(InputIterator1                             keys_first1,
+                            InputIterator1                             keys_last1,
+                            InputIterator2                             keys_first2,
+                            InputIterator2                             keys_last2,
+                            InputIterator3                             values_first1,
+                            OutputIterator1                            keys_result,
+                            OutputIterator2                            values_result);
+
+
+/*! \p set_intersection_by_key performs a key-value intersection operation from set theory.
+ *  \p set_intersection_by_key constructs a sorted range that is the intersection of the sorted
+ *  ranges <tt>[keys_first1, keys_last1)</tt> and <tt>[keys_first2, keys_last2)</tt>. Associated
+ *  with each element from the input and output key ranges is a value element. The associated input
+ *  value ranges need not be sorted.
+ *
+ *  In the simplest case, \p set_intersection_by_key performs the "intersection" operation from set
+ *  theory: the keys output range contains a copy of every element that is contained in both
+ *  <tt>[keys_first1, keys_last1)</tt> <tt>[keys_first2, keys_last2)</tt>.
+ *  The general case is more complicated, because the input ranges may contain duplicate elements.
+ *  The generalization is that if an element appears \c m times in <tt>[keys_first1, keys_last1)</tt>
+ *  and \c n times in <tt>[keys_first2, keys_last2)</tt> (where \c m may be zero), then it
+ *  appears <tt>min(m,n)</tt> times in the keys output range.
+ *  \p set_intersection_by_key is stable, meaning both that elements are copied from the first
+ *  input range rather than the second, and that the relative order of elements in the output range
+ *  is the same as the first input range.
+ *
+ *  Each time a key element is copied from <tt>[keys_first1, keys_last1)</tt> to the keys output range,
+ *  the corresponding value element is copied from <tt>[values_first1, values_last1)</tt> to the values
+ *  output range.
+ *
+ *  This version of \p set_intersection_by_key compares objects using a function object \p comp.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param keys_first1 The beginning of the first input range of keys.
+ *  \param keys_last1 The end of the first input range of keys.
+ *  \param keys_first2 The beginning of the second input range of keys.
+ *  \param keys_last2 The end of the second input range of keys.
+ *  \param values_first1 The beginning of the first input range of values.
+ *  \param keys_result The beginning of the output range of keys.
+ *  \param values_result The beginning of the output range of values.
+ *  \param comp Comparison operator.
+ *  \return A \p pair \c p such that <tt>p.first</tt> is the end of the output range of keys,
+ *          and such that <tt>p.second</tt> is the end of the output range of values.
+ *
+ *  \note Unlike the other key-value set operations, \p set_intersection_by_key is unique in that it has no
+ *        \c values_first2 parameter because elements from the second input range are never copied to the output range.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator1 and \p InputIterator2 have the same \c value_type,
+ *          \p InputIterator1's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a>,
+ *          the ordering on \p InputIterator1's \c value_type is a strict weak ordering, as defined in the <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a> requirements,
+ *          and \p InputIterator1's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator2 and \p InputIterator1 have the same \c value_type,
+ *          \p InputIterator2's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a>,
+ *          the ordering on \p InputIterator2's \c value_type is a strict weak ordering, as defined in the <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a> requirements,
+ *          and \p InputIterator2's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam InputIterator3 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator3's \c value_type is convertible to a type in \p OutputIterator2's set of \c value_types.
+ *  \tparam OutputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam OutputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam StrictWeakCompare is a model of <a href="http://www.sgi.com/tech/stl/StrictWeakOrdering.html">Strict Weak Ordering</a>.
+ *
+ *  \pre The ranges <tt>[keys_first1, keys_last1)</tt> and <tt>[keys_first2, keys_last2)</tt> shall be sorted with respect to \p comp.
+ *  \pre The resulting ranges shall not overlap with any input range.
+ *
+ *  The following code snippet demonstrates how to use \p set_intersection_by_key to compute the
+ *  set intersection of two sets of integers sorted in descending order with their values using the
+ *  \p thrust::host execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/set_operations.h>
+ *  #include <thrust/functional.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  int A_keys[6] = {11, 9, 7, 5, 3, 1};
+ *  int A_vals[6] = { 0, 0, 0, 0, 0, 0};
+ *  
+ *  int B_keys[7] = {13, 8, 5, 3, 2, 1, 1};
+ *
+ *  int keys_result[7];
+ *  int vals_result[7];
+ *
+ *  thrust::pair<int*,int*> end = thrust::set_intersection_by_key(thrust::host, A_keys, A_keys + 6, B_keys, B_keys + 7, A_vals, keys_result, vals_result, thrust::greater<int>());
+ *
+ *  // keys_result is now {5, 3, 1}
+ *  // vals_result is now {0, 0, 0}
+ *  \endcode
+ *
+ *  \see \p set_union_by_key
+ *  \see \p set_difference_by_key
+ *  \see \p set_symmetric_difference_by_key
+ *  \see \p sort_by_key
+ *  \see \p is_sorted
+ */
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename StrictWeakCompare>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    set_intersection_by_key(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                            InputIterator1                                              keys_first1,
+                            InputIterator1                                              keys_last1,
+                            InputIterator2                                              keys_first2,
+                            InputIterator2                                              keys_last2,
+                            InputIterator3                                              values_first1,
+                            OutputIterator1                                             keys_result,
+                            OutputIterator2                                             values_result,
+                            StrictWeakCompare                                           comp);
+
+
+/*! \p set_intersection_by_key performs a key-value intersection operation from set theory.
+ *  \p set_intersection_by_key constructs a sorted range that is the intersection of the sorted
+ *  ranges <tt>[keys_first1, keys_last1)</tt> and <tt>[keys_first2, keys_last2)</tt>. Associated
+ *  with each element from the input and output key ranges is a value element. The associated input
+ *  value ranges need not be sorted.
+ *
+ *  In the simplest case, \p set_intersection_by_key performs the "intersection" operation from set
+ *  theory: the keys output range contains a copy of every element that is contained in both
+ *  <tt>[keys_first1, keys_last1)</tt> <tt>[keys_first2, keys_last2)</tt>.
+ *  The general case is more complicated, because the input ranges may contain duplicate elements.
+ *  The generalization is that if an element appears \c m times in <tt>[keys_first1, keys_last1)</tt>
+ *  and \c n times in <tt>[keys_first2, keys_last2)</tt> (where \c m may be zero), then it
+ *  appears <tt>min(m,n)</tt> times in the keys output range.
+ *  \p set_intersection_by_key is stable, meaning both that elements are copied from the first
+ *  input range rather than the second, and that the relative order of elements in the output range
+ *  is the same as the first input range.
+ *
+ *  Each time a key element is copied from <tt>[keys_first1, keys_last1)</tt> to the keys output range,
+ *  the corresponding value element is copied from <tt>[values_first1, values_last1)</tt> to the values
+ *  output range.
+ *
+ *  This version of \p set_intersection_by_key compares objects using a function object \p comp.
+ *
+ *  \param keys_first1 The beginning of the first input range of keys.
+ *  \param keys_last1 The end of the first input range of keys.
+ *  \param keys_first2 The beginning of the second input range of keys.
+ *  \param keys_last2 The end of the second input range of keys.
+ *  \param values_first1 The beginning of the first input range of values.
+ *  \param keys_result The beginning of the output range of keys.
+ *  \param values_result The beginning of the output range of values.
+ *  \param comp Comparison operator.
+ *  \return A \p pair \c p such that <tt>p.first</tt> is the end of the output range of keys,
+ *          and such that <tt>p.second</tt> is the end of the output range of values.
+ *
+ *  \note Unlike the other key-value set operations, \p set_intersection_by_key is unique in that it has no
+ *        \c values_first2 parameter because elements from the second input range are never copied to the output range.
+ *
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator1 and \p InputIterator2 have the same \c value_type,
+ *          \p InputIterator1's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a>,
+ *          the ordering on \p InputIterator1's \c value_type is a strict weak ordering, as defined in the <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a> requirements,
+ *          and \p InputIterator1's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator2 and \p InputIterator1 have the same \c value_type,
+ *          \p InputIterator2's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a>,
+ *          the ordering on \p InputIterator2's \c value_type is a strict weak ordering, as defined in the <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a> requirements,
+ *          and \p InputIterator2's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam InputIterator3 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator3's \c value_type is convertible to a type in \p OutputIterator2's set of \c value_types.
+ *  \tparam OutputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam OutputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam StrictWeakCompare is a model of <a href="http://www.sgi.com/tech/stl/StrictWeakOrdering.html">Strict Weak Ordering</a>.
+ *
+ *  \pre The ranges <tt>[keys_first1, keys_last1)</tt> and <tt>[keys_first2, keys_last2)</tt> shall be sorted with respect to \p comp.
+ *  \pre The resulting ranges shall not overlap with any input range.
+ *
+ *  The following code snippet demonstrates how to use \p set_intersection_by_key to compute the
+ *  set intersection of two sets of integers sorted in descending order with their values.
+ *
+ *  \code
+ *  #include <thrust/set_operations.h>
+ *  #include <thrust/functional.h>
+ *  ...
+ *  int A_keys[6] = {11, 9, 7, 5, 3, 1};
+ *  int A_vals[6] = { 0, 0, 0, 0, 0, 0};
+ *  
+ *  int B_keys[7] = {13, 8, 5, 3, 2, 1, 1};
+ *
+ *  int keys_result[7];
+ *  int vals_result[7];
+ *
+ *  thrust::pair<int*,int*> end = thrust::set_intersection_by_key(A_keys, A_keys + 6, B_keys, B_keys + 7, A_vals, keys_result, vals_result, thrust::greater<int>());
+ *
+ *  // keys_result is now {5, 3, 1}
+ *  // vals_result is now {0, 0, 0}
+ *  \endcode
+ *
+ *  \see \p set_union_by_key
+ *  \see \p set_difference_by_key
+ *  \see \p set_symmetric_difference_by_key
+ *  \see \p sort_by_key
+ *  \see \p is_sorted
+ */
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename StrictWeakCompare>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    set_intersection_by_key(InputIterator1                             keys_first1,
+                            InputIterator1                             keys_last1,
+                            InputIterator2                             keys_first2,
+                            InputIterator2                             keys_last2,
+                            InputIterator3                             values_first1,
+                            OutputIterator1                            keys_result,
+                            OutputIterator2                            values_result,
+                            StrictWeakCompare                          comp);
+
+
+/*! \p set_symmetric_difference_by_key performs a key-value symmetric difference operation from set theory.
+ *  \p set_difference_by_key constructs a sorted range that is the symmetric difference of the sorted
+ *  ranges <tt>[keys_first1, keys_last1)</tt> and <tt>[keys_first2, keys_last2)</tt>. Associated
+ *  with each element from the input and output key ranges is a value element. The associated input
+ *  value ranges need not be sorted.
+ *
+ *  In the simplest case, \p set_symmetric_difference_by_key performs a set theoretic calculation:
+ *  it constructs the union of the two sets A - B and B - A, where A and B are the two
+ *  input ranges. That is, the output range contains a copy of every element that is
+ *  contained in <tt>[keys_first1, keys_last1)</tt> but not <tt>[keys_first2, keys_last1)</tt>, and a copy of
+ *  every element that is contained in <tt>[keys_first2, keys_last2)</tt> but not <tt>[keys_first1, keys_last1)</tt>.
+ *  The general case is more complicated, because the input ranges may contain duplicate elements.
+ *  The generalization is that if <tt>[keys_first1, keys_last1)</tt> contains \c m elements that are
+ *  equivalent to each other and <tt>[keys_first2, keys_last1)</tt> contains \c n elements that are
+ *  equivalent to them, then <tt>|m - n|</tt> of those elements shall be copied to the output
+ *  range: the last <tt>m - n</tt> elements from <tt>[keys_first1, keys_last1)</tt> if <tt>m > n</tt>, and
+ *  the last <tt>n - m</tt> of these elements from <tt>[keys_first2, keys_last2)</tt> if <tt>m < n</tt>.
+ *
+ *  Each time a key element is copied from <tt>[keys_first1, keys_last1)</tt> or
+ *  <tt>[keys_first2, keys_last2)</tt> is copied to the keys output range, the
+ *  corresponding value element is copied from the corresponding values input range (beginning at
+ *  \p values_first1 or \p values_first2) to the values output range.
+ *
+ *  This version of \p set_symmetric_difference_by_key compares key elements using \c operator<.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param keys_first1 The beginning of the first input range of keys.
+ *  \param keys_last1 The end of the first input range of keys.
+ *  \param keys_first2 The beginning of the second input range of keys.
+ *  \param keys_last2 The end of the second input range of keys.
+ *  \param values_first1 The beginning of the first input range of values.
+ *  \param values_first2 The beginning of the first input range of values.
+ *  \param keys_result The beginning of the output range of keys.
+ *  \param values_result The beginning of the output range of values.
+ *  \return A \p pair \c p such that <tt>p.first</tt> is the end of the output range of keys,
+ *          and such that <tt>p.second</tt> is the end of the output range of values.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator1 and \p InputIterator2 have the same \c value_type,
+ *          \p InputIterator1's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a>,
+ *          the ordering on \p InputIterator1's \c value_type is a strict weak ordering, as defined in the <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a> requirements,
+ *          and \p InputIterator1's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator2 and \p InputIterator1 have the same \c value_type,
+ *          \p InputIterator2's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a>,
+ *          the ordering on \p InputIterator2's \c value_type is a strict weak ordering, as defined in the <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a> requirements,
+ *          and \p InputIterator2's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam InputIterator3 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator3's \c value_type is convertible to a type in \p OutputIterator2's set of \c value_types.
+ *  \tparam InputIterator4 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator4's \c value_type is convertible to a type in \p OutputIterator2's set of \c value_types.
+ *  \tparam OutputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam OutputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *
+ *  \pre The ranges <tt>[keys_first1, keys_last1)</tt> and <tt>[keys_first2, keys_last2)</tt> shall be sorted with respect to <tt>operator<</tt>.
+ *  \pre The resulting ranges shall not overlap with any input range.
+ *
+ *  The following code snippet demonstrates how to use \p set_symmetric_difference_by_key to compute the
+ *  symmetric difference of two sets of integers sorted in ascending order with their values using the
+ *  \p thrust::host execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/set_operations.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  int A_keys[6] = {0, 1, 2, 2, 4, 6, 7};
+ *  int A_vals[6] = {0, 0, 0, 0, 0, 0, 0};
+ *
+ *  int B_keys[5] = {1, 1, 2, 5, 8};
+ *  int B_vals[5] = {1, 1, 1, 1, 1};
+ *
+ *  int keys_result[6];
+ *  int vals_result[6];
+ *
+ *  thrust::pair<int*,int*> end = thrust::set_symmetric_difference_by_key(thrust::host, A_keys, A_keys + 6, B_keys, B_keys + 5, A_vals, B_vals, keys_result, vals_result);
+ *  // keys_result is now {0, 4, 5, 6, 7, 8}
+ *  // vals_result is now {0, 0, 1, 0, 0, 1}
+ *  \endcode
+ *
+ *  \see \p set_union_by_key
+ *  \see \p set_intersection_by_key
+ *  \see \p set_difference_by_key
+ *  \see \p sort_by_key
+ *  \see \p is_sorted
+ */
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename InputIterator4,
+         typename OutputIterator1,
+         typename OutputIterator2>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    set_symmetric_difference_by_key(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                    InputIterator1                                              keys_first1,
+                                    InputIterator1                                              keys_last1,
+                                    InputIterator2                                              keys_first2,
+                                    InputIterator2                                              keys_last2,
+                                    InputIterator3                                              values_first1,
+                                    InputIterator4                                              values_first2,
+                                    OutputIterator1                                             keys_result,
+                                    OutputIterator2                                             values_result);
+
+
+/*! \p set_symmetric_difference_by_key performs a key-value symmetric difference operation from set theory.
+ *  \p set_difference_by_key constructs a sorted range that is the symmetric difference of the sorted
+ *  ranges <tt>[keys_first1, keys_last1)</tt> and <tt>[keys_first2, keys_last2)</tt>. Associated
+ *  with each element from the input and output key ranges is a value element. The associated input
+ *  value ranges need not be sorted.
+ *
+ *  In the simplest case, \p set_symmetric_difference_by_key performs a set theoretic calculation:
+ *  it constructs the union of the two sets A - B and B - A, where A and B are the two
+ *  input ranges. That is, the output range contains a copy of every element that is
+ *  contained in <tt>[keys_first1, keys_last1)</tt> but not <tt>[keys_first2, keys_last1)</tt>, and a copy of
+ *  every element that is contained in <tt>[keys_first2, keys_last2)</tt> but not <tt>[keys_first1, keys_last1)</tt>.
+ *  The general case is more complicated, because the input ranges may contain duplicate elements.
+ *  The generalization is that if <tt>[keys_first1, keys_last1)</tt> contains \c m elements that are
+ *  equivalent to each other and <tt>[keys_first2, keys_last1)</tt> contains \c n elements that are
+ *  equivalent to them, then <tt>|m - n|</tt> of those elements shall be copied to the output
+ *  range: the last <tt>m - n</tt> elements from <tt>[keys_first1, keys_last1)</tt> if <tt>m > n</tt>, and
+ *  the last <tt>n - m</tt> of these elements from <tt>[keys_first2, keys_last2)</tt> if <tt>m < n</tt>.
+ *
+ *  Each time a key element is copied from <tt>[keys_first1, keys_last1)</tt> or
+ *  <tt>[keys_first2, keys_last2)</tt> is copied to the keys output range, the
+ *  corresponding value element is copied from the corresponding values input range (beginning at
+ *  \p values_first1 or \p values_first2) to the values output range.
+ *
+ *  This version of \p set_symmetric_difference_by_key compares key elements using \c operator<.
+ *
+ *  \param keys_first1 The beginning of the first input range of keys.
+ *  \param keys_last1 The end of the first input range of keys.
+ *  \param keys_first2 The beginning of the second input range of keys.
+ *  \param keys_last2 The end of the second input range of keys.
+ *  \param values_first1 The beginning of the first input range of values.
+ *  \param values_first2 The beginning of the first input range of values.
+ *  \param keys_result The beginning of the output range of keys.
+ *  \param values_result The beginning of the output range of values.
+ *  \return A \p pair \c p such that <tt>p.first</tt> is the end of the output range of keys,
+ *          and such that <tt>p.second</tt> is the end of the output range of values.
+ *
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator1 and \p InputIterator2 have the same \c value_type,
+ *          \p InputIterator1's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a>,
+ *          the ordering on \p InputIterator1's \c value_type is a strict weak ordering, as defined in the <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a> requirements,
+ *          and \p InputIterator1's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator2 and \p InputIterator1 have the same \c value_type,
+ *          \p InputIterator2's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a>,
+ *          the ordering on \p InputIterator2's \c value_type is a strict weak ordering, as defined in the <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a> requirements,
+ *          and \p InputIterator2's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam InputIterator3 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator3's \c value_type is convertible to a type in \p OutputIterator2's set of \c value_types.
+ *  \tparam InputIterator4 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator4's \c value_type is convertible to a type in \p OutputIterator2's set of \c value_types.
+ *  \tparam OutputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam OutputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *
+ *  \pre The ranges <tt>[keys_first1, keys_last1)</tt> and <tt>[keys_first2, keys_last2)</tt> shall be sorted with respect to <tt>operator<</tt>.
+ *  \pre The resulting ranges shall not overlap with any input range.
+ *
+ *  The following code snippet demonstrates how to use \p set_symmetric_difference_by_key to compute the
+ *  symmetric difference of two sets of integers sorted in ascending order with their values.
+ *
+ *  \code
+ *  #include <thrust/set_operations.h>
+ *  ...
+ *  int A_keys[6] = {0, 1, 2, 2, 4, 6, 7};
+ *  int A_vals[6] = {0, 0, 0, 0, 0, 0, 0};
+ *
+ *  int B_keys[5] = {1, 1, 2, 5, 8};
+ *  int B_vals[5] = {1, 1, 1, 1, 1};
+ *
+ *  int keys_result[6];
+ *  int vals_result[6];
+ *
+ *  thrust::pair<int*,int*> end = thrust::set_symmetric_difference_by_key(A_keys, A_keys + 6, B_keys, B_keys + 5, A_vals, B_vals, keys_result, vals_result);
+ *  // keys_result is now {0, 4, 5, 6, 7, 8}
+ *  // vals_result is now {0, 0, 1, 0, 0, 1}
+ *  \endcode
+ *
+ *  \see \p set_union_by_key
+ *  \see \p set_intersection_by_key
+ *  \see \p set_difference_by_key
+ *  \see \p sort_by_key
+ *  \see \p is_sorted
+ */
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename InputIterator4,
+         typename OutputIterator1,
+         typename OutputIterator2>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    set_symmetric_difference_by_key(InputIterator1                             keys_first1,
+                                    InputIterator1                             keys_last1,
+                                    InputIterator2                             keys_first2,
+                                    InputIterator2                             keys_last2,
+                                    InputIterator3                             values_first1,
+                                    InputIterator4                             values_first2,
+                                    OutputIterator1                            keys_result,
+                                    OutputIterator2                            values_result);
+
+
+/*! \p set_symmetric_difference_by_key performs a key-value symmetric difference operation from set theory.
+ *  \p set_difference_by_key constructs a sorted range that is the symmetric difference of the sorted
+ *  ranges <tt>[keys_first1, keys_last1)</tt> and <tt>[keys_first2, keys_last2)</tt>. Associated
+ *  with each element from the input and output key ranges is a value element. The associated input
+ *  value ranges need not be sorted.
+ *
+ *  In the simplest case, \p set_symmetric_difference_by_key performs a set theoretic calculation:
+ *  it constructs the union of the two sets A - B and B - A, where A and B are the two
+ *  input ranges. That is, the output range contains a copy of every element that is
+ *  contained in <tt>[keys_first1, keys_last1)</tt> but not <tt>[keys_first2, keys_last1)</tt>, and a copy of
+ *  every element that is contained in <tt>[keys_first2, keys_last2)</tt> but not <tt>[keys_first1, keys_last1)</tt>.
+ *  The general case is more complicated, because the input ranges may contain duplicate elements.
+ *  The generalization is that if <tt>[keys_first1, keys_last1)</tt> contains \c m elements that are
+ *  equivalent to each other and <tt>[keys_first2, keys_last1)</tt> contains \c n elements that are
+ *  equivalent to them, then <tt>|m - n|</tt> of those elements shall be copied to the output
+ *  range: the last <tt>m - n</tt> elements from <tt>[keys_first1, keys_last1)</tt> if <tt>m > n</tt>, and
+ *  the last <tt>n - m</tt> of these elements from <tt>[keys_first2, keys_last2)</tt> if <tt>m < n</tt>.
+ *
+ *  Each time a key element is copied from <tt>[keys_first1, keys_last1)</tt> or
+ *  <tt>[keys_first2, keys_last2)</tt> is copied to the keys output range, the
+ *  corresponding value element is copied from the corresponding values input range (beginning at
+ *  \p values_first1 or \p values_first2) to the values output range.
+ *
+ *  This version of \p set_symmetric_difference_by_key compares key elements using a function object \c comp.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param keys_first1 The beginning of the first input range of keys.
+ *  \param keys_last1 The end of the first input range of keys.
+ *  \param keys_first2 The beginning of the second input range of keys.
+ *  \param keys_last2 The end of the second input range of keys.
+ *  \param values_first1 The beginning of the first input range of values.
+ *  \param values_first2 The beginning of the first input range of values.
+ *  \param keys_result The beginning of the output range of keys.
+ *  \param values_result The beginning of the output range of values.
+ *  \param comp Comparison operator.
+ *  \return A \p pair \c p such that <tt>p.first</tt> is the end of the output range of keys,
+ *          and such that <tt>p.second</tt> is the end of the output range of values.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator1 and \p InputIterator2 have the same \c value_type,
+ *          \p InputIterator1's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a>,
+ *          the ordering on \p InputIterator1's \c value_type is a strict weak ordering, as defined in the <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a> requirements,
+ *          and \p InputIterator1's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator2 and \p InputIterator1 have the same \c value_type,
+ *          \p InputIterator2's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a>,
+ *          the ordering on \p InputIterator2's \c value_type is a strict weak ordering, as defined in the <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a> requirements,
+ *          and \p InputIterator2's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam InputIterator3 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator3's \c value_type is convertible to a type in \p OutputIterator2's set of \c value_types.
+ *  \tparam InputIterator4 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator4's \c value_type is convertible to a type in \p OutputIterator2's set of \c value_types.
+ *  \tparam OutputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam OutputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam StrictWeakCompare is a model of <a href="http://www.sgi.com/tech/stl/StrictWeakOrdering.html">Strict Weak Ordering</a>.
+ *
+ *  \pre The ranges <tt>[keys_first1, keys_last1)</tt> and <tt>[keys_first2, keys_last2)</tt> shall be sorted with respect to \p comp.
+ *  \pre The resulting ranges shall not overlap with any input range.
+ *
+ *  The following code snippet demonstrates how to use \p set_symmetric_difference_by_key to compute the
+ *  symmetric difference of two sets of integers sorted in descending order with their values using the
+ *  \p thrust::host execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/set_operations.h>
+ *  #include <thrust/functional.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  int A_keys[6] = {7, 6, 4, 2, 2, 1, 0};
+ *  int A_vals[6] = {0, 0, 0, 0, 0, 0, 0};
+ *
+ *  int B_keys[5] = {8, 5, 2, 1, 1};
+ *  int B_vals[5] = {1, 1, 1, 1, 1};
+ *
+ *  int keys_result[6];
+ *  int vals_result[6];
+ *
+ *  thrust::pair<int*,int*> end = thrust::set_symmetric_difference_by_key(thrust::host, A_keys, A_keys + 6, B_keys, B_keys + 5, A_vals, B_vals, keys_result, vals_result);
+ *  // keys_result is now {8, 7, 6, 5, 4, 0}
+ *  // vals_result is now {1, 0, 0, 1, 0, 0}
+ *  \endcode
+ *
+ *  \see \p set_union_by_key
+ *  \see \p set_intersection_by_key
+ *  \see \p set_difference_by_key
+ *  \see \p sort_by_key
+ *  \see \p is_sorted
+ */
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename InputIterator4,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename StrictWeakCompare>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    set_symmetric_difference_by_key(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                    InputIterator1                                              keys_first1,
+                                    InputIterator1                                              keys_last1,
+                                    InputIterator2                                              keys_first2,
+                                    InputIterator2                                              keys_last2,
+                                    InputIterator3                                              values_first1,
+                                    InputIterator4                                              values_first2,
+                                    OutputIterator1                                             keys_result,
+                                    OutputIterator2                                             values_result,
+                                    StrictWeakCompare                                           comp);
+
+
+/*! \p set_symmetric_difference_by_key performs a key-value symmetric difference operation from set theory.
+ *  \p set_difference_by_key constructs a sorted range that is the symmetric difference of the sorted
+ *  ranges <tt>[keys_first1, keys_last1)</tt> and <tt>[keys_first2, keys_last2)</tt>. Associated
+ *  with each element from the input and output key ranges is a value element. The associated input
+ *  value ranges need not be sorted.
+ *
+ *  In the simplest case, \p set_symmetric_difference_by_key performs a set theoretic calculation:
+ *  it constructs the union of the two sets A - B and B - A, where A and B are the two
+ *  input ranges. That is, the output range contains a copy of every element that is
+ *  contained in <tt>[keys_first1, keys_last1)</tt> but not <tt>[keys_first2, keys_last1)</tt>, and a copy of
+ *  every element that is contained in <tt>[keys_first2, keys_last2)</tt> but not <tt>[keys_first1, keys_last1)</tt>.
+ *  The general case is more complicated, because the input ranges may contain duplicate elements.
+ *  The generalization is that if <tt>[keys_first1, keys_last1)</tt> contains \c m elements that are
+ *  equivalent to each other and <tt>[keys_first2, keys_last1)</tt> contains \c n elements that are
+ *  equivalent to them, then <tt>|m - n|</tt> of those elements shall be copied to the output
+ *  range: the last <tt>m - n</tt> elements from <tt>[keys_first1, keys_last1)</tt> if <tt>m > n</tt>, and
+ *  the last <tt>n - m</tt> of these elements from <tt>[keys_first2, keys_last2)</tt> if <tt>m < n</tt>.
+ *
+ *  Each time a key element is copied from <tt>[keys_first1, keys_last1)</tt> or
+ *  <tt>[keys_first2, keys_last2)</tt> is copied to the keys output range, the
+ *  corresponding value element is copied from the corresponding values input range (beginning at
+ *  \p values_first1 or \p values_first2) to the values output range.
+ *
+ *  This version of \p set_symmetric_difference_by_key compares key elements using a function object \c comp.
+ *
+ *  \param keys_first1 The beginning of the first input range of keys.
+ *  \param keys_last1 The end of the first input range of keys.
+ *  \param keys_first2 The beginning of the second input range of keys.
+ *  \param keys_last2 The end of the second input range of keys.
+ *  \param values_first1 The beginning of the first input range of values.
+ *  \param values_first2 The beginning of the first input range of values.
+ *  \param keys_result The beginning of the output range of keys.
+ *  \param values_result The beginning of the output range of values.
+ *  \param comp Comparison operator.
+ *  \return A \p pair \c p such that <tt>p.first</tt> is the end of the output range of keys,
+ *          and such that <tt>p.second</tt> is the end of the output range of values.
+ *
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator1 and \p InputIterator2 have the same \c value_type,
+ *          \p InputIterator1's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a>,
+ *          the ordering on \p InputIterator1's \c value_type is a strict weak ordering, as defined in the <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a> requirements,
+ *          and \p InputIterator1's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator2 and \p InputIterator1 have the same \c value_type,
+ *          \p InputIterator2's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a>,
+ *          the ordering on \p InputIterator2's \c value_type is a strict weak ordering, as defined in the <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a> requirements,
+ *          and \p InputIterator2's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam InputIterator3 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator3's \c value_type is convertible to a type in \p OutputIterator2's set of \c value_types.
+ *  \tparam InputIterator4 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator4's \c value_type is convertible to a type in \p OutputIterator2's set of \c value_types.
+ *  \tparam OutputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam OutputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam StrictWeakCompare is a model of <a href="http://www.sgi.com/tech/stl/StrictWeakOrdering.html">Strict Weak Ordering</a>.
+ *
+ *  \pre The ranges <tt>[keys_first1, keys_last1)</tt> and <tt>[keys_first2, keys_last2)</tt> shall be sorted with respect to \p comp.
+ *  \pre The resulting ranges shall not overlap with any input range.
+ *
+ *  The following code snippet demonstrates how to use \p set_symmetric_difference_by_key to compute the
+ *  symmetric difference of two sets of integers sorted in descending order with their values.
+ *
+ *  \code
+ *  #include <thrust/set_operations.h>
+ *  #include <thrust/functional.h>
+ *  ...
+ *  int A_keys[6] = {7, 6, 4, 2, 2, 1, 0};
+ *  int A_vals[6] = {0, 0, 0, 0, 0, 0, 0};
+ *
+ *  int B_keys[5] = {8, 5, 2, 1, 1};
+ *  int B_vals[5] = {1, 1, 1, 1, 1};
+ *
+ *  int keys_result[6];
+ *  int vals_result[6];
+ *
+ *  thrust::pair<int*,int*> end = thrust::set_symmetric_difference_by_key(A_keys, A_keys + 6, B_keys, B_keys + 5, A_vals, B_vals, keys_result, vals_result);
+ *  // keys_result is now {8, 7, 6, 5, 4, 0}
+ *  // vals_result is now {1, 0, 0, 1, 0, 0}
+ *  \endcode
+ *
+ *  \see \p set_union_by_key
+ *  \see \p set_intersection_by_key
+ *  \see \p set_difference_by_key
+ *  \see \p sort_by_key
+ *  \see \p is_sorted
+ */
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename InputIterator4,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename StrictWeakCompare>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    set_symmetric_difference_by_key(InputIterator1                             keys_first1,
+                                    InputIterator1                             keys_last1,
+                                    InputIterator2                             keys_first2,
+                                    InputIterator2                             keys_last2,
+                                    InputIterator3                             values_first1,
+                                    InputIterator4                             values_first2,
+                                    OutputIterator1                            keys_result,
+                                    OutputIterator2                            values_result,
+                                    StrictWeakCompare                          comp);
+
+
+/*! \p set_union_by_key performs a key-value union operation from set theory.
+ *  \p set_union_by_key constructs a sorted range that is the union of the sorted
+ *  ranges <tt>[keys_first1, keys_last1)</tt> and <tt>[keys_first2, keys_last2)</tt>. Associated
+ *  with each element from the input and output key ranges is a value element. The associated input
+ *  value ranges need not be sorted.
+ *
+ *  In the simplest case, \p set_union_by_key performs the "union" operation from set theory:
+ *  the output range contains a copy of every element that is contained in
+ *  <tt>[keys_first1, keys_last1)</tt>, <tt>[keys_first2, keys_last1)</tt>, or both. The general case
+ *  is more complicated, because the input ranges may contain duplicate elements.
+ *  The generalization is that if <tt>[keys_first1, keys_last1)</tt> contains \c m elements
+ *  that are equivalent to each other and if <tt>[keys_first2, keys_last2)</tt> contains \c n
+ *  elements that are equivalent to them, then all \c m elements from the first
+ *  range shall be copied to the output range, in order, and then <tt>max(n - m, 0)</tt>
+ *  elements from the second range shall be copied to the output, in order.
+ *
+ *  Each time a key element is copied from <tt>[keys_first1, keys_last1)</tt> or
+ *  <tt>[keys_first2, keys_last2)</tt> is copied to the keys output range, the
+ *  corresponding value element is copied from the corresponding values input range (beginning at
+ *  \p values_first1 or \p values_first2) to the values output range.
+ *
+ *  This version of \p set_union_by_key compares key elements using \c operator<.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param keys_first1 The beginning of the first input range of keys.
+ *  \param keys_last1 The end of the first input range of keys.
+ *  \param keys_first2 The beginning of the second input range of keys.
+ *  \param keys_last2 The end of the second input range of keys.
+ *  \param values_first1 The beginning of the first input range of values.
+ *  \param values_first2 The beginning of the first input range of values.
+ *  \param keys_result The beginning of the output range of keys.
+ *  \param values_result The beginning of the output range of values.
+ *  \return A \p pair \c p such that <tt>p.first</tt> is the end of the output range of keys,
+ *          and such that <tt>p.second</tt> is the end of the output range of values.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator1 and \p InputIterator2 have the same \c value_type,
+ *          \p InputIterator1's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a>,
+ *          the ordering on \p InputIterator1's \c value_type is a strict weak ordering, as defined in the <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a> requirements,
+ *          and \p InputIterator1's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator2 and \p InputIterator1 have the same \c value_type,
+ *          \p InputIterator2's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a>,
+ *          the ordering on \p InputIterator2's \c value_type is a strict weak ordering, as defined in the <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a> requirements,
+ *          and \p InputIterator2's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam InputIterator3 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator3's \c value_type is convertible to a type in \p OutputIterator2's set of \c value_types.
+ *  \tparam InputIterator4 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator4's \c value_type is convertible to a type in \p OutputIterator2's set of \c value_types.
+ *  \tparam OutputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam OutputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *
+ *  \pre The ranges <tt>[keys_first1, keys_last1)</tt> and <tt>[keys_first2, keys_last2)</tt> shall be sorted with respect to <tt>operator<</tt>.
+ *  \pre The resulting ranges shall not overlap with any input range.
+ *
+ *  The following code snippet demonstrates how to use \p set_symmetric_difference_by_key to compute the
+ *  symmetric difference of two sets of integers sorted in ascending order with their values using the
+ *  \p thrust::host execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/set_operations.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  int A_keys[6] = {0, 2, 4, 6, 8, 10, 12};
+ *  int A_vals[6] = {0, 0, 0, 0, 0,  0,  0};
+ *
+ *  int B_keys[5] = {1, 3, 5, 7, 9};
+ *  int B_vals[5] = {1, 1, 1, 1, 1};
+ *
+ *  int keys_result[11];
+ *  int vals_result[11];
+ *
+ *  thrust::pair<int*,int*> end = thrust::set_symmetric_difference_by_key(thrust::host, A_keys, A_keys + 6, B_keys, B_keys + 5, A_vals, B_vals, keys_result, vals_result);
+ *  // keys_result is now {0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 12}
+ *  // vals_result is now {0, 1, 0, 1, 0, 1, 0, 1, 0, 1,  0,  0}
+ *  \endcode
+ *
+ *  \see \p set_symmetric_difference_by_key
+ *  \see \p set_intersection_by_key
+ *  \see \p set_difference_by_key
+ *  \see \p sort_by_key
+ *  \see \p is_sorted
+ */
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename InputIterator4,
+         typename OutputIterator1,
+         typename OutputIterator2>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    set_union_by_key(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                     InputIterator1                                              keys_first1,
+                     InputIterator1                                              keys_last1,
+                     InputIterator2                                              keys_first2,
+                     InputIterator2                                              keys_last2,
+                     InputIterator3                                              values_first1,
+                     InputIterator4                                              values_first2,
+                     OutputIterator1                                             keys_result,
+                     OutputIterator2                                             values_result);
+
+
+/*! \p set_union_by_key performs a key-value union operation from set theory.
+ *  \p set_union_by_key constructs a sorted range that is the union of the sorted
+ *  ranges <tt>[keys_first1, keys_last1)</tt> and <tt>[keys_first2, keys_last2)</tt>. Associated
+ *  with each element from the input and output key ranges is a value element. The associated input
+ *  value ranges need not be sorted.
+ *
+ *  In the simplest case, \p set_union_by_key performs the "union" operation from set theory:
+ *  the output range contains a copy of every element that is contained in
+ *  <tt>[keys_first1, keys_last1)</tt>, <tt>[keys_first2, keys_last1)</tt>, or both. The general case
+ *  is more complicated, because the input ranges may contain duplicate elements.
+ *  The generalization is that if <tt>[keys_first1, keys_last1)</tt> contains \c m elements
+ *  that are equivalent to each other and if <tt>[keys_first2, keys_last2)</tt> contains \c n
+ *  elements that are equivalent to them, then all \c m elements from the first
+ *  range shall be copied to the output range, in order, and then <tt>max(n - m, 0)</tt>
+ *  elements from the second range shall be copied to the output, in order.
+ *
+ *  Each time a key element is copied from <tt>[keys_first1, keys_last1)</tt> or
+ *  <tt>[keys_first2, keys_last2)</tt> is copied to the keys output range, the
+ *  corresponding value element is copied from the corresponding values input range (beginning at
+ *  \p values_first1 or \p values_first2) to the values output range.
+ *
+ *  This version of \p set_union_by_key compares key elements using \c operator<.
+ *
+ *  \param keys_first1 The beginning of the first input range of keys.
+ *  \param keys_last1 The end of the first input range of keys.
+ *  \param keys_first2 The beginning of the second input range of keys.
+ *  \param keys_last2 The end of the second input range of keys.
+ *  \param values_first1 The beginning of the first input range of values.
+ *  \param values_first2 The beginning of the first input range of values.
+ *  \param keys_result The beginning of the output range of keys.
+ *  \param values_result The beginning of the output range of values.
+ *  \return A \p pair \c p such that <tt>p.first</tt> is the end of the output range of keys,
+ *          and such that <tt>p.second</tt> is the end of the output range of values.
+ *
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator1 and \p InputIterator2 have the same \c value_type,
+ *          \p InputIterator1's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a>,
+ *          the ordering on \p InputIterator1's \c value_type is a strict weak ordering, as defined in the <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a> requirements,
+ *          and \p InputIterator1's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator2 and \p InputIterator1 have the same \c value_type,
+ *          \p InputIterator2's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a>,
+ *          the ordering on \p InputIterator2's \c value_type is a strict weak ordering, as defined in the <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a> requirements,
+ *          and \p InputIterator2's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam InputIterator3 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator3's \c value_type is convertible to a type in \p OutputIterator2's set of \c value_types.
+ *  \tparam InputIterator4 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator4's \c value_type is convertible to a type in \p OutputIterator2's set of \c value_types.
+ *  \tparam OutputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam OutputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *
+ *  \pre The ranges <tt>[keys_first1, keys_last1)</tt> and <tt>[keys_first2, keys_last2)</tt> shall be sorted with respect to <tt>operator<</tt>.
+ *  \pre The resulting ranges shall not overlap with any input range.
+ *
+ *  The following code snippet demonstrates how to use \p set_symmetric_difference_by_key to compute the
+ *  symmetric difference of two sets of integers sorted in ascending order with their values.
+ *
+ *  \code
+ *  #include <thrust/set_operations.h>
+ *  ...
+ *  int A_keys[6] = {0, 2, 4, 6, 8, 10, 12};
+ *  int A_vals[6] = {0, 0, 0, 0, 0,  0,  0};
+ *
+ *  int B_keys[5] = {1, 3, 5, 7, 9};
+ *  int B_vals[5] = {1, 1, 1, 1, 1};
+ *
+ *  int keys_result[11];
+ *  int vals_result[11];
+ *
+ *  thrust::pair<int*,int*> end = thrust::set_symmetric_difference_by_key(A_keys, A_keys + 6, B_keys, B_keys + 5, A_vals, B_vals, keys_result, vals_result);
+ *  // keys_result is now {0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 12}
+ *  // vals_result is now {0, 1, 0, 1, 0, 1, 0, 1, 0, 1,  0,  0}
+ *  \endcode
+ *
+ *  \see \p set_symmetric_difference_by_key
+ *  \see \p set_intersection_by_key
+ *  \see \p set_difference_by_key
+ *  \see \p sort_by_key
+ *  \see \p is_sorted
+ */
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename InputIterator4,
+         typename OutputIterator1,
+         typename OutputIterator2>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    set_union_by_key(InputIterator1                             keys_first1,
+                     InputIterator1                             keys_last1,
+                     InputIterator2                             keys_first2,
+                     InputIterator2                             keys_last2,
+                     InputIterator3                             values_first1,
+                     InputIterator4                             values_first2,
+                     OutputIterator1                            keys_result,
+                     OutputIterator2                            values_result);
+
+
+/*! \p set_union_by_key performs a key-value union operation from set theory.
+ *  \p set_union_by_key constructs a sorted range that is the union of the sorted
+ *  ranges <tt>[keys_first1, keys_last1)</tt> and <tt>[keys_first2, keys_last2)</tt>. Associated
+ *  with each element from the input and output key ranges is a value element. The associated input
+ *  value ranges need not be sorted.
+ *
+ *  In the simplest case, \p set_union_by_key performs the "union" operation from set theory:
+ *  the output range contains a copy of every element that is contained in
+ *  <tt>[keys_first1, keys_last1)</tt>, <tt>[keys_first2, keys_last1)</tt>, or both. The general case
+ *  is more complicated, because the input ranges may contain duplicate elements.
+ *  The generalization is that if <tt>[keys_first1, keys_last1)</tt> contains \c m elements
+ *  that are equivalent to each other and if <tt>[keys_first2, keys_last2)</tt> contains \c n
+ *  elements that are equivalent to them, then all \c m elements from the first
+ *  range shall be copied to the output range, in order, and then <tt>max(n - m, 0)</tt>
+ *  elements from the second range shall be copied to the output, in order.
+ *
+ *  Each time a key element is copied from <tt>[keys_first1, keys_last1)</tt> or
+ *  <tt>[keys_first2, keys_last2)</tt> is copied to the keys output range, the
+ *  corresponding value element is copied from the corresponding values input range (beginning at
+ *  \p values_first1 or \p values_first2) to the values output range.
+ *
+ *  This version of \p set_union_by_key compares key elements using a function object \c comp.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param keys_first1 The beginning of the first input range of keys.
+ *  \param keys_last1 The end of the first input range of keys.
+ *  \param keys_first2 The beginning of the second input range of keys.
+ *  \param keys_last2 The end of the second input range of keys.
+ *  \param values_first1 The beginning of the first input range of values.
+ *  \param values_first2 The beginning of the first input range of values.
+ *  \param keys_result The beginning of the output range of keys.
+ *  \param values_result The beginning of the output range of values.
+ *  \param comp Comparison operator.
+ *  \return A \p pair \c p such that <tt>p.first</tt> is the end of the output range of keys,
+ *          and such that <tt>p.second</tt> is the end of the output range of values.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator1 and \p InputIterator2 have the same \c value_type,
+ *          \p InputIterator1's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a>,
+ *          the ordering on \p InputIterator1's \c value_type is a strict weak ordering, as defined in the <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a> requirements,
+ *          and \p InputIterator1's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator2 and \p InputIterator1 have the same \c value_type,
+ *          \p InputIterator2's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a>,
+ *          the ordering on \p InputIterator2's \c value_type is a strict weak ordering, as defined in the <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a> requirements,
+ *          and \p InputIterator2's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam InputIterator3 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator3's \c value_type is convertible to a type in \p OutputIterator2's set of \c value_types.
+ *  \tparam InputIterator4 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator4's \c value_type is convertible to a type in \p OutputIterator2's set of \c value_types.
+ *  \tparam OutputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam OutputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam StrictWeakCompare is a model of <a href="http://www.sgi.com/tech/stl/StrictWeakOrdering.html">Strict Weak Ordering</a>.
+ *
+ *  \pre The ranges <tt>[keys_first1, keys_last1)</tt> and <tt>[keys_first2, keys_last2)</tt> shall be sorted with respect to \p comp.
+ *  \pre The resulting ranges shall not overlap with any input range.
+ *
+ *  The following code snippet demonstrates how to use \p set_symmetric_difference_by_key to compute the
+ *  symmetric difference of two sets of integers sorted in descending order with their values using the
+ *  \p thrust::host execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/set_operations.h>
+ *  #include <thrust/functional.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  int A_keys[6] = {12, 10, 8, 6, 4, 2, 0};
+ *  int A_vals[6] = { 0,  0, 0, 0, 0, 0, 0};
+ *
+ *  int B_keys[5] = {9, 7, 5, 3, 1};
+ *  int B_vals[5] = {1, 1, 1, 1, 1};
+ *
+ *  int keys_result[11];
+ *  int vals_result[11];
+ *
+ *  thrust::pair<int*,int*> end = thrust::set_symmetric_difference_by_key(thrust::host, A_keys, A_keys + 6, B_keys, B_keys + 5, A_vals, B_vals, keys_result, vals_result, thrust::greater<int>());
+ *  // keys_result is now {12, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0}
+ *  // vals_result is now { 0,  1, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0}
+ *  \endcode
+ *
+ *  \see \p set_symmetric_difference_by_key
+ *  \see \p set_intersection_by_key
+ *  \see \p set_difference_by_key
+ *  \see \p sort_by_key
+ *  \see \p is_sorted
+ */
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename InputIterator4,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename StrictWeakCompare>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    set_union_by_key(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                     InputIterator1                                              keys_first1,
+                     InputIterator1                                              keys_last1,
+                     InputIterator2                                              keys_first2,
+                     InputIterator2                                              keys_last2,
+                     InputIterator3                                              values_first1,
+                     InputIterator4                                              values_first2,
+                     OutputIterator1                                             keys_result,
+                     OutputIterator2                                             values_result,
+                     StrictWeakCompare                                           comp);
+
+
+/*! \p set_union_by_key performs a key-value union operation from set theory.
+ *  \p set_union_by_key constructs a sorted range that is the union of the sorted
+ *  ranges <tt>[keys_first1, keys_last1)</tt> and <tt>[keys_first2, keys_last2)</tt>. Associated
+ *  with each element from the input and output key ranges is a value element. The associated input
+ *  value ranges need not be sorted.
+ *
+ *  In the simplest case, \p set_union_by_key performs the "union" operation from set theory:
+ *  the output range contains a copy of every element that is contained in
+ *  <tt>[keys_first1, keys_last1)</tt>, <tt>[keys_first2, keys_last1)</tt>, or both. The general case
+ *  is more complicated, because the input ranges may contain duplicate elements.
+ *  The generalization is that if <tt>[keys_first1, keys_last1)</tt> contains \c m elements
+ *  that are equivalent to each other and if <tt>[keys_first2, keys_last2)</tt> contains \c n
+ *  elements that are equivalent to them, then all \c m elements from the first
+ *  range shall be copied to the output range, in order, and then <tt>max(n - m, 0)</tt>
+ *  elements from the second range shall be copied to the output, in order.
+ *
+ *  Each time a key element is copied from <tt>[keys_first1, keys_last1)</tt> or
+ *  <tt>[keys_first2, keys_last2)</tt> is copied to the keys output range, the
+ *  corresponding value element is copied from the corresponding values input range (beginning at
+ *  \p values_first1 or \p values_first2) to the values output range.
+ *
+ *  This version of \p set_union_by_key compares key elements using a function object \c comp.
+ *
+ *  \param keys_first1 The beginning of the first input range of keys.
+ *  \param keys_last1 The end of the first input range of keys.
+ *  \param keys_first2 The beginning of the second input range of keys.
+ *  \param keys_last2 The end of the second input range of keys.
+ *  \param values_first1 The beginning of the first input range of values.
+ *  \param values_first2 The beginning of the first input range of values.
+ *  \param keys_result The beginning of the output range of keys.
+ *  \param values_result The beginning of the output range of values.
+ *  \param comp Comparison operator.
+ *  \return A \p pair \c p such that <tt>p.first</tt> is the end of the output range of keys,
+ *          and such that <tt>p.second</tt> is the end of the output range of values.
+ *
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator1 and \p InputIterator2 have the same \c value_type,
+ *          \p InputIterator1's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a>,
+ *          the ordering on \p InputIterator1's \c value_type is a strict weak ordering, as defined in the <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a> requirements,
+ *          and \p InputIterator1's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          \p InputIterator2 and \p InputIterator1 have the same \c value_type,
+ *          \p InputIterator2's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a>,
+ *          the ordering on \p InputIterator2's \c value_type is a strict weak ordering, as defined in the <a href="http://www.sgi.com/tech/stl/LessThanComparable">LessThan Comparable</a> requirements,
+ *          and \p InputIterator2's \c value_type is convertable to a type in \p OutputIterator's set of \c value_types.
+ *  \tparam InputIterator3 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator3's \c value_type is convertible to a type in \p OutputIterator2's set of \c value_types.
+ *  \tparam InputIterator4 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator4's \c value_type is convertible to a type in \p OutputIterator2's set of \c value_types.
+ *  \tparam OutputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam OutputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam StrictWeakCompare is a model of <a href="http://www.sgi.com/tech/stl/StrictWeakOrdering.html">Strict Weak Ordering</a>.
+ *
+ *  \pre The ranges <tt>[keys_first1, keys_last1)</tt> and <tt>[keys_first2, keys_last2)</tt> shall be sorted with respect to \p comp.
+ *  \pre The resulting ranges shall not overlap with any input range.
+ *
+ *  The following code snippet demonstrates how to use \p set_symmetric_difference_by_key to compute the
+ *  symmetric difference of two sets of integers sorted in descending order with their values.
+ *
+ *  \code
+ *  #include <thrust/set_operations.h>
+ *  #include <thrust/functional.h>
+ *  ...
+ *  int A_keys[6] = {12, 10, 8, 6, 4, 2, 0};
+ *  int A_vals[6] = { 0,  0, 0, 0, 0, 0, 0};
+ *
+ *  int B_keys[5] = {9, 7, 5, 3, 1};
+ *  int B_vals[5] = {1, 1, 1, 1, 1};
+ *
+ *  int keys_result[11];
+ *  int vals_result[11];
+ *
+ *  thrust::pair<int*,int*> end = thrust::set_symmetric_difference_by_key(A_keys, A_keys + 6, B_keys, B_keys + 5, A_vals, B_vals, keys_result, vals_result, thrust::greater<int>());
+ *  // keys_result is now {12, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0}
+ *  // vals_result is now { 0,  1, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0}
+ *  \endcode
+ *
+ *  \see \p set_symmetric_difference_by_key
+ *  \see \p set_intersection_by_key
+ *  \see \p set_difference_by_key
+ *  \see \p sort_by_key
+ *  \see \p is_sorted
+ */
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename InputIterator4,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename StrictWeakCompare>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    set_union_by_key(InputIterator1                             keys_first1,
+                     InputIterator1                             keys_last1,
+                     InputIterator2                             keys_first2,
+                     InputIterator2                             keys_last2,
+                     InputIterator3                             values_first1,
+                     InputIterator4                             values_first2,
+                     OutputIterator1                            keys_result,
+                     OutputIterator2                            values_result,
+                     StrictWeakCompare                          comp);
+
+
+/*! \} // end set_operations
+ */
+
+
+} // end thrust
+
+#include <thrust/detail/set_operations.inl>
+
diff --git a/compat/thrust/sort.h b/compat/thrust/sort.h
new file mode 100644
index 0000000000..e8edfcd876
--- /dev/null
+++ b/compat/thrust/sort.h
@@ -0,0 +1,1349 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file sort.h
+ *  \brief Functions for reorganizing ranges into sorted order
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/execution_policy.h>
+
+namespace thrust
+{
+
+
+/*! \addtogroup sorting
+ *  \ingroup algorithms
+ *  \{
+ */
+
+
+/*! \p sort sorts the elements in <tt>[first, last)</tt> into
+ *  ascending order, meaning that if \c i and \c j are any two valid
+ *  iterators in <tt>[first, last)</tt> such that \c i precedes \c j,
+ *  then \c *j is not less than \c *i. Note: \c sort is not guaranteed
+ *  to be stable. That is, suppose that \c *i and \c *j are equivalent:
+ *  neither one is less than the other. It is not guaranteed that the
+ *  relative order of these two elements will be preserved by \p sort.
+ *
+ *  This version of \p sort compares objects using \c operator<.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the sequence.
+ *  \param last The end of the sequence.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam RandomAccessIterator is a model of <a href="http://www.sgi.com/tech/stl/RandomAccessIterator.html">Random Access Iterator</a>,
+ *          \p RandomAccessIterator is mutable,
+ *          and \p RandomAccessIterator's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable.html">LessThan Comparable</a>,
+ *          and the ordering relation on \p RandomAccessIterator's \c value_type is a <em>strict weak ordering</em>, as defined in the
+ *          <a href="http://www.sgi.com/tech/stl/LessThanComparable.html">LessThan Comparable</a> requirements.
+ *
+ *  The following code snippet demonstrates how to use \p sort to sort
+ *  a sequence of integers using the \p thrust::host execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/sort.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  const int N = 6;
+ *  int A[N] = {1, 4, 2, 8, 5, 7};
+ *  thrust::sort(thrust::host, A, A + N);
+ *  // A is now {1, 2, 4, 5, 7, 8}
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/sort.html
+ *  \see \p stable_sort
+ *  \see \p sort_by_key
+ */
+template<typename DerivedPolicy, typename RandomAccessIterator>
+  void sort(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+            RandomAccessIterator first,
+            RandomAccessIterator last);
+
+
+/*! \p sort sorts the elements in <tt>[first, last)</tt> into
+ *  ascending order, meaning that if \c i and \c j are any two valid
+ *  iterators in <tt>[first, last)</tt> such that \c i precedes \c j,
+ *  then \c *j is not less than \c *i. Note: \c sort is not guaranteed
+ *  to be stable. That is, suppose that \c *i and \c *j are equivalent:
+ *  neither one is less than the other. It is not guaranteed that the
+ *  relative order of these two elements will be preserved by \p sort.
+ *
+ *  This version of \p sort compares objects using \c operator<.
+ *
+ *  \param first The beginning of the sequence.
+ *  \param last The end of the sequence.
+ *
+ *  \tparam RandomAccessIterator is a model of <a href="http://www.sgi.com/tech/stl/RandomAccessIterator.html">Random Access Iterator</a>,
+ *          \p RandomAccessIterator is mutable,
+ *          and \p RandomAccessIterator's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable.html">LessThan Comparable</a>,
+ *          and the ordering relation on \p RandomAccessIterator's \c value_type is a <em>strict weak ordering</em>, as defined in the
+ *          <a href="http://www.sgi.com/tech/stl/LessThanComparable.html">LessThan Comparable</a> requirements.
+ *
+ *  The following code snippet demonstrates how to use \p sort to sort
+ *  a sequence of integers.
+ *
+ *  \code
+ *  #include <thrust/sort.h>
+ *  ...
+ *  const int N = 6;
+ *  int A[N] = {1, 4, 2, 8, 5, 7};
+ *  thrust::sort(A, A + N);
+ *  // A is now {1, 2, 4, 5, 7, 8}
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/sort.html
+ *  \see \p stable_sort
+ *  \see \p sort_by_key
+ */
+template<typename RandomAccessIterator>
+  void sort(RandomAccessIterator first,
+            RandomAccessIterator last);
+
+
+/*! \p sort sorts the elements in <tt>[first, last)</tt> into
+ *  ascending order, meaning that if \c i and \c j are any two valid
+ *  iterators in <tt>[first, last)</tt> such that \c i precedes \c j,
+ *  then \c *j is not less than \c *i. Note: \c sort is not guaranteed
+ *  to be stable. That is, suppose that \c *i and \c *j are equivalent:
+ *  neither one is less than the other. It is not guaranteed that the
+ *  relative order of these two elements will be preserved by \p sort.
+ *
+ *  This version of \p sort compares objects using a function object
+ *  \p comp.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the sequence.
+ *  \param last The end of the sequence.
+ *  \param comp  Comparison operator.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam RandomAccessIterator is a model of <a href="http://www.sgi.com/tech/stl/RandomAccessIterator.html">Random Access Iterator</a>,
+ *          \p RandomAccessIterator is mutable,
+ *          and \p RandomAccessIterator's \c value_type is convertible to \p StrictWeakOrdering's
+ *          \c first_argument_type and \c second_argument_type.
+ *  \tparam StrictWeakOrdering is a model of <a href="http://www.sgi.com/tech/stl/StrictWeakOrdering.html">Strict Weak Ordering</a>.
+ *
+ *  The following code demonstrates how to sort integers in descending order
+ *  using the greater<int> comparison operator using the \p thrust::host execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/sort.h>
+ *  #include <thrust/functional.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  const int N = 6;
+ *  int A[N] = {1, 4, 2, 8, 5, 7};
+ *  thrust::sort(thrust::host, A, A + N, thrust::greater<int>());
+ *  // A is now {8, 7, 5, 4, 2, 1};
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/sort.html
+ *  \see \p stable_sort
+ *  \see \p sort_by_key
+ */
+template<typename DerivedPolicy,
+         typename RandomAccessIterator,
+         typename StrictWeakOrdering>
+  void sort(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+            RandomAccessIterator first,
+            RandomAccessIterator last,
+            StrictWeakOrdering comp);
+
+
+/*! \p sort sorts the elements in <tt>[first, last)</tt> into
+ *  ascending order, meaning that if \c i and \c j are any two valid
+ *  iterators in <tt>[first, last)</tt> such that \c i precedes \c j,
+ *  then \c *j is not less than \c *i. Note: \c sort is not guaranteed
+ *  to be stable. That is, suppose that \c *i and \c *j are equivalent:
+ *  neither one is less than the other. It is not guaranteed that the
+ *  relative order of these two elements will be preserved by \p sort.
+ *
+ *  This version of \p sort compares objects using a function object
+ *  \p comp.
+ *
+ *  \param first The beginning of the sequence.
+ *  \param last The end of the sequence.
+ *  \param comp  Comparison operator.
+ *
+ *  \tparam RandomAccessIterator is a model of <a href="http://www.sgi.com/tech/stl/RandomAccessIterator.html">Random Access Iterator</a>,
+ *          \p RandomAccessIterator is mutable,
+ *          and \p RandomAccessIterator's \c value_type is convertible to \p StrictWeakOrdering's
+ *          \c first_argument_type and \c second_argument_type.
+ *  \tparam StrictWeakOrdering is a model of <a href="http://www.sgi.com/tech/stl/StrictWeakOrdering.html">Strict Weak Ordering</a>.
+ *
+ *  The following code demonstrates how to sort integers in descending order
+ *  using the greater<int> comparison operator.
+ *
+ *  \code
+ *  #include <thrust/sort.h>
+ *  #include <thrust/functional.h>
+ *  ...
+ *  const int N = 6;
+ *  int A[N] = {1, 4, 2, 8, 5, 7};
+ *  thrust::sort(A, A + N, thrust::greater<int>());
+ *  // A is now {8, 7, 5, 4, 2, 1};
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/sort.html
+ *  \see \p stable_sort
+ *  \see \p sort_by_key
+ */
+template<typename RandomAccessIterator,
+         typename StrictWeakOrdering>
+  void sort(RandomAccessIterator first,
+            RandomAccessIterator last,
+            StrictWeakOrdering comp);
+
+
+/*! \p stable_sort is much like \c sort: it sorts the elements in
+ *  <tt>[first, last)</tt> into ascending order, meaning that if \c i
+ *  and \c j are any two valid iterators in <tt>[first, last)</tt> such
+ *  that \c i precedes \c j, then \c *j is not less than \c *i.
+ *
+ *  As the name suggests, \p stable_sort is stable: it preserves the
+ *  relative ordering of equivalent elements. That is, if \c x and \c y
+ *  are elements in <tt>[first, last)</tt> such that \c x precedes \c y,
+ *  and if the two elements are equivalent (neither <tt>x < y</tt> nor
+ *  <tt>y < x</tt>) then a postcondition of \p stable_sort is that \c x
+ *  still precedes \c y.
+ *
+ *  This version of \p stable_sort compares objects using \c operator<.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the sequence.
+ *  \param last The end of the sequence.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam RandomAccessIterator is a model of <a href="http://www.sgi.com/tech/stl/RandomAccessIterator.html">Random Access Iterator</a>,
+ *          \p RandomAccessIterator is mutable,
+ *          and \p RandomAccessIterator's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable.html">LessThan Comparable</a>,
+ *          and the ordering relation on \p RandomAccessIterator's \c value_type is a <em>strict weak ordering</em>, as defined in the
+ *          <a href="http://www.sgi.com/tech/stl/LessThanComparable.html">LessThan Comparable</a> requirements.
+ *
+ *  The following code snippet demonstrates how to use \p sort to sort
+ *  a sequence of integers using the \p thrust::host execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/sort.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  const int N = 6;
+ *  int A[N] = {1, 4, 2, 8, 5, 7};
+ *  thrust::stable_sort(thrust::host, A, A + N);
+ *  // A is now {1, 2, 4, 5, 7, 8}
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/stable_sort.html
+ *  \see \p sort
+ *  \see \p stable_sort_by_key
+ */
+template<typename DerivedPolicy, typename RandomAccessIterator>
+  void stable_sort(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                   RandomAccessIterator first,
+                   RandomAccessIterator last);
+
+
+/*! \p stable_sort is much like \c sort: it sorts the elements in
+ *  <tt>[first, last)</tt> into ascending order, meaning that if \c i
+ *  and \c j are any two valid iterators in <tt>[first, last)</tt> such
+ *  that \c i precedes \c j, then \c *j is not less than \c *i.
+ *
+ *  As the name suggests, \p stable_sort is stable: it preserves the
+ *  relative ordering of equivalent elements. That is, if \c x and \c y
+ *  are elements in <tt>[first, last)</tt> such that \c x precedes \c y,
+ *  and if the two elements are equivalent (neither <tt>x < y</tt> nor
+ *  <tt>y < x</tt>) then a postcondition of \p stable_sort is that \c x
+ *  still precedes \c y.
+ *
+ *  This version of \p stable_sort compares objects using \c operator<.
+ *
+ *  \param first The beginning of the sequence.
+ *  \param last The end of the sequence.
+ *
+ *  \tparam RandomAccessIterator is a model of <a href="http://www.sgi.com/tech/stl/RandomAccessIterator.html">Random Access Iterator</a>,
+ *          \p RandomAccessIterator is mutable,
+ *          and \p RandomAccessIterator's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable.html">LessThan Comparable</a>,
+ *          and the ordering relation on \p RandomAccessIterator's \c value_type is a <em>strict weak ordering</em>, as defined in the
+ *          <a href="http://www.sgi.com/tech/stl/LessThanComparable.html">LessThan Comparable</a> requirements.
+ *
+ *  The following code snippet demonstrates how to use \p sort to sort
+ *  a sequence of integers.
+ *
+ *  \code
+ *  #include <thrust/sort.h>
+ *  ...
+ *  const int N = 6;
+ *  int A[N] = {1, 4, 2, 8, 5, 7};
+ *  thrust::stable_sort(A, A + N);
+ *  // A is now {1, 2, 4, 5, 7, 8}
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/stable_sort.html
+ *  \see \p sort
+ *  \see \p stable_sort_by_key
+ */
+template<typename RandomAccessIterator>
+  void stable_sort(RandomAccessIterator first,
+                   RandomAccessIterator last);
+
+
+/*! \p stable_sort is much like \c sort: it sorts the elements in
+ *  <tt>[first, last)</tt> into ascending order, meaning that if \c i
+ *  and \c j are any two valid iterators in <tt>[first, last)</tt> such
+ *  that \c i precedes \c j, then \c *j is not less than \c *i.
+ *
+ *  As the name suggests, \p stable_sort is stable: it preserves the
+ *  relative ordering of equivalent elements. That is, if \c x and \c y
+ *  are elements in <tt>[first, last)</tt> such that \c x precedes \c y,
+ *  and if the two elements are equivalent (neither <tt>x < y</tt> nor
+ *  <tt>y < x</tt>) then a postcondition of \p stable_sort is that \c x
+ *  still precedes \c y.
+ *
+ *  This version of \p stable_sort compares objects using a function object
+ *  \p comp.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the sequence.
+ *  \param last The end of the sequence.
+ *  \param comp Comparison operator.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam RandomAccessIterator is a model of <a href="http://www.sgi.com/tech/stl/RandomAccessIterator.html">Random Access Iterator</a>,
+ *          \p RandomAccessIterator is mutable,
+ *          and \p RandomAccessIterator's \c value_type is convertible to \p StrictWeakOrdering's
+ *          \c first_argument_type and \c second_argument_type.
+ *  \tparam StrictWeakOrdering is a model of <a href="http://www.sgi.com/tech/stl/StrictWeakOrdering.html">Strict Weak Ordering</a>.
+ *
+ *  The following code demonstrates how to sort integers in descending order
+ *  using the greater<int> comparison operator using the \p thrust::host execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/sort.h>
+ *  #include <thrust/functional.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  const int N = 6;
+ *  int A[N] = {1, 4, 2, 8, 5, 7};
+ *  thrust::sort(A, A + N, thrust::greater<int>());
+ *  // A is now {8, 7, 5, 4, 2, 1};
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/stable_sort.html
+ *  \see \p sort
+ *  \see \p stable_sort_by_key
+ */
+template<typename DerivedPolicy,
+         typename RandomAccessIterator,
+         typename StrictWeakOrdering>
+  void stable_sort(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                   RandomAccessIterator first,
+                   RandomAccessIterator last,
+                   StrictWeakOrdering comp);
+
+
+/*! \p stable_sort is much like \c sort: it sorts the elements in
+ *  <tt>[first, last)</tt> into ascending order, meaning that if \c i
+ *  and \c j are any two valid iterators in <tt>[first, last)</tt> such
+ *  that \c i precedes \c j, then \c *j is not less than \c *i.
+ *
+ *  As the name suggests, \p stable_sort is stable: it preserves the
+ *  relative ordering of equivalent elements. That is, if \c x and \c y
+ *  are elements in <tt>[first, last)</tt> such that \c x precedes \c y,
+ *  and if the two elements are equivalent (neither <tt>x < y</tt> nor
+ *  <tt>y < x</tt>) then a postcondition of \p stable_sort is that \c x
+ *  still precedes \c y.
+ *
+ *  This version of \p stable_sort compares objects using a function object
+ *  \p comp.
+ *
+ *  \param first The beginning of the sequence.
+ *  \param last The end of the sequence.
+ *  \param comp Comparison operator.
+ *
+ *  \tparam RandomAccessIterator is a model of <a href="http://www.sgi.com/tech/stl/RandomAccessIterator.html">Random Access Iterator</a>,
+ *          \p RandomAccessIterator is mutable,
+ *          and \p RandomAccessIterator's \c value_type is convertible to \p StrictWeakOrdering's
+ *          \c first_argument_type and \c second_argument_type.
+ *  \tparam StrictWeakOrdering is a model of <a href="http://www.sgi.com/tech/stl/StrictWeakOrdering.html">Strict Weak Ordering</a>.
+ *
+ *  The following code demonstrates how to sort integers in descending order
+ *  using the greater<int> comparison operator.
+ *
+ *  \code
+ *  #include <thrust/sort.h>
+ *  #include <thrust/functional.h>
+ *  ...
+ *  const int N = 6;
+ *  int A[N] = {1, 4, 2, 8, 5, 7};
+ *  thrust::sort(A, A + N, thrust::greater<int>());
+ *  // A is now {8, 7, 5, 4, 2, 1};
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/stable_sort.html
+ *  \see \p sort
+ *  \see \p stable_sort_by_key
+ */
+template<typename RandomAccessIterator,
+         typename StrictWeakOrdering>
+  void stable_sort(RandomAccessIterator first,
+                   RandomAccessIterator last,
+                   StrictWeakOrdering comp);
+
+
+///////////////
+// Key Value //
+///////////////
+
+
+/*! \p sort_by_key performs a key-value sort. That is, \p sort_by_key sorts the
+ *  elements in <tt>[keys_first, keys_last)</tt> and <tt>[values_first,
+ *  values_first + (keys_last - keys_first))</tt> into ascending key order,
+ *  meaning that if \c i and \c j are any two valid iterators in <tt>[keys_first,
+ *  keys_last)</tt> such that \c i precedes \c j, and \c p and \c q are iterators
+ *  in <tt>[values_first, values_first + (keys_last - keys_first))</tt>
+ *  corresponding to \c i and \c j respectively, then \c *j is not less than
+ *  \c *i.
+ *
+ *  Note: \c sort_by_key is not guaranteed to be stable. That is, suppose that
+ *  \c *i and \c *j are equivalent: neither one is less than the other. It is not
+ *  guaranteed that the relative order of these two keys or the relative
+ *  order of their corresponding values will be preserved by \p sort_by_key.
+ *
+ *  This version of \p sort_by_key compares key objects using \c operator<.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param keys_first The beginning of the key sequence.
+ *  \param keys_last The end of the key sequence.
+ *  \param values_first The beginning of the value sequence.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam RandomAccessIterator1 is a model of <a href="http://www.sgi.com/tech/stl/RandomAccessIterator.html">Random Access Iterator</a>,
+ *          \p RandomAccessIterator1 is mutable,
+ *          and \p RandomAccessIterator1's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable.html">LessThan Comparable</a>,
+ *          and the ordering relation on \p RandomAccessIterator1's \c value_type is a <em>strict weak ordering</em>, as defined in the
+ *          <a href="http://www.sgi.com/tech/stl/LessThanComparable.html">LessThan Comparable</a> requirements.
+ *  \tparam RandomAccessIterator2 is a model of <a href="http://www.sgi.com/tech/stl/RandomAccessIterator.hml">Random Access Iterator</a>,
+ *          and \p RandomAccessIterator2 is mutable.
+ *
+ *  \pre The range <tt>[keys_first, keys_last))</tt> shall not overlap the range <tt>[values_first, values_first + (keys_last - keys_first))</tt>.
+ *
+ *  The following code snippet demonstrates how to use \p sort_by_key to sort
+ *  an array of character values using integers as sorting keys using the \p thrust::host execution policy
+ *  for parallelization:
+ *
+ *  \code
+ *  #include <thrust/sort.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  const int N = 6;
+ *  int    keys[N] = {  1,   4,   2,   8,   5,   7};
+ *  char values[N] = {'a', 'b', 'c', 'd', 'e', 'f'};
+ *  thrust::sort_by_key(thrust::host, keys, keys + N, values);
+ *  // keys is now   {  1,   2,   4,   5,   7,   8}
+ *  // values is now {'a', 'c', 'b', 'e', 'f', 'd'}
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/sort.html
+ *  \see \p stable_sort_by_key
+ *  \see \p sort
+ */
+template<typename DerivedPolicy,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2>
+  void sort_by_key(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                   RandomAccessIterator1 keys_first,
+                   RandomAccessIterator1 keys_last,
+                   RandomAccessIterator2 values_first);
+
+
+/*! \p sort_by_key performs a key-value sort. That is, \p sort_by_key sorts the
+ *  elements in <tt>[keys_first, keys_last)</tt> and <tt>[values_first,
+ *  values_first + (keys_last - keys_first))</tt> into ascending key order,
+ *  meaning that if \c i and \c j are any two valid iterators in <tt>[keys_first,
+ *  keys_last)</tt> such that \c i precedes \c j, and \c p and \c q are iterators
+ *  in <tt>[values_first, values_first + (keys_last - keys_first))</tt>
+ *  corresponding to \c i and \c j respectively, then \c *j is not less than
+ *  \c *i.
+ *
+ *  Note: \c sort_by_key is not guaranteed to be stable. That is, suppose that
+ *  \c *i and \c *j are equivalent: neither one is less than the other. It is not
+ *  guaranteed that the relative order of these two keys or the relative
+ *  order of their corresponding values will be preserved by \p sort_by_key.
+ *
+ *  This version of \p sort_by_key compares key objects using \c operator<.
+ *
+ *  \param keys_first The beginning of the key sequence.
+ *  \param keys_last The end of the key sequence.
+ *  \param values_first The beginning of the value sequence.
+ *
+ *  \tparam RandomAccessIterator1 is a model of <a href="http://www.sgi.com/tech/stl/RandomAccessIterator.html">Random Access Iterator</a>,
+ *          \p RandomAccessIterator1 is mutable,
+ *          and \p RandomAccessIterator1's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable.html">LessThan Comparable</a>,
+ *          and the ordering relation on \p RandomAccessIterator1's \c value_type is a <em>strict weak ordering</em>, as defined in the
+ *          <a href="http://www.sgi.com/tech/stl/LessThanComparable.html">LessThan Comparable</a> requirements.
+ *  \tparam RandomAccessIterator2 is a model of <a href="http://www.sgi.com/tech/stl/RandomAccessIterator.hml">Random Access Iterator</a>,
+ *          and \p RandomAccessIterator2 is mutable.
+ *
+ *  \pre The range <tt>[keys_first, keys_last))</tt> shall not overlap the range <tt>[values_first, values_first + (keys_last - keys_first))</tt>.
+ *
+ *  The following code snippet demonstrates how to use \p sort_by_key to sort
+ *  an array of character values using integers as sorting keys.
+ *
+ *  \code
+ *  #include <thrust/sort.h>
+ *  ...
+ *  const int N = 6;
+ *  int    keys[N] = {  1,   4,   2,   8,   5,   7};
+ *  char values[N] = {'a', 'b', 'c', 'd', 'e', 'f'};
+ *  thrust::sort_by_key(keys, keys + N, values);
+ *  // keys is now   {  1,   2,   4,   5,   7,   8}
+ *  // values is now {'a', 'c', 'b', 'e', 'f', 'd'}
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/sort.html
+ *  \see \p stable_sort_by_key
+ *  \see \p sort
+ */
+template<typename RandomAccessIterator1,
+         typename RandomAccessIterator2>
+  void sort_by_key(RandomAccessIterator1 keys_first,
+                   RandomAccessIterator1 keys_last,
+                   RandomAccessIterator2 values_first);
+
+
+/*! \p sort_by_key performs a key-value sort. That is, \p sort_by_key sorts the
+ *  elements in <tt>[keys_first, keys_last)</tt> and <tt>[values_first,
+ *  values_first + (keys_last - keys_first))</tt> into ascending key order,
+ *  meaning that if \c i and \c j are any two valid iterators in <tt>[keys_first,
+ *  keys_last)</tt> such that \c i precedes \c j, and \c p and \c q are iterators
+ *  in <tt>[values_first, values_first + (keys_last - keys_first))</tt>
+ *  corresponding to \c i and \c j respectively, then \c *j is not less than
+ *  \c *i.
+ *
+ *  Note: \c sort_by_key is not guaranteed to be stable. That is, suppose that
+ *  \c *i and \c *j are equivalent: neither one is less than the other. It is not
+ *  guaranteed that the relative order of these two keys or the relative
+ *  order of their corresponding values will be preserved by \p sort_by_key.
+ *
+ *  This version of \p sort_by_key compares key objects using a function object
+ *  \c comp.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param keys_first The beginning of the key sequence.
+ *  \param keys_last The end of the key sequence.
+ *  \param values_first The beginning of the value sequence.
+ *  \param comp Comparison operator.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam RandomAccessIterator1 is a model of <a href="http://www.sgi.com/tech/stl/RandomAccessIterator.html">Random Access Iterator</a>,
+ *          \p RandomAccessIterator1 is mutable,
+ *          and \p RandomAccessIterator1's \c value_type is convertible to \p StrictWeakOrdering's
+ *          \c first_argument_type and \c second_argument_type.
+ *  \tparam RandomAccessIterator2 is a model of <a href="http://www.sgi.com/tech/stl/RandomAccessIterator.hml">Random Access Iterator</a>,
+ *          and \p RandomAccessIterator2 is mutable.
+ *  \tparam StrictWeakOrdering is a model of <a href="http://www.sgi.com/tech/stl/StrictWeakOrdering.html">Strict Weak Ordering</a>.
+ *
+ *  \pre The range <tt>[keys_first, keys_last))</tt> shall not overlap the range <tt>[values_first, values_first + (keys_last - keys_first))</tt>.
+ *
+ *  The following code snippet demonstrates how to use \p sort_by_key to sort
+ *  an array of character values using integers as sorting keys using the \p thrust::host execution policy
+ *  for parallelization.The keys are sorted in descending order using the <tt>greater<int></tt> comparison operator.
+ *
+ *  \code
+ *  #include <thrust/sort.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  const int N = 6;
+ *  int    keys[N] = {  1,   4,   2,   8,   5,   7};
+ *  char values[N] = {'a', 'b', 'c', 'd', 'e', 'f'};
+ *  thrust::sort_by_key(thrust::host, keys, keys + N, values, thrust::greater<int>());
+ *  // keys is now   {  8,   7,   5,   4,   2,   1}
+ *  // values is now {'d', 'f', 'e', 'b', 'c', 'a'}
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/sort.html
+ *  \see \p stable_sort_by_key
+ *  \see \p sort
+ */
+template<typename DerivedPolicy,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2,
+         typename StrictWeakOrdering>
+  void sort_by_key(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                   RandomAccessIterator1 keys_first,
+                   RandomAccessIterator1 keys_last,
+                   RandomAccessIterator2 values_first,
+                   StrictWeakOrdering comp);
+
+
+/*! \p sort_by_key performs a key-value sort. That is, \p sort_by_key sorts the
+ *  elements in <tt>[keys_first, keys_last)</tt> and <tt>[values_first,
+ *  values_first + (keys_last - keys_first))</tt> into ascending key order,
+ *  meaning that if \c i and \c j are any two valid iterators in <tt>[keys_first,
+ *  keys_last)</tt> such that \c i precedes \c j, and \c p and \c q are iterators
+ *  in <tt>[values_first, values_first + (keys_last - keys_first))</tt>
+ *  corresponding to \c i and \c j respectively, then \c *j is not less than
+ *  \c *i.
+ *
+ *  Note: \c sort_by_key is not guaranteed to be stable. That is, suppose that
+ *  \c *i and \c *j are equivalent: neither one is less than the other. It is not
+ *  guaranteed that the relative order of these two keys or the relative
+ *  order of their corresponding values will be preserved by \p sort_by_key.
+ *
+ *  This version of \p sort_by_key compares key objects using a function object
+ *  \c comp.
+ *
+ *  \param keys_first The beginning of the key sequence.
+ *  \param keys_last The end of the key sequence.
+ *  \param values_first The beginning of the value sequence.
+ *  \param comp Comparison operator.
+ *
+ *  \tparam RandomAccessIterator1 is a model of <a href="http://www.sgi.com/tech/stl/RandomAccessIterator.html">Random Access Iterator</a>,
+ *          \p RandomAccessIterator1 is mutable,
+ *          and \p RandomAccessIterator1's \c value_type is convertible to \p StrictWeakOrdering's
+ *          \c first_argument_type and \c second_argument_type.
+ *  \tparam RandomAccessIterator2 is a model of <a href="http://www.sgi.com/tech/stl/RandomAccessIterator.hml">Random Access Iterator</a>,
+ *          and \p RandomAccessIterator2 is mutable.
+ *  \tparam StrictWeakOrdering is a model of <a href="http://www.sgi.com/tech/stl/StrictWeakOrdering.html">Strict Weak Ordering</a>.
+ *
+ *  \pre The range <tt>[keys_first, keys_last))</tt> shall not overlap the range <tt>[values_first, values_first + (keys_last - keys_first))</tt>.
+ *
+ *  The following code snippet demonstrates how to use \p sort_by_key to sort
+ *  an array of character values using integers as sorting keys.  The keys
+ *  are sorted in descending order using the greater<int> comparison operator.
+ *
+ *  \code
+ *  #include <thrust/sort.h>
+ *  ...
+ *  const int N = 6;
+ *  int    keys[N] = {  1,   4,   2,   8,   5,   7};
+ *  char values[N] = {'a', 'b', 'c', 'd', 'e', 'f'};
+ *  thrust::sort_by_key(keys, keys + N, values, thrust::greater<int>());
+ *  // keys is now   {  8,   7,   5,   4,   2,   1}
+ *  // values is now {'d', 'f', 'e', 'b', 'c', 'a'}
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/sort.html
+ *  \see \p stable_sort_by_key
+ *  \see \p sort
+ */
+template<typename RandomAccessIterator1,
+         typename RandomAccessIterator2,
+         typename StrictWeakOrdering>
+  void sort_by_key(RandomAccessIterator1 keys_first,
+                   RandomAccessIterator1 keys_last,
+                   RandomAccessIterator2 values_first,
+                   StrictWeakOrdering comp);
+
+
+/*! \p stable_sort_by_key performs a key-value sort. That is, \p stable_sort_by_key
+ *  sorts the elements in <tt>[keys_first, keys_last)</tt> and <tt>[values_first,
+ *  values_first + (keys_last - keys_first))</tt> into ascending key order,
+ *  meaning that if \c i and \c j are any two valid iterators in <tt>[keys_first,
+ *  keys_last)</tt> such that \c i precedes \c j, and \c p and \c q are iterators
+ *  in <tt>[values_first, values_first + (keys_last - keys_first))</tt>
+ *  corresponding to \c i and \c j respectively, then \c *j is not less than
+ *  \c *i.
+ *
+ *  As the name suggests, \p stable_sort_by_key is stable: it preserves the
+ *  relative ordering of equivalent elements. That is, if \c x and \c y
+ *  are elements in <tt>[keys_first, keys_last)</tt> such that \c x precedes \c y,
+ *  and if the two elements are equivalent (neither <tt>x < y</tt> nor
+ *  <tt>y < x</tt>) then a postcondition of \p stable_sort_by_key is that \c x
+ *  still precedes \c y.
+ *
+ *  This version of \p stable_sort_by_key compares key objects using \c operator<.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param keys_first The beginning of the key sequence.
+ *  \param keys_last The end of the key sequence.
+ *  \param values_first The beginning of the value sequence.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam RandomAccessIterator1 is a model of <a href="http://www.sgi.com/tech/stl/RandomAccessIterator.html">Random Access Iterator</a>,
+ *          \p RandomAccessIterator1 is mutable,
+ *          and \p RandomAccessIterator1's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable.html">LessThan Comparable</a>,
+ *          and the ordering relation on \p RandomAccessIterator1's \c value_type is a <em>strict weak ordering</em>, as defined in the
+ *          <a href="http://www.sgi.com/tech/stl/LessThanComparable.html">LessThan Comparable</a> requirements.
+ *  \tparam RandomAccessIterator2 is a model of <a href="http://www.sgi.com/tech/stl/RandomAccessIterator.hml">Random Access Iterator</a>,
+ *          and \p RandomAccessIterator2 is mutable.
+ *
+ *  \pre The range <tt>[keys_first, keys_last))</tt> shall not overlap the range <tt>[values_first, values_first + (keys_last - keys_first))</tt>.
+ *
+ *  The following code snippet demonstrates how to use \p stable_sort_by_key to sort
+ *  an array of characters using integers as sorting keys using the \p thrust::host execution policy for
+ *  parallelization:
+ *
+ *  \code
+ *  #include <thrust/sort.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  const int N = 6;
+ *  int    keys[N] = {  1,   4,   2,   8,   5,   7};
+ *  char values[N] = {'a', 'b', 'c', 'd', 'e', 'f'};
+ *  thrust::stable_sort_by_key(thrust::host, keys, keys + N, values);
+ *  // keys is now   {  1,   2,   4,   5,   7,   8}
+ *  // values is now {'a', 'c', 'b', 'e', 'f', 'd'}
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/sort.html
+ *  \see \p sort_by_key
+ *  \see \p stable_sort
+ */
+template<typename DerivedPolicy,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2>
+  void stable_sort_by_key(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                          RandomAccessIterator1 keys_first,
+                          RandomAccessIterator1 keys_last,
+                          RandomAccessIterator2 values_first);
+
+
+/*! \p stable_sort_by_key performs a key-value sort. That is, \p stable_sort_by_key
+ *  sorts the elements in <tt>[keys_first, keys_last)</tt> and <tt>[values_first,
+ *  values_first + (keys_last - keys_first))</tt> into ascending key order,
+ *  meaning that if \c i and \c j are any two valid iterators in <tt>[keys_first,
+ *  keys_last)</tt> such that \c i precedes \c j, and \c p and \c q are iterators
+ *  in <tt>[values_first, values_first + (keys_last - keys_first))</tt>
+ *  corresponding to \c i and \c j respectively, then \c *j is not less than
+ *  \c *i.
+ *
+ *  As the name suggests, \p stable_sort_by_key is stable: it preserves the
+ *  relative ordering of equivalent elements. That is, if \c x and \c y
+ *  are elements in <tt>[keys_first, keys_last)</tt> such that \c x precedes \c y,
+ *  and if the two elements are equivalent (neither <tt>x < y</tt> nor
+ *  <tt>y < x</tt>) then a postcondition of \p stable_sort_by_key is that \c x
+ *  still precedes \c y.
+ *
+ *  This version of \p stable_sort_by_key compares key objects using \c operator<.
+ *
+ *  \param keys_first The beginning of the key sequence.
+ *  \param keys_last The end of the key sequence.
+ *  \param values_first The beginning of the value sequence.
+ *
+ *  \tparam RandomAccessIterator1 is a model of <a href="http://www.sgi.com/tech/stl/RandomAccessIterator.html">Random Access Iterator</a>,
+ *          \p RandomAccessIterator1 is mutable,
+ *          and \p RandomAccessIterator1's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable.html">LessThan Comparable</a>,
+ *          and the ordering relation on \p RandomAccessIterator1's \c value_type is a <em>strict weak ordering</em>, as defined in the
+ *          <a href="http://www.sgi.com/tech/stl/LessThanComparable.html">LessThan Comparable</a> requirements.
+ *  \tparam RandomAccessIterator2 is a model of <a href="http://www.sgi.com/tech/stl/RandomAccessIterator.hml">Random Access Iterator</a>,
+ *          and \p RandomAccessIterator2 is mutable.
+ *
+ *  \pre The range <tt>[keys_first, keys_last))</tt> shall not overlap the range <tt>[values_first, values_first + (keys_last - keys_first))</tt>.
+ *
+ *  The following code snippet demonstrates how to use \p stable_sort_by_key to sort
+ *  an array of characters using integers as sorting keys.
+ *
+ *  \code
+ *  #include <thrust/sort.h>
+ *  ...
+ *  const int N = 6;
+ *  int    keys[N] = {  1,   4,   2,   8,   5,   7};
+ *  char values[N] = {'a', 'b', 'c', 'd', 'e', 'f'};
+ *  thrust::stable_sort_by_key(keys, keys + N, values);
+ *  // keys is now   {  1,   2,   4,   5,   7,   8}
+ *  // values is now {'a', 'c', 'b', 'e', 'f', 'd'}
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/sort.html
+ *  \see \p sort_by_key
+ *  \see \p stable_sort
+ */
+template<typename RandomAccessIterator1,
+         typename RandomAccessIterator2>
+  void stable_sort_by_key(RandomAccessIterator1 keys_first,
+                          RandomAccessIterator1 keys_last,
+                          RandomAccessIterator2 values_first);
+
+
+/*! \p stable_sort_by_key performs a key-value sort. That is, \p stable_sort_by_key
+ *  sorts the elements in <tt>[keys_first, keys_last)</tt> and <tt>[values_first,
+ *  values_first + (keys_last - keys_first))</tt> into ascending key order,
+ *  meaning that if \c i and \c j are any two valid iterators in <tt>[keys_first,
+ *  keys_last)</tt> such that \c i precedes \c j, and \c p and \c q are iterators
+ *  in <tt>[values_first, values_first + (keys_last - keys_first))</tt>
+ *  corresponding to \c i and \c j respectively, then \c *j is not less than
+ *  \c *i.
+ *
+ *  As the name suggests, \p stable_sort_by_key is stable: it preserves the
+ *  relative ordering of equivalent elements. That is, if \c x and \c y
+ *  are elements in <tt>[keys_first, keys_last)</tt> such that \c x precedes \c y,
+ *  and if the two elements are equivalent (neither <tt>x < y</tt> nor
+ *  <tt>y < x</tt>) then a postcondition of \p stable_sort_by_key is that \c x
+ *  still precedes \c y.
+ *
+ *  This version of \p stable_sort_by_key compares key objects using the function
+ *  object \p comp.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param keys_first The beginning of the key sequence.
+ *  \param keys_last The end of the key sequence.
+ *  \param values_first The beginning of the value sequence.
+ *  \param comp Comparison operator.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam RandomAccessIterator1 is a model of <a href="http://www.sgi.com/tech/stl/RandomAccessIterator.html">Random Access Iterator</a>,
+ *          \p RandomAccessIterator1 is mutable,
+ *          and \p RandomAccessIterator1's \c value_type is convertible to \p StrictWeakOrdering's
+ *          \c first_argument_type and \c second_argument_type.
+ *  \tparam RandomAccessIterator2 is a model of <a href="http://www.sgi.com/tech/stl/RandomAccessIterator.hml">Random Access Iterator</a>,
+ *          and \p RandomAccessIterator2 is mutable.
+ *  \tparam StrictWeakOrdering is a model of <a href="http://www.sgi.com/tech/stl/StrictWeakOrdering.html">Strict Weak Ordering</a>.
+ *
+ *  \pre The range <tt>[keys_first, keys_last))</tt> shall not overlap the range <tt>[values_first, values_first + (keys_last - keys_first))</tt>.
+ *
+ *  The following code snippet demonstrates how to use \p sort_by_key to sort
+ *  an array of character values using integers as sorting keys using the \p thrust::host execution policy for
+ *  parallelization. The keys are sorted in descending order using the <tt>greater<int></tt> comparison operator.
+ *
+ *  \code
+ *  #include <thrust/sort.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  const int N = 6;
+ *  int    keys[N] = {  1,   4,   2,   8,   5,   7};
+ *  char values[N] = {'a', 'b', 'c', 'd', 'e', 'f'};
+ *  thrust::stable_sort_by_key(thrust::host, keys, keys + N, values, thrust::greater<int>());
+ *  // keys is now   {  8,   7,   5,   4,   2,   1}
+ *  // values is now {'d', 'f', 'e', 'b', 'c', 'a'}
+ *  \endcode
+ *
+ *
+ *  \see http://www.sgi.com/tech/stl/sort.html
+ *  \see \p sort_by_key
+ *  \see \p stable_sort
+ */
+template<typename DerivedPolicy,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2,
+         typename StrictWeakOrdering>
+  void stable_sort_by_key(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                          RandomAccessIterator1 keys_first,
+                          RandomAccessIterator1 keys_last,
+                          RandomAccessIterator2 values_first,
+                          StrictWeakOrdering comp);
+
+
+/*! \p stable_sort_by_key performs a key-value sort. That is, \p stable_sort_by_key
+ *  sorts the elements in <tt>[keys_first, keys_last)</tt> and <tt>[values_first,
+ *  values_first + (keys_last - keys_first))</tt> into ascending key order,
+ *  meaning that if \c i and \c j are any two valid iterators in <tt>[keys_first,
+ *  keys_last)</tt> such that \c i precedes \c j, and \c p and \c q are iterators
+ *  in <tt>[values_first, values_first + (keys_last - keys_first))</tt>
+ *  corresponding to \c i and \c j respectively, then \c *j is not less than
+ *  \c *i.
+ *
+ *  As the name suggests, \p stable_sort_by_key is stable: it preserves the
+ *  relative ordering of equivalent elements. That is, if \c x and \c y
+ *  are elements in <tt>[keys_first, keys_last)</tt> such that \c x precedes \c y,
+ *  and if the two elements are equivalent (neither <tt>x < y</tt> nor
+ *  <tt>y < x</tt>) then a postcondition of \p stable_sort_by_key is that \c x
+ *  still precedes \c y.
+ *
+ *  This version of \p stable_sort_by_key compares key objects using the function
+ *  object \p comp.
+ *
+ *  \param keys_first The beginning of the key sequence.
+ *  \param keys_last The end of the key sequence.
+ *  \param values_first The beginning of the value sequence.
+ *  \param comp Comparison operator.
+ *
+ *  \tparam RandomAccessIterator1 is a model of <a href="http://www.sgi.com/tech/stl/RandomAccessIterator.html">Random Access Iterator</a>,
+ *          \p RandomAccessIterator1 is mutable,
+ *          and \p RandomAccessIterator1's \c value_type is convertible to \p StrictWeakOrdering's
+ *          \c first_argument_type and \c second_argument_type.
+ *  \tparam RandomAccessIterator2 is a model of <a href="http://www.sgi.com/tech/stl/RandomAccessIterator.hml">Random Access Iterator</a>,
+ *          and \p RandomAccessIterator2 is mutable.
+ *  \tparam StrictWeakOrdering is a model of <a href="http://www.sgi.com/tech/stl/StrictWeakOrdering.html">Strict Weak Ordering</a>.
+ *
+ *  \pre The range <tt>[keys_first, keys_last))</tt> shall not overlap the range <tt>[values_first, values_first + (keys_last - keys_first))</tt>.
+ *
+ *  The following code snippet demonstrates how to use \p sort_by_key to sort
+ *  an array of character values using integers as sorting keys.  The keys
+ *  are sorted in descending order using the greater<int> comparison operator.
+ *
+ *  \code
+ *  #include <thrust/sort.h>
+ *  ...
+ *  const int N = 6;
+ *  int    keys[N] = {  1,   4,   2,   8,   5,   7};
+ *  char values[N] = {'a', 'b', 'c', 'd', 'e', 'f'};
+ *  thrust::stable_sort_by_key(keys, keys + N, values, thrust::greater<int>());
+ *  // keys is now   {  8,   7,   5,   4,   2,   1}
+ *  // values is now {'d', 'f', 'e', 'b', 'c', 'a'}
+ *  \endcode
+ *
+ *
+ *  \see http://www.sgi.com/tech/stl/sort.html
+ *  \see \p sort_by_key
+ *  \see \p stable_sort
+ */
+template<typename RandomAccessIterator1,
+         typename RandomAccessIterator2,
+         typename StrictWeakOrdering>
+  void stable_sort_by_key(RandomAccessIterator1 keys_first,
+                          RandomAccessIterator1 keys_last,
+                          RandomAccessIterator2 values_first,
+                          StrictWeakOrdering comp);
+
+
+/*! \} // end sorting
+ */
+
+
+/*! \addtogroup reductions
+ *  \{
+ *  \addtogroup predicates
+ *  \{
+ */
+
+
+/*! \p is_sorted returns \c true if the range <tt>[first, last)</tt> is
+ *  sorted in ascending order, and \c false otherwise.
+ *
+ *  Specifically, this version of \p is_sorted returns \c false if for
+ *  some iterator \c i in the range <tt>[first, last - 1)</tt> the
+ *  expression <tt>*(i + 1) < *i</tt> is \c true.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the sequence.
+ *  \param last  The end of the sequence.
+ *  \return \c true, if the sequence is sorted; \c false, otherwise.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>,
+ *          \p ForwardIterator's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable.html">LessThan Comparable</a>,
+ *          and the ordering on objects of \p ForwardIterator's \c value_type is a <em>strict weak ordering</em>, as defined
+ *          in the <a href="http://www.sgi.com/tech/stl/LessThanComparable.html">LessThan Comparable</a> requirements.
+ *
+ *
+ *  The following code demonstrates how to use \p is_sorted to test whether the
+ *  contents of a \c device_vector are stored in ascending order using the \p thrust::device execution policy
+ *  for parallelization:
+ *
+ *  \code
+ *  #include <thrust/sort.h>
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/sort.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  thrust::device_vector<int> v(6);
+ *  v[0] = 1;
+ *  v[1] = 4;
+ *  v[2] = 2;
+ *  v[3] = 8;
+ *  v[4] = 5;
+ *  v[5] = 7;
+ *
+ *  bool result = thrust::is_sorted(thrust::device, v.begin(), v.end());
+ *
+ *  // result == false
+ *
+ *  thrust::sort(v.begin(), v.end());
+ *  result = thrust::is_sorted(thrust::device, v.begin(), v.end());
+ *
+ *  // result == true
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/is_sorted.html
+ *  \see is_sorted_until
+ *  \see \c sort
+ *  \see \c stable_sort
+ *  \see \c less<T>
+ */
+template<typename DerivedPolicy, typename ForwardIterator>
+  bool is_sorted(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                 ForwardIterator first,
+                 ForwardIterator last);
+
+
+/*! \p is_sorted returns \c true if the range <tt>[first, last)</tt> is
+ *  sorted in ascending order, and \c false otherwise.
+ *
+ *  Specifically, this version of \p is_sorted returns \c false if for
+ *  some iterator \c i in the range <tt>[first, last - 1)</tt> the
+ *  expression <tt>*(i + 1) < *i</tt> is \c true.
+ *
+ *  \param first The beginning of the sequence.
+ *  \param last  The end of the sequence.
+ *  \return \c true, if the sequence is sorted; \c false, otherwise.
+ *
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>,
+ *          \p ForwardIterator's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable.html">LessThan Comparable</a>,
+ *          and the ordering on objects of \p ForwardIterator's \c value_type is a <em>strict weak ordering</em>, as defined
+ *          in the <a href="http://www.sgi.com/tech/stl/LessThanComparable.html">LessThan Comparable</a> requirements.
+ *
+ *
+ *  The following code demonstrates how to use \p is_sorted to test whether the
+ *  contents of a \c device_vector are stored in ascending order.
+ *
+ *  \code
+ *  #include <thrust/sort.h>
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/sort.h>
+ *  ...
+ *  thrust::device_vector<int> v(6);
+ *  v[0] = 1;
+ *  v[1] = 4;
+ *  v[2] = 2;
+ *  v[3] = 8;
+ *  v[4] = 5;
+ *  v[5] = 7;
+ *
+ *  bool result = thrust::is_sorted(v.begin(), v.end());
+ *
+ *  // result == false
+ *
+ *  thrust::sort(v.begin(), v.end());
+ *  result = thrust::is_sorted(v.begin(), v.end());
+ *
+ *  // result == true
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/is_sorted.html
+ *  \see is_sorted_until
+ *  \see \c sort
+ *  \see \c stable_sort
+ *  \see \c less<T>
+ */
+template<typename ForwardIterator>
+  bool is_sorted(ForwardIterator first,
+                 ForwardIterator last);
+
+
+/*! \p is_sorted returns \c true if the range <tt>[first, last)</tt> is sorted in ascending 
+ *  order accoring to a user-defined comparison operation, and \c false otherwise.
+ *
+ *  Specifically, this version of \p is_sorted returns \c false if for some iterator \c i in
+ *  the range <tt>[first, last - 1)</tt> the expression <tt>comp(*(i + 1), *i)</tt> is \c true.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the sequence.
+ *  \param last  The end of the sequence.
+ *  \param comp  Comparison operator.
+ *  \return \c true, if the sequence is sorted according to comp; \c false, otherwise.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>,
+ *          and \p ForwardIterator's \c value_type is convertible to both \c StrictWeakOrdering's \c first_argument_type
+ *          and \c second_argument_type.
+ *  \tparam Compare is a model of <a href="http://www.sgi.com/tech/stl/StrictWeakOrdering.html">Strict Weak Ordering</a>.
+ *
+ *  The following code snippet demonstrates how to use \p is_sorted to test whether the
+ *  contents of a \c device_vector are stored in descending order using the \p thrust::device execution
+ *  policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/sort.h>
+ *  #include <thrust/functional.h>
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  thrust::device_vector<int> v(6);
+ *  v[0] = 1;
+ *  v[1] = 4;
+ *  v[2] = 2;
+ *  v[3] = 8;
+ *  v[4] = 5;
+ *  v[5] = 7;
+ *
+ *  thrust::greater<int> comp;
+ *  bool result = thrust::is_sorted(thrust::device, v.begin(), v.end(), comp);
+ *
+ *  // result == false
+ *
+ *  thrust::sort(v.begin(), v.end(), comp);
+ *  result = thrust::is_sorted(thrust::device, v.begin(), v.end(), comp);
+ *
+ *  // result == true
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/is_sorted.html
+ *  \see \c sort
+ *  \see \c stable_sort
+ *  \see \c less<T>
+ */
+template<typename DerivedPolicy, typename ForwardIterator, typename Compare>
+  bool is_sorted(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                 ForwardIterator first,
+                 ForwardIterator last,
+                 Compare comp);
+
+
+/*! \p is_sorted returns \c true if the range <tt>[first, last)</tt> is sorted in ascending 
+ *  order accoring to a user-defined comparison operation, and \c false otherwise.
+ *
+ *  Specifically, this version of \p is_sorted returns \c false if for some iterator \c i in
+ *  the range <tt>[first, last - 1)</tt> the expression <tt>comp(*(i + 1), *i)</tt> is \c true.
+ *
+ *  \param first The beginning of the sequence.
+ *  \param last  The end of the sequence.
+ *  \param comp  Comparison operator.
+ *  \return \c true, if the sequence is sorted according to comp; \c false, otherwise.
+ *
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>,
+ *          and \p ForwardIterator's \c value_type is convertible to both \c StrictWeakOrdering's \c first_argument_type
+ *          and \c second_argument_type.
+ *  \tparam Compare is a model of <a href="http://www.sgi.com/tech/stl/StrictWeakOrdering.html">Strict Weak Ordering</a>.
+ *
+ *  The following code snippet demonstrates how to use \p is_sorted to test whether the
+ *  contents of a \c device_vector are stored in descending order.
+ *
+ *  \code
+ *  #include <thrust/sort.h>
+ *  #include <thrust/functional.h>
+ *  #include <thrust/device_vector.h>
+ *  ...
+ *  thrust::device_vector<int> v(6);
+ *  v[0] = 1;
+ *  v[1] = 4;
+ *  v[2] = 2;
+ *  v[3] = 8;
+ *  v[4] = 5;
+ *  v[5] = 7;
+ *
+ *  thrust::greater<int> comp;
+ *  bool result = thrust::is_sorted(v.begin(), v.end(), comp);
+ *
+ *  // result == false
+ *
+ *  thrust::sort(v.begin(), v.end(), comp);
+ *  result = thrust::is_sorted(v.begin(), v.end(), comp);
+ *
+ *  // result == true
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/is_sorted.html
+ *  \see \c sort
+ *  \see \c stable_sort
+ *  \see \c less<T>
+ */
+template<typename ForwardIterator, typename Compare>
+  bool is_sorted(ForwardIterator first,
+                 ForwardIterator last,
+                 Compare comp);
+
+
+/*! This version of \p is_sorted_until returns the last iterator \c i in <tt>[first,last]</tt> for
+ *  which the range <tt>[first,last)</tt> is sorted using \c operator<. If <tt>distance(first,last) < 2</tt>,
+ *  \p is_sorted_until simply returns \p last.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the range of interest.
+ *  \param last The end of the range of interest.
+ *  \return The last iterator in the input range for which it is sorted.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a> and
+ *          \p ForwardIterator's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable.html">LessThan Comparable</a>.
+ *
+ *  The following code snippet demonstrates how to use \p is_sorted_until to find the first position
+ *  in an array where the data becomes unsorted using the \p thrust::host execution policy for
+ *  parallelization:
+ *  
+ *  \code
+ *  #include <thrust/sort.h>
+ *  #include <thrust/execution_policy.h>
+ *
+ *  ...
+ *   
+ *  int A[8] = {0, 1, 2, 3, 0, 1, 2, 3};
+ *  
+ *  int * B = thrust::is_sorted_until(thrust::host, A, A + 8);
+ *  
+ *  // B - A is 4
+ *  // [A, B) is sorted
+ *  \endcode
+ *
+ *  \see \p is_sorted
+ *  \see \p sort
+ *  \see \p sort_by_key
+ *  \see \p stable_sort
+ *  \see \p stable_sort_by_key
+ */
+template<typename DerivedPolicy, typename ForwardIterator>
+  ForwardIterator is_sorted_until(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                  ForwardIterator first,
+                                  ForwardIterator last);
+
+
+/*! This version of \p is_sorted_until returns the last iterator \c i in <tt>[first,last]</tt> for
+ *  which the range <tt>[first,last)</tt> is sorted using \c operator<. If <tt>distance(first,last) < 2</tt>,
+ *  \p is_sorted_until simply returns \p last.
+ *
+ *  \param first The beginning of the range of interest.
+ *  \param last The end of the range of interest.
+ *  \return The last iterator in the input range for which it is sorted.
+ *
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a> and
+ *          \p ForwardIterator's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/LessThanComparable.html">LessThan Comparable</a>.
+ *
+ *  The following code snippet demonstrates how to use \p is_sorted_until to find the first position
+ *  in an array where the data becomes unsorted:
+ *  
+ *  \code
+ *  #include <thrust/sort.h>
+ *
+ *  ...
+ *   
+ *  int A[8] = {0, 1, 2, 3, 0, 1, 2, 3};
+ *  
+ *  int * B = thrust::is_sorted_until(A, A + 8);
+ *  
+ *  // B - A is 4
+ *  // [A, B) is sorted
+ *  \endcode
+ *
+ *  \see \p is_sorted
+ *  \see \p sort
+ *  \see \p sort_by_key
+ *  \see \p stable_sort
+ *  \see \p stable_sort_by_key
+ */
+template<typename ForwardIterator>
+  ForwardIterator is_sorted_until(ForwardIterator first,
+                                  ForwardIterator last);
+
+
+/*! This version of \p is_sorted_until returns the last iterator \c i in <tt>[first,last]</tt> for
+ *  which the range <tt>[first,last)</tt> is sorted using the function object \c comp. If <tt>distance(first,last) < 2</tt>,
+ *  \p is_sorted_until simply returns \p last.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization:
+ *  \param first The beginning of the range of interest.
+ *  \param last The end of the range of interest.
+ *  \param comp The function object to use for comparison.
+ *  \return The last iterator in the input range for which it is sorted.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a> and
+ *          \p ForwardIterator's \c value_type is convertible to \p Compare's \c argument_type.
+ *  \tparam Compare is a model of <a href="http://www.sgi.com/tech/stl/StrictWeakOrdering.html">Strict Weak Ordering</a>.
+ *
+ *  The following code snippet demonstrates how to use \p is_sorted_until to find the first position
+ *  in an array where the data becomes unsorted in descending order using the \p thrust::host execution
+ *  policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/sort.h>
+ *  #include <thrust/functional.h>
+ *  #include <thrust/execution_policy.h>
+ *
+ *  ...
+ *   
+ *  int A[8] = {3, 2, 1, 0, 3, 2, 1, 0};
+ *  
+ *  thrust::greater<int> comp;
+ *  int * B = thrust::is_sorted_until(thrust::host, A, A + 8, comp);
+ *  
+ *  // B - A is 4
+ *  // [A, B) is sorted in descending order
+ *  \endcode
+ *
+ *  \see \p is_sorted
+ *  \see \p sort
+ *  \see \p sort_by_key
+ *  \see \p stable_sort
+ *  \see \p stable_sort_by_key
+ */
+template<typename DerivedPolicy, typename ForwardIterator, typename Compare>
+  ForwardIterator is_sorted_until(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                  ForwardIterator first,
+                                  ForwardIterator last,
+                                  Compare comp);
+
+
+/*! This version of \p is_sorted_until returns the last iterator \c i in <tt>[first,last]</tt> for
+ *  which the range <tt>[first,last)</tt> is sorted using the function object \c comp. If <tt>distance(first,last) < 2</tt>,
+ *  \p is_sorted_until simply returns \p last.
+ *
+ *  \param first The beginning of the range of interest.
+ *  \param last The end of the range of interest.
+ *  \param comp The function object to use for comparison.
+ *  \return The last iterator in the input range for which it is sorted.
+ *
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a> and
+ *          \p ForwardIterator's \c value_type is convertible to \p Compare's \c argument_type.
+ *  \tparam Compare is a model of <a href="http://www.sgi.com/tech/stl/StrictWeakOrdering.html">Strict Weak Ordering</a>.
+ *
+ *  The following code snippet demonstrates how to use \p is_sorted_until to find the first position
+ *  in an array where the data becomes unsorted in descending order:
+ *
+ *  \code
+ *  #include <thrust/sort.h>
+ *  #include <thrust/functional.h>
+ *
+ *  ...
+ *   
+ *  int A[8] = {3, 2, 1, 0, 3, 2, 1, 0};
+ *  
+ *  thrust::greater<int> comp;
+ *  int * B = thrust::is_sorted_until(A, A + 8, comp);
+ *  
+ *  // B - A is 4
+ *  // [A, B) is sorted in descending order
+ *  \endcode
+ *
+ *  \see \p is_sorted
+ *  \see \p sort
+ *  \see \p sort_by_key
+ *  \see \p stable_sort
+ *  \see \p stable_sort_by_key
+ */
+template<typename ForwardIterator, typename Compare>
+  ForwardIterator is_sorted_until(ForwardIterator first,
+                                  ForwardIterator last,
+                                  Compare comp);
+
+
+/*! \} // end predicates
+ *  \} // end reductions
+ */
+
+
+} // end namespace thrust
+
+#include <thrust/detail/sort.inl>
+
diff --git a/compat/thrust/swap.h b/compat/thrust/swap.h
new file mode 100644
index 0000000000..085e546930
--- /dev/null
+++ b/compat/thrust/swap.h
@@ -0,0 +1,190 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+/*! \file swap.h
+ *  \brief Functions for swapping the value of elements
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/execution_policy.h>
+
+// empty Doxygen comment below so namespace thrust's documentation will be extracted
+
+/*!
+ */
+namespace thrust
+{
+
+/*! \addtogroup utility
+ *  \{
+ */
+
+/*! \addtogroup swap
+ *  \{
+ */
+
+/*! \p swap assigns the contents of \c a to \c b and the
+ *  contents of \c b to \c a. This is used as a primitive operation
+ *  by many other algorithms.
+ *  
+ *  \param a The first value of interest. After completion,
+ *           the value of b will be returned here.
+ *  \param b The second value of interest. After completion,
+ *           the value of a will be returned here.
+ *
+ *  \tparam Assignable is a model of <a href="http://www.sgi.com/tech/stl/Assignable.html">Assignable</a>.
+ *
+ *  The following code snippet demonstrates how to use \p swap to
+ *  swap the contents of two variables.
+ *
+ *  \code
+ *  #include <thrust/swap.h>
+ *  ...
+ *  int x = 1;
+ *  int y = 2;
+ *  thrust::swap(x,h);
+ *
+ *  // x == 2, y == 1
+ *  \endcode
+ */
+template<typename Assignable1, typename Assignable2>
+__host__ __device__ 
+inline void swap(Assignable1 &a, Assignable2 &b);
+
+/*! \} // swap
+ */
+
+/*! \} // utility
+ */
+
+
+/*! \addtogroup copying
+ *  \{
+ */
+
+
+/*! \p swap_ranges swaps each of the elements in the range <tt>[first1, last1)</tt>
+ *  with the corresponding element in the range <tt>[first2, first2 + (last1 - first1))</tt>.
+ *  That is, for each integer \c n such that <tt>0 <= n < (last1 - first1)</tt>, it swaps
+ *  <tt>*(first1 + n)</tt> and <tt>*(first2 + n)</tt>. The return value is
+ *  <tt>first2 + (last1 - first1)</tt>.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first1 The beginning of the first sequence to swap.
+ *  \param last1 One position past the last element of the first sequence to swap.
+ *  \param first2 The beginning of the second sequence to swap.
+ *  \return An iterator pointing to one position past the last element of the second
+ *          sequence to swap.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam ForwardIterator1 is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>,
+ *          and \p ForwardIterator1's \c value_type must be convertible to \p ForwardIterator2's \c value_type.
+ *  \tparam ForwardIterator2 is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>,
+ *          and \p ForwardIterator2's \c value_type must be convertible to \p ForwardIterator1's \c value_type.
+ *
+ *  \pre \p first1 may equal \p first2, but the range <tt>[first1, last1)</tt> shall not overlap the range <tt>[first2, first2 + (last1 - first1))</tt> otherwise.
+ *
+ *  The following code snippet demonstrates how to use \p swap_ranges to
+ *  swap the contents of two \c thrust::device_vectors using the \p thrust::device execution
+ *  policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/swap.h>
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  thrust::device_vector<int> v1(2), v2(2);
+ *  v1[0] = 1;
+ *  v1[1] = 2;
+ *  v2[0] = 3;
+ *  v2[1] = 4;
+ *
+ *  thrust::swap_ranges(thrust::device, v1.begin(), v1.end(), v2.begin());
+ *
+ *  // v1[0] == 3, v1[1] == 4, v2[0] == 1, v2[1] == 2
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/swap_ranges.html
+ *  \see \c swap
+ */
+template<typename DerivedPolicy,
+         typename ForwardIterator1,
+         typename ForwardIterator2>
+  ForwardIterator2 swap_ranges(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                               ForwardIterator1 first1,
+                               ForwardIterator1 last1,
+                               ForwardIterator2 first2);
+
+
+/*! \p swap_ranges swaps each of the elements in the range <tt>[first1, last1)</tt>
+ *  with the corresponding element in the range <tt>[first2, first2 + (last1 - first1))</tt>.
+ *  That is, for each integer \c n such that <tt>0 <= n < (last1 - first1)</tt>, it swaps
+ *  <tt>*(first1 + n)</tt> and <tt>*(first2 + n)</tt>. The return value is
+ *  <tt>first2 + (last1 - first1)</tt>.
+ *
+ *  \param first1 The beginning of the first sequence to swap.
+ *  \param last1 One position past the last element of the first sequence to swap.
+ *  \param first2 The beginning of the second sequence to swap.
+ *  \return An iterator pointing to one position past the last element of the second
+ *          sequence to swap.
+ *
+ *  \tparam ForwardIterator1 is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>,
+ *          and \p ForwardIterator1's \c value_type must be convertible to \p ForwardIterator2's \c value_type.
+ *  \tparam ForwardIterator2 is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>,
+ *          and \p ForwardIterator2's \c value_type must be convertible to \p ForwardIterator1's \c value_type.
+ *
+ *  \pre \p first1 may equal \p first2, but the range <tt>[first1, last1)</tt> shall not overlap the range <tt>[first2, first2 + (last1 - first1))</tt> otherwise.
+ *
+ *  The following code snippet demonstrates how to use \p swap_ranges to
+ *  swap the contents of two \c thrust::device_vectors.
+ *
+ *  \code
+ *  #include <thrust/swap.h>
+ *  #include <thrust/device_vector.h>
+ *  ...
+ *  thrust::device_vector<int> v1(2), v2(2);
+ *  v1[0] = 1;
+ *  v1[1] = 2;
+ *  v2[0] = 3;
+ *  v2[1] = 4;
+ *
+ *  thrust::swap_ranges(v1.begin(), v1.end(), v2.begin());
+ *
+ *  // v1[0] == 3, v1[1] == 4, v2[0] == 1, v2[1] == 2
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/swap_ranges.html
+ *  \see \c swap
+ */
+template<typename ForwardIterator1,
+         typename ForwardIterator2>
+  ForwardIterator2 swap_ranges(ForwardIterator1 first1,
+                               ForwardIterator1 last1,
+                               ForwardIterator2 first2);
+
+
+/*! \} // copying
+ */
+
+
+} // end thrust
+
+#include <thrust/detail/swap.inl>
+
diff --git a/compat/thrust/system/cpp/detail/adjacent_difference.h b/compat/thrust/system/cpp/detail/adjacent_difference.h
new file mode 100644
index 0000000000..ea212ffcd9
--- /dev/null
+++ b/compat/thrust/system/cpp/detail/adjacent_difference.h
@@ -0,0 +1,54 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file adjacent_difference.h
+ *  \brief C++ implementation of adjacent_difference.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/cpp/detail/execution_policy.h>
+#include <thrust/system/detail/internal/scalar/adjacent_difference.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cpp
+{
+namespace detail
+{
+
+template <typename DerivedPolicy,
+          typename InputIterator,
+          typename OutputIterator,
+          typename BinaryFunction>
+OutputIterator adjacent_difference(execution_policy<DerivedPolicy> &,
+                                   InputIterator first,
+                                   InputIterator last,
+                                   OutputIterator result,
+                                   BinaryFunction binary_op)
+{
+  return thrust::system::detail::internal::scalar::adjacent_difference(first, last, result, binary_op);
+}
+
+} // end namespace detail
+} // end namespace cpp
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/cpp/detail/assign_value.h b/compat/thrust/system/cpp/detail/assign_value.h
new file mode 100644
index 0000000000..847fc97afd
--- /dev/null
+++ b/compat/thrust/system/cpp/detail/assign_value.h
@@ -0,0 +1,42 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/system/cpp/detail/execution_policy.h>
+#include <thrust/detail/raw_pointer_cast.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cpp
+{
+namespace detail
+{
+
+template<typename DerivedPolicy, typename Pointer1, typename Pointer2>
+__host__ __device__
+  void assign_value(thrust::system::cpp::detail::execution_policy<DerivedPolicy> &, Pointer1 dst, Pointer2 src)
+{
+  *thrust::raw_pointer_cast(dst) = *thrust::raw_pointer_cast(src);
+} // end assign_value()
+
+} // end detail
+} // end cpp
+} // end system
+} // end thrust
+
diff --git a/compat/thrust/system/cpp/detail/binary_search.h b/compat/thrust/system/cpp/detail/binary_search.h
new file mode 100644
index 0000000000..37af539e0d
--- /dev/null
+++ b/compat/thrust/system/cpp/detail/binary_search.h
@@ -0,0 +1,77 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file binary_search.h
+ *  \brief C++ implementation of binary search algorithms.
+ */
+
+#pragma once
+
+#include <thrust/system/cpp/detail/execution_policy.h>
+#include <thrust/system/detail/internal/scalar/binary_search.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cpp
+{
+namespace detail
+{
+
+template <typename ForwardIterator,
+          typename T,
+          typename StrictWeakOrdering>
+ForwardIterator lower_bound(tag,
+                            ForwardIterator first,
+                            ForwardIterator last,
+                            const T& val,
+                            StrictWeakOrdering comp)
+{
+  return thrust::system::detail::internal::scalar::lower_bound(first, last, val, comp);
+}
+
+
+template <typename ForwardIterator,
+          typename T,
+          typename StrictWeakOrdering>
+ForwardIterator upper_bound(tag,
+                            ForwardIterator first,
+                            ForwardIterator last,
+                            const T& val, 
+                            StrictWeakOrdering comp)
+{
+  return thrust::system::detail::internal::scalar::upper_bound(first, last, val, comp);
+}
+
+template <typename ForwardIterator,
+          typename T,
+          typename StrictWeakOrdering>
+bool binary_search(tag,
+                   ForwardIterator first,
+                   ForwardIterator last,
+                   const T& val, 
+                   StrictWeakOrdering comp)
+{
+  return thrust::system::detail::internal::scalar::binary_search(first, last, val, comp);
+}
+
+} // end namespace detail
+} // end namespace cpp
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/cpp/detail/copy.h b/compat/thrust/system/cpp/detail/copy.h
new file mode 100644
index 0000000000..7299bbbd7a
--- /dev/null
+++ b/compat/thrust/system/cpp/detail/copy.h
@@ -0,0 +1,61 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+/*! \file copy.h
+ *  \brief C++ implementations of copy functions.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/cpp/detail/execution_policy.h>
+#include <thrust/system/detail/internal/scalar/copy.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cpp
+{
+namespace detail
+{
+
+template<typename InputIterator,
+         typename OutputIterator>
+  OutputIterator copy(tag,
+                      InputIterator first,
+                      InputIterator last,
+                      OutputIterator result)
+{
+  return thrust::system::detail::internal::scalar::copy(first, last, result);
+}
+
+template<typename InputIterator,
+         typename Size,
+         typename OutputIterator>
+  OutputIterator copy_n(tag,
+                        InputIterator first,
+                        Size n,
+                        OutputIterator result)
+{
+  return thrust::system::detail::internal::scalar::copy_n(first, n, result);
+}
+
+} // end namespace detail
+} // end namespace cpp
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/cpp/detail/copy_if.h b/compat/thrust/system/cpp/detail/copy_if.h
new file mode 100644
index 0000000000..2faadfa1b7
--- /dev/null
+++ b/compat/thrust/system/cpp/detail/copy_if.h
@@ -0,0 +1,51 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/cpp/detail/execution_policy.h>
+#include <thrust/system/detail/internal/scalar/copy_if.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cpp
+{
+namespace detail
+{
+
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename Predicate>
+  OutputIterator copy_if(tag,
+                         InputIterator1 first,
+                         InputIterator1 last,
+                         InputIterator2 stencil,
+                         OutputIterator result,
+                         Predicate pred)
+{
+  return thrust::system::detail::internal::scalar::copy_if(first, last, stencil, result, pred);
+}
+
+} // end detail
+} // end cpp
+} // end system
+} // end thrust
+
diff --git a/compat/thrust/system/cpp/detail/count.h b/compat/thrust/system/cpp/detail/count.h
new file mode 100644
index 0000000000..a307fc5fc8
--- /dev/null
+++ b/compat/thrust/system/cpp/detail/count.h
@@ -0,0 +1,22 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system has no special version of this algorithm 
+
diff --git a/compat/thrust/system/cpp/detail/dispatch/sort.h b/compat/thrust/system/cpp/detail/dispatch/sort.h
new file mode 100644
index 0000000000..2a03cf62bc
--- /dev/null
+++ b/compat/thrust/system/cpp/detail/dispatch/sort.h
@@ -0,0 +1,119 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/reverse.h>
+
+#include <thrust/detail/type_traits.h>
+#include <thrust/iterator/iterator_traits.h>
+
+#include <thrust/system/detail/internal/scalar/stable_merge_sort.h>
+#include <thrust/system/detail/internal/scalar/stable_radix_sort.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cpp
+{
+namespace detail
+{
+namespace dispatch
+{
+
+////////////////
+// Radix Sort //
+////////////////
+
+template<typename RandomAccessIterator,
+         typename StrictWeakOrdering>
+void stable_sort(RandomAccessIterator first,
+                 RandomAccessIterator last,
+                 StrictWeakOrdering comp,
+                 thrust::detail::true_type)
+{
+  thrust::system::detail::internal::scalar::stable_radix_sort(first, last);
+        
+  // if comp is greater<T> then reverse the keys
+  typedef typename thrust::iterator_traits<RandomAccessIterator>::value_type KeyType;
+  const static bool reverse = thrust::detail::is_same<StrictWeakOrdering, typename thrust::greater<KeyType> >::value;
+
+  if (reverse)
+    thrust::reverse(first, last);
+}
+
+template<typename RandomAccessIterator1,
+         typename RandomAccessIterator2,
+         typename StrictWeakOrdering>
+void stable_sort_by_key(RandomAccessIterator1 first1,
+                        RandomAccessIterator1 last1,
+                        RandomAccessIterator2 first2,
+                        StrictWeakOrdering comp,
+                        thrust::detail::true_type)
+{
+  // if comp is greater<T> then reverse the keys and values
+  typedef typename thrust::iterator_traits<RandomAccessIterator1>::value_type KeyType;
+  const static bool reverse = thrust::detail::is_same<StrictWeakOrdering, typename thrust::greater<KeyType> >::value;
+
+  // note, we also have to reverse the (unordered) input to preserve stability
+  if (reverse)
+  {
+    thrust::reverse(first1,  last1);
+    thrust::reverse(first2, first2 + (last1 - first1));
+  }
+
+  thrust::system::detail::internal::scalar::stable_radix_sort_by_key(first1, last1, first2);
+
+  if (reverse)
+  {
+    thrust::reverse(first1,  last1);
+    thrust::reverse(first2, first2 + (last1 - first1));
+  }
+}
+
+////////////////
+// Merge Sort //
+////////////////
+
+template<typename RandomAccessIterator,
+         typename StrictWeakOrdering>
+void stable_sort(RandomAccessIterator first,
+                 RandomAccessIterator last,
+                 StrictWeakOrdering comp,
+                 thrust::detail::false_type)
+{
+  thrust::system::detail::internal::scalar::stable_merge_sort(first, last, comp);
+}
+
+template<typename RandomAccessIterator1,
+         typename RandomAccessIterator2,
+         typename StrictWeakOrdering>
+void stable_sort_by_key(RandomAccessIterator1 first1,
+                        RandomAccessIterator1 last1,
+                        RandomAccessIterator2 first2,
+                        StrictWeakOrdering comp,
+                        thrust::detail::false_type)
+{
+  thrust::system::detail::internal::scalar::stable_merge_sort_by_key(first1, last1, first2, comp);
+}
+
+} // end namespace dispatch
+} // end namespace detail
+} // end namespace cpp
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/cpp/detail/equal.h b/compat/thrust/system/cpp/detail/equal.h
new file mode 100644
index 0000000000..a307fc5fc8
--- /dev/null
+++ b/compat/thrust/system/cpp/detail/equal.h
@@ -0,0 +1,22 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system has no special version of this algorithm 
+
diff --git a/compat/thrust/system/cpp/detail/execution_policy.h b/compat/thrust/system/cpp/detail/execution_policy.h
new file mode 100644
index 0000000000..229ff5c6c8
--- /dev/null
+++ b/compat/thrust/system/cpp/detail/execution_policy.h
@@ -0,0 +1,84 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/execution_policy.h>
+
+namespace thrust
+{
+namespace system
+{
+// put the canonical tag in the same ns as the backend's entry points
+namespace cpp
+{
+namespace detail
+{
+
+// this awkward sequence of definitions arise
+// from the desire both for tag to derive
+// from execution_policy and for execution_policy
+// to convert to tag (when execution_policy is not
+// an ancestor of tag)
+
+// forward declaration of tag
+struct tag;
+
+// forward declaration of execution_policy
+template<typename> struct execution_policy;
+
+// specialize execution_policy for tag
+template<>
+  struct execution_policy<tag>
+    : thrust::execution_policy<tag>
+{};
+
+// tag's definition comes before the
+// generic definition of execution_policy
+struct tag : execution_policy<tag> {};
+
+// allow conversion to tag when it is not a successor
+template<typename Derived>
+  struct execution_policy
+    : thrust::execution_policy<Derived>
+{
+  // allow conversion to tag
+  inline operator tag () const
+  {
+    return tag();
+  }
+};
+
+} // end detail
+
+// alias execution_policy and tag here
+using thrust::system::cpp::detail::execution_policy;
+using thrust::system::cpp::detail::tag;
+
+} // end cpp
+} // end system
+
+// alias items at top-level
+namespace cpp
+{
+
+using thrust::system::cpp::execution_policy;
+using thrust::system::cpp::tag;
+
+} // end cpp
+} // end thrust
+
diff --git a/compat/thrust/system/cpp/detail/extrema.h b/compat/thrust/system/cpp/detail/extrema.h
new file mode 100644
index 0000000000..3eab6d406d
--- /dev/null
+++ b/compat/thrust/system/cpp/detail/extrema.h
@@ -0,0 +1,72 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file extrema.h
+ *  \brief C++ implementations of extrema functions.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/pair.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/system/cpp/detail/execution_policy.h>
+#include <thrust/system/detail/internal/scalar/extrema.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cpp
+{
+namespace detail
+{
+
+template <typename DerivedPolicy, typename ForwardIterator, typename BinaryPredicate>
+ForwardIterator min_element(execution_policy<DerivedPolicy> &,
+                            ForwardIterator first, 
+                            ForwardIterator last,
+                            BinaryPredicate comp)
+{
+  return thrust::system::detail::internal::scalar::min_element(first, last, comp);
+}
+
+
+template <typename DerivedPolicy, typename ForwardIterator, typename BinaryPredicate>
+ForwardIterator max_element(execution_policy<DerivedPolicy> &,
+                            ForwardIterator first, 
+                            ForwardIterator last,
+                            BinaryPredicate comp)
+{
+  return thrust::system::detail::internal::scalar::max_element(first, last, comp);
+}
+
+
+template <typename DerivedPolicy, typename ForwardIterator, typename BinaryPredicate>
+thrust::pair<ForwardIterator,ForwardIterator> minmax_element(execution_policy<DerivedPolicy> &,
+                                                             ForwardIterator first, 
+                                                             ForwardIterator last,
+                                                             BinaryPredicate comp)
+{
+  return thrust::system::detail::internal::scalar::minmax_element(first, last, comp);
+}
+
+} // end namespace detail
+} // end namespace cpp
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/cpp/detail/fill.h b/compat/thrust/system/cpp/detail/fill.h
new file mode 100644
index 0000000000..a307fc5fc8
--- /dev/null
+++ b/compat/thrust/system/cpp/detail/fill.h
@@ -0,0 +1,22 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system has no special version of this algorithm 
+
diff --git a/compat/thrust/system/cpp/detail/find.h b/compat/thrust/system/cpp/detail/find.h
new file mode 100644
index 0000000000..9698524ed8
--- /dev/null
+++ b/compat/thrust/system/cpp/detail/find.h
@@ -0,0 +1,52 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file find.h
+ *  \brief C++ implementation of find_if. 
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/system/cpp/detail/execution_policy.h>
+#include <thrust/system/detail/internal/scalar/find.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cpp
+{
+namespace detail
+{
+
+template <typename InputIterator,
+          typename Predicate>
+InputIterator find_if(tag,
+                      InputIterator first,
+                      InputIterator last,
+                      Predicate pred)
+{
+  return thrust::system::detail::internal::scalar::find_if(first, last, pred);
+}
+
+} // end namespace detail
+} // end namespace cpp
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/cpp/detail/for_each.h b/compat/thrust/system/cpp/detail/for_each.h
new file mode 100644
index 0000000000..8d4e1c730b
--- /dev/null
+++ b/compat/thrust/system/cpp/detail/for_each.h
@@ -0,0 +1,60 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/cpp/detail/execution_policy.h>
+#include <thrust/system/detail/internal/scalar/for_each.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cpp
+{
+namespace detail
+{
+
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename UnaryFunction>
+InputIterator for_each(thrust::system::cpp::detail::execution_policy<DerivedPolicy> &,
+                       InputIterator first,
+                       InputIterator last,
+                       UnaryFunction f)
+{
+  return thrust::system::detail::internal::scalar::for_each(first, last, f);
+}
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename Size,
+         typename UnaryFunction>
+InputIterator for_each_n(thrust::system::cpp::detail::execution_policy<DerivedPolicy> &,
+                         InputIterator first,
+                         Size n,
+                         UnaryFunction f)
+{
+  return thrust::system::detail::internal::scalar::for_each_n(first, n, f);
+}
+
+} // end namespace detail
+} // end namespace cpp
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/cpp/detail/gather.h b/compat/thrust/system/cpp/detail/gather.h
new file mode 100644
index 0000000000..a307fc5fc8
--- /dev/null
+++ b/compat/thrust/system/cpp/detail/gather.h
@@ -0,0 +1,22 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system has no special version of this algorithm 
+
diff --git a/compat/thrust/system/cpp/detail/generate.h b/compat/thrust/system/cpp/detail/generate.h
new file mode 100644
index 0000000000..a307fc5fc8
--- /dev/null
+++ b/compat/thrust/system/cpp/detail/generate.h
@@ -0,0 +1,22 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system has no special version of this algorithm 
+
diff --git a/compat/thrust/system/cpp/detail/get_value.h b/compat/thrust/system/cpp/detail/get_value.h
new file mode 100644
index 0000000000..5ddb2c8349
--- /dev/null
+++ b/compat/thrust/system/cpp/detail/get_value.h
@@ -0,0 +1,45 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/system/cpp/detail/execution_policy.h>
+#include <thrust/detail/raw_pointer_cast.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cpp
+{
+namespace detail
+{
+
+
+template<typename DerivedPolicy, typename Pointer>
+__host__ __device__
+  typename thrust::iterator_value<Pointer>::type
+    get_value(thrust::system::cpp::detail::execution_policy<DerivedPolicy> &, Pointer ptr)
+{
+  return *thrust::raw_pointer_cast(ptr);
+} // end get_value()
+
+
+} // end detail
+} // end cpp
+} // end system
+} // end thrust
+
diff --git a/compat/thrust/system/cpp/detail/inner_product.h b/compat/thrust/system/cpp/detail/inner_product.h
new file mode 100644
index 0000000000..a307fc5fc8
--- /dev/null
+++ b/compat/thrust/system/cpp/detail/inner_product.h
@@ -0,0 +1,22 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system has no special version of this algorithm 
+
diff --git a/compat/thrust/system/cpp/detail/iter_swap.h b/compat/thrust/system/cpp/detail/iter_swap.h
new file mode 100644
index 0000000000..257276ffea
--- /dev/null
+++ b/compat/thrust/system/cpp/detail/iter_swap.h
@@ -0,0 +1,46 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/system/cpp/detail/execution_policy.h>
+#include <thrust/detail/raw_pointer_cast.h>
+#include <thrust/detail/swap.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cpp
+{
+namespace detail
+{
+
+
+template<typename Pointer1, typename Pointer2>
+__host__ __device__
+  void iter_swap(tag, Pointer1 a, Pointer2 b)
+{
+  using thrust::swap;
+  swap(*thrust::raw_pointer_cast(a), *thrust::raw_pointer_cast(b));
+} // end iter_swap()
+
+
+} // end detail
+} // end cpp
+} // end system
+} // end thrust
+
diff --git a/compat/thrust/system/cpp/detail/logical.h b/compat/thrust/system/cpp/detail/logical.h
new file mode 100644
index 0000000000..a307fc5fc8
--- /dev/null
+++ b/compat/thrust/system/cpp/detail/logical.h
@@ -0,0 +1,22 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system has no special version of this algorithm 
+
diff --git a/compat/thrust/system/cpp/detail/malloc_and_free.h b/compat/thrust/system/cpp/detail/malloc_and_free.h
new file mode 100644
index 0000000000..4f8ae82092
--- /dev/null
+++ b/compat/thrust/system/cpp/detail/malloc_and_free.h
@@ -0,0 +1,54 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/raw_pointer_cast.h>
+#include <cstdlib> // for malloc & free
+#include <thrust/system/cpp/detail/execution_policy.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cpp
+{
+namespace detail
+{
+
+
+// note that malloc returns a raw pointer to avoid
+// depending on the heavyweight thrust/system/cpp/memory.h header
+template<typename DerivedPolicy>
+  void *malloc(execution_policy<DerivedPolicy> &, std::size_t n)
+{
+  return std::malloc(n);
+} // end malloc()
+
+
+template<typename DerivedPolicy, typename Pointer>
+  void free(execution_policy<DerivedPolicy> &, Pointer ptr)
+{
+  std::free(thrust::raw_pointer_cast(ptr));
+} // end free()
+
+
+} // end detail
+} // end cpp
+} // end system
+} // end thrust
+
diff --git a/compat/thrust/system/cpp/detail/memory.inl b/compat/thrust/system/cpp/detail/memory.inl
new file mode 100644
index 0000000000..7f9a48dee7
--- /dev/null
+++ b/compat/thrust/system/cpp/detail/memory.inl
@@ -0,0 +1,92 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/system/cpp/memory.h>
+#include <thrust/system/cpp/detail/malloc_and_free.h>
+#include <limits>
+
+namespace thrust
+{
+
+// XXX WAR an issue with MSVC 2005 (cl v14.00) incorrectly implementing
+//     pointer_raw_pointer for pointer by specializing it here
+#if (THRUST_HOST_COMPILER == THRUST_HOST_COMPILER_MSVC) && (_MSC_VER <= 1400)
+namespace detail
+{
+
+template<typename T>
+  struct pointer_raw_pointer< thrust::cpp::pointer<T> >
+{
+  typedef typename thrust::cpp::pointer<T>::raw_pointer type;
+}; // end pointer_raw_pointer
+
+} // end detail
+#endif
+
+namespace system
+{
+namespace cpp
+{
+
+
+template<typename T>
+  template<typename OtherT>
+    reference<T> &
+      reference<T>
+        ::operator=(const reference<OtherT> &other)
+{
+  return super_t::operator=(other);
+} // end reference::operator=()
+
+template<typename T>
+  reference<T> &
+    reference<T>
+      ::operator=(const value_type &x)
+{
+  return super_t::operator=(x);
+} // end reference::operator=()
+
+template<typename T>
+__host__ __device__
+void swap(reference<T> a, reference<T> b)
+{
+  a.swap(b);
+} // end swap()
+
+pointer<void> malloc(std::size_t n)
+{
+  tag t;
+  return pointer<void>(thrust::system::cpp::detail::malloc(t, n));
+} // end malloc()
+
+template<typename T>
+pointer<T> malloc(std::size_t n)
+{
+  pointer<void> raw_ptr = thrust::system::cpp::malloc(sizeof(T) * n);
+  return pointer<T>(reinterpret_cast<T*>(raw_ptr.get()));
+} // end malloc()
+
+void free(pointer<void> ptr)
+{
+  tag t;
+  return thrust::system::cpp::detail::free(t, ptr);
+} // end free()
+
+} // end cpp
+} // end system
+} // end thrust
+
diff --git a/compat/thrust/system/cpp/detail/merge.h b/compat/thrust/system/cpp/detail/merge.h
new file mode 100644
index 0000000000..7f01c0713a
--- /dev/null
+++ b/compat/thrust/system/cpp/detail/merge.h
@@ -0,0 +1,76 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/pair.h>
+#include <thrust/system/cpp/detail/execution_policy.h>
+#include <thrust/system/detail/internal/scalar/merge.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cpp
+{
+namespace detail
+{
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename StrictWeakOrdering>
+OutputIterator merge(execution_policy<DerivedPolicy> &,
+                     InputIterator1 first1,
+                     InputIterator1 last1,
+                     InputIterator2 first2,
+                     InputIterator2 last2,
+                     OutputIterator result,
+                     StrictWeakOrdering comp)
+{
+  return thrust::system::detail::internal::scalar::merge(first1, last1, first2, last2, result, comp);
+}
+
+template <typename DerivedPolicy,
+          typename InputIterator1,
+          typename InputIterator2,
+          typename InputIterator3,
+          typename InputIterator4,
+          typename OutputIterator1,
+          typename OutputIterator2,
+          typename StrictWeakOrdering>
+thrust::pair<OutputIterator1,OutputIterator2>
+  merge_by_key(execution_policy<DerivedPolicy> &,
+               InputIterator1 keys_first1,
+               InputIterator1 keys_last1,
+               InputIterator2 keys_first2,
+               InputIterator2 keys_last2,
+               InputIterator3 values_first1,
+               InputIterator4 values_first2,
+               OutputIterator1 keys_result,
+               OutputIterator2 values_result,
+               StrictWeakOrdering comp)
+{
+  return thrust::system::detail::internal::scalar::merge_by_key(keys_first1, keys_last1, keys_first2, keys_last2, values_first1, values_first2, keys_result, values_result, comp);
+}
+
+} // end namespace detail
+} // end namespace cpp
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/cpp/detail/mismatch.h b/compat/thrust/system/cpp/detail/mismatch.h
new file mode 100644
index 0000000000..a307fc5fc8
--- /dev/null
+++ b/compat/thrust/system/cpp/detail/mismatch.h
@@ -0,0 +1,22 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system has no special version of this algorithm 
+
diff --git a/compat/thrust/system/cpp/detail/par.h b/compat/thrust/system/cpp/detail/par.h
new file mode 100644
index 0000000000..953e5274db
--- /dev/null
+++ b/compat/thrust/system/cpp/detail/par.h
@@ -0,0 +1,66 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/cpp/detail/execution_policy.h>
+#include <thrust/detail/execute_with_allocator.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cpp
+{
+namespace detail
+{
+
+
+struct par_t : thrust::system::cpp::detail::execution_policy<par_t>
+{
+  par_t() : thrust::system::cpp::detail::execution_policy<par_t>() {}
+
+  template<typename Allocator>
+    thrust::detail::execute_with_allocator<Allocator, thrust::system::cpp::detail::execution_policy>
+      operator()(Allocator &alloc) const
+  {
+    return thrust::detail::execute_with_allocator<Allocator, thrust::system::cpp::detail::execution_policy>(alloc);
+  }
+};
+
+
+} // end detail
+
+
+static const detail::par_t par;
+
+
+} // end cpp
+} // end system
+
+
+// alias par here
+namespace cpp
+{
+
+
+using thrust::system::cpp::par;
+
+
+} // end cpp
+} // end thrust
+
diff --git a/compat/thrust/system/cpp/detail/partition.h b/compat/thrust/system/cpp/detail/partition.h
new file mode 100644
index 0000000000..25a4f1c346
--- /dev/null
+++ b/compat/thrust/system/cpp/detail/partition.h
@@ -0,0 +1,95 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file partition.h
+ *  \brief cpp implementations of partition functions
+ */
+
+#pragma once
+
+#include <thrust/pair.h>
+#include <thrust/system/cpp/detail/execution_policy.h>
+#include <thrust/system/detail/internal/scalar/partition.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cpp
+{
+namespace detail
+{
+
+template<typename ForwardIterator,
+         typename Predicate>
+  ForwardIterator stable_partition(tag,
+                                   ForwardIterator first,
+                                   ForwardIterator last,
+                                   Predicate pred)
+{
+  return thrust::system::detail::internal::scalar::stable_partition(first, last, pred);
+}
+
+template<typename ForwardIterator,
+         typename InputIterator,
+         typename Predicate>
+  ForwardIterator stable_partition(tag,
+                                   ForwardIterator first,
+                                   ForwardIterator last,
+                                   InputIterator stencil,
+                                   Predicate pred)
+{
+  return thrust::system::detail::internal::scalar::stable_partition(first, last, stencil, pred);
+}
+
+template<typename InputIterator,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename Predicate>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    stable_partition_copy(tag,
+                          InputIterator first,
+                          InputIterator last,
+                          OutputIterator1 out_true,
+                          OutputIterator2 out_false,
+                          Predicate pred)
+{
+  return thrust::system::detail::internal::scalar::stable_partition_copy(first, last, out_true, out_false, pred);
+}
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename Predicate>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    stable_partition_copy(tag,
+                          InputIterator1 first,
+                          InputIterator1 last,
+                          InputIterator2 stencil,
+                          OutputIterator1 out_true,
+                          OutputIterator2 out_false,
+                          Predicate pred)
+{
+  return thrust::system::detail::internal::scalar::stable_partition_copy(first, last, stencil, out_true, out_false, pred);
+}
+
+} // end namespace detail
+} // end namespace cpp
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/cpp/detail/reduce.h b/compat/thrust/system/cpp/detail/reduce.h
new file mode 100644
index 0000000000..5428206ba3
--- /dev/null
+++ b/compat/thrust/system/cpp/detail/reduce.h
@@ -0,0 +1,54 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file reduce.h
+ *  \brief C++ implementation of reduce algorithms.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/cpp/detail/execution_policy.h>
+#include <thrust/system/detail/internal/scalar/reduce.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cpp
+{
+namespace detail
+{
+
+template<typename DerivedPolicy,
+         typename InputIterator, 
+         typename OutputType,
+         typename BinaryFunction>
+  OutputType reduce(execution_policy<DerivedPolicy> &,
+                    InputIterator begin,
+                    InputIterator end,
+                    OutputType init,
+                    BinaryFunction binary_op)
+{
+  return thrust::system::detail::internal::scalar::reduce(begin, end, init, binary_op);
+}
+
+} // end namespace detail
+} // end namespace cpp
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/cpp/detail/reduce_by_key.h b/compat/thrust/system/cpp/detail/reduce_by_key.h
new file mode 100644
index 0000000000..22dc2d9d3d
--- /dev/null
+++ b/compat/thrust/system/cpp/detail/reduce_by_key.h
@@ -0,0 +1,57 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/cpp/detail/execution_policy.h>
+#include <thrust/pair.h>
+#include <thrust/system/detail/internal/scalar/reduce_by_key.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cpp
+{
+namespace detail
+{
+
+template <typename DerivedPolicy,
+          typename InputIterator1,
+          typename InputIterator2,
+          typename OutputIterator1,
+          typename OutputIterator2,
+          typename BinaryPredicate,
+          typename BinaryFunction>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    reduce_by_key(execution_policy<DerivedPolicy> &,
+                  InputIterator1 keys_first, 
+                  InputIterator1 keys_last,
+                  InputIterator2 values_first,
+                  OutputIterator1 keys_output,
+                  OutputIterator2 values_output,
+                  BinaryPredicate binary_pred,
+                  BinaryFunction binary_op)
+{
+  return thrust::system::detail::internal::scalar::reduce_by_key(keys_first, keys_last, values_first, keys_output, values_output, binary_pred, binary_op);
+}
+
+} // end namespace detail
+} // end namespace cpp
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/cpp/detail/remove.h b/compat/thrust/system/cpp/detail/remove.h
new file mode 100644
index 0000000000..cf2202bee3
--- /dev/null
+++ b/compat/thrust/system/cpp/detail/remove.h
@@ -0,0 +1,88 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/cpp/detail/execution_policy.h>
+#include <thrust/system/detail/internal/scalar/remove.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cpp
+{
+namespace detail
+{
+
+template<typename ForwardIterator,
+         typename Predicate>
+  ForwardIterator remove_if(tag,
+                            ForwardIterator first,
+                            ForwardIterator last,
+                            Predicate pred)
+{
+  return thrust::system::detail::internal::scalar::remove_if(first, last, pred);
+}
+
+
+template<typename ForwardIterator,
+         typename InputIterator,
+         typename Predicate>
+  ForwardIterator remove_if(tag,
+                            ForwardIterator first,
+                            ForwardIterator last,
+                            InputIterator stencil,
+                            Predicate pred)
+{
+  return thrust::system::detail::internal::scalar::remove_if(first, last, stencil, pred);
+}
+
+
+template<typename InputIterator,
+         typename OutputIterator,
+         typename Predicate>
+  OutputIterator remove_copy_if(tag,
+                                InputIterator first,
+                                InputIterator last,
+                                OutputIterator result,
+                                Predicate pred)
+{
+  return thrust::system::detail::internal::scalar::remove_copy_if(first, last, result, pred);
+}
+
+
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename Predicate>
+  OutputIterator remove_copy_if(tag,
+                                InputIterator1 first,
+                                InputIterator1 last,
+                                InputIterator2 stencil,
+                                OutputIterator result,
+                                Predicate pred)
+{
+  return thrust::system::detail::internal::scalar::remove_copy_if(first, last, stencil, result, pred);
+}
+
+} // end namespace detail
+} // end namespace cpp
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/cpp/detail/replace.h b/compat/thrust/system/cpp/detail/replace.h
new file mode 100644
index 0000000000..a307fc5fc8
--- /dev/null
+++ b/compat/thrust/system/cpp/detail/replace.h
@@ -0,0 +1,22 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system has no special version of this algorithm 
+
diff --git a/compat/thrust/system/cpp/detail/reverse.h b/compat/thrust/system/cpp/detail/reverse.h
new file mode 100644
index 0000000000..a307fc5fc8
--- /dev/null
+++ b/compat/thrust/system/cpp/detail/reverse.h
@@ -0,0 +1,22 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system has no special version of this algorithm 
+
diff --git a/compat/thrust/system/cpp/detail/scan.h b/compat/thrust/system/cpp/detail/scan.h
new file mode 100644
index 0000000000..d4bae1e739
--- /dev/null
+++ b/compat/thrust/system/cpp/detail/scan.h
@@ -0,0 +1,70 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file scan.h
+ *  \brief C++ implementations of scan functions.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/cpp/detail/execution_policy.h>
+#include <thrust/system/detail/internal/scalar/scan.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cpp
+{
+namespace detail
+{
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator,
+         typename BinaryFunction>
+  OutputIterator inclusive_scan(execution_policy<DerivedPolicy> &,
+                                InputIterator first,
+                                InputIterator last,
+                                OutputIterator result,
+                                BinaryFunction binary_op)
+{
+  return thrust::system::detail::internal::scalar::inclusive_scan(first, last, result, binary_op);
+}
+
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator,
+         typename T,
+         typename BinaryFunction>
+  OutputIterator exclusive_scan(execution_policy<DerivedPolicy> &,
+                                InputIterator first,
+                                InputIterator last,
+                                OutputIterator result,
+                                T init,
+                                BinaryFunction binary_op)
+{
+  return thrust::system::detail::internal::scalar::exclusive_scan(first, last, result, init, binary_op);
+}
+
+} // end namespace detail
+} // end namespace cpp
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/cpp/detail/scan_by_key.h b/compat/thrust/system/cpp/detail/scan_by_key.h
new file mode 100644
index 0000000000..4165d842fd
--- /dev/null
+++ b/compat/thrust/system/cpp/detail/scan_by_key.h
@@ -0,0 +1,71 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/cpp/detail/execution_policy.h>
+#include <thrust/system/detail/internal/scalar/scan_by_key.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cpp
+{
+namespace detail
+{
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename BinaryPredicate,
+         typename BinaryFunction>
+  OutputIterator inclusive_scan_by_key(tag,
+                                       InputIterator1 first1,
+                                       InputIterator1 last1,
+                                       InputIterator2 first2,
+                                       OutputIterator result,
+                                       BinaryPredicate binary_pred,
+                                       BinaryFunction binary_op)
+{
+  return thrust::system::detail::internal::scalar::inclusive_scan_by_key(first1, last1, first2, result, binary_pred, binary_op);
+}
+
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename T,
+         typename BinaryPredicate,
+         typename BinaryFunction>
+  OutputIterator exclusive_scan_by_key(tag,
+                                       InputIterator1 first1,
+                                       InputIterator1 last1,
+                                       InputIterator2 first2,
+                                       OutputIterator result,
+                                       T init,
+                                       BinaryPredicate binary_pred,
+                                       BinaryFunction binary_op)
+{
+  return thrust::system::detail::internal::scalar::exclusive_scan_by_key(first1, last1, first2, result, init, binary_pred, binary_op);
+}
+
+} // end namespace detail
+} // end namespace cpp
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/cpp/detail/scatter.h b/compat/thrust/system/cpp/detail/scatter.h
new file mode 100644
index 0000000000..a307fc5fc8
--- /dev/null
+++ b/compat/thrust/system/cpp/detail/scatter.h
@@ -0,0 +1,22 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system has no special version of this algorithm 
+
diff --git a/compat/thrust/system/cpp/detail/sequence.h b/compat/thrust/system/cpp/detail/sequence.h
new file mode 100644
index 0000000000..a307fc5fc8
--- /dev/null
+++ b/compat/thrust/system/cpp/detail/sequence.h
@@ -0,0 +1,22 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system has no special version of this algorithm 
+
diff --git a/compat/thrust/system/cpp/detail/set_operations.h b/compat/thrust/system/cpp/detail/set_operations.h
new file mode 100644
index 0000000000..07ce71257b
--- /dev/null
+++ b/compat/thrust/system/cpp/detail/set_operations.h
@@ -0,0 +1,105 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/cpp/detail/execution_policy.h>
+#include <thrust/system/detail/internal/scalar/set_operations.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cpp
+{
+namespace detail
+{
+
+
+template<typename ExecutionPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename StrictWeakOrdering>
+  OutputIterator set_difference(execution_policy<ExecutionPolicy> &,
+                                InputIterator1 first1,
+                                InputIterator1 last1,
+                                InputIterator2 first2,
+                                InputIterator2 last2,
+                                OutputIterator result,
+                                StrictWeakOrdering comp)
+{
+  return thrust::system::detail::internal::scalar::set_difference(first1, last1, first2, last2, result, comp);
+}
+
+
+template<typename ExecutionPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename StrictWeakOrdering>
+  OutputIterator set_intersection(execution_policy<ExecutionPolicy> &,
+                                  InputIterator1 first1,
+                                  InputIterator1 last1,
+                                  InputIterator2 first2,
+                                  InputIterator2 last2,
+                                  OutputIterator result,
+                                  StrictWeakOrdering comp)
+{
+  return thrust::system::detail::internal::scalar::set_intersection(first1, last1, first2, last2, result, comp);
+}
+
+
+template<typename ExecutionPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename StrictWeakOrdering>
+  OutputIterator set_symmetric_difference(execution_policy<ExecutionPolicy> &,
+                                          InputIterator1 first1,
+                                          InputIterator1 last1,
+                                          InputIterator2 first2,
+                                          InputIterator2 last2,
+                                          OutputIterator result,
+                                          StrictWeakOrdering comp)
+{
+  return thrust::system::detail::internal::scalar::set_symmetric_difference(first1, last1, first2, last2, result, comp);
+}
+
+
+template<typename ExecutionPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename StrictWeakOrdering>
+  OutputIterator set_union(execution_policy<ExecutionPolicy> &,
+                           InputIterator1 first1,
+                           InputIterator1 last1,
+                           InputIterator2 first2,
+                           InputIterator2 last2,
+                           OutputIterator result,
+                           StrictWeakOrdering comp)
+{
+  return thrust::system::detail::internal::scalar::set_union(first1, last1, first2, last2, result, comp);
+}
+
+
+} // end namespace detail
+} // end namespace cpp
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/cpp/detail/sort.h b/compat/thrust/system/cpp/detail/sort.h
new file mode 100644
index 0000000000..60244e22a3
--- /dev/null
+++ b/compat/thrust/system/cpp/detail/sort.h
@@ -0,0 +1,60 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/cpp/detail/execution_policy.h>
+#include <thrust/system/detail/internal/scalar/sort.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cpp
+{
+namespace detail
+{
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator,
+         typename StrictWeakOrdering>
+  void stable_sort(execution_policy<DerivedPolicy> &,
+                   RandomAccessIterator first,
+                   RandomAccessIterator last,
+                   StrictWeakOrdering comp)
+{
+  thrust::system::detail::internal::scalar::stable_sort(first, last, comp);
+}
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2,
+         typename StrictWeakOrdering>
+  void stable_sort_by_key(execution_policy<DerivedPolicy> &,
+                          RandomAccessIterator1 keys_first,
+                          RandomAccessIterator1 keys_last,
+                          RandomAccessIterator2 values_first,
+                          StrictWeakOrdering comp)
+{
+  thrust::system::detail::internal::scalar::stable_sort_by_key(keys_first, keys_last, values_first, comp);
+}
+
+} // end namespace detail
+} // end namespace cpp
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/cpp/detail/swap_ranges.h b/compat/thrust/system/cpp/detail/swap_ranges.h
new file mode 100644
index 0000000000..a834a2c0ed
--- /dev/null
+++ b/compat/thrust/system/cpp/detail/swap_ranges.h
@@ -0,0 +1,22 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// cpp has no special swap_ranges
+
diff --git a/compat/thrust/system/cpp/detail/tabulate.h b/compat/thrust/system/cpp/detail/tabulate.h
new file mode 100644
index 0000000000..a307fc5fc8
--- /dev/null
+++ b/compat/thrust/system/cpp/detail/tabulate.h
@@ -0,0 +1,22 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system has no special version of this algorithm 
+
diff --git a/compat/thrust/system/cpp/detail/temporary_buffer.h b/compat/thrust/system/cpp/detail/temporary_buffer.h
new file mode 100644
index 0000000000..628bd75719
--- /dev/null
+++ b/compat/thrust/system/cpp/detail/temporary_buffer.h
@@ -0,0 +1,22 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system has no special temporary buffer functions
+
diff --git a/compat/thrust/system/cpp/detail/transform.h b/compat/thrust/system/cpp/detail/transform.h
new file mode 100644
index 0000000000..5909d4a4fb
--- /dev/null
+++ b/compat/thrust/system/cpp/detail/transform.h
@@ -0,0 +1,22 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// cpp has no special transform
+
diff --git a/compat/thrust/system/cpp/detail/transform_reduce.h b/compat/thrust/system/cpp/detail/transform_reduce.h
new file mode 100644
index 0000000000..a307fc5fc8
--- /dev/null
+++ b/compat/thrust/system/cpp/detail/transform_reduce.h
@@ -0,0 +1,22 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system has no special version of this algorithm 
+
diff --git a/compat/thrust/system/cpp/detail/transform_scan.h b/compat/thrust/system/cpp/detail/transform_scan.h
new file mode 100644
index 0000000000..a307fc5fc8
--- /dev/null
+++ b/compat/thrust/system/cpp/detail/transform_scan.h
@@ -0,0 +1,22 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system has no special version of this algorithm 
+
diff --git a/compat/thrust/system/cpp/detail/uninitialized_copy.h b/compat/thrust/system/cpp/detail/uninitialized_copy.h
new file mode 100644
index 0000000000..a307fc5fc8
--- /dev/null
+++ b/compat/thrust/system/cpp/detail/uninitialized_copy.h
@@ -0,0 +1,22 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system has no special version of this algorithm 
+
diff --git a/compat/thrust/system/cpp/detail/uninitialized_fill.h b/compat/thrust/system/cpp/detail/uninitialized_fill.h
new file mode 100644
index 0000000000..a307fc5fc8
--- /dev/null
+++ b/compat/thrust/system/cpp/detail/uninitialized_fill.h
@@ -0,0 +1,22 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system has no special version of this algorithm 
+
diff --git a/compat/thrust/system/cpp/detail/unique.h b/compat/thrust/system/cpp/detail/unique.h
new file mode 100644
index 0000000000..cf740498e1
--- /dev/null
+++ b/compat/thrust/system/cpp/detail/unique.h
@@ -0,0 +1,61 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/cpp/detail/execution_policy.h>
+#include <thrust/pair.h>
+#include <thrust/system/detail/internal/scalar/unique.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cpp
+{
+namespace detail
+{
+
+template<typename DerivedPolicy,
+         typename ForwardIterator,
+         typename BinaryPredicate>
+  ForwardIterator unique(execution_policy<DerivedPolicy> &,
+                         ForwardIterator first,
+                         ForwardIterator last,
+                         BinaryPredicate binary_pred)
+{
+  return thrust::system::detail::internal::scalar::unique(first, last, binary_pred);
+}
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator,
+         typename BinaryPredicate>
+  OutputIterator unique_copy(execution_policy<DerivedPolicy> &,
+                             InputIterator first,
+                             InputIterator last,
+                             OutputIterator output,
+                             BinaryPredicate binary_pred)
+{
+  return thrust::system::detail::internal::scalar::unique_copy(first, last, output, binary_pred);
+}
+
+} // end namespace detail
+} // end namespace cpp 
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/cpp/detail/unique_by_key.h b/compat/thrust/system/cpp/detail/unique_by_key.h
new file mode 100644
index 0000000000..a9f13d6a27
--- /dev/null
+++ b/compat/thrust/system/cpp/detail/unique_by_key.h
@@ -0,0 +1,70 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/cpp/detail/execution_policy.h>
+#include <thrust/pair.h>
+#include <thrust/system/detail/internal/scalar/unique_by_key.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cpp
+{
+namespace detail
+{
+
+template<typename DerivedPolicy,
+         typename ForwardIterator1,
+         typename ForwardIterator2,
+         typename BinaryPredicate>
+  thrust::pair<ForwardIterator1,ForwardIterator2>
+    unique_by_key(execution_policy<DerivedPolicy> &,
+                  ForwardIterator1 keys_first, 
+                  ForwardIterator1 keys_last,
+                  ForwardIterator2 values_first,
+                  BinaryPredicate binary_pred)
+{
+  return thrust::system::detail::internal::scalar::unique_by_key(keys_first, keys_last, values_first, binary_pred);
+}
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename BinaryPredicate>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    unique_by_key_copy(execution_policy<DerivedPolicy> &,
+                       InputIterator1 keys_first, 
+                       InputIterator1 keys_last,
+                       InputIterator2 values_first,
+                       OutputIterator1 keys_output,
+                       OutputIterator2 values_output,
+                       BinaryPredicate binary_pred)
+{
+  return thrust::system::detail::internal::scalar::unique_by_key_copy(keys_first, keys_last, values_first, keys_output, values_output, binary_pred);
+}
+
+} // end namespace detail
+} // end namespace cpp 
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/cpp/detail/vector.inl b/compat/thrust/system/cpp/detail/vector.inl
new file mode 100644
index 0000000000..03bffcd8aa
--- /dev/null
+++ b/compat/thrust/system/cpp/detail/vector.inl
@@ -0,0 +1,97 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/cpp/vector.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cpp
+{
+
+template<typename T, typename Allocator>
+  vector<T,Allocator>
+    ::vector()
+      : super_t()
+{}
+
+template<typename T, typename Allocator>
+  vector<T,Allocator>
+    ::vector(size_type n)
+      : super_t(n)
+{}
+
+template<typename T, typename Allocator>
+  vector<T,Allocator>
+    ::vector(size_type n, const value_type &value)
+      : super_t(n,value)
+{}
+
+template<typename T, typename Allocator>
+  vector<T,Allocator>
+    ::vector(const vector &x)
+      : super_t(x)
+{}
+
+template<typename T, typename Allocator>
+  template<typename OtherT, typename OtherAllocator>
+    vector<T,Allocator>
+      ::vector(const thrust::detail::vector_base<OtherT,OtherAllocator> &x)
+        : super_t(x)
+{}
+
+template<typename T, typename Allocator>
+  template<typename OtherT, typename OtherAllocator>
+    vector<T,Allocator>
+      ::vector(const std::vector<OtherT,OtherAllocator> &x)
+        : super_t(x)
+{}
+
+template<typename T, typename Allocator>
+  template<typename InputIterator>
+    vector<T,Allocator>
+      ::vector(InputIterator first, InputIterator last)
+        : super_t(first,last)
+{}
+
+template<typename T, typename Allocator>
+  template<typename OtherT, typename OtherAllocator>
+    vector<T,Allocator> &
+      vector<T,Allocator>
+        ::operator=(const std::vector<OtherT,OtherAllocator> &x)
+{
+  super_t::operator=(x);
+  return *this;
+}
+
+template<typename T, typename Allocator>
+  template<typename OtherT, typename OtherAllocator>
+    vector<T,Allocator> &
+      vector<T,Allocator>
+        ::operator=(const thrust::detail::vector_base<OtherT,OtherAllocator> &x)
+{
+  super_t::operator=(x);
+  return *this;
+}
+      
+} // end cpp
+} // end system
+} // end thrust
+
diff --git a/compat/thrust/system/cpp/execution_policy.h b/compat/thrust/system/cpp/execution_policy.h
new file mode 100644
index 0000000000..f192eb9659
--- /dev/null
+++ b/compat/thrust/system/cpp/execution_policy.h
@@ -0,0 +1,157 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+/*! \file thrust/system/cpp/execution_policy.h
+ *  \brief Execution policies for Thrust's standard C++ system.
+ */
+
+#include <thrust/detail/config.h>
+
+// get the execution policies definitions first
+#include <thrust/system/cpp/detail/execution_policy.h>
+
+// get the definition of par
+#include <thrust/system/cpp/detail/par.h>
+
+// now get all the algorithm definitions
+
+#include <thrust/system/cpp/detail/adjacent_difference.h>
+#include <thrust/system/cpp/detail/assign_value.h>
+#include <thrust/system/cpp/detail/binary_search.h>
+#include <thrust/system/cpp/detail/copy.h>
+#include <thrust/system/cpp/detail/copy_if.h>
+#include <thrust/system/cpp/detail/count.h>
+#include <thrust/system/cpp/detail/equal.h>
+#include <thrust/system/cpp/detail/extrema.h>
+#include <thrust/system/cpp/detail/fill.h>
+#include <thrust/system/cpp/detail/find.h>
+#include <thrust/system/cpp/detail/for_each.h>
+#include <thrust/system/cpp/detail/gather.h>
+#include <thrust/system/cpp/detail/generate.h>
+#include <thrust/system/cpp/detail/get_value.h>
+#include <thrust/system/cpp/detail/inner_product.h>
+#include <thrust/system/cpp/detail/iter_swap.h>
+#include <thrust/system/cpp/detail/logical.h>
+#include <thrust/system/cpp/detail/malloc_and_free.h>
+#include <thrust/system/cpp/detail/merge.h>
+#include <thrust/system/cpp/detail/mismatch.h>
+#include <thrust/system/cpp/detail/partition.h>
+#include <thrust/system/cpp/detail/reduce.h>
+#include <thrust/system/cpp/detail/reduce_by_key.h>
+#include <thrust/system/cpp/detail/remove.h>
+#include <thrust/system/cpp/detail/replace.h>
+#include <thrust/system/cpp/detail/reverse.h>
+#include <thrust/system/cpp/detail/scan.h>
+#include <thrust/system/cpp/detail/scan_by_key.h>
+#include <thrust/system/cpp/detail/scatter.h>
+#include <thrust/system/cpp/detail/sequence.h>
+#include <thrust/system/cpp/detail/set_operations.h>
+#include <thrust/system/cpp/detail/sort.h>
+#include <thrust/system/cpp/detail/swap_ranges.h>
+#include <thrust/system/cpp/detail/tabulate.h>
+#include <thrust/system/cpp/detail/transform.h>
+#include <thrust/system/cpp/detail/transform_reduce.h>
+#include <thrust/system/cpp/detail/transform_scan.h>
+#include <thrust/system/cpp/detail/uninitialized_copy.h>
+#include <thrust/system/cpp/detail/uninitialized_fill.h>
+#include <thrust/system/cpp/detail/unique.h>
+#include <thrust/system/cpp/detail/unique_by_key.h>
+
+
+// define these entities here for the purpose of Doxygenating them
+// they are actually defined elsewhere
+#if 0
+namespace thrust
+{
+namespace system
+{
+namespace cpp
+{
+
+
+/*! \addtogroup execution_policies
+ *  \{
+ */
+
+
+/*! \p thrust::system::cpp::execution_policy is the base class for all Thrust parallel execution
+ *  policies which are derived from Thrust's standard C++ backend system.
+ */
+template<typename DerivedPolicy>
+struct execution_policy : thrust::execution_policy<DerivedPolicy>
+{};
+
+
+/*! \p thrust::system::cpp::tag is a type representing Thrust's standard C++ backend system in C++'s type system.
+ *  Iterators "tagged" with a type which is convertible to \p cpp::tag assert that they may be
+ *  "dispatched" to algorithm implementations in the \p cpp system.
+ */
+struct tag : thrust::system::cpp::execution_policy<tag> { unspecified };
+
+
+/*! 
+ *  \p thrust::system::cpp::par is the parallel execution policy associated with Thrust's standard
+ *  C++ backend system.
+ *
+ *  Instead of relying on implicit algorithm dispatch through iterator system tags, users may
+ *  directly target Thrust's C++ backend system by providing \p thrust::cpp::par as an algorithm
+ *  parameter.
+ *
+ *  Explicit dispatch can be useful in avoiding the introduction of data copies into containers such
+ *  as \p thrust::cpp::vector.
+ *
+ *  The type of \p thrust::cpp::par is implementation-defined.
+ *
+ *  The following code snippet demonstrates how to use \p thrust::cpp::par to explicitly dispatch an
+ *  invocation of \p thrust::for_each to the standard C++ backend system:
+ *
+ *  \code
+ *  #include <thrust/for_each.h>
+ *  #include <thrust/system/cpp/execution_policy.h>
+ *  #include <cstdio>
+ *
+ *  struct printf_functor
+ *  {
+ *    __host__ __device__
+ *    void operator()(int x)
+ *    {
+ *      printf("%d\n");
+ *    }
+ *  };
+ *  ...
+ *  int vec[3];
+ *  vec[0] = 0; vec[1] = 1; vec[2] = 2;
+ *
+ *  thrust::for_each(thrust::cpp::par, vec.begin(), vec.end(), printf_functor());
+ *
+ *  // 0 1 2 is printed to standard output in some unspecified order
+ *  \endcode
+ */
+static const unspecified par;
+
+
+/*! \}
+ */
+
+
+} // end cpp
+} // end system
+} // end thrust
+#endif
+
+
diff --git a/compat/thrust/system/cpp/memory.h b/compat/thrust/system/cpp/memory.h
new file mode 100644
index 0000000000..f3a58b8c32
--- /dev/null
+++ b/compat/thrust/system/cpp/memory.h
@@ -0,0 +1,414 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+/*! \file thrust/system/cpp/memory.h
+ *  \brief Managing memory associated with Thrust's standard C++ system.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/cpp/detail/execution_policy.h>
+#include <thrust/memory.h>
+#include <thrust/detail/type_traits.h>
+#include <thrust/detail/allocator/malloc_allocator.h>
+#include <ostream>
+
+namespace thrust
+{
+namespace system
+{
+namespace cpp
+{
+
+template<typename> class pointer;
+
+} // end cpp
+} // end system
+} // end thrust
+
+
+/*! \cond
+ */
+
+// specialize std::iterator_traits to avoid problems with the name of
+// pointer's constructor shadowing its nested pointer type
+// do this before pointer is defined so the specialization is correctly
+// used inside the definition
+namespace std
+{
+
+template<typename Element>
+  struct iterator_traits<thrust::system::cpp::pointer<Element> >
+{
+  private:
+    typedef thrust::system::cpp::pointer<Element> ptr;
+
+  public:
+    typedef typename ptr::iterator_category       iterator_category;
+    typedef typename ptr::value_type              value_type;
+    typedef typename ptr::difference_type         difference_type;
+    typedef ptr                                   pointer;
+    typedef typename ptr::reference               reference;
+}; // end iterator_traits
+
+} // end std
+
+/*! \endcond
+ */
+
+
+namespace thrust
+{
+namespace system
+{
+
+/*! \addtogroup system_backends Systems
+ *  \ingroup system
+ *  \{
+ */
+
+/*! \namespace thrust::system::cpp
+ *  \brief \p thrust::system::cpp is the namespace containing functionality for allocating, manipulating,
+ *         and deallocating memory available to Thrust's standard C++ backend system.
+ *         The identifiers are provided in a separate namespace underneath <tt>thrust::system</tt>
+ *         for import convenience but are also aliased in the top-level <tt>thrust::cpp</tt>
+ *         namespace for easy access.
+ *
+ */
+namespace cpp
+{
+
+// forward declaration of reference for pointer
+template<typename Element> class reference;
+
+/*! \cond
+ */
+
+// XXX nvcc + msvc have trouble instantiating reference below
+//     this is a workaround
+namespace detail
+{
+
+template<typename Element>
+  struct reference_msvc_workaround
+{
+  typedef thrust::system::cpp::reference<Element> type;
+}; // end reference_msvc_workaround
+
+} // end detail
+
+/*! \endcond
+ */
+
+
+/*! \p pointer stores a pointer to an object allocated in memory available to the cpp system.
+ *  This type provides type safety when dispatching standard algorithms on ranges resident
+ *  in cpp memory.
+ *
+ *  \p pointer has pointer semantics: it may be dereferenced and manipulated with pointer arithmetic.
+ *
+ *  \p pointer can be created with the function \p cpp::malloc, or by explicitly calling its constructor
+ *  with a raw pointer.
+ *
+ *  The raw pointer encapsulated by a \p pointer may be obtained by eiter its <tt>get</tt> member function
+ *  or the \p raw_pointer_cast function.
+ *
+ *  \note \p pointer is not a "smart" pointer; it is the programmer's responsibility to deallocate memory
+ *  pointed to by \p pointer.
+ *
+ *  \tparam T specifies the type of the pointee.
+ *
+ *  \see cpp::malloc
+ *  \see cpp::free
+ *  \see raw_pointer_cast
+ */
+template<typename T>
+  class pointer
+    : public thrust::pointer<
+               T,
+               thrust::system::cpp::tag,
+               thrust::system::cpp::reference<T>,
+               thrust::system::cpp::pointer<T>
+             >
+{
+  /*! \cond
+   */
+
+  private:
+    typedef thrust::pointer<
+      T,
+      thrust::system::cpp::tag,
+      //thrust::system::cpp::reference<T>,
+      typename detail::reference_msvc_workaround<T>::type,
+      thrust::system::cpp::pointer<T>
+    > super_t;
+
+  /*! \endcond
+   */
+
+  public:
+    // note that cpp::pointer's member functions need __host__ __device__
+    // to interoperate with nvcc + iterators' dereference member function
+
+    /*! \p pointer's no-argument constructor initializes its encapsulated pointer to \c 0.
+     */
+    __host__ __device__
+    pointer() : super_t() {}
+
+    /*! This constructor allows construction of a <tt>pointer<const T></tt> from a <tt>T*</tt>.
+     *
+     *  \param ptr A raw pointer to copy from, presumed to point to a location in memory
+     *         accessible by the \p cpp system.
+     *  \tparam OtherT \p OtherT shall be convertible to \p T.
+     */
+    template<typename OtherT>
+    __host__ __device__
+    explicit pointer(OtherT *ptr) : super_t(ptr) {}
+
+    /*! This constructor allows construction from another pointer-like object with related type.
+     *
+     *  \param other The \p OtherPointer to copy.
+     *  \tparam OtherPointer The system tag associated with \p OtherPointer shall be convertible
+     *          to \p thrust::system::cpp::tag and its element type shall be convertible to \p T.
+     */
+    template<typename OtherPointer>
+    __host__ __device__
+    pointer(const OtherPointer &other,
+            typename thrust::detail::enable_if_pointer_is_convertible<
+              OtherPointer,
+              pointer
+            >::type * = 0) : super_t(other) {}
+
+    /*! Assignment operator allows assigning from another pointer-like object with related type.
+     *
+     *  \param other The other pointer-like object to assign from.
+     *  \tparam OtherPointer The system tag associated with \p OtherPointer shall be convertible
+     *          to \p thrust::system::cpp::tag and its element type shall be convertible to \p T.
+     */
+    template<typename OtherPointer>
+    __host__ __device__
+    typename thrust::detail::enable_if_pointer_is_convertible<
+      OtherPointer,
+      pointer,
+      pointer &
+    >::type
+    operator=(const OtherPointer &other)
+    {
+      return super_t::operator=(other);
+    }
+}; // end pointer
+
+
+/*! \p reference is a wrapped reference to an object stored in memory available to the \p cpp system.
+ *  \p reference is the type of the result of dereferencing a \p cpp::pointer.
+ *
+ *  \tparam T Specifies the type of the referenced object.
+ */
+template<typename T>
+  class reference
+    : public thrust::reference<
+               T,
+               thrust::system::cpp::pointer<T>,
+               thrust::system::cpp::reference<T>
+             >
+{
+  /*! \cond
+   */
+
+  private:
+    typedef thrust::reference<
+      T,
+      thrust::system::cpp::pointer<T>,
+      thrust::system::cpp::reference<T>
+    > super_t;
+
+  /*! \endcond
+   */
+
+  public:
+    /*! \cond
+     */
+
+    typedef typename super_t::value_type value_type;
+    typedef typename super_t::pointer    pointer;
+
+    /*! \endcond
+     */
+
+    /*! This constructor initializes this \p reference to refer to an object
+     *  pointed to by the given \p pointer. After this \p reference is constructed,
+     *  it shall refer to the object pointed to by \p ptr.
+     *
+     *  \param ptr A \p pointer to copy from.
+     */
+    __host__ __device__
+    explicit reference(const pointer &ptr)
+      : super_t(ptr)
+    {}
+
+    /*! This constructor accepts a const reference to another \p reference of related type.
+     *  After this \p reference is constructed, it shall refer to the same object as \p other.
+     *
+     *  \param other A \p reference to copy from.
+     *  \tparam OtherT The element type of the other \p reference.
+     *
+     *  \note This constructor is templated primarily to allow initialization of <tt>reference<const T></tt>
+     *        from <tt>reference<T></tt>.
+     */
+    template<typename OtherT>
+    __host__ __device__
+    reference(const reference<OtherT> &other,
+              typename thrust::detail::enable_if_convertible<
+                typename reference<OtherT>::pointer,
+                pointer
+              >::type * = 0)
+      : super_t(other)
+    {}
+
+    /*! Copy assignment operator copy assigns from another \p reference of related type.
+     *
+     *  \param other The other \p reference to assign from.
+     *  \return <tt>*this</tt>
+     *  \tparam OtherT The element type of the other \p reference.
+     */
+    template<typename OtherT>
+    reference &operator=(const reference<OtherT> &other);
+
+    /*! Assignment operator assigns from a \p value_type.
+     *
+     *  \param x The \p value_type to assign from.
+     *  \return <tt>*this</tt>
+     */
+    reference &operator=(const value_type &x);
+}; // end reference
+
+/*! Exchanges the values of two objects referred to by \p reference.
+ *  \p x The first \p reference of interest.
+ *  \p y The second \p reference ot interest.
+ */
+template<typename T>
+__host__ __device__
+void swap(reference<T> x, reference<T> y);
+
+/*! Allocates an area of memory available to Thrust's <tt>cpp</tt> system.
+ *  \param n Number of bytes to allocate.
+ *  \return A <tt>cpp::pointer<void></tt> pointing to the beginning of the newly
+ *          allocated memory. A null <tt>cpp::pointer<void></tt> is returned if
+ *          an error occurs.
+ *  \note The <tt>cpp::pointer<void></tt> returned by this function must be
+ *        deallocated with \p cpp::free.
+ *  \see cpp::free
+ *  \see std::malloc
+ */
+inline pointer<void> malloc(std::size_t n);
+
+/*! Allocates a typed area of memory available to Thrust's <tt>cpp</tt> system.
+ *  \param n Number of elements to allocate.
+ *  \return A <tt>cpp::pointer<T></tt> pointing to the beginning of the newly
+ *          allocated elements. A null <tt>cpp::pointer<T></tt> is returned if
+ *          an error occurs.
+ *  \note The <tt>cpp::pointer<T></tt> returned by this function must be
+ *        deallocated with \p cpp::free.
+ *  \see cpp::free
+ *  \see std::malloc
+ */
+template<typename T>
+inline pointer<T> malloc(std::size_t n);
+
+/*! Deallocates an area of memory previously allocated by <tt>cpp::malloc</tt>.
+ *  \param ptr A <tt>cpp::pointer<void></tt> pointing to the beginning of an area
+ *         of memory previously allocated with <tt>cpp::malloc</tt>.
+ *  \see cpp::malloc
+ *  \see std::free
+ */
+inline void free(pointer<void> ptr);
+
+// XXX upon c++11
+// template<typename T> using allocator = thrust::detail::malloc_allocator<T,tag,pointer<T> >;
+
+/*! \p cpp::allocator is the default allocator used by the \p cpp system's containers such as
+ *  <tt>cpp::vector</tt> if no user-specified allocator is provided. \p cpp::allocator allocates
+ *  (deallocates) storage with \p cpp::malloc (\p cpp::free).
+ */
+template<typename T>
+  struct allocator
+    : thrust::detail::malloc_allocator<
+        T,
+        tag,
+        pointer<T>
+      >
+{
+  /*! The \p rebind metafunction provides the type of an \p allocator
+   *  instantiated with another type.
+   *
+   *  \tparam U The other type to use for instantiation.
+   */
+  template<typename U>
+    struct rebind
+  {
+    /*! The typedef \p other gives the type of the rebound \p allocator.
+     */
+    typedef allocator<U> other;
+  };
+
+  /*! No-argument constructor has no effect.
+   */
+  __host__ __device__
+  inline allocator() {}
+
+  /*! Copy constructor has no effect.
+   */
+  __host__ __device__
+  inline allocator(const allocator &) {}
+
+  /*! Constructor from other \p allocator has no effect.
+   */
+  template<typename U>
+  __host__ __device__
+  inline allocator(const allocator<U> &) {}
+
+  /*! Destructor has no effect.
+   */
+  __host__ __device__
+  inline ~allocator() {}
+}; // end allocator
+
+} // end cpp
+
+/*! \}
+ */
+
+} // end system
+
+/*! \namespace thrust::cpp
+ *  \brief \p thrust::cpp is a top-level alias for thrust::system::cpp.
+ */
+namespace cpp
+{
+
+using thrust::system::cpp::pointer;
+using thrust::system::cpp::reference;
+using thrust::system::cpp::malloc;
+using thrust::system::cpp::free;
+using thrust::system::cpp::allocator;
+
+} // end cpp
+
+} // end thrust
+
+#include <thrust/system/cpp/detail/memory.inl>
+
diff --git a/compat/thrust/system/cpp/vector.h b/compat/thrust/system/cpp/vector.h
new file mode 100644
index 0000000000..4282df991a
--- /dev/null
+++ b/compat/thrust/system/cpp/vector.h
@@ -0,0 +1,149 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+/*! \file thrust/system/cpp/vector.h
+ *  \brief A dynamically-sizable array of elements which reside in memory available to
+ *         Thrust's standard C++ system.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/cpp/memory.h>
+#include <thrust/detail/vector_base.h>
+#include <vector>
+
+namespace thrust
+{
+
+// forward declaration of host_vector
+template<typename T, typename Allocator> class host_vector;
+
+namespace system
+{
+namespace cpp
+{
+
+// XXX upon c++11
+// template<typename T, typename Allocator = allocator<T> > using vector = thrust::detail::vector_base<T,Allocator>;
+
+/*! \p cpp::vector is a container that supports random access to elements,
+ *  constant time removal of elements at the end, and linear time insertion
+ *  and removal of elements at the beginning or in the middle. The number of
+ *  elements in a \p cpp::vector may vary dynamically; memory management is
+ *  automatic. The elements contained in a \p cpp::vector reside in memory
+ *  available to the \p cpp system.
+ *
+ *  \tparam T The element type of the \p cpp::vector.
+ *  \tparam Allocator The allocator type of the \p cpp::vector. Defaults to \p cpp::allocator.
+ *
+ *  \see http://www.sgi.com/tech/stl/Vector.html
+ *  \see host_vector For the documentation of the complete interface which is
+ *                   shared by \p cpp::vector
+ *  \see device_vector
+ */
+template<typename T, typename Allocator = allocator<T> >
+  class vector
+    : public thrust::detail::vector_base<T,Allocator>
+{
+  /*! \cond
+   */
+  private:
+    typedef thrust::detail::vector_base<T,Allocator> super_t;
+  /*! \endcond
+   */
+
+  public:
+
+  /*! \cond
+   */
+    typedef typename super_t::size_type  size_type;
+    typedef typename super_t::value_type value_type;
+
+  /*! \endcond
+   */
+
+    /*! This constructor creates an empty \p cpp::vector.
+     */
+    vector();
+
+    /*! This constructor creates a \p cpp::vector with \p n default-constructed elements.
+     *  \param n The size of the \p cpp::vector to create.
+     */
+    explicit vector(size_type n);
+
+    /*! This constructor creates a \p cpp::vector with \p n copies of \p value.
+     *  \param n The size of the \p cpp::vector to create.
+     *  \param value An element to copy.
+     */
+    explicit vector(size_type n, const value_type &value);
+
+    /*! Copy constructor copies from another \p cpp::vector.
+     *  \param x The other \p cpp::vector to copy.
+     */
+    vector(const vector &x);
+
+    /*! This constructor copies from another Thrust vector-like object.
+     *  \param x The other object to copy from.
+     */
+    template<typename OtherT, typename OtherAllocator>
+    vector(const thrust::detail::vector_base<OtherT,OtherAllocator> &x);
+
+    /*! This constructor copies from a \c std::vector.
+     *  \param x The \c std::vector to copy from.
+     */
+    template<typename OtherT, typename OtherAllocator>
+    vector(const std::vector<OtherT,OtherAllocator> &x);
+
+    /*! This constructor creates a \p cpp::vector by copying from a range.
+     *  \param first The beginning of the range.
+     *  \param last The end of the range.
+     */
+    template<typename InputIterator>
+    vector(InputIterator first, InputIterator last);
+
+    // XXX vector_base should take a Derived type so we don't have to define these superfluous assigns
+
+    /*! Assignment operator assigns from a \c std::vector.
+     *  \param x The \c std::vector to assign from.
+     *  \return <tt>*this</tt>
+     */
+    template<typename OtherT, typename OtherAllocator>
+    vector &operator=(const std::vector<OtherT,OtherAllocator> &x);
+
+    /*! Assignment operator assigns from another Thrust vector-like object.
+     *  \param x The other object to assign from.
+     *  \return <tt>*this</tt>
+     */
+    template<typename OtherT, typename OtherAllocator>
+    vector &operator=(const thrust::detail::vector_base<OtherT,OtherAllocator> &x);
+}; // end vector
+
+} // end cpp
+} // end system
+
+// alias system::cpp names at top-level
+namespace cpp
+{
+
+using thrust::system::cpp::vector;
+
+} // end cpp
+
+} // end thrust
+
+#include <thrust/system/cpp/detail/vector.inl>
+
diff --git a/compat/thrust/system/cuda/detail/adjacent_difference.h b/compat/thrust/system/cuda/detail/adjacent_difference.h
new file mode 100644
index 0000000000..ec51794ff9
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/adjacent_difference.h
@@ -0,0 +1,51 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file adjacent_difference.h
+ *  \brief CUDA implementation of adjacent_difference.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/cuda/detail/execution_policy.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+
+template <typename DerivedPolicy,
+          typename InputIterator,
+          typename OutputIterator,
+          typename BinaryFunction>
+OutputIterator adjacent_difference(execution_policy<DerivedPolicy> &exec,
+                                   InputIterator first, InputIterator last,
+                                   OutputIterator result,
+                                   BinaryFunction binary_op);
+
+} // end namespace detail
+} // end namespace cuda
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/cuda/detail/adjacent_difference.inl>
+
diff --git a/compat/thrust/system/cuda/detail/adjacent_difference.inl b/compat/thrust/system/cuda/detail/adjacent_difference.inl
new file mode 100644
index 0000000000..9e4756a5a3
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/adjacent_difference.inl
@@ -0,0 +1,197 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/detail/config.h>
+
+#include <thrust/gather.h>
+#include <thrust/functional.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/iterator/counting_iterator.h>
+#include <thrust/iterator/transform_iterator.h>
+#include <thrust/detail/temporary_array.h>
+#include <thrust/system/detail/internal/decompose.h>
+#include <thrust/system/cuda/detail/default_decomposition.h>
+#include <thrust/system/cuda/detail/detail/launch_closure.h>
+#include <thrust/system/cuda/detail/detail/launch_calculator.h>
+#include <thrust/system/cuda/detail/execution_policy.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+namespace detail
+{
+
+template <typename Decomposition>
+struct last_index_in_each_interval : public thrust::unary_function<typename Decomposition::index_type, typename Decomposition::index_type>
+{
+  typedef typename Decomposition::index_type index_type;
+
+  Decomposition decomp;
+
+  last_index_in_each_interval(Decomposition decomp) : decomp(decomp) {}
+
+  __host__ __device__
+  index_type operator()(index_type interval)
+  {
+    return decomp[interval].end() - 1;
+  }
+};
+
+template <typename InputIterator1,
+          typename InputIterator2,
+          typename OutputIterator,
+          typename BinaryFunction,
+          typename Decomposition,
+          typename Context>
+struct adjacent_difference_closure
+{
+  InputIterator1 input;
+  InputIterator2 input_copy;
+  OutputIterator output;
+  BinaryFunction binary_op;
+  Decomposition  decomp;
+  Context        context;
+
+  typedef Context context_type;
+  
+  adjacent_difference_closure(InputIterator1 input,
+                              InputIterator2 input_copy,
+                              OutputIterator output,
+                              BinaryFunction binary_op,
+                              Decomposition  decomp,
+                              Context        context = Context())
+    : input(input), input_copy(input_copy), output(output), binary_op(binary_op), decomp(decomp), context(context) {}
+
+  __device__ __thrust_forceinline__
+  void operator()(void)
+  {
+    typedef typename thrust::iterator_value<InputIterator1>::type  InputType;
+    typedef typename Decomposition::index_type index_type;
+
+    // this block processes results in [range.begin(), range.end())
+    thrust::system::detail::internal::index_range<index_type> range = decomp[context.block_index()];
+    
+    input_copy += context.block_index() - 1;
+      
+    // prime the temp values for all threads so we don't need to launch a default constructor
+    InputType next_left = (context.block_index() == 0) ? *input : *input_copy;
+
+    index_type base = range.begin();
+    index_type i    = range.begin() + context.thread_index();
+    
+    if (i < range.end())
+    {
+      if (context.thread_index() > 0)
+      {
+        InputIterator1 temp = input + (i - 1);
+        next_left = *temp;
+      }              
+    }
+    
+    input  += i;
+    output += i;
+
+    while (base < range.end())
+    {
+      InputType curr_left = next_left;
+
+      if (i + context.block_dimension() < range.end())
+      {
+        InputIterator1 temp = input + (context.block_dimension() - 1);
+        next_left = *temp;
+      }
+
+      context.barrier();
+
+      if (i < range.end())
+      {
+        if (i == 0)
+          *output = *input;
+        else
+        {
+          InputType x = *input;
+          *output = binary_op(x, curr_left);
+        }
+      }
+
+      i      += context.block_dimension();
+      base   += context.block_dimension();
+      input  += context.block_dimension();
+      output += context.block_dimension();
+    }
+  }
+};
+
+} // end namespace detail
+
+
+__THRUST_DISABLE_MSVC_POSSIBLE_LOSS_OF_DATA_WARNING_BEGIN
+
+template <typename DerivedPolicy,
+          typename InputIterator,
+          typename OutputIterator,
+          typename BinaryFunction>
+OutputIterator adjacent_difference(execution_policy<DerivedPolicy> &exec,
+                                   InputIterator first, InputIterator last,
+                                   OutputIterator result,
+                                   BinaryFunction binary_op)
+{
+  typedef typename thrust::iterator_value<InputIterator>::type                        InputType;
+  typedef typename thrust::iterator_difference<InputIterator>::type                   IndexType;
+  typedef          thrust::system::detail::internal::uniform_decomposition<IndexType> Decomposition;
+
+  IndexType n = last - first;
+
+  if (n == 0)
+    return result;
+
+  Decomposition decomp = default_decomposition(last - first);
+
+  // allocate temporary storage
+  thrust::detail::temporary_array<InputType,DerivedPolicy> temp(exec, decomp.size() - 1);
+
+  // gather last value in each interval
+  detail::last_index_in_each_interval<Decomposition> unary_op(decomp);
+  thrust::gather(exec,
+                 thrust::make_transform_iterator(thrust::counting_iterator<IndexType>(0), unary_op),
+                 thrust::make_transform_iterator(thrust::counting_iterator<IndexType>(0), unary_op) + (decomp.size() - 1),
+                 first,
+                 temp.begin());
+
+  
+  typedef typename thrust::detail::temporary_array<InputType,DerivedPolicy>::iterator InputIterator2;
+  typedef detail::blocked_thread_array Context;
+  typedef detail::adjacent_difference_closure<InputIterator,InputIterator2,OutputIterator,BinaryFunction,Decomposition,Context> Closure;
+
+  Closure closure(first, temp.begin(), result, binary_op, decomp); 
+
+  detail::launch_closure(closure, decomp.size());
+  
+  return result + n;
+}
+
+__THRUST_DISABLE_MSVC_POSSIBLE_LOSS_OF_DATA_WARNING_END
+
+} // end namespace detail
+} // end namespace cuda
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/cuda/detail/assign_value.h b/compat/thrust/system/cuda/detail/assign_value.h
new file mode 100644
index 0000000000..c90cf65b79
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/assign_value.h
@@ -0,0 +1,198 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/cuda/detail/execution_policy.h>
+#include <thrust/detail/raw_pointer_cast.h>
+#include <thrust/detail/copy.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+
+
+// XXX WAR an issue with msvc 2005 (cl v14.00) which creates multiply-defined
+//     symbols resulting from assign_value
+#if (THRUST_HOST_COMPILER == THRUST_HOST_COMPILER_MSVC) && (_MSC_VER <= 1400)
+
+namespace
+{
+
+template<typename DerivedPolicy, typename Pointer1, typename Pointer2>
+inline __host__ __device__
+  void assign_value_msvc2005_war(thrust::cuda::execution_policy<DerivedPolicy> &exec, Pointer1 dst, Pointer2 src)
+{
+  // XXX war nvbugs/881631
+  struct war_nvbugs_881631
+  {
+    __host__ inline static void host_path(thrust::cuda::execution_policy<DerivedPolicy> &exec, Pointer1 dst, Pointer2 src)
+    {
+      thrust::copy(exec, src, src + 1, dst);
+    }
+
+    __device__ inline static void device_path(thrust::cuda::execution_policy<DerivedPolicy> &, Pointer1 dst, Pointer2 src)
+    {
+      *thrust::raw_pointer_cast(dst) = *thrust::raw_pointer_cast(src);
+    }
+  };
+
+#ifndef __CUDA_ARCH__
+  war_nvbugs_881631::host_path(exec,dst,src);
+#else
+  war_nvbugs_881631::device_path(exec,dst,src);
+#endif // __CUDA_ARCH__
+} // end assign_value_msvc2005_war()
+
+} // end anon namespace
+
+template<typename DerivedPolicy, typename Pointer1, typename Pointer2>
+inline __host__ __device__
+  void assign_value(thrust::cuda::execution_policy<DerivedPolicy> &exec, Pointer1 dst, Pointer2 src)
+{
+  return assign_value_msvc2005_war(exec,dst,src);
+} // end assign_value()
+
+#else
+
+template<typename DerivedPolicy, typename Pointer1, typename Pointer2>
+inline __host__ __device__
+  void assign_value(thrust::cuda::execution_policy<DerivedPolicy> &exec, Pointer1 dst, Pointer2 src)
+{
+  // XXX war nvbugs/881631
+  struct war_nvbugs_881631
+  {
+    __host__ inline static void host_path(thrust::cuda::execution_policy<DerivedPolicy> &exec, Pointer1 dst, Pointer2 src)
+    {
+      thrust::copy(exec, src, src + 1, dst);
+    }
+
+    __device__ inline static void device_path(thrust::cuda::execution_policy<DerivedPolicy> &, Pointer1 dst, Pointer2 src)
+    {
+      *thrust::raw_pointer_cast(dst) = *thrust::raw_pointer_cast(src);
+    }
+  };
+
+#ifndef __CUDA_ARCH__
+  war_nvbugs_881631::host_path(exec,dst,src);
+#else
+  war_nvbugs_881631::device_path(exec,dst,src);
+#endif // __CUDA_ARCH__
+} // end assign_value()
+
+#endif // msvc 2005 WAR
+
+
+// XXX WAR an issue with msvc 2005 (cl v14.00) which creates multiply-defined
+//     symbols resulting from assign_value
+#if (THRUST_HOST_COMPILER == THRUST_HOST_COMPILER_MSVC) && (_MSC_VER <= 1400)
+
+namespace
+{
+
+
+template<typename System1, typename System2, typename Pointer1, typename Pointer2>
+inline __host__ __device__
+  void assign_value_msvc2005_war(cross_system<System1,System2> &systems, Pointer1 dst, Pointer2 src)
+{
+  // XXX war nvbugs/881631
+  struct war_nvbugs_881631
+  {
+    __host__ inline static void host_path(cross_system<System1,System2> &systems, Pointer1 dst, Pointer2 src)
+    {
+      // rotate the systems so that they are ordered the same as (src, dst)
+      // for the call to thrust::copy
+      cross_system<System2,System1> rotated_systems = systems.rotate();
+      thrust::copy(rotated_systems, src, src + 1, dst);
+    }
+
+    __device__ inline static void device_path(cross_system<System1,System2> &systems, Pointer1 dst, Pointer2 src)
+    {
+      // XXX forward the true cuda::execution_policy inside systems here
+      //     instead of materializing a tag
+      thrust::cuda::tag cuda_tag;
+      thrust::system::cuda::detail::assign_value(cuda_tag, dst, src);
+    }
+  };
+
+#if __CUDA_ARCH__
+  war_nvbugs_881631::device_path(systems,dst,src);
+#else
+  war_nvbugs_881631::host_path(systems,dst,src);
+#endif
+} // end assign_value_msvc2005_war
+
+
+} // end anon namespace
+
+
+template<typename System1, typename System2, typename Pointer1, typename Pointer2>
+inline __host__ __device__
+  void assign_value(cross_system<System1,System2> &systems, Pointer1 dst, Pointer2 src)
+{
+  return assign_value_msvc2005_war(systems,dst,src);
+} // end assign_value()
+
+
+#else
+
+
+template<typename System1, typename System2, typename Pointer1, typename Pointer2>
+inline __host__ __device__
+  void assign_value(cross_system<System1,System2> &systems, Pointer1 dst, Pointer2 src)
+{
+  // XXX war nvbugs/881631
+  struct war_nvbugs_881631
+  {
+    __host__ inline static void host_path(cross_system<System1,System2> &systems, Pointer1 dst, Pointer2 src)
+    {
+      // rotate the systems so that they are ordered the same as (src, dst)
+      // for the call to thrust::copy
+      cross_system<System2,System1> rotated_systems = systems.rotate();
+      thrust::copy(rotated_systems, src, src + 1, dst);
+    }
+
+    __device__ inline static void device_path(cross_system<System1,System2> &systems, Pointer1 dst, Pointer2 src)
+    {
+      // XXX forward the true cuda::execution_policy inside systems here
+      //     instead of materializing a tag
+      thrust::cuda::tag cuda_tag;
+      thrust::system::cuda::detail::assign_value(cuda_tag, dst, src);
+    }
+  };
+
+#if __CUDA_ARCH__
+  war_nvbugs_881631::device_path(systems,dst,src);
+#else
+  war_nvbugs_881631::host_path(systems,dst,src);
+#endif
+} // end assign_value()
+
+
+#endif // msvc 2005 WAR
+
+  
+} // end detail
+} // end cuda
+} // end system
+} // end thrust
+
diff --git a/compat/thrust/system/cuda/detail/binary_search.h b/compat/thrust/system/cuda/detail/binary_search.h
new file mode 100644
index 0000000000..a307fc5fc8
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/binary_search.h
@@ -0,0 +1,22 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system has no special version of this algorithm 
+
diff --git a/compat/thrust/system/cuda/detail/block/copy.h b/compat/thrust/system/cuda/detail/block/copy.h
new file mode 100644
index 0000000000..9cc786bfcf
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/block/copy.h
@@ -0,0 +1,223 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+/*! \file copy.h
+ *  \brief CUDA implementation of device-to-device copy,
+ *         based on Gregory Diamos' memcpy code.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+#include <thrust/pair.h>
+
+#include <thrust/detail/type_traits.h>
+#include <thrust/detail/dispatch/is_trivial_copy.h>
+#include <thrust/detail/raw_reference_cast.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+namespace block
+{
+
+namespace trivial_copy_detail
+{
+
+
+template<typename Size>
+  inline __device__ thrust::pair<Size,Size> quotient_and_remainder(Size n, Size d)
+{
+  Size quotient  = n / d;
+  Size remainder = n - d * quotient; 
+  return thrust::make_pair(quotient,remainder);
+} // end quotient_and_remainder()
+
+
+// assumes the addresses dst & src are aligned to T boundaries
+template<typename Context,
+         typename T>
+__device__ __thrust_forceinline__
+void aligned_copy(Context context, T *dst, const T *src, unsigned int num_elements)
+{
+  for(unsigned int i = context.thread_index();
+      i < num_elements;
+      i += context.block_dimension())
+  {
+    dst[i] = src[i];
+  }
+} // end aligned_copy()
+
+
+} // end namespace trivial_copy_detail
+
+
+template <typename Context>
+__device__ __thrust_forceinline__
+void trivial_copy(Context context, void* destination_, const void* source_, size_t num_bytes)
+{
+  // reinterpret at bytes
+  char* destination  = reinterpret_cast<char*>(destination_);
+  const char* source = reinterpret_cast<const char*>(source_);
+ 
+  // TODO replace this with uint64
+#if THRUST_DEVICE_COMPILER != THRUST_DEVICE_COMPILER_NVCC
+  typedef long long  int2;
+  typedef long long uint2;
+#endif // THRUST_DEVICE_COMPILER_NVCC
+
+  // check alignment
+  // XXX can we do this in three steps?
+  //     1. copy until alignment is met
+  //     2. go hog wild
+  //     3. get the remainder
+  if(reinterpret_cast<size_t>(destination) % sizeof(uint2) != 0 || reinterpret_cast<size_t>(source) % sizeof(uint2) != 0)
+  {
+    for(unsigned int i = context.thread_index(); i < num_bytes; i += context.block_dimension())
+    {
+      destination[i] = source[i];
+    }
+  }
+  else
+  {
+    // it's aligned; do a wide copy
+
+    // this pair stores the number of int2s in the aligned portion of the arrays
+    // and the number of bytes in the remainder
+    const thrust::pair<size_t,size_t> num_wide_elements_and_remainder_bytes = trivial_copy_detail::quotient_and_remainder(num_bytes, sizeof(int2));
+
+    // copy int2 elements
+    trivial_copy_detail::aligned_copy(context,
+                                      reinterpret_cast<int2*>(destination),
+                                      reinterpret_cast<const int2*>(source),
+                                      num_wide_elements_and_remainder_bytes.first);
+
+    // XXX we could copy int elements here
+
+    // copy remainder byte by byte
+
+    // to find the beginning of the remainder arrays, we need to point at the beginning, and then skip the number of bytes in the aligned portion
+    // this is sizeof(int2) times the number of int2s comprising the aligned portion
+    const char *remainder_first  = reinterpret_cast<const char*>(source + sizeof(int2) * num_wide_elements_and_remainder_bytes.first);
+          char *remainder_result = reinterpret_cast<char*>(destination  + sizeof(int2) * num_wide_elements_and_remainder_bytes.first);
+
+    trivial_copy_detail::aligned_copy(context, remainder_result, remainder_first, num_wide_elements_and_remainder_bytes.second);
+  }
+} // end trivial_copy()
+
+
+namespace detail
+{
+namespace dispatch
+{
+
+template<typename Context,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2>
+  __thrust_forceinline__ __device__
+  RandomAccessIterator2 copy(Context context,
+                             RandomAccessIterator1 first,
+                             RandomAccessIterator1 last,
+                             RandomAccessIterator2 result,
+                             thrust::detail::true_type is_trivial_copy)
+{
+  typedef typename thrust::iterator_value<RandomAccessIterator1>::type T;
+
+  const T *src = &thrust::raw_reference_cast(*first);
+        T *dst = &thrust::raw_reference_cast(*result);
+
+  size_t n = (last - first);
+  thrust::system::cuda::detail::block::trivial_copy(context, dst, src, n * sizeof(T));
+  return result + n;
+} // end copy()
+
+template<typename Context,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2>
+  __thrust_forceinline__ __device__
+  RandomAccessIterator2 copy(Context context, 
+                             RandomAccessIterator1 first,
+                             RandomAccessIterator1 last,
+                             RandomAccessIterator2 result,
+                             thrust::detail::false_type is_trivial_copy)
+{
+  RandomAccessIterator2 end_of_output = result + (last - first);
+  
+  // advance iterators
+  first  += context.thread_index();
+  result += context.thread_index();
+
+  for(;
+      first < last;
+      first  += context.block_dimension(),
+      result += context.block_dimension())
+  {
+    *result = *first;
+  } // end for
+
+  return end_of_output;
+} // end copy()
+
+} // end namespace dispatch
+} // end namespace detail
+
+template<typename Context, 
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2>
+  __thrust_forceinline__ __device__
+  RandomAccessIterator2 copy(Context context,
+                             RandomAccessIterator1 first,
+                             RandomAccessIterator1 last,
+                             RandomAccessIterator2 result)
+{
+  return detail::dispatch::copy(context, first, last, result,
+#if __CUDA_ARCH__ < 200
+      // does not work reliably on pre-Fermi due to "Warning: ... assuming global memory space" issues
+      thrust::detail::false_type()
+#else
+      typename thrust::detail::dispatch::is_trivial_copy<RandomAccessIterator1,RandomAccessIterator2>::type()
+#endif
+      );
+} // end copy()
+
+
+template<typename Context, typename RandomAccessIterator1, typename Size, typename RandomAccessIterator2>
+inline __device__
+RandomAccessIterator2 copy_n(Context &ctx, RandomAccessIterator1 first, Size n, RandomAccessIterator2 result)
+{
+  for(Size i = ctx.thread_index(); i < n; i += ctx.block_dimension())
+  {
+    result[i] = first[i];
+  }
+
+  ctx.barrier();
+
+  return result + n;
+}
+
+
+} // end namespace block
+} // end namespace detail
+} // end namespace cuda
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/cuda/detail/block/exclusive_scan.h b/compat/thrust/system/cuda/detail/block/exclusive_scan.h
new file mode 100644
index 0000000000..580a7578bf
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/block/exclusive_scan.h
@@ -0,0 +1,74 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/functional.h>
+#include <thrust/system/cuda/detail/block/inclusive_scan.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+namespace block
+{
+
+
+template<typename Context, typename RandomAccessIterator, typename T, typename BinaryFunction>
+inline __device__
+typename thrust::iterator_value<RandomAccessIterator>::type
+  inplace_exclusive_scan(Context &ctx, RandomAccessIterator first, T init, BinaryFunction op)
+{
+  // perform an inclusive scan, then shift right
+  block::inplace_inclusive_scan(ctx, first, op);
+
+  typename thrust::iterator_value<RandomAccessIterator>::type carry = first[ctx.block_dimension() - 1];
+
+  ctx.barrier();
+
+  typename thrust::iterator_value<RandomAccessIterator>::type left = (ctx.thread_index() == 0) ? init : first[ctx.thread_index() - 1];
+
+  ctx.barrier();
+
+  first[ctx.thread_index()] = left;
+
+  ctx.barrier();
+
+  return carry;
+}
+
+
+template<typename Context, typename Iterator, typename T>
+inline __device__
+  typename thrust::iterator_value<Iterator>::type
+    inplace_exclusive_scan(Context &ctx, Iterator first, T init)
+{
+  return block::inplace_exclusive_scan(ctx, first, init, thrust::plus<typename thrust::iterator_value<Iterator>::type>());
+}
+
+
+} // end namespace block
+} // end namespace detail
+} // end namespace cuda
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/cuda/detail/block/inclusive_scan.h b/compat/thrust/system/cuda/detail/block/inclusive_scan.h
new file mode 100644
index 0000000000..012f7cd2f7
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/block/inclusive_scan.h
@@ -0,0 +1,191 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/iterator/iterator_traits.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+namespace block
+{
+
+template<typename Context,
+         typename InputIterator,
+         typename BinaryFunction>
+__device__ __thrust_forceinline__
+void inclusive_scan(Context context,
+                    InputIterator first,
+                    BinaryFunction binary_op)
+{
+  // TODO generalize to arbitrary n
+  // TODO support dynamic block_size
+  const unsigned int block_size = Context::ThreadsPerBlock::value;
+
+  typename thrust::iterator_value<InputIterator>::type val = first[context.thread_index()];
+
+  if(block_size >    1) { if (context.thread_index() >=    1) { val = binary_op(first[context.thread_index() -    1], val); } context.barrier(); first[context.thread_index()] = val; context.barrier(); }
+  if(block_size >    2) { if (context.thread_index() >=    2) { val = binary_op(first[context.thread_index() -    2], val); } context.barrier(); first[context.thread_index()] = val; context.barrier(); } 
+  if(block_size >    4) { if (context.thread_index() >=    4) { val = binary_op(first[context.thread_index() -    4], val); } context.barrier(); first[context.thread_index()] = val; context.barrier(); }
+  if(block_size >    8) { if (context.thread_index() >=    8) { val = binary_op(first[context.thread_index() -    8], val); } context.barrier(); first[context.thread_index()] = val; context.barrier(); }
+  if(block_size >   16) { if (context.thread_index() >=   16) { val = binary_op(first[context.thread_index() -   16], val); } context.barrier(); first[context.thread_index()] = val; context.barrier(); }
+  if(block_size >   32) { if (context.thread_index() >=   32) { val = binary_op(first[context.thread_index() -   32], val); } context.barrier(); first[context.thread_index()] = val; context.barrier(); }
+  if(block_size >   64) { if (context.thread_index() >=   64) { val = binary_op(first[context.thread_index() -   64], val); } context.barrier(); first[context.thread_index()] = val; context.barrier(); }
+  if(block_size >  128) { if (context.thread_index() >=  128) { val = binary_op(first[context.thread_index() -  128], val); } context.barrier(); first[context.thread_index()] = val; context.barrier(); }
+  if(block_size >  256) { if (context.thread_index() >=  256) { val = binary_op(first[context.thread_index() -  256], val); } context.barrier(); first[context.thread_index()] = val; context.barrier(); }
+  if(block_size >  512) { if (context.thread_index() >=  512) { val = binary_op(first[context.thread_index() -  512], val); } context.barrier(); first[context.thread_index()] = val; context.barrier(); }
+  if(block_size > 1024) { if (context.thread_index() >= 1024) { val = binary_op(first[context.thread_index() - 1024], val); } context.barrier(); first[context.thread_index()] = val; context.barrier(); }
+} // end inclusive_scan()
+
+
+template<typename Context,
+         typename InputIterator,
+         typename Size,
+         typename BinaryFunction>
+__device__ __thrust_forceinline__
+void inclusive_scan_n(Context context,
+                      InputIterator first,
+                      Size n,
+                      BinaryFunction binary_op)
+{
+  // TODO support n > context.block_dimension()
+  typename thrust::iterator_value<InputIterator>::type val = first[context.thread_index()];
+
+  for (unsigned int i = 1; i < n; i <<= 1)
+  {
+    if (context.thread_index() < n && context.thread_index() >= i)
+      val = binary_op(first[context.thread_index() - i], val);
+
+    context.barrier();
+    
+    first[context.thread_index()] = val;
+    
+    context.barrier();
+  }
+} // end inclusive_scan()
+
+
+template<typename Context,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename BinaryFunction>
+__device__ __thrust_forceinline__
+void inclusive_scan_by_flag(Context context,
+                            InputIterator1 first1,
+                            InputIterator2 first2,
+                            BinaryFunction binary_op)
+{
+  // TODO generalize to arbitrary n
+  // TODO support dynamic block_size
+  const unsigned int block_size = Context::ThreadsPerBlock::value;
+
+  typename thrust::iterator_value<InputIterator1>::type flg = first1[context.thread_index()];
+  typename thrust::iterator_value<InputIterator2>::type val = first2[context.thread_index()];
+
+  if(block_size >    1) { if (context.thread_index() >=    1) { if (!flg) { flg |= first1[context.thread_index() -    1]; val = binary_op(first2[context.thread_index() -    1], val); } } context.barrier(); first1[context.thread_index()] = flg; first2[context.thread_index()] = val; context.barrier(); }
+  if(block_size >    2) { if (context.thread_index() >=    2) { if (!flg) { flg |= first1[context.thread_index() -    2]; val = binary_op(first2[context.thread_index() -    2], val); } } context.barrier(); first1[context.thread_index()] = flg; first2[context.thread_index()] = val; context.barrier(); } 
+  if(block_size >    4) { if (context.thread_index() >=    4) { if (!flg) { flg |= first1[context.thread_index() -    4]; val = binary_op(first2[context.thread_index() -    4], val); } } context.barrier(); first1[context.thread_index()] = flg; first2[context.thread_index()] = val; context.barrier(); }
+  if(block_size >    8) { if (context.thread_index() >=    8) { if (!flg) { flg |= first1[context.thread_index() -    8]; val = binary_op(first2[context.thread_index() -    8], val); } } context.barrier(); first1[context.thread_index()] = flg; first2[context.thread_index()] = val; context.barrier(); }
+  if(block_size >   16) { if (context.thread_index() >=   16) { if (!flg) { flg |= first1[context.thread_index() -   16]; val = binary_op(first2[context.thread_index() -   16], val); } } context.barrier(); first1[context.thread_index()] = flg; first2[context.thread_index()] = val; context.barrier(); }
+  if(block_size >   32) { if (context.thread_index() >=   32) { if (!flg) { flg |= first1[context.thread_index() -   32]; val = binary_op(first2[context.thread_index() -   32], val); } } context.barrier(); first1[context.thread_index()] = flg; first2[context.thread_index()] = val; context.barrier(); }
+  if(block_size >   64) { if (context.thread_index() >=   64) { if (!flg) { flg |= first1[context.thread_index() -   64]; val = binary_op(first2[context.thread_index() -   64], val); } } context.barrier(); first1[context.thread_index()] = flg; first2[context.thread_index()] = val; context.barrier(); }
+  if(block_size >  128) { if (context.thread_index() >=  128) { if (!flg) { flg |= first1[context.thread_index() -  128]; val = binary_op(first2[context.thread_index() -  128], val); } } context.barrier(); first1[context.thread_index()] = flg; first2[context.thread_index()] = val; context.barrier(); }
+  if(block_size >  256) { if (context.thread_index() >=  256) { if (!flg) { flg |= first1[context.thread_index() -  256]; val = binary_op(first2[context.thread_index() -  256], val); } } context.barrier(); first1[context.thread_index()] = flg; first2[context.thread_index()] = val; context.barrier(); }
+  if(block_size >  512) { if (context.thread_index() >=  512) { if (!flg) { flg |= first1[context.thread_index() -  512]; val = binary_op(first2[context.thread_index() -  512], val); } } context.barrier(); first1[context.thread_index()] = flg; first2[context.thread_index()] = val; context.barrier(); }
+  if(block_size > 1024) { if (context.thread_index() >= 1024) { if (!flg) { flg |= first1[context.thread_index() - 1024]; val = binary_op(first2[context.thread_index() - 1024], val); } } context.barrier(); first1[context.thread_index()] = flg; first2[context.thread_index()] = val; context.barrier(); }
+} // end inclusive_scan_by_flag()
+
+
+template<typename Context,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename Size,
+         typename BinaryFunction>
+__device__ __thrust_forceinline__
+void inclusive_scan_by_flag_n(Context context,
+                              InputIterator1 first1,
+                              InputIterator2 first2,
+                              Size n,
+                              BinaryFunction binary_op)
+{
+  // TODO support n > context.block_dimension()
+  typename thrust::iterator_value<InputIterator1>::type flg = first1[context.thread_index()];
+  typename thrust::iterator_value<InputIterator2>::type val = first2[context.thread_index()];
+  
+  for (unsigned int i = 1; i < n; i <<= 1)
+  {
+    if (context.thread_index() < n && context.thread_index() >= i) 
+    {
+      if (!flg)
+      { 
+        flg |= first1[context.thread_index() - i];
+        val  = binary_op(first2[context.thread_index() - i], val);
+      }
+    }
+
+    context.barrier();
+    
+    first1[context.thread_index()] = flg;
+    first2[context.thread_index()] = val;
+    
+    context.barrier();
+  }
+} // end inclusive_scan_by_flag()
+
+
+template<typename Context, typename RandomAccessIterator, typename BinaryFunction>
+__device__ __thrust_forceinline__
+void inplace_inclusive_scan(Context &ctx, RandomAccessIterator first, BinaryFunction op)
+{
+  typename thrust::iterator_value<RandomAccessIterator>::type x = first[ctx.thread_index()];
+
+  for(unsigned int offset = 1; offset < ctx.block_dimension(); offset *= 2)
+  {
+    if(ctx.thread_index() >= offset)
+    {
+      x = op(first[ctx.thread_index() - offset], x);
+    }
+
+    ctx.barrier();
+
+    first[ctx.thread_index()] = x;
+
+    ctx.barrier();
+  }
+}
+
+
+template<typename Context, typename RandomAccessIterator>
+__device__ __thrust_forceinline__
+void inplace_inclusive_scan(Context &ctx, RandomAccessIterator first)
+{
+  block::inplace_inclusive_scan(ctx, first, thrust::plus<typename thrust::iterator_value<RandomAccessIterator>::type>());
+}
+
+
+} // end namespace block
+} // end namespace detail
+} // end namespace cuda
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/cuda/detail/block/merge.h b/compat/thrust/system/cuda/detail/block/merge.h
new file mode 100644
index 0000000000..9af0b7bfbb
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/block/merge.h
@@ -0,0 +1,74 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+namespace block
+{
+
+template<typename Context,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2,
+         typename RandomAccessIterator3,
+         typename StrictWeakOrdering>
+__device__ __thrust_forceinline__
+  RandomAccessIterator3 merge(Context context,
+                              RandomAccessIterator1 first1,
+                              RandomAccessIterator1 last1,
+                              RandomAccessIterator2 first2,
+                              RandomAccessIterator2 last2,
+                              RandomAccessIterator3 result,
+                              StrictWeakOrdering comp);
+
+// XXX assumes that context.block_dimension() <= n1 and
+//                  context.block_dimension() <= n2
+// This algorithm is analogous to inplace_merge
+// but instead of working on the ranges
+// [first, middle) and [middle, last)
+// it works on the ranges
+// [first, first + n1) and [first + n1, first + n1 + n2)
+template<typename Context,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2,
+         typename Size1,
+         typename Size2,
+         typename StrictWeakOrdering>
+__device__ __thrust_forceinline__
+  void inplace_merge_by_key_n(Context context,
+                              RandomAccessIterator1 keys_first,
+                              RandomAccessIterator2 values_first,
+                              Size1 n1,
+                              Size2 n2,
+                              StrictWeakOrdering comp);
+
+} // end namespace block
+} // end namespace detail
+} // end namespace cuda
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/cuda/detail/block/merge.inl>
+
diff --git a/compat/thrust/system/cuda/detail/block/merge.inl b/compat/thrust/system/cuda/detail/block/merge.inl
new file mode 100644
index 0000000000..5eae2b58f3
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/block/merge.inl
@@ -0,0 +1,168 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/detail/raw_reference_cast.h>
+#include <thrust/system/detail/generic/scalar/binary_search.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+namespace block
+{
+
+template<typename Context,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2,
+         typename RandomAccessIterator3,
+         typename StrictWeakOrdering>
+__device__ __thrust_forceinline__
+  RandomAccessIterator3 merge(Context context,
+                              RandomAccessIterator1 first1,
+                              RandomAccessIterator1 last1,
+                              RandomAccessIterator2 first2,
+                              RandomAccessIterator2 last2,
+                              RandomAccessIterator3 result,
+                              StrictWeakOrdering comp)
+{
+  typedef typename thrust::iterator_difference<RandomAccessIterator1>::type difference1;
+  typedef typename thrust::iterator_difference<RandomAccessIterator2>::type difference2;
+
+  difference1 n1 = last1 - first1;
+  difference2 n2 = last2 - first2;
+
+  // find the rank of each element in the other array
+  difference2 rank2 = 0;
+  if(context.thread_index() < n1)
+  {
+    RandomAccessIterator1 x = first1;
+    x += context.thread_index();
+
+    // lower_bound ensures that x sorts before any equivalent element of input2
+    // this ensures stability
+    rank2 = thrust::system::detail::generic::scalar::lower_bound(first2, last2, raw_reference_cast(*x), comp) - first2;
+  } // end if
+
+  difference1 rank1 = 0;
+  if(context.thread_index() < n2)
+  {
+    RandomAccessIterator2 x = first2 + context.thread_index();
+
+    // upper_bound ensures that x sorts before any equivalent element of input1
+    // this ensures stability
+    rank1 = thrust::system::detail::generic::scalar::upper_bound(first1, last1, raw_reference_cast(*x), comp) - first1;
+  } // end if
+
+  if(context.thread_index() < n1)
+  {
+    // scatter each element from input1
+    RandomAccessIterator1 src = first1 + context.thread_index();
+    RandomAccessIterator3 dst = result + context.thread_index() + rank2;
+
+    *dst = *src;
+  }
+
+  if(context.thread_index() < n2)
+  {
+    // scatter each element from input2
+    RandomAccessIterator2 src = first2 + context.thread_index();
+    RandomAccessIterator3 dst = result + context.thread_index() + rank1;
+
+    *dst = *src;
+  }
+
+  return result + n1 + n2;
+} // end merge
+
+
+template<typename Context,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2,
+         typename Size1,
+         typename Size2,
+         typename StrictWeakOrdering>
+__device__ __thrust_forceinline__
+  void inplace_merge_by_key_n(Context context,
+                              RandomAccessIterator1 keys_first,
+                              RandomAccessIterator2 values_first,
+                              Size1 n1,
+                              Size2 n2,
+                              StrictWeakOrdering comp)
+{
+  RandomAccessIterator1 input1 = keys_first;
+  RandomAccessIterator1 input2 = keys_first + n1;
+
+  RandomAccessIterator2 input1val = values_first;
+  RandomAccessIterator2 input2val = values_first + n1;
+  
+  typedef typename thrust::iterator_value<RandomAccessIterator1>::type KeyType;
+  typedef typename thrust::iterator_value<RandomAccessIterator2>::type ValueType;
+
+  // XXX use uninitialized here
+  KeyType inp1 = input1[context.thread_index()]; ValueType inp1val = input1val[context.thread_index()];
+  KeyType inp2 = input2[context.thread_index()]; ValueType inp2val = input2val[context.thread_index()];
+  
+  // to merge input1 and input2, use binary search to find the rank of inp1 & inp2 in arrays input2 & input1, respectively
+  // as before, the "end" variables point to one element after the last element of the arrays
+  
+  // start by looking through input2 for inp1's rank
+  unsigned int start_1 = 0;
+  
+  // don't do the search if our value is beyond the end of input1
+  if(context.thread_index() < n1)
+  {
+    start_1 = thrust::system::detail::generic::scalar::lower_bound_n(input2, n2, inp1, comp) - input2;
+  } // end if
+  
+  // now look through input1 for inp2's rank
+  unsigned int start_2 = 0;
+  
+  // don't do the search if our value is beyond the end of input2
+  if(context.thread_index() < n2)
+  {
+    // upper_bound ensures that equivalent elements in the first range sort before the second
+    start_2 = thrust::system::detail::generic::scalar::upper_bound_n(input1, n1, inp2, comp) - input1;
+  } // end if
+
+  context.barrier();
+  
+  // Write back into the right position to the input arrays; can be done in place since we read in
+  // the input arrays into registers before.
+  if(context.thread_index() < n1)
+  {
+    input1[start_1 + context.thread_index()] = inp1;
+    input1val[start_1 + context.thread_index()] = inp1val;
+  } // end if
+  
+  if(context.thread_index() < n2)
+  {
+    input1[start_2 + context.thread_index()] = inp2;
+    input1val[start_2 + context.thread_index()] = inp2val;
+  } // end if
+} // end inplace_merge_by_key_n()
+
+
+} // end namespace block
+} // end namespace detail
+} // end namespace cuda
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/cuda/detail/block/merging_sort.h b/compat/thrust/system/cuda/detail/block/merging_sort.h
new file mode 100644
index 0000000000..8f8f999ec5
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/block/merging_sort.h
@@ -0,0 +1,199 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file merging_sort.h
+ *  \brief Block version of merge sort
+ */
+
+#pragma once
+
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/detail/type_traits.h>
+#include <thrust/system/detail/generic/scalar/binary_search.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+namespace block
+{
+
+
+template<typename RandomAccessIterator1, typename RandomAccessIterator2, typename Compare>
+__device__ void conditional_swap(RandomAccessIterator1 keys_first,
+                                 RandomAccessIterator2 values_first,
+                                 const unsigned int i,
+                                 const unsigned int end,
+                                 bool pred,
+                                 Compare comp)
+{
+  typedef typename thrust::iterator_traits<RandomAccessIterator1>::value_type KeyType;
+  typedef typename thrust::iterator_traits<RandomAccessIterator2>::value_type ValueType;
+
+  if(pred && i+1<end)
+  {
+    KeyType xi = keys_first[i];
+    KeyType xj = keys_first[i+1];
+
+    // swap if xj sorts before xi
+    if(comp(xj, xi))
+    {
+      // XXX this implementation should really dispatch swap via ADL
+      ValueType yi;
+      yi = values_first[i];
+      ValueType yj;
+      yj = values_first[i+1];
+
+      keys_first[i]     = xj;
+      keys_first[i+1]   = xi;
+      values_first[i]   = yj;
+      values_first[i+1] = yi;
+    }
+  }
+}
+
+
+template<typename Context,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2,
+         typename Compare>
+__device__ void transposition_sort(Context context,
+                                   RandomAccessIterator1 keys_first,
+                                   RandomAccessIterator2 values_first,
+                                   const unsigned int i,
+                                   const unsigned int end,
+                                   const unsigned int size,
+                                   Compare comp)
+{
+  const bool is_odd = i&0x1;
+  
+  for(unsigned int round=size/2; round>0; --round)
+  {
+    // ODDS
+    conditional_swap(keys_first, values_first, i, end, is_odd, comp);
+    context.barrier();
+  
+    // EVENS
+    conditional_swap(keys_first, values_first, i, end, !is_odd, comp);
+    context.barrier();
+  }
+}
+
+template<typename Context,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2,
+         typename StrictWeakOrdering>
+__device__ void merge(Context context,
+                      RandomAccessIterator1 keys_first, 
+                      RandomAccessIterator2 values_first,
+                      const unsigned int i,
+                      const unsigned int n,
+                      unsigned int begin,
+                      unsigned int end,
+                      unsigned int h,
+                      StrictWeakOrdering cmp)
+{
+  // INVARIANT: Every element i resides within a sequence [begin,end)
+  //            of length h which is already sorted
+  while( h<n )
+  {
+    h *= 2;
+
+    unsigned int new_begin = i&(~(h-1));
+    unsigned int new_end   = min(n,new_begin+h);
+
+    typedef typename thrust::iterator_traits<RandomAccessIterator1>::value_type KeyType;
+    typedef typename thrust::iterator_traits<RandomAccessIterator2>::value_type ValueType;
+
+    KeyType key;
+    ValueType value;
+
+    unsigned int rank = i - begin;
+
+    // prevent out-of-bounds access
+    if(i < new_end)
+    {
+      key = keys_first[i];
+
+      if(begin==new_begin)  // in the left side of merging pair
+      {
+        RandomAccessIterator1 result = thrust::system::detail::generic::scalar::lower_bound_n(keys_first+end, new_end-end, key, cmp);
+        rank += (result - (keys_first+end));
+      }
+      else                  // in the right side of merging pair
+      {
+        RandomAccessIterator1 result = thrust::system::detail::generic::scalar::upper_bound_n(keys_first+new_begin, begin-new_begin, key, cmp);
+        rank += (result - (keys_first+new_begin));
+      }
+
+      value = values_first[i];
+    }
+
+    context.barrier();
+
+    if(i < new_end)
+    {
+      keys_first[new_begin+rank] = key;
+      values_first[new_begin+rank] = value;
+    }
+    
+    context.barrier();
+
+    begin = new_begin;
+    end   = new_end;
+  }
+}
+
+
+/*! Block-wise implementation of merge sort.
+ *  It provides the same external interface as odd_even_sort.
+ */
+template<typename Context,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2,
+         typename StrictWeakOrdering>
+__device__ void merging_sort(Context context,
+                             RandomAccessIterator1 keys_first,
+                             RandomAccessIterator2 values_first,
+                             const unsigned int n,
+                             StrictWeakOrdering comp)
+{
+  // Phase 1: Sort subsequences of length 32 using odd-even
+  //          transposition sort.  The code below assumes that h is a
+  //          power of 2.  Empirically, 32 delivers best results,
+  //          which is not surprising since that's the warp width.
+  unsigned int i = context.thread_index();
+  unsigned int h = 32;
+  unsigned int begin=i&(~(h-1)),  end=min(n,begin+h);
+  
+  transposition_sort(context, keys_first, values_first, i, end, h, comp);
+  
+  // Phase 2: Apply merge tree to produce final sorted results
+  merge(context, keys_first, values_first, i, n, begin, end, h, comp);
+} // end merging_sort()
+
+
+} // end namespace block
+} // end namespace detail
+} // end namespace cuda
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/cuda/detail/block/odd_even_sort.h b/compat/thrust/system/cuda/detail/block/odd_even_sort.h
new file mode 100644
index 0000000000..0fa0ea069a
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/block/odd_even_sort.h
@@ -0,0 +1,151 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file odd_even_sort.h
+ *  \brief Block versions of Batcher's Odd-Even Merge Sort
+ */
+
+#pragma once
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+namespace block
+{
+
+
+/*! Block-wise implementation of Batcher's Odd-Even Merge Sort
+ *  This implementation is based on Nadathur Satish's.
+ */
+template<typename KeyType,
+         typename ValueType,
+         typename StrictWeakOrdering>
+  __device__ void odd_even_sort(KeyType *keys,
+                                ValueType *data,
+                                const unsigned int n,
+                                StrictWeakOrdering comp)
+{
+  for(unsigned int p = blockDim.x>>1; p > 0; p >>= 1)
+  {
+    unsigned int q = blockDim.x>>1, r = 0, d = p;
+
+    while(q >= p)
+    {
+      unsigned int j = threadIdx.x + d;
+
+      // if j lies beyond the end of the array, we consider it "sorted" wrt i
+      // regardless of whether i lies beyond the end of the array 
+      if(threadIdx.x < (blockDim.x-d) && (threadIdx.x & p) == r && j < n)
+      {
+        KeyType xikey = keys[threadIdx.x];
+        KeyType xjkey = keys[j];
+
+        ValueType xivalue = data[threadIdx.x];
+        ValueType xjvalue = data[j];
+
+        // does xj sort before xi?
+        if(comp(xjkey, xikey))
+        {
+          keys[threadIdx.x] = xjkey;
+          keys[j] = xikey;
+
+          data[threadIdx.x] = xjvalue;
+          data[j] = xivalue;
+        } // end if
+      } // end if
+
+      d = q - p;
+      q >>= 1;
+      r = p;
+
+      __syncthreads();
+    } // end while
+  } // end for p
+} // end odd_even_sort()
+
+template<typename KeyType,
+         typename ValueType,
+         typename StrictWeakOrdering>
+  __device__ void stable_odd_even_sort(KeyType *keys,
+                                       ValueType *data,
+                                       const unsigned int n,
+                                       StrictWeakOrdering comp)
+{
+  for(unsigned int i = 0;
+      i < blockDim.x>>1;
+      ++i)
+  {
+    bool thread_is_odd = threadIdx.x & 0x1;
+
+    // do odds first
+    if(thread_is_odd && threadIdx.x + 1 < n)
+    {
+      KeyType xikey = keys[threadIdx.x];
+      KeyType xjkey = keys[threadIdx.x + 1];
+
+      ValueType xivalue = data[threadIdx.x];
+      ValueType xjvalue = data[threadIdx.x + 1];
+
+      // does xj sort before xi?
+      if(comp(xjkey, xikey))
+      {
+        keys[threadIdx.x] = xjkey;
+        keys[threadIdx.x + 1] = xikey;
+
+        data[threadIdx.x] = xjvalue;
+        data[threadIdx.x + 1] = xivalue;
+      } // end if
+    } // end if
+
+    __syncthreads();
+
+    // do evens second
+    if(!thread_is_odd && threadIdx.x + 1 < n)
+    {
+      KeyType xikey = keys[threadIdx.x];
+      KeyType xjkey = keys[threadIdx.x + 1];
+
+      ValueType xivalue = data[threadIdx.x];
+      ValueType xjvalue = data[threadIdx.x + 1];
+
+      // does xj sort before xi?
+      if(comp(xjkey, xikey))
+      {
+        keys[threadIdx.x] = xjkey;
+        keys[threadIdx.x + 1] = xikey;
+
+        data[threadIdx.x] = xjvalue;
+        data[threadIdx.x + 1] = xivalue;
+      } // end if
+    } // end if
+
+    __syncthreads();
+  } // end for i
+} // end stable_odd_even_sort()
+
+
+} // end namespace block
+} // end namespace detail
+} // end namespace cuda
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/cuda/detail/block/reduce.h b/compat/thrust/system/cuda/detail/block/reduce.h
new file mode 100644
index 0000000000..e0a1901b28
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/block/reduce.h
@@ -0,0 +1,67 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/iterator/iterator_traits.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+namespace block
+{
+
+/* Reduces [data, data + n) using binary_op and stores the result in data[0]
+ *
+ * Upon return the elements in [data + 1, data + n) have unspecified values.
+ */
+template <typename Context, typename ValueIterator, typename BinaryFunction>
+__device__ __thrust_forceinline__
+void reduce_n(Context context, ValueIterator data, unsigned int n, BinaryFunction binary_op)
+{
+  if (context.block_dimension() < n)
+  {
+    for (unsigned int i = context.block_dimension() + context.thread_index(); i < n; i += context.block_dimension())
+      data[context.thread_index()] = binary_op(data[context.thread_index()], data[i]);
+
+    context.barrier();
+  }
+
+  while (n > 1)
+  {
+    unsigned int half = n / 2;
+
+    if (context.thread_index() < half)
+      data[context.thread_index()] = binary_op(data[context.thread_index()], data[n - context.thread_index() - 1]);
+
+    context.barrier();
+
+    n = n - half;
+  }
+}
+
+} // end namespace block
+} // end namespace detail
+} // end namespace cuda
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/cuda/detail/copy.h b/compat/thrust/system/cuda/detail/copy.h
new file mode 100644
index 0000000000..8f7ee97c24
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/copy.h
@@ -0,0 +1,79 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/cuda/detail/execution_policy.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator>
+  OutputIterator copy(execution_policy<DerivedPolicy> &exec,
+                      InputIterator first,
+                      InputIterator last,
+                      OutputIterator result);
+
+
+template<typename System1,
+         typename System2,
+         typename InputIterator,
+         typename OutputIterator>
+  OutputIterator copy(cross_system<System1,System2> exec,
+                      InputIterator first,
+                      InputIterator last,
+                      OutputIterator result);
+
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename Size,
+         typename OutputIterator>
+  OutputIterator copy_n(execution_policy<DerivedPolicy> &exec,
+                        InputIterator first,
+                        Size n,
+                        OutputIterator result);
+
+
+template<typename System1,
+         typename System2,
+         typename InputIterator,
+         typename Size,
+         typename OutputIterator>
+  OutputIterator copy_n(cross_system<System1,System2> exec,
+                        InputIterator first,
+                        Size n,
+                        OutputIterator result);
+
+
+} // end detail
+} // end cuda
+} // end system
+} // end thrust
+
+#include <thrust/system/cuda/detail/copy.inl>
+
diff --git a/compat/thrust/system/cuda/detail/copy.inl b/compat/thrust/system/cuda/detail/copy.inl
new file mode 100644
index 0000000000..125eebdaa5
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/copy.inl
@@ -0,0 +1,88 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/system/cuda/detail/copy.h>
+#include <thrust/system/cuda/detail/copy_device_to_device.h>
+#include <thrust/system/cuda/detail/copy_cross_system.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+
+
+template<typename System,
+         typename InputIterator,
+         typename OutputIterator>
+  OutputIterator copy(execution_policy<System> &system,
+                      InputIterator first,
+                      InputIterator last,
+                      OutputIterator result)
+{
+  return thrust::system::cuda::detail::copy_device_to_device(system,first,last,result);
+} // end copy()
+
+
+template<typename System1,
+         typename System2,
+         typename InputIterator,
+         typename OutputIterator>
+  OutputIterator copy(cross_system<System1,System2> systems,
+                      InputIterator first,
+                      InputIterator last,
+                      OutputIterator result)
+{
+  return thrust::system::cuda::detail::copy_cross_system(systems,first,last,result);
+} // end copy()
+
+
+template<typename System,
+         typename InputIterator,
+         typename Size,
+         typename OutputIterator>
+  OutputIterator copy_n(execution_policy<System> &system,
+                        InputIterator first,
+                        Size n,
+                        OutputIterator result)
+{
+  return thrust::system::cuda::detail::copy_device_to_device(system,first,first+n,result);
+} // end copy_n()
+
+
+template<typename System1,
+         typename System2,
+         typename InputIterator,
+         typename Size,
+         typename OutputIterator>
+  OutputIterator copy_n(cross_system<System1,System2> systems,
+                        InputIterator first,
+                        Size n,
+                        OutputIterator result)
+{
+  return thrust::system::cuda::detail::copy_cross_system_n(systems,first,n,result);
+} // end copy_n()
+
+
+} // end detail
+} // end cuda
+} // end system
+} // end thrust
+
diff --git a/compat/thrust/system/cuda/detail/copy_cross_system.h b/compat/thrust/system/cuda/detail/copy_cross_system.h
new file mode 100644
index 0000000000..f68ea3c88a
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/copy_cross_system.h
@@ -0,0 +1,59 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/cuda/detail/execution_policy.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+
+
+template<typename System1,
+         typename System2,
+         typename InputIterator,
+         typename OutputIterator>
+  OutputIterator copy_cross_system(cross_system<System1,System2> systems,
+                                   InputIterator begin, 
+                                   InputIterator end, 
+                                   OutputIterator result);
+
+
+template<typename System1,
+         typename System2,
+         typename InputIterator,
+         typename Size,
+         typename OutputIterator>
+  OutputIterator copy_cross_system_n(cross_system<System1,System2> systems,
+                                     InputIterator begin, 
+                                     Size n, 
+                                     OutputIterator result);
+
+
+} // end detail
+} // end cuda
+} // end system
+} // end thrust
+
+#include <thrust/system/cuda/detail/copy_cross_system.inl>
+
diff --git a/compat/thrust/system/cuda/detail/copy_cross_system.inl b/compat/thrust/system/cuda/detail/copy_cross_system.inl
new file mode 100644
index 0000000000..861cb2c2cd
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/copy_cross_system.inl
@@ -0,0 +1,301 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/system/cuda/detail/copy_cross_system.h>
+#include <thrust/detail/copy.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/detail/temporary_array.h>
+#include <thrust/detail/dispatch/is_trivial_copy.h>
+#include <thrust/system/cuda/detail/trivial_copy.h>
+
+namespace thrust
+{
+namespace detail
+{
+
+// XXX WAR circular #inclusion problem
+template<typename,typename> class temporary_array;
+
+} // end detail
+
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+
+
+// general input to random access case
+template<typename System1,
+         typename System2,
+         typename InputIterator,
+         typename RandomAccessIterator>
+  RandomAccessIterator copy_cross_system(cross_system<System1,System2> systems,
+                                         InputIterator begin,
+                                         InputIterator end,
+                                         RandomAccessIterator result,
+                                         thrust::incrementable_traversal_tag, 
+                                         thrust::random_access_traversal_tag)
+{
+  //std::cerr << std::endl;
+  //std::cerr << "general copy_host_to_device(): InputIterator: " << typeid(InputIterator).name() << std::endl;
+  //std::cerr << "general copy_host_to_device(): OutputIterator: " << typeid(OutputIterator).name() << std::endl;
+
+  typedef typename thrust::iterator_value<InputIterator>::type InputType;
+
+  // allocate temporary storage in System1
+  thrust::detail::temporary_array<InputType, System1> temp(systems.system1,begin,end);
+  return thrust::copy(systems, temp.begin(), temp.end(), result);
+}
+
+template<typename System1,
+         typename System2,
+         typename InputIterator,
+         typename Size,
+         typename RandomAccessIterator>
+  RandomAccessIterator copy_cross_system_n(cross_system<System1,System2> systems,
+                                           InputIterator first,
+                                           Size n,
+                                           RandomAccessIterator result,
+                                           thrust::incrementable_traversal_tag, 
+                                           thrust::random_access_traversal_tag)
+{
+  typedef typename thrust::iterator_value<InputIterator>::type InputType;
+
+  // allocate and copy to temporary storage System1
+  thrust::detail::temporary_array<InputType, System1> temp(systems.system1, first, n);
+
+  // recurse
+  return copy_cross_system(systems, temp.begin(), temp.end(), result);
+}
+
+
+// random access to general output case
+template<typename System1,
+         typename System2,
+         typename RandomAccessIterator,
+         typename OutputIterator>
+  OutputIterator copy_cross_system(cross_system<System1,System2> systems,
+                                   RandomAccessIterator begin,
+                                   RandomAccessIterator end,
+                                   OutputIterator result,
+                                   thrust::random_access_traversal_tag, 
+                                   thrust::incrementable_traversal_tag)
+{
+  typedef typename thrust::iterator_value<RandomAccessIterator>::type InputType;
+
+  // copy to temporary storage in System2
+  thrust::detail::temporary_array<InputType,System2> temp(systems.system2, systems.system1, begin, end);
+
+  return thrust::copy(systems.system2, temp.begin(), temp.end(), result);
+}
+
+template<typename System1,
+         typename System2,
+         typename RandomAccessIterator,
+         typename Size,
+         typename OutputIterator>
+  OutputIterator copy_cross_system_n(cross_system<System1,System2> systems,
+                                     RandomAccessIterator first,
+                                     Size n,
+                                     OutputIterator result,
+                                     thrust::random_access_traversal_tag, 
+                                     thrust::incrementable_traversal_tag)
+{
+  typedef typename thrust::iterator_value<RandomAccessIterator>::type InputType;
+
+  // copy to temporary storage in System2
+  thrust::detail::temporary_array<InputType,System2> temp(systems.system2, systems.system1, first, n);
+
+  // copy temp to result
+  return thrust::copy(systems.system2, temp.begin(), temp.end(), result);
+}
+
+
+// trivial copy
+template<typename System1,
+         typename System2,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2>
+  RandomAccessIterator2 copy_cross_system(cross_system<System1,System2> systems,
+                                          RandomAccessIterator1 begin,
+                                          RandomAccessIterator1 end,
+                                          RandomAccessIterator2 result,
+                                          thrust::random_access_traversal_tag,
+                                          thrust::random_access_traversal_tag,
+                                          thrust::detail::true_type) // trivial copy
+{
+//  std::cerr << std::endl;
+//  std::cerr << "random access copy_device_to_host(): trivial" << std::endl;
+//  std::cerr << "general copy_device_to_host(): RandomAccessIterator1: " << typeid(RandomAccessIterator1).name() << std::endl;
+//  std::cerr << "general copy_device_to_host(): RandomAccessIterator2: " << typeid(RandomAccessIterator2).name() << std::endl;
+  
+  // how many elements to copy?
+  typename thrust::iterator_traits<RandomAccessIterator1>::difference_type n = end - begin;
+
+  thrust::system::cuda::detail::trivial_copy_n(systems, begin, n, result);
+
+  return result + n;
+}
+
+
+namespace detail
+{
+
+// random access non-trivial iterator to random access iterator
+template<typename System1,
+         typename System2,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2>
+  RandomAccessIterator2 non_trivial_random_access_copy_cross_system(cross_system<System1,System2> systems,
+                                                                    RandomAccessIterator1 begin,
+                                                                    RandomAccessIterator1 end,
+                                                                    RandomAccessIterator2 result,
+                                                                    thrust::detail::false_type) // InputIterator is non-trivial
+{
+  // copy the input to a temporary input system buffer of OutputType
+  typedef typename thrust::iterator_value<RandomAccessIterator2>::type OutputType;
+
+  // allocate temporary storage in System1
+  thrust::detail::temporary_array<OutputType,System1> temp(systems.system1, begin, end);
+
+  // recurse
+  return copy_cross_system(systems, temp.begin(), temp.end(), result);
+}
+
+template<typename System1,
+         typename System2,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2>
+  RandomAccessIterator2 non_trivial_random_access_copy_cross_system(cross_system<System1,System2> systems,
+                                                                    RandomAccessIterator1 begin,
+                                                                    RandomAccessIterator1 end,
+                                                                    RandomAccessIterator2 result,
+                                                                    thrust::detail::true_type) // InputIterator is trivial
+{
+  typename thrust::iterator_difference<RandomAccessIterator1>::type n = thrust::distance(begin, end);
+
+  // allocate temporary storage in System2
+  // retain the input's type for the intermediate storage
+  // do not initialize the storage (the 0 does this)
+  typedef typename thrust::iterator_value<RandomAccessIterator1>::type InputType;
+  thrust::detail::temporary_array<InputType,System2> temp(0, systems.system2, n);
+
+  // force a trivial (memcpy) copy of the input to the temporary
+  // note that this will not correctly account for copy constructors
+  // but there's nothing we can do about that
+  // XXX one thing we might try is to use pinned memory for the temporary storage
+  //     this might allow us to correctly account for copy constructors
+  thrust::system::cuda::detail::trivial_copy_n(systems, begin, n, temp.begin());
+
+  // finally, copy to the result
+  return thrust::copy(systems.system2, temp.begin(), temp.end(), result);
+}
+
+} // end detail
+
+
+// random access iterator to random access host iterator with non-trivial copy
+template<typename System1,
+         typename System2,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2>
+  RandomAccessIterator2 copy_cross_system(cross_system<System1,System2> systems,
+                                          RandomAccessIterator1 begin,
+                                          RandomAccessIterator1 end,
+                                          RandomAccessIterator2 result,
+                                          thrust::random_access_traversal_tag,
+                                          thrust::random_access_traversal_tag,
+                                          thrust::detail::false_type) // is_trivial_copy
+{
+  // dispatch a non-trivial random access cross system copy based on whether or not the InputIterator is trivial
+  return detail::non_trivial_random_access_copy_cross_system(systems, begin, end, result,
+      typename thrust::detail::is_trivial_iterator<RandomAccessIterator1>::type());
+}
+
+// random access iterator to random access iterator
+template<typename System1,
+         typename System2,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2>
+  RandomAccessIterator2 copy_cross_system(cross_system<System1,System2> systems,
+                                          RandomAccessIterator1 begin,
+                                          RandomAccessIterator1 end,
+                                          RandomAccessIterator2 result,
+                                          thrust::random_access_traversal_tag input_traversal,
+                                          thrust::random_access_traversal_tag output_traversal)
+{
+  // dispatch on whether this is a trivial copy
+  return copy_cross_system(systems, begin, end, result, input_traversal, output_traversal,
+          typename thrust::detail::dispatch::is_trivial_copy<RandomAccessIterator1,RandomAccessIterator2>::type());
+}
+
+template<typename System1,
+         typename System2,
+         typename RandomAccessIterator1,
+         typename Size,
+         typename RandomAccessIterator2>
+  RandomAccessIterator2 copy_cross_system_n(cross_system<System1,System2> systems,
+                                            RandomAccessIterator1 first,
+                                            Size n,
+                                            RandomAccessIterator2 result,
+                                            thrust::random_access_traversal_tag input_traversal,
+                                            thrust::random_access_traversal_tag output_traversal)
+{
+  // implement with copy_cross_system
+  return copy_cross_system(systems, first, first + n, result, input_traversal, output_traversal);
+}
+
+/////////////////
+// Entry Point //
+/////////////////
+
+template<typename System1,
+         typename System2,
+         typename InputIterator,
+         typename OutputIterator>
+  OutputIterator copy_cross_system(cross_system<System1,System2> systems,
+                                   InputIterator begin, 
+                                   InputIterator end, 
+                                   OutputIterator result)
+{
+  return copy_cross_system(systems, begin, end, result, 
+          typename thrust::iterator_traversal<InputIterator>::type(),
+          typename thrust::iterator_traversal<OutputIterator>::type());
+}
+
+template<typename System1,
+         typename System2,
+         typename InputIterator,
+         typename Size,
+         typename OutputIterator>
+  OutputIterator copy_cross_system_n(cross_system<System1,System2> systems,
+                                     InputIterator begin, 
+                                     Size n, 
+                                     OutputIterator result)
+{
+  return copy_cross_system_n(systems, begin, n, result, 
+          typename thrust::iterator_traversal<InputIterator>::type(),
+          typename thrust::iterator_traversal<OutputIterator>::type());
+}
+
+} // end detail
+} // end cuda
+} // end system
+} // end thrust
+
diff --git a/compat/thrust/system/cuda/detail/copy_device_to_device.h b/compat/thrust/system/cuda/detail/copy_device_to_device.h
new file mode 100644
index 0000000000..a7d8df8613
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/copy_device_to_device.h
@@ -0,0 +1,51 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file copy_device_to_device.h
+ *  \brief Device implementations for copying on the device.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/cuda/detail/execution_policy.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator>
+  OutputIterator copy_device_to_device(execution_policy<DerivedPolicy> &exec,
+                                       InputIterator begin, 
+                                       InputIterator end, 
+                                       OutputIterator result);
+
+} // end namespace detail
+} // end namespace cuda
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/cuda/detail/copy_device_to_device.inl>
+
diff --git a/compat/thrust/system/cuda/detail/copy_device_to_device.inl b/compat/thrust/system/cuda/detail/copy_device_to_device.inl
new file mode 100644
index 0000000000..c8263c5d55
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/copy_device_to_device.inl
@@ -0,0 +1,127 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/system/cuda/detail/copy_device_to_device.h>
+#include <thrust/system/cuda/detail/copy_cross_system.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/distance.h>
+#include <thrust/transform.h>
+#include <thrust/functional.h>
+#include <thrust/detail/type_traits.h>
+#include <thrust/detail/temporary_array.h>
+#include <thrust/system/cuda/detail/trivial_copy.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+namespace detail
+{
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator>
+  OutputIterator copy_device_to_device(execution_policy<DerivedPolicy> &exec,
+                                       InputIterator begin, 
+                                       InputIterator end, 
+                                       OutputIterator result,
+                                       thrust::detail::false_type)
+{
+    // general case (mixed types)
+    typedef typename thrust::iterator_traits<InputIterator>::value_type InputType;
+
+#if THRUST_DEVICE_COMPILER == THRUST_DEVICE_COMPILER_NVCC
+    return thrust::transform(exec, begin, end, result, thrust::identity<InputType>());
+#else
+    // we're not compiling with nvcc: copy [begin, end) to temp host memory
+    typename thrust::iterator_traits<InputIterator>::difference_type n = thrust::distance(begin, end);
+
+    thrust::host_system_tag temp_exec;
+    thrust::detail::temporary_array<InputType, thrust::host_system_tag> temp1(temp_exec, begin, end);
+
+    // transform temp1 to OutputType in host memory
+    typedef typename thrust::iterator_traits<OutputIterator>::value_type OutputType;
+
+    thrust::detail::temporary_array<OutputType, thrust::host_system_tag> temp2(temp_exec, temp1.begin(), temp1.end());
+
+    // copy temp2 to device
+    result = thrust::system::cuda::detail::copy_cross_system(temp2.begin(), temp2.end(), result);
+
+    return result;
+#endif // THRUST_DEVICE_COMPILER_NVCC
+}
+
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator>
+  OutputIterator copy_device_to_device(execution_policy<DerivedPolicy> &exec,
+                                       InputIterator begin, 
+                                       InputIterator end, 
+                                       OutputIterator result,
+                                       thrust::detail::true_type)
+{
+    // specialization for device to device when the value_types match, operator= is not overloaded,
+    // and the iterators are pointers
+
+    // how many elements to copy?
+    typename thrust::iterator_traits<OutputIterator>::difference_type n = end - begin;
+
+    thrust::system::cuda::detail::trivial_copy_n(exec, begin, n, result);
+
+    return result + n;
+}
+
+} // end namespace detail
+
+/////////////////
+// Entry Point //
+/////////////////
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator>
+  OutputIterator copy_device_to_device(execution_policy<DerivedPolicy> &exec,
+                                       InputIterator begin, 
+                                       InputIterator end, 
+                                       OutputIterator result)
+{
+    typedef typename thrust::iterator_traits<InputIterator>::value_type InputType;
+    typedef typename thrust::iterator_traits<OutputIterator>::value_type OutputType;
+
+    const bool use_trivial_copy = 
+        thrust::detail::is_same<InputType, OutputType>::value
+        && thrust::detail::is_trivial_iterator<InputIterator>::value 
+        && thrust::detail::is_trivial_iterator<OutputIterator>::value;
+
+    // XXX WAR unused variable warning
+    (void) use_trivial_copy;
+
+    return detail::copy_device_to_device(exec, begin, end, result,
+            thrust::detail::integral_constant<bool, use_trivial_copy>());
+
+}
+
+} // end namespace detail
+} // end namespace cuda
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/cuda/detail/copy_if.h b/compat/thrust/system/cuda/detail/copy_if.h
new file mode 100644
index 0000000000..5ed0f6c9c4
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/copy_if.h
@@ -0,0 +1,49 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/cuda/detail/execution_policy.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename Predicate>
+   OutputIterator copy_if(execution_policy<DerivedPolicy> &exec,
+                          InputIterator1 first,
+                          InputIterator1 last,
+                          InputIterator2 stencil,
+                          OutputIterator result,
+                          Predicate pred);
+
+} // end namespace detail
+} // end namespace cuda
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/cuda/detail/copy_if.inl>
+
diff --git a/compat/thrust/system/cuda/detail/copy_if.inl b/compat/thrust/system/cuda/detail/copy_if.inl
new file mode 100644
index 0000000000..15ea7faa82
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/copy_if.inl
@@ -0,0 +1,212 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/iterator/transform_iterator.h>
+#include <thrust/detail/minmax.h>
+#include <thrust/detail/temporary_array.h>
+#include <thrust/detail/internal_functional.h>
+#include <thrust/system/detail/internal/decompose.h>
+#include <thrust/scan.h>
+#include <thrust/system/cuda/detail/default_decomposition.h>
+#include <thrust/system/cuda/detail/reduce_intervals.h>
+#include <thrust/system/cuda/detail/block/inclusive_scan.h>
+#include <thrust/system/cuda/detail/detail/launch_closure.h>
+#include <thrust/system/cuda/detail/execution_policy.h>
+#include <thrust/functional.h>
+
+
+__THRUST_DISABLE_MSVC_POSSIBLE_LOSS_OF_DATA_WARNING_BEGIN
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+
+template <typename InputIterator1,
+          typename InputIterator2,
+          typename InputIterator3,
+          typename Decomposition,
+          typename OutputIterator,
+          typename Context>
+struct copy_if_intervals_closure
+{
+  InputIterator1 input;
+  InputIterator2 stencil;
+  InputIterator3 offsets;
+  Decomposition decomp;
+  OutputIterator output;
+
+  typedef Context context_type;
+  context_type context;
+  
+  copy_if_intervals_closure(InputIterator1 input,
+                            InputIterator2 stencil,
+                            InputIterator3 offsets,
+                            Decomposition decomp,
+                            OutputIterator output,
+                            Context context = Context())
+    : input(input), stencil(stencil), offsets(offsets), decomp(decomp), output(output), context(context) {}
+
+  __device__ __thrust_forceinline__
+  void operator()(void)
+  {
+    typedef typename thrust::iterator_value<OutputIterator>::type OutputType;
+   
+    typedef unsigned int PredicateType;
+    
+    const unsigned int CTA_SIZE = context_type::ThreadsPerBlock::value;
+
+    thrust::plus<PredicateType> binary_op;
+
+    __shared__ PredicateType sdata[CTA_SIZE];  context.barrier();
+    
+    typedef typename Decomposition::index_type IndexType;
+
+    // this block processes results in [range.begin(), range.end())
+    thrust::system::detail::internal::index_range<IndexType> range = decomp[context.block_index()];
+
+    IndexType base = range.begin();
+
+    PredicateType predicate = 0;
+    
+    // advance input iterators to this thread's starting position
+    input   += base + context.thread_index();
+    stencil += base + context.thread_index();
+
+    // advance output to this interval's starting position
+    if (context.block_index() != 0)
+    {
+        InputIterator3 temp = offsets + (context.block_index() - 1);
+        output += *temp;
+    }
+
+    // process full blocks
+    while(base + CTA_SIZE <= range.end())
+    {
+        // read data
+        sdata[context.thread_index()] = predicate = *stencil;
+      
+        context.barrier();
+
+        // scan block
+        block::inclusive_scan(context, sdata, binary_op);
+       
+        // write data
+        if (predicate)
+        {
+            OutputIterator temp2 = output + (sdata[context.thread_index()] - 1);
+            *temp2 = *input;
+        }
+
+        // advance inputs by CTA_SIZE
+        base    += CTA_SIZE;
+        input   += CTA_SIZE;
+        stencil += CTA_SIZE;
+
+        // advance output by number of true predicates
+        output += sdata[CTA_SIZE - 1];
+
+        context.barrier();
+    }
+
+    // process partially full block at end of input (if necessary)
+    if (base < range.end())
+    {
+        // read data
+        if (base + context.thread_index() < range.end())
+            sdata[context.thread_index()] = predicate = *stencil;
+        else
+            sdata[context.thread_index()] = predicate = 0;
+       
+        context.barrier();
+
+        // scan block
+        block::inclusive_scan(context, sdata, binary_op);
+       
+        // write data
+        if (predicate) // expects predicate=false for >= interval_end
+        {
+            OutputIterator temp2 = output + (sdata[context.thread_index()] - 1);
+            *temp2 = *input;
+        }
+    }
+  }
+}; // copy_if_intervals_closure
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename Predicate>
+   OutputIterator copy_if(execution_policy<DerivedPolicy> &exec,
+                          InputIterator1 first,
+                          InputIterator1 last,
+                          InputIterator2 stencil,
+                          OutputIterator output,
+                          Predicate pred)
+{
+  typedef typename thrust::iterator_difference<InputIterator1>::type IndexType;
+  typedef typename thrust::iterator_value<OutputIterator>::type      OutputType;
+
+  if (first == last)
+      return output;
+
+  typedef thrust::system::detail::internal::uniform_decomposition<IndexType> Decomposition;
+  typedef thrust::detail::temporary_array<IndexType, DerivedPolicy>          IndexArray;
+
+  Decomposition decomp = default_decomposition(last - first);
+
+  // storage for per-block predicate counts
+  IndexArray block_results(exec, decomp.size());
+
+  // convert stencil into an iterator that produces integral values in {0,1}
+  typedef typename thrust::detail::predicate_to_integral<Predicate,IndexType>              PredicateToIndexTransform;
+  typedef thrust::transform_iterator<PredicateToIndexTransform, InputIterator2, IndexType> PredicateToIndexIterator;
+
+  PredicateToIndexIterator predicate_stencil(stencil, PredicateToIndexTransform(pred));
+
+  // compute number of true values in each interval
+  thrust::system::cuda::detail::reduce_intervals(exec, predicate_stencil, block_results.begin(), thrust::plus<IndexType>(), decomp);
+
+  // scan the partial sums
+  thrust::inclusive_scan(exec, block_results.begin(), block_results.end(), block_results.begin(), thrust::plus<IndexType>());
+
+  // copy values to output
+  const unsigned int ThreadsPerBlock = 256;
+  typedef typename IndexArray::iterator InputIterator3;
+  typedef detail::statically_blocked_thread_array<ThreadsPerBlock> Context;
+  typedef copy_if_intervals_closure<InputIterator1,PredicateToIndexIterator,InputIterator3,Decomposition,OutputIterator,Context> Closure;
+  Closure closure(first, predicate_stencil, block_results.begin(), decomp, output);
+  detail::launch_closure(closure, decomp.size(), ThreadsPerBlock);
+
+  return output + block_results[decomp.size() - 1];
+} // end copy_if()
+
+
+} // end namespace detail
+} // end namespace cuda
+} // end namespace system
+} // end namespace thrust
+
+__THRUST_DISABLE_MSVC_POSSIBLE_LOSS_OF_DATA_WARNING_END
+
diff --git a/compat/thrust/system/cuda/detail/count.h b/compat/thrust/system/cuda/detail/count.h
new file mode 100644
index 0000000000..a307fc5fc8
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/count.h
@@ -0,0 +1,22 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system has no special version of this algorithm 
+
diff --git a/compat/thrust/system/cuda/detail/cuda_launch_config.h b/compat/thrust/system/cuda/detail/cuda_launch_config.h
new file mode 100644
index 0000000000..b7f0ca2409
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/cuda_launch_config.h
@@ -0,0 +1,384 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <cstddef>
+#include <thrust/detail/config.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+
+
+// XXX define our own device_properties_t to avoid errors when #including
+//     this file in the absence of a CUDA installation
+struct device_properties_t
+{
+  // mirror the type and spelling of cudaDeviceProp's members
+  // keep these alphabetized
+  int    major;
+  int    maxGridSize[3];
+  int    maxThreadsPerBlock;
+  int    maxThreadsPerMultiProcessor;
+  int    minor;
+  int    multiProcessorCount;
+  int    regsPerBlock;
+  size_t sharedMemPerBlock;
+  int    warpSize;
+};
+
+
+// XXX define our own device_properties_t to avoid errors when #including
+//     this file in the absence of a CUDA installation
+struct function_attributes_t
+{
+  // mirror the type and spelling of cudaFuncAttributes' members
+  // keep these alphabetized
+  size_t constSizeBytes;
+  size_t localSizeBytes;
+  int    maxThreadsPerBlock;
+  int    numRegs;
+  size_t sharedSizeBytes;
+};
+
+
+/*! Computes a block size in number of threads for a CUDA kernel using a occupancy-promoting heuristic.
+ *  \param attributes The cudaFuncAttributes corresponding to a __global__ function of interest on a GPU of interest.
+ *  \param properties The cudaDeviceProp corresponding to a GPU on which to launch the __global__ function of interest.
+ *  \return A CUDA block size, in number of threads, which the resources of the GPU's streaming multiprocessor can
+ *          accomodate and which is intended to promote occupancy. The result is equivalent to the one performed by
+ *          the "CUDA Occupancy Calculator". 
+ *  \note The __global__ function of interest is presumed to use 0 bytes of dynamically-allocated __shared__ memory.
+ */
+inline __host__ __device__
+std::size_t block_size_with_maximum_potential_occupancy(const function_attributes_t &attributes,
+                                                        const device_properties_t   &properties);
+
+/*! Computes a block size in number of threads for a CUDA kernel using a occupancy-promoting heuristic.
+ *  Use this version of the function when a CUDA block's dynamically-allocated __shared__ memory requirements
+ *  vary with the size of the block.
+ *  \param attributes The cudaFuncAttributes corresponding to a __global__ function of interest on a GPU of interest.
+ *  \param properties The cudaDeviceProp corresponding to a GPU on which to launch the __global__ function of interest.
+ *  \param block_size_to_dynamic_smem_bytes A unary function which maps an integer CUDA block size to the number of bytes
+ *         of dynamically-allocated __shared__ memory required by a CUDA block of that size.
+ *  \return A CUDA block size, in number of threads, which the resources of the GPU's streaming multiprocessor can
+ *          accomodate and which is intended to promote occupancy. The result is equivalent to the one performed by
+ *          the "CUDA Occupancy Calculator". 
+ */
+template<typename UnaryFunction>
+inline __host__ __device__
+std::size_t block_size_with_maximum_potential_occupancy(const function_attributes_t &attributes,
+                                                        const device_properties_t   &properties,
+                                                        UnaryFunction block_size_to_dynamic_smem_size);
+
+
+/*! Returns the maximum amount of dynamic shared memory each block
+ *  can utilize without reducing thread occupancy.
+ *
+ *  \param properties CUDA device properties
+ *  \param attributes CUDA function attributes
+ *  \param blocks_per_processor Number of blocks per streaming multiprocessor
+ */
+inline __host__ __device__
+size_t proportional_smem_allocation(const device_properties_t   &properties,
+                                    const function_attributes_t &attributes,
+                                    size_t blocks_per_processor);
+
+
+template<typename UnaryFunction>
+inline __host__ __device__
+size_t max_blocksize_subject_to_smem_usage(const device_properties_t   &properties,
+                                           const function_attributes_t &attributes,
+                                           UnaryFunction blocksize_to_dynamic_smem_usage);
+
+
+
+namespace cuda_launch_config_detail
+{
+
+using std::size_t;
+
+namespace util
+{
+
+
+template<typename T>
+inline __host__ __device__
+T min_(const T &lhs, const T &rhs)
+{
+  return rhs < lhs ? rhs : lhs;
+}
+
+
+template <typename T>
+struct zero_function
+{
+  inline __host__ __device__
+  T operator()(T)
+  {
+    return 0;
+  }
+};
+
+
+// x/y rounding towards +infinity for integers, used to determine # of blocks/warps etc.
+template<typename L, typename R>
+  inline __host__ __device__ L divide_ri(const L x, const R y)
+{
+    return (x + (y - 1)) / y;
+}
+
+// x/y rounding towards zero for integers, used to determine # of blocks/warps etc.
+template<typename L, typename R>
+  inline __host__ __device__ L divide_rz(const L x, const R y)
+{
+    return x / y;
+}
+
+// round x towards infinity to the next multiple of y
+template<typename L, typename R>
+  inline __host__ __device__ L round_i(const L x, const R y){ return y * divide_ri(x, y); }
+
+// round x towards zero to the next multiple of y
+template<typename L, typename R>
+  inline __host__ __device__ L round_z(const L x, const R y){ return y * divide_rz(x, y); }
+
+} // end namespace util
+
+
+
+// granularity of shared memory allocation
+inline __host__ __device__
+size_t smem_allocation_unit(const device_properties_t &properties)
+{
+  switch(properties.major)
+  {
+    case 1:  return 512;
+    case 2:  return 128;
+    case 3:  return 256;
+    default: return 256; // unknown GPU; have to guess
+  }
+}
+
+
+// granularity of register allocation
+inline __host__ __device__
+size_t reg_allocation_unit(const device_properties_t &properties, const size_t regsPerThread)
+{
+  switch(properties.major)
+  {
+    case 1:  return (properties.minor <= 1) ? 256 : 512;
+    case 2:  switch(regsPerThread)
+             {
+               case 21:
+               case 22:
+               case 29:
+               case 30:
+               case 37:
+               case 38:
+               case 45:
+               case 46:
+                 return 128;
+               default:
+                 return 64;
+             }
+    case 3:  return 256;
+    default: return 256; // unknown GPU; have to guess
+  }
+}
+
+
+// granularity of warp allocation
+inline __host__ __device__
+size_t warp_allocation_multiple(const device_properties_t &properties)
+{
+  return (properties.major <= 1) ? 2 : 1;
+}
+
+// number of "sides" into which the multiprocessor is partitioned
+inline __host__ __device__
+size_t num_sides_per_multiprocessor(const device_properties_t &properties)
+{
+  switch(properties.major)
+  {
+    case 1:  return 1;
+    case 2:  return 2;
+    case 3:  return 4;
+    default: return 4; // unknown GPU; have to guess
+  }
+}
+
+
+inline __host__ __device__
+size_t max_blocks_per_multiprocessor(const device_properties_t &properties)
+{
+  return (properties.major <= 2) ? 8 : 16;
+}
+
+
+inline __host__ __device__
+size_t max_active_blocks_per_multiprocessor(const device_properties_t    &properties,
+                                            const function_attributes_t  &attributes,
+                                            int CTA_SIZE,
+                                            size_t dynamic_smem_bytes)
+{
+  // Determine the maximum number of CTAs that can be run simultaneously per SM
+  // This is equivalent to the calculation done in the CUDA Occupancy Calculator spreadsheet
+
+  //////////////////////////////////////////
+  // Limits due to threads/SM or blocks/SM
+  //////////////////////////////////////////
+  const size_t maxThreadsPerSM = properties.maxThreadsPerMultiProcessor;  // 768, 1024, 1536, etc.
+  const size_t maxBlocksPerSM  = max_blocks_per_multiprocessor(properties);
+
+  // Calc limits
+  const size_t ctaLimitThreads = (CTA_SIZE <= properties.maxThreadsPerBlock) ? maxThreadsPerSM / CTA_SIZE : 0;
+  const size_t ctaLimitBlocks  = maxBlocksPerSM;
+
+  //////////////////////////////////////////
+  // Limits due to shared memory/SM
+  //////////////////////////////////////////
+  const size_t smemAllocationUnit     = smem_allocation_unit(properties);
+  const size_t smemBytes  = attributes.sharedSizeBytes + dynamic_smem_bytes;
+  const size_t smemPerCTA = util::round_i(smemBytes, smemAllocationUnit);
+
+  // Calc limit
+  const size_t ctaLimitSMem = smemPerCTA > 0 ? properties.sharedMemPerBlock / smemPerCTA : maxBlocksPerSM;
+
+  //////////////////////////////////////////
+  // Limits due to registers/SM
+  //////////////////////////////////////////
+  const size_t regAllocationUnit      = reg_allocation_unit(properties, attributes.numRegs);
+  const size_t warpAllocationMultiple = warp_allocation_multiple(properties);
+  const size_t numWarps = util::round_i(util::divide_ri(CTA_SIZE, properties.warpSize), warpAllocationMultiple);
+
+  // Calc limit
+  size_t ctaLimitRegs;
+  if(properties.major <= 1)
+  {
+    // GPUs of compute capability 1.x allocate registers to CTAs
+    // Number of regs per block is regs per thread times number of warps times warp size, rounded up to allocation unit
+    const size_t regsPerCTA = util::round_i(attributes.numRegs * properties.warpSize * numWarps, regAllocationUnit);
+    ctaLimitRegs = regsPerCTA > 0 ? properties.regsPerBlock / regsPerCTA : maxBlocksPerSM;
+  }
+  else
+  {
+    // GPUs of compute capability 2.x and higher allocate registers to warps
+    // Number of regs per warp is regs per thread times times warp size, rounded up to allocation unit
+    const size_t regsPerWarp = util::round_i(attributes.numRegs * properties.warpSize, regAllocationUnit);
+    const size_t numSides = num_sides_per_multiprocessor(properties);
+    const size_t numRegsPerSide = properties.regsPerBlock / numSides;
+    ctaLimitRegs = regsPerWarp > 0 ? ((numRegsPerSide / regsPerWarp) * numSides) / numWarps : maxBlocksPerSM;
+  }
+
+  //////////////////////////////////////////
+  // Overall limit is min() of limits due to above reasons
+  //////////////////////////////////////////
+  return util::min_(ctaLimitRegs, util::min_(ctaLimitSMem, util::min_(ctaLimitThreads, ctaLimitBlocks)));
+}
+
+
+} // end namespace cuda_launch_config_detail
+
+
+template<typename UnaryFunction>
+inline __host__ __device__
+std::size_t block_size_with_maximum_potential_occupancy(const function_attributes_t &attributes,
+                                                        const device_properties_t   &properties,
+                                                        UnaryFunction block_size_to_dynamic_smem_size)
+{
+  size_t max_occupancy      = properties.maxThreadsPerMultiProcessor;
+  size_t largest_blocksize  = cuda_launch_config_detail::util::min_(properties.maxThreadsPerBlock, attributes.maxThreadsPerBlock);
+  size_t granularity        = properties.warpSize;
+  size_t max_blocksize      = 0;
+  size_t highest_occupancy  = 0;
+
+  for(size_t blocksize = largest_blocksize; blocksize != 0; blocksize -= granularity)
+  {
+    size_t occupancy = blocksize * cuda_launch_config_detail::max_active_blocks_per_multiprocessor(properties, attributes, blocksize, block_size_to_dynamic_smem_size(blocksize));
+
+    if(occupancy > highest_occupancy)
+    {
+      max_blocksize = blocksize;
+      highest_occupancy = occupancy;
+    }
+
+    // early out, can't do better
+    if(highest_occupancy == max_occupancy)
+      break;
+  }
+
+  return max_blocksize;
+}
+
+
+inline __host__ __device__
+std::size_t block_size_with_maximum_potential_occupancy(const function_attributes_t &attributes,
+                                                        const device_properties_t   &properties)
+{
+  return block_size_with_maximum_potential_occupancy(attributes, properties, cuda_launch_config_detail::util::zero_function<std::size_t>());
+}
+
+
+inline __host__ __device__
+size_t proportional_smem_allocation(const device_properties_t   &properties,
+                                    const function_attributes_t &attributes,
+                                    size_t blocks_per_processor)
+{
+  size_t smem_per_processor    = properties.sharedMemPerBlock;
+  size_t smem_allocation_unit  = cuda_launch_config_detail::smem_allocation_unit(properties);
+
+  size_t total_smem_per_block  = cuda_launch_config_detail::util::round_z(smem_per_processor / blocks_per_processor, smem_allocation_unit);
+  size_t static_smem_per_block = attributes.sharedSizeBytes;
+  
+  return total_smem_per_block - static_smem_per_block;
+}
+
+
+template<typename UnaryFunction>
+inline __host__ __device__
+size_t max_blocksize_subject_to_smem_usage(const device_properties_t   &properties,
+                                           const function_attributes_t &attributes,
+                                           UnaryFunction blocksize_to_dynamic_smem_usage)
+{
+  size_t largest_blocksize = (thrust::min)(properties.maxThreadsPerBlock, attributes.maxThreadsPerBlock);
+  size_t granularity = properties.warpSize;
+  
+  for(int blocksize = largest_blocksize; blocksize > 0; blocksize -= granularity)
+  {
+    size_t total_smem_usage = blocksize_to_dynamic_smem_usage(blocksize) + attributes.sharedSizeBytes;
+
+    if(total_smem_usage <= properties.sharedMemPerBlock)
+    {
+      return blocksize;
+    }
+  }
+
+  return 0;
+}
+
+
+} // end detail
+} // end cuda
+} // end system
+} // end thrust
+
diff --git a/compat/thrust/system/cuda/detail/default_decomposition.h b/compat/thrust/system/cuda/detail/default_decomposition.h
new file mode 100644
index 0000000000..1ed6bcfe20
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/default_decomposition.h
@@ -0,0 +1,45 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file default_decomposition.h
+ *  \brief Return a decomposition that is appropriate for the CUDA backend.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/internal/decompose.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+
+template <typename IndexType>
+thrust::system::detail::internal::uniform_decomposition<IndexType> default_decomposition(IndexType n);
+
+} // end namespace detail
+} // end namespace cuda
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/cuda/detail/default_decomposition.inl>
+
diff --git a/compat/thrust/system/cuda/detail/default_decomposition.inl b/compat/thrust/system/cuda/detail/default_decomposition.inl
new file mode 100644
index 0000000000..3f0879ac93
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/default_decomposition.inl
@@ -0,0 +1,41 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/system/cuda/detail/runtime_introspection.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+
+template <typename IndexType>
+thrust::system::detail::internal::uniform_decomposition<IndexType> default_decomposition(IndexType n)
+{
+  // TODO eliminate magical constant
+  device_properties_t properties = device_properties();
+  return thrust::system::detail::internal::uniform_decomposition<IndexType>(n, properties.maxThreadsPerBlock, 10 * properties.multiProcessorCount);
+}
+
+} // end namespace detail
+} // end namespace cuda
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/cuda/detail/detail/alignment.h b/compat/thrust/system/cuda/detail/detail/alignment.h
new file mode 100644
index 0000000000..31fdaaf422
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/detail/alignment.h
@@ -0,0 +1,223 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+namespace detail
+{
+namespace alignment_of_detail
+{
+
+
+template<typename T> class alignment_of_impl;
+
+template<typename T, std::size_t size_diff>
+  struct helper
+{
+  static const std::size_t value = size_diff;
+};
+
+template<typename T>
+  class helper<T,0>
+{
+  public:
+    static const std::size_t value = alignment_of_impl<T>::value;
+};
+
+template<typename T>
+  class alignment_of_impl
+{
+  private:
+    struct big { T x; char c; };
+
+  public:
+    static const std::size_t value = helper<big, sizeof(big) - sizeof(T)>::value;
+};
+
+
+} // end alignment_of_detail
+
+
+template<typename T>
+  struct alignment_of
+    : alignment_of_detail::alignment_of_impl<T>
+{};
+
+
+template<std::size_t Align> struct aligned_type;
+
+// __align__ is CUDA-specific, so guard it
+#if THRUST_DEVICE_COMPILER == THRUST_DEVICE_COMPILER_NVCC
+
+// implementing aligned_type portably is tricky:
+
+#  if THRUST_HOST_COMPILER == THRUST_HOST_COMPILER_MSVC
+// implement aligned_type with specialization because MSVC
+// requires literals as arguments to declspec(align(n))
+template<> struct aligned_type<1>
+{
+  struct __align__(1) type { };
+};
+
+template<> struct aligned_type<2>
+{
+  struct __align__(2) type { };
+};
+
+template<> struct aligned_type<4>
+{
+  struct __align__(4) type { };
+};
+
+template<> struct aligned_type<8>
+{
+  struct __align__(8) type { };
+};
+
+template<> struct aligned_type<16>
+{
+  struct __align__(16) type { };
+};
+
+template<> struct aligned_type<32>
+{
+  struct __align__(32) type { };
+};
+
+template<> struct aligned_type<64>
+{
+  struct __align__(64) type { };
+};
+
+template<> struct aligned_type<128>
+{
+  struct __align__(128) type { };
+};
+
+template<> struct aligned_type<256>
+{
+  struct __align__(256) type { };
+};
+
+template<> struct aligned_type<512>
+{
+  struct __align__(512) type { };
+};
+
+template<> struct aligned_type<1024>
+{
+  struct __align__(1024) type { };
+};
+
+template<> struct aligned_type<2048>
+{
+  struct __align__(2048) type { };
+};
+
+template<> struct aligned_type<4096>
+{
+  struct __align__(4096) type { };
+};
+
+template<> struct aligned_type<8192>
+{
+  struct __align__(8192) type { };
+};
+#  elif (THRUST_HOST_COMPILER == THRUST_HOST_COMPILER_GCC) && (THRUST_GCC_VERSION < 40300)
+// implement aligned_type with specialization because gcc 4.2
+// requires literals as arguments to __attribute__(aligned(n))
+template<> struct aligned_type<1>
+{
+  struct __align__(1) type { };
+};
+
+template<> struct aligned_type<2>
+{
+  struct __align__(2) type { };
+};
+
+template<> struct aligned_type<4>
+{
+  struct __align__(4) type { };
+};
+
+template<> struct aligned_type<8>
+{
+  struct __align__(8) type { };
+};
+
+template<> struct aligned_type<16>
+{
+  struct __align__(16) type { };
+};
+
+template<> struct aligned_type<32>
+{
+  struct __align__(32) type { };
+};
+
+template<> struct aligned_type<64>
+{
+  struct __align__(64) type { };
+};
+
+template<> struct aligned_type<128>
+{
+  struct __align__(128) type { };
+};
+
+#  else
+// assume the compiler allows template parameters as
+// arguments to __align__ 
+template<std::size_t Align> struct aligned_type
+{
+  struct __align__(Align) type { };
+};
+#  endif // THRUST_HOST_COMPILER
+#else
+template<std::size_t Align> struct aligned_type
+{
+  struct type { };
+};
+#endif // THRUST_DEVICE_COMPILER
+
+
+template<std::size_t Len, std::size_t Align>
+  struct aligned_storage
+{
+  union type
+  {
+    unsigned char data[Len];
+
+    typename aligned_type<Align>::type align;
+  };
+};
+
+
+} // end detail
+} // end detail
+} // end cuda
+} // end system
+} // end thrust
+
diff --git a/compat/thrust/system/cuda/detail/detail/b40c/kernel_utils.h b/compat/thrust/system/cuda/detail/detail/b40c/kernel_utils.h
new file mode 100644
index 0000000000..e2c5a44941
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/detail/b40c/kernel_utils.h
@@ -0,0 +1,284 @@
+/**
+ * Copyright 2010 Duane Merrill
+ * 
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ * 
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License. 
+ * 
+ * For more information, see our Google Code project site: 
+ * http://code.google.com/p/back40computing/
+ * 
+ * Thanks!
+ */
+
+
+//------------------------------------------------------------------------------
+// Common B40C Defines, Properties, and Routines 
+//------------------------------------------------------------------------------
+
+
+#pragma once
+
+#include <cuda.h>
+#include <thrust/system/cuda/detail/synchronize.h>
+
+namespace thrust  {
+namespace system  {
+namespace cuda    {
+namespace detail  {
+namespace detail  {
+namespace b40c_thrust   {
+
+//------------------------------------------------------------------------------
+// Device properties 
+//------------------------------------------------------------------------------
+
+
+#ifndef __CUDA_ARCH__
+	#define __CUDA_ARCH__ 0
+#endif
+
+#define B40C_FERMI(version)								(version >= 200)
+#define B40C_LOG_WARP_THREADS							5									// 32 threads in a warp
+#define B40C_WARP_THREADS								(1 << B40C_LOG_WARP_THREADS)
+#define B40C_LOG_MEM_BANKS(version) 					((version >= 200) ? 5 : 4)			// 32 banks on fermi, 16 on tesla
+#define B40C_MEM_BANKS(version)							(1 << B40C_LOG_MEM_BANKS(version))
+
+// TODO refactor these
+#if __CUDA_ARCH__ >= 200
+	#define FastMul(a, b) (a * b)
+#else
+	#define FastMul(a, b) (__umul24(a, b))
+#endif	
+
+#if __CUDA_ARCH__ >= 120
+	#define WarpVoteAll(active_threads, predicate) (__all(predicate))
+#else 
+	#define WarpVoteAll(active_threads, predicate) (EmulatedWarpVoteAll<active_threads>(predicate))
+#endif
+
+#if __CUDA_ARCH__ >= 200
+	#define TallyWarpVote(active_threads, predicate, storage) (__popc(__ballot(predicate)))
+#else 
+	#define TallyWarpVote(active_threads, predicate, storage) (TallyWarpVoteSm10<active_threads>(predicate, storage))
+#endif
+
+#ifdef __LP64__
+	#define _B40C_LP64_ true
+#else
+	#define _B40C_LP64_ false
+#endif
+
+#define _B40C_REG_MISER_QUALIFIER_ __shared__
+
+
+//------------------------------------------------------------------------------
+// Handy routines 
+//------------------------------------------------------------------------------
+
+
+/**
+ * Select maximum
+ */
+#define B40C_MAX(a, b) ((a > b) ? a : b)
+
+
+/**
+ * MagnitudeShift().  Allows you to shift left for positive magnitude values, 
+ * right for negative.   
+ * 
+ * N.B. This code is a little strange; we are using this meta-programming 
+ * pattern of partial template specialization for structures in order to 
+ * decide whether to shift left or right.  Normally we would just use a 
+ * conditional to decide if something was negative or not and then shift 
+ * accordingly, knowing that the compiler will elide the untaken branch, 
+ * i.e., the out-of-bounds shift during dead code elimination. However, 
+ * the pass for bounds-checking shifts seems to happen before the DCE 
+ * phase, which results in a an unsightly number of compiler warnings, so 
+ * we force the issue earlier using structural template specialization.
+ */
+
+template <typename K, int magnitude, bool shift_left> struct MagnitudeShiftOp;
+
+template <typename K, int magnitude> 
+struct MagnitudeShiftOp<K, magnitude, true> {
+	__device__ __forceinline__ static K Shift(K key) {
+		return key << magnitude;
+	}
+};
+
+template <typename K, int magnitude> 
+struct MagnitudeShiftOp<K, magnitude, false> {
+	__device__ __forceinline__ static K Shift(K key) {
+		return key >> magnitude;
+	}
+};
+
+template <typename K, int magnitude> 
+__device__ __forceinline__ K MagnitudeShift(K key) {
+	return MagnitudeShiftOp<K, (magnitude > 0) ? magnitude : magnitude * -1, (magnitude > 0)>::Shift(key);
+}
+
+
+/**
+ * Supress warnings for unused constants
+ */
+template <typename T>
+__device__ __forceinline__ void SuppressUnusedConstantWarning(const T) {}
+
+
+
+
+//------------------------------------------------------------------------------
+// Common device routines
+//------------------------------------------------------------------------------
+
+
+/**
+ * Perform a warp-synchrounous prefix scan.  Allows for diverting a warp's
+ * threads into separate scan problems (multi-scan). 
+ */
+template <int NUM_ELEMENTS, bool MULTI_SCAN>
+__device__ __forceinline__ int WarpScan(
+	volatile int warpscan[][NUM_ELEMENTS],
+	int partial_reduction,
+	int copy_section) {
+	
+	int warpscan_idx;
+	if (MULTI_SCAN) {
+		warpscan_idx = threadIdx.x & (NUM_ELEMENTS - 1);
+	} else {
+		warpscan_idx = threadIdx.x;
+	}
+
+	warpscan[1][warpscan_idx] = partial_reduction;
+
+	if (NUM_ELEMENTS > 1) warpscan[1][warpscan_idx] = partial_reduction = 
+			partial_reduction + warpscan[1][warpscan_idx - 1];
+	if (NUM_ELEMENTS > 2) warpscan[1][warpscan_idx] = partial_reduction = 
+			partial_reduction + warpscan[1][warpscan_idx - 2];
+	if (NUM_ELEMENTS > 4) warpscan[1][warpscan_idx] = partial_reduction = 
+			partial_reduction + warpscan[1][warpscan_idx - 4];
+	if (NUM_ELEMENTS > 8) warpscan[1][warpscan_idx] = partial_reduction = 
+			partial_reduction + warpscan[1][warpscan_idx - 8];
+	if (NUM_ELEMENTS > 16) warpscan[1][warpscan_idx] = partial_reduction = 
+			partial_reduction + warpscan[1][warpscan_idx - 16];
+	
+	if (copy_section > 0) {
+		warpscan[1 + copy_section][warpscan_idx] = partial_reduction;
+	}
+	
+	return warpscan[1][warpscan_idx - 1];
+}
+
+/**
+ * Perform a warp-synchronous reduction
+ */
+template <int NUM_ELEMENTS>
+__device__ __forceinline__ void WarpReduce(
+	int idx,
+	volatile int *storage,
+	int partial_reduction)
+{
+	storage[idx] = partial_reduction;
+
+	if (NUM_ELEMENTS > 16) storage[idx] = partial_reduction = partial_reduction + storage[idx + 16];
+	if (NUM_ELEMENTS > 8) storage[idx] = partial_reduction = partial_reduction + storage[idx + 8];
+	if (NUM_ELEMENTS > 4) storage[idx] = partial_reduction = partial_reduction + storage[idx + 4];
+	if (NUM_ELEMENTS > 2) storage[idx] = partial_reduction = partial_reduction + storage[idx + 2];
+	if (NUM_ELEMENTS > 1) storage[idx] = partial_reduction = partial_reduction + storage[idx + 1];
+}
+
+
+/**
+ * Tally a warp-vote regarding the given predicate using the supplied storage
+ */
+template <int ACTIVE_THREADS>
+__device__ __forceinline__ int TallyWarpVoteSm10(int predicate, int storage[]) {
+	WarpReduce<ACTIVE_THREADS>(threadIdx.x, storage, predicate);
+	return storage[0];
+}
+
+
+/**
+ * Tally a warp-vote regarding the given predicate
+ */
+template <int ACTIVE_THREADS>
+__device__ __forceinline__ int TallyWarpVoteSm10(int predicate) {
+  __shared__ int vote_reduction[B40C_WARP_THREADS];
+  return TallyWarpVoteSm10<ACTIVE_THREADS>(predicate, vote_reduction);
+}
+
+/**
+ * Emulate the __all() warp vote instruction
+ */
+template <int ACTIVE_THREADS>
+__device__ __forceinline__ int EmulatedWarpVoteAll(int predicate) {
+	return (TallyWarpVoteSm10<ACTIVE_THREADS>(predicate) == ACTIVE_THREADS);
+}
+
+
+/**
+ * Have each thread concurrently perform a serial reduction over its specified segment 
+ */
+template <int LENGTH>
+__device__ __forceinline__ int
+SerialReduce(int segment[]) {
+	
+	int reduce = segment[0];
+
+	#pragma unroll
+	for (int i = 1; i < (int) LENGTH; i++) {
+		reduce += segment[i];
+	}
+	
+	return reduce;
+}
+
+
+/**
+ * Have each thread concurrently perform a serial scan over its specified segment
+ */
+template <int LENGTH>
+__device__ __forceinline__
+void SerialScan(int segment[], int seed0) {
+	
+	int seed1;
+
+	#pragma unroll	
+	for (int i = 0; i < (int) LENGTH; i += 2) {
+		seed1 = segment[i] + seed0;
+		segment[i] = seed0;
+		seed0 = seed1 + segment[i + 1];
+		segment[i + 1] = seed1;
+	}
+}
+
+
+
+
+//------------------------------------------------------------------------------
+// Empty Kernels
+//------------------------------------------------------------------------------
+
+template <typename T>
+__global__ void FlushKernel(void)
+{
+}
+
+
+} // end namespace b40c_thrust
+} // end namespace detail
+} // end namespace detail
+} // end namespace cuda
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/cuda/detail/detail/b40c/radixsort_api.h b/compat/thrust/system/cuda/detail/detail/b40c/radixsort_api.h
new file mode 100644
index 0000000000..2b199bb08a
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/detail/b40c/radixsort_api.h
@@ -0,0 +1,807 @@
+/******************************************************************************
+ * Copyright 2010 Duane Merrill
+ * 
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ * 
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License. 
+ * 
+ * 
+ * 
+ * 
+ * AUTHORS' REQUEST: 
+ * 
+ * 		If you use|reference|benchmark this code, please cite our Technical 
+ * 		Report (http://www.cs.virginia.edu/~dgm4d/papers/RadixSortTR.pdf):
+ * 
+ *		@TechReport{ Merrill:Sorting:2010,
+ *        	author = "Duane Merrill and Andrew Grimshaw",
+ *        	title = "Revisiting Sorting for GPGPU Stream Architectures",
+ *        	year = "2010",
+ *        	institution = "University of Virginia, Department of Computer Science",
+ *        	address = "Charlottesville, VA, USA",
+ *        	number = "CS2010-03"
+ *		}
+ * 
+ * For more information, see our Google Code project site: 
+ * http://code.google.com/p/back40computing/
+ * 
+ * Thanks!
+ ******************************************************************************/
+
+
+
+/******************************************************************************
+ * Radix Sorting API
+ *
+ * USAGE:
+ * 
+ * Using the B40C radix sorting implementation is easy.  Just #include this API 
+ * file and its kernel include dependencies within your source.  Below are two
+ * examples for using: 
+ *
+ * (1) A keys-only example for sorting floats:
+ * 
+ *		// Create storage-management structure
+ * 		RadixSortStorage<float> device_storage(d_float_keys);			
+ *
+ *		// Create and enact sorter
+ * 		RadixSortingEnactor sorter<float>(d_float_keys_len);
+ *		sorter.EnactSort(device_storage);
+ *
+ *		// Re-acquire pointer to sorted keys, free unused/temp storage 
+ *		d_float_keys = device_storage.d_keys;
+ *		device_storage.CleanupTempStorage();
+ *
+ * (2) And a key-value example for sorting ints paired with doubles:
+ *
+ *		// Create storage-management structure
+ * 		RadixSortStorage<int, double> device_storage(d_int_keys, d_double_values);			
+ *
+ *		// Create and enact sorter
+ * 		RadixSortingEnactor sorter<int, double>(d_int_keys_len);
+ *		sorter.EnactSort(device_storage);
+ *
+ *		// Re-acquire pointer to sorted keys and values, free unused/temp storage 
+ *		d_int_keys = device_storage.d_keys;
+ *		d_double_values = device_storage.d_values;
+ *		device_storage.CleanupTempStorage();
+ *
+ *
+ ******************************************************************************/
+
+#pragma once
+
+#include <stdlib.h> 
+#include <stdio.h> 
+#include <string.h> 
+#include <math.h> 
+#include <float.h>
+
+#include "radixsort_reduction_kernel.h"
+#include "radixsort_spine_kernel.h"
+#include "radixsort_scanscatter_kernel.h"
+
+#include <thrust/swap.h>
+
+namespace thrust  {
+namespace system  {
+namespace cuda    {
+namespace detail  {
+namespace detail  {
+namespace b40c_thrust   {
+
+
+/******************************************************************************
+ * Debugging options
+ ******************************************************************************/
+
+static bool RADIXSORT_DEBUG = false;
+
+
+
+/******************************************************************************
+ * Structures for mananging device-side sorting state
+ ******************************************************************************/
+
+/**
+ * Sorting storage-management structure for device vectors
+ */
+template <typename K, typename V = KeysOnlyType>
+struct RadixSortStorage {
+
+	// Device vector of keys to sort
+	K* d_keys;
+	
+	// Device vector of values to sort
+	V* d_values;
+
+	// Ancillary device vector for key storage 
+	K* d_alt_keys;
+
+	// Ancillary device vector for value storage
+	V* d_alt_values;
+
+	// Temporary device storage needed for radix sorting histograms
+	int *d_spine;
+	
+	// Flip-flopping temporary device storage denoting which digit place 
+	// pass should read from which input source (i.e., false if reading from 
+	// keys, true if reading from alternate_keys
+	bool *d_from_alt_storage;
+
+	// Host-side boolean whether or not an odd number of sorting passes left the 
+	// results in alternate storage.  If so, the d_keys (and d_values) pointers 
+	// will have been swapped with the d_alt_keys (and d_alt_values) pointers in order to 
+	// point to the final results.
+	bool using_alternate_storage;
+	
+	// Constructor
+	RadixSortStorage(K* keys = NULL, V* values = NULL) 
+	{ 
+		d_keys = keys; 
+		d_values = values; 
+		d_alt_keys = NULL; 
+		d_alt_values = NULL; 
+		d_spine = NULL;
+		d_from_alt_storage = NULL;
+		
+		using_alternate_storage = false;
+	}
+
+	// Clean up non-results storage (may include freeing original storage if 
+	// primary pointers were swizzled as per using_alternate_storage) 
+	cudaError_t CleanupTempStorage() 
+	{
+		if (d_alt_keys) cudaFree(d_alt_keys);
+		if (d_alt_values) cudaFree(d_alt_values);
+		if (d_spine) cudaFree(d_spine);
+		if (d_from_alt_storage) cudaFree(d_from_alt_storage);
+		
+		return cudaSuccess;
+	}
+};
+
+
+
+/******************************************************************************
+ * Base class for sorting enactors
+ ******************************************************************************/
+
+
+/**
+ * Base class for SRTS radix sorting enactors.
+ */
+template <typename K, typename V>
+class BaseRadixSortingEnactor 
+{
+public:
+	
+	// Unsigned integer type suitable for radix sorting of keys
+	typedef typename KeyConversion<K>::UnsignedBits ConvertedKeyType;
+
+protected:
+
+	//
+	// Information about our problem configuration
+	//
+	
+	bool				_keys_only;
+	unsigned int 		_num_elements;
+	int 				_cycle_elements;
+	int 				_spine_elements;
+	int 				_grid_size;
+	CtaDecomposition 	_work_decomposition;
+	int 				_passes;
+	bool 				_swizzle_pointers_for_odd_passes;
+
+	// Information about our target device
+	cudaDeviceProp 		_device_props;
+	int 				_device_sm_version;
+	
+	// Information about our kernel assembly
+	int 				_kernel_ptx_version;
+	cudaFuncAttributes 	_spine_scan_kernel_attrs;
+	
+protected:
+	
+	/**
+	 * Constructor.
+	 */
+	BaseRadixSortingEnactor(int passes, int radix_bits, unsigned int num_elements, int max_grid_size, bool swizzle_pointers_for_odd_passes = true); 
+	
+	/**
+	 * Heuristic for determining the number of CTAs to launch.
+	 *   
+	 * @param[in] 		max_grid_size  
+	 * 		Maximum allowable number of CTAs to launch.  A value of 0 indicates 
+	 * 		that the default value should be used.
+	 * 
+	 * @return The actual number of CTAs that should be launched
+	 */
+	int GridSize(int max_grid_size);
+
+	/**
+	 * Performs a distribution sorting pass over a single digit place
+	 */
+	template <int PASS, int RADIX_BITS, int BIT, typename PreprocessFunctor, typename PostprocessFunctor>
+	cudaError_t DigitPlacePass(const RadixSortStorage<ConvertedKeyType, V> &converted_storage); 
+	
+	/**
+	 * Enacts a sorting operation by performing the the appropriate 
+	 * digit-place passes.  To be overloaded by specialized subclasses.
+	 */
+	virtual cudaError_t EnactDigitPlacePasses(const RadixSortStorage<ConvertedKeyType, V> &converted_storage) = 0;
+	
+public:
+	
+	/**
+	 * Returns the length (in unsigned ints) of the device vector needed for  
+	 * temporary storage of the reduction spine.  Useful if pre-allocating 
+	 * your own device storage (as opposed to letting EnactSort() allocate it
+	 * for you).
+	 */
+	int SpineElements() { return _spine_elements; }
+
+	/**
+	 * Returns whether or not the problem will fit on the device.
+	 */
+	bool CanFit();
+
+	/**
+	 * Enacts a radix sorting operation on the specified device data.
+	 * 
+	 * IMPORTANT NOTES: The device storage backing the specified input vectors of 
+	 * keys (and data) will be modified.  (I.e., treat this as an in-place sort.)  
+	 * 
+	 * Additionally, the pointers in the problem_storage structure may be updated 
+	 * (a) depending upon the number of digit-place sorting passes needed, and (b) 
+	 * whether or not the caller has already allocated temporary storage.  
+	 * 
+	 * The sorted results will always be referenced by problem_storage.d_keys (and 
+	 * problem_storage.d_values).  However, for an odd number of sorting passes (uncommon)
+	 * these results will actually be backed by the storage initially allocated for 
+	 * by problem_storage.d_alt_keys (and problem_storage.d_alt_values).  If so, 
+	 * problem_storage.d_alt_keys and problem_storage.d_alt_keys will be updated to 
+	 * reference the original problem_storage.d_keys and problem_storage.d_values in order 
+	 * to facilitate cleanup.  
+	 * 
+	 * This means it is important to avoid keeping stale copies of device pointers 
+	 * to keys/data; you will want to re-reference the pointers in problem_storage.
+	 * 
+	 * @param[in/out] 	problem_storage 
+	 * 		Device vectors of keys and values to sort, and ancillary storage 
+	 * 		needed by the sorting kernels. See the IMPORTANT NOTES above. 
+	 * 
+	 * 		The problem_storage.[alternate_keys|alternate_values|d_spine] fields are 
+	 * 		temporary storage needed by the sorting kernels.  To facilitate 
+	 * 		speed, callers are welcome to re-use this storage for same-sized 
+	 * 		(or smaller) sortign problems. If NULL, these storage vectors will be 
+	 *      allocated by this routine (and must be subsequently cuda-freed by 
+	 *      the caller).
+	 *
+	 * @return cudaSuccess on success, error enumeration otherwise
+	 */
+	cudaError_t EnactSort(RadixSortStorage<K, V> &problem_storage);	
+
+    /*
+     * Destructor
+     */
+    virtual ~BaseRadixSortingEnactor() {}
+};
+
+
+
+template <typename K, typename V>
+BaseRadixSortingEnactor<K, V>::BaseRadixSortingEnactor(
+	int passes, 
+	int max_radix_bits, 
+	unsigned int num_elements, 
+	int max_grid_size,
+	bool swizzle_pointers_for_odd_passes) 
+{
+	//
+	// Get current device properties 
+	//
+
+	int current_device;
+	cudaGetDevice(&current_device);
+	cudaGetDeviceProperties(&_device_props, current_device);
+	_device_sm_version = _device_props.major * 100 + _device_props.minor * 10;
+
+	
+	//
+	// Get SM version of compiled kernel assembly
+	//
+	cudaFuncGetAttributes(&_spine_scan_kernel_attrs, SrtsScanSpine<void>);
+	_kernel_ptx_version = _spine_scan_kernel_attrs.ptxVersion * 10;
+	
+
+	//
+	// Determine number of CTAs to launch, shared memory, cycle elements, etc.
+	//
+
+	_passes								= passes;
+	_num_elements 						= num_elements;
+	_keys_only 							= IsKeysOnly<V>();
+	_cycle_elements 					= B40C_RADIXSORT_CYCLE_ELEMENTS(_kernel_ptx_version , ConvertedKeyType, V);
+	_grid_size 							= GridSize(max_grid_size);
+	_swizzle_pointers_for_odd_passes	= swizzle_pointers_for_odd_passes;
+	
+	int total_cycles 			= _num_elements / _cycle_elements;
+	unsigned int cycles_per_block 		= total_cycles / _grid_size;						
+	unsigned int extra_cycles 			= total_cycles - (cycles_per_block * _grid_size);
+
+	CtaDecomposition work_decomposition = {
+		extra_cycles,										// num_big_blocks
+		(cycles_per_block + 1) * _cycle_elements,			// big_block_elements
+		cycles_per_block * _cycle_elements,					// normal_block_elements
+		_num_elements - (total_cycles * _cycle_elements),	// extra_elements_last_block
+		_num_elements};										// num_elements
+	
+	_work_decomposition = work_decomposition;
+	
+	int spine_cycles = ((_grid_size * (1 << max_radix_bits)) + B40C_RADIXSORT_SPINE_CYCLE_ELEMENTS - 1) / B40C_RADIXSORT_SPINE_CYCLE_ELEMENTS;
+	_spine_elements = spine_cycles * B40C_RADIXSORT_SPINE_CYCLE_ELEMENTS;
+}
+
+
+
+template <typename K, typename V>
+int BaseRadixSortingEnactor<K, V>::GridSize(int max_grid_size)
+{
+	const int SINGLE_CTA_CUTOFF = 0;		// right now zero; we have no single-cta sorting
+
+	// find maximum number of threadblocks if "use-default"
+	if (max_grid_size == 0) {
+
+		if (_num_elements <= static_cast<unsigned int>(SINGLE_CTA_CUTOFF)) {
+
+			// The problem size is too small to warrant a two-level reduction: 
+			// use only one stream-processor
+			max_grid_size = 1;
+
+		} else {
+
+			if (_device_sm_version <= 120) {
+				
+				// G80/G90
+				max_grid_size = _device_props.multiProcessorCount * 4;
+				
+			} else if (_device_sm_version < 200) {
+				
+				// GT200 (has some kind of TLB or icache drama)
+				int orig_max_grid_size = _device_props.multiProcessorCount * B40C_RADIXSORT_SCAN_SCATTER_CTA_OCCUPANCY(_kernel_ptx_version);
+				if (_keys_only) { 
+					orig_max_grid_size *= (_num_elements + (1024 * 1024 * 96) - 1) / (1024 * 1024 * 96);
+				} else {
+					orig_max_grid_size *= (_num_elements + (1024 * 1024 * 64) - 1) / (1024 * 1024 * 64);
+				}
+				max_grid_size = orig_max_grid_size;
+
+				if (_num_elements / _cycle_elements > static_cast<unsigned int>(max_grid_size)) {
+	
+					double multiplier1 = 4.0;
+					double multiplier2 = 16.0;
+
+					double delta1 = 0.068;
+					double delta2 = 0.127;	
+	
+					int dividend = (_num_elements + _cycle_elements - 1) / _cycle_elements;
+	
+					while(true) {
+	
+						double quotient = ((double) dividend) / (multiplier1 * max_grid_size);
+						quotient -= (int) quotient;
+
+						if ((quotient > delta1) && (quotient < 1 - delta1)) {
+
+							quotient = ((double) dividend) / (multiplier2 * max_grid_size / 3.0);
+							quotient -= (int) quotient;
+
+							if ((quotient > delta2) && (quotient < 1 - delta2)) {
+								break;
+							}
+						}
+						
+						if (max_grid_size == orig_max_grid_size - 2) {
+							max_grid_size = orig_max_grid_size - 30;
+						} else {
+							max_grid_size -= 1;
+						}
+					}
+				}
+			} else {
+				
+				// GF100
+				max_grid_size = 418;
+			}
+		}
+	}
+
+	// Calculate the actual number of threadblocks to launch.  Initially
+	// assume that each threadblock will do only one cycle_elements worth 
+	// of work, but then clamp it by the "max" restriction derived above
+	// in order to accomodate the "single-sp" and "saturated" cases.
+
+	int grid_size = _num_elements / _cycle_elements;
+	if (grid_size == 0) {
+		grid_size = 1;
+	}
+	if (grid_size > max_grid_size) {
+		grid_size = max_grid_size;
+	} 
+
+	return grid_size;
+}
+
+
+
+template <typename K, typename V>
+bool BaseRadixSortingEnactor<K, V>::
+CanFit() 
+{
+	long long bytes = (_num_elements * sizeof(K) * 2) + (_spine_elements * sizeof(int));
+	if (!_keys_only) bytes += _num_elements * sizeof(V) * 2;
+
+	if (_device_props.totalGlobalMem < 1024 * 1024 * 513) {
+		return (bytes < ((double) _device_props.totalGlobalMem) * 0.81); 	// allow up to 81% capacity for 512MB   
+	}
+	
+	return (bytes < ((double) _device_props.totalGlobalMem) * 0.89); 	// allow up to 90% capacity 
+}
+
+
+
+template <typename K, typename V>
+template <int PASS, int RADIX_BITS, int BIT, typename PreprocessFunctor, typename PostprocessFunctor>
+cudaError_t BaseRadixSortingEnactor<K, V>::
+DigitPlacePass(const RadixSortStorage<ConvertedKeyType, V> &converted_storage)
+{
+	int threads = B40C_RADIXSORT_THREADS;
+	int dynamic_smem;
+
+	cudaFuncAttributes reduce_kernel_attrs, scan_scatter_attrs;
+	cudaFuncGetAttributes(&reduce_kernel_attrs, RakingReduction<ConvertedKeyType, V, PASS, RADIX_BITS, BIT, PreprocessFunctor>);
+	cudaFuncGetAttributes(&scan_scatter_attrs, ScanScatterDigits<ConvertedKeyType, V, PASS, RADIX_BITS, BIT, PreprocessFunctor, PostprocessFunctor>);
+	
+	//
+	// Counting Reduction
+	//
+
+	// Run tesla flush kernel if we have two or more threadblocks for each of the SMs
+	if ((_device_sm_version == 130) && (_work_decomposition.num_elements > static_cast<unsigned int>(_device_props.multiProcessorCount * _cycle_elements * 2))) { 
+		FlushKernel<void><<<_grid_size, B40C_RADIXSORT_THREADS, scan_scatter_attrs.sharedSizeBytes>>>();
+		synchronize_if_enabled("FlushKernel");
+	}
+
+	// GF100 and GT200 get the same smem allocation for every kernel launch (pad the reduction/top-level-scan kernels)
+	dynamic_smem = (_kernel_ptx_version >= 130) ? scan_scatter_attrs.sharedSizeBytes - reduce_kernel_attrs.sharedSizeBytes : 0;
+
+	RakingReduction<ConvertedKeyType, V, PASS, RADIX_BITS, BIT, PreprocessFunctor> <<<_grid_size, threads, dynamic_smem>>>(
+		converted_storage.d_from_alt_storage,
+		converted_storage.d_spine,
+		converted_storage.d_keys,
+		converted_storage.d_alt_keys,
+		_work_decomposition);
+    synchronize_if_enabled("RakingReduction");
+
+	
+	//
+	// Spine
+	//
+	
+	// GF100 and GT200 get the same smem allocation for every kernel launch (pad the reduction/top-level-scan kernels)
+	dynamic_smem = (_kernel_ptx_version >= 130) ? scan_scatter_attrs.sharedSizeBytes - _spine_scan_kernel_attrs.sharedSizeBytes : 0;
+	
+	SrtsScanSpine<void><<<_grid_size, B40C_RADIXSORT_SPINE_THREADS, dynamic_smem>>>(
+		converted_storage.d_spine,
+		converted_storage.d_spine,
+		_spine_elements);
+    synchronize_if_enabled("SrtsScanSpine");
+
+	
+	//
+	// Scanning Scatter
+	//
+	
+	// Run tesla flush kernel if we have two or more threadblocks for each of the SMs
+	if ((_device_sm_version == 130) && (_work_decomposition.num_elements > static_cast<unsigned int>(_device_props.multiProcessorCount * _cycle_elements * 2))) { 
+		FlushKernel<void><<<_grid_size, B40C_RADIXSORT_THREADS, scan_scatter_attrs.sharedSizeBytes>>>();
+		synchronize_if_enabled("FlushKernel");
+	}
+
+	ScanScatterDigits<ConvertedKeyType, V, PASS, RADIX_BITS, BIT, PreprocessFunctor, PostprocessFunctor> <<<_grid_size, threads, 0>>>(
+		converted_storage.d_from_alt_storage,
+		converted_storage.d_spine,
+		converted_storage.d_keys,
+		converted_storage.d_alt_keys,
+		converted_storage.d_values,
+		converted_storage.d_alt_values,
+		_work_decomposition);
+    synchronize_if_enabled("ScanScatterDigits");
+
+	return cudaSuccess;
+}
+
+
+
+template <typename K, typename V>
+cudaError_t BaseRadixSortingEnactor<K, V>::
+EnactSort(RadixSortStorage<K, V> &problem_storage) 
+{
+	//
+	// Allocate device memory for temporary storage (if necessary)
+	//
+
+	if (problem_storage.d_alt_keys == NULL) {
+		cudaMalloc((void**) &problem_storage.d_alt_keys, _num_elements * sizeof(K));
+	}
+	if (!_keys_only && (problem_storage.d_alt_values == NULL)) {
+		cudaMalloc((void**) &problem_storage.d_alt_values, _num_elements * sizeof(V));
+	}
+	if (problem_storage.d_spine == NULL) {
+		cudaMalloc((void**) &problem_storage.d_spine, _spine_elements * sizeof(int));
+	}
+	if (problem_storage.d_from_alt_storage == NULL) {
+		cudaMalloc((void**) &problem_storage.d_from_alt_storage, 2 * sizeof(bool));
+	}
+
+	// Determine suitable type of unsigned byte storage to use for keys 
+	typedef typename KeyConversion<K>::UnsignedBits ConvertedKeyType;
+	
+	// Copy storage pointers to an appropriately typed stucture 
+	RadixSortStorage<ConvertedKeyType, V> converted_storage;
+	memcpy(&converted_storage, &problem_storage, sizeof(RadixSortStorage<K, V>));
+
+	// 
+	// Enact the sorting operation
+	//
+	
+	if (RADIXSORT_DEBUG) {
+		
+		printf("_device_sm_version: %d, _kernel_ptx_version: %d\n", _device_sm_version, _kernel_ptx_version);
+		printf("Bottom-level reduction & scan kernels:\n\tgrid_size: %d, \n\tthreads: %d, \n\tcycle_elements: %d, \n\tnum_big_blocks: %d, \n\tbig_block_elements: %d, \n\tnormal_block_elements: %d\n\textra_elements_last_block: %d\n\n",
+			_grid_size, B40C_RADIXSORT_THREADS, _cycle_elements, _work_decomposition.num_big_blocks, _work_decomposition.big_block_elements, _work_decomposition.normal_block_elements, _work_decomposition.extra_elements_last_block);
+		printf("Top-level spine scan:\n\tgrid_size: %d, \n\tthreads: %d, \n\tspine_block_elements: %d\n\n", 
+			_grid_size, B40C_RADIXSORT_SPINE_THREADS, _spine_elements);
+	}	
+
+	cudaError_t retval = EnactDigitPlacePasses(converted_storage);
+
+	
+	//
+	// Swizzle pointers if we left our sorted output in temp storage 
+	//
+	
+	if (_swizzle_pointers_for_odd_passes) {
+	
+		cudaMemcpy(
+			&problem_storage.using_alternate_storage, 
+			&problem_storage.d_from_alt_storage[_passes & 0x1], 
+			sizeof(bool), 
+			cudaMemcpyDeviceToHost);
+	
+		if (problem_storage.using_alternate_storage) {
+            thrust::swap<K*>(problem_storage.d_keys, problem_storage.d_alt_keys);
+			if (!_keys_only) {
+                thrust::swap<V*>(problem_storage.d_values, problem_storage.d_alt_values);
+			}
+		}
+	}
+	
+	return retval;
+}
+
+
+
+
+
+/******************************************************************************
+ * Sorting enactor classes
+ ******************************************************************************/
+
+/**
+ * Generic sorting enactor class.  Simply create an instance of this class
+ * with your key-type K (and optionally value-type V if sorting with satellite 
+ * values).
+ * 
+ * Template specialization provides the appropriate enactor instance to handle 
+ * the specified data types. 
+ * 
+ * @template-param K
+ * 		Type of keys to be sorted
+ *
+ * @template-param V
+ * 		Type of values to be sorted.
+ *
+ * @template-param ConvertedKeyType
+ * 		Leave as default to effect necessary enactor specialization.
+ */
+template <typename K, typename V = KeysOnlyType, typename ConvertedKeyType = typename KeyConversion<K>::UnsignedBits>
+class RadixSortingEnactor;
+
+
+
+/**
+ * Sorting enactor that is specialized for for 8-bit key types
+ */
+template <typename K, typename V>
+class RadixSortingEnactor<K, V, unsigned char> : public BaseRadixSortingEnactor<K, V>
+{
+protected:
+
+	typedef BaseRadixSortingEnactor<K, V> Base; 
+	typedef typename Base::ConvertedKeyType ConvertedKeyType;
+
+	cudaError_t EnactDigitPlacePasses(const RadixSortStorage<ConvertedKeyType, V> &converted_storage)
+	{
+		Base::template DigitPlacePass<0, 4, 0, PreprocessKeyFunctor<K>,      NopFunctor<ConvertedKeyType> >(converted_storage);
+		Base::template DigitPlacePass<1, 4, 4, NopFunctor<ConvertedKeyType>, PostprocessKeyFunctor<K> >    (converted_storage); 
+
+		return cudaSuccess;
+	}
+
+public:
+	
+	/**
+	 * Constructor.
+	 * 
+	 * @param[in] 		num_elements 
+	 * 		Length (in elements) of the input to a sorting operation
+	 * 
+	 * @param[in] 		max_grid_size  
+	 * 		Maximum allowable number of CTAs to launch.  The default value of 0 indicates 
+	 * 		that the dispatch logic should select an appropriate value for the target device.
+	 */	
+	RadixSortingEnactor(unsigned int num_elements, int max_grid_size = 0) : Base::BaseRadixSortingEnactor(2, 4, num_elements, max_grid_size) {}
+
+};
+
+
+
+/**
+ * Sorting enactor that is specialized for for 16-bit key types
+ */
+template <typename K, typename V>
+class RadixSortingEnactor<K, V, unsigned short> : public BaseRadixSortingEnactor<K, V>
+{
+protected:
+
+	typedef BaseRadixSortingEnactor<K, V> Base; 
+	typedef typename Base::ConvertedKeyType ConvertedKeyType;
+
+	cudaError_t EnactDigitPlacePasses(const RadixSortStorage<ConvertedKeyType, V> &converted_storage)
+	{
+		Base::template DigitPlacePass<0, 4, 0,  PreprocessKeyFunctor<K>,      NopFunctor<ConvertedKeyType> >(converted_storage);
+		Base::template DigitPlacePass<1, 4, 4,  NopFunctor<ConvertedKeyType>, NopFunctor<ConvertedKeyType> >(converted_storage); 
+		Base::template DigitPlacePass<2, 4, 8,  NopFunctor<ConvertedKeyType>, NopFunctor<ConvertedKeyType> >(converted_storage); 
+		Base::template DigitPlacePass<3, 4, 12, NopFunctor<ConvertedKeyType>, PostprocessKeyFunctor<K> >    (converted_storage); 
+
+		return cudaSuccess;
+	}
+
+public:
+	
+	/**
+	 * Constructor.
+	 * 
+	 * @param[in] 		num_elements 
+	 * 		Length (in elements) of the input to a sorting operation
+	 * 
+	 * @param[in] 		max_grid_size  
+	 * 		Maximum allowable number of CTAs to launch.  The default value of 0 indicates 
+	 * 		that the dispatch logic should select an appropriate value for the target device.
+	 */	
+	RadixSortingEnactor(unsigned int num_elements, int max_grid_size = 0) : Base::BaseRadixSortingEnactor(4, 4, num_elements, max_grid_size) {}
+
+};
+
+
+/**
+ * Sorting enactor that is specialized for for 32-bit key types
+ */
+template <typename K, typename V>
+class RadixSortingEnactor<K, V, unsigned int> : public BaseRadixSortingEnactor<K, V>
+{
+protected:
+
+	typedef BaseRadixSortingEnactor<K, V> Base; 
+	typedef typename Base::ConvertedKeyType ConvertedKeyType;
+
+	cudaError_t EnactDigitPlacePasses(const RadixSortStorage<ConvertedKeyType, V> &converted_storage)
+	{
+		Base::template DigitPlacePass<0, 4, 0,  PreprocessKeyFunctor<K>,      NopFunctor<ConvertedKeyType> >(converted_storage);
+		Base::template DigitPlacePass<1, 4, 4,  NopFunctor<ConvertedKeyType>, NopFunctor<ConvertedKeyType> >(converted_storage); 
+		Base::template DigitPlacePass<2, 4, 8,  NopFunctor<ConvertedKeyType>, NopFunctor<ConvertedKeyType> >(converted_storage); 
+		Base::template DigitPlacePass<3, 4, 12, NopFunctor<ConvertedKeyType>, NopFunctor<ConvertedKeyType> >(converted_storage); 
+		Base::template DigitPlacePass<4, 4, 16, NopFunctor<ConvertedKeyType>, NopFunctor<ConvertedKeyType> >(converted_storage); 
+		Base::template DigitPlacePass<5, 4, 20, NopFunctor<ConvertedKeyType>, NopFunctor<ConvertedKeyType> >(converted_storage); 
+		Base::template DigitPlacePass<6, 4, 24, NopFunctor<ConvertedKeyType>, NopFunctor<ConvertedKeyType> >(converted_storage); 
+		Base::template DigitPlacePass<7, 4, 28, NopFunctor<ConvertedKeyType>, PostprocessKeyFunctor<K> >    (converted_storage); 
+
+		return cudaSuccess;
+	}
+
+public:
+	
+	/**
+	 * Constructor.
+	 * 
+	 * @param[in] 		num_elements 
+	 * 		Length (in elements) of the input to a sorting operation
+	 * 
+	 * @param[in] 		max_grid_size  
+	 * 		Maximum allowable number of CTAs to launch.  The default value of 0 indicates 
+	 * 		that the dispatch logic should select an appropriate value for the target device.
+	 */	
+	RadixSortingEnactor(unsigned int num_elements, int max_grid_size = 0) : Base::BaseRadixSortingEnactor(8, 4, num_elements, max_grid_size) {}
+
+};
+
+
+
+/**
+ * Sorting enactor that is specialized for for 64-bit key types
+ */
+template <typename K, typename V>
+class RadixSortingEnactor<K, V, unsigned long long> : public BaseRadixSortingEnactor<K, V>
+{
+protected:
+
+	typedef BaseRadixSortingEnactor<K, V> Base; 
+	typedef typename Base::ConvertedKeyType ConvertedKeyType;
+
+	cudaError_t EnactDigitPlacePasses(const RadixSortStorage<ConvertedKeyType, V> &converted_storage)
+	{
+		Base::template DigitPlacePass<0,  4, 0,  PreprocessKeyFunctor<K>,      NopFunctor<ConvertedKeyType> >(converted_storage);
+		Base::template DigitPlacePass<1,  4, 4,  NopFunctor<ConvertedKeyType>, NopFunctor<ConvertedKeyType> >(converted_storage); 
+		Base::template DigitPlacePass<2,  4, 8,  NopFunctor<ConvertedKeyType>, NopFunctor<ConvertedKeyType> >(converted_storage); 
+		Base::template DigitPlacePass<3,  4, 12, NopFunctor<ConvertedKeyType>, NopFunctor<ConvertedKeyType> >(converted_storage); 
+		Base::template DigitPlacePass<4,  4, 16, NopFunctor<ConvertedKeyType>, NopFunctor<ConvertedKeyType> >(converted_storage); 
+		Base::template DigitPlacePass<5,  4, 20, NopFunctor<ConvertedKeyType>, NopFunctor<ConvertedKeyType> >(converted_storage); 
+		Base::template DigitPlacePass<6,  4, 24, NopFunctor<ConvertedKeyType>, NopFunctor<ConvertedKeyType> >(converted_storage); 
+		Base::template DigitPlacePass<7,  4, 28, NopFunctor<ConvertedKeyType>, NopFunctor<ConvertedKeyType> >(converted_storage); 
+		Base::template DigitPlacePass<8,  4, 32, NopFunctor<ConvertedKeyType>, NopFunctor<ConvertedKeyType> >(converted_storage);
+		Base::template DigitPlacePass<9,  4, 36, NopFunctor<ConvertedKeyType>, NopFunctor<ConvertedKeyType> >(converted_storage); 
+		Base::template DigitPlacePass<10, 4, 40, NopFunctor<ConvertedKeyType>, NopFunctor<ConvertedKeyType> >(converted_storage); 
+		Base::template DigitPlacePass<11, 4, 44, NopFunctor<ConvertedKeyType>, NopFunctor<ConvertedKeyType> >(converted_storage); 
+		Base::template DigitPlacePass<12, 4, 48, NopFunctor<ConvertedKeyType>, NopFunctor<ConvertedKeyType> >(converted_storage); 
+		Base::template DigitPlacePass<13, 4, 52, NopFunctor<ConvertedKeyType>, NopFunctor<ConvertedKeyType> >(converted_storage); 
+		Base::template DigitPlacePass<14, 4, 56, NopFunctor<ConvertedKeyType>, NopFunctor<ConvertedKeyType> >(converted_storage); 
+		Base::template DigitPlacePass<15, 4, 60, NopFunctor<ConvertedKeyType>, PostprocessKeyFunctor<K> >    (converted_storage); 
+
+		return cudaSuccess;
+	}
+
+public:
+	
+	/**
+	 * Constructor.
+	 * 
+	 * @param[in] 		num_elements 
+	 * 		Length (in elements) of the input to a sorting operation
+	 * 
+	 * @param[in] 		max_grid_size  
+	 * 		Maximum allowable number of CTAs to launch.  The default value of 0 indicates 
+	 * 		that the dispatch logic should select an appropriate value for the target device.
+	 */	
+	RadixSortingEnactor(unsigned int num_elements, int max_grid_size = 0) : Base::BaseRadixSortingEnactor(16, 4, num_elements, max_grid_size) {}
+
+};
+
+
+} // end namespace b40c_thrust
+} // end namespace detail
+} // end namespace detail
+} // end namespace cuda
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/cuda/detail/detail/b40c/radixsort_kernel_common.h b/compat/thrust/system/cuda/detail/detail/b40c/radixsort_kernel_common.h
new file mode 100644
index 0000000000..7899dc3c0e
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/detail/b40c/radixsort_kernel_common.h
@@ -0,0 +1,173 @@
+/******************************************************************************
+ * 
+ * Copyright 2010 Duane Merrill
+ * 
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ * 
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License. 
+ * 
+ * 
+ * 
+ * 
+ * AUTHORS' REQUEST: 
+ * 
+ * 		If you use|reference|benchmark this code, please cite our Technical 
+ * 		Report (http://www.cs.virginia.edu/~dgm4d/papers/RadixSortTR.pdf):
+ * 
+ *		@TechReport{ Merrill:Sorting:2010,
+ *        	author = "Duane Merrill and Andrew Grimshaw",
+ *        	title = "Revisiting Sorting for GPGPU Stream Architectures",
+ *        	year = "2010",
+ *        	institution = "University of Virginia, Department of Computer Science",
+ *        	address = "Charlottesville, VA, USA",
+ *        	number = "CS2010-03"
+ *		}
+ * 
+ * For more information, see our Google Code project site: 
+ * http://code.google.com/p/back40computing/
+ * 
+ * Thanks!
+ * 
+ ******************************************************************************/
+
+
+/******************************************************************************
+ * Configuration management for B40C radix sorting kernels  
+ ******************************************************************************/
+
+#pragma once
+
+#include "kernel_utils.h"
+#include "vector_types.h"
+#include "radixsort_key_conversion.h"
+
+namespace thrust  {
+namespace system  {
+namespace cuda    {
+namespace detail  {
+namespace detail  {
+namespace b40c_thrust   {
+
+
+/******************************************************************************
+ * Radix sorting configuration  
+ ******************************************************************************/
+
+// 128 threads
+#define B40C_RADIXSORT_LOG_THREADS						7								
+#define B40C_RADIXSORT_THREADS							(1 << B40C_RADIXSORT_LOG_THREADS)	
+
+// Target threadblock occupancy for counting/reduction kernel
+#define B40C_SM20_REDUCE_CTA_OCCUPANCY()					(8)			// 8 threadblocks on GF100
+#define B40C_SM12_REDUCE_CTA_OCCUPANCY()					(5)			// 5 threadblocks on GT200
+#define B40C_SM10_REDUCE_CTA_OCCUPANCY()					(3)			// 4 threadblocks on G80
+#define B40C_RADIXSORT_REDUCE_CTA_OCCUPANCY(version)		((version >= 200) ? B40C_SM20_REDUCE_CTA_OCCUPANCY() : 	\
+			        										 (version >= 120) ? B40C_SM12_REDUCE_CTA_OCCUPANCY() : 	\
+					        													B40C_SM10_REDUCE_CTA_OCCUPANCY())		
+													                    
+// Target threadblock occupancy for bulk scan/scatter kernel
+#define B40C_SM20_SCAN_SCATTER_CTA_OCCUPANCY()				(7)			// 7 threadblocks on GF100
+#define B40C_SM12_SCAN_SCATTER_CTA_OCCUPANCY()				(5)			// 5 threadblocks on GT200
+#define B40C_SM10_SCAN_SCATTER_CTA_OCCUPANCY()				(2)			// 2 threadblocks on G80
+#define B40C_RADIXSORT_SCAN_SCATTER_CTA_OCCUPANCY(version)	((version >= 200) ? B40C_SM20_SCAN_SCATTER_CTA_OCCUPANCY() : 	\
+			    											 (version >= 120) ? B40C_SM12_SCAN_SCATTER_CTA_OCCUPANCY() : 	\
+				    															B40C_SM10_SCAN_SCATTER_CTA_OCCUPANCY())		
+
+// Number of 256-element sets to rake per raking pass
+#define B40C_SM20_LOG_SETS_PER_PASS()					(1)			// 2 sets on GF100
+#define B40C_SM12_LOG_SETS_PER_PASS()					(0)			// 1 set on GT200
+#define B40C_SM10_LOG_SETS_PER_PASS()					(1)			// 2 sets on G80
+#define B40C_RADIXSORT_LOG_SETS_PER_PASS(version)		((version >= 200) ? B40C_SM20_LOG_SETS_PER_PASS() : 	\
+			     										 (version >= 120) ? B40C_SM12_LOG_SETS_PER_PASS() : 	\
+				    														B40C_SM10_LOG_SETS_PER_PASS())		
+
+// Number of raking passes per cycle
+#define B40C_SM20_LOG_PASSES_PER_CYCLE(K, V)					(((B40C_MAX(sizeof(K), sizeof(V)) > 4) || _B40C_LP64_) ? 0 : 1)	// 2 passes on GF100 (only one for large keys/values, or for 64-bit device pointers)
+#define B40C_SM12_LOG_PASSES_PER_CYCLE(K, V)					(B40C_MAX(sizeof(K), sizeof(V)) > 4 ? 0 : 1)					// 2 passes on GT200 (only for large keys/values)
+#define B40C_SM10_LOG_PASSES_PER_CYCLE(K, V)					(0)																// 1 pass on G80
+#define B40C_RADIXSORT_LOG_PASSES_PER_CYCLE(version, K, V)	((version >= 200) ? B40C_SM20_LOG_PASSES_PER_CYCLE(K, V) : 	\
+				    										 (version >= 120) ? B40C_SM12_LOG_PASSES_PER_CYCLE(K, V) : 	\
+					    														B40C_SM10_LOG_PASSES_PER_CYCLE(K, V))		
+
+
+// Number of raking threads per raking pass
+#define B40C_SM20_LOG_RAKING_THREADS_PER_PASS()				(B40C_LOG_WARP_THREADS + 1)		// 2 raking warps on GF100
+#define B40C_SM12_LOG_RAKING_THREADS_PER_PASS()				(B40C_LOG_WARP_THREADS)			// 1 raking warp on GT200
+#define B40C_SM10_LOG_RAKING_THREADS_PER_PASS()				(B40C_LOG_WARP_THREADS + 2)		// 4 raking warps on G80
+#define B40C_RADIXSORT_LOG_RAKING_THREADS_PER_PASS(version)	((version >= 200) ? B40C_SM20_LOG_RAKING_THREADS_PER_PASS() : 	\
+				    										 (version >= 120) ? B40C_SM12_LOG_RAKING_THREADS_PER_PASS() : 	\
+					    														B40C_SM10_LOG_RAKING_THREADS_PER_PASS())		
+
+
+// Number of elements per cycle
+#define B40C_RADIXSORT_LOG_CYCLE_ELEMENTS(version, K, V)		(B40C_RADIXSORT_LOG_SETS_PER_PASS(version) + B40C_RADIXSORT_LOG_PASSES_PER_CYCLE(version, K, V) + B40C_RADIXSORT_LOG_THREADS + 1)
+#define B40C_RADIXSORT_CYCLE_ELEMENTS(version, K, V)			(1 << B40C_RADIXSORT_LOG_CYCLE_ELEMENTS(version, K, V))
+
+// Number of warps per CTA
+#define B40C_RADIXSORT_LOG_WARPS								(B40C_RADIXSORT_LOG_THREADS - B40C_LOG_WARP_THREADS)
+#define B40C_RADIXSORT_WARPS									(1 << B40C_RADIXSORT_LOG_WARPS)
+
+// Number of threads for spine-scanning kernel
+#define B40C_RADIXSORT_LOG_SPINE_THREADS						7		// 128 threads
+#define B40C_RADIXSORT_SPINE_THREADS							(1 << B40C_RADIXSORT_LOG_SPINE_THREADS)	
+
+// Number of elements per spine-scanning cycle
+#define B40C_RADIXSORT_LOG_SPINE_CYCLE_ELEMENTS  				9		// 512 elements
+#define B40C_RADIXSORT_SPINE_CYCLE_ELEMENTS		    			(1 << B40C_RADIXSORT_LOG_SPINE_CYCLE_ELEMENTS)
+
+
+
+/******************************************************************************
+ * SRTS Control Structures
+ ******************************************************************************/
+
+
+/**
+ * Value-type structure denoting keys-only sorting
+ */
+struct KeysOnlyType {};
+
+/**
+ * Returns whether or not the templated type indicates keys-only sorting
+ */
+template <typename V>
+inline __host__ __device__ bool IsKeysOnly() {return false;}
+
+
+/**
+ * Returns whether or not the templated type indicates keys-only sorting
+ */
+template <>
+inline __host__ __device__ bool IsKeysOnly<KeysOnlyType>() {return true;}
+
+
+/**
+ * A given threadblock may receive one of three different amounts of 
+ * work: "big", "normal", and "last".  The big workloads are one
+ * cycle_elements greater than the normal, and the last workload 
+ * does the extra (problem-size % cycle_elements) work.
+ */
+struct CtaDecomposition {
+	unsigned int num_big_blocks;
+	unsigned int big_block_elements;
+	unsigned int normal_block_elements;
+	unsigned int extra_elements_last_block;
+	unsigned int num_elements;
+};
+
+
+} // end namespace b40c_thrust
+} // end namespace detail
+} // end namespace detail
+} // end namespace cuda
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/cuda/detail/detail/b40c/radixsort_key_conversion.h b/compat/thrust/system/cuda/detail/detail/b40c/radixsort_key_conversion.h
new file mode 100644
index 0000000000..a170f95e6c
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/detail/b40c/radixsort_key_conversion.h
@@ -0,0 +1,352 @@
+/******************************************************************************
+ * 
+ * Copyright 2010 Duane Merrill
+ * 
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ * 
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License. 
+ * 
+ * 
+ * 
+ * 
+ * AUTHORS' REQUEST: 
+ * 
+ * 		If you use|reference|benchmark this code, please cite our Technical 
+ * 		Report (http://www.cs.virginia.edu/~dgm4d/papers/RadixSortTR.pdf):
+ * 
+ *		@TechReport{ Merrill:Sorting:2010,
+ *        	author = "Duane Merrill and Andrew Grimshaw",
+ *        	title = "Revisiting Sorting for GPGPU Stream Architectures",
+ *        	year = "2010",
+ *        	institution = "University of Virginia, Department of Computer Science",
+ *        	address = "Charlottesville, VA, USA",
+ *        	number = "CS2010-03"
+ *		}
+ * 
+ * For more information, see our Google Code project site: 
+ * http://code.google.com/p/back40computing/
+ * 
+ * Thanks!
+ * 
+ ******************************************************************************/
+
+
+/******************************************************************************
+ * Functors for converting signed and floating point types to unsigned types
+ * suitable for radix sorting  
+ ******************************************************************************/
+
+#pragma once
+
+namespace thrust  {
+namespace system  {
+namespace cuda    {
+namespace detail  {
+namespace detail  {
+namespace b40c_thrust   {
+
+
+//
+// Do-nothing functors
+//
+
+template <typename T>
+struct NopFunctor{
+    template <typename ConvertedKeyType>
+	__device__ __host__ __forceinline__ void operator()(ConvertedKeyType &converted_key) {}
+	__device__ __host__ __forceinline__ static bool MustApply(){ return false;}
+};
+
+//
+// Do-nothing functors that indicate a mandatory pass
+//
+
+template <typename T>
+struct MandatoryPassNopFunctor{
+    template <typename ConvertedKeyType>
+	__device__ __host__ __forceinline__ void operator()(ConvertedKeyType &converted_key) {}
+	__device__ __host__ __forceinline__ static bool MustApply(){ return false;}
+};
+
+
+//
+// Conversion for generic unsigned types
+//
+
+template <typename T> struct KeyConversion {
+	typedef T UnsignedBits;
+};
+
+template <typename T>
+struct PreprocessKeyFunctor{
+    template <typename ConvertedKeyType>
+	__device__ __host__ __forceinline__ void operator()(ConvertedKeyType &converted_key) {}
+	__device__ __host__ __forceinline__ static bool MustApply(){ return false;}
+};
+
+template <typename T>
+struct PostprocessKeyFunctor {
+    template <typename ConvertedKeyType>
+	__device__ __host__ __forceinline__ void operator()(ConvertedKeyType &converted_key) {}
+	__device__ __host__ __forceinline__ static bool MustApply(){ return false;}
+};
+
+
+
+//
+// Conversion for floats
+//
+
+template <> struct KeyConversion<float> {
+	typedef unsigned int UnsignedBits;
+};
+
+template <>
+struct PreprocessKeyFunctor<float> {
+	__device__ __host__ __forceinline__ void operator()(unsigned int &converted_key) {
+
+		unsigned int mask = (converted_key & 0x80000000) ? 0xffffffff : 0x80000000; 
+		converted_key ^= mask;
+	}
+	__device__ __host__ __forceinline__ static bool MustApply(){ return true;}
+};
+
+template <>
+struct PostprocessKeyFunctor<float> {
+	__device__ __host__ __forceinline__ void operator()(unsigned int &converted_key) {
+
+		unsigned int mask = (converted_key & 0x80000000) ? 0x80000000 : 0xffffffff; 
+		converted_key ^= mask;
+    }
+	__device__ __host__ __forceinline__ static bool MustApply(){ return true;}
+};
+
+
+
+//
+// Conversion for doubles
+//
+
+template <> struct KeyConversion<double> {
+	typedef unsigned long long UnsignedBits;
+};
+
+template <>
+struct PreprocessKeyFunctor<double> {
+	__device__ __host__ __forceinline__ void operator()(unsigned long long &converted_key) {
+
+		unsigned long long mask = (converted_key & 0x8000000000000000) ? 0xffffffffffffffff : 0x8000000000000000; 
+		converted_key ^= mask;
+	}
+	__device__ __host__ __forceinline__ static bool MustApply(){ return true;}
+};
+
+template <>
+struct PostprocessKeyFunctor<double> {
+	__device__ __host__ __forceinline__ void operator()(unsigned long long &converted_key)  {
+		unsigned long long mask = (converted_key & 0x8000000000000000) ? 0x8000000000000000 : 0xffffffffffffffff; 
+        converted_key ^= mask;
+    }
+	__device__ __host__ __forceinline__ static bool MustApply(){ return true;}
+};
+
+
+//
+// Conversion for signed chars
+//
+
+template <> struct KeyConversion<char> {
+  typedef unsigned char UnsignedBits;
+};
+
+template <>
+struct PreprocessKeyFunctor<char> {
+  __device__ __host__ __forceinline__ void operator()(unsigned char &converted_key) {
+    // char is unsigned on some platforms, so we have to check
+    if(std::numeric_limits<char>::is_signed)
+    {
+      const unsigned int SIGN_MASK = 1u << ((sizeof(char) * 8) - 1);
+      converted_key ^= SIGN_MASK;	
+    }
+  }
+  __device__ __host__ __forceinline__ static bool MustApply(){ return std::numeric_limits<char>::is_signed;}
+};
+
+template <>
+struct PostprocessKeyFunctor<char> {
+  __device__ __host__ __forceinline__ void operator()(unsigned char &converted_key)  {
+    // char is unsigned on some platforms, so we have to check
+    if(std::numeric_limits<char>::is_signed)
+    {
+      const unsigned int SIGN_MASK = 1u << ((sizeof(char) * 8) - 1);
+      converted_key ^= SIGN_MASK;	
+    }
+  }
+  __device__ __host__ __forceinline__ static bool MustApply(){ return std::numeric_limits<char>::is_signed;}
+};
+
+
+// TODO handle this more gracefully
+template <> struct KeyConversion<signed char> {
+	typedef unsigned char UnsignedBits;
+};
+
+template <>
+struct PreprocessKeyFunctor<signed char> {
+	__device__ __host__ __forceinline__ void operator()(unsigned char &converted_key) {
+		const unsigned int SIGN_MASK = 1u << ((sizeof(char) * 8) - 1);
+		converted_key ^= SIGN_MASK;	
+	}
+	__device__ __host__ __forceinline__ static bool MustApply(){ return true;}
+};
+
+template <>
+struct PostprocessKeyFunctor<signed char> {
+	__device__ __host__ __forceinline__ void operator()(unsigned char &converted_key)  {
+		const unsigned int SIGN_MASK = 1u << ((sizeof(char) * 8) - 1);
+		converted_key ^= SIGN_MASK;	
+    }
+	__device__ __host__ __forceinline__ static bool MustApply(){ return true;}
+};
+
+
+//
+// Conversion for signed shorts
+//
+
+template <> struct KeyConversion<short> {
+	typedef unsigned short UnsignedBits;
+};
+
+template <>
+struct PreprocessKeyFunctor<short> {
+	__device__ __host__ __forceinline__ void operator()(unsigned short &converted_key) {
+		const unsigned int SIGN_MASK = 1u << ((sizeof(short) * 8) - 1);
+		converted_key ^= SIGN_MASK;	
+	}
+	__device__ __host__ __forceinline__ static bool MustApply(){ return true;}
+};
+
+template <>
+struct PostprocessKeyFunctor<short> {
+	__device__ __host__ __forceinline__ void operator()(unsigned short &converted_key)  {
+		const unsigned int SIGN_MASK = 1u << ((sizeof(short) * 8) - 1);
+		converted_key ^= SIGN_MASK;	
+    }
+	__device__ __host__ __forceinline__ static bool MustApply(){ return true;}
+};
+
+
+
+//
+// Conversion for signed ints
+//
+
+template <> struct KeyConversion<int> {
+	typedef unsigned int UnsignedBits;
+};
+
+template <>
+struct PreprocessKeyFunctor<int> {
+	__device__ __host__ __forceinline__ void operator()(unsigned int &converted_key) {
+		const unsigned int SIGN_MASK = 1u << ((sizeof(int) * 8) - 1);
+		converted_key ^= SIGN_MASK;	
+	}
+	__device__ __host__ __forceinline__ static bool MustApply(){ return true;}
+};
+
+template <>
+struct PostprocessKeyFunctor<int> {
+	__device__ __host__ __forceinline__ void operator()(unsigned int &converted_key)  {
+		const unsigned int SIGN_MASK = 1u << ((sizeof(int) * 8) - 1);
+		converted_key ^= SIGN_MASK;	
+    }
+	__device__ __host__ __forceinline__ static bool MustApply(){ return true;}
+};
+
+
+
+//
+// Conversion for signed longs
+//
+
+// TODO rework this with metaprogramming
+template <> struct KeyConversion<unsigned long> {
+#if ULONG_MAX == UINT_MAX
+    typedef unsigned int UnsignedBits;
+#else
+    typedef unsigned long long UnsignedBits;
+#endif
+};
+
+// TODO rework this with metaprogramming
+template <> struct KeyConversion<long> {
+#if ULONG_MAX == UINT_MAX
+    typedef unsigned int UnsignedBits;
+#else
+    typedef unsigned long long UnsignedBits;
+#endif
+};
+
+template <>
+struct PreprocessKeyFunctor<long> {
+	__device__ __host__ __forceinline__ void operator()(typename KeyConversion<long>::UnsignedBits& converted_key) {
+		const typename KeyConversion<long>::UnsignedBits SIGN_MASK = 1ul << ((sizeof(long) * 8) - 1);
+		converted_key ^= SIGN_MASK;	
+	}
+	__device__ __host__ __forceinline__ static bool MustApply(){ return true;}
+};
+
+template <>
+struct PostprocessKeyFunctor<long> {
+	__device__ __host__ __forceinline__ void operator()(typename KeyConversion<long>::UnsignedBits& converted_key) {
+		const typename KeyConversion<long>::UnsignedBits SIGN_MASK = 1ul << ((sizeof(long) * 8) - 1);
+		converted_key ^= SIGN_MASK;	
+    }
+	__device__ __host__ __forceinline__ static bool MustApply(){ return true;}
+};
+
+
+
+//
+// Conversion for signed long longs 
+//
+
+template <> struct KeyConversion<long long> {
+	typedef unsigned long long UnsignedBits;
+};
+
+template <>
+struct PreprocessKeyFunctor<long long> {
+	__device__ __host__ __forceinline__ void operator()(unsigned long long &converted_key) {
+		const unsigned long long SIGN_MASK = 1ull << ((sizeof(long long) * 8) - 1);
+		converted_key ^= SIGN_MASK;	
+	}
+	__device__ __host__ __forceinline__ static bool MustApply(){ return true;}
+};
+
+template <>
+struct PostprocessKeyFunctor<long long> {
+	__device__ __host__ __forceinline__ void operator()(unsigned long long &converted_key)  {
+		const unsigned long long SIGN_MASK = 1ull << ((sizeof(long long) * 8) - 1);
+		converted_key ^= SIGN_MASK;	
+    }
+	__device__ __host__ __forceinline__ static bool MustApply(){ return true;}
+};
+
+
+} // end namespace b40c_thrust
+} // end namespace detail
+} // end namespace detail
+} // end namespace cuda
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/cuda/detail/detail/b40c/radixsort_reduction_kernel.h b/compat/thrust/system/cuda/detail/detail/b40c/radixsort_reduction_kernel.h
new file mode 100644
index 0000000000..a8f91d3d24
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/detail/b40c/radixsort_reduction_kernel.h
@@ -0,0 +1,439 @@
+/******************************************************************************
+ * 
+ * Copyright 2010 Duane Merrill
+ * 
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ * 
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License. 
+ * 
+ * 
+ * 
+ * 
+ * AUTHORS' REQUEST: 
+ * 
+ * 		If you use|reference|benchmark this code, please cite our Technical 
+ * 		Report (http://www.cs.virginia.edu/~dgm4d/papers/RadixSortTR.pdf):
+ * 
+ *		@TechReport{ Merrill:Sorting:2010,
+ *        	author = "Duane Merrill and Andrew Grimshaw",
+ *        	title = "Revisiting Sorting for GPGPU Stream Architectures",
+ *        	year = "2010",
+ *        	institution = "University of Virginia, Department of Computer Science",
+ *        	address = "Charlottesville, VA, USA",
+ *        	number = "CS2010-03"
+ *		}
+ * 
+ * For more information, see our Google Code project site: 
+ * http://code.google.com/p/back40computing/
+ * 
+ * Thanks!
+ * 
+ ******************************************************************************/
+
+
+/******************************************************************************
+ * Bottom-level digit-reduction/counting kernel
+ ******************************************************************************/
+
+#pragma once
+
+#include "radixsort_kernel_common.h"
+
+namespace thrust  {
+namespace system  {
+namespace cuda    {
+namespace detail  {
+namespace detail  {
+namespace b40c_thrust   {
+
+/******************************************************************************
+ * Defines
+ ******************************************************************************/
+
+const int BYTE_ENCODE_SHIFT = 0x3;
+
+
+/******************************************************************************
+ * Cycle-processing Routines
+ ******************************************************************************/
+
+__device__ __forceinline__ int DecodeInt(int encoded, int quad_byte){
+	return (encoded >> quad_byte) & 0xff;		// shift right 8 bits per digit and return rightmost 8 bits
+}
+
+
+__device__ __forceinline__ int EncodeInt(int count, int quad_byte) {
+	return count << quad_byte;					// shift left 8 bits per digit
+}
+
+
+template <typename K, long long RADIX_DIGITS, int BIT>
+__device__ __forceinline__ void DecodeDigit(
+	K key, 
+	int &lane, 
+	int &quad_shift) 
+{
+	const K DIGIT_MASK = RADIX_DIGITS - 1;
+	lane = (key & (DIGIT_MASK << BIT)) >> (BIT + 2);
+	
+	const K QUAD_MASK = (RADIX_DIGITS < 4) ? 0x1 : 0x3;
+	if (BIT == 32) {
+		// N.B.: This takes one more instruction than the code below it, but 
+		// otherwise the compiler goes nuts and shoves hundreds of bytes 
+		// to lmem when bit = 32 on 64-bit keys.		
+		quad_shift = ((key >> BIT) & QUAD_MASK) << BYTE_ENCODE_SHIFT;	
+	} else {
+		quad_shift = MagnitudeShift<K, BYTE_ENCODE_SHIFT - BIT>(key & (QUAD_MASK << BIT));
+	}
+}
+
+
+template <int RADIX_DIGITS, int SCAN_LANES, int LANES_PER_WARP, int BIT, bool FINAL_REDUCE>
+__device__ __forceinline__ void ReduceEncodedCounts(
+	int local_counts[LANES_PER_WARP][4],
+	int encoded_carry[SCAN_LANES][B40C_RADIXSORT_THREADS]) 
+{
+	const int LOG_PARTIALS_PER_THREAD = B40C_RADIXSORT_LOG_THREADS - B40C_LOG_WARP_THREADS;
+	const int PARTIALS_PER_THREAD = 1 << LOG_PARTIALS_PER_THREAD;
+	
+	int encoded;
+	int idx = threadIdx.x & (B40C_WARP_THREADS - 1);
+	
+	
+	__syncthreads();
+
+	#pragma unroll
+	for (int j = 0; j < (int) LANES_PER_WARP; j++) {
+		
+		int warp_id = (threadIdx.x >> B40C_LOG_WARP_THREADS) + (j * B40C_RADIXSORT_WARPS);
+		if (warp_id < SCAN_LANES) {
+
+			// rest of my elements
+			#pragma unroll
+			for (int i = 0; i < (int) PARTIALS_PER_THREAD; i++) {
+				encoded = encoded_carry[warp_id][idx + (i * B40C_WARP_THREADS)];		
+				local_counts[j][0] += DecodeInt(encoded, 0u << BYTE_ENCODE_SHIFT);
+				local_counts[j][1] += DecodeInt(encoded, 1u << BYTE_ENCODE_SHIFT);
+				local_counts[j][2] += DecodeInt(encoded, 2u << BYTE_ENCODE_SHIFT);
+				local_counts[j][3] += DecodeInt(encoded, 3u << BYTE_ENCODE_SHIFT);
+			}
+			
+			if (FINAL_REDUCE) {
+				// reduce all four packed fields, leaving them in the first four elements of our row
+				WarpReduce<B40C_WARP_THREADS>(idx, &encoded_carry[warp_id][0], local_counts[j][0]);
+				WarpReduce<B40C_WARP_THREADS>(idx, &encoded_carry[warp_id][1], local_counts[j][1]);
+				WarpReduce<B40C_WARP_THREADS>(idx, &encoded_carry[warp_id][2], local_counts[j][2]);
+				WarpReduce<B40C_WARP_THREADS>(idx, &encoded_carry[warp_id][3], local_counts[j][3]);
+			}
+		}
+	}	
+
+	__syncthreads();
+	
+}
+	
+
+template <typename K, int RADIX_DIGITS, int SCAN_LANES, int BIT, typename PreprocessFunctor>
+__device__ __forceinline__ void Bucket(
+	K input, 
+	int encoded_carry[SCAN_LANES][B40C_RADIXSORT_THREADS],
+	PreprocessFunctor preprocess = PreprocessFunctor()) 
+{
+	int lane, quad_shift;
+	preprocess(input);
+	DecodeDigit<K, RADIX_DIGITS, BIT>(input, lane, quad_shift);
+	encoded_carry[lane][threadIdx.x] += EncodeInt(1, quad_shift);
+}
+
+
+template <typename K, int RADIX_DIGITS, int SCAN_LANES, int BIT, typename PreprocessFunctor, int CYCLES>
+struct LoadOp;
+
+template <typename K, int RADIX_DIGITS, int SCAN_LANES, int BIT, typename PreprocessFunctor>
+struct LoadOp<K, RADIX_DIGITS, SCAN_LANES, BIT, PreprocessFunctor, 1> 
+{
+	static __device__ __forceinline__  void BlockOfLoads(K *d_in_keys, int offset, int encoded_carry[SCAN_LANES][B40C_RADIXSORT_THREADS])
+	{
+		K key = d_in_keys[offset + threadIdx.x];
+		Bucket<K, RADIX_DIGITS, SCAN_LANES, BIT, PreprocessFunctor>(key, encoded_carry);
+	}
+};
+
+template <typename K, int RADIX_DIGITS, int SCAN_LANES, int BIT, typename PreprocessFunctor>
+struct LoadOp<K, RADIX_DIGITS, SCAN_LANES, BIT, PreprocessFunctor, 2> 
+{
+	static __device__ __forceinline__  void BlockOfLoads(K *d_in_keys, int offset, int encoded_carry[SCAN_LANES][B40C_RADIXSORT_THREADS])
+	{
+		LoadOp<K, RADIX_DIGITS, SCAN_LANES, BIT, PreprocessFunctor, 1>::BlockOfLoads(d_in_keys, offset + (B40C_RADIXSORT_THREADS * 0), encoded_carry);
+		LoadOp<K, RADIX_DIGITS, SCAN_LANES, BIT, PreprocessFunctor, 1>::BlockOfLoads(d_in_keys, offset + (B40C_RADIXSORT_THREADS * 1), encoded_carry);
+	}
+};
+
+template <typename K, int RADIX_DIGITS, int SCAN_LANES, int BIT, typename PreprocessFunctor>
+struct LoadOp<K, RADIX_DIGITS, SCAN_LANES, BIT, PreprocessFunctor, 4> 
+{
+	static __device__ __forceinline__  void BlockOfLoads(K *d_in_keys, int offset, int encoded_carry[SCAN_LANES][B40C_RADIXSORT_THREADS])
+	{
+		LoadOp<K, RADIX_DIGITS, SCAN_LANES, BIT, PreprocessFunctor, 2>::BlockOfLoads(d_in_keys, offset + (B40C_RADIXSORT_THREADS * 0), encoded_carry);
+		LoadOp<K, RADIX_DIGITS, SCAN_LANES, BIT, PreprocessFunctor, 2>::BlockOfLoads(d_in_keys, offset + (B40C_RADIXSORT_THREADS * 2), encoded_carry);
+	}
+};
+
+template <typename K, int RADIX_DIGITS, int SCAN_LANES, int BIT, typename PreprocessFunctor>
+struct LoadOp<K, RADIX_DIGITS, SCAN_LANES, BIT, PreprocessFunctor, 8> 
+{
+	static __device__ __forceinline__  void BlockOfLoads(K *d_in_keys, int offset, int encoded_carry[SCAN_LANES][B40C_RADIXSORT_THREADS])
+	{
+			K keys[8];
+				
+			keys[0] = d_in_keys[offset + (B40C_RADIXSORT_THREADS * 0) + threadIdx.x];
+			keys[1] = d_in_keys[offset + (B40C_RADIXSORT_THREADS * 1) + threadIdx.x];
+			keys[2] = d_in_keys[offset + (B40C_RADIXSORT_THREADS * 2) + threadIdx.x];
+			keys[3] = d_in_keys[offset + (B40C_RADIXSORT_THREADS * 3) + threadIdx.x];
+
+			if (B40C_FERMI(__CUDA_ARCH__)) __syncthreads();
+			
+			keys[4] = d_in_keys[offset + (B40C_RADIXSORT_THREADS * 4) + threadIdx.x];
+			keys[5] = d_in_keys[offset + (B40C_RADIXSORT_THREADS * 5) + threadIdx.x];
+			keys[6] = d_in_keys[offset + (B40C_RADIXSORT_THREADS * 6) + threadIdx.x];
+			keys[7] = d_in_keys[offset + (B40C_RADIXSORT_THREADS * 7) + threadIdx.x];
+			
+			Bucket<K, RADIX_DIGITS, SCAN_LANES, BIT, PreprocessFunctor>(keys[0], encoded_carry);
+			Bucket<K, RADIX_DIGITS, SCAN_LANES, BIT, PreprocessFunctor>(keys[1], encoded_carry);
+			Bucket<K, RADIX_DIGITS, SCAN_LANES, BIT, PreprocessFunctor>(keys[2], encoded_carry);
+			Bucket<K, RADIX_DIGITS, SCAN_LANES, BIT, PreprocessFunctor>(keys[3], encoded_carry);
+			Bucket<K, RADIX_DIGITS, SCAN_LANES, BIT, PreprocessFunctor>(keys[4], encoded_carry);
+			Bucket<K, RADIX_DIGITS, SCAN_LANES, BIT, PreprocessFunctor>(keys[5], encoded_carry);
+			Bucket<K, RADIX_DIGITS, SCAN_LANES, BIT, PreprocessFunctor>(keys[6], encoded_carry);
+			Bucket<K, RADIX_DIGITS, SCAN_LANES, BIT, PreprocessFunctor>(keys[7], encoded_carry);
+	}
+};
+
+template <typename K, int RADIX_DIGITS, int SCAN_LANES, int BIT, typename PreprocessFunctor>
+struct LoadOp<K, RADIX_DIGITS, SCAN_LANES, BIT, PreprocessFunctor, 16> {
+
+	static __device__ __forceinline__  void BlockOfLoads(K *d_in_keys, int offset, int encoded_carry[SCAN_LANES][B40C_RADIXSORT_THREADS])
+	{
+		LoadOp<K, RADIX_DIGITS, SCAN_LANES, BIT, PreprocessFunctor, 8>::BlockOfLoads(d_in_keys, offset + (B40C_RADIXSORT_THREADS * 0), encoded_carry);
+		LoadOp<K, RADIX_DIGITS, SCAN_LANES, BIT, PreprocessFunctor, 8>::BlockOfLoads(d_in_keys, offset + (B40C_RADIXSORT_THREADS * 8), encoded_carry);
+	}
+};
+
+template <typename K, int RADIX_DIGITS, int SCAN_LANES, int BIT, typename PreprocessFunctor>
+struct LoadOp<K, RADIX_DIGITS, SCAN_LANES, BIT, PreprocessFunctor, 32> {
+
+	static __device__ __forceinline__  void BlockOfLoads(K *d_in_keys, int offset, int encoded_carry[SCAN_LANES][B40C_RADIXSORT_THREADS])
+	{
+		LoadOp<K, RADIX_DIGITS, SCAN_LANES, BIT, PreprocessFunctor, 16>::BlockOfLoads(d_in_keys, offset + (B40C_RADIXSORT_THREADS * 0), encoded_carry);
+		LoadOp<K, RADIX_DIGITS, SCAN_LANES, BIT, PreprocessFunctor, 16>::BlockOfLoads(d_in_keys, offset + (B40C_RADIXSORT_THREADS * 16), encoded_carry);
+	}
+};
+
+template <typename K, int RADIX_DIGITS, int SCAN_LANES, int BIT, typename PreprocessFunctor>
+struct LoadOp<K, RADIX_DIGITS, SCAN_LANES, BIT, PreprocessFunctor, 64> {
+
+	static __device__ __forceinline__  void BlockOfLoads(K *d_in_keys, int offset, int encoded_carry[SCAN_LANES][B40C_RADIXSORT_THREADS])
+	{
+		LoadOp<K, RADIX_DIGITS, SCAN_LANES, BIT, PreprocessFunctor, 32>::BlockOfLoads(d_in_keys, offset + (B40C_RADIXSORT_THREADS * 0), encoded_carry);
+		LoadOp<K, RADIX_DIGITS, SCAN_LANES, BIT, PreprocessFunctor, 32>::BlockOfLoads(d_in_keys, offset + (B40C_RADIXSORT_THREADS * 32), encoded_carry);
+	}
+};
+
+template <typename K, int RADIX_DIGITS, int SCAN_LANES, int BIT, typename PreprocessFunctor>
+struct LoadOp<K, RADIX_DIGITS, SCAN_LANES, BIT, PreprocessFunctor, 128> {
+
+	static __device__ __forceinline__  void BlockOfLoads(K *d_in_keys, int offset, int encoded_carry[SCAN_LANES][B40C_RADIXSORT_THREADS])
+	{
+		LoadOp<K, RADIX_DIGITS, SCAN_LANES, BIT, PreprocessFunctor, 64>::BlockOfLoads(d_in_keys, offset + (B40C_RADIXSORT_THREADS * 0), encoded_carry);
+		LoadOp<K, RADIX_DIGITS, SCAN_LANES, BIT, PreprocessFunctor, 64>::BlockOfLoads(d_in_keys, offset + (B40C_RADIXSORT_THREADS * 64), encoded_carry);
+	}
+};
+
+template <typename K, int RADIX_DIGITS, int SCAN_LANES, int BIT, typename PreprocessFunctor>
+struct LoadOp<K, RADIX_DIGITS, SCAN_LANES, BIT, PreprocessFunctor, 252> {
+
+	static __device__ __forceinline__  void BlockOfLoads(K *d_in_keys, int offset, int encoded_carry[SCAN_LANES][B40C_RADIXSORT_THREADS])
+	{
+		LoadOp<K, RADIX_DIGITS, SCAN_LANES, BIT, PreprocessFunctor, 128>::BlockOfLoads(d_in_keys, offset + (B40C_RADIXSORT_THREADS * 0), encoded_carry);
+		LoadOp<K, RADIX_DIGITS, SCAN_LANES, BIT, PreprocessFunctor, 64>::BlockOfLoads(d_in_keys, offset + (B40C_RADIXSORT_THREADS * 128), encoded_carry);
+		LoadOp<K, RADIX_DIGITS, SCAN_LANES, BIT, PreprocessFunctor, 32>::BlockOfLoads(d_in_keys, offset + (B40C_RADIXSORT_THREADS * 192), encoded_carry);
+		LoadOp<K, RADIX_DIGITS, SCAN_LANES, BIT, PreprocessFunctor, 16>::BlockOfLoads(d_in_keys, offset + (B40C_RADIXSORT_THREADS * 224), encoded_carry);
+		LoadOp<K, RADIX_DIGITS, SCAN_LANES, BIT, PreprocessFunctor, 8>::BlockOfLoads(d_in_keys, offset + (B40C_RADIXSORT_THREADS * 240), encoded_carry);
+		LoadOp<K, RADIX_DIGITS, SCAN_LANES, BIT, PreprocessFunctor, 4>::BlockOfLoads(d_in_keys, offset + (B40C_RADIXSORT_THREADS * 248), encoded_carry);
+	}
+};
+
+
+template <int SCAN_LANES>
+__device__ __forceinline__ void ResetEncodedCarry(
+	int encoded_carry[SCAN_LANES][B40C_RADIXSORT_THREADS])
+{
+	#pragma unroll
+	for (int SCAN_LANE = 0; SCAN_LANE < (int) SCAN_LANES; SCAN_LANE++) {
+		encoded_carry[SCAN_LANE][threadIdx.x] = 0;
+	}
+}
+
+
+template <typename K, int RADIX_DIGITS, int SCAN_LANES, int LANES_PER_WARP, int BIT, typename PreprocessFunctor>
+__device__ __forceinline__ int ProcessLoads(
+	K *d_in_keys,
+	int loads,
+	int &offset,
+	int encoded_carry[SCAN_LANES][B40C_RADIXSORT_THREADS],
+	int local_counts[LANES_PER_WARP][4])
+{
+	// Unroll batches of loads with occasional reduction to avoid overflow
+	while (loads >= 32) {
+	
+		LoadOp<K, RADIX_DIGITS, SCAN_LANES, BIT, PreprocessFunctor, 32>::BlockOfLoads(d_in_keys, offset, encoded_carry);
+		offset += B40C_RADIXSORT_THREADS * 32;
+		loads -= 32;
+
+		// Reduce int local count registers to prevent overflow
+		ReduceEncodedCounts<RADIX_DIGITS, SCAN_LANES, LANES_PER_WARP, BIT, false>(
+				local_counts, 
+				encoded_carry);
+		
+		// Reset encoded counters
+		ResetEncodedCarry<SCAN_LANES>(encoded_carry);
+	} 
+	
+	int retval = loads;
+	
+	// Wind down loads in decreasing batch sizes
+
+	while (loads >= 4) {
+		LoadOp<K, RADIX_DIGITS, SCAN_LANES, BIT, PreprocessFunctor, 4>::BlockOfLoads(d_in_keys, offset, encoded_carry);
+		offset += B40C_RADIXSORT_THREADS * 4;
+		loads -= 4;
+	} 
+
+	while (loads) {
+		LoadOp<K, RADIX_DIGITS, SCAN_LANES, BIT, PreprocessFunctor, 1>::BlockOfLoads(d_in_keys, offset, encoded_carry);
+		offset += B40C_RADIXSORT_THREADS * 1;
+		loads--;
+	}
+	
+	return retval;
+}
+
+
+/******************************************************************************
+ * Reduction/counting Kernel Entry Point
+ ******************************************************************************/
+
+template <typename K, typename V, int PASS, int RADIX_BITS, int BIT, typename PreprocessFunctor>
+__launch_bounds__ (B40C_RADIXSORT_THREADS, B40C_RADIXSORT_REDUCE_CTA_OCCUPANCY(__CUDA_ARCH__))
+__global__ 
+void RakingReduction(
+	bool *d_from_alt_storage,
+	int *d_spine,
+	K *d_in_keys,
+	K *d_out_keys,
+	CtaDecomposition work_decomposition)
+{
+	const int RADIX_DIGITS 		= 1 << RADIX_BITS;
+
+	const int LOG_SCAN_LANES 		= (RADIX_BITS >= 2) ? RADIX_BITS - 2 : 0;	// Always at least one fours group
+	const int SCAN_LANES 			= 1 << LOG_SCAN_LANES;
+
+	const int LOG_LANES_PER_WARP 	= (SCAN_LANES > B40C_RADIXSORT_WARPS) ? LOG_SCAN_LANES - B40C_RADIXSORT_LOG_WARPS : 0;	// Always at least one fours group per warp
+	const int LANES_PER_WARP 		= 1 << LOG_LANES_PER_WARP;
+	
+	
+	// Each thread gets its own column of fours-groups (for conflict-free updates)
+	__shared__ int encoded_carry[SCAN_LANES][B40C_RADIXSORT_THREADS];			
+
+	// Each thread is also responsible for aggregating an unencoded segment of a fours-group
+	int local_counts[LANES_PER_WARP][4];								
+
+	// Determine where to read our input
+	bool from_alt_storage = (PASS == 0) ? false : d_from_alt_storage[PASS & 0x1];
+	if (from_alt_storage) d_in_keys = d_out_keys;
+	
+	// Calculate our threadblock's range
+	int offset, block_elements;
+	if (blockIdx.x < work_decomposition.num_big_blocks) {
+		offset = work_decomposition.big_block_elements * blockIdx.x;
+		block_elements = work_decomposition.big_block_elements;
+	} else {
+		offset = (work_decomposition.normal_block_elements * blockIdx.x) + (work_decomposition.num_big_blocks * B40C_RADIXSORT_CYCLE_ELEMENTS(__CUDA_ARCH__, K, V));
+		block_elements = work_decomposition.normal_block_elements;
+	}
+	
+	// Initialize local counts
+	#pragma unroll 
+	for (int LANE = 0; LANE < (int) LANES_PER_WARP; LANE++) {
+		local_counts[LANE][0] = 0;
+		local_counts[LANE][1] = 0;
+		local_counts[LANE][2] = 0;
+		local_counts[LANE][3] = 0;
+	}
+	
+	// Reset encoded counters
+	ResetEncodedCarry<SCAN_LANES>(encoded_carry);
+	
+	// Process loads
+	int loads = block_elements >> B40C_RADIXSORT_LOG_THREADS;
+	int unreduced_loads = ProcessLoads<K, RADIX_DIGITS, SCAN_LANES, LANES_PER_WARP, BIT, PreprocessFunctor>(
+		d_in_keys,
+		loads,
+		offset,
+		encoded_carry,
+		local_counts);
+	
+	// Cleanup if we're the last block  
+	if ((blockIdx.x == gridDim.x - 1) && (work_decomposition.extra_elements_last_block)) {
+
+		const int LOADS_PER_CYCLE = B40C_RADIXSORT_CYCLE_ELEMENTS(__CUDA_ARCH__, K, V) / B40C_RADIXSORT_THREADS;
+		
+		// If extra guarded loads may cause overflow, reduce now and reset counters
+		if (unreduced_loads + LOADS_PER_CYCLE > 255) {
+		
+			ReduceEncodedCounts<RADIX_DIGITS, SCAN_LANES, LANES_PER_WARP, BIT, false>(
+					local_counts, 
+					encoded_carry);
+			
+			ResetEncodedCarry<SCAN_LANES>(encoded_carry);
+		}
+		
+		// perform up to LOADS_PER_CYCLE extra guarded loads
+		#pragma unroll
+		for (int EXTRA_LOAD = 0; EXTRA_LOAD < (int) LOADS_PER_CYCLE; EXTRA_LOAD++) {
+			if (threadIdx.x + (B40C_RADIXSORT_THREADS * EXTRA_LOAD) < work_decomposition.extra_elements_last_block) {
+				K key = d_in_keys[offset + (B40C_RADIXSORT_THREADS * EXTRA_LOAD) + threadIdx.x];
+				Bucket<K, RADIX_DIGITS, SCAN_LANES, BIT, PreprocessFunctor>(key, encoded_carry);
+			}
+		}
+	}
+	
+	// Aggregate 
+	ReduceEncodedCounts<RADIX_DIGITS, SCAN_LANES, LANES_PER_WARP, BIT, true>(
+		local_counts, 
+		encoded_carry);
+
+	// Write carry in parallel (carries per row are in the first four bytes of each row) 
+	if (threadIdx.x < RADIX_DIGITS) {
+
+		int row = threadIdx.x >> 2;		
+		int col = threadIdx.x & 3;			 
+		d_spine[(gridDim.x * threadIdx.x) + blockIdx.x] = encoded_carry[row][col];
+	}
+} 
+
+} // end namespace b40c_thrust
+} // end namespace detail
+} // end namespace detail
+} // end namespace cuda
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/cuda/detail/detail/b40c/radixsort_scanscatter_kernel.h b/compat/thrust/system/cuda/detail/detail/b40c/radixsort_scanscatter_kernel.h
new file mode 100644
index 0000000000..1377999c76
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/detail/b40c/radixsort_scanscatter_kernel.h
@@ -0,0 +1,1207 @@
+/******************************************************************************
+ * 
+ * Copyright 2010 Duane Merrill
+ * 
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ * 
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License. 
+ * 
+ * 
+ * 
+ * 
+ * AUTHORS' REQUEST: 
+ * 
+ * 		If you use|reference|benchmark this code, please cite our Technical 
+ * 		Report (http://www.cs.virginia.edu/~dgm4d/papers/RadixSortTR.pdf):
+ * 
+ *		@TechReport{ Merrill:Sorting:2010,
+ *        	author = "Duane Merrill and Andrew Grimshaw",
+ *        	title = "Revisiting Sorting for GPGPU Stream Architectures",
+ *        	year = "2010",
+ *        	institution = "University of Virginia, Department of Computer Science",
+ *        	address = "Charlottesville, VA, USA",
+ *        	number = "CS2010-03"
+ *		}
+ * 
+ * For more information, see our Google Code project site: 
+ * http://code.google.com/p/back40computing/
+ * 
+ * Thanks!
+ * 
+ ******************************************************************************/
+
+
+/******************************************************************************
+// Bottom-level digit scanning/scattering kernel
+ ******************************************************************************/
+
+#pragma once
+
+#include "radixsort_kernel_common.h"
+
+namespace thrust  {
+namespace system  {
+namespace cuda    {
+namespace detail  {
+namespace detail  {
+namespace b40c_thrust   {
+
+/******************************************************************************
+ * Appropriate substitutes to use for out-of-bounds key (and value) offsets 
+ ******************************************************************************/
+
+template <typename T> 
+__device__ __forceinline__ T DefaultextraValue() {
+	return T();
+}
+
+template <> 
+__device__ __forceinline__ unsigned char DefaultextraValue<unsigned char>() {
+	return (unsigned char) -1;
+}
+
+template <> 
+__device__ __forceinline__ unsigned short DefaultextraValue<unsigned short>() {
+	return (unsigned short) -1;
+}
+
+template <> 
+__device__ __forceinline__ unsigned int DefaultextraValue<unsigned int>() {
+	return (unsigned int) -1u;
+}
+
+template <> 
+__device__ __forceinline__ unsigned long DefaultextraValue<unsigned long>() {
+	return (unsigned long) -1ul;
+}
+
+template <> 
+__device__ __forceinline__ unsigned long long DefaultextraValue<unsigned long long>() {
+	return (unsigned long long) -1ull;
+}
+
+
+/******************************************************************************
+ * Cycle-processing Routines
+ ******************************************************************************/
+
+template <typename K, long long RADIX_DIGITS, int BIT>
+__device__ __forceinline__ int DecodeDigit(K key) 
+{
+	const K DIGIT_MASK = RADIX_DIGITS - 1;
+	return (key >> BIT) & DIGIT_MASK;
+}
+
+
+template <typename K, long long RADIX_DIGITS, int BIT, int PADDED_PARTIALS_PER_LANE>
+__device__ __forceinline__ void DecodeDigit(
+	K key, 
+	int &digit, 
+	int &flag_offset,		// in bytes
+	const int SET_OFFSET)
+{
+	const int PADDED_BYTES_PER_LANE 	= PADDED_PARTIALS_PER_LANE * 4;
+	const int SET_OFFSET_BYTES 		= SET_OFFSET * 4;
+	const K QUAD_MASK 							= (RADIX_DIGITS < 4) ? 0x1 : 0x3;
+	
+	digit = DecodeDigit<K, RADIX_DIGITS, BIT>(key);
+	int lane = digit >> 2;
+	int quad_byte = digit & QUAD_MASK;
+
+	flag_offset = SET_OFFSET_BYTES + FastMul(lane, PADDED_BYTES_PER_LANE) + quad_byte;
+}
+
+
+template <typename K, long long RADIX_DIGITS, int BIT, int SETS_PER_PASS, int SCAN_LANES_PER_SET, int PADDED_PARTIALS_PER_LANE>
+__device__ __forceinline__ void DecodeDigits(
+	typename VecType<K, 2>::Type keypairs[SETS_PER_PASS],
+	int2 digits[SETS_PER_PASS],
+	int2 flag_offsets[SETS_PER_PASS])		// in bytes 
+{
+
+	#pragma unroll
+	for (int SET = 0; SET < (int) SETS_PER_PASS; SET++) {
+		
+		const int SET_OFFSET = SET * SCAN_LANES_PER_SET * PADDED_PARTIALS_PER_LANE;
+
+		DecodeDigit<K, RADIX_DIGITS, BIT, PADDED_PARTIALS_PER_LANE>(
+				keypairs[SET].x, digits[SET].x, flag_offsets[SET].x, SET_OFFSET);
+		
+		DecodeDigit<K, RADIX_DIGITS, BIT, PADDED_PARTIALS_PER_LANE>(
+				keypairs[SET].y, digits[SET].y, flag_offsets[SET].y, SET_OFFSET);
+	}
+}
+
+
+template <typename T, typename PreprocessFunctor>
+__device__ __forceinline__ void GuardedReadSet(
+	T *in, 
+	typename VecType<T, 2>::Type &pair,
+	int offset,
+	int extra[1],
+	PreprocessFunctor preprocess = PreprocessFunctor())				
+{
+	if (offset - extra[0] < 0) {
+		pair.x = in[offset];
+		preprocess(pair.x);
+	} else {
+		pair.x = DefaultextraValue<T>();
+	}
+	
+	if (offset + 1 - extra[0] < 0) {
+		pair.y = in[offset + 1];
+		preprocess(pair.y);
+	} else {
+		pair.y = DefaultextraValue<T>();
+	}
+}
+
+
+template <typename T, bool UNGUARDED_IO, int SETS_PER_PASS, typename PreprocessFunctor>
+__device__ __forceinline__ void ReadSets(
+	typename VecType<T, 2>::Type *d_in, 
+	typename VecType<T, 2>::Type pairs[SETS_PER_PASS],
+	const int BASE2,
+	int extra[1],
+	PreprocessFunctor preprocess = PreprocessFunctor())				
+{
+	if (UNGUARDED_IO) {
+
+		// N.B. -- I wish we could do some pragma unrolling here too, but the compiler makes it 1% slower
+		if (SETS_PER_PASS > 0) pairs[0] = d_in[threadIdx.x + BASE2 + (B40C_RADIXSORT_THREADS * 0)];
+		if (SETS_PER_PASS > 1) pairs[1] = d_in[threadIdx.x + BASE2 + (B40C_RADIXSORT_THREADS * 1)];
+		if (SETS_PER_PASS > 2) pairs[2] = d_in[threadIdx.x + BASE2 + (B40C_RADIXSORT_THREADS * 2)];
+		if (SETS_PER_PASS > 3) pairs[3] = d_in[threadIdx.x + BASE2 + (B40C_RADIXSORT_THREADS * 3)];
+
+		#pragma unroll 
+		for (int SET = 0; SET < (int) SETS_PER_PASS; SET++) {
+			preprocess(pairs[SET].x);
+			preprocess(pairs[SET].y);
+		}
+		
+	} else {
+
+		T* in = (T*) d_in;
+		
+		// N.B. --  I wish we could do some pragma unrolling here, but the compiler won't let 
+		// us with user-defined value types (e.g., Fribbitz): "Advisory: Loop was not unrolled, cannot deduce loop trip count"
+		
+		if (SETS_PER_PASS > 0) GuardedReadSet<T, PreprocessFunctor>(in, pairs[0], (threadIdx.x << 1) + (BASE2 << 1) + (B40C_RADIXSORT_THREADS * 2 * 0), extra);
+		if (SETS_PER_PASS > 1) GuardedReadSet<T, PreprocessFunctor>(in, pairs[1], (threadIdx.x << 1) + (BASE2 << 1) + (B40C_RADIXSORT_THREADS * 2 * 1), extra);
+		if (SETS_PER_PASS > 2) GuardedReadSet<T, PreprocessFunctor>(in, pairs[2], (threadIdx.x << 1) + (BASE2 << 1) + (B40C_RADIXSORT_THREADS * 2 * 2), extra);
+		if (SETS_PER_PASS > 3) GuardedReadSet<T, PreprocessFunctor>(in, pairs[3], (threadIdx.x << 1) + (BASE2 << 1) + (B40C_RADIXSORT_THREADS * 2 * 3), extra);
+	}
+}
+
+
+template <int SETS_PER_PASS>
+__device__ __forceinline__ void PlacePartials(
+	unsigned char * base_partial,
+	int2 digits[SETS_PER_PASS],
+	int2 flag_offsets[SETS_PER_PASS]) 
+{
+	#pragma unroll
+	for (int SET = 0; SET < (int) SETS_PER_PASS; SET++) {
+		base_partial[flag_offsets[SET].x] = 1;
+		base_partial[flag_offsets[SET].y] = 1 + (digits[SET].x == digits[SET].y);
+	}
+}
+
+
+template <int SETS_PER_PASS>
+__device__ __forceinline__ void ExtractRanks(
+	unsigned char * base_partial,
+	int2 digits[SETS_PER_PASS],
+	int2 flag_offsets[SETS_PER_PASS],
+	int2 ranks[SETS_PER_PASS]) 
+{
+	#pragma unroll
+	for (int SET = 0; SET < (int) SETS_PER_PASS; SET++) {
+		ranks[SET].x = base_partial[flag_offsets[SET].x];
+		ranks[SET].y = base_partial[flag_offsets[SET].y] + (digits[SET].x == digits[SET].y);
+	}
+}
+
+
+template <int RADIX_DIGITS, int SETS_PER_PASS>
+__device__ __forceinline__ void UpdateRanks(
+	int2 digits[SETS_PER_PASS],
+	int2 ranks[SETS_PER_PASS],
+	int digit_counts[SETS_PER_PASS][RADIX_DIGITS])
+{
+	// N.B.: I wish we could pragma unroll here, but doing so currently 
+	// results in the 3.1 compilier on 64-bit platforms generating bad
+	// code for SM1.3, resulting in incorrect sorting (e.g., problem size 16)
+	
+	if (SETS_PER_PASS > 0) {
+		ranks[0].x += digit_counts[0][digits[0].x];
+		ranks[0].y += digit_counts[0][digits[0].y]; 
+	}	
+	if (SETS_PER_PASS > 1) {
+		ranks[1].x += digit_counts[1][digits[1].x];
+		ranks[1].y += digit_counts[1][digits[1].y]; 
+	}	
+	if (SETS_PER_PASS > 2) {
+		ranks[2].x += digit_counts[2][digits[2].x];
+		ranks[2].y += digit_counts[2][digits[2].y]; 
+	}	
+	if (SETS_PER_PASS > 3) {
+		ranks[3].x += digit_counts[3][digits[3].x];
+		ranks[3].y += digit_counts[3][digits[3].y]; 
+	}	
+}
+
+template <int RADIX_DIGITS, int PASSES_PER_CYCLE, int SETS_PER_PASS>
+__device__ __forceinline__ void UpdateRanks(
+	int2 digits[PASSES_PER_CYCLE][SETS_PER_PASS],
+	int2 ranks[PASSES_PER_CYCLE][SETS_PER_PASS],
+	int digit_counts[PASSES_PER_CYCLE][SETS_PER_PASS][RADIX_DIGITS])
+{
+	// N.B.: I wish we could pragma unroll here, but doing so currently 
+	// results in the 3.1 compilier on 64-bit platforms generating bad
+	// code for SM1.3, resulting in incorrect sorting (e.g., problem size 16)
+	
+	if (PASSES_PER_CYCLE > 0) UpdateRanks<RADIX_DIGITS, SETS_PER_PASS>(digits[0], ranks[0], digit_counts[0]);
+	if (PASSES_PER_CYCLE > 1) UpdateRanks<RADIX_DIGITS, SETS_PER_PASS>(digits[1], ranks[1], digit_counts[1]);
+	if (PASSES_PER_CYCLE > 2) UpdateRanks<RADIX_DIGITS, SETS_PER_PASS>(digits[2], ranks[2], digit_counts[2]);
+	if (PASSES_PER_CYCLE > 3) UpdateRanks<RADIX_DIGITS, SETS_PER_PASS>(digits[3], ranks[3], digit_counts[3]);
+}
+
+
+
+template <int SCAN_LANES_PER_PASS, int LOG_RAKING_THREADS_PER_LANE, int RAKING_THREADS_PER_LANE, int PARTIALS_PER_SEG>
+__device__ __forceinline__ void PrefixScanOverLanes(
+	int 	raking_segment[],
+	int 	warpscan[SCAN_LANES_PER_PASS][3][RAKING_THREADS_PER_LANE],
+	int 	copy_section)
+{
+	// Upsweep rake
+	int partial_reduction = SerialReduce<PARTIALS_PER_SEG>(raking_segment);
+
+	// Warpscan reduction in digit warpscan_lane
+	int warpscan_lane = threadIdx.x >> LOG_RAKING_THREADS_PER_LANE;
+	int group_prefix = WarpScan<RAKING_THREADS_PER_LANE, true>(
+		warpscan[warpscan_lane], 
+		partial_reduction,
+		copy_section);
+
+	// Downsweep rake
+	SerialScan<PARTIALS_PER_SEG>(raking_segment, group_prefix);
+	
+}
+
+
+template <int SCAN_LANES_PER_PASS, int RAKING_THREADS_PER_LANE, int SETS_PER_PASS, int SCAN_LANES_PER_SET>
+__device__ __forceinline__ void RecoverDigitCounts(
+	int warpscan[SCAN_LANES_PER_PASS][3][RAKING_THREADS_PER_LANE],
+	int counts[SETS_PER_PASS],
+	int copy_section)
+{
+	int my_lane = threadIdx.x >> 2;
+	int my_quad_byte = threadIdx.x & 3;
+	
+	#pragma unroll
+	for (int SET = 0; SET < (int) SETS_PER_PASS; SET++) {
+		unsigned char *warpscan_count = (unsigned char *) &warpscan[my_lane + (SCAN_LANES_PER_SET * SET)][1 + copy_section][RAKING_THREADS_PER_LANE - 1];
+		counts[SET] = warpscan_count[my_quad_byte];
+	}
+}
+
+template<int RADIX_DIGITS>
+__device__ __forceinline__ void CorrectUnguardedSetOverflow(
+	int2 			set_digits,
+	int 	&set_count)				
+{
+	if (WarpVoteAll(RADIX_DIGITS, set_count <= 1)) {
+		// All first-pass, first set keys have same digit. 
+		set_count = (threadIdx.x == set_digits.x) ? 256 : 0;
+	}
+}
+
+template <int RADIX_DIGITS, int SETS_PER_PASS>
+__device__ __forceinline__ void CorrectUnguardedPassOverflow(
+	int2 			pass_digits[SETS_PER_PASS],
+	int 	pass_counts[SETS_PER_PASS])				
+{
+	// N.B. -- I wish we could do some pragma unrolling here too, but the compiler won't comply, 
+	// telling me "Advisory: Loop was not unrolled, unexpected call OPs"
+
+	if (SETS_PER_PASS > 0) CorrectUnguardedSetOverflow<RADIX_DIGITS>(pass_digits[0], pass_counts[0]);
+	if (SETS_PER_PASS > 1) CorrectUnguardedSetOverflow<RADIX_DIGITS>(pass_digits[1], pass_counts[1]);
+	if (SETS_PER_PASS > 2) CorrectUnguardedSetOverflow<RADIX_DIGITS>(pass_digits[2], pass_counts[2]);
+	if (SETS_PER_PASS > 3) CorrectUnguardedSetOverflow<RADIX_DIGITS>(pass_digits[3], pass_counts[3]);
+}
+
+
+template <int RADIX_DIGITS, int PASSES_PER_CYCLE, int SETS_PER_PASS>
+__device__ __forceinline__ void CorrectUnguardedCycleOverflow(
+	int2 			cycle_digits[PASSES_PER_CYCLE][SETS_PER_PASS],
+	int 	cycle_counts[PASSES_PER_CYCLE][SETS_PER_PASS])
+{
+	// N.B. -- I wish we could do some pragma unrolling here too, but the compiler won't comply, 
+	// telling me "Advisory: Loop was not unrolled, unexpected call OPs"
+
+	if (PASSES_PER_CYCLE > 0) CorrectUnguardedPassOverflow<RADIX_DIGITS, SETS_PER_PASS>(cycle_digits[0], cycle_counts[0]);
+	if (PASSES_PER_CYCLE > 1) CorrectUnguardedPassOverflow<RADIX_DIGITS, SETS_PER_PASS>(cycle_digits[1], cycle_counts[1]);
+}
+
+
+template <int RADIX_DIGITS>
+__device__ __forceinline__ void CorrectLastLaneOverflow(int &count, int extra[1]) 
+{
+	if (WarpVoteAll(RADIX_DIGITS, count == 0) && (threadIdx.x == RADIX_DIGITS - 1)) {
+		// We're 'f' and we overflowed b/c of invalid 'f' placemarkers; the number of valid items in this set is the count of valid f's 
+		count = extra[0] & 255;
+	}
+}
+		
+
+template <int RADIX_DIGITS, int PASSES_PER_CYCLE, int SETS_PER_PASS, int SETS_PER_CYCLE, bool UNGUARDED_IO>
+__device__ __forceinline__ void CorrectForOverflows(
+	int2 digits[PASSES_PER_CYCLE][SETS_PER_PASS],
+	int counts[PASSES_PER_CYCLE][SETS_PER_PASS], 
+	int extra[1])				
+{
+	if (!UNGUARDED_IO) {
+
+		// Correct any overflow in the partially-filled last lane
+		int *linear_counts = (int *) counts;
+		CorrectLastLaneOverflow<RADIX_DIGITS>(linear_counts[SETS_PER_CYCLE - 1], extra);
+	}
+
+	CorrectUnguardedCycleOverflow<RADIX_DIGITS, PASSES_PER_CYCLE, SETS_PER_PASS>(digits, counts);
+}
+
+
+template <
+	typename K,
+	int BIT, 
+	int RADIX_DIGITS,
+	int SCAN_LANES_PER_SET,
+	int SETS_PER_PASS,
+	int RAKING_THREADS_PER_PASS,
+	int SCAN_LANES_PER_PASS,
+	int LOG_RAKING_THREADS_PER_LANE,
+	int RAKING_THREADS_PER_LANE,
+	int PARTIALS_PER_SEG,
+	int PADDED_PARTIALS_PER_LANE,
+	int PASSES_PER_CYCLE>
+__device__ __forceinline__ void ScanPass(
+	int *base_partial,
+	int	*raking_partial,
+	int warpscan[SCAN_LANES_PER_PASS][3][RAKING_THREADS_PER_LANE],
+	typename VecType<K, 2>::Type keypairs[SETS_PER_PASS],
+	int2 digits[SETS_PER_PASS],
+	int2 flag_offsets[SETS_PER_PASS],
+	int2 ranks[SETS_PER_PASS],
+	int copy_section)
+{
+	// Reset smem
+	#pragma unroll
+	for (int SCAN_LANE = 0; SCAN_LANE < (int) SCAN_LANES_PER_PASS; SCAN_LANE++) {
+		base_partial[SCAN_LANE * PADDED_PARTIALS_PER_LANE] = 0;
+	}
+	
+	// Decode digits for first pass
+	DecodeDigits<K, RADIX_DIGITS, BIT, SETS_PER_PASS, SCAN_LANES_PER_SET, PADDED_PARTIALS_PER_LANE>(
+		keypairs, digits, flag_offsets);
+	
+	// Encode counts into smem for first pass
+	PlacePartials<SETS_PER_PASS>(
+		(unsigned char *) base_partial,
+		digits,
+		flag_offsets); 
+	
+	__syncthreads();
+	
+	// Intra-group prefix scans for first pass
+	if (threadIdx.x < RAKING_THREADS_PER_PASS) {
+	
+		PrefixScanOverLanes<SCAN_LANES_PER_PASS, LOG_RAKING_THREADS_PER_LANE, RAKING_THREADS_PER_LANE, PARTIALS_PER_SEG>(		// first pass is offset right by one
+			raking_partial,
+			warpscan, 
+			copy_section);
+	}
+	
+	__syncthreads();
+
+	// Extract ranks
+	ExtractRanks<SETS_PER_PASS>(
+		(unsigned char *) base_partial, 
+		digits, 
+		flag_offsets, 
+		ranks); 	
+}	
+	
+
+/******************************************************************************
+ * SM1.3 Local Exchange Routines
+ * 
+ * Routines for exchanging keys (and values) in shared memory (i.e., local 
+ * scattering) in order to to facilitate coalesced global scattering
+ ******************************************************************************/
+
+template <typename T, bool UNGUARDED_IO, int PASSES_PER_CYCLE, int SETS_PER_PASS, typename PostprocessFunctor>
+__device__ __forceinline__ void ScatterSets(
+	T *d_out, 
+	typename VecType<T, 2>::Type pairs[SETS_PER_PASS],
+	int2 offsets[SETS_PER_PASS],
+	const int BASE4,
+	int extra[1],
+	PostprocessFunctor postprocess = PostprocessFunctor())				
+{
+	#pragma unroll 
+	for (int SET = 0; SET < (int) SETS_PER_PASS; SET++) {
+		postprocess(pairs[SET].x);
+		postprocess(pairs[SET].y);
+	}
+
+	// N.B. -- I wish we could do some pragma unrolling here too, but the compiler makes it 1% slower 
+		
+	if (SETS_PER_PASS > 0) { 
+		if (UNGUARDED_IO || (threadIdx.x + BASE4 + (B40C_RADIXSORT_THREADS * 0) < extra[0])) 
+			d_out[offsets[0].x] = pairs[0].x;
+		if (UNGUARDED_IO || (threadIdx.x + BASE4 + (B40C_RADIXSORT_THREADS * 1) < extra[0])) 
+			d_out[offsets[0].y] = pairs[0].y;
+	}
+
+	if (SETS_PER_PASS > 1) { 
+		if (UNGUARDED_IO || (threadIdx.x + BASE4 + (B40C_RADIXSORT_THREADS * 2) < extra[0])) 
+			d_out[offsets[1].x] = pairs[1].x;
+		if (UNGUARDED_IO || (threadIdx.x + BASE4 + (B40C_RADIXSORT_THREADS * 3) < extra[0])) 
+			d_out[offsets[1].y] = pairs[1].y;
+	}
+
+	if (SETS_PER_PASS > 2) { 
+		if (UNGUARDED_IO || (threadIdx.x + BASE4 + (B40C_RADIXSORT_THREADS * 4) < extra[0])) 
+			d_out[offsets[2].x] = pairs[2].x;
+		if (UNGUARDED_IO || (threadIdx.x + BASE4 + (B40C_RADIXSORT_THREADS * 5) < extra[0])) 
+			d_out[offsets[2].y] = pairs[2].y;
+	}
+
+	if (SETS_PER_PASS > 3) { 
+		if (UNGUARDED_IO || (threadIdx.x + BASE4 + (B40C_RADIXSORT_THREADS * 6) < extra[0])) 
+			d_out[offsets[3].x] = pairs[3].x;
+		if (UNGUARDED_IO || (threadIdx.x + BASE4 + (B40C_RADIXSORT_THREADS * 7) < extra[0])) 
+			d_out[offsets[3].y] = pairs[3].y;
+	}
+}
+
+template <typename T, int PASSES_PER_CYCLE, int SETS_PER_PASS>
+__device__ __forceinline__ void PushPairs(
+	T *swap, 
+	typename VecType<T, 2>::Type pairs[PASSES_PER_CYCLE][SETS_PER_PASS],
+	int2 ranks[PASSES_PER_CYCLE][SETS_PER_PASS])				
+{
+	#pragma unroll 
+	for (int PASS = 0; PASS < (int) PASSES_PER_CYCLE; PASS++) {
+	
+		#pragma unroll 
+		for (int SET = 0; SET < (int) SETS_PER_PASS; SET++) {
+			swap[ranks[PASS][SET].x] = pairs[PASS][SET].x;
+			swap[ranks[PASS][SET].y] = pairs[PASS][SET].y;
+		}
+	}
+}
+	
+template <typename T, int PASSES_PER_CYCLE, int SETS_PER_PASS>
+__device__ __forceinline__ void ExchangePairs(
+	T *swap, 
+	typename VecType<T, 2>::Type pairs[PASSES_PER_CYCLE][SETS_PER_PASS],
+	int2 ranks[PASSES_PER_CYCLE][SETS_PER_PASS])				
+{
+	// Push in Pairs
+	PushPairs<T, PASSES_PER_CYCLE, SETS_PER_PASS>(swap, pairs, ranks);
+	
+	__syncthreads();
+	
+	// Extract pairs
+	#pragma unroll 
+	for (int PASS = 0; PASS < (int) PASSES_PER_CYCLE; PASS++) {
+		
+		#pragma unroll 
+		for (int SET = 0; SET < (int) SETS_PER_PASS; SET++) {
+			const int BLOCK = ((PASS * SETS_PER_PASS) + SET) * 2;
+			pairs[PASS][SET].x = swap[threadIdx.x + (B40C_RADIXSORT_THREADS * (BLOCK + 0))];
+			pairs[PASS][SET].y = swap[threadIdx.x + (B40C_RADIXSORT_THREADS * (BLOCK + 1))];
+		}
+	}
+}
+
+
+template <
+	typename K,
+	typename V,	
+	int RADIX_DIGITS, 
+	int BIT, 
+	int PASSES_PER_CYCLE,
+	int SETS_PER_PASS,
+	bool UNGUARDED_IO,
+	typename PostprocessFunctor>
+__device__ __forceinline__ void SwapAndScatterSm13(
+	typename VecType<K, 2>::Type keypairs[PASSES_PER_CYCLE][SETS_PER_PASS], 
+	int2 ranks[PASSES_PER_CYCLE][SETS_PER_PASS],
+	int4 *exchange,
+	typename VecType<V, 2>::Type *d_in_values, 
+	K *d_out_keys, 
+	V *d_out_values, 
+	int carry[RADIX_DIGITS], 
+	int extra[1])				
+{
+	int2 offsets[PASSES_PER_CYCLE][SETS_PER_PASS];
+	
+	// Swap keys according to ranks
+	ExchangePairs<K, PASSES_PER_CYCLE, SETS_PER_PASS>((K*) exchange, keypairs, ranks);				
+	
+	// Calculate scatter offsets (re-decode digits from keys: it's less work than making a second exchange of digits) 
+	#pragma unroll 
+	for (int PASS = 0; PASS < (int) PASSES_PER_CYCLE; PASS++) {
+		
+		#pragma unroll 
+		for (int SET = 0; SET < (int) SETS_PER_PASS; SET++) {
+			const int BLOCK = ((PASS * SETS_PER_PASS) + SET) * 2;
+			offsets[PASS][SET].x = threadIdx.x + (B40C_RADIXSORT_THREADS * (BLOCK + 0)) + carry[DecodeDigit<K, RADIX_DIGITS, BIT>(keypairs[PASS][SET].x)];
+			offsets[PASS][SET].y = threadIdx.x + (B40C_RADIXSORT_THREADS * (BLOCK + 1)) + carry[DecodeDigit<K, RADIX_DIGITS, BIT>(keypairs[PASS][SET].y)];
+		}
+	}
+	
+	// Scatter keys
+	#pragma unroll 
+	for (int PASS = 0; PASS < (int) PASSES_PER_CYCLE; PASS++) {
+		const int BLOCK = PASS * SETS_PER_PASS * 2;
+		ScatterSets<K, UNGUARDED_IO, PASSES_PER_CYCLE, SETS_PER_PASS, PostprocessFunctor>(d_out_keys, keypairs[PASS], offsets[PASS], B40C_RADIXSORT_THREADS * BLOCK, extra);
+	}
+
+	if (!IsKeysOnly<V>()) {
+	
+		__syncthreads();
+
+		// Read input data
+		typename VecType<V, 2>::Type datapairs[PASSES_PER_CYCLE][SETS_PER_PASS];
+
+		// N.B. -- I wish we could do some pragma unrolling here too, but the compiler won't comply, 
+		// telling me "Advisory: Loop was not unrolled, unexpected control flow"
+
+		if (PASSES_PER_CYCLE > 0) ReadSets<V, UNGUARDED_IO, SETS_PER_PASS, NopFunctor<V> >(d_in_values, datapairs[0], B40C_RADIXSORT_THREADS * SETS_PER_PASS * 0, extra);
+		if (PASSES_PER_CYCLE > 1) ReadSets<V, UNGUARDED_IO, SETS_PER_PASS, NopFunctor<V> >(d_in_values, datapairs[1], B40C_RADIXSORT_THREADS * SETS_PER_PASS * 1, extra);
+		
+		// Swap data according to ranks
+		ExchangePairs<V, PASSES_PER_CYCLE, SETS_PER_PASS>((V*) exchange, datapairs, ranks);
+		
+		// Scatter data
+		#pragma unroll 
+		for (int PASS = 0; PASS < (int) PASSES_PER_CYCLE; PASS++) {
+			const int BLOCK = PASS * SETS_PER_PASS * 2;
+			ScatterSets<V, UNGUARDED_IO, PASSES_PER_CYCLE, SETS_PER_PASS, NopFunctor<V> >(d_out_values, datapairs[PASS], offsets[PASS], B40C_RADIXSORT_THREADS * BLOCK, extra);
+		}
+	}
+}
+
+
+/******************************************************************************
+ * SM1.0 Local Exchange Routines
+ *
+ * Routines for exchanging keys (and values) in shared memory (i.e., local 
+ * scattering) in order to to facilitate coalesced global scattering
+ ******************************************************************************/
+
+template <
+	typename T, 
+	int RADIX_DIGITS,
+	bool UNGUARDED_IO,
+	typename PostprocessFunctor> 
+__device__ __forceinline__ void ScatterPass(
+	T *swapmem,
+	T *d_out, 
+	int digit_scan[2][RADIX_DIGITS], 
+	int carry[RADIX_DIGITS], 
+	int extra[1],
+	int base_digit,				
+	PostprocessFunctor postprocess = PostprocessFunctor())				
+{
+	const int LOG_STORE_TXN_THREADS = B40C_LOG_MEM_BANKS(__CUDA_ARCH__);
+	const int STORE_TXN_THREADS = 1 << LOG_STORE_TXN_THREADS;
+	
+	int store_txn_idx = threadIdx.x & (STORE_TXN_THREADS - 1);
+	int store_txn_digit = threadIdx.x >> LOG_STORE_TXN_THREADS;
+	
+	int my_digit = base_digit + store_txn_digit;
+	if (my_digit < RADIX_DIGITS) {
+	
+		int my_exclusive_scan = digit_scan[1][my_digit - 1];
+		int my_inclusive_scan = digit_scan[1][my_digit];
+		int my_digit_count = my_inclusive_scan - my_exclusive_scan;
+
+		int my_carry = carry[my_digit] + my_exclusive_scan;
+		int my_aligned_offset = store_txn_idx - (my_carry & (STORE_TXN_THREADS - 1));
+		
+		while (my_aligned_offset < my_digit_count) {
+
+			if ((my_aligned_offset >= 0) && (UNGUARDED_IO || (my_exclusive_scan + my_aligned_offset < extra[0]))) { 
+			
+				T datum = swapmem[my_exclusive_scan + my_aligned_offset];
+				postprocess(datum);
+				d_out[my_carry + my_aligned_offset] = datum;
+			}
+			my_aligned_offset += STORE_TXN_THREADS;
+		}
+	}
+}
+
+template <
+	typename T,
+	int RADIX_DIGITS, 
+	int PASSES_PER_CYCLE,
+	int SETS_PER_PASS,
+	bool UNGUARDED_IO,
+	typename PostprocessFunctor>
+__device__ __forceinline__ void SwapAndScatterPairs(
+	typename VecType<T, 2>::Type pairs[PASSES_PER_CYCLE][SETS_PER_PASS], 
+	int2 ranks[PASSES_PER_CYCLE][SETS_PER_PASS],
+	T *exchange,
+	T *d_out, 
+	int carry[RADIX_DIGITS], 
+	int digit_scan[2][RADIX_DIGITS], 
+	int extra[1])				
+{
+	const int SCATTER_PASS_DIGITS = B40C_RADIXSORT_WARPS * (B40C_WARP_THREADS / B40C_MEM_BANKS(__CUDA_ARCH__));
+	const int SCATTER_PASSES = RADIX_DIGITS / SCATTER_PASS_DIGITS;
+
+	// Push in pairs
+	PushPairs<T, PASSES_PER_CYCLE, SETS_PER_PASS>(exchange, pairs, ranks);
+
+	__syncthreads();
+
+	// N.B. -- I wish we could do some pragma unrolling here too, but the compiler won't comply, 
+	// telling me "Advisory: Loop was not unrolled, not an innermost loop"
+
+	if (SCATTER_PASSES > 0) ScatterPass<T, RADIX_DIGITS, UNGUARDED_IO, PostprocessFunctor>(exchange, d_out, digit_scan, carry, extra, SCATTER_PASS_DIGITS * 0);
+	if (SCATTER_PASSES > 1) ScatterPass<T, RADIX_DIGITS, UNGUARDED_IO, PostprocessFunctor>(exchange, d_out, digit_scan, carry, extra, SCATTER_PASS_DIGITS * 1);
+	if (SCATTER_PASSES > 2) ScatterPass<T, RADIX_DIGITS, UNGUARDED_IO, PostprocessFunctor>(exchange, d_out, digit_scan, carry, extra, SCATTER_PASS_DIGITS * 2);
+	if (SCATTER_PASSES > 3) ScatterPass<T, RADIX_DIGITS, UNGUARDED_IO, PostprocessFunctor>(exchange, d_out, digit_scan, carry, extra, SCATTER_PASS_DIGITS * 3);
+	if (SCATTER_PASSES > 4) ScatterPass<T, RADIX_DIGITS, UNGUARDED_IO, PostprocessFunctor>(exchange, d_out, digit_scan, carry, extra, SCATTER_PASS_DIGITS * 4);
+	if (SCATTER_PASSES > 5) ScatterPass<T, RADIX_DIGITS, UNGUARDED_IO, PostprocessFunctor>(exchange, d_out, digit_scan, carry, extra, SCATTER_PASS_DIGITS * 5);
+	if (SCATTER_PASSES > 6) ScatterPass<T, RADIX_DIGITS, UNGUARDED_IO, PostprocessFunctor>(exchange, d_out, digit_scan, carry, extra, SCATTER_PASS_DIGITS * 6);
+	if (SCATTER_PASSES > 7) ScatterPass<T, RADIX_DIGITS, UNGUARDED_IO, PostprocessFunctor>(exchange, d_out, digit_scan, carry, extra, SCATTER_PASS_DIGITS * 7);
+}
+
+
+template <
+	typename K,
+	typename V,	
+	int RADIX_DIGITS, 
+	int PASSES_PER_CYCLE,
+	int SETS_PER_PASS,
+	bool UNGUARDED_IO,
+	typename PostprocessFunctor>
+__device__ __forceinline__ void SwapAndScatterSm10(
+	typename VecType<K, 2>::Type keypairs[PASSES_PER_CYCLE][SETS_PER_PASS], 
+	int2 ranks[PASSES_PER_CYCLE][SETS_PER_PASS],
+	int4 *exchange,
+	typename VecType<V, 2>::Type *d_in_values, 
+	K *d_out_keys, 
+	V *d_out_values, 
+	int carry[RADIX_DIGITS], 
+	int digit_scan[2][RADIX_DIGITS], 
+	int extra[1])				
+{
+	// Swap and scatter keys
+	SwapAndScatterPairs<K, RADIX_DIGITS, PASSES_PER_CYCLE, SETS_PER_PASS, UNGUARDED_IO, PostprocessFunctor>(
+		keypairs, ranks, (K*) exchange, d_out_keys, carry, digit_scan, extra);				
+	
+	if (!IsKeysOnly<V>()) {
+
+		__syncthreads();
+		
+		// N.B. -- I wish we could do some pragma unrolling here too, but the compiler won't comply, 
+		// telling me "Advisory: Loop was not unrolled, unexpected control flow"
+
+		// Read input data
+		typename VecType<V, 2>::Type datapairs[PASSES_PER_CYCLE][SETS_PER_PASS];
+		if (PASSES_PER_CYCLE > 0) ReadSets<V, UNGUARDED_IO, SETS_PER_PASS, NopFunctor<V> >(d_in_values, datapairs[0], B40C_RADIXSORT_THREADS * SETS_PER_PASS * 0, extra);
+		if (PASSES_PER_CYCLE > 1) ReadSets<V, UNGUARDED_IO, SETS_PER_PASS, NopFunctor<V> >(d_in_values, datapairs[1], B40C_RADIXSORT_THREADS * SETS_PER_PASS * 1, extra);
+
+		// Swap and scatter data
+		SwapAndScatterPairs<V, RADIX_DIGITS, PASSES_PER_CYCLE, SETS_PER_PASS, UNGUARDED_IO, NopFunctor<V> >(
+			datapairs, ranks, (V*) exchange, d_out_values, carry, digit_scan, extra);				
+	}
+}
+
+
+/******************************************************************************
+ * Cycle of RADIXSORT_CYCLE_ELEMENTS keys (and values)
+ ******************************************************************************/
+
+template <
+	typename K,
+	typename V,	
+	int BIT, 
+	bool UNGUARDED_IO,
+	int RADIX_DIGITS,
+	int LOG_SCAN_LANES_PER_SET,
+	int SCAN_LANES_PER_SET,
+	int SETS_PER_PASS,
+	int PASSES_PER_CYCLE,
+	int LOG_SCAN_LANES_PER_PASS,
+	int SCAN_LANES_PER_PASS,
+	int LOG_PARTIALS_PER_LANE,
+	int LOG_PARTIALS_PER_PASS,
+	int LOG_RAKING_THREADS_PER_PASS,
+	int RAKING_THREADS_PER_PASS,
+	int LOG_RAKING_THREADS_PER_LANE,
+	int RAKING_THREADS_PER_LANE,
+	int LOG_PARTIALS_PER_SEG,
+	int PARTIALS_PER_SEG,
+	int LOG_PARTIALS_PER_ROW,
+	int PARTIALS_PER_ROW,
+	int LOG_SEGS_PER_ROW,	
+	int SEGS_PER_ROW,
+	int LOG_ROWS_PER_SET,
+	int LOG_ROWS_PER_LANE,
+	int ROWS_PER_LANE,
+	int LOG_ROWS_PER_PASS,
+	int ROWS_PER_PASS,
+	int MAX_EXCHANGE_BYTES,
+	typename PreprocessFunctor,
+	typename PostprocessFunctor>
+
+__device__ __forceinline__ void SrtsScanDigitCycle(
+	typename VecType<K, 2>::Type *d_in_keys, 
+	typename VecType<V, 2>::Type *d_in_values, 
+	K *d_out_keys, 
+	V *d_out_values, 
+	int4 *exchange,								
+	int	warpscan[SCAN_LANES_PER_PASS][3][RAKING_THREADS_PER_LANE],
+	int	carry[RADIX_DIGITS],
+	int	digit_scan[2][RADIX_DIGITS],						 
+	int	digit_counts[PASSES_PER_CYCLE][SETS_PER_PASS][RADIX_DIGITS],
+	int	extra[1],
+	int	*base_partial,
+	int	*raking_partial)		
+{
+	
+	const int PADDED_PARTIALS_PER_LANE 		= ROWS_PER_LANE * (PARTIALS_PER_ROW + 1);	 
+	const int SETS_PER_CYCLE 				= PASSES_PER_CYCLE * SETS_PER_PASS;
+
+	// N.B.: We use the following voodoo incantations to elide the compiler's miserable 
+	// "declared but never referenced" warnings for these (which are actually used for 
+	// template instantiation)	
+	SuppressUnusedConstantWarning(PADDED_PARTIALS_PER_LANE);
+	SuppressUnusedConstantWarning(SETS_PER_CYCLE);
+	
+	typename VecType<K, 2>::Type 	keypairs[PASSES_PER_CYCLE][SETS_PER_PASS];
+	int2 							digits[PASSES_PER_CYCLE][SETS_PER_PASS];
+	int2 							flag_offsets[PASSES_PER_CYCLE][SETS_PER_PASS];		// a byte offset
+	int2 							ranks[PASSES_PER_CYCLE][SETS_PER_PASS];
+
+	
+	//-------------------------------------------------------------------------
+	// Read keys
+	//-------------------------------------------------------------------------
+
+	// N.B. -- I wish we could do some pragma unrolling here too, but the compiler won't comply, 
+	// telling me "Advisory: Loop was not unrolled, unexpected control flow construct"
+	
+	// Read Keys
+	if (PASSES_PER_CYCLE > 0) ReadSets<K, UNGUARDED_IO, SETS_PER_PASS, PreprocessFunctor>(d_in_keys, keypairs[0], B40C_RADIXSORT_THREADS * SETS_PER_PASS * 0, extra);		 
+	if (PASSES_PER_CYCLE > 1) ReadSets<K, UNGUARDED_IO, SETS_PER_PASS, PreprocessFunctor>(d_in_keys, keypairs[1], B40C_RADIXSORT_THREADS * SETS_PER_PASS * 1, extra); 	
+	
+	//-------------------------------------------------------------------------
+	// Lane-scanning Passes
+	//-------------------------------------------------------------------------
+
+	#pragma unroll
+	for (int PASS = 0; PASS < (int) PASSES_PER_CYCLE; PASS++) {
+	
+		// First Pass
+		ScanPass<K, BIT, RADIX_DIGITS, SCAN_LANES_PER_SET, SETS_PER_PASS, RAKING_THREADS_PER_PASS, SCAN_LANES_PER_PASS, LOG_RAKING_THREADS_PER_LANE, RAKING_THREADS_PER_LANE, PARTIALS_PER_SEG, PADDED_PARTIALS_PER_LANE, PASSES_PER_CYCLE>(
+			base_partial,
+			raking_partial,
+			warpscan,
+			keypairs[PASS],
+			digits[PASS],
+			flag_offsets[PASS],
+			ranks[PASS],
+			PASSES_PER_CYCLE - PASS - 1);		// lower passes get copied right
+	}
+	
+	//-------------------------------------------------------------------------
+	// Digit-scanning 
+	//-------------------------------------------------------------------------
+
+	// Recover second-half digit-counts, scan across all digit-counts
+	if (threadIdx.x < RADIX_DIGITS) {
+
+		int counts[PASSES_PER_CYCLE][SETS_PER_PASS];
+
+		// Recover digit-counts
+
+		#pragma unroll
+		for (int PASS = 0; PASS < (int) PASSES_PER_CYCLE; PASS++) {
+			RecoverDigitCounts<SCAN_LANES_PER_PASS, RAKING_THREADS_PER_LANE, SETS_PER_PASS, SCAN_LANES_PER_SET>(		// first pass, offset by 1			
+				warpscan, 
+				counts[PASS],
+				PASSES_PER_CYCLE - PASS - 1);		// lower passes get copied right
+		}
+		
+		// Check for overflows
+		CorrectForOverflows<RADIX_DIGITS, PASSES_PER_CYCLE, SETS_PER_PASS, SETS_PER_CYCLE, UNGUARDED_IO>(
+				digits, counts, extra);
+
+		// Scan across my digit counts for each set 
+		int exclusive_total = 0;
+		int inclusive_total = 0;
+		
+		#pragma unroll
+		for (int PASS = 0; PASS < (int) PASSES_PER_CYCLE; PASS++) {
+		
+			#pragma unroll
+			for (int SET = 0; SET < (int) SETS_PER_PASS; SET++) {
+				inclusive_total += counts[PASS][SET];
+				counts[PASS][SET] = exclusive_total;
+				exclusive_total = inclusive_total;
+			}
+		}
+
+		// second half of carry update
+		int my_carry = carry[threadIdx.x] + digit_scan[1][threadIdx.x];
+
+		// Perform overflow-free SIMD Kogge-Stone across digits
+		int digit_prefix = WarpScan<RADIX_DIGITS, false>(
+				digit_scan, 
+				inclusive_total,
+				0);
+
+		// first-half of carry update 
+		carry[threadIdx.x] = my_carry - digit_prefix;
+		
+		#pragma unroll
+		for (int PASS = 0; PASS < (int) PASSES_PER_CYCLE; PASS++) {
+
+			#pragma unroll
+			for (int SET = 0; SET < (int) SETS_PER_PASS; SET++) {
+				digit_counts[PASS][SET][threadIdx.x] = counts[PASS][SET] + digit_prefix;
+			}
+		}
+	}
+	
+	__syncthreads();
+
+	//-------------------------------------------------------------------------
+	// Update Ranks
+	//-------------------------------------------------------------------------
+
+	UpdateRanks<RADIX_DIGITS, PASSES_PER_CYCLE, SETS_PER_PASS>(digits, ranks, digit_counts);
+	
+	
+	//-------------------------------------------------------------------------
+	// Scatter 
+	//-------------------------------------------------------------------------
+
+#if ((__CUDA_ARCH__ < 130) || FERMI_ECC)		
+
+	SwapAndScatterSm10<K, V, RADIX_DIGITS, PASSES_PER_CYCLE, SETS_PER_PASS, UNGUARDED_IO, PostprocessFunctor>(
+		keypairs, 
+		ranks,
+		exchange,
+		d_in_values, 
+		d_out_keys, 
+		d_out_values, 
+		carry, 
+		digit_scan,
+		extra);
+	
+#else 
+
+	SwapAndScatterSm13<K, V, RADIX_DIGITS, BIT, PASSES_PER_CYCLE, SETS_PER_PASS, UNGUARDED_IO, PostprocessFunctor>(
+		keypairs, 
+		ranks,
+		exchange,
+		d_in_values, 
+		d_out_keys, 
+		d_out_values, 
+		carry, 
+		extra);
+	
+#endif
+
+	__syncthreads();
+
+}
+
+
+
+/******************************************************************************
+ * Scan/Scatter Kernel Entry Point
+ ******************************************************************************/
+
+template <
+	typename K, 
+	typename V, 
+	int PASS, 
+	int RADIX_BITS, 
+	int BIT, 
+	typename PreprocessFunctor, 
+	typename PostprocessFunctor>
+__launch_bounds__ (B40C_RADIXSORT_THREADS, B40C_RADIXSORT_SCAN_SCATTER_CTA_OCCUPANCY(__CUDA_ARCH__))
+__global__ 
+void ScanScatterDigits(
+	bool *d_from_alt_storage,
+	int* d_spine,
+	K* d_in_keys,
+	K* d_out_keys,
+	V* d_in_values,
+	V* d_out_values,
+	CtaDecomposition work_decomposition)
+{
+
+	const int RADIX_DIGITS 				= 1 << RADIX_BITS;
+	
+	const int LOG_SCAN_LANES_PER_SET	= (RADIX_BITS > 2) ? RADIX_BITS - 2 : 0;					// Always at one lane per set
+	const int SCAN_LANES_PER_SET		= 1 << LOG_SCAN_LANES_PER_SET;								// N.B.: we have "declared but never referenced" warnings for these, but they're actually used for template instantiation
+	
+	const int LOG_SETS_PER_PASS			= B40C_RADIXSORT_LOG_SETS_PER_PASS(__CUDA_ARCH__);			
+	const int SETS_PER_PASS				= 1 << LOG_SETS_PER_PASS;
+	
+	const int LOG_PASSES_PER_CYCLE		= B40C_RADIXSORT_LOG_PASSES_PER_CYCLE(__CUDA_ARCH__, K, V);			
+	const int PASSES_PER_CYCLE			= 1 << LOG_PASSES_PER_CYCLE;
+
+	const int LOG_SCAN_LANES_PER_PASS	= LOG_SETS_PER_PASS + LOG_SCAN_LANES_PER_SET;
+	const int SCAN_LANES_PER_PASS		= 1 << LOG_SCAN_LANES_PER_PASS;
+	
+	const int LOG_PARTIALS_PER_LANE 	= B40C_RADIXSORT_LOG_THREADS;
+	
+	const int LOG_PARTIALS_PER_PASS		= LOG_SCAN_LANES_PER_PASS + LOG_PARTIALS_PER_LANE;
+
+	const int LOG_RAKING_THREADS_PER_PASS 		= B40C_RADIXSORT_LOG_RAKING_THREADS_PER_PASS(__CUDA_ARCH__);
+	const int RAKING_THREADS_PER_PASS			= 1 << LOG_RAKING_THREADS_PER_PASS;
+
+	const int LOG_RAKING_THREADS_PER_LANE 		= LOG_RAKING_THREADS_PER_PASS - LOG_SCAN_LANES_PER_PASS;
+	const int RAKING_THREADS_PER_LANE 			= 1 << LOG_RAKING_THREADS_PER_LANE;
+
+	const int LOG_PARTIALS_PER_SEG 		= LOG_PARTIALS_PER_LANE - LOG_RAKING_THREADS_PER_LANE;
+	const int PARTIALS_PER_SEG 			= 1 << LOG_PARTIALS_PER_SEG;
+
+	const int LOG_PARTIALS_PER_ROW		= (LOG_PARTIALS_PER_SEG < B40C_LOG_MEM_BANKS(__CUDA_ARCH__)) ? B40C_LOG_MEM_BANKS(__CUDA_ARCH__) : LOG_PARTIALS_PER_SEG;		// floor of MEM_BANKS partials per row
+	const int PARTIALS_PER_ROW			= 1 << LOG_PARTIALS_PER_ROW;
+	const int PADDED_PARTIALS_PER_ROW 	= PARTIALS_PER_ROW + 1;
+
+	const int LOG_SEGS_PER_ROW 			= LOG_PARTIALS_PER_ROW - LOG_PARTIALS_PER_SEG;	
+	const int SEGS_PER_ROW				= 1 << LOG_SEGS_PER_ROW;
+
+	const int LOG_ROWS_PER_SET 			= LOG_PARTIALS_PER_PASS - LOG_PARTIALS_PER_ROW;
+
+	const int LOG_ROWS_PER_LANE 		= LOG_PARTIALS_PER_LANE - LOG_PARTIALS_PER_ROW;
+	const int ROWS_PER_LANE 			= 1 << LOG_ROWS_PER_LANE;
+
+	const int LOG_ROWS_PER_PASS 		= LOG_SCAN_LANES_PER_PASS + LOG_ROWS_PER_LANE;
+	const int ROWS_PER_PASS 			= 1 << LOG_ROWS_PER_PASS;
+	
+	const int SCAN_LANE_BYTES			= ROWS_PER_PASS * PADDED_PARTIALS_PER_ROW * sizeof(int);
+	const int MAX_EXCHANGE_BYTES		= (sizeof(K) > sizeof(V)) ? 
+													B40C_RADIXSORT_CYCLE_ELEMENTS(__CUDA_ARCH__, K, V) * sizeof(K) : 
+													B40C_RADIXSORT_CYCLE_ELEMENTS(__CUDA_ARCH__, K, V) * sizeof(V);
+	const int SCAN_LANE_INT4S         = (B40C_MAX(MAX_EXCHANGE_BYTES, SCAN_LANE_BYTES) + sizeof(int4) - 1) / sizeof(int4);
+
+
+	// N.B.: We use the following voodoo incantations to elide the compiler's miserable 
+	// "declared but never referenced" warnings for these (which are actually used for 
+	// template instantiation)	
+	SuppressUnusedConstantWarning(SCAN_LANES_PER_SET);
+	SuppressUnusedConstantWarning(PARTIALS_PER_SEG);
+	SuppressUnusedConstantWarning(LOG_ROWS_PER_SET);
+	SuppressUnusedConstantWarning(ROWS_PER_LANE);
+
+    // scan_lanes is a int4[] to avoid alignment issues when casting to (K *) and/or (V *)
+	__shared__ int4		scan_lanes[SCAN_LANE_INT4S];
+	__shared__ int 		warpscan[SCAN_LANES_PER_PASS][3][RAKING_THREADS_PER_LANE];		// One warpscan per fours-group
+	__shared__ int 		carry[RADIX_DIGITS];
+	__shared__ int 		digit_scan[2][RADIX_DIGITS];						 
+	__shared__ int 		digit_counts[PASSES_PER_CYCLE][SETS_PER_PASS][RADIX_DIGITS];
+	__shared__ bool 	non_trivial_digit_pass;
+	__shared__ bool		from_alt_storage;
+	
+	_B40C_REG_MISER_QUALIFIER_ int extra[1];
+	_B40C_REG_MISER_QUALIFIER_ int oob[1];
+
+	extra[0] = (blockIdx.x == gridDim.x - 1) ? work_decomposition.extra_elements_last_block : 0;
+
+	// calculate our threadblock's range
+	int block_elements, block_offset;
+	if (blockIdx.x < work_decomposition.num_big_blocks) {
+		block_offset = work_decomposition.big_block_elements * blockIdx.x;
+		block_elements = work_decomposition.big_block_elements;
+	} else {
+		block_offset = (work_decomposition.normal_block_elements * blockIdx.x) + (work_decomposition.num_big_blocks * B40C_RADIXSORT_CYCLE_ELEMENTS(__CUDA_ARCH__, K, V));
+		block_elements = work_decomposition.normal_block_elements;
+	}
+	oob[0] = block_offset + block_elements;	// out-of-bounds
+
+	
+	// location for placing 2-element partial reductions in the first lane of a pass	
+	int row = threadIdx.x >> LOG_PARTIALS_PER_ROW; 
+	int col = threadIdx.x & (PARTIALS_PER_ROW - 1); 
+	int *base_partial = reinterpret_cast<int *>(scan_lanes) + (row * PADDED_PARTIALS_PER_ROW) + col; 								
+	
+	// location for raking across all sets within a pass
+	int *raking_partial = 0;										
+
+	if (threadIdx.x < RAKING_THREADS_PER_PASS) {
+
+		// initalize lane warpscans
+		if (threadIdx.x < RAKING_THREADS_PER_LANE) {
+			
+			#pragma unroll
+			for (int SCAN_LANE = 0; SCAN_LANE < (int) SCAN_LANES_PER_PASS; SCAN_LANE++) {
+				warpscan[SCAN_LANE][0][threadIdx.x] = 0;
+			}
+		}
+
+		// initialize digit warpscans
+		if (threadIdx.x < RADIX_DIGITS) {
+
+			// Initialize digit_scan
+			digit_scan[0][threadIdx.x] = 0;
+			digit_scan[1][threadIdx.x] = 0;
+
+			// Determine where to read our input
+			from_alt_storage = (PASS == 0) ? false : d_from_alt_storage[PASS & 0x1];
+
+			// Read carry in parallel 
+			int spine_digit_offset = FastMul(gridDim.x, threadIdx.x);
+			int my_digit_carry = d_spine[spine_digit_offset + blockIdx.x];
+			carry[threadIdx.x] = my_digit_carry;
+
+			// Determine whether or not we have work to do and setup the next round 
+			// accordingly.  Everybody but the first threadblock can determine this 
+			// from the number of non-zero-and-non-oob digit carries.  First block 
+			// needs someone else's because he always writes the zero offset.
+			
+			int predicate;
+			if (PreprocessFunctor::MustApply() || PostprocessFunctor::MustApply()) {
+
+				non_trivial_digit_pass = true;
+
+			} else {
+
+				if (blockIdx.x > 0) {
+					// Non-first CTA : use digit-carry from first block
+					my_digit_carry = d_spine[spine_digit_offset];
+				}
+				
+				predicate = ((my_digit_carry > 0) && (my_digit_carry < work_decomposition.num_elements));
+				non_trivial_digit_pass = (TallyWarpVote(RADIX_DIGITS, predicate, reinterpret_cast<int *>(scan_lanes)) > 0);
+			}
+
+			// Let the next round know which set of buffers to use
+			if (blockIdx.x == 0) d_from_alt_storage[(PASS + 1) & 0x1] = from_alt_storage ^ non_trivial_digit_pass;
+		}
+
+		// initialize raking segment
+		row = threadIdx.x >> LOG_SEGS_PER_ROW;
+		col = (threadIdx.x & (SEGS_PER_ROW - 1)) << LOG_PARTIALS_PER_SEG;
+		raking_partial = reinterpret_cast<int *>(scan_lanes) + (row * PADDED_PARTIALS_PER_ROW) + col; 
+	}
+
+	// Sync to acquire non_trivial_digit_pass and from_temp_storage
+	__syncthreads();
+	
+	// Short-circuit this entire pass
+	if (!non_trivial_digit_pass) return; 
+
+	if (!from_alt_storage) {
+	
+		// Scan in tiles of cycle_elements
+		while (block_offset < oob[0]) {
+	
+			SrtsScanDigitCycle<K, V, BIT, true, RADIX_DIGITS, LOG_SCAN_LANES_PER_SET, SCAN_LANES_PER_SET, SETS_PER_PASS, PASSES_PER_CYCLE, LOG_SCAN_LANES_PER_PASS, SCAN_LANES_PER_PASS, LOG_PARTIALS_PER_LANE, LOG_PARTIALS_PER_PASS, LOG_RAKING_THREADS_PER_PASS, RAKING_THREADS_PER_PASS, LOG_RAKING_THREADS_PER_LANE, RAKING_THREADS_PER_LANE, LOG_PARTIALS_PER_SEG, PARTIALS_PER_SEG, LOG_PARTIALS_PER_ROW, PARTIALS_PER_ROW, LOG_SEGS_PER_ROW, SEGS_PER_ROW, LOG_ROWS_PER_SET, LOG_ROWS_PER_LANE, ROWS_PER_LANE, LOG_ROWS_PER_PASS, ROWS_PER_PASS, MAX_EXCHANGE_BYTES, PreprocessFunctor, PostprocessFunctor>(	
+				reinterpret_cast<typename VecType<K, 2>::Type *>((void *) &d_in_keys[block_offset]), 
+				reinterpret_cast<typename VecType<V, 2>::Type *>((void *) &d_in_values[block_offset]), 
+				d_out_keys, 
+				d_out_values, 
+				scan_lanes,
+				warpscan,
+				carry,
+				digit_scan,						 
+				digit_counts,
+				extra,
+				base_partial,
+				raking_partial);		
+	
+			block_offset += B40C_RADIXSORT_CYCLE_ELEMENTS(__CUDA_ARCH__, K, V);
+		}
+	
+		if (extra[0]) {
+			
+			SrtsScanDigitCycle<K, V, BIT, false, RADIX_DIGITS, LOG_SCAN_LANES_PER_SET, SCAN_LANES_PER_SET, SETS_PER_PASS, PASSES_PER_CYCLE, LOG_SCAN_LANES_PER_PASS, SCAN_LANES_PER_PASS, LOG_PARTIALS_PER_LANE, LOG_PARTIALS_PER_PASS, LOG_RAKING_THREADS_PER_PASS, RAKING_THREADS_PER_PASS, LOG_RAKING_THREADS_PER_LANE, RAKING_THREADS_PER_LANE, LOG_PARTIALS_PER_SEG, PARTIALS_PER_SEG, LOG_PARTIALS_PER_ROW, PARTIALS_PER_ROW, LOG_SEGS_PER_ROW, SEGS_PER_ROW, LOG_ROWS_PER_SET, LOG_ROWS_PER_LANE, ROWS_PER_LANE, LOG_ROWS_PER_PASS, ROWS_PER_PASS, MAX_EXCHANGE_BYTES, PreprocessFunctor, PostprocessFunctor>(	
+				reinterpret_cast<typename VecType<K, 2>::Type *>((void *) &d_in_keys[block_offset]), 
+				reinterpret_cast<typename VecType<V, 2>::Type *>((void *) &d_in_values[block_offset]), 
+				d_out_keys, 
+				d_out_values, 
+				scan_lanes,
+				warpscan,
+				carry,
+				digit_scan,						 
+				digit_counts,
+				extra,
+				base_partial,
+				raking_partial);		
+		}
+
+	} else {
+		
+		// Scan in tiles of cycle_elements
+		while (block_offset < oob[0]) {
+
+			SrtsScanDigitCycle<K, V, BIT, true, RADIX_DIGITS, LOG_SCAN_LANES_PER_SET, SCAN_LANES_PER_SET, SETS_PER_PASS, PASSES_PER_CYCLE, LOG_SCAN_LANES_PER_PASS, SCAN_LANES_PER_PASS, LOG_PARTIALS_PER_LANE, LOG_PARTIALS_PER_PASS, LOG_RAKING_THREADS_PER_PASS, RAKING_THREADS_PER_PASS, LOG_RAKING_THREADS_PER_LANE, RAKING_THREADS_PER_LANE, LOG_PARTIALS_PER_SEG, PARTIALS_PER_SEG, LOG_PARTIALS_PER_ROW, PARTIALS_PER_ROW, LOG_SEGS_PER_ROW, SEGS_PER_ROW, LOG_ROWS_PER_SET, LOG_ROWS_PER_LANE, ROWS_PER_LANE, LOG_ROWS_PER_PASS, ROWS_PER_PASS, MAX_EXCHANGE_BYTES, PreprocessFunctor, PostprocessFunctor>(	
+				reinterpret_cast<typename VecType<K, 2>::Type *>((void *) &d_out_keys[block_offset]), 
+				reinterpret_cast<typename VecType<V, 2>::Type *>((void *) &d_out_values[block_offset]), 
+				d_in_keys, 
+				d_in_values, 
+				scan_lanes,
+				warpscan,
+				carry,
+				digit_scan,						 
+				digit_counts,
+				extra,
+				base_partial,
+				raking_partial);		
+
+			block_offset += B40C_RADIXSORT_CYCLE_ELEMENTS(__CUDA_ARCH__, K, V);
+		}
+
+		if (extra[0]) {
+			
+			SrtsScanDigitCycle<K, V, BIT, false, RADIX_DIGITS, LOG_SCAN_LANES_PER_SET, SCAN_LANES_PER_SET, SETS_PER_PASS, PASSES_PER_CYCLE, LOG_SCAN_LANES_PER_PASS, SCAN_LANES_PER_PASS, LOG_PARTIALS_PER_LANE, LOG_PARTIALS_PER_PASS, LOG_RAKING_THREADS_PER_PASS, RAKING_THREADS_PER_PASS, LOG_RAKING_THREADS_PER_LANE, RAKING_THREADS_PER_LANE, LOG_PARTIALS_PER_SEG, PARTIALS_PER_SEG, LOG_PARTIALS_PER_ROW, PARTIALS_PER_ROW, LOG_SEGS_PER_ROW, SEGS_PER_ROW, LOG_ROWS_PER_SET, LOG_ROWS_PER_LANE, ROWS_PER_LANE, LOG_ROWS_PER_PASS, ROWS_PER_PASS, MAX_EXCHANGE_BYTES, PreprocessFunctor, PostprocessFunctor>(	
+				reinterpret_cast<typename VecType<K, 2>::Type *>((void *) &d_out_keys[block_offset]), 
+				reinterpret_cast<typename VecType<V, 2>::Type *>((void *) &d_out_values[block_offset]), 
+				d_in_keys, 
+				d_in_values, 
+				scan_lanes,
+				warpscan,
+				carry,
+				digit_scan,						 
+				digit_counts,
+				extra,
+				base_partial,
+				raking_partial);		
+		}
+		
+	}
+}
+
+} // end namespace b40c_thrust
+} // end namespace detail
+} // end namespace detail
+} // end namespace cuda
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/cuda/detail/detail/b40c/radixsort_spine_kernel.h b/compat/thrust/system/cuda/detail/detail/b40c/radixsort_spine_kernel.h
new file mode 100644
index 0000000000..3d20f4aa79
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/detail/b40c/radixsort_spine_kernel.h
@@ -0,0 +1,187 @@
+/******************************************************************************
+ * 
+ * Copyright 2010 Duane Merrill
+ * 
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ * 
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License. 
+ * 
+ * 
+ * 
+ * 
+ * AUTHORS' REQUEST: 
+ * 
+ * 		If you use|reference|benchmark this code, please cite our Technical 
+ * 		Report (http://www.cs.virginia.edu/~dgm4d/papers/RadixSortTR.pdf):
+ * 
+ *		@TechReport{ Merrill:Sorting:2010,
+ *        	author = "Duane Merrill and Andrew Grimshaw",
+ *        	title = "Revisiting Sorting for GPGPU Stream Architectures",
+ *        	year = "2010",
+ *        	institution = "University of Virginia, Department of Computer Science",
+ *        	address = "Charlottesville, VA, USA",
+ *        	number = "CS2010-03"
+ *		}
+ * 
+ * For more information, see our Google Code project site: 
+ * http://code.google.com/p/back40computing/
+ * 
+ * Thanks!
+ * 
+ ******************************************************************************/
+
+
+/******************************************************************************
+ * Top-level histogram/spine scanning kernel
+ ******************************************************************************/
+
+#pragma once
+
+#include "radixsort_kernel_common.h"
+
+namespace thrust  {
+namespace system  {
+namespace cuda    {
+namespace detail  {
+namespace detail  {
+namespace b40c_thrust   {
+
+/******************************************************************************
+ * Scans a cycle of RADIXSORT_CYCLE_ELEMENTS elements 
+ ******************************************************************************/
+
+template<int PARTIALS_PER_SEG>
+__device__ __forceinline__ void SrtsScanCycle(
+	int *smem_offset,
+	int *smem_segment,
+	int warpscan[2][B40C_WARP_THREADS],
+	int4 *in, 
+	int4 *out,
+	int &carry)
+{
+	int4 datum; 
+
+	// read input data
+	datum = in[threadIdx.x];
+
+	smem_offset[0] = datum.x + datum.y + datum.z + datum.w;
+
+	__syncthreads();
+
+	if (threadIdx.x < B40C_WARP_THREADS) {
+
+		int partial_reduction = SerialReduce<PARTIALS_PER_SEG>(smem_segment);
+
+		int seed = WarpScan<B40C_WARP_THREADS, false>(warpscan, partial_reduction, 0);
+		seed += carry;		
+		
+		SerialScan<PARTIALS_PER_SEG>(smem_segment, seed);
+
+		carry += warpscan[1][B40C_WARP_THREADS - 1];	
+	}
+
+	__syncthreads();
+
+	int part0 = smem_offset[0];
+	int part1;
+
+	part1 = datum.x + part0;
+	datum.x = part0;
+	part0 = part1 + datum.y;
+	datum.y = part1;
+
+	part1 = datum.z + part0;
+	datum.z = part0;
+	part0 = part1 + datum.w;
+	datum.w = part1;
+	
+	out[threadIdx.x] = datum;
+}
+
+
+/******************************************************************************
+ * Spine/histogram Scan Kernel Entry Point
+ ******************************************************************************/
+
+template <typename T>
+__global__ void SrtsScanSpine(
+	int *d_ispine,
+	int *d_ospine,
+	int normal_block_elements)
+{
+	const int LOG_PARTIALS				= B40C_RADIXSORT_LOG_THREADS;				
+	const int PARTIALS			 		= 1 << LOG_PARTIALS;
+	
+	const int LOG_PARTIALS_PER_SEG 		= LOG_PARTIALS - B40C_LOG_WARP_THREADS;
+	const int PARTIALS_PER_SEG 			= 1 << LOG_PARTIALS_PER_SEG;
+
+	const int LOG_PARTIALS_PER_ROW		= (LOG_PARTIALS_PER_SEG < B40C_LOG_MEM_BANKS(__CUDA_ARCH__)) ? B40C_LOG_MEM_BANKS(__CUDA_ARCH__) : LOG_PARTIALS_PER_SEG;		// floor of 32 elts per row
+	const int PARTIALS_PER_ROW			= 1 << LOG_PARTIALS_PER_ROW;
+	
+	const int LOG_SEGS_PER_ROW 			= LOG_PARTIALS_PER_ROW - LOG_PARTIALS_PER_SEG;	
+	const int SEGS_PER_ROW				= 1 << LOG_SEGS_PER_ROW;
+
+	const int SMEM_ROWS 				= PARTIALS / PARTIALS_PER_ROW;
+	
+	__shared__ int smem[SMEM_ROWS][PARTIALS_PER_ROW + 1];
+	__shared__ int warpscan[2][B40C_WARP_THREADS];
+
+  // WAR spurious unused constant warning
+  SuppressUnusedConstantWarning(PARTIALS_PER_SEG);
+
+	int *smem_segment = 0;
+	int carry = 0;
+
+	int row = threadIdx.x >> LOG_PARTIALS_PER_ROW;		
+	int col = threadIdx.x & (PARTIALS_PER_ROW - 1);			
+	int *smem_offset = &smem[row][col];
+
+	if (blockIdx.x > 0) {
+		return;
+	}
+	
+	if (threadIdx.x < B40C_WARP_THREADS) {
+		
+		// two segs per row, odd segs are offset by 8
+		row = threadIdx.x >> LOG_SEGS_PER_ROW;
+		col = (threadIdx.x & (SEGS_PER_ROW - 1)) << LOG_PARTIALS_PER_SEG;
+		smem_segment = &smem[row][col];
+	
+		if (threadIdx.x < B40C_WARP_THREADS) {
+			carry = 0;
+			warpscan[0][threadIdx.x] = 0;
+		}
+	}
+
+	// scan the spine in blocks of cycle_elements
+	int block_offset = 0;
+	while (block_offset < normal_block_elements) {
+		
+		SrtsScanCycle<PARTIALS_PER_SEG>(	
+			smem_offset, 
+			smem_segment, 
+			warpscan,
+			reinterpret_cast<int4 *>((void *) &d_ispine[block_offset]), 
+			reinterpret_cast<int4 *>((void *) &d_ospine[block_offset]), 
+			carry);
+
+		block_offset += B40C_RADIXSORT_SPINE_CYCLE_ELEMENTS;
+	}
+} 
+
+
+} // end namespace b40c_thrust
+} // end namespace detail
+} // end namespace detail
+} // end namespace cuda
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/cuda/detail/detail/b40c/vector_types.h b/compat/thrust/system/cuda/detail/detail/b40c/vector_types.h
new file mode 100644
index 0000000000..6db7931078
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/detail/b40c/vector_types.h
@@ -0,0 +1,96 @@
+/**
+ * Copyright 2010 Duane Merrill
+ * 
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ * 
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License. 
+ * 
+ * For more information, see our Google Code project site: 
+ * http://code.google.com/p/back40computing/
+ * 
+ * Thanks!
+ */
+
+#pragma once
+
+#include <vector_types.h>
+
+namespace thrust  {
+namespace system  {
+namespace cuda    {
+namespace detail  {
+namespace detail  {
+namespace b40c_thrust   {
+
+//------------------------------------------------------------------------------
+// Vector types
+//------------------------------------------------------------------------------
+
+template <typename K, int vec_elements> struct VecType;
+
+
+//
+// Define general vector types
+//
+
+template <typename K> 
+struct VecType<K, 1> {
+	K x;
+	typedef K Type;
+};
+
+template <typename K> 
+struct VecType<K, 2> {
+	K x;
+	K y;
+	typedef VecType<K, 2> Type;
+};
+
+template <typename K> 
+struct VecType<K, 4> {
+	K x;
+	K y;
+	K z;
+	K w;
+	typedef VecType<K, 4> Type;
+};
+
+//
+// Specialize certain built-in vector types
+//
+
+#define B40C_DEFINE_VECTOR_TYPE(base_type,short_type)                           \
+  template<> struct VecType<base_type, 1> { typedef short_type##1 Type; };      \
+  template<> struct VecType<base_type, 2> { typedef short_type##2 Type; };      \
+  template<> struct VecType<base_type, 4> { typedef short_type##4 Type; };     
+
+B40C_DEFINE_VECTOR_TYPE(char,               char)
+B40C_DEFINE_VECTOR_TYPE(short,              short)
+B40C_DEFINE_VECTOR_TYPE(int,                int)
+B40C_DEFINE_VECTOR_TYPE(long,               long)
+B40C_DEFINE_VECTOR_TYPE(long long,          longlong)
+B40C_DEFINE_VECTOR_TYPE(unsigned char,      uchar)
+B40C_DEFINE_VECTOR_TYPE(unsigned short,     ushort)
+B40C_DEFINE_VECTOR_TYPE(unsigned int,       uint)
+B40C_DEFINE_VECTOR_TYPE(unsigned long,      ulong)
+B40C_DEFINE_VECTOR_TYPE(unsigned long long, ulonglong)
+B40C_DEFINE_VECTOR_TYPE(float,              float)
+B40C_DEFINE_VECTOR_TYPE(double,             double)
+
+#undef B40C_DEFINE_VECTOR_TYPE
+
+} // end namespace b40c_thrust
+} // end namespace detail
+} // end namespace detail
+} // end namespace cuda
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/cuda/detail/detail/balanced_path.h b/compat/thrust/system/cuda/detail/detail/balanced_path.h
new file mode 100644
index 0000000000..51e4f5b767
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/detail/balanced_path.h
@@ -0,0 +1,156 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/pair.h>
+#include <thrust/detail/minmax.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+namespace detail
+{
+namespace balanced_path_detail
+{
+
+template<bool UpperBound, typename IntT, typename It, typename T, typename Comp>
+__host__ __device__ void BinarySearchIteration(It data, int& begin, int& end,
+	T key, int shift, Comp comp) {
+
+	IntT scale = (1<< shift) - 1;
+	int mid = (int)((begin + scale * end)>> shift);
+
+	T key2 = data[mid];
+	bool pred = UpperBound ? !comp(key, key2) : comp(key2, key);
+	if(pred) begin = (int)mid + 1;
+	else end = mid;
+}
+
+template<bool UpperBound, typename T, typename It, typename Comp>
+__host__ __device__ int BinarySearch(It data, int count, T key, Comp comp) {
+	int begin = 0;
+	int end = count;
+	while(begin < end) 
+		BinarySearchIteration<UpperBound, int>(data, begin, end, key, 1, comp);
+	return begin;
+}
+
+template<bool UpperBound, typename IntT, typename T, typename It, typename Comp>
+__host__ __device__ int BiasedBinarySearch(It data, int count, T key, 
+	IntT levels, Comp comp) {
+	int begin = 0;
+	int end = count;
+
+	if(levels >= 4 && begin < end)
+		BinarySearchIteration<UpperBound, IntT>(data, begin, end, key, 9, comp);
+	if(levels >= 3 && begin < end)
+		BinarySearchIteration<UpperBound, IntT>(data, begin, end, key, 7, comp);
+	if(levels >= 2 && begin < end)
+		BinarySearchIteration<UpperBound, IntT>(data, begin, end, key, 5, comp);
+	if(levels >= 1 && begin < end)
+		BinarySearchIteration<UpperBound, IntT>(data, begin, end, key, 4, comp);
+
+	while(begin < end)
+		BinarySearchIteration<UpperBound, IntT>(data, begin, end, key, 1, comp);
+	return begin;
+}
+
+template<bool UpperBound, typename It1, typename It2, typename Comp>
+__host__ __device__ int MergePath(It1 a, int aCount, It2 b, int bCount, int diag, Comp comp)
+{
+  typedef typename thrust::iterator_traits<It1>::value_type T;
+  
+  int begin = thrust::max(0, diag - bCount);
+  int end   = thrust::min(diag, aCount);
+  
+  while(begin < end) 
+  {
+    int mid = (begin + end)>> 1;
+    T aKey = a[mid];
+    T bKey = b[diag - 1 - mid];
+    bool pred = UpperBound ? comp(aKey, bKey) : !comp(bKey, aKey);
+    if(pred) begin = mid + 1;
+    else end = mid;
+  }
+  return begin;
+}
+
+
+} // end namespace balanced_path_detail
+
+
+template<typename RandomAccessIterator1, typename Size1, typename RandomAccessIterator2, typename Size2, typename Compare>
+__host__ __device__
+thrust::pair<Size1,Size1>
+  balanced_path(RandomAccessIterator1 first1, Size1 n1,
+                RandomAccessIterator2 first2, Size1 n2,
+                Size1 diag,
+                Size2 levels,
+                Compare comp)
+{
+  typedef typename thrust::iterator_traits<RandomAccessIterator1>::value_type T;
+
+  Size1 aIndex = balanced_path_detail::MergePath<false>(first1, n1, first2, n2, diag, comp);
+  Size1 bIndex = diag - aIndex;
+  
+  bool star = false;
+  if(bIndex < n2)
+  {
+    T x = first2[bIndex];
+    
+    // Search for the beginning of the duplicate run in both A and B.
+    Size1 aStart = balanced_path_detail::BiasedBinarySearch<false>(first1, aIndex, x, levels, comp);
+    Size1 bStart = balanced_path_detail::BiasedBinarySearch<false>(first2, bIndex, x, levels, comp);
+    
+    // The distance between x's merge path and its lower_bound is its rank.
+    // We add up the a and b ranks and evenly distribute them to
+    // get a stairstep path.
+    Size1 aRun = aIndex - aStart;
+    Size1 bRun = bIndex - bStart;
+    Size1 xCount = aRun + bRun;
+    
+    // Attempt to advance b and regress a.
+    Size1 bAdvance = thrust::max(xCount >> 1, xCount - aRun);
+    Size1 bEnd     = thrust::min<Size1>(n2, bStart + bAdvance + 1);
+    Size1 bRunEnd  = balanced_path_detail::BinarySearch<true>(first2 + bIndex, bEnd - bIndex, x, comp) + bIndex;
+    bRun = bRunEnd - bStart;
+    
+    bAdvance = thrust::min(bAdvance, bRun);
+    Size1 aAdvance = xCount - bAdvance;
+    
+    bool roundUp = (aAdvance == bAdvance + 1) && (bAdvance < bRun);
+    aIndex = aStart + aAdvance;
+    
+    if(roundUp) star = true;
+  }
+
+  return thrust::make_pair(aIndex, (diag - aIndex) + star);
+}
+
+
+} // end namespace detail
+} // end namespace detail
+} // end namespace cuda
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/cuda/detail/detail/cached_temporary_allocator.h b/compat/thrust/system/cuda/detail/detail/cached_temporary_allocator.h
new file mode 100644
index 0000000000..2bbd658456
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/detail/cached_temporary_allocator.h
@@ -0,0 +1,156 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/raw_pointer_cast.h>
+#include <thrust/detail/allocator/allocator_traits.h>
+#include <thrust/detail/allocator/temporary_allocator.h>
+#include <thrust/pair.h>
+#include <map>
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+namespace detail
+{
+
+
+template<typename DerivedPolicy, template<typename> class BasePolicy>
+  class cached_temporary_allocator
+    : public BasePolicy<cached_temporary_allocator<DerivedPolicy,BasePolicy> >
+{
+  private:
+    typedef thrust::detail::temporary_allocator<char,DerivedPolicy> base_allocator_type;
+    typedef thrust::detail::allocator_traits<base_allocator_type>   traits;
+    typedef typename traits::pointer                                  allocator_pointer;
+    typedef std::multimap<std::ptrdiff_t, void*>                      free_blocks_type;
+    typedef std::map<void *, std::ptrdiff_t>                          allocated_blocks_type;
+
+    base_allocator_type   m_base_allocator;
+    free_blocks_type      free_blocks;
+    allocated_blocks_type allocated_blocks;
+
+    void free_all()
+    {
+      // deallocate all outstanding blocks in both lists
+      for(free_blocks_type::iterator i = free_blocks.begin();
+          i != free_blocks.end();
+          ++i)
+      {
+        // transform the pointer to allocator_pointer before calling deallocate
+        traits::deallocate(m_base_allocator, allocator_pointer(reinterpret_cast<char*>(i->second)), i->first);
+      }
+
+      for(allocated_blocks_type::iterator i = allocated_blocks.begin();
+          i != allocated_blocks.end();
+          ++i)
+      {
+        // transform the pointer to allocator_pointer before calling deallocate
+        traits::deallocate(m_base_allocator, allocator_pointer(reinterpret_cast<char*>(i->first)), i->second);
+      }
+    }
+
+  public:
+    cached_temporary_allocator(thrust::execution_policy<DerivedPolicy> &system)
+      : m_base_allocator(system)
+    {}
+
+    ~cached_temporary_allocator()
+    {
+      // free all allocations when cached_allocator goes out of scope
+      free_all();
+    }
+
+    void *allocate(std::ptrdiff_t num_bytes)
+    {
+      void *result = 0;
+
+      // search the cache for a free block
+      free_blocks_type::iterator free_block = free_blocks.find(num_bytes);
+
+      if(free_block != free_blocks.end())
+      {
+        // get the pointer
+        result = free_block->second;
+
+        // erase from the free_blocks map
+        free_blocks.erase(free_block);
+      }
+      else
+      {
+        // no allocation of the right size exists
+        // create a new one with m_base_allocator
+        // allocate memory and convert to raw pointer
+        result = thrust::raw_pointer_cast(traits::allocate(m_base_allocator, num_bytes));
+      }
+
+      // insert the allocated pointer into the allocated_blocks map
+      allocated_blocks.insert(std::make_pair(result, num_bytes));
+
+      return result;
+    }
+
+    void deallocate(void *ptr)
+    {
+      // erase the allocated block from the allocated blocks map
+      allocated_blocks_type::iterator iter = allocated_blocks.find(ptr);
+      std::ptrdiff_t num_bytes = iter->second;
+      allocated_blocks.erase(iter);
+
+      // insert the block into the free blocks map
+      free_blocks.insert(std::make_pair(num_bytes, ptr));
+    }
+};
+
+
+// overload get_temporary_buffer on cached_temporary_allocator
+// note that we take a reference to cached_temporary_allocator
+template<typename T, typename DerivedPolicy, template<typename> class BasePolicy>
+  thrust::pair<T*, std::ptrdiff_t>
+    get_temporary_buffer(cached_temporary_allocator<DerivedPolicy,BasePolicy> &alloc, std::ptrdiff_t n)
+{
+  // ask the allocator for sizeof(T) * n bytes
+  T* result = reinterpret_cast<T*>(alloc.allocate(sizeof(T) * n));
+
+  // return the pointer and the number of elements allocated
+  return thrust::make_pair(result,n);
+}
+
+
+// overload return_temporary_buffer on cached_temporary_allocator
+// an overloaded return_temporary_buffer should always accompany
+// an overloaded get_temporary_buffer
+template<typename Pointer, typename DerivedPolicy, template<typename> class BasePolicy>
+  void return_temporary_buffer(cached_temporary_allocator<DerivedPolicy,BasePolicy> &alloc, Pointer p)
+{
+  // return the pointer to the allocator
+  alloc.deallocate(thrust::raw_pointer_cast(p));
+}
+
+
+} // end detail
+} // end detail
+} // end cuda
+} // end system
+} // end thrust
+
diff --git a/compat/thrust/system/cuda/detail/detail/fast_scan.h b/compat/thrust/system/cuda/detail/detail/fast_scan.h
new file mode 100644
index 0000000000..d095a4a2db
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/detail/fast_scan.h
@@ -0,0 +1,67 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file fast_scan.h
+ *  \brief A fast scan for primitive types.
+ */
+
+#pragma once
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+namespace detail
+{
+namespace fast_scan
+{
+
+template <typename ExecutionPolicy,
+          typename InputIterator,
+          typename OutputIterator,
+          typename BinaryFunction>
+OutputIterator inclusive_scan(execution_policy<ExecutionPolicy> &exec,
+                              InputIterator first,
+                              InputIterator last,
+                              OutputIterator output,
+                              BinaryFunction binary_op);
+
+template <typename ExecutionPolicy,
+          typename InputIterator,
+          typename OutputIterator,
+          typename T,
+          typename BinaryFunction>
+OutputIterator exclusive_scan(execution_policy<ExecutionPolicy> &exec,
+                              InputIterator first,
+                              InputIterator last,
+                              OutputIterator output,
+                              const T init,
+                              BinaryFunction binary_op);
+
+} // end namespace fast_scan
+} // end namespace detail
+} // end namespace detail
+} // end namespace cuda
+} // end namespace system
+} // end namespace thrust
+
+#include "fast_scan.inl"
+
diff --git a/compat/thrust/system/cuda/detail/detail/fast_scan.inl b/compat/thrust/system/cuda/detail/detail/fast_scan.inl
new file mode 100644
index 0000000000..b02763d8a9
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/detail/fast_scan.inl
@@ -0,0 +1,753 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+#include <thrust/detail/config.h>
+
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/system/detail/generic/select_system.h>
+#include <thrust/detail/temporary_array.h>
+
+#include <thrust/detail/type_traits.h>
+#include <thrust/detail/type_traits/function_traits.h>
+#include <thrust/detail/type_traits/iterator/is_output_iterator.h>
+
+#include <thrust/system/cuda/detail/reduce_intervals.h>
+#include <thrust/system/cuda/detail/synchronize.h>
+#include <thrust/system/cuda/detail/default_decomposition.h>
+#include <thrust/system/cuda/detail/detail/launch_closure.h>
+#include <thrust/system/cuda/detail/detail/uninitialized.h>
+#include <thrust/detail/raw_pointer_cast.h>
+
+__THRUST_DISABLE_MSVC_POSSIBLE_LOSS_OF_DATA_WARNING_BEGIN
+
+
+namespace thrust
+{
+namespace detail
+{
+
+// forward declaration of temporary_array
+template<typename,typename> class temporary_array;
+
+} // end detail
+
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+namespace detail
+{
+namespace fast_scan
+{
+namespace fast_scan_detail
+{
+
+
+// TODO tune this
+template <typename ValueType>
+struct inclusive_scan_block_size
+{
+  private:
+  static const unsigned int max_memory         = 16384 - 256 - 2 * sizeof(ValueType);
+  static const unsigned int max_block_size     = max_memory / sizeof(ValueType);
+  static const unsigned int default_block_size = 7 * 32;
+  static const unsigned int block_size         = (max_block_size < default_block_size) ? max_block_size : default_block_size;
+
+  public:
+  static const unsigned int pass1 = block_size;
+  static const unsigned int pass2 = block_size;
+  static const unsigned int pass3 = block_size;
+};
+
+// TODO tune this
+template <typename ValueType>
+struct exclusive_scan_block_size
+{
+  private:
+  static const unsigned int max_memory         = 16384 - 256 - 2 * sizeof(ValueType);
+  static const unsigned int max_block_size     = max_memory / sizeof(ValueType);
+  static const unsigned int default_block_size = 5 * 32;
+  static const unsigned int block_size         = (max_block_size < default_block_size) ? max_block_size : default_block_size;
+
+  public:
+  static const unsigned int pass1 = block_size;
+  static const unsigned int pass2 = block_size;
+  static const unsigned int pass3 = block_size;
+};
+
+
+template <unsigned int CTA_SIZE,
+          typename Context,
+          typename SharedArray,
+          typename BinaryFunction>
+__device__ __thrust_forceinline__
+void scan_block(Context context, SharedArray array, BinaryFunction binary_op)
+{
+    typedef typename thrust::iterator_value<SharedArray>::type T;
+
+    T val = array[context.thread_index()];
+
+    if (CTA_SIZE >    1) { if(context.thread_index() >=    1) { T tmp = array[context.thread_index() -    1]; val = binary_op(tmp, val); } context.barrier(); array[context.thread_index()] = val; context.barrier(); }
+    if (CTA_SIZE >    2) { if(context.thread_index() >=    2) { T tmp = array[context.thread_index() -    2]; val = binary_op(tmp, val); } context.barrier(); array[context.thread_index()] = val; context.barrier(); }
+    if (CTA_SIZE >    4) { if(context.thread_index() >=    4) { T tmp = array[context.thread_index() -    4]; val = binary_op(tmp, val); } context.barrier(); array[context.thread_index()] = val; context.barrier(); }
+    if (CTA_SIZE >    8) { if(context.thread_index() >=    8) { T tmp = array[context.thread_index() -    8]; val = binary_op(tmp, val); } context.barrier(); array[context.thread_index()] = val; context.barrier(); }
+    if (CTA_SIZE >   16) { if(context.thread_index() >=   16) { T tmp = array[context.thread_index() -   16]; val = binary_op(tmp, val); } context.barrier(); array[context.thread_index()] = val; context.barrier(); }
+    if (CTA_SIZE >   32) { if(context.thread_index() >=   32) { T tmp = array[context.thread_index() -   32]; val = binary_op(tmp, val); } context.barrier(); array[context.thread_index()] = val; context.barrier(); }
+    if (CTA_SIZE >   64) { if(context.thread_index() >=   64) { T tmp = array[context.thread_index() -   64]; val = binary_op(tmp, val); } context.barrier(); array[context.thread_index()] = val; context.barrier(); }
+    if (CTA_SIZE >  128) { if(context.thread_index() >=  128) { T tmp = array[context.thread_index() -  128]; val = binary_op(tmp, val); } context.barrier(); array[context.thread_index()] = val; context.barrier(); }
+    if (CTA_SIZE >  256) { if(context.thread_index() >=  256) { T tmp = array[context.thread_index() -  256]; val = binary_op(tmp, val); } context.barrier(); array[context.thread_index()] = val; context.barrier(); }  
+    if (CTA_SIZE >  512) { if(context.thread_index() >=  512) { T tmp = array[context.thread_index() -  512]; val = binary_op(tmp, val); } context.barrier(); array[context.thread_index()] = val; context.barrier(); }  
+    if (CTA_SIZE > 1024) { if(context.thread_index() >= 1024) { T tmp = array[context.thread_index() - 1024]; val = binary_op(tmp, val); } context.barrier(); array[context.thread_index()] = val; context.barrier(); }  
+}
+
+template <unsigned int CTA_SIZE,
+          typename Context,
+          typename SharedArray,
+          typename BinaryFunction>
+__device__ __thrust_forceinline__
+void scan_block_n(Context context, SharedArray array, const unsigned int n, BinaryFunction binary_op)
+{
+    typedef typename thrust::iterator_value<SharedArray>::type T;
+
+    T val = array[context.thread_index()];
+
+    if (CTA_SIZE >    1) { if(context.thread_index() < n && context.thread_index() >=    1) { T tmp = array[context.thread_index() -    1]; val = binary_op(tmp, val); } context.barrier(); array[context.thread_index()] = val; context.barrier(); }
+    if (CTA_SIZE >    2) { if(context.thread_index() < n && context.thread_index() >=    2) { T tmp = array[context.thread_index() -    2]; val = binary_op(tmp, val); } context.barrier(); array[context.thread_index()] = val; context.barrier(); }
+    if (CTA_SIZE >    4) { if(context.thread_index() < n && context.thread_index() >=    4) { T tmp = array[context.thread_index() -    4]; val = binary_op(tmp, val); } context.barrier(); array[context.thread_index()] = val; context.barrier(); }
+    if (CTA_SIZE >    8) { if(context.thread_index() < n && context.thread_index() >=    8) { T tmp = array[context.thread_index() -    8]; val = binary_op(tmp, val); } context.barrier(); array[context.thread_index()] = val; context.barrier(); }
+    if (CTA_SIZE >   16) { if(context.thread_index() < n && context.thread_index() >=   16) { T tmp = array[context.thread_index() -   16]; val = binary_op(tmp, val); } context.barrier(); array[context.thread_index()] = val; context.barrier(); }
+    if (CTA_SIZE >   32) { if(context.thread_index() < n && context.thread_index() >=   32) { T tmp = array[context.thread_index() -   32]; val = binary_op(tmp, val); } context.barrier(); array[context.thread_index()] = val; context.barrier(); }
+    if (CTA_SIZE >   64) { if(context.thread_index() < n && context.thread_index() >=   64) { T tmp = array[context.thread_index() -   64]; val = binary_op(tmp, val); } context.barrier(); array[context.thread_index()] = val; context.barrier(); }
+    if (CTA_SIZE >  128) { if(context.thread_index() < n && context.thread_index() >=  128) { T tmp = array[context.thread_index() -  128]; val = binary_op(tmp, val); } context.barrier(); array[context.thread_index()] = val; context.barrier(); }
+    if (CTA_SIZE >  256) { if(context.thread_index() < n && context.thread_index() >=  256) { T tmp = array[context.thread_index() -  256]; val = binary_op(tmp, val); } context.barrier(); array[context.thread_index()] = val; context.barrier(); }
+    if (CTA_SIZE >  512) { if(context.thread_index() < n && context.thread_index() >=  512) { T tmp = array[context.thread_index() -  512]; val = binary_op(tmp, val); } context.barrier(); array[context.thread_index()] = val; context.barrier(); }
+    if (CTA_SIZE > 1024) { if(context.thread_index() < n && context.thread_index() >= 1024) { T tmp = array[context.thread_index() - 1024]; val = binary_op(tmp, val); } context.barrier(); array[context.thread_index()] = val; context.barrier(); }
+}
+
+template <unsigned int CTA_SIZE,
+          unsigned int K,
+          bool FullBlock,
+          typename Context,
+          typename InputIterator,
+          typename ValueType>
+__device__ __thrust_forceinline__
+void load_block(Context context,
+                const unsigned int n,
+                InputIterator input,
+                ValueType (&sdata)[K][CTA_SIZE + 1])
+{
+  for(unsigned int k = 0; k < K; k++)
+  {
+    const unsigned int offset = k*CTA_SIZE + context.thread_index();
+
+    if (FullBlock || offset < n)
+    {
+      InputIterator temp = input + offset;
+      sdata[offset % K][offset / K] = *temp;
+    }
+  }
+
+  context.barrier();
+}
+
+template <unsigned int CTA_SIZE,
+          unsigned int K,
+          bool Inclusive,
+          bool FullBlock,
+          typename Context,
+          typename OutputIterator,
+          typename ValueType>
+__device__ __thrust_forceinline__
+void store_block(Context context,
+                 const unsigned int n,
+                 OutputIterator output,
+                 ValueType (&sdata)[K][CTA_SIZE + 1],
+                 ValueType& carry)
+{
+  if (Inclusive)
+  {
+    for(unsigned int k = 0; k < K; k++)
+    {
+      const unsigned int offset = k*CTA_SIZE + context.thread_index();
+
+      if (FullBlock || offset < n)
+      {
+        OutputIterator temp = output + offset;
+        *temp = sdata[offset % K][offset / K];
+      }
+    }   
+  }
+  else
+  {
+    for(unsigned int k = 0; k < K; k++)
+    {
+      const unsigned int offset = k*CTA_SIZE + context.thread_index();
+
+      if (FullBlock || offset < n)
+      {
+        OutputIterator temp = output + offset;
+        *temp = (offset == 0) ? carry : sdata[(offset - 1) % K][(offset - 1) / K];
+      }
+    }   
+  }
+}
+
+template <unsigned int CTA_SIZE,
+          unsigned int K,
+          bool FullBlock,
+          typename Context,
+          typename InputIterator,
+          typename BinaryFunction,
+          typename ValueType>
+__device__ __thrust_forceinline__
+void upsweep_body(Context context,
+                  const unsigned int n,
+                  const bool carry_in,
+                  InputIterator input,
+                  BinaryFunction binary_op,
+                  ValueType (&sdata)[K][CTA_SIZE + 1],
+                  ValueType& carry)
+{
+  // read data
+  load_block<CTA_SIZE,K,FullBlock>(context, n, input, sdata);
+ 
+  // copy into local array
+  ValueType ldata[K];
+  for (unsigned int k = 0; k < K; k++)
+    ldata[k] = sdata[k][context.thread_index()];
+
+  // carry in
+  if (context.thread_index() == 0 && carry_in)
+  {
+    // XXX WAR sm_10 issue
+    ValueType tmp = carry;
+    ldata[0] = binary_op(tmp, ldata[0]);
+  }
+
+  // scan local values
+  for(unsigned int k = 1; k < K; k++)
+  {
+    const unsigned int offset = K * context.thread_index() + k;
+
+    if (FullBlock || offset < n)
+      ldata[k] = binary_op(ldata[k-1],ldata[k]);
+  }
+
+  sdata[K - 1][context.thread_index()] = ldata[K - 1];
+
+  context.barrier();
+
+  // second level scan
+  if (FullBlock && sizeof(ValueType) > 1) // TODO investigate why this WAR is necessary
+    scan_block<CTA_SIZE>(context, sdata[K - 1], binary_op); 
+  else
+    scan_block_n<CTA_SIZE>(context, sdata[K - 1], n / K, binary_op);
+
+  // store carry out
+  if (FullBlock)
+  {
+     if (context.thread_index() == CTA_SIZE - 1)
+        carry = sdata[K - 1][context.thread_index()];
+  }
+  else
+  {
+    if (context.thread_index() == (n - 1) / K)
+    {
+      ValueType sum;
+
+      for (unsigned int k = 0; k < K; k++)
+          if ((n - 1) % K == k)
+              sum = ldata[k];
+
+      if (context.thread_index() > 0)
+      {
+        // WAR sm_10 issue
+        ValueType tmp = sdata[K - 1][context.thread_index() - 1];
+        sum = binary_op(tmp, sum);
+      }
+
+      carry = sum;
+    }
+  }
+
+  context.barrier();
+}
+
+template <unsigned int CTA_SIZE,
+          unsigned int K,
+          bool Inclusive,
+          bool FullBlock,
+          typename Context,
+          typename InputIterator,
+          typename OutputIterator,
+          typename BinaryFunction,
+          typename ValueType>
+__device__ __thrust_forceinline__
+void scan_body(Context context,
+               const unsigned int n,
+               const bool carry_in,
+               InputIterator input,
+               OutputIterator output,
+               BinaryFunction binary_op,
+               ValueType (&sdata)[K][CTA_SIZE + 1],
+               ValueType& carry)
+{
+  // read data
+  load_block<CTA_SIZE,K,FullBlock>(context, n, input, sdata);
+
+  // copy into local array
+  ValueType ldata[K];
+  for (unsigned int k = 0; k < K; k++)
+    ldata[k] = sdata[k][context.thread_index()];
+
+  // carry in
+  if (context.thread_index() == 0 && carry_in)
+  {
+    // XXX WAR sm_10 issue
+    ValueType tmp = carry;
+    ldata[0] = binary_op(tmp, ldata[0]);
+  }
+
+  // scan local values
+  for(unsigned int k = 1; k < K; k++)
+  {
+    const unsigned int offset = K * context.thread_index() + k;
+
+    if (FullBlock || offset < n)
+      ldata[k] = binary_op(ldata[k-1],ldata[k]);
+  }
+
+  sdata[K - 1][context.thread_index()] = ldata[K - 1];
+
+  context.barrier();
+
+  // second level scan
+  if (FullBlock)
+    scan_block<CTA_SIZE>(context, sdata[K - 1], binary_op);
+  else
+    scan_block_n<CTA_SIZE>(context, sdata[K - 1], n / K, binary_op);
+  
+  // update local values
+  if (context.thread_index() > 0)
+  {
+    ValueType left = sdata[K - 1][context.thread_index() - 1];
+
+    for(unsigned int k = 0; k < K; k++)
+    {
+      const unsigned int offset = K * context.thread_index() + k;
+
+      if (FullBlock || offset < n)
+        ldata[k] = binary_op(left, ldata[k]);
+    }
+  }
+
+  for (unsigned int k = 0; k < K; k++)
+    sdata[k][context.thread_index()] = ldata[k];
+
+  context.barrier();
+
+  // write data
+  store_block<CTA_SIZE, K, Inclusive, FullBlock>(context, n, output, sdata, carry);
+  
+  // store carry out
+  if (context.thread_index() == 0)
+  {
+    if (FullBlock)
+      carry = sdata[K - 1][CTA_SIZE - 1];
+    else
+      carry = sdata[(n - 1) % K][(n - 1) / K]; // note: this must come after the local update
+  }
+
+  context.barrier();
+}
+
+template <typename InputIterator,
+          typename ValueType,
+          typename BinaryFunction,
+          typename Decomposition,
+          typename Context>
+struct upsweep_intervals_closure
+{
+  InputIterator  input;
+  ValueType *    block_results; // TODO change this to ValueIterator
+  BinaryFunction binary_op;
+  Decomposition  decomp;
+  Context        context;
+  
+  typedef Context context_type;
+
+  upsweep_intervals_closure(InputIterator input,
+                            ValueType * block_results,
+                            BinaryFunction binary_op,
+                            Decomposition decomp,
+                            Context context = Context())
+    : input(input), block_results(block_results), binary_op(binary_op), decomp(decomp), context(context) {}
+
+  __device__ __thrust_forceinline__
+  void operator()(void)
+  {
+    typedef typename Decomposition::index_type  IndexType;
+
+    const unsigned int CTA_SIZE = context_type::ThreadsPerBlock::value;
+
+#if __CUDA_ARCH__ >= 200
+    const unsigned int SMEM = (48 * 1024);
+#else
+    const unsigned int SMEM = (16 * 1024) - 256;
+#endif
+    const unsigned int MAX_K = ((SMEM - 1 * sizeof(ValueType)) / (sizeof(ValueType) * (CTA_SIZE + 1)));
+    const unsigned int K     = (MAX_K < 6) ? MAX_K : 6;
+
+    __shared__ uninitialized<ValueType[K][CTA_SIZE + 1]> sdata; // padded to avoid bank conflicts
+    
+    __shared__ uninitialized<ValueType> carry; // storage for carry out
+    if(context.thread_index() == 0) carry.construct();
+    
+    context.barrier();
+    
+    thrust::system::detail::internal::index_range<IndexType> interval = decomp[context.block_index()];
+
+    IndexType base = interval.begin();
+
+    input += base;
+
+    const unsigned int unit_size = K * CTA_SIZE;
+
+    bool carry_in = false;
+
+    // process full units
+    while (base + unit_size <= interval.end())
+    {
+      const unsigned int n = unit_size;
+      upsweep_body<CTA_SIZE,K,true>(context, n, carry_in, input, binary_op, sdata.get(), carry.get());
+      base   += unit_size;
+      input  += unit_size;
+      carry_in = true;
+    }
+
+    // process partially full unit at end of input (if necessary)
+    if (base < interval.end())
+    {
+      const unsigned int n = interval.end() - base;
+      upsweep_body<CTA_SIZE,K,false>(context, n, carry_in, input, binary_op, sdata.get(), carry.get());
+    }
+
+    // write interval sum
+    if (context.thread_index() == 0)
+      block_results[context.block_index()] = carry;
+  }
+};
+
+
+template <bool Inclusive,
+          typename InputIterator,
+          typename OutputIterator,
+          typename ValueType,
+          typename BinaryFunction,
+          typename Decomposition,
+          typename Context>
+struct downsweep_intervals_closure
+{
+  InputIterator  input;
+  OutputIterator output;
+  ValueType *    block_results;
+  BinaryFunction binary_op;
+  Decomposition  decomp;
+  Context        context;
+
+  typedef Context context_type;
+
+  downsweep_intervals_closure(InputIterator input,
+                              OutputIterator output,
+                              ValueType * block_results,
+                              BinaryFunction binary_op,
+                              Decomposition decomp,
+                              Context context = Context())
+    : input(input), output(output), block_results(block_results), binary_op(binary_op), decomp(decomp), context(context) {}
+
+  __device__ __thrust_forceinline__
+  void operator()(void)
+  {
+    typedef typename Decomposition::index_type IndexType;
+    
+    const unsigned int CTA_SIZE = context_type::ThreadsPerBlock::value;
+
+#if __CUDA_ARCH__ >= 200
+    const unsigned int SMEM = (48 * 1024);
+#else
+    const unsigned int SMEM = (16 * 1024) - 256;
+#endif
+    const unsigned int MAX_K = ((SMEM - 1 * sizeof(ValueType))/ (sizeof(ValueType) * (CTA_SIZE + 1)));
+    const unsigned int K     = (MAX_K < 6) ? MAX_K : 6;
+
+    __shared__ uninitialized<ValueType[K][CTA_SIZE + 1]> sdata;  // padded to avoid bank conflicts
+    
+    __shared__ uninitialized<ValueType> carry; // storage for carry in and carry out
+    if(context.thread_index() == 0) carry.construct();
+
+    context.barrier();
+
+    thrust::system::detail::internal::index_range<IndexType> interval = decomp[context.block_index()];
+
+    IndexType base = interval.begin();
+
+    input  += base;
+    output += base;
+
+    const unsigned int unit_size = K * CTA_SIZE;
+
+    bool carry_in  = (Inclusive && context.block_index() == 0) ? false : true;
+
+    if (carry_in)
+    {
+      if (context.thread_index() == 0)
+        carry = block_results[context.block_index()];
+      context.barrier();
+    }
+
+    // process full units
+    while (base + unit_size <= interval.end())
+    {
+      const unsigned int n = unit_size;
+      scan_body<CTA_SIZE,K,Inclusive,true>(context, n, carry_in, input, output, binary_op, sdata.get(), carry.get());
+      base   += K * CTA_SIZE;
+      input  += K * CTA_SIZE;
+      output += K * CTA_SIZE;
+      carry_in = true;
+    }
+
+    // process partially full unit at end of input (if necessary)
+    if (base < interval.end())
+    {
+      const unsigned int n = interval.end() - base;
+      scan_body<CTA_SIZE,K,Inclusive,false>(context, n, carry_in, input, output, binary_op, sdata.get(), carry.get());
+    }
+  }
+};
+
+
+} // end namespace fast_scan_detail
+
+
+template <typename DerivedPolicy,
+          typename InputIterator,
+          typename OutputIterator,
+          typename BinaryFunction>
+OutputIterator inclusive_scan(execution_policy<DerivedPolicy> &exec,
+                              InputIterator first,
+                              InputIterator last,
+                              OutputIterator output,
+                              BinaryFunction binary_op)
+{
+  using namespace fast_scan_detail;
+
+  // the pseudocode for deducing the type of the temporary used below:
+  // 
+  // if BinaryFunction is AdaptableBinaryFunction
+  //   TemporaryType = AdaptableBinaryFunction::result_type
+  // else if OutputIterator is a "pure" output iterator
+  //   TemporaryType = InputIterator::value_type
+  // else
+  //   TemporaryType = OutputIterator::value_type
+  //
+  // XXX upon c++0x, TemporaryType needs to be:
+  // result_of<BinaryFunction>::type
+
+  typedef typename thrust::detail::eval_if<
+    thrust::detail::has_result_type<BinaryFunction>::value,
+    thrust::detail::result_type<BinaryFunction>,
+    thrust::detail::eval_if<
+      thrust::detail::is_output_iterator<OutputIterator>::value,
+      thrust::iterator_value<InputIterator>,
+      thrust::iterator_value<OutputIterator>
+    >
+  >::type ValueType;
+
+  typedef unsigned int                                                       IndexType;
+  typedef thrust::system::detail::internal::uniform_decomposition<IndexType> Decomposition;
+  typedef thrust::detail::temporary_array<ValueType,DerivedPolicy>           ValueArray;
+
+  if (first == last)
+      return output;
+
+  Decomposition decomp = thrust::system::cuda::detail::default_decomposition<IndexType>(last - first);
+
+  ValueArray block_results(exec, decomp.size());
+  
+  // compute sum over each interval
+  if (thrust::detail::is_commutative<BinaryFunction>::value)
+  {
+    // use reduce_intervals for commutative operators
+    thrust::system::cuda::detail::reduce_intervals(exec, first, block_results.begin(), binary_op, decomp);
+  }
+  else
+  {
+    const static unsigned int ThreadsPerBlock = inclusive_scan_block_size<ValueType>::pass1;
+    typedef detail::statically_blocked_thread_array<ThreadsPerBlock> Context;
+
+    typedef upsweep_intervals_closure<InputIterator,ValueType,BinaryFunction,Decomposition,Context> Closure;
+    Closure closure(first,
+                    thrust::raw_pointer_cast(&block_results[0]),
+                    binary_op,
+                    decomp);
+    detail::launch_closure(closure, decomp.size(), ThreadsPerBlock);
+  }
+
+  // second level inclusive scan of per-block results
+  {
+    const static unsigned int ThreadsPerBlock = inclusive_scan_block_size<ValueType>::pass2;
+    typedef detail::statically_blocked_thread_array<ThreadsPerBlock> Context;
+
+    typedef downsweep_intervals_closure<true,ValueType*,ValueType*,ValueType,BinaryFunction,Decomposition,Context> Closure;
+    Closure closure(thrust::raw_pointer_cast(&block_results[0]),
+                    thrust::raw_pointer_cast(&block_results[0]),
+                    thrust::raw_pointer_cast(&block_results[0]), // not used
+                    binary_op,
+                    Decomposition(decomp.size(), 1, 1));
+    detail::launch_closure(closure, 1, ThreadsPerBlock);
+  }
+  
+  // update intervals with result of second level scan
+  {
+    const static unsigned int ThreadsPerBlock = inclusive_scan_block_size<ValueType>::pass3;
+    typedef detail::statically_blocked_thread_array<ThreadsPerBlock> Context;
+
+    typedef downsweep_intervals_closure<true,InputIterator,OutputIterator,ValueType,BinaryFunction,Decomposition,Context> Closure;
+    Closure closure(first,
+                    output,
+                    thrust::raw_pointer_cast(&block_results[0]) - 1, // shift block results
+                    binary_op,
+                    decomp);
+    detail::launch_closure(closure, decomp.size(), ThreadsPerBlock);
+  }
+  
+  return output + (last - first);
+}
+
+
+template <typename DerivedPolicy,
+          typename InputIterator,
+          typename OutputIterator,
+          typename T,
+          typename BinaryFunction>
+OutputIterator exclusive_scan(execution_policy<DerivedPolicy> &exec,
+                              InputIterator first,
+                              InputIterator last,
+                              OutputIterator output,
+                              const T init,
+                              BinaryFunction binary_op)
+{
+  using namespace fast_scan_detail;
+
+  // the pseudocode for deducing the type of the temporary used below:
+  // 
+  // if BinaryFunction is AdaptableBinaryFunction
+  //   TemporaryType = AdaptableBinaryFunction::result_type
+  // else if OutputIterator is a "pure" output iterator
+  //   TemporaryType = InputIterator::value_type
+  // else
+  //   TemporaryType = OutputIterator::value_type
+  //
+  // XXX upon c++0x, TemporaryType needs to be:
+  // result_of<BinaryFunction>::type
+
+  typedef typename thrust::detail::eval_if<
+    thrust::detail::has_result_type<BinaryFunction>::value,
+    thrust::detail::result_type<BinaryFunction>,
+    thrust::detail::eval_if<
+      thrust::detail::is_output_iterator<OutputIterator>::value,
+      thrust::iterator_value<InputIterator>,
+      thrust::iterator_value<OutputIterator>
+    >
+  >::type ValueType;
+
+  typedef unsigned int                                                       IndexType;
+  typedef thrust::system::detail::internal::uniform_decomposition<IndexType> Decomposition;
+  typedef thrust::detail::temporary_array<ValueType,DerivedPolicy>           ValueArray;
+
+  if (first == last)
+      return output;
+
+  Decomposition decomp = thrust::system::cuda::detail::default_decomposition<IndexType>(last - first);
+
+  ValueArray block_results(exec, decomp.size() + 1);
+  
+  // compute sum over each interval
+  if (thrust::detail::is_commutative<BinaryFunction>::value)
+  {
+    // use reduce_intervals for commutative operators
+    thrust::system::cuda::detail::reduce_intervals(exec, first, block_results.begin() + 1, binary_op, decomp);
+  }
+  else
+  {
+    const static unsigned int ThreadsPerBlock = exclusive_scan_block_size<ValueType>::pass1;
+    typedef detail::statically_blocked_thread_array<ThreadsPerBlock> Context;
+
+    typedef upsweep_intervals_closure<InputIterator,ValueType,BinaryFunction,Decomposition,Context> Closure;
+    Closure closure(first,
+                    thrust::raw_pointer_cast(&block_results[0]) + 1,
+                    binary_op,
+                    decomp);
+    detail::launch_closure(closure, decomp.size(), ThreadsPerBlock);
+  }
+
+  // place init before per-block results
+  block_results[0] = init;
+  
+  // second level inclusive scan of per-block results
+  {
+    const static unsigned int ThreadsPerBlock = exclusive_scan_block_size<ValueType>::pass2;
+    typedef detail::statically_blocked_thread_array<ThreadsPerBlock> Context;
+
+    typedef downsweep_intervals_closure<true,ValueType*,ValueType*,ValueType,BinaryFunction,Decomposition,Context> Closure;
+    Closure closure(thrust::raw_pointer_cast(&block_results[0]),
+                    thrust::raw_pointer_cast(&block_results[0]),
+                    thrust::raw_pointer_cast(&block_results[0]), // not used
+                    binary_op,
+                    Decomposition(decomp.size() + 1, 1, 1));
+    detail::launch_closure(closure, 1, ThreadsPerBlock);
+  }
+  
+  // update intervals with result of second level scan
+  {
+    const static unsigned int ThreadsPerBlock = exclusive_scan_block_size<ValueType>::pass3;
+    typedef detail::statically_blocked_thread_array<ThreadsPerBlock> Context;
+
+    typedef downsweep_intervals_closure<false,InputIterator,OutputIterator,ValueType,BinaryFunction,Decomposition,Context> Closure;
+    Closure closure(first,
+                    output,
+                    thrust::raw_pointer_cast(&block_results[0]), // shift block results
+                    binary_op,
+                    decomp);
+    detail::launch_closure(closure, decomp.size(), ThreadsPerBlock);
+  }
+  
+  return output + (last - first);
+}
+
+
+} // end namespace fast_scan
+} // end namespace detail
+} // end namespace detail
+} // end namespace cuda
+} // end namespace system
+} // end namespace thrust
+
+__THRUST_DISABLE_MSVC_POSSIBLE_LOSS_OF_DATA_WARNING_END
+
diff --git a/compat/thrust/system/cuda/detail/detail/launch_calculator.h b/compat/thrust/system/cuda/detail/detail/launch_calculator.h
new file mode 100644
index 0000000000..5126aa6f1a
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/detail/launch_calculator.h
@@ -0,0 +1,82 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+#include <thrust/system/cuda/detail/cuda_launch_config.h>
+#include <thrust/tuple.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+namespace detail
+{
+
+template <typename Closure>
+class launch_calculator
+{
+  device_properties_t   properties;
+  function_attributes_t attributes;
+
+  public:
+  
+  launch_calculator(void);
+
+  launch_calculator(const device_properties_t& properties, const function_attributes_t& attributes);
+
+  thrust::tuple<size_t,size_t,size_t> with_variable_block_size(void) const;
+
+  template <typename UnaryFunction>
+  thrust::tuple<size_t,size_t,size_t> with_variable_block_size(UnaryFunction block_size_to_smem_size) const;
+  
+  thrust::tuple<size_t,size_t,size_t> with_variable_block_size_available_smem(void) const;
+
+  private:
+
+  /*! Returns a pair (num_threads_per_block, num_blocks_per_multiprocessor)
+   *  where num_threads_per_block is a valid block size for an instance of Closure
+   *  chosen by a heuristic and num_blocks_per_multiprocessor is the maximum
+   *  number of such blocks that can execute on a streaming multiprocessor at once.
+   */
+  thrust::pair<size_t, size_t> default_block_configuration() const;
+
+  /*! Returns a pair (num_threads_per_block, num_blocks_per_multiprocessor)
+   *  where num_threads_per_block is a valid block size for an instance of Closure
+   *  chosen by a heuristic and num_blocks_per_multiprocessor is the maximum
+   *  number of such blocks that can execute on a streaming multiprocessor at once.
+   *
+   *  \param block_size_to_smem_size Mapping from num_threads_per_block to number of
+   *                                 dynamically-allocated bytes of shared memory
+   */
+  template<typename UnaryFunction>
+  thrust::pair<size_t, size_t> default_block_configuration(UnaryFunction block_size_to_smem_size) const;
+};
+
+} // end namespace detail
+} // end namespace detail
+} // end namespace cuda
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/cuda/detail/detail/launch_calculator.inl>
+
diff --git a/compat/thrust/system/cuda/detail/detail/launch_calculator.inl b/compat/thrust/system/cuda/detail/detail/launch_calculator.inl
new file mode 100644
index 0000000000..b851d5fe13
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/detail/launch_calculator.inl
@@ -0,0 +1,103 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+// do not attempt to compile this file with any other compiler
+#if THRUST_DEVICE_COMPILER == THRUST_DEVICE_COMPILER_NVCC
+
+#include <thrust/system/cuda/detail/runtime_introspection.h>
+#include <thrust/system/cuda/detail/cuda_launch_config.h>
+#include <thrust/system/cuda/detail/detail/launch_closure.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+namespace detail
+{
+
+template <typename Closure>
+launch_calculator<Closure>::launch_calculator(void)
+  : properties(device_properties()),
+    attributes(closure_attributes<Closure>())
+{}
+  
+template <typename Closure>
+launch_calculator<Closure>::launch_calculator(const device_properties_t& properties, const function_attributes_t& attributes)
+  : properties(properties),
+    attributes(attributes)
+{}
+
+template <typename Closure>
+  template <typename UnaryFunction>
+thrust::pair<size_t, size_t> launch_calculator<Closure>::default_block_configuration(UnaryFunction block_size_to_smem_size) const
+{
+  // choose a block size
+  std::size_t num_threads_per_block = block_size_with_maximum_potential_occupancy(attributes, properties, block_size_to_smem_size);
+
+  // choose a subscription rate
+  std::size_t num_blocks_per_multiprocessor = properties.maxThreadsPerMultiProcessor / num_threads_per_block;
+
+  return thrust::make_pair(num_threads_per_block, num_blocks_per_multiprocessor);
+}
+
+
+template <typename Closure>
+thrust::pair<size_t, size_t> launch_calculator<Closure>::default_block_configuration(void) const
+{
+  // choose a block size
+  std::size_t num_threads_per_block = block_size_with_maximum_potential_occupancy(attributes, properties);
+
+  // choose a subscription rate
+  std::size_t num_blocks_per_multiprocessor = properties.maxThreadsPerMultiProcessor / num_threads_per_block;
+
+  return thrust::make_pair(num_threads_per_block, num_blocks_per_multiprocessor);
+}
+
+template <typename Closure>
+thrust::tuple<size_t,size_t,size_t> launch_calculator<Closure>::with_variable_block_size(void) const
+{
+  thrust::pair<size_t, size_t> config = default_block_configuration();
+  return thrust::tuple<size_t,size_t,size_t>(config.second * properties.multiProcessorCount, config.first, 0);
+}
+
+template <typename Closure>
+  template <typename UnaryFunction>
+thrust::tuple<size_t,size_t,size_t> launch_calculator<Closure>::with_variable_block_size(UnaryFunction block_size_to_smem_size) const
+{
+  thrust::pair<size_t, size_t> config = default_block_configuration(block_size_to_smem_size);
+  return thrust::tuple<size_t,size_t,size_t>(config.second * properties.multiProcessorCount, config.first, block_size_to_smem_size(config.first));
+}
+  
+template <typename Closure>
+thrust::tuple<size_t,size_t,size_t> launch_calculator<Closure>::with_variable_block_size_available_smem(void) const
+{
+  thrust::pair<size_t, size_t> config = default_block_configuration();
+  size_t smem_per_block = proportional_smem_allocation(properties, attributes, config.second);
+  return thrust::tuple<size_t,size_t,size_t>(config.second * properties.multiProcessorCount, config.first, smem_per_block);
+}
+
+} // end detail
+} // end namespace detail
+} // end namespace cuda
+} // end namespace system
+} // end namespace thrust
+
+#endif // THRUST_DEVICE_COMPILER == THRUST_DEVICE_COMPILER_NVCC
+
diff --git a/compat/thrust/system/cuda/detail/detail/launch_closure.h b/compat/thrust/system/cuda/detail/detail/launch_closure.h
new file mode 100644
index 0000000000..c2e6c4344f
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/detail/launch_closure.h
@@ -0,0 +1,114 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/type_traits.h>
+#include <thrust/system/cuda/detail/cuda_launch_config.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+namespace detail
+{
+
+template <unsigned int _ThreadsPerBlock = 0,
+          unsigned int _BlocksPerMultiprocessor = 0>
+struct launch_bounds
+{
+  typedef thrust::detail::integral_constant<unsigned int, _ThreadsPerBlock>         ThreadsPerBlock;
+  typedef thrust::detail::integral_constant<unsigned int, _BlocksPerMultiprocessor> BlocksPerMultiprocessor;
+};
+
+struct thread_array : public launch_bounds<>
+{
+// CUDA built-in variables require nvcc
+#if THRUST_DEVICE_COMPILER == THRUST_DEVICE_COMPILER_NVCC
+  __device__ __thrust_forceinline__ unsigned int thread_index(void) const { return threadIdx.x; }
+  __device__ __thrust_forceinline__ unsigned int thread_count(void) const { return blockDim.x * gridDim.x; } 
+#else
+  __device__ __thrust_forceinline__ unsigned int thread_index(void) const { return 0; }
+  __device__ __thrust_forceinline__ unsigned int thread_count(void) const { return 0; } 
+#endif // THRUST_DEVICE_COMPILER_NVCC
+};
+
+struct blocked_thread_array : public launch_bounds<>
+{
+// CUDA built-in variables require nvcc
+#if THRUST_DEVICE_COMPILER == THRUST_DEVICE_COMPILER_NVCC
+  __device__ __thrust_forceinline__ unsigned int thread_index(void)    const { return threadIdx.x; }
+  __device__ __thrust_forceinline__ unsigned int block_dimension(void) const { return blockDim.x;  } 
+  __device__ __thrust_forceinline__ unsigned int block_index(void)     const { return blockIdx.x;  }
+  __device__ __thrust_forceinline__ unsigned int grid_dimension(void)  const { return gridDim.x;   }
+  __device__ __thrust_forceinline__ unsigned int linear_index(void)    const { return block_dimension() * block_index() + thread_index(); }
+  __device__ __thrust_forceinline__ void         barrier(void)               { __syncthreads();    }
+#else
+  __device__ __thrust_forceinline__ unsigned int thread_index(void)    const { return 0; }
+  __device__ __thrust_forceinline__ unsigned int block_dimension(void) const { return 0; }
+  __device__ __thrust_forceinline__ unsigned int block_index(void)     const { return 0; }
+  __device__ __thrust_forceinline__ unsigned int grid_dimension(void)  const { return 0; }
+  __device__ __thrust_forceinline__ unsigned int linear_index(void)    const { return 0; }
+  __device__ __thrust_forceinline__ void         barrier(void)               {           }
+#endif // THRUST_DEVICE_COMPILER_NVCC
+};
+
+template <unsigned int _ThreadsPerBlock>
+struct statically_blocked_thread_array : public launch_bounds<_ThreadsPerBlock,1>
+{
+// CUDA built-in variables require nvcc
+#if THRUST_DEVICE_COMPILER == THRUST_DEVICE_COMPILER_NVCC
+  __device__ __thrust_forceinline__ unsigned int thread_index(void)    const { return threadIdx.x;      }
+  __device__ __thrust_forceinline__ unsigned int block_dimension(void) const { return _ThreadsPerBlock; } // minor optimization
+  __device__ __thrust_forceinline__ unsigned int block_index(void)     const { return blockIdx.x;       }
+  __device__ __thrust_forceinline__ unsigned int grid_dimension(void)  const { return gridDim.x;        }
+  __device__ __thrust_forceinline__ unsigned int linear_index(void)    const { return block_dimension() * block_index() + thread_index(); }
+  __device__ __thrust_forceinline__ void         barrier(void)               { __syncthreads();    }
+#else
+  __device__ __thrust_forceinline__ unsigned int thread_index(void)    const { return 0; }
+  __device__ __thrust_forceinline__ unsigned int block_dimension(void) const { return 0; }
+  __device__ __thrust_forceinline__ unsigned int block_index(void)     const { return 0; }
+  __device__ __thrust_forceinline__ unsigned int grid_dimension(void)  const { return 0; }
+  __device__ __thrust_forceinline__ unsigned int linear_index(void)    const { return 0; }
+  __device__ __thrust_forceinline__ void         barrier(void)               {           }
+#endif // THRUST_DEVICE_COMPILER_NVCC
+};
+
+template<typename Closure, typename Size1, typename Size2>
+  void launch_closure(Closure f, Size1 num_blocks, Size2 block_size);
+
+template<typename Closure, typename Size1, typename Size2, typename Size3>
+  void launch_closure(Closure f, Size1 num_blocks, Size2 block_size, Size3 smem_size);
+
+/*! Returns a copy of the cudaFuncAttributes structure
+ *  that is associated with a given Closure
+ */
+template <typename Closure>
+function_attributes_t closure_attributes(void);
+
+} // end namespace detail
+} // end namespace detail
+} // end namespace cuda
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/cuda/detail/detail/launch_closure.inl>
+
diff --git a/compat/thrust/system/cuda/detail/detail/launch_closure.inl b/compat/thrust/system/cuda/detail/detail/launch_closure.inl
new file mode 100644
index 0000000000..ce39cfc136
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/detail/launch_closure.inl
@@ -0,0 +1,207 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/detail/minmax.h>
+#include <thrust/detail/type_traits.h>
+#include <thrust/detail/temporary_array.h>
+#include <thrust/system/cuda/detail/runtime_introspection.h>
+#include <thrust/system/cuda/detail/synchronize.h>
+#include <thrust/system/cuda/detail/detail/launch_calculator.h>
+#include <thrust/system/cuda/detail/execution_policy.h>
+
+namespace thrust
+{
+namespace detail
+{
+
+// XXX WAR circular inclusion problems with this forward declaration
+template<typename, typename> class temporary_array;
+
+} // end detail
+
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+namespace detail
+{
+
+#if THRUST_DEVICE_COMPILER == THRUST_DEVICE_COMPILER_NVCC
+template<typename Closure>
+__global__ __launch_bounds__(Closure::context_type::ThreadsPerBlock::value, Closure::context_type::BlocksPerMultiprocessor::value)
+void launch_closure_by_value(Closure f)
+{
+  f();
+}
+
+template<typename Closure>
+__global__ __launch_bounds__(Closure::context_type::ThreadsPerBlock::value, Closure::context_type::BlocksPerMultiprocessor::value)
+void launch_closure_by_pointer(const Closure *f)
+{
+  // copy to registers
+  Closure f_reg = *f;
+  f_reg();
+}
+#else
+template<typename Closure>
+void launch_closure_by_value(Closure) {}
+
+template<typename Closure>
+void launch_closure_by_pointer(const Closure *) {}
+
+#endif // THRUST_DEVICE_COMPILER_NVCC
+
+template<typename Closure,
+         bool launch_by_value = sizeof(Closure) <= 256>
+  struct closure_launcher_base
+{
+  typedef void (*launch_function_t)(Closure); 
+ 
+  static launch_function_t get_launch_function(void)
+  {
+    return launch_closure_by_value<Closure>;
+  }
+
+  template<typename Size1, typename Size2, typename Size3>
+  static void launch(Closure f, Size1 num_blocks, Size2 block_size, Size3 smem_size)
+  {
+#if THRUST_DEVICE_COMPILER == THRUST_DEVICE_COMPILER_NVCC
+    if(num_blocks > 0)
+    {
+      launch_closure_by_value<<<(unsigned int) num_blocks, (unsigned int) block_size, (unsigned int) smem_size>>>(f);
+      synchronize_if_enabled("launch_closure_by_value");
+    }
+#endif // THRUST_DEVICE_COMPILER_NVCC
+  }
+}; // end closure_launcher_base
+
+
+template<typename Closure>
+  struct closure_launcher_base<Closure,false>
+{
+  typedef void (*launch_function_t)(const Closure *); 
+ 
+  static launch_function_t get_launch_function(void)
+  {
+    return launch_closure_by_pointer<Closure>;
+  }
+
+  template<typename Size1, typename Size2, typename Size3>
+  static void launch(Closure f, Size1 num_blocks, Size2 block_size, Size3 smem_size)
+  {
+#if THRUST_DEVICE_COMPILER == THRUST_DEVICE_COMPILER_NVCC
+    if(num_blocks > 0)
+    {
+      // use temporary storage for the closure
+      // XXX use of cuda::tag is too specific here
+      thrust::cuda::tag cuda_tag;
+      thrust::host_system_tag host_tag;
+      thrust::detail::temporary_array<Closure,thrust::cuda::tag> closure_storage(cuda_tag, host_tag, &f, &f + 1);
+
+      // launch
+      detail::launch_closure_by_pointer<<<(unsigned int) num_blocks, (unsigned int) block_size, (unsigned int) smem_size>>>((&closure_storage[0]).get());
+      synchronize_if_enabled("launch_closure_by_pointer");
+    }
+#endif // THRUST_DEVICE_COMPILER_NVCC
+  }
+};
+
+
+template<typename Closure>
+  struct closure_launcher
+    : public closure_launcher_base<Closure>
+{
+  typedef closure_launcher_base<Closure> super_t;
+  
+  static inline const device_properties_t& device_properties(void)
+  {
+    return device_properties();
+  }
+  
+  static inline function_attributes_t function_attributes(void)
+  {
+    return thrust::system::cuda::detail::function_attributes(super_t::get_launch_function());
+  }
+
+  template<typename Size1, typename Size2, typename Size3>
+  static void launch(Closure f, Size1 num_blocks, Size2 block_size, Size3 smem_size)
+  {
+    super_t::launch(f,num_blocks,block_size,smem_size);
+  }
+};
+
+template<typename Closure, typename Size>
+  void launch_closure(Closure f, Size num_blocks)
+{
+  launch_calculator<Closure> calculator;
+  launch_closure(f, num_blocks, thrust::get<1>(calculator.with_variable_block_size()));
+} // end launch_closure()
+
+template<typename Closure, typename Size1, typename Size2>
+  void launch_closure(Closure f, Size1 num_blocks, Size2 block_size)
+{
+  launch_closure(f, num_blocks, block_size, 0u);
+} // end launch_closure()
+
+template<typename Closure, typename Size1, typename Size2, typename Size3>
+  void launch_closure(Closure f, Size1 num_blocks, Size2 block_size, Size3 smem_size)
+{
+  closure_launcher<Closure>::launch(f, num_blocks, block_size, smem_size);
+} // end launch_closure()
+
+  
+template <typename Closure>
+function_attributes_t closure_attributes(void)
+{
+  typedef closure_launcher<Closure> Launcher;
+
+  // cache the result of function_attributes(), because it is slow
+  // only cache the first few devices
+  static const int max_num_devices                                  = 16;
+
+  static bool attributes_exist[max_num_devices]                     = {0};
+  static function_attributes_t function_attributes[max_num_devices] = {};
+
+  // XXX device_id ought to be an argument to this function
+  int device_id = current_device();
+
+  if(device_id >= max_num_devices)
+  {
+    return thrust::system::cuda::detail::function_attributes(Launcher::get_launch_function());
+  }
+
+  if(!attributes_exist[device_id])
+  {
+    function_attributes[device_id] = thrust::system::cuda::detail::function_attributes(Launcher::get_launch_function());
+
+    // disallow the compiler to move the write to attributes_exist[device_id]
+    // before the initialization of function_attributes[device_id]
+    __thrust_compiler_fence();
+
+    attributes_exist[device_id] = true;
+  }
+
+  return function_attributes[device_id];
+}
+
+} // end namespace detail
+} // end namespace detail
+} // end namespace cuda
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/cuda/detail/detail/set_operation.h b/compat/thrust/system/cuda/detail/detail/set_operation.h
new file mode 100644
index 0000000000..5475731edb
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/detail/set_operation.h
@@ -0,0 +1,56 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/cuda/detail/execution_policy.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+namespace detail
+{
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2,
+         typename RandomAccessIterator3,
+         typename Compare,
+         typename SetOperation>
+  RandomAccessIterator3 set_operation(execution_policy<DerivedPolicy> &exec,
+                                      RandomAccessIterator1 first1,
+                                      RandomAccessIterator1 last1,
+                                      RandomAccessIterator2 first2,
+                                      RandomAccessIterator2 last2,
+                                      RandomAccessIterator3 result,
+                                      Compare comp,
+                                      SetOperation set_op);
+
+
+} // end detail
+} // end detail
+} // end cuda
+} // end system
+} // end thrust
+
+#include <thrust/system/cuda/detail/detail/set_operation.inl>
+
diff --git a/compat/thrust/system/cuda/detail/detail/set_operation.inl b/compat/thrust/system/cuda/detail/detail/set_operation.inl
new file mode 100644
index 0000000000..3f14379628
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/detail/set_operation.inl
@@ -0,0 +1,639 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/system/cuda/detail/detail/set_operation.h>
+#include <thrust/system/cuda/detail/detail/balanced_path.h>
+#include <thrust/system/cuda/detail/block/inclusive_scan.h>
+#include <thrust/system/cuda/detail/block/exclusive_scan.h>
+#include <thrust/system/cuda/detail/block/copy.h>
+#include <thrust/iterator/counting_iterator.h>
+#include <thrust/transform.h>
+#include <thrust/scan.h>
+#include <thrust/pair.h>
+#include <thrust/detail/util/blocking.h>
+#include <thrust/detail/temporary_array.h>
+#include <thrust/detail/cstdint.h>
+#include <thrust/detail/minmax.h>
+
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+namespace detail
+{
+namespace set_operation_detail
+{
+
+
+using thrust::system::cuda::detail::detail::statically_blocked_thread_array;
+using thrust::detail::uint16_t;
+using thrust::detail::uint32_t;
+
+
+// empirically determined on sm_20
+// value_types larger than this will fail to launch if placed in smem
+template<typename T>
+  struct stage_through_smem
+{
+  static const bool value = sizeof(T) <= 6 * sizeof(uint32_t);
+};
+
+
+// max_input_size <= 32
+template<typename Size, typename InputIterator, typename OutputIterator>
+inline __device__
+  OutputIterator serial_bounded_copy_if(Size max_input_size,
+                                        InputIterator first,
+                                        uint32_t mask,
+                                        OutputIterator result)
+{
+  for(Size i = 0; i < max_input_size; ++i, ++first)
+  {
+    if((1<<i) & mask)
+    {
+      *result = *first;
+      ++result;
+    }
+  }
+
+  return result;
+}
+
+
+template<typename Size, typename InputIterator1, typename InputIterator2, typename Compare>
+  struct find_partition_offsets_functor
+{
+  Size partition_size;
+  InputIterator1 first1;
+  InputIterator2 first2;
+  Size n1, n2;
+  Compare comp;
+
+  find_partition_offsets_functor(Size partition_size,
+                                 InputIterator1 first1, InputIterator1 last1,
+                                 InputIterator2 first2, InputIterator2 last2,
+                                 Compare comp)
+    : partition_size(partition_size),
+      first1(first1), first2(first2),
+      n1(last1 - first1), n2(last2 - first2),
+      comp(comp)
+  {}
+
+  inline __host__ __device__
+  thrust::pair<Size,Size> operator()(Size i) const
+  {
+    Size diag = thrust::min(n1 + n2, i * partition_size);
+
+    // XXX the correctness of balanced_path depends critically on the ll suffix below
+    //     why???
+    return balanced_path(first1, n1, first2, n2, diag, 4ll, comp);
+  }
+};
+
+
+template<typename Size, typename DerivedPolicy, typename InputIterator1, typename InputIterator2, typename OutputIterator, typename Compare>
+  OutputIterator find_partition_offsets(thrust::cuda::execution_policy<DerivedPolicy> &exec,
+                                        Size num_partitions,
+                                        Size partition_size,
+                                        InputIterator1 first1, InputIterator1 last1,
+                                        InputIterator2 first2, InputIterator2 last2,
+                                        OutputIterator result,
+                                        Compare comp)
+{
+  find_partition_offsets_functor<Size,InputIterator1,InputIterator2,Compare> f(partition_size, first1, last1, first2, last2, comp);
+
+  return thrust::transform(exec,
+                           thrust::counting_iterator<Size>(0),
+                           thrust::counting_iterator<Size>(num_partitions),
+                           result,
+                           f);
+}
+
+
+namespace block
+{
+
+
+template<unsigned int block_size, typename T>
+inline __device__
+T right_neighbor(statically_blocked_thread_array<block_size> &ctx, const T &x, const T &boundary)
+{
+  // stage this shift to conserve smem
+  const unsigned int storage_size = block_size / 2;
+  __shared__ uninitialized_array<T,storage_size> shared;
+
+  T result = x;
+
+  unsigned int tid = ctx.thread_index();
+
+  if(0 < tid && tid <= storage_size)
+  {
+    shared[tid - 1] = x;
+  }
+
+  ctx.barrier();
+
+  if(tid < storage_size)
+  {
+    result = shared[tid];
+  }
+
+  ctx.barrier();
+  
+  tid -= storage_size;
+  if(0 < tid && tid <= storage_size)
+  {
+    shared[tid - 1] = x;
+  }
+  else if(tid == 0)
+  {
+    shared[storage_size-1] = boundary;
+  }
+
+  ctx.barrier();
+
+  if(tid < storage_size)
+  {
+    result = shared[tid];
+  }
+
+  ctx.barrier();
+
+  return result;
+}
+
+
+template<uint16_t block_size, uint16_t work_per_thread, typename InputIterator1, typename InputIterator2, typename Compare, typename SetOperation>
+inline __device__
+  unsigned int bounded_count_set_operation_n(statically_blocked_thread_array<block_size> &ctx,
+                                             InputIterator1 first1, uint16_t n1,
+                                             InputIterator2 first2, uint16_t n2,
+                                             Compare comp,
+                                             SetOperation set_op)
+{
+  unsigned int thread_idx = ctx.thread_index();
+
+  // find partition offsets
+  uint16_t diag = thrust::min<uint16_t>(n1 + n2, thread_idx * work_per_thread);
+  thrust::pair<uint16_t,uint16_t> thread_input_begin = balanced_path(first1, n1, first2, n2, diag, 2, comp);
+  thrust::pair<uint16_t,uint16_t> thread_input_end   = block::right_neighbor<block_size>(ctx, thread_input_begin, thrust::make_pair(n1,n2));
+
+  __shared__ uint16_t s_thread_output_size[block_size];
+
+  // work_per_thread + 1 to accomodate a "starred" partition returned from balanced_path above
+  s_thread_output_size[thread_idx] =
+    set_op.count(work_per_thread + 1,
+                 first1 + thread_input_begin.first,  first1 + thread_input_end.first,
+                 first2 + thread_input_begin.second, first2 + thread_input_end.second,
+                 comp);
+
+  ctx.barrier();
+
+  // reduce per-thread counts
+  thrust::system::cuda::detail::block::inplace_inclusive_scan(ctx, s_thread_output_size);
+  return s_thread_output_size[ctx.block_dimension() - 1];
+}
+
+
+inline __device__ int pop_count(unsigned int x)
+{
+// guard use of __popc from other compilers
+#if THRUST_DEVICE_COMPILER == THRUST_DEVICE_COMPILER_NVCC
+  return __popc(x);
+#else
+  return x;
+#endif
+}
+
+
+
+template<uint16_t block_size, uint16_t work_per_thread, typename InputIterator1, typename InputIterator2, typename OutputIterator, typename Compare, typename SetOperation>
+inline __device__
+  OutputIterator bounded_set_operation_n(statically_blocked_thread_array<block_size> &ctx,
+                                         InputIterator1 first1, uint16_t n1,
+                                         InputIterator2 first2, uint16_t n2,
+                                         OutputIterator result,
+                                         Compare comp,
+                                         SetOperation set_op)
+{
+  unsigned int thread_idx = ctx.thread_index();
+  
+  // find partition offsets
+  uint16_t diag = thrust::min<uint16_t>(n1 + n2, thread_idx * work_per_thread);
+  thrust::pair<uint16_t,uint16_t> thread_input_begin = balanced_path(first1, n1, first2, n2, diag, 2, comp);
+  thrust::pair<uint16_t,uint16_t> thread_input_end   = block::right_neighbor<block_size>(ctx, thread_input_begin, thrust::make_pair(n1,n2));
+
+  typedef typename thrust::iterator_value<InputIterator1>::type value_type;
+  // +1 to accomodate a "starred" partition returned from balanced_path above
+  uninitialized_array<value_type, work_per_thread + 1> sparse_result;
+  uint32_t active_mask =
+    set_op(work_per_thread + 1,
+           first1 + thread_input_begin.first,  first1 + thread_input_end.first,
+           first2 + thread_input_begin.second, first2 + thread_input_end.second,
+           sparse_result.begin(),
+           comp);
+
+  __shared__ uint16_t s_thread_output_size[block_size];
+  s_thread_output_size[thread_idx] = pop_count(active_mask);
+
+  ctx.barrier();
+
+  // scan to turn per-thread counts into output indices
+  uint16_t block_output_size = thrust::system::cuda::detail::block::inplace_exclusive_scan(ctx, s_thread_output_size, 0u);
+
+  serial_bounded_copy_if(work_per_thread + 1, sparse_result.begin(), active_mask, result + s_thread_output_size[thread_idx]);
+
+  ctx.barrier();
+
+  return result + block_output_size;
+}
+
+
+template<uint16_t block_size, uint16_t work_per_thread, typename InputIterator1, typename InputIterator2, typename Compare, typename SetOperation>
+inline __device__
+  typename thrust::iterator_difference<InputIterator1>::type
+    count_set_operation(statically_blocked_thread_array<block_size> &ctx,
+                        InputIterator1 first1, InputIterator1 last1,
+                        InputIterator2 first2, InputIterator2 last2,
+                        Compare comp,
+                        SetOperation set_op)
+{
+  typedef typename thrust::iterator_difference<InputIterator1>::type difference;
+
+  difference result = 0;
+
+  thrust::pair<difference,difference> remaining_input_size = thrust::make_pair(last1 - first1, last2 - first2);
+
+  // iterate until the input is consumed
+  while(remaining_input_size.first + remaining_input_size.second > 0)
+  {
+    // find the end of this subpartition's input
+    // -1 to accomodate "starred" partitions
+    uint16_t max_subpartition_size = block_size * work_per_thread - 1;
+    difference diag = thrust::min<difference>(remaining_input_size.first + remaining_input_size.second, max_subpartition_size);
+    thrust::pair<uint16_t,uint16_t> subpartition_size = balanced_path(first1, remaining_input_size.first, first2, remaining_input_size.second, diag, 4ll, comp);
+  
+    typedef typename thrust::iterator_value<InputIterator2>::type value_type;
+    if(stage_through_smem<value_type>::value)
+    {
+      // load the input into __shared__ storage
+      __shared__ uninitialized_array<value_type, block_size * work_per_thread> s_input;
+  
+      value_type *s_input_end1 = thrust::system::cuda::detail::block::copy_n(ctx, first1, subpartition_size.first,  s_input.begin());
+      value_type *s_input_end2 = thrust::system::cuda::detail::block::copy_n(ctx, first2, subpartition_size.second, s_input_end1);
+  
+      result += block::bounded_count_set_operation_n<block_size,work_per_thread>(ctx,
+                                                                                 s_input.begin(), subpartition_size.first,
+                                                                                 s_input_end1,    subpartition_size.second,
+                                                                                 comp,
+                                                                                 set_op);
+    }
+    else
+    {
+      result += block::bounded_count_set_operation_n<block_size,work_per_thread>(ctx,
+                                                                                 first1, subpartition_size.first,
+                                                                                 first2, subpartition_size.second,
+                                                                                 comp,
+                                                                                 set_op);
+    }
+
+    // advance input
+    first1 += subpartition_size.first;
+    first2 += subpartition_size.second;
+
+    // decrement remaining size
+    remaining_input_size.first  -= subpartition_size.first;
+    remaining_input_size.second -= subpartition_size.second;
+  }
+
+  return result;
+}
+
+
+template<uint16_t block_size, uint16_t work_per_thread, typename InputIterator1, typename InputIterator2, typename OutputIterator, typename Compare, typename SetOperation>
+inline __device__
+OutputIterator set_operation(statically_blocked_thread_array<block_size> &ctx,
+                             InputIterator1 first1, InputIterator1 last1,
+                             InputIterator2 first2, InputIterator2 last2,
+                             OutputIterator result,
+                             Compare comp,
+                             SetOperation set_op)
+{
+  typedef typename thrust::iterator_difference<InputIterator1>::type difference;
+
+  thrust::pair<difference,difference> remaining_input_size = thrust::make_pair(last1 - first1, last2 - first2);
+
+  // iterate until the input is consumed
+  while(remaining_input_size.first + remaining_input_size.second > 0)
+  {
+    // find the end of this subpartition's input
+    // -1 to accomodate "starred" partitions
+    uint16_t max_subpartition_size = block_size * work_per_thread - 1;
+    difference diag = thrust::min<difference>(remaining_input_size.first + remaining_input_size.second, max_subpartition_size);
+    thrust::pair<uint16_t,uint16_t> subpartition_size = balanced_path(first1, remaining_input_size.first, first2, remaining_input_size.second, diag, 4ll, comp);
+    
+    typedef typename thrust::iterator_value<InputIterator2>::type value_type;
+    if(stage_through_smem<value_type>::value)
+    {
+      // load the input into __shared__ storage
+      __shared__ uninitialized_array<value_type, block_size * work_per_thread> s_input;
+  
+      value_type *s_input_end1 = thrust::system::cuda::detail::block::copy_n(ctx, first1, subpartition_size.first,  s_input.begin());
+      value_type *s_input_end2 = thrust::system::cuda::detail::block::copy_n(ctx, first2, subpartition_size.second, s_input_end1);
+  
+      result = block::bounded_set_operation_n<block_size,work_per_thread>(ctx,
+                                                                          s_input.begin(), subpartition_size.first,
+                                                                          s_input_end1,    subpartition_size.second,
+                                                                          result,
+                                                                          comp,
+                                                                          set_op);
+    }
+    else
+    {
+      result = block::bounded_set_operation_n<block_size,work_per_thread>(ctx,
+                                                                          first1, subpartition_size.first,
+                                                                          first2, subpartition_size.second,
+                                                                          result,
+                                                                          comp,
+                                                                          set_op);
+    }
+  
+    // advance input
+    first1 += subpartition_size.first;
+    first2 += subpartition_size.second;
+
+    // decrement remaining size
+    remaining_input_size.first  -= subpartition_size.first;
+    remaining_input_size.second -= subpartition_size.second;
+  }
+
+  return result;
+}
+
+
+} // end namespace block
+
+
+template<uint16_t threads_per_block, uint16_t work_per_thread, typename InputIterator1, typename Size, typename InputIterator2, typename InputIterator3, typename OutputIterator, typename Compare, typename SetOperation>
+  inline __device__ void count_set_operation(statically_blocked_thread_array<threads_per_block> &ctx,
+                                             InputIterator1                                      input_partition_offsets,
+                                             Size                                                num_partitions,
+                                             InputIterator2                                      first1,
+                                             InputIterator3                                      first2,
+                                             OutputIterator                                      result,
+                                             Compare                                             comp,
+                                             SetOperation                                        set_op)
+{
+  // consume partitions
+  for(Size partition_idx = ctx.block_index();
+      partition_idx < num_partitions;
+      partition_idx += ctx.grid_dimension())
+  {
+    typedef typename thrust::iterator_difference<InputIterator2>::type difference;
+
+    // find the partition
+    thrust::pair<difference,difference> block_input_begin = input_partition_offsets[partition_idx];
+    thrust::pair<difference,difference> block_input_end   = input_partition_offsets[partition_idx + 1];
+
+    // count the size of the set operation
+    difference count = block::count_set_operation<threads_per_block,work_per_thread>(ctx,
+                                                                                     first1 + block_input_begin.first,  first1 + block_input_end.first,
+                                                                                     first2 + block_input_begin.second, first2 + block_input_end.second,
+                                                                                     comp,
+                                                                                     set_op);
+
+    if(ctx.thread_index() == 0)
+    {
+      result[partition_idx] = count;
+    }
+  }
+}
+
+
+template<uint16_t threads_per_block, uint16_t work_per_thread, typename InputIterator1, typename Size, typename InputIterator2, typename InputIterator3, typename OutputIterator, typename Compare, typename SetOperation>
+  struct count_set_operation_closure
+{
+  typedef statically_blocked_thread_array<threads_per_block> context_type;
+
+  InputIterator1 input_partition_offsets;
+  Size           num_partitions;
+  InputIterator2 first1;
+  InputIterator3 first2;
+  OutputIterator result;
+  Compare        comp;
+  SetOperation   set_op;
+
+  count_set_operation_closure(InputIterator1 input_partition_offsets,
+                              Size           num_partitions,
+                              InputIterator2 first1,
+                              InputIterator3 first2,
+                              OutputIterator result,
+                              Compare        comp,
+                              SetOperation   set_op)
+    : input_partition_offsets(input_partition_offsets),
+      num_partitions(num_partitions),
+      first1(first1),
+      first2(first2),
+      result(result),
+      comp(comp),
+      set_op(set_op)
+  {}
+
+  inline __device__ void operator()() const
+  {
+    context_type ctx;
+    count_set_operation<threads_per_block,work_per_thread>(ctx, input_partition_offsets, num_partitions, first1, first2, result, comp, set_op);
+  }
+};
+
+
+template<uint16_t threads_per_block, uint16_t work_per_thread, typename InputIterator1, typename Size, typename InputIterator2, typename InputIterator3, typename OutputIterator, typename Compare, typename SetOperation>
+  count_set_operation_closure<threads_per_block,work_per_thread,InputIterator1,Size,InputIterator2,InputIterator3,OutputIterator,Compare,SetOperation>
+    make_count_set_operation_closure(InputIterator1 input_partition_offsets,
+                                     Size           num_partitions,
+                                     InputIterator2 first1,
+                                     InputIterator3 first2,
+                                     OutputIterator result,
+                                     Compare        comp,
+                                     SetOperation   set_op)
+{
+  typedef count_set_operation_closure<threads_per_block,work_per_thread,InputIterator1,Size,InputIterator2,InputIterator3,OutputIterator,Compare,SetOperation> result_type;
+  return result_type(input_partition_offsets,num_partitions,first1,first2,result,comp,set_op);
+}
+
+
+template<uint16_t threads_per_block, uint16_t work_per_thread, typename InputIterator1, typename Size, typename InputIterator2, typename InputIterator3, typename InputIterator4, typename OutputIterator, typename Compare, typename SetOperation>
+inline __device__
+  void set_operation(statically_blocked_thread_array<threads_per_block> &ctx,
+                     InputIterator1                                      input_partition_offsets,
+                     Size                                                num_partitions,
+                     InputIterator2                                      first1,
+                     InputIterator3                                      first2,
+                     InputIterator4                                      output_partition_offsets,
+                     OutputIterator                                      result,
+                     Compare                                             comp,
+                     SetOperation                                        set_op)
+{
+  // consume partitions
+  for(Size partition_idx = ctx.block_index();
+      partition_idx < num_partitions;
+      partition_idx += ctx.grid_dimension())
+  {
+    typedef typename thrust::iterator_difference<InputIterator2>::type difference;
+
+    // find the partition
+    thrust::pair<difference,difference> block_input_begin = input_partition_offsets[partition_idx];
+    thrust::pair<difference,difference> block_input_end   = input_partition_offsets[partition_idx + 1];
+
+    // do the set operation across the partition
+    block::set_operation<threads_per_block,work_per_thread>(ctx,
+                                                            first1 + block_input_begin.first,  first1 + block_input_end.first,
+                                                            first2 + block_input_begin.second, first2 + block_input_end.second,
+                                                            result + output_partition_offsets[partition_idx],
+                                                            comp,
+                                                            set_op);
+  }
+}
+
+
+template<uint16_t threads_per_block, uint16_t work_per_thread, typename InputIterator1, typename Size, typename InputIterator2, typename InputIterator3, typename InputIterator4, typename OutputIterator, typename Compare, typename SetOperation>
+  struct set_operation_closure
+{
+  typedef statically_blocked_thread_array<threads_per_block> context_type;
+
+  InputIterator1 input_partition_offsets;
+  Size           num_partitions;
+  InputIterator2 first1;
+  InputIterator3 first2;
+  InputIterator4 output_partition_offsets;
+  OutputIterator result;
+  Compare        comp;
+  SetOperation   set_op;
+
+  set_operation_closure(InputIterator1 input_partition_offsets,
+                        Size           num_partitions,
+                        InputIterator2 first1,
+                        InputIterator3 first2,
+                        InputIterator4 output_partition_offsets,
+                        OutputIterator result,
+                        Compare        comp,
+                        SetOperation   set_op)
+    : input_partition_offsets(input_partition_offsets),
+      num_partitions(num_partitions),
+      first1(first1),
+      first2(first2),
+      output_partition_offsets(output_partition_offsets),
+      result(result),
+      comp(comp),
+      set_op(set_op)
+  {}
+
+  inline __device__ void operator()() const
+  {
+    context_type ctx;
+    set_operation<threads_per_block,work_per_thread>(ctx, input_partition_offsets, num_partitions, first1, first2, output_partition_offsets, result, comp, set_op);
+  }
+};
+
+
+template<uint16_t threads_per_block, uint16_t work_per_thread, typename InputIterator1, typename Size, typename InputIterator2, typename InputIterator3, typename InputIterator4, typename OutputIterator, typename Compare, typename SetOperation>
+  set_operation_closure<threads_per_block,work_per_thread,InputIterator1,Size,InputIterator2,InputIterator3,InputIterator4,OutputIterator,Compare,SetOperation>
+    make_set_operation_closure(InputIterator1 input_partition_offsets,
+                               Size           num_partitions,
+                               InputIterator2 first1,
+                               InputIterator3 first2,
+                               InputIterator4 output_partition_offsets,
+                               OutputIterator result,
+                               Compare        comp,
+                               SetOperation   set_op)
+{
+  typedef set_operation_closure<threads_per_block,work_per_thread,InputIterator1,Size,InputIterator2,InputIterator3,InputIterator4,OutputIterator,Compare,SetOperation> result_type;
+  return result_type(input_partition_offsets,num_partitions,first1,first2,output_partition_offsets,result,comp,set_op);
+}
+
+
+} // end namespace set_operation_detail
+
+
+template<typename DerivedPolicy, typename InputIterator1, typename InputIterator2, typename OutputIterator, typename Compare, typename SetOperation>
+  OutputIterator set_operation(thrust::cuda::execution_policy<DerivedPolicy> &exec,
+                               InputIterator1 first1, InputIterator1 last1,
+                               InputIterator2 first2, InputIterator2 last2,
+                               OutputIterator result,
+                               Compare comp,
+                               SetOperation set_op)
+{
+  using thrust::system::cuda::detail::device_properties;
+  using thrust::system::cuda::detail::detail::launch_closure;
+  namespace d = thrust::system::cuda::detail::detail::set_operation_detail;
+
+  typedef typename thrust::iterator_difference<InputIterator1>::type difference;
+
+  const difference n1 = last1 - first1;
+  const difference n2 = last2 - first2;
+
+  // handle empty input
+  if(n1 == 0 && n2 == 0)
+  {
+    return result;
+  }
+
+  const thrust::detail::uint16_t work_per_thread   = 15;
+  const thrust::detail::uint16_t threads_per_block = 128;
+  const thrust::detail::uint16_t work_per_block    = threads_per_block * work_per_thread;
+
+  // -1 because balanced_path adds a single element to the end of a "starred" partition, increasing its size by one
+  const thrust::detail::uint16_t maximum_partition_size = work_per_block - 1;
+  const difference num_partitions = thrust::detail::util::divide_ri(n1 + n2, maximum_partition_size);
+
+  // find input partition offsets
+  // +1 to handle the end of the input elegantly
+  thrust::detail::temporary_array<thrust::pair<difference,difference>, DerivedPolicy> input_partition_offsets(0, exec, num_partitions + 1);
+  d::find_partition_offsets<difference>(exec, input_partition_offsets.size(), maximum_partition_size, first1, last1, first2, last2, input_partition_offsets.begin(), comp);
+
+  const difference num_blocks = thrust::min<difference>(device_properties().maxGridSize[0], num_partitions);
+
+  // find output partition offsets
+  // +1 to store the total size of the total
+  thrust::detail::temporary_array<difference, DerivedPolicy> output_partition_offsets(0, exec, num_partitions + 1);
+  launch_closure(d::make_count_set_operation_closure<threads_per_block,work_per_thread>(input_partition_offsets.begin(), num_partitions, first1, first2, output_partition_offsets.begin(), comp, set_op),
+                 num_blocks,
+                 threads_per_block);
+
+  // turn the output partition counts into offsets to output partitions
+  thrust::exclusive_scan(exec, output_partition_offsets.begin(), output_partition_offsets.end(), output_partition_offsets.begin());
+
+  // run the set op kernel
+  launch_closure(d::make_set_operation_closure<threads_per_block,work_per_thread>(input_partition_offsets.begin(), num_partitions, first1, first2, output_partition_offsets.begin(), result, comp, set_op),
+                 num_blocks,
+                 threads_per_block);
+
+  return result + output_partition_offsets[num_partitions];
+}
+
+
+} // end namespace detail
+} // end namespace detail
+} // end namespace cuda
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/cuda/detail/detail/stable_merge_sort.h b/compat/thrust/system/cuda/detail/detail/stable_merge_sort.h
new file mode 100644
index 0000000000..23f32545af
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/detail/stable_merge_sort.h
@@ -0,0 +1,63 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file stable_merge_sort_dev.h
+ *  \brief Defines the interface for a stable merge implementation on CUDA
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/cuda/detail/execution_policy.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+namespace detail
+{
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator,
+         typename StrictWeakOrdering>
+void stable_merge_sort(execution_policy<DerivedPolicy> &exec,
+                       RandomAccessIterator first,
+                       RandomAccessIterator last,
+                       StrictWeakOrdering comp);
+    
+template<typename DerivedPolicy,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2,
+         typename StrictWeakOrdering>
+void stable_merge_sort_by_key(execution_policy<DerivedPolicy> &exec,
+                              RandomAccessIterator1 keys_begin,
+                              RandomAccessIterator1 keys_end,
+                              RandomAccessIterator2 values_begin,
+                              StrictWeakOrdering comp);
+
+} // end namespace detail
+} // end namespace detail
+} // end namespace cuda
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/cuda/detail/detail/stable_merge_sort.inl>
+
diff --git a/compat/thrust/system/cuda/detail/detail/stable_merge_sort.inl b/compat/thrust/system/cuda/detail/detail/stable_merge_sort.inl
new file mode 100644
index 0000000000..0c69803294
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/detail/stable_merge_sort.inl
@@ -0,0 +1,1103 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file stable_merge_sort.inl
+ *  \brief Inline file for stable_merge_sort.h.
+ *  \note This algorithm is based on the one described
+ *        in "Designing Efficient Sorting Algorithms for
+ *        Manycore GPUs", by Satish, Harris, and Garland.
+ */
+
+#include <thrust/detail/config.h>
+
+#include <thrust/functional.h>
+#include <thrust/detail/copy.h>
+
+#include <thrust/detail/function.h>
+
+#include <thrust/detail/mpl/math.h> // for log2<N>
+#include <thrust/detail/util/blocking.h>
+#include <thrust/iterator/iterator_traits.h>
+
+#include <thrust/system/cuda/detail/detail/stable_sort_by_count.h>
+#include <thrust/system/cuda/detail/runtime_introspection.h>
+#include <thrust/detail/temporary_array.h>
+#include <thrust/system/cuda/detail/detail/launch_closure.h>
+#include <thrust/system/cuda/detail/detail/uninitialized.h>
+#include <thrust/system/cuda/detail/detail/cached_temporary_allocator.h>
+#include <thrust/system/cuda/detail/block/merge.h>
+#include <thrust/system/cuda/detail/block/copy.h>
+#include <thrust/pair.h>
+#include <thrust/tuple.h>
+#include <thrust/detail/minmax.h>
+#include <thrust/iterator/counting_iterator.h>
+#include <thrust/iterator/transform_iterator.h>
+#include <thrust/gather.h>
+
+__THRUST_DISABLE_MSVC_POSSIBLE_LOSS_OF_DATA_WARNING_BEGIN
+
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+namespace detail
+{
+namespace stable_merge_sort_detail
+{
+
+
+template<unsigned int log_block_size, typename Key, typename Value>
+  struct is_block_size_valid
+{
+  // assume sm_10 limits
+  static const unsigned int max_num_smem_bytes = 16384;
+
+  // CUDA steals 256 for itself for kernel parms
+  static const unsigned int num_reserved_smem_bytes = 256;
+
+  // the number of bytes available to our kernels
+  static const unsigned int num_available_smem_bytes = max_num_smem_bytes - num_reserved_smem_bytes;
+
+  // merge_small_tiles_by_key_closure is the hungriest kernel
+  // the block_size it uses is 2x the size of all the other kernels
+  // this merge_small_tiles_by_key_closure's smem requirements:
+  //   2 * block_size_x2 * sizeof(Key)
+  // + 2 * block_size_x2 * sizeof(Key)
+  // + 2 * block_size_x2 * sizeof(Value)
+  // ================================
+  // 4 * (block_size) * (2 * sizeof(Key) + sizeof(Value))
+  static const unsigned int num_needed_smem_bytes = 4 * (1 << log_block_size) * (2 * sizeof(Key) + sizeof(Value));
+
+  static const bool value = num_needed_smem_bytes <= num_available_smem_bytes;
+};
+
+
+
+// choose a (log) block_size to use for our kernels
+template<unsigned int log_preferred_block_size, typename Key, typename Value>
+  struct select_log_block_size
+    : thrust::detail::eval_if<
+        is_block_size_valid<log_preferred_block_size, Key, Value>::value,
+        thrust::detail::integral_constant<unsigned int, log_preferred_block_size>,
+        select_log_block_size<log_preferred_block_size - 1, Key, Value>
+      >::type
+{};
+
+
+// don't recurse lower than block_size < 128
+template<typename Key, typename Value>
+  struct select_log_block_size<6, Key, Value>
+{
+  // no block size exists which can satisfy the storage demands
+};
+
+
+template<typename Key, typename Value>
+  struct block_size
+{
+  // prefer block_size == 512, go lower if we need to
+  static const unsigned int value = 1 << select_log_block_size<8, Key, Value>::value;
+};
+
+
+template <typename Size>
+inline unsigned int max_grid_size(Size block_size)
+{
+  const device_properties_t& properties = device_properties();
+
+  const unsigned int max_threads = properties.maxThreadsPerMultiProcessor * properties.multiProcessorCount;
+  const unsigned int max_blocks  = properties.maxGridSize[0];
+  
+  return std::min<unsigned int>(max_blocks, 3 * max_threads / block_size);
+} // end max_grid_size()
+
+
+// Base case for the merge algorithm: merges data where tile_size <= block_size. 
+// Works by loading two or more tiles into shared memory and doing a binary search.
+template<unsigned int block_size,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2,
+         typename RandomAccessIterator3,
+         typename RandomAccessIterator4,
+         typename StrictWeakOrdering,
+         typename Context>
+struct merge_small_tiles_by_key_closure
+{
+  typedef Context context_type;
+
+  RandomAccessIterator1 keys_first;
+  RandomAccessIterator2 values_first;
+  const unsigned int n;
+  const unsigned int log_tile_size;
+  RandomAccessIterator3 keys_result;
+  RandomAccessIterator4 values_result;
+  StrictWeakOrdering comp;
+  context_type context;
+
+  // these members are derivable from block_size, n, and log_tile_size
+  unsigned int index_of_last_block;
+  unsigned int index_of_last_tile_in_last_block;
+  unsigned int size_of_last_tile;
+
+  merge_small_tiles_by_key_closure
+    (RandomAccessIterator1 keys_first,
+     RandomAccessIterator2 values_first,
+     const unsigned int n,
+     const unsigned int log_tile_size,
+     RandomAccessIterator3 keys_result,
+     RandomAccessIterator4 values_result,
+     StrictWeakOrdering comp,
+     Context context = Context())
+    : keys_first(keys_first), values_first(values_first),
+      n(n), 
+      log_tile_size(log_tile_size),
+      keys_result(keys_result), values_result(values_result),
+      comp(comp),
+      context(context)
+  {
+    // compute the number of tiles, including a possible partial tile
+    unsigned int tile_size = 1 << log_tile_size;
+    unsigned int num_tiles = thrust::detail::util::divide_ri(n, tile_size);
+    unsigned int partial_tile_size = n % tile_size;
+
+    // compute the number of logical thread blocks, including a possible partial block
+    unsigned int tiles_per_block = block_size / tile_size;
+    unsigned int num_blocks = thrust::detail::util::divide_ri(num_tiles, tiles_per_block);
+    unsigned int partial_block_size = num_tiles % tiles_per_block;
+
+    // compute the number of tiles in the last block, which might be of partial size
+    unsigned int number_of_tiles_in_last_block = partial_block_size ? partial_block_size : tiles_per_block;
+
+    size_of_last_tile = partial_tile_size ? partial_tile_size : tile_size;
+    index_of_last_tile_in_last_block = number_of_tiles_in_last_block - 1;
+    index_of_last_block = num_blocks - 1;
+  }
+
+  unsigned int grid_size() const
+  {
+    const unsigned int max_num_blocks = max_grid_size(block_size);
+    const unsigned int num_logical_blocks = index_of_last_block + 1;
+    return thrust::min<unsigned int>(num_logical_blocks, max_num_blocks);
+  }
+
+  __device__ __thrust_forceinline__
+  void operator()(void)
+  {
+    typedef typename iterator_value<RandomAccessIterator3>::type KeyType;
+    typedef typename iterator_value<RandomAccessIterator4>::type ValueType;
+
+    // load (2*block_size) elements into shared memory. These (2*block_size) elements belong to (2*block_size)/tile_size different tiles.
+    __shared__ uninitialized_array<KeyType, 2 * block_size>   key;
+    __shared__ uninitialized_array<KeyType, 2 * block_size>   outkey;
+    __shared__ uninitialized_array<ValueType, 2 * block_size> outvalue;
+
+    const unsigned int grid_size = context.grid_dimension() * context.block_dimension();
+
+    unsigned int block_idx = context.block_index();
+    
+    // the global index of this task
+    unsigned int i = context.thread_index() + context.block_index() * context.block_dimension();
+
+    // advance iterators
+    keys_first    += i;
+    values_first  += i;
+    keys_result   += i;
+    values_result += i;
+
+    for(;
+        block_idx <= index_of_last_block;
+        block_idx += context.grid_dimension(), i += grid_size, keys_first += grid_size, values_first += grid_size, keys_result += grid_size, values_result += grid_size)
+    {
+      KeyType my_key;
+      
+      // copy over inputs to shared memory
+      if(i < n)
+      {
+        key[context.thread_index()] = my_key = *keys_first;
+      } // end if
+      
+      // the tile to which the element belongs
+      unsigned int tile_index = context.thread_index()>>log_tile_size;
+
+      // figure out the index and size of the other tile
+      unsigned int other_tile_index = tile_index^1;
+      unsigned int other_tile_size = (1<<log_tile_size);
+
+      // if the other tile is the final tile, it is potentially
+      // smaller than the rest
+      if(block_idx == index_of_last_block
+         && other_tile_index == index_of_last_tile_in_last_block)
+      {
+        other_tile_size = size_of_last_tile;
+      } // end if
+      
+      // figure out where the other tile begins in shared memory
+      KeyType *other = key.data() + (other_tile_index<<log_tile_size);
+
+      context.barrier();
+      if(i < n)
+      {
+        // to compute the rank of my element in the merged sequence
+        // add the rank of the element in the other tile
+        // plus the rank of the element in this tile
+        // the computation for the rank of the element in this tile 
+        // differs depending on if we're in the odd or even tile
+        unsigned int rank;
+        if(tile_index & 1)
+        {
+          rank = thrust::system::detail::generic::scalar::upper_bound_n(other, other_tile_size, my_key, comp) - other;
+          rank += context.thread_index() - (1<<log_tile_size);
+        }
+        else
+        {
+          rank = thrust::system::detail::generic::scalar::lower_bound_n(other, other_tile_size, my_key, comp) - other;
+          rank += context.thread_index();
+        }
+
+        // store my key and value to the output arrays in smem
+        outkey[rank] = my_key;
+        outvalue[rank] = *values_first;
+      } // end if
+      context.barrier();
+      
+      if(i < n)
+      {
+        // coalesced writes to global memory
+        *keys_result   = outkey[context.thread_index()];
+        *values_result = outvalue[context.thread_index()];
+      } // end if
+      context.barrier();
+    } // end for
+  } // end operator()
+}; // merge_small_tiles_by_key_closure
+
+
+template<unsigned int stride>
+  class static_strided_integer_range
+{
+  // XXX cudafe doesn't like this private for some reason
+  //private:
+  public:
+    typedef typename thrust::counting_iterator<unsigned int> counting_iterator;
+
+    struct stride_functor
+      : public thrust::unary_function<unsigned int,unsigned int>
+    {
+      inline __host__ __device__
+      unsigned int operator()(unsigned int i) const
+      {
+        return stride * i;
+      }
+    };
+
+  public:
+    typedef typename thrust::transform_iterator<stride_functor, counting_iterator> iterator;
+
+    static_strided_integer_range(unsigned int num_strides)
+      : m_begin(iterator(counting_iterator(0), stride_functor())),
+        m_end(iterator(counting_iterator(num_strides), stride_functor()))
+    {}
+
+    iterator begin() const
+    {
+      return m_begin;
+    }
+
+    iterator end() const
+    {
+      return m_end;
+    }
+
+  private:
+    iterator m_begin, m_end;
+};
+
+
+///////////////////// Find the rank of each extracted element in both arrays ////////////////////////////////////////
+///////////////////// This breaks up the array into independent segments to merge ////////////////////////////////////////
+// Inputs: d_splitters, d_splittes_pos: the merged array of splitters with corresponding positions.
+//		   d_srcData: input data, datasize: number of entries in d_srcData
+//		   N_SPLITTERS the number of splitters, log_blocksize: log of the size of each block of sorted data
+//		   log_num_merged_splitters_per_tile = log of the number of merged splitters. ( = log_blocksize - 7). 
+// Output: d_rank1, d_rank2: ranks of each splitter in d_splitters in the block to which it belongs
+//		   (say i) and its corresponding block (block i+1).
+template<unsigned int block_size,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2,
+         typename RandomAccessIterator3,
+         typename RandomAccessIterator4,
+         typename StrictWeakOrdering,
+         typename Context>
+struct rank_splitters_closure
+{
+  typedef Context context_type;
+
+  static const unsigned int log_block_size = thrust::detail::mpl::math::log2<block_size>::value;
+
+  RandomAccessIterator1 splitters_first;
+  RandomAccessIterator2 splitters_pos_first;
+  RandomAccessIterator3 keys_first;
+  RandomAccessIterator4 ranks_result1;
+  RandomAccessIterator4 ranks_result2;
+  unsigned int num_splitters;
+  unsigned int num_keys;
+  unsigned int log_tile_size;
+  thrust::detail::device_function<
+    StrictWeakOrdering,
+    bool
+  > comp;
+  context_type context;
+
+  // this member is derivable from those received in the constructor
+  unsigned int log_num_merged_splitters_per_tile;
+
+  rank_splitters_closure(RandomAccessIterator1 splitters_first,
+                         RandomAccessIterator2 splitters_pos_first, 
+                         RandomAccessIterator3 keys_first,
+                         unsigned int num_splitters,
+                         unsigned int num_keys, 
+                         unsigned int log_tile_size, 
+                         RandomAccessIterator4 ranks_result1,
+                         RandomAccessIterator4 ranks_result2, 
+                         StrictWeakOrdering comp,
+                         context_type context = context_type())
+    : splitters_first(splitters_first), splitters_pos_first(splitters_pos_first),
+      keys_first(keys_first),
+      ranks_result1(ranks_result1), ranks_result2(ranks_result2),
+      num_splitters(num_splitters), num_keys(num_keys),
+      log_tile_size(log_tile_size),
+      comp(comp), context(context)
+  {
+    // the number of splitters in each tile before merging
+    const unsigned int log_num_splitters_per_tile = log_tile_size - log_block_size;
+
+    // the number of splitters in each merged tile
+    log_num_merged_splitters_per_tile = log_num_splitters_per_tile + 1;
+  }
+
+  inline unsigned int grid_size() const
+  {
+    unsigned int num_blocks = num_splitters / block_size;
+    if(num_splitters % block_size) ++num_blocks;
+
+    // compute the maximum number of block_size we can launch on this arch
+    const unsigned int max_num_blocks = max_grid_size(block_size);
+
+    return min<unsigned int>(num_blocks, max_num_blocks);
+  }
+
+  /*! this member function returns the index of the (odd,even) block pair
+   *  that the splitter of interest belongs to
+   *  \param splitter_idx The index of the splitter in the splitters list
+   *  \return The global index of the (odd,even) block pair
+   */
+  __device__ __thrust_forceinline__
+  unsigned int block_pair_idx(unsigned int splitter_idx) const
+  {
+    return splitter_idx >> log_num_merged_splitters_per_tile;
+  }
+
+  /*! This member function returns the end of the search range in the other tile in
+   *  which the splitter of interest needs to be ranked.
+   *  \param splitter_idx The index of the splitter in the splitters array
+   *  \param splitter_global_idx The index of the splitter in the global array of elements
+   *  \param tile_idx The index of the tile to which the splitter belongs.
+   *  \return The half-open interval in the other tile in which the splitter needs to be ranked.
+   *          [first_index_to_search, size_of_interval)
+   */
+  __device__ __thrust_forceinline__
+  thrust::pair<unsigned int,unsigned int> search_interval(unsigned int splitter_idx, unsigned int splitter_global_idx, unsigned int tile_idx) const
+  {
+    // We want to compute the ranks of the splitter in d_srcData1 and d_srcData2
+    // for instance, if the splitter belongs to d_srcData1, then 
+    // (1) the rank in d_srcData1 is simply given by its splitter_global_idx
+    // (2) to find the rank in d_srcData2, we first find the block in d_srcData2 where inp appears.
+    //     We do this by noting that we have already merged/sorted splitters, and thus the rank
+    //     of inp in the elements of d_srcData2 that are present in splitters is given by 
+    //        position of inp in d_splitters - rank of inp in elements of d_srcData1 in splitters
+    //        = i - splitter_global_idx
+    //     This also gives us the block of d_srcData2 that the splitter belongs in, since we have one
+    //     element in splitters per block of d_srcData2.
+    
+    //     We now perform a binary search over this block of d_srcData2 to find the rank of inp in d_srcData2.
+    //     start and end are the start and end indices of this block in d_srcData2, forming the bounds of the binary search.
+    //     Note that this binary search is in global memory with uncoalesced loads. However, we only find the ranks 
+    //     of a small set of elements, one per splitter: thus it is not the performance bottleneck.
+    
+    // the local index of the splitter within the (odd, even) block pair.
+    const unsigned int splitter_block_pair_idx = splitter_idx - (block_pair_idx(splitter_idx)<<log_num_merged_splitters_per_tile);
+
+    // the index of the splitter within its tile
+    const unsigned int splitter_tile_idx = splitter_global_idx - (tile_idx<<log_tile_size);
+
+    // the index of the splitter's block within its tile
+    const unsigned int block_tile_idx = splitter_tile_idx >> log_block_size;
+    
+    // find the end of the search range in the other tile
+    unsigned int end = (( splitter_block_pair_idx - block_tile_idx) << log_block_size);
+
+    // begin by assuming the search range is the size of a full block
+    unsigned int other_block_size = block_size;
+
+    // the index of the other tile can be found with
+    const unsigned int other_tile_idx = tile_idx ^ 1;
+    
+    // the size of the other tile can be less than tile_size if the it is the last tile.
+    unsigned int other_tile_size = min<unsigned int>(1 << log_tile_size, num_keys - (other_tile_idx<<log_tile_size));
+
+    if(end > other_tile_size)
+    {
+      // the other block has partial size
+      end = other_tile_size;
+      other_block_size = num_keys % block_size;
+    }
+    else if(end == 0)
+    {
+      // when the search range is empty
+      // the other_block_size is 0
+      other_block_size = 0;
+    }
+
+    // the search range begins other_block_size elements before the end
+    unsigned int start = end - other_block_size;
+
+    return thrust::make_pair(start,other_block_size);
+  }
+
+  __device__ __thrust_forceinline__
+  void operator()(void)
+  {
+    typedef typename iterator_value<RandomAccessIterator1>::type KeyType;
+    typedef typename iterator_value<RandomAccessIterator2>::type IndexType;
+  
+    const unsigned int grid_size = context.grid_dimension() * context.block_dimension();
+  
+    unsigned int splitter_idx = context.thread_index() + context.block_index() * context.block_dimension();
+  
+    // advance iterators
+    splitters_first     += splitter_idx;
+    splitters_pos_first += splitter_idx;
+    ranks_result1       += splitter_idx;
+    ranks_result2       += splitter_idx;
+    
+    for(;
+        splitter_idx < num_splitters;
+        splitter_idx += grid_size, splitters_first += grid_size, splitters_pos_first += grid_size, ranks_result1 += grid_size, ranks_result2 += grid_size)
+    {
+      // the index of the splitter within the global array of elements
+      IndexType splitter_global_idx = *splitters_pos_first;
+
+      // the tile to which the splitter belongs.
+      unsigned int tile_idx = (splitter_global_idx >> log_tile_size);
+      
+      // the index of the "other" tile which which tile_idx must be merged.
+      unsigned int other_tile_idx = tile_idx^1;
+
+      // compute the interval in the other tile to search
+      unsigned int start, n;
+      thrust::tie(start,n) = search_interval(splitter_idx, splitter_global_idx, tile_idx);
+
+      // point to the beginning of the other tile
+      RandomAccessIterator3 other_tile_begin = keys_first + (other_tile_idx<<log_tile_size);
+
+      // offset the pointer to the other tile by the search range's offset
+      RandomAccessIterator3 search_range_begin = other_tile_begin + start;
+      
+      // find the rank of our splitter in the other tile
+      KeyType splitter = *splitters_first;
+
+      // the index of the splitter within its tile
+      // this is one of the output ranks
+      const unsigned int splitter_tile_idx = splitter_global_idx - (tile_idx<<log_tile_size);
+
+      // branch depending on whether or not our splitter is in the odd tile
+      if(tile_idx & 1)
+      {
+        unsigned int result = thrust::system::detail::generic::scalar::upper_bound_n(search_range_begin, n, splitter, comp) - search_range_begin;
+
+        *ranks_result1 = start + result;
+
+        *ranks_result2 = splitter_tile_idx;
+      } // end if
+      else
+      {
+        unsigned int result = thrust::system::detail::generic::scalar::lower_bound_n(search_range_begin, n, splitter, comp) - search_range_begin;
+
+        *ranks_result1 = splitter_tile_idx;
+
+        *ranks_result2 = start + result;
+      } // end else
+    } // end for
+  } // end operator()
+}; // rank_splitters_closure
+
+
+template<unsigned int block_size,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2,
+         typename RandomAccessIterator3,
+         typename RandomAccessIterator4,
+         typename StrictWeakOrdering>
+  void rank_splitters(RandomAccessIterator1 splitters_first,
+                      RandomAccessIterator1 splitters_last,
+                      RandomAccessIterator2 splitter_positions_first,
+                      RandomAccessIterator3 keys_first,
+                      RandomAccessIterator3 keys_last,
+                      size_t log_tile_size,
+                      RandomAccessIterator4 ranks_result1,
+                      RandomAccessIterator4 ranks_result2,
+                      StrictWeakOrdering comp)
+{
+  typedef rank_splitters_closure<
+    block_size,
+    RandomAccessIterator1,
+    RandomAccessIterator2,
+    RandomAccessIterator3,
+    RandomAccessIterator4,
+    StrictWeakOrdering,
+    detail::statically_blocked_thread_array<block_size>
+  > Closure;
+
+  Closure closure(splitters_first,
+                  splitter_positions_first,
+                  keys_first,
+                  splitters_last - splitters_first,
+                  keys_last - keys_first,
+                  log_tile_size,
+                  ranks_result1,
+                  ranks_result2,
+                  comp);
+
+  detail::launch_closure(closure, closure.grid_size(), block_size);
+}
+
+
+template<typename Context,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2,
+         typename Size,
+         typename RandomAccessIterator3,
+         typename RandomAccessIterator4>
+__device__
+  void copy_n(Context context,
+              RandomAccessIterator1 first1,
+              RandomAccessIterator2 first2,
+              Size n,
+              RandomAccessIterator3 result1,
+              RandomAccessIterator4 result2)
+{
+  for(Size i = context.thread_index();
+      i < n;
+      i += context.block_dimension())
+  {
+    result1[i] = first1[i];
+    result2[i] = first2[i];
+  }
+}
+
+
+///////////////////// MERGE TWO INDEPENDENT SEGMENTS USING BINARY SEARCH IN SHARED MEMORY ////////////////////////////////////////
+// NOTE: This is the most compute-intensive part of the algorithm. 
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+// Thread block i merges entries between rank[i] and rank[i+1]. These can be independently
+// merged and concatenated, as noted above. 
+// Each thread in the thread block i does a binary search of one element between rank[i] -> rank[i+1] in the 
+// other array. 
+
+// Inputs: srcdatakey, value: inputs
+//         log_blocksize, log_num_merged_splitters_per_tile: as in previous functions
+// Outputs: resultdatakey, resultdatavalue: output merged arrays are written here.
+template<unsigned int block_size,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2,
+         typename RandomAccessIterator3,
+         typename RandomAccessIterator4,
+         typename RandomAccessIterator5,
+         typename RandomAccessIterator6,
+         typename StrictWeakOrdering,
+         typename Context>
+struct merge_subtiles_by_key_closure
+{	
+  typedef Context context_type;
+  static const unsigned int log_block_size = thrust::detail::mpl::math::log2<block_size>::value;
+
+  RandomAccessIterator1 keys_first;
+  RandomAccessIterator2 values_first;
+  unsigned int n;
+  RandomAccessIterator3 ranks_first1;
+  RandomAccessIterator4 ranks_first2; 
+  const unsigned int tile_size;
+  const unsigned int num_splitters;
+  RandomAccessIterator5 keys_result;
+  RandomAccessIterator6 values_result;
+  StrictWeakOrdering comp;
+  Context context;
+
+  // this member is derivable from the constructor parameters
+  unsigned int log_num_merged_splitters_per_tile;
+
+  merge_subtiles_by_key_closure
+    (RandomAccessIterator1 keys_first,
+     RandomAccessIterator2 values_first,
+     unsigned int n, 
+     RandomAccessIterator3 ranks_first1,
+     RandomAccessIterator4 ranks_first2, 
+     const unsigned int log_tile_size, 
+     const unsigned int num_splitters,
+     RandomAccessIterator5 keys_result,
+     RandomAccessIterator6 values_result,
+     StrictWeakOrdering comp,
+     Context context = Context())
+    : keys_first(keys_first), values_first(values_first), n(n),
+      ranks_first1(ranks_first1), ranks_first2(ranks_first2),
+      tile_size(1 << log_tile_size),
+      num_splitters(num_splitters),
+      keys_result(keys_result), values_result(values_result),
+      comp(comp), context(context)
+  {
+    const unsigned int log_num_splitters_per_tile = log_tile_size - log_block_size;
+    log_num_merged_splitters_per_tile = log_num_splitters_per_tile + 1;
+  }
+
+  unsigned int grid_size() const
+  {
+    const unsigned int max_num_blocks = max_grid_size(block_size);
+    return thrust::min<unsigned int>(num_splitters, max_num_blocks);
+  }
+
+  __device__ __thrust_forceinline__
+  unsigned int even_offset(unsigned int oddeven_blockid) const
+  {
+    return oddeven_blockid << (log_num_merged_splitters_per_tile + log_block_size);
+  }
+
+  __device__ __thrust_forceinline__
+  void get_partition(unsigned int partition_idx, unsigned int oddeven_blockid,
+                     unsigned int &rank1, unsigned int &size1,
+                     unsigned int &rank2, unsigned int &size2) const
+  {
+    // XXX this logic would be much improved if we were guaranteed that there was 
+    //     an element at ranks_first[1]
+    // XXX we could eliminate the need for local_blockIdx, log_num_merged_splitters_per_block, tile_size, and n
+    
+    // the index of the merged splitter within the splitters for the odd-even block pair.
+    unsigned int local_blockIdx = partition_idx - (oddeven_blockid<<log_num_merged_splitters_per_tile);
+
+    rank1 = *ranks_first1;
+    rank2 = *ranks_first2;
+  
+    // get the rank of the next splitter if we aren't processing the very last splitter of a partially full tile
+    // or if we aren't processing the last splitter in our tile
+    if((partition_idx == num_splitters - 1) || (local_blockIdx == ((1<<log_num_merged_splitters_per_tile)-1)))
+    {
+      // we're at the end
+      size1 = size2 = tile_size;
+    } // end if
+    else
+    {
+      // dereference the rank of the *next* splitter
+      size1 = ranks_first1[1];
+      size2 = ranks_first2[1];
+    } // end else
+    
+    // Adjust size2 to account for the last block possibly not being full.
+    // check if size2 would fall off the end of the array
+    if((even_offset(oddeven_blockid) + tile_size + size2) > n)
+    {
+      size2 = n - tile_size - even_offset(oddeven_blockid);
+    } // end if
+  
+    // measure each array relative to its beginning
+    size1 -= rank1;
+    size2 -= rank2;
+  }
+
+  template<typename KeyType, typename ValueType>
+  __device__ __thrust_forceinline__
+  void do_it(KeyType *s_keys, ValueType *s_values)
+  {
+    // advance iterators
+    unsigned int i = context.block_index();
+    ranks_first1 += i;
+    ranks_first2 += i;
+    
+    // Thread Block i merges the sub-block associated with splitter i: rank[i] -> rank[i+1] in a particular odd-even block pair.
+    for(;
+        i < num_splitters;
+        i += context.grid_dimension(), ranks_first1 += context.grid_dimension(), ranks_first2 += context.grid_dimension())
+    {
+      // the (odd, even) block pair that the splitter belongs to.
+      unsigned int oddeven_blockid = i >> log_num_merged_splitters_per_tile;
+      
+      // start1 & start2 store rank[i] and rank[i+1] indices in arrays 1 and 2.
+      // size1 & size2 store the number of of elements between rank[i] & rank[i+1] in arrays 1 & 2.
+      unsigned int rank1, rank2, size1, size2;
+      get_partition(i, oddeven_blockid, rank1, size1, rank2, size2);
+  
+      // find where the odd,even arrays begin
+      RandomAccessIterator1 even_keys_first = keys_first + even_offset(oddeven_blockid);
+      RandomAccessIterator1 odd_keys_first  = even_keys_first + tile_size;
+  
+      RandomAccessIterator2 even_values_first = values_first + even_offset(oddeven_blockid);
+      RandomAccessIterator2 odd_values_first  = even_values_first + tile_size;
+      
+      // load tiles into smem
+      copy_n(context, even_keys_first + rank1, even_values_first + rank1, size1, s_keys, s_values);
+      copy_n(context, odd_keys_first  + rank2, odd_values_first  + rank2, size2, s_keys + size1, s_values + size1);
+
+      context.barrier();
+  
+      // merge the arrays in-place
+      block::inplace_merge_by_key_n(context, s_keys, s_values, size1, size2, comp);
+
+      context.barrier();
+      
+      // write tiles to gmem
+      unsigned int dst_offset = even_offset(oddeven_blockid) + rank1 + rank2;
+      copy_n(context, s_keys, s_values, size1 + size2, keys_result + dst_offset, values_result + dst_offset);
+
+      context.barrier();
+    } // end for i
+  }
+
+  __device__ __thrust_forceinline__
+  void operator()(void)
+  {
+    typedef typename iterator_value<RandomAccessIterator5>::type KeyType;
+    typedef typename iterator_value<RandomAccessIterator6>::type ValueType;
+  
+    __shared__ uninitialized_array<KeyType,   2 * block_size> s_keys;
+    __shared__ uninitialized_array<ValueType, 2 * block_size> s_values;
+  
+    do_it(s_keys.data(), s_values.data());
+  }
+}; // merge_subtiles_by_key_closure
+
+// merge_subtiles_by_key() merges each sub-tile independently. As explained in rank_splitters(), 
+// the sub-tiles are defined by the ranks of the splitter elements d_rank1 and d_rank2 in the odd and even tiles resp.
+// It can be easily shown that each sub-tile cannot contain more than block_size elements of either the odd or even tile.
+
+// the function calls merge_subblocks_binarysearch_kernel() for the remaining N_splitterS sub-tiles
+// We use 1 thread block per splitter: For instance, thread block 0 will merge rank1[0] -> rank1[1] of array i with
+// rank2[0] -> rank2[1] of array i^1, with i being the thread block to which the splitter belongs.
+
+// We implement each sub-tile merge using a binary search. We compute the rank of each element belonging to a sub-tile 
+// of an odd numbered tile in the corresponding sub-tile of its even numbered pair. It then adds this rank to 
+// the index of the element in its own sub-tile to find the output index of the element in the merged sub-tile.
+
+template<typename RandomAccessIterator1,
+         typename RandomAccessIterator2,
+         typename RandomAccessIterator3,
+         typename RandomAccessIterator4,
+         typename RandomAccessIterator5,
+         typename RandomAccessIterator6,
+         typename RandomAccessIterator7,
+         typename StrictWeakOrdering>
+  void merge_subtiles_by_key(RandomAccessIterator1 keys_first,
+                             RandomAccessIterator1 keys_last,
+                             RandomAccessIterator2 values_first,
+                             RandomAccessIterator3 splitters_pos_first, 
+                             RandomAccessIterator3 splitters_pos_last,
+                             RandomAccessIterator4 ranks_first1,
+                             RandomAccessIterator5 ranks_first2, 
+                             RandomAccessIterator6 keys_result,
+                             RandomAccessIterator7 values_result, 
+                             unsigned int log_tile_size, 
+                             StrictWeakOrdering comp)
+{
+  typedef typename iterator_value<RandomAccessIterator6>::type KeyType;
+  typedef typename iterator_value<RandomAccessIterator7>::type ValueType;
+
+  const unsigned int block_size = stable_merge_sort_detail::block_size<KeyType,ValueType>::value;
+
+  typedef merge_subtiles_by_key_closure<
+    block_size,
+    RandomAccessIterator1,
+    RandomAccessIterator2,
+    RandomAccessIterator4,
+    RandomAccessIterator5,
+    RandomAccessIterator6,
+    RandomAccessIterator7,
+    StrictWeakOrdering,
+    detail::statically_blocked_thread_array<block_size>
+  > Closure;
+
+  Closure closure(keys_first,
+                  values_first,
+                  keys_last - keys_first, 
+                  ranks_first1,
+                  ranks_first2, 
+                  log_tile_size,
+                  splitters_pos_last - splitters_pos_first,
+  	          keys_result,
+                  values_result,
+                  comp);
+
+  detail::launch_closure(closure, closure.grid_size(), block_size);
+}
+
+
+template<unsigned int block_size,
+         typename DerivedPolicy,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2,
+         typename RandomAccessIterator3,
+         typename RandomAccessIterator4,
+         typename StrictWeakOrdering>
+  void merge_small_tiles_by_key(execution_policy<DerivedPolicy> &,
+                                RandomAccessIterator1 keys_first,
+                                RandomAccessIterator1 keys_last,
+                                RandomAccessIterator2 values_first,
+                                size_t log_tile_size,
+                                RandomAccessIterator3 keys_result,
+                                RandomAccessIterator4 values_result,
+                                StrictWeakOrdering comp)
+{
+  typedef merge_small_tiles_by_key_closure<
+    block_size,
+    RandomAccessIterator1,
+    RandomAccessIterator2,
+    RandomAccessIterator3,
+    RandomAccessIterator4,
+    StrictWeakOrdering,
+    detail::statically_blocked_thread_array<block_size>
+  > Closure;
+
+  Closure closure(keys_first, values_first, keys_last - keys_first, log_tile_size, keys_result, values_result, comp);
+
+  detail::launch_closure(closure, closure.grid_size(), block_size);
+} // end merge_small_tiles_by_key()
+
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2,
+         typename RandomAccessIterator3,
+         typename RandomAccessIterator4,
+         typename StrictWeakOrdering>
+  void merge_tiles_by_key_recursive(execution_policy<DerivedPolicy> &exec,
+                                    RandomAccessIterator1 keys_first,
+                                    RandomAccessIterator1 keys_last,
+                                    RandomAccessIterator2 values_first,
+                                    RandomAccessIterator3 keys_result,
+                                    RandomAccessIterator4 values_result,
+                                    size_t log_tile_size,
+                                    StrictWeakOrdering comp)
+{
+  typedef typename iterator_value<RandomAccessIterator3>::type KeyType;
+  typedef typename iterator_value<RandomAccessIterator4>::type ValueType;
+
+  const size_t tile_size = 1<<log_tile_size;
+
+  // Compute the block_size based on the types to sort
+  const unsigned int block_size = stable_merge_sort_detail::block_size<KeyType,ValueType>::value;
+
+  // Case (a): tile_size <= block_size
+  if(tile_size <= block_size)
+  {
+    return merge_small_tiles_by_key<2*block_size>(exec, keys_first, keys_last, values_first, log_tile_size, keys_result, values_result, comp);
+  } // end if
+
+  // Case (b) tile_size >= block_size
+
+  // step 1 of the recursive case: gather one splitter per block_size entries in each odd-even tile pair.
+  thrust::detail::temporary_array<KeyType, DerivedPolicy> splitters(exec, thrust::detail::util::divide_ri(keys_last - keys_first, block_size));
+  static_strided_integer_range<block_size>                splitters_pos(splitters.size());
+  thrust::gather(exec, splitters_pos.begin(), splitters_pos.end(), keys_first, splitters.begin());
+                            
+  // step 2 of the recursive case: merge the splitters & their positions
+  thrust::detail::temporary_array<KeyType,      DerivedPolicy> merged_splitters(exec, splitters.size());
+  thrust::detail::temporary_array<unsigned int, DerivedPolicy> merged_splitters_pos(exec, splitters.size());
+
+  const unsigned int log_block_size = thrust::detail::mpl::math::log2<block_size>::value;
+  size_t log_num_splitters_per_tile = log_tile_size - log_block_size;
+  merge_tiles_by_key_recursive(exec,
+                               splitters.begin(),
+                               splitters.end(),
+                               splitters_pos.begin(),
+                               merged_splitters.begin(),
+                               merged_splitters_pos.begin(),
+                               log_num_splitters_per_tile,
+                               comp);
+
+  // step 3 of the recursive case: find the ranks of each splitter in the respective two tiles.
+  // reuse the merged_splitters_pos storage
+  thrust::detail::temporary_array<unsigned int, DerivedPolicy> &rank1 = merged_splitters_pos;
+  thrust::detail::temporary_array<unsigned int, DerivedPolicy> rank2(exec, rank1.size());
+
+  rank_splitters<block_size>(merged_splitters.begin(),
+                             merged_splitters.end(),
+                             merged_splitters_pos.begin(),
+                             keys_first,
+                             keys_last,
+                             log_tile_size,
+                             rank1.begin(),
+                             rank2.begin(),
+                             comp);
+
+  // step 4 of the recursive case: merge each sub-tile independently in parallel.
+  merge_subtiles_by_key(keys_first,
+                        keys_last,
+                        values_first,
+                        merged_splitters_pos.begin(),
+                        merged_splitters_pos.end(),
+                        rank1.begin(),
+                        rank2.begin(),
+                        keys_result,
+                        values_result,
+                        log_tile_size,
+                        comp);
+}
+
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2,
+         typename RandomAccessIterator3,
+         typename RandomAccessIterator4,
+         typename StrictWeakOrdering>
+  void merge_tiles_by_key(execution_policy<DerivedPolicy> &exec,
+                          RandomAccessIterator1 keys_first,
+                          RandomAccessIterator2 values_first,
+                          size_t n,
+                          RandomAccessIterator3 keys_result,
+                          RandomAccessIterator4 values_result,
+                          unsigned int log_tile_size,
+                          StrictWeakOrdering comp)
+{
+  const unsigned int tile_size = 1 << log_tile_size;
+  const size_t num_tiles = thrust::detail::util::divide_ri(n, tile_size);
+
+  // if there is an odd number of tiles, we should exclude the last one
+  // without a twin in merge_recursive
+  const size_t last_tile_offset = (num_tiles%2)?((num_tiles-1)*tile_size):n;
+
+  merge_tiles_by_key_recursive(exec,
+                               keys_first,
+                               keys_first + last_tile_offset,
+                               values_first,
+                               keys_result,
+                               values_result,
+                               log_tile_size,
+                               comp);
+
+  // copy the last tile without a twin, should it exist
+  if(last_tile_offset < n)
+  {
+    thrust::copy(exec, keys_first + last_tile_offset, keys_first + n, keys_result + last_tile_offset);
+    thrust::copy(exec, values_first + last_tile_offset, values_first + n, values_result + last_tile_offset);
+  } // end if
+} // end merge_tiles_by_key()
+
+
+} // end stable_merge_sort_detail
+
+
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator,
+         typename StrictWeakOrdering>
+void stable_merge_sort(execution_policy<DerivedPolicy> &exec,
+                       RandomAccessIterator first,
+                       RandomAccessIterator last,
+                       StrictWeakOrdering comp)
+{
+  // XXX it's potentially unsafe to pass the same array for keys & values
+  thrust::system::cuda::detail::detail::stable_merge_sort_by_key(exec, first, last, first, comp);
+}
+
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2,
+         typename StrictWeakOrdering>
+  void stable_merge_sort_by_key(execution_policy<DerivedPolicy> &exec,
+                                RandomAccessIterator1 keys_first,
+                                RandomAccessIterator1 keys_last,
+                                RandomAccessIterator2 values_first,
+                                StrictWeakOrdering comp)
+{
+  typedef typename thrust::iterator_traits<RandomAccessIterator1>::value_type KeyType;
+  typedef typename thrust::iterator_traits<RandomAccessIterator2>::value_type ValueType;
+
+  // compute the block_size based on the types we're sorting
+  const unsigned int block_size = stable_merge_sort_detail::block_size<KeyType,ValueType>::value;
+
+  // XXX WAR unused variable warning issued by nvcc
+  (void) block_size;
+
+  // first, sort each tile of block_size elements
+  stable_sort_by_count<block_size>(exec, keys_first, keys_last, values_first, comp);
+
+  // merge tiles if there is more than one
+  const size_t n = keys_last - keys_first;
+  if(n > block_size)
+  {
+    // allocate scratch space
+    using namespace thrust::detail;
+    using namespace stable_merge_sort_detail;
+    temporary_array<KeyType,   DerivedPolicy> temp_keys(exec, n);
+    temporary_array<ValueType, DerivedPolicy> temp_values(exec, n);
+
+    // use a caching allocator for the calls to merge_tiles_by_key
+    // XXX unfortunately g++-4.2 can't deal with this special execution policy
+#if defined(THRUST_GCC_VERSION) && THRUST_GCC_VERSION < 40300
+    execution_policy<DerivedPolicy> &merge_exec = exec;
+#else
+    cached_temporary_allocator<DerivedPolicy,thrust::cuda::execution_policy> merge_exec(exec);
+#endif
+
+    // The log(n) iterations start here. Each call to 'merge' merges an odd-even pair of tiles
+    unsigned int log_tile_size = thrust::detail::mpl::math::log2<block_size>::value;
+    bool ping = true;
+    for(; (1u << log_tile_size) < n; ++log_tile_size, ping = !ping)
+    {
+      // we ping-pong back and forth
+      if(ping)
+      {
+        merge_tiles_by_key(merge_exec, keys_first, values_first, n, temp_keys.begin(), temp_values.begin(), log_tile_size, comp);
+      } // end if
+      else
+      {
+        merge_tiles_by_key(merge_exec, temp_keys.begin(), temp_values.begin(), n, keys_first, values_first, log_tile_size, comp);
+      } // end else
+    } // end for
+
+    // this is to make sure that our data is finally in the data and keys arrays
+    // and not in the temporary arrays
+    if(!ping)
+    {
+      thrust::copy(exec, temp_keys.begin(), temp_keys.end(), keys_first);
+      thrust::copy(exec, temp_values.begin(), temp_values.end(), values_first);
+    } // end if
+  } // end if
+} // end stable_merge_sort_by_key()
+
+
+} // end namespace detail
+} // end namespace detail
+} // end namespace cuda
+} // end namespace system
+} // end namespace thrust
+
+__THRUST_DISABLE_MSVC_POSSIBLE_LOSS_OF_DATA_WARNING_END
+
diff --git a/compat/thrust/system/cuda/detail/detail/stable_primitive_sort.h b/compat/thrust/system/cuda/detail/detail/stable_primitive_sort.h
new file mode 100644
index 0000000000..8449a17b0d
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/detail/stable_primitive_sort.h
@@ -0,0 +1,54 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/cuda/detail/execution_policy.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+namespace detail
+{
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator>
+void stable_primitive_sort(execution_policy<DerivedPolicy> &exec,
+                           RandomAccessIterator first,
+                           RandomAccessIterator last);
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2>
+void stable_primitive_sort_by_key(execution_policy<DerivedPolicy> &exec,
+                                  RandomAccessIterator1 keys_first,
+                                  RandomAccessIterator1 keys_last,
+                                  RandomAccessIterator2 values_first);
+
+} // end namespace detail
+} // end namespace detail
+} // end namespace cuda
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/cuda/detail/detail/stable_primitive_sort.inl>
+
diff --git a/compat/thrust/system/cuda/detail/detail/stable_primitive_sort.inl b/compat/thrust/system/cuda/detail/detail/stable_primitive_sort.inl
new file mode 100644
index 0000000000..d6f4c775b6
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/detail/stable_primitive_sort.inl
@@ -0,0 +1,159 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/cuda/detail/detail/stable_primitive_sort.h>
+#include <thrust/system/cuda/detail/detail/stable_radix_sort.h>
+#include <thrust/functional.h>
+#include <thrust/partition.h>
+#include <thrust/iterator/zip_iterator.h>
+#include <thrust/detail/type_traits.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+namespace detail
+{
+namespace stable_primitive_sort_detail
+{
+
+
+template<typename Iterator>
+  struct enable_if_bool_sort
+    : thrust::detail::enable_if<
+        thrust::detail::is_same<
+          bool,
+          typename thrust::iterator_value<Iterator>::type
+        >::value
+      >
+{};
+
+
+template<typename Iterator>
+  struct disable_if_bool_sort
+    : thrust::detail::disable_if<
+        thrust::detail::is_same<
+          bool,
+          typename thrust::iterator_value<Iterator>::type
+        >::value
+      >
+{};
+
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator>
+  typename enable_if_bool_sort<RandomAccessIterator>::type
+    stable_primitive_sort(execution_policy<DerivedPolicy> &exec,
+                          RandomAccessIterator first,
+                          RandomAccessIterator last)
+{
+  // use stable_partition if we're sorting bool
+  // stable_partition puts true values first, so we need to logical_not
+  thrust::stable_partition(exec, first, last, thrust::logical_not<bool>());
+}
+
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator>
+  typename disable_if_bool_sort<RandomAccessIterator>::type
+    stable_primitive_sort(execution_policy<DerivedPolicy> &exec,
+                          RandomAccessIterator first,
+                          RandomAccessIterator last)
+{
+  // call stable_radix_sort
+  thrust::system::cuda::detail::detail::stable_radix_sort(exec,first,last);
+}
+
+
+struct logical_not_first
+{
+  template<typename Tuple>
+  __host__ __device__
+  bool operator()(Tuple t)
+  {
+    return !thrust::get<0>(t);
+  }
+};
+
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2>
+  typename enable_if_bool_sort<RandomAccessIterator1>::type
+    stable_primitive_sort_by_key(execution_policy<DerivedPolicy> &exec,
+                                 RandomAccessIterator1 keys_first,
+                                 RandomAccessIterator1 keys_last,
+                                 RandomAccessIterator2 values_first)
+{
+  // use stable_partition if we're sorting bool
+  // stable_partition puts true values first, so we need to logical_not
+  thrust::stable_partition(exec,
+                           thrust::make_zip_iterator(thrust::make_tuple(keys_first, values_first)),
+                           thrust::make_zip_iterator(thrust::make_tuple(keys_last, values_first)),
+                           logical_not_first());
+}
+
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2>
+  typename disable_if_bool_sort<RandomAccessIterator1>::type
+    stable_primitive_sort_by_key(execution_policy<DerivedPolicy> &exec,
+                                 RandomAccessIterator1 keys_first,
+                                 RandomAccessIterator1 keys_last,
+                                 RandomAccessIterator2 values_first)
+{
+  // call stable_radix_sort_by_key
+  thrust::system::cuda::detail::detail::stable_radix_sort_by_key(exec, keys_first, keys_last, values_first);
+}
+    
+  
+
+}
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator>
+void stable_primitive_sort(execution_policy<DerivedPolicy> &exec,
+                           RandomAccessIterator first,
+                           RandomAccessIterator last)
+{
+  thrust::system::cuda::detail::detail::stable_primitive_sort_detail::stable_primitive_sort(exec,first,last);
+}
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2>
+void stable_primitive_sort_by_key(execution_policy<DerivedPolicy> &exec,
+                                  RandomAccessIterator1 keys_first,
+                                  RandomAccessIterator1 keys_last,
+                                  RandomAccessIterator2 values_first)
+{
+  thrust::system::cuda::detail::detail::stable_primitive_sort_detail::stable_primitive_sort_by_key(exec, keys_first, keys_last, values_first);
+}
+
+} // end namespace detail
+} // end namespace detail
+} // end namespace cuda
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/cuda/detail/detail/stable_radix_sort.h b/compat/thrust/system/cuda/detail/detail/stable_radix_sort.h
new file mode 100644
index 0000000000..7a8b9964c0
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/detail/stable_radix_sort.h
@@ -0,0 +1,59 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file stable_radix_sort_dev.h
+ *  \brief Defines the interface for a stable radix sort implementation on CUDA
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/cuda/detail/execution_policy.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+namespace detail
+{
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator>
+void stable_radix_sort(execution_policy<DerivedPolicy> &exec,
+                       RandomAccessIterator first,
+                       RandomAccessIterator last);
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2>
+void stable_radix_sort_by_key(execution_policy<DerivedPolicy> &exec,
+                              RandomAccessIterator1 keys_first,
+                              RandomAccessIterator1 keys_last,
+                              RandomAccessIterator2 values_first);
+
+} // end namespace detail
+} // end namespace detail
+} // end namespace cuda
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/cuda/detail/detail/stable_radix_sort.inl>
+
diff --git a/compat/thrust/system/cuda/detail/detail/stable_radix_sort.inl b/compat/thrust/system/cuda/detail/detail/stable_radix_sort.inl
new file mode 100644
index 0000000000..9ea197702c
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/detail/stable_radix_sort.inl
@@ -0,0 +1,220 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/detail/config.h>
+
+// do not attempt to compile this file with any other compiler
+#if THRUST_DEVICE_COMPILER == THRUST_DEVICE_COMPILER_NVCC
+
+#include <thrust/detail/copy.h>
+#include <thrust/gather.h>
+#include <thrust/sequence.h>
+#include <thrust/iterator/iterator_traits.h>
+
+#include <thrust/detail/temporary_array.h>
+#include <thrust/detail/type_traits.h>
+#include <thrust/detail/util/align.h>
+#include <thrust/detail/raw_pointer_cast.h>
+
+
+__THRUST_DISABLE_MSVC_POSSIBLE_LOSS_OF_DATA_WARNING_BEGIN
+
+
+#include <thrust/system/cuda/detail/detail/b40c/radixsort_api.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+namespace detail
+{
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator>
+void stable_radix_sort(execution_policy<DerivedPolicy> &exec,
+                       RandomAccessIterator first,
+                       RandomAccessIterator last)
+{
+    typedef typename thrust::iterator_value<RandomAccessIterator>::type K;
+    
+    unsigned int num_elements = last - first;
+
+    // ensure data is properly aligned
+    if (!thrust::detail::util::is_aligned(thrust::raw_pointer_cast(&*first), 2*sizeof(K)))
+    {
+        thrust::detail::temporary_array<K, DerivedPolicy> aligned_keys(exec, first, last);
+        stable_radix_sort(exec, aligned_keys.begin(), aligned_keys.end());
+        thrust::copy(exec, aligned_keys.begin(), aligned_keys.end(), first);
+        return;
+    }
+    
+    thrust::system::cuda::detail::detail::b40c_thrust::RadixSortingEnactor<K> sorter(num_elements);
+    thrust::system::cuda::detail::detail::b40c_thrust::RadixSortStorage<K>    storage;
+    
+    // allocate temporary buffers
+    thrust::detail::temporary_array<K,    DerivedPolicy> temp_keys(exec, num_elements);
+    thrust::detail::temporary_array<int,  DerivedPolicy> temp_spine(exec, sorter.SpineElements());
+    thrust::detail::temporary_array<bool, DerivedPolicy> temp_from_alt(exec, 2);
+
+    // define storage
+    storage.d_keys             = thrust::raw_pointer_cast(&*first);
+    storage.d_alt_keys         = thrust::raw_pointer_cast(&temp_keys[0]);
+    storage.d_spine            = thrust::raw_pointer_cast(&temp_spine[0]);
+    storage.d_from_alt_storage = thrust::raw_pointer_cast(&temp_from_alt[0]);
+
+    // perform the sort
+    sorter.EnactSort(storage);
+    
+    // radix sort sometimes leaves results in the alternate buffers
+    if (storage.using_alternate_storage)
+    {
+        thrust::copy(exec, temp_keys.begin(), temp_keys.end(), first);
+    }
+}
+
+///////////////////////
+// Key-Value Sorting //
+///////////////////////
+
+// sort values directly
+template<typename DerivedPolicy,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2>
+void stable_radix_sort_by_key(execution_policy<DerivedPolicy> &exec,
+                              RandomAccessIterator1 first1,
+                              RandomAccessIterator1 last1,
+                              RandomAccessIterator2 first2,
+                              thrust::detail::true_type)
+{
+    typedef typename thrust::iterator_value<RandomAccessIterator1>::type K;
+    typedef typename thrust::iterator_value<RandomAccessIterator2>::type V;
+    
+    unsigned int num_elements = last1 - first1;
+
+    // ensure data is properly aligned
+    if (!thrust::detail::util::is_aligned(thrust::raw_pointer_cast(&*first1), 2*sizeof(K)))
+    {
+        thrust::detail::temporary_array<K,DerivedPolicy> aligned_keys(exec, first1, last1);
+        stable_radix_sort_by_key(exec, aligned_keys.begin(), aligned_keys.end(), first2);
+        thrust::copy(exec, aligned_keys.begin(), aligned_keys.end(), first1);
+        return;
+    }
+    if (!thrust::detail::util::is_aligned(thrust::raw_pointer_cast(&*first2), 2*sizeof(V)))
+    {
+        thrust::detail::temporary_array<V,DerivedPolicy> aligned_values(exec, first2, first2 + num_elements);
+        stable_radix_sort_by_key(exec, first1, last1, aligned_values.begin());
+        thrust::copy(exec, aligned_values.begin(), aligned_values.end(), first2);
+        return;
+    }
+   
+    thrust::system::cuda::detail::detail::b40c_thrust::RadixSortingEnactor<K,V> sorter(num_elements);
+    thrust::system::cuda::detail::detail::b40c_thrust::RadixSortStorage<K,V>    storage;
+    
+    // allocate temporary buffers
+    thrust::detail::temporary_array<K,    DerivedPolicy> temp_keys(exec, num_elements);
+    thrust::detail::temporary_array<V,    DerivedPolicy> temp_values(exec, num_elements);
+    thrust::detail::temporary_array<int,  DerivedPolicy> temp_spine(exec, sorter.SpineElements());
+    thrust::detail::temporary_array<bool, DerivedPolicy> temp_from_alt(exec, 2);
+
+    // define storage
+    storage.d_keys             = thrust::raw_pointer_cast(&*first1);
+    storage.d_values           = thrust::raw_pointer_cast(&*first2);
+    storage.d_alt_keys         = thrust::raw_pointer_cast(&temp_keys[0]);
+    storage.d_alt_values       = thrust::raw_pointer_cast(&temp_values[0]);
+    storage.d_spine            = thrust::raw_pointer_cast(&temp_spine[0]);
+    storage.d_from_alt_storage = thrust::raw_pointer_cast(&temp_from_alt[0]);
+
+    // perform the sort
+    sorter.EnactSort(storage);
+    
+    // radix sort sometimes leaves results in the alternate buffers
+    if (storage.using_alternate_storage)
+    {
+        thrust::copy(exec, temp_keys.begin(),   temp_keys.end(),   first1);
+        thrust::copy(exec, temp_values.begin(), temp_values.end(), first2);
+    }
+}
+
+
+// sort values indirectly
+template<typename DerivedPolicy, 
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2>
+void stable_radix_sort_by_key(execution_policy<DerivedPolicy> &exec,
+                              RandomAccessIterator1 first1,
+                              RandomAccessIterator1 last1,
+                              RandomAccessIterator2 first2,
+                              thrust::detail::false_type)
+{
+    typedef typename thrust::iterator_value<RandomAccessIterator2>::type V;
+    
+    unsigned int num_elements = last1 - first1;
+
+    // sort with integer values and then permute the real values accordingly
+    thrust::detail::temporary_array<unsigned int,DerivedPolicy> permutation(exec, num_elements);
+    thrust::sequence(exec, permutation.begin(), permutation.end());
+
+    stable_radix_sort_by_key(exec, first1, last1, permutation.begin());
+    
+    // copy values into temp vector and then permute
+    thrust::detail::temporary_array<V,DerivedPolicy> temp_values(exec, first2, first2 + num_elements);
+   
+    // permute values
+    thrust::gather(exec,
+                   permutation.begin(), permutation.end(),
+                   temp_values.begin(),
+                   first2);
+}
+
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2>
+void stable_radix_sort_by_key(execution_policy<DerivedPolicy> &exec,
+                              RandomAccessIterator1 first1,
+                              RandomAccessIterator1 last1,
+                              RandomAccessIterator2 first2)
+{
+    typedef typename thrust::iterator_value<RandomAccessIterator2>::type V;
+
+    // decide how to handle values
+    static const bool sort_values_directly = thrust::detail::is_trivial_iterator<RandomAccessIterator2>::value &&
+                                             thrust::detail::is_arithmetic<V>::value &&
+                                             sizeof(V) <= 8;    // TODO profile this
+
+    // XXX WAR unused variable warning
+    (void) sort_values_directly;
+
+    stable_radix_sort_by_key(exec, first1, last1, first2, 
+                             thrust::detail::integral_constant<bool, sort_values_directly>());
+}
+
+} // end namespace detail
+} // end namespace detail
+} // end namespace cuda
+} // end namespace system
+} // end namespace thrust
+
+
+__THRUST_DISABLE_MSVC_POSSIBLE_LOSS_OF_DATA_WARNING_END
+
+
+#endif // THRUST_DEVICE_COMPILER == THRUST_DEVICE_COMPILER_NVCC
+
diff --git a/compat/thrust/system/cuda/detail/detail/stable_sort_by_count.h b/compat/thrust/system/cuda/detail/detail/stable_sort_by_count.h
new file mode 100644
index 0000000000..b563654895
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/detail/stable_sort_by_count.h
@@ -0,0 +1,53 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/cuda/detail/execution_policy.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+namespace detail
+{
+
+
+template<unsigned int count,
+         typename DerivedPolicy,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2,
+         typename Compare>
+void stable_sort_by_count(execution_policy<DerivedPolicy> &exec,
+                          RandomAccessIterator1 keys_first,
+                          RandomAccessIterator1 keys_last,
+                          RandomAccessIterator2 values_first,
+                          Compare comp);
+
+
+} // end detail
+} // end detail
+} // end cuda
+} // end system
+} // end thrust
+
+#include <thrust/system/cuda/detail/detail/stable_sort_by_count.inl>
+
diff --git a/compat/thrust/system/cuda/detail/detail/stable_sort_by_count.inl b/compat/thrust/system/cuda/detail/detail/stable_sort_by_count.inl
new file mode 100644
index 0000000000..5efb36b9b0
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/detail/stable_sort_by_count.inl
@@ -0,0 +1,179 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+#include <thrust/detail/config.h>
+#include <thrust/system/cuda/detail/detail/stable_sort_by_count.h>
+#include <thrust/system/cuda/detail/detail/launch_closure.h>
+#include <thrust/system/cuda/detail/detail/uninitialized.h>
+#include <thrust/system/cuda/detail/block/merging_sort.h>
+#include <thrust/detail/minmax.h>
+#include <thrust/system/cuda/detail/runtime_introspection.h>
+
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+namespace detail
+{
+namespace stable_sort_by_count_detail
+{
+
+
+template<unsigned int block_size,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2,
+         typename StrictWeakOrdering,
+         typename Context>
+struct stable_sort_by_count_closure
+{
+  typedef Context context_type;
+
+  RandomAccessIterator1 keys_first;
+  RandomAccessIterator2 values_first;
+  StrictWeakOrdering comp; // XXX this should probably be thrust::detail::device_function
+  const unsigned int n;
+  context_type context;
+
+  stable_sort_by_count_closure(RandomAccessIterator1 keys_first,
+                               RandomAccessIterator2 values_first,
+                               StrictWeakOrdering comp,
+                               const unsigned int n,
+                               context_type context = context_type())
+    : keys_first(keys_first),
+      values_first(values_first),
+      comp(comp),
+      n(n),
+      context(context)
+  {}
+
+  __device__ __thrust_forceinline__
+  void operator()(void)
+  {
+    typedef typename iterator_value<RandomAccessIterator1>::type KeyType;
+    typedef typename iterator_value<RandomAccessIterator2>::type ValueType;
+  
+    __shared__ uninitialized_array<KeyType,block_size>   s_keys;
+    __shared__ uninitialized_array<ValueType,block_size> s_data;
+  
+    const unsigned int grid_size = context.grid_dimension() * context.block_dimension();
+  
+    // block_offset records the global index of this block's 0th thread
+    unsigned int block_offset = context.block_index() * block_size;
+    unsigned int i = context.thread_index() + block_offset;
+  
+    // advance iterators
+    keys_first   += i;
+    values_first += i;
+  
+    for(;
+        block_offset < n;
+        block_offset += grid_size, i += grid_size, keys_first += grid_size, values_first += grid_size)
+    {
+      context.barrier();
+      // copy input to shared
+      if(i < n)
+      {
+        s_keys[context.thread_index()] = *keys_first;
+        s_data[context.thread_index()] = *values_first;
+      } // end if
+      context.barrier();
+  
+      // this block could be partially full
+      unsigned int length = block_size;
+      if(block_offset + block_size > n)
+      {
+        length = n - block_offset;
+      } // end if
+  
+      // run merge_sort over the block
+      block::merging_sort(context, s_keys.begin(), s_data.begin(), length, comp);
+  
+      // write result
+      if(i < n)
+      {
+        *keys_first   = s_keys[context.thread_index()];
+        *values_first = s_data[context.thread_index()];
+      } // end if
+    } // end for i
+  }
+
+
+  static size_t max_grid_size()
+  {
+    const device_properties_t& properties = device_properties();
+
+    const unsigned int max_threads = properties.maxThreadsPerMultiProcessor * properties.multiProcessorCount;
+    const unsigned int max_blocks  = properties.maxGridSize[0];
+    
+    return thrust::min<size_t>(max_blocks, 3 * max_threads / block_size);
+  } // end max_grid_size()
+
+
+  size_t grid_size() const
+  {
+    // compute the maximum number of blocks we can launch on this arch
+    const unsigned int max_num_blocks = max_grid_size();
+
+    // first, sort within each block
+    size_t num_blocks = n / block_size;
+    if(n % block_size) ++num_blocks;
+
+    return thrust::min<size_t>(num_blocks, max_num_blocks);
+  } // end grid_size()
+}; // stable_sort_by_count_closure
+
+
+} // end stable_sort_by_count_detail
+
+
+template<unsigned int count,
+         typename DerivedPolicy,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2,
+         typename Compare>
+void stable_sort_by_count(execution_policy<DerivedPolicy> &,
+                          RandomAccessIterator1 keys_first,
+                          RandomAccessIterator1 keys_last,
+                          RandomAccessIterator2 values_first,
+                          Compare comp)
+{
+  typedef stable_sort_by_count_detail::stable_sort_by_count_closure<
+    count,
+    RandomAccessIterator1,
+    RandomAccessIterator2,
+    Compare,
+    detail::statically_blocked_thread_array<count>
+  > Closure;
+
+  Closure closure(keys_first, values_first, comp, keys_last - keys_first);
+ 
+  // do an odd-even sort per block of data
+  detail::launch_closure(closure, closure.grid_size(), count);
+} // end stable_sort_by_count()
+
+
+} // end detail
+} // end detail
+} // end cuda
+} // end system
+} // end thrust
+
diff --git a/compat/thrust/system/cuda/detail/detail/uninitialized.h b/compat/thrust/system/cuda/detail/detail/uninitialized.h
new file mode 100644
index 0000000000..a3e3dd2e76
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/detail/uninitialized.h
@@ -0,0 +1,261 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/cuda/detail/detail/alignment.h>
+#include <cstddef>
+#include <new>
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+namespace detail
+{
+
+
+template<typename T>
+  class uninitialized
+{
+  private:
+    typename aligned_storage<
+      sizeof(T),
+      alignment_of<T>::value
+    >::type storage;
+
+    __device__ __thrust_forceinline__ const T* ptr() const
+    {
+      return reinterpret_cast<const T*>(storage.data);
+    }
+
+    __device__ __thrust_forceinline__ T* ptr()
+    {
+      return reinterpret_cast<T*>(storage.data);
+    }
+
+  public:
+    // copy assignment
+    __device__ __thrust_forceinline__ uninitialized<T> &operator=(const T &other)
+    {
+      T& self = *this;
+      self = other;
+      return *this;
+    }
+
+    __device__ __thrust_forceinline__ T& get()
+    {
+      return *ptr();
+    }
+
+    __device__ __thrust_forceinline__ const T& get() const
+    {
+      return *ptr();
+    }
+
+    __device__ __thrust_forceinline__ operator T& ()
+    {
+      return get();
+    }
+
+    __device__ __thrust_forceinline__ operator const T&() const
+    {
+      return get();
+    }
+
+    __thrust_forceinline__ __device__ void construct()
+    {
+      ::new(ptr()) T();
+    }
+
+    template<typename Arg>
+    __thrust_forceinline__ __device__ void construct(const Arg &a)
+    {
+      ::new(ptr()) T(a);
+    }
+
+    template<typename Arg1, typename Arg2>
+    __thrust_forceinline__ __device__ void construct(const Arg1 &a1, const Arg2 &a2)
+    {
+      ::new(ptr()) T(a1,a2);
+    }
+
+    template<typename Arg1, typename Arg2, typename Arg3>
+    __thrust_forceinline__ __device__ void construct(const Arg1 &a1, const Arg2 &a2, const Arg3 &a3)
+    {
+      ::new(ptr()) T(a1,a2,a3);
+    }
+
+    template<typename Arg1, typename Arg2, typename Arg3, typename Arg4>
+    __thrust_forceinline__ __device__ void construct(const Arg1 &a1, const Arg2 &a2, const Arg3 &a3, const Arg4 &a4)
+    {
+      ::new(ptr()) T(a1,a2,a3,a4);
+    }
+
+    template<typename Arg1, typename Arg2, typename Arg3, typename Arg4, typename Arg5>
+    __thrust_forceinline__ __device__ void construct(const Arg1 &a1, const Arg2 &a2, const Arg3 &a3, const Arg4 &a4, const Arg5 &a5)
+    {
+      ::new(ptr()) T(a1,a2,a3,a4,a5);
+    }
+
+    template<typename Arg1, typename Arg2, typename Arg3, typename Arg4, typename Arg5, typename Arg6>
+    __thrust_forceinline__ __device__ void construct(const Arg1 &a1, const Arg2 &a2, const Arg3 &a3, const Arg4 &a4, const Arg5 &a5, const Arg6 &a6)
+    {
+      ::new(ptr()) T(a1,a2,a3,a4,a5,a6);
+    }
+
+    template<typename Arg1, typename Arg2, typename Arg3, typename Arg4, typename Arg5, typename Arg6, typename Arg7>
+    __thrust_forceinline__ __device__ void construct(const Arg1 &a1, const Arg2 &a2, const Arg3 &a3, const Arg4 &a4, const Arg5 &a5, const Arg6 &a6, const Arg7 &a7)
+    {
+      ::new(ptr()) T(a1,a2,a3,a4,a5,a6,a7);
+    }
+
+    template<typename Arg1, typename Arg2, typename Arg3, typename Arg4, typename Arg5, typename Arg6, typename Arg7, typename Arg8>
+    __thrust_forceinline__ __device__ void construct(const Arg1 &a1, const Arg2 &a2, const Arg3 &a3, const Arg4 &a4, const Arg5 &a5, const Arg6 &a6, const Arg7 &a7, const Arg8 &a8)
+    {
+      ::new(ptr()) T(a1,a2,a3,a4,a5,a6,a7,a8);
+    }
+
+    template<typename Arg1, typename Arg2, typename Arg3, typename Arg4, typename Arg5, typename Arg6, typename Arg7, typename Arg8, typename Arg9>
+    __thrust_forceinline__ __device__ void construct(const Arg1 &a1, const Arg2 &a2, const Arg3 &a3, const Arg4 &a4, const Arg5 &a5, const Arg6 &a6, const Arg7 &a7, const Arg8 &a8, const Arg9 &a9)
+    {
+      ::new(ptr()) T(a1,a2,a3,a4,a5,a6,a7,a8,a9);
+    }
+
+    template<typename Arg1, typename Arg2, typename Arg3, typename Arg4, typename Arg5, typename Arg6, typename Arg7, typename Arg8, typename Arg9, typename Arg10>
+    __thrust_forceinline__ __device__ void construct(const Arg1 &a1, const Arg2 &a2, const Arg3 &a3, const Arg4 &a4, const Arg5 &a5, const Arg6 &a6, const Arg7 &a7, const Arg8 &a8, const Arg9 &a9, const Arg10 &a10)
+    {
+      ::new(ptr()) T(a1,a2,a3,a4,a5,a6,a7,a8,a9,a10);
+    }
+
+    __thrust_forceinline__ __device__ void destroy()
+    {
+      T& self = *this;
+      self.~T();
+    }
+};
+
+
+template<typename T, std::size_t N>
+  class uninitialized_array
+{
+  public:
+    typedef T             value_type; 
+    typedef T&            reference;
+    typedef const T&      const_reference;
+    typedef T*            pointer;
+    typedef const T*      const_pointer;
+    typedef pointer       iterator;
+    typedef const_pointer const_iterator;
+    typedef std::size_t   size_type;
+
+    __thrust_forceinline__ __device__ iterator begin()
+    {
+      return data();
+    }
+
+    __thrust_forceinline__ __device__ const_iterator begin() const
+    {
+      return data();
+    }
+
+    __thrust_forceinline__ __device__ iterator end()
+    {
+      return begin() + size();
+    }
+
+    __thrust_forceinline__ __device__ const_iterator end() const
+    {
+      return begin() + size();
+    }
+
+    __thrust_forceinline__ __device__ const_iterator cbegin() const
+    {
+      return begin();
+    }
+
+    __thrust_forceinline__ __device__ const_iterator cend() const
+    {
+      return end();
+    }
+
+    __thrust_forceinline__ __device__ size_type size() const
+    {
+      return N;
+    }
+
+    __thrust_forceinline__ __device__ bool empty() const
+    {
+      return false;
+    }
+
+    __thrust_forceinline__ __device__ T* data()
+    {
+      return impl.get();
+    }
+
+    __thrust_forceinline__ __device__ const T* data() const
+    {
+      return impl.get();
+    }
+
+    // element access
+    __thrust_forceinline__ __device__ reference operator[](size_type n)
+    {
+      return data()[n];
+    }
+
+    __thrust_forceinline__ __device__ const_reference operator[](size_type n) const
+    {
+      return data()[n];
+    }
+
+    __thrust_forceinline__ __device__ reference front()
+    {
+      return *data();
+    }
+
+    __thrust_forceinline__ __device__ const_reference front() const
+    {
+      return *data();
+    }
+
+    __thrust_forceinline__ __device__ reference back()
+    {
+      return data()[size() - size_type(1)];
+    }
+
+    __thrust_forceinline__ __device__ const_reference back() const
+    {
+      return data()[size() - size_type(1)];
+    }
+
+  private:
+    uninitialized<T[N]> impl;
+};
+
+
+} // end detail
+} // end detail
+} // end cuda
+} // end system
+} // end thrust
+
diff --git a/compat/thrust/system/cuda/detail/equal.h b/compat/thrust/system/cuda/detail/equal.h
new file mode 100644
index 0000000000..a307fc5fc8
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/equal.h
@@ -0,0 +1,22 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system has no special version of this algorithm 
+
diff --git a/compat/thrust/system/cuda/detail/error.inl b/compat/thrust/system/cuda/detail/error.inl
new file mode 100644
index 0000000000..41b928fa32
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/error.inl
@@ -0,0 +1,95 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+#pragma once
+
+#include <thrust/system/cuda/error.h>
+#include <thrust/system/cuda/detail/guarded_cuda_runtime_api.h>
+
+namespace thrust
+{
+
+namespace system
+{
+
+
+error_code make_error_code(cuda::errc::errc_t e)
+{
+  return error_code(static_cast<int>(e), cuda_category());
+} // end make_error_code()
+
+
+error_condition make_error_condition(cuda::errc::errc_t e)
+{
+  return error_condition(static_cast<int>(e), cuda_category());
+} // end make_error_condition()
+
+
+namespace cuda
+{
+
+namespace detail
+{
+
+
+class cuda_error_category
+  : public error_category
+{
+  public:
+    inline cuda_error_category(void) {}
+
+    inline virtual const char *name(void) const
+    {
+      return "cuda";
+    }
+
+    inline virtual std::string message(int ev) const
+    {
+      static const std::string unknown_err("Unknown error");
+      const char *c_str = ::cudaGetErrorString(static_cast<cudaError_t>(ev));
+      return c_str ? std::string(c_str) : unknown_err;
+    }
+
+    inline virtual error_condition default_error_condition(int ev) const
+    {
+      using namespace cuda::errc;
+
+      if(ev < ::cudaErrorApiFailureBase)
+      {
+        return make_error_condition(static_cast<errc_t>(ev));
+      }
+
+      return system_category().default_error_condition(ev);
+    }
+}; // end cuda_error_category
+
+} // end detail
+
+} // end namespace cuda
+
+
+const error_category &cuda_category(void)
+{
+  static const cuda::detail::cuda_error_category result;
+  return result;
+}
+
+
+} // end namespace system
+
+} // end namespace thrust
+
diff --git a/compat/thrust/system/cuda/detail/execution_policy.h b/compat/thrust/system/cuda/detail/execution_policy.h
new file mode 100644
index 0000000000..7dae04c1eb
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/execution_policy.h
@@ -0,0 +1,131 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/execution_policy.h>
+#include <thrust/system/cpp/detail/execution_policy.h>
+#include <thrust/iterator/detail/any_system_tag.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+// put the canonical tag in the same ns as the backend's entry points
+namespace detail
+{
+
+// this awkward sequence of definitions arise
+// from the desire both for tag to derive
+// from execution_policy and for execution_policy
+// to convert to tag (when execution_policy is not
+// an ancestor of tag)
+
+// forward declaration of tag
+struct tag;
+
+// forward declaration of execution_policy
+template<typename> struct execution_policy;
+
+// specialize execution_policy for tag
+template<>
+  struct execution_policy<tag>
+    : thrust::execution_policy<tag>
+{};
+
+// tag's definition comes before the
+// generic definition of execution_policy
+struct tag : execution_policy<tag> {};
+
+// allow conversion to tag when it is not a successor
+template<typename Derived>
+  struct execution_policy
+    : thrust::execution_policy<Derived>
+{
+  // allow conversion to tag
+  inline operator tag () const
+  {
+    return tag();
+  }
+};
+
+
+template<typename System1, typename System2>
+  struct cross_system
+    : thrust::execution_policy<cross_system<System1,System2> >
+{
+  inline __host__ __device__
+  cross_system(thrust::execution_policy<System1> &system1,
+               thrust::execution_policy<System2> &system2)
+    : system1(system1), system2(system2)
+  {}
+
+  thrust::execution_policy<System1> &system1;
+  thrust::execution_policy<System2> &system2;
+
+  inline __host__ __device__
+  cross_system<System2,System1> rotate() const
+  {
+    return cross_system<System2,System1>(system2,system1);
+  }
+};
+
+
+// overloads of select_system
+
+// cpp interop
+template<typename System1, typename System2>
+inline __host__ __device__
+cross_system<System1,System2> select_system(const execution_policy<System1> &system1, const thrust::cpp::execution_policy<System2> &system2)
+{
+  thrust::execution_policy<System1> &non_const_system1 = const_cast<execution_policy<System1>&>(system1);
+  thrust::cpp::execution_policy<System2> &non_const_system2 = const_cast<thrust::cpp::execution_policy<System2>&>(system2);
+  return cross_system<System1,System2>(non_const_system1,non_const_system2);
+}
+
+
+template<typename System1, typename System2>
+inline __host__ __device__
+cross_system<System1,System2> select_system(const thrust::cpp::execution_policy<System1> &system1, execution_policy<System2> &system2)
+{
+  thrust::cpp::execution_policy<System1> &non_const_system1 = const_cast<thrust::cpp::execution_policy<System1>&>(system1);
+  thrust::execution_policy<System2> &non_const_system2 = const_cast<execution_policy<System2>&>(system2);
+  return cross_system<System1,System2>(non_const_system1,non_const_system2);
+}
+
+
+} // end detail
+
+// alias execution_policy and tag here
+using thrust::system::cuda::detail::execution_policy;
+using thrust::system::cuda::detail::tag;
+
+} // end cuda
+} // end system
+
+// alias items at top-level
+namespace cuda
+{
+
+using thrust::system::cuda::execution_policy;
+using thrust::system::cuda::tag;
+
+} // end cuda
+} // end thrust
+
diff --git a/compat/thrust/system/cuda/detail/extern_shared_ptr.h b/compat/thrust/system/cuda/detail/extern_shared_ptr.h
new file mode 100644
index 0000000000..5f34cc89ba
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/extern_shared_ptr.h
@@ -0,0 +1,58 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+
+template<typename T>
+  class extern_shared_ptr
+{
+// don't attempt to compile with any compiler other than nvcc
+// due to use of __shared__ below
+#if THRUST_DEVICE_COMPILER == THRUST_DEVICE_COMPILER_NVCC
+  public:
+    __device__
+    inline operator T * (void)
+    {
+      extern __shared__ int4 smem[];
+      return reinterpret_cast<T*>(smem);
+    }
+
+    __device__
+    inline operator const T * (void) const
+    {
+      extern __shared__ int4 smem[];
+      return reinterpret_cast<const T*>(smem);
+    }
+#endif // THRUST_DEVICE_COMPILER_NVCC
+}; // end extern_shared_ptr
+
+} // end detail
+} // end cuda
+} // end system
+} // end thrust
+
diff --git a/compat/thrust/system/cuda/detail/extrema.h b/compat/thrust/system/cuda/detail/extrema.h
new file mode 100644
index 0000000000..a307fc5fc8
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/extrema.h
@@ -0,0 +1,22 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system has no special version of this algorithm 
+
diff --git a/compat/thrust/system/cuda/detail/fill.h b/compat/thrust/system/cuda/detail/fill.h
new file mode 100644
index 0000000000..9c753bb9aa
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/fill.h
@@ -0,0 +1,54 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file fill.h
+ *  \brief Device implementation of fill.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/cuda/detail/execution_policy.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+
+template<typename DerivedPolicy, typename ForwardIterator, typename T>
+  void fill(execution_policy<DerivedPolicy> &exec,
+            ForwardIterator first,
+            ForwardIterator last,
+            const T &value);
+
+template<typename DerivedPolicy, typename OutputIterator, typename Size, typename T>
+  OutputIterator fill_n(execution_policy<DerivedPolicy> &exec,
+                        OutputIterator first,
+                        Size n,
+                        const T &value);
+
+} // end namespace detail
+} // end namespace cuda
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/cuda/detail/fill.inl>
+
diff --git a/compat/thrust/system/cuda/detail/fill.inl b/compat/thrust/system/cuda/detail/fill.inl
new file mode 100644
index 0000000000..3c1feb8ac8
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/fill.inl
@@ -0,0 +1,178 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file fill.inl
+ *  \brief Inline file for fill.h.
+ */
+
+#include <thrust/detail/config.h>
+
+#include <thrust/distance.h>
+#include <thrust/detail/util/align.h>
+#include <thrust/generate.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/detail/type_traits.h>
+#include <thrust/detail/raw_pointer_cast.h>
+#include <thrust/detail/minmax.h>
+#include <thrust/detail/internal_functional.h>
+
+#include <thrust/system/cuda/detail/runtime_introspection.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+namespace detail
+{
+
+
+template<typename WidePtr, typename T>
+  WidePtr widen_raw_ptr(T *ptr)
+{
+  typedef thrust::detail::pointer_traits<WidePtr> WideTraits;
+  typedef typename WideTraits::element_type       WideT;
+
+  // carefully widen the pointer to avoid warnings about conversions between differently aligned types on ARM
+  WideT *wide_raw_ptr = static_cast<WideT*>(static_cast<void*>(ptr));
+
+  return WideTraits::pointer_to(*wide_raw_ptr);
+}
+
+
+template<typename WideType, typename DerivedPolicy, typename Pointer, typename Size, typename T>
+  Pointer wide_fill_n(execution_policy<DerivedPolicy> &exec,
+                      Pointer first,
+                      Size n,
+                      const T &value)
+{
+  typedef typename thrust::iterator_value<Pointer>::type OutputType;
+
+  size_t ALIGNMENT_BOUNDARY = 128; // begin copying blocks at this byte boundary
+
+  WideType   wide_exemplar;
+  OutputType narrow_exemplars[sizeof(WideType) / sizeof(OutputType)];
+
+  for (size_t i = 0; i < sizeof(WideType) / sizeof(OutputType); i++)
+      narrow_exemplars[i] = static_cast<OutputType>(value);
+
+  // cast through char * to avoid type punning warnings
+  for (size_t i = 0; i < sizeof(WideType); i++)
+      reinterpret_cast<char *>(&wide_exemplar)[i] = reinterpret_cast<char *>(narrow_exemplars)[i];
+
+  OutputType *first_raw = thrust::raw_pointer_cast(first);
+  OutputType *last_raw  = first_raw + n;
+
+  OutputType *block_first_raw = (thrust::min)(first_raw + n,   thrust::detail::util::align_up(first_raw, ALIGNMENT_BOUNDARY));
+  OutputType *block_last_raw  = (thrust::max)(block_first_raw, thrust::detail::util::align_down(last_raw, sizeof(WideType)));
+
+  // rebind Pointer to WideType
+  typedef typename thrust::detail::rebind_pointer<Pointer,WideType>::type WidePtr;
+
+  // point to the widened range
+  // XXX since we've got an execution policy, we probably don't even need to deal with rebinding pointers
+  WidePtr block_first_wide = widen_raw_ptr<WidePtr>(block_first_raw);
+  WidePtr block_last_wide  = widen_raw_ptr<WidePtr>(block_last_raw);
+
+  thrust::generate(exec, first,                   Pointer(block_first_raw),    thrust::detail::fill_functor<OutputType>(value));
+  thrust::generate(exec, block_first_wide,        block_last_wide,             thrust::detail::fill_functor<WideType>(wide_exemplar));
+  thrust::generate(exec, Pointer(block_last_raw), first + n,                   thrust::detail::fill_functor<OutputType>(value));
+
+  return first + n;
+}
+
+template<typename DerivedPolicy, typename OutputIterator, typename Size, typename T>
+  OutputIterator fill_n(execution_policy<DerivedPolicy> &exec,
+                        OutputIterator first,
+                        Size n,
+                        const T &value,
+                        thrust::detail::false_type)
+{
+  thrust::detail::fill_functor<T> func(value); 
+  return thrust::generate_n(exec, first, n, func);
+}
+
+template<typename DerivedPolicy, typename OutputIterator, typename Size, typename T>
+  OutputIterator fill_n(execution_policy<DerivedPolicy> &exec,
+                        OutputIterator first,
+                        Size n,
+                        const T &value,
+                        thrust::detail::true_type)
+{
+  typedef typename thrust::iterator_traits<OutputIterator>::value_type OutputType;
+  
+  if ( thrust::detail::util::is_aligned<OutputType>(thrust::raw_pointer_cast(&*first)) )
+  {
+      if (compute_capability() < 20)
+      {
+        // 32-bit writes are faster on G80 and GT200
+        typedef unsigned int WideType;
+        wide_fill_n<WideType>(exec, &*first, n, value);
+      }
+      else
+      {
+        // 64-bit writes are faster on Fermi
+        typedef unsigned long long WideType;
+        wide_fill_n<WideType>(exec, &*first, n, value);
+      }
+
+      return first + n;
+  }
+  else
+  {
+    return fill_n(exec, first, n, value, thrust::detail::false_type());
+  }
+}
+
+} // end detail
+
+template<typename DerivedPolicy, typename OutputIterator, typename Size, typename T>
+  OutputIterator fill_n(execution_policy<DerivedPolicy> &exec,
+                        OutputIterator first,
+                        Size n,
+                        const T &value)
+{
+  typedef typename thrust::iterator_traits<OutputIterator>::value_type      OutputType;
+
+  // we're compiling with nvcc, launch a kernel
+  const bool use_wide_fill = thrust::detail::is_trivial_iterator<OutputIterator>::value
+      && thrust::detail::has_trivial_assign<OutputType>::value
+      && (sizeof(OutputType) == 1 || sizeof(OutputType) == 2 || sizeof(OutputType) == 4);
+
+  // XXX WAR usused variable warning
+  (void)use_wide_fill;
+
+  return detail::fill_n(exec, first, n, value, thrust::detail::integral_constant<bool, use_wide_fill>());
+}
+
+template<typename DerivedPolicy, typename ForwardIterator, typename T>
+  void fill(execution_policy<DerivedPolicy> &exec,
+            ForwardIterator first,
+            ForwardIterator last,
+            const T &value)
+{
+  thrust::system::cuda::detail::fill_n(exec, first, thrust::distance(first,last), value);
+} // end fill()
+
+} // end namespace detail
+} // end namespace cuda
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/cuda/detail/find.h b/compat/thrust/system/cuda/detail/find.h
new file mode 100644
index 0000000000..a307fc5fc8
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/find.h
@@ -0,0 +1,22 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system has no special version of this algorithm 
+
diff --git a/compat/thrust/system/cuda/detail/for_each.h b/compat/thrust/system/cuda/detail/for_each.h
new file mode 100644
index 0000000000..56be13b177
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/for_each.h
@@ -0,0 +1,60 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file for_each.h
+ *  \brief Defines the interface for a function that executes a 
+ *  function or functional for each value in a given range.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/cuda/detail/execution_policy.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator,
+         typename UnaryFunction>
+  RandomAccessIterator for_each(execution_policy<DerivedPolicy> &s,
+                                RandomAccessIterator first,
+                                RandomAccessIterator last,
+                                UnaryFunction f);
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator,
+         typename Size,
+         typename UnaryFunction>
+  RandomAccessIterator for_each_n(execution_policy<DerivedPolicy> &s,
+                                  RandomAccessIterator first,
+                                  Size n,
+                                  UnaryFunction f);
+
+} // end namespace detail
+} // end namespace cuda
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/cuda/detail/for_each.inl>
+
diff --git a/compat/thrust/system/cuda/detail/for_each.inl b/compat/thrust/system/cuda/detail/for_each.inl
new file mode 100644
index 0000000000..be6e56131f
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/for_each.inl
@@ -0,0 +1,199 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file for_each.inl
+ *  \brief Inline file for for_each.h.
+ */
+
+#include <thrust/detail/config.h>
+
+#include <thrust/detail/minmax.h>
+#include <thrust/detail/static_assert.h>
+
+#include <thrust/distance.h>
+#include <thrust/for_each.h>
+#include <thrust/system/cuda/detail/detail/launch_closure.h>
+#include <thrust/system/cuda/detail/detail/launch_calculator.h>
+#include <thrust/detail/util/blocking.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/detail/function.h>
+
+#include <limits>
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+namespace for_each_n_detail
+{
+
+
+template<typename RandomAccessIterator,
+         typename Size,
+         typename UnaryFunction,
+         typename Context>
+struct for_each_n_closure
+{
+  typedef void result_type;
+  typedef Context context_type;
+
+  RandomAccessIterator first;
+  Size n;
+  thrust::detail::device_function<UnaryFunction,void> f;
+  Context context;
+
+  for_each_n_closure(RandomAccessIterator first,
+                     Size n,
+                     UnaryFunction f,
+                     Context context = Context())
+    : first(first), n(n), f(f), context(context)
+  {}
+
+  __device__ __thrust_forceinline__
+  result_type operator()(void)
+  {
+    const Size grid_size = context.block_dimension() * context.grid_dimension();
+
+    Size i = context.linear_index();
+
+    // advance iterator
+    first += i;
+
+    while(i < n)
+    {
+      f(*first);
+      i += grid_size;
+      first += grid_size;
+    }
+  }
+}; // end for_each_n_closure
+
+
+template<typename Closure, typename Size>
+thrust::tuple<size_t,size_t> configure_launch(Size n)
+{
+  // calculate launch configuration
+  detail::launch_calculator<Closure> calculator;
+  
+  thrust::tuple<size_t, size_t, size_t> config = calculator.with_variable_block_size();
+  size_t max_blocks = thrust::get<0>(config);
+  size_t block_size = thrust::get<1>(config);
+  size_t num_blocks = thrust::min(max_blocks, thrust::detail::util::divide_ri<size_t>(n, block_size));
+
+  return thrust::make_tuple(num_blocks, block_size);
+}
+
+
+template<typename Size>
+bool use_big_closure(Size n, unsigned int little_grid_size)
+{
+  // use the big closure when n will not fit within an unsigned int
+  // or if incrementing an unsigned int by little_grid_size would overflow
+  // the counter
+  
+  Size threshold = std::numeric_limits<unsigned int>::max();
+
+  bool result = (sizeof(Size) > sizeof(unsigned int)) && (n > threshold);
+
+  if(!result)
+  {
+    // check if we'd overflow the little closure's counter
+    unsigned int little_n = static_cast<unsigned int>(n);
+
+    if((little_n - 1u) + little_grid_size < little_n)
+    {
+      result = true;
+    }
+  }
+
+  return result;
+}
+
+
+} // end for_each_n_detail
+
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator,
+         typename Size,
+         typename UnaryFunction>
+RandomAccessIterator for_each_n(execution_policy<DerivedPolicy> &,
+                                RandomAccessIterator first,
+                                Size n,
+                                UnaryFunction f)
+{
+  // we're attempting to launch a kernel, assert we're compiling with nvcc
+  // ========================================================================
+  // X Note to the user: If you've found this line due to a compiler error, X
+  // X you need to compile your code using nvcc, rather than g++ or cl.exe  X
+  // ========================================================================
+  THRUST_STATIC_ASSERT( (thrust::detail::depend_on_instantiation<RandomAccessIterator, THRUST_DEVICE_COMPILER == THRUST_DEVICE_COMPILER_NVCC>::value) );
+
+  if(n <= 0) return first;  // empty range
+  
+  // create two candidate closures to implement the for_each
+  // choose between them based on the whether we can fit n into a smaller integer
+  // and whether or not we'll overflow the closure's counter
+
+  typedef detail::blocked_thread_array Context;
+  typedef for_each_n_detail::for_each_n_closure<RandomAccessIterator, Size, UnaryFunction, Context>         BigClosure;
+  typedef for_each_n_detail::for_each_n_closure<RandomAccessIterator, unsigned int, UnaryFunction, Context> LittleClosure;
+
+  BigClosure    big_closure(first, n, f);
+  LittleClosure little_closure(first, static_cast<unsigned int>(n), f);
+
+  thrust::tuple<size_t, size_t> little_config = for_each_n_detail::configure_launch<LittleClosure>(n);
+
+  unsigned int little_grid_size = thrust::get<0>(little_config) * thrust::get<1>(little_config);
+
+  if(for_each_n_detail::use_big_closure(n, little_grid_size))
+  {
+    // launch the big closure
+    thrust::tuple<size_t, size_t> big_config = for_each_n_detail::configure_launch<BigClosure>(n);
+    detail::launch_closure(big_closure, thrust::get<0>(big_config), thrust::get<1>(big_config));
+  }
+  else
+  {
+    // launch the little closure
+    detail::launch_closure(little_closure, thrust::get<0>(little_config), thrust::get<1>(little_config));
+  }
+
+  return first + n;
+} 
+
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename UnaryFunction>
+  InputIterator for_each(execution_policy<DerivedPolicy> &exec,
+                         InputIterator first,
+                         InputIterator last,
+                         UnaryFunction f)
+{
+  return cuda::detail::for_each_n(exec, first, thrust::distance(first,last), f);
+} // end for_each()
+
+
+} // end namespace detail
+} // end namespace cuda
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/cuda/detail/gather.h b/compat/thrust/system/cuda/detail/gather.h
new file mode 100644
index 0000000000..a307fc5fc8
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/gather.h
@@ -0,0 +1,22 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system has no special version of this algorithm 
+
diff --git a/compat/thrust/system/cuda/detail/generate.h b/compat/thrust/system/cuda/detail/generate.h
new file mode 100644
index 0000000000..a307fc5fc8
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/generate.h
@@ -0,0 +1,22 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system has no special version of this algorithm 
+
diff --git a/compat/thrust/system/cuda/detail/get_value.h b/compat/thrust/system/cuda/detail/get_value.h
new file mode 100644
index 0000000000..273023f612
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/get_value.h
@@ -0,0 +1,93 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/cuda/detail/execution_policy.h>
+#include <thrust/system/cuda/detail/assign_value.h>
+#include <thrust/detail/raw_pointer_cast.h>
+#include <thrust/iterator/iterator_traits.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+
+
+namespace
+{
+
+
+template<typename DerivedPolicy, typename Pointer>
+inline __host__ __device__
+  typename thrust::iterator_value<Pointer>::type
+    get_value_msvc2005_war(execution_policy<DerivedPolicy> &exec, Pointer ptr)
+{
+  typedef typename thrust::iterator_value<Pointer>::type result_type;
+
+  // XXX war nvbugs/881631
+  struct war_nvbugs_881631
+  {
+    __host__ inline static result_type host_path(execution_policy<DerivedPolicy> &exec, Pointer ptr)
+    {
+      // when called from host code, implement with assign_value
+      // note that this requires a type with default constructor
+      result_type result;
+
+      thrust::host_system_tag host_tag;
+      cross_system<thrust::host_system_tag, DerivedPolicy> systems(host_tag, exec);
+      assign_value(systems, &result, ptr);
+
+      return result;
+    }
+
+    __device__ inline static result_type device_path(execution_policy<DerivedPolicy> &, Pointer ptr)
+    {
+      // when called from device code, just do simple deref
+      return *thrust::raw_pointer_cast(ptr);
+    }
+  };
+
+#ifndef __CUDA_ARCH__
+  return war_nvbugs_881631::host_path(exec, ptr);
+#else
+  return war_nvbugs_881631::device_path(exec, ptr);
+#endif // __CUDA_ARCH__
+} // end get_value_msvc2005_war()
+
+
+} // end anon namespace
+
+
+template<typename DerivedPolicy, typename Pointer>
+inline __host__ __device__
+  typename thrust::iterator_value<Pointer>::type
+    get_value(execution_policy<DerivedPolicy> &exec, Pointer ptr)
+{
+  return get_value_msvc2005_war(exec,ptr);
+} // end get_value()
+
+
+} // end detail
+} // end cuda
+} // end system
+} // end thrust
+
diff --git a/compat/thrust/system/cuda/detail/guarded_cuda_runtime_api.h b/compat/thrust/system/cuda/detail/guarded_cuda_runtime_api.h
new file mode 100644
index 0000000000..e6c0d2812e
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/guarded_cuda_runtime_api.h
@@ -0,0 +1,39 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// the purpose of this header is to check for the existence of macros
+// such as __host__ and __device__, which may already be defined by thrust
+// and to undefine them before entering cuda_runtime_api.h (which will redefine them)
+
+// we only try to do this stuff if cuda/include/host_defines.h has been included
+#if !defined(__HOST_DEFINES_H__)
+
+#ifdef __host__
+#undef __host__
+#endif // __host__
+
+#ifdef __device__
+#undef __device__
+#endif // __device__
+
+#endif // __HOST_DEFINES_H__
+
+#include <cuda_runtime_api.h>
+
diff --git a/compat/thrust/system/cuda/detail/inner_product.h b/compat/thrust/system/cuda/detail/inner_product.h
new file mode 100644
index 0000000000..a307fc5fc8
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/inner_product.h
@@ -0,0 +1,22 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system has no special version of this algorithm 
+
diff --git a/compat/thrust/system/cuda/detail/iter_swap.h b/compat/thrust/system/cuda/detail/iter_swap.h
new file mode 100644
index 0000000000..9b2bcf069f
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/iter_swap.h
@@ -0,0 +1,65 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/raw_pointer_cast.h>
+#include <thrust/swap.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+
+
+template<typename Pointer1, typename Pointer2>
+inline __host__ __device__
+void iter_swap(tag, Pointer1 a, Pointer2 b)
+{
+  // XXX war nvbugs/881631
+  struct war_nvbugs_881631
+  {
+    __host__ inline static void host_path(Pointer1 a, Pointer2 b)
+    {
+      thrust::swap_ranges(a, a + 1, b);
+    }
+
+    __device__ inline static void device_path(Pointer1 a, Pointer2 b)
+    {
+      using thrust::swap;
+      swap(*thrust::raw_pointer_cast(a),
+           *thrust::raw_pointer_cast(b));
+    }
+  };
+
+#ifndef __CUDA_ARCH__
+  return war_nvbugs_881631::host_path(a,b);
+#else
+  return war_nvbugs_881631::device_path(a,b);
+#endif // __CUDA_ARCH__
+} // end iter_swap()
+
+
+} // end detail
+} // end cuda
+} // end system
+} // end thrust
+
diff --git a/compat/thrust/system/cuda/detail/logical.h b/compat/thrust/system/cuda/detail/logical.h
new file mode 100644
index 0000000000..a307fc5fc8
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/logical.h
@@ -0,0 +1,22 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system has no special version of this algorithm 
+
diff --git a/compat/thrust/system/cuda/detail/malloc_and_free.h b/compat/thrust/system/cuda/detail/malloc_and_free.h
new file mode 100644
index 0000000000..676dd7cd5a
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/malloc_and_free.h
@@ -0,0 +1,71 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/cuda/detail/execution_policy.h>
+#include <thrust/detail/raw_pointer_cast.h>
+#include <thrust/system/cuda/detail/guarded_cuda_runtime_api.h>
+#include <thrust/system/system_error.h>
+#include <thrust/system/cuda/error.h>
+#include <thrust/system/detail/bad_alloc.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+
+
+// note that malloc returns a raw pointer to avoid
+// depending on the heavyweight thrust/system/cuda/memory.h header
+template<typename DerivedPolicy>
+  void *malloc(execution_policy<DerivedPolicy> &, std::size_t n)
+{
+  void *result = 0;
+
+  cudaError_t error = cudaMalloc(reinterpret_cast<void**>(&result), n);
+
+  if(error)
+  {
+    throw thrust::system::detail::bad_alloc(thrust::cuda_category().message(error).c_str());
+  } // end if
+
+  return result;
+} // end malloc()
+
+
+template<typename DerivedPolicy, typename Pointer>
+  void free(execution_policy<DerivedPolicy> &, Pointer ptr)
+{
+  cudaError_t error = cudaFree(thrust::raw_pointer_cast(ptr));
+
+  if(error)
+  {
+    throw thrust::system_error(error, thrust::cuda_category());
+  } // end error
+} // end free()
+
+
+} // end detail
+} // end cuda
+} // end system
+} // end thrust
+
diff --git a/compat/thrust/system/cuda/detail/memory.inl b/compat/thrust/system/cuda/detail/memory.inl
new file mode 100644
index 0000000000..998b54e345
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/memory.inl
@@ -0,0 +1,94 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/system/cuda/memory.h>
+#include <thrust/system/cuda/detail/malloc_and_free.h>
+#include <limits>
+
+namespace thrust
+{
+
+// XXX WAR an issue with MSVC 2005 (cl v14.00) incorrectly implementing
+//     pointer_raw_pointer for pointer by specializing it here
+//     note that we specialize it here, before the use of raw_pointer_cast
+//     below, which causes pointer_raw_pointer's instantiation
+#if (THRUST_HOST_COMPILER == THRUST_HOST_COMPILER_MSVC) && (_MSC_VER <= 1400)
+namespace detail
+{
+
+template<typename T>
+  struct pointer_raw_pointer< thrust::cuda::pointer<T> >
+{
+  typedef typename thrust::cuda::pointer<T>::raw_pointer type;
+}; // end pointer_raw_pointer
+
+} // end detail
+#endif
+
+namespace system
+{
+namespace cuda
+{
+
+
+template<typename T>
+  template<typename OtherT>
+    reference<T> &
+      reference<T>
+        ::operator=(const reference<OtherT> &other)
+{
+  return super_t::operator=(other);
+} // end reference::operator=()
+
+template<typename T>
+  reference<T> &
+    reference<T>
+      ::operator=(const value_type &x)
+{
+  return super_t::operator=(x);
+} // end reference::operator=()
+
+template<typename T>
+__host__ __device__
+void swap(reference<T> a, reference<T> b)
+{
+  a.swap(b);
+} // end swap()
+
+pointer<void> malloc(std::size_t n)
+{
+  tag cuda_tag;
+  return pointer<void>(thrust::system::cuda::detail::malloc(cuda_tag, n));
+} // end malloc()
+
+template<typename T>
+pointer<T> malloc(std::size_t n)
+{
+  pointer<void> raw_ptr = thrust::system::cuda::malloc(sizeof(T) * n);
+  return pointer<T>(reinterpret_cast<T*>(raw_ptr.get()));
+} // end malloc()
+
+void free(pointer<void> ptr)
+{
+  tag cuda_tag;
+  return thrust::system::cuda::detail::free(cuda_tag, ptr.get());
+} // end free()
+
+} // end cuda
+} // end system
+} // end thrust
+
diff --git a/compat/thrust/system/cuda/detail/merge.h b/compat/thrust/system/cuda/detail/merge.h
new file mode 100644
index 0000000000..e01b705470
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/merge.h
@@ -0,0 +1,50 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/cuda/detail/execution_policy.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2,
+         typename RandomAccessIterator3,
+         typename StrictWeakOrdering>
+  RandomAccessIterator3 merge(execution_policy<DerivedPolicy> &exec,
+                              RandomAccessIterator1 first1,
+                              RandomAccessIterator1 last1,
+                              RandomAccessIterator2 first2,
+                              RandomAccessIterator2 last2,
+                              RandomAccessIterator3 result,
+                              StrictWeakOrdering comp);
+
+} // end detail
+} // end cuda
+} // end system
+} // end thrust
+
+#include <thrust/system/cuda/detail/merge.inl>
+
diff --git a/compat/thrust/system/cuda/detail/merge.inl b/compat/thrust/system/cuda/detail/merge.inl
new file mode 100644
index 0000000000..bf7516fde1
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/merge.inl
@@ -0,0 +1,285 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/system/cuda/detail/merge.h>
+#include <thrust/pair.h>
+#include <thrust/tuple.h>
+#include <thrust/detail/minmax.h>
+#include <thrust/detail/function.h>
+#include <thrust/system/cuda/detail/detail/uninitialized.h>
+#include <thrust/system/cuda/detail/detail/launch_closure.h>
+#include <thrust/detail/util/blocking.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+namespace merge_detail
+{
+
+
+template<typename RandomAccessIterator1,
+         typename RandomAccessIterator2,
+         typename Size,
+         typename Compare>
+__device__ __thrust_forceinline__
+thrust::pair<Size,Size>
+  partition_search(RandomAccessIterator1 first1,
+                   RandomAccessIterator2 first2,
+                   Size diag,
+                   Size lower_bound1,
+                   Size upper_bound1,
+                   Size lower_bound2,
+                   Size upper_bound2,
+                   Compare comp)
+{
+  Size begin = thrust::max<Size>(lower_bound1, diag - upper_bound2);
+  Size end   = thrust::min<Size>(diag - lower_bound2, upper_bound1);
+
+  while(begin < end)
+  {
+    Size mid = (begin + end) / 2;
+    Size index1 = mid;
+    Size index2 = diag - mid - 1;
+
+    if(comp(first2[index2], first1[index1]))
+    {
+      end = mid;
+    }
+    else
+    {
+      begin = mid + 1;
+    }
+  }
+
+  return thrust::make_pair(begin, diag - begin);
+}
+
+
+template<typename Context, typename RandomAccessIterator1, typename Size, typename RandomAccessIterator2, typename RandomAccessIterator3, typename Compare>
+__device__ __thrust_forceinline__
+void merge_n(Context &ctx,
+             RandomAccessIterator1 first1,
+             Size n1,
+             RandomAccessIterator2 first2,
+             Size n2,
+             RandomAccessIterator3 result,
+             Compare comp_,
+             unsigned int work_per_thread)
+{
+  const unsigned int block_size = ctx.block_dimension();
+  thrust::detail::device_function<Compare,bool> comp(comp_);
+  typedef typename thrust::iterator_value<RandomAccessIterator1>::type value_type1;
+  typedef typename thrust::iterator_value<RandomAccessIterator2>::type value_type2;
+
+  Size result_size = n1 + n2;
+
+  // this is just oversubscription_rate * block_size * work_per_thread
+  // but it makes no sense to send oversubscription_rate as an extra parameter
+  Size work_per_block = thrust::detail::util::divide_ri(result_size, ctx.grid_dimension());
+
+  using thrust::system::cuda::detail::detail::uninitialized;
+  __shared__ uninitialized<thrust::pair<Size,Size> > s_block_input_begin;
+
+  Size result_block_offset = ctx.block_index() * work_per_block;
+
+  // find where this block's input begins in both input sequences
+  if(ctx.thread_index() == 0)
+  {
+    s_block_input_begin = (ctx.block_index() == 0) ?
+      thrust::pair<Size,Size>(0,0) :
+      partition_search(first1, first2,
+                       result_block_offset,
+                       Size(0), n1,
+                       Size(0), n2,
+                       comp);
+  }
+
+  ctx.barrier();
+
+  // iterate to consume this block's input
+  Size work_per_iteration = block_size * work_per_thread;
+  thrust::pair<Size,Size> block_input_end = s_block_input_begin;
+  block_input_end.first  += work_per_iteration;
+  block_input_end.second += work_per_iteration;
+  Size result_block_offset_last = result_block_offset + thrust::min<Size>(work_per_block, result_size - result_block_offset);
+
+  for(;
+      result_block_offset < result_block_offset_last;
+      result_block_offset += work_per_iteration,
+      block_input_end.first  += work_per_iteration,
+      block_input_end.second += work_per_iteration
+     )
+  {
+    // find where this thread's input begins in both input sequences for this iteration
+    thrust::pair<Size,Size> thread_input_begin =
+      partition_search(first1, first2,
+                       Size(result_block_offset + ctx.thread_index() * work_per_thread),
+                       s_block_input_begin.get().first,  thrust::min<Size>(block_input_end.first , n1),
+                       s_block_input_begin.get().second, thrust::min<Size>(block_input_end.second, n2),
+                       comp);
+
+    ctx.barrier();
+
+    // XXX the performance impact of not keeping x1 & x2
+    //     in registers is about 10% for int32
+    uninitialized<value_type1> x1;
+    uninitialized<value_type2> x2;
+
+    // XXX this is just a serial merge -- try to simplify or abstract this loop
+    Size i = result_block_offset + ctx.thread_index() * work_per_thread;
+    Size last_i = i + thrust::min<Size>(work_per_thread, result_size - thread_input_begin.first - thread_input_begin.second);
+    for(;
+        i < last_i;
+        ++i)
+    {
+      // optionally load x1 & x2
+      bool output_x2 = true;
+      if(thread_input_begin.second < n2)
+      {
+        x2 = first2[thread_input_begin.second];
+      }
+      else
+      {
+        output_x2 = false;
+      }
+
+      if(thread_input_begin.first < n1)
+      {
+        x1 = first1[thread_input_begin.first];
+
+        if(output_x2)
+        {
+          output_x2 = comp(x2.get(), x1.get());
+        }
+      }
+
+      result[i] = output_x2 ? x2.get() : x1.get();
+
+      if(output_x2)
+      {
+        ++thread_input_begin.second;
+      }
+      else
+      {
+        ++thread_input_begin.first;
+      }
+    } // end for
+
+    // the block's last thread has conveniently located the
+    // beginning of the next iteration's input
+    if(ctx.thread_index() == block_size-1)
+    {
+      s_block_input_begin = thread_input_begin;
+    }
+    ctx.barrier();
+  } // end for
+} // end merge_n
+
+
+template<typename RandomAccessIterator1, typename Size, typename RandomAccessIterator2, typename RandomAccessIterator3, typename Compare>
+  struct merge_n_closure
+{
+  typedef thrust::system::cuda::detail::detail::blocked_thread_array context_type;
+
+  RandomAccessIterator1 first1;
+  Size n1;
+  RandomAccessIterator2 first2;
+  Size n2;
+  RandomAccessIterator3 result;
+  Compare comp;
+  Size work_per_thread;
+
+  merge_n_closure(RandomAccessIterator1 first1, Size n1, RandomAccessIterator2 first2, Size n2, RandomAccessIterator3 result, Compare comp, Size work_per_thread)
+    : first1(first1), n1(n1), first2(first2), n2(n2), result(result), comp(comp), work_per_thread(work_per_thread)
+  {}
+
+  __device__ __forceinline__
+  void operator()()
+  {
+    context_type ctx;
+    merge_n(ctx, first1, n1, first2, n2, result, comp, work_per_thread);
+  }
+};
+
+
+// returns (work_per_thread, threads_per_block, oversubscription_factor)
+template<typename RandomAccessIterator1, typename RandomAccessIterator2, typename RandomAccessIterator3, typename Compare>
+  thrust::tuple<unsigned int,unsigned int,unsigned int>
+    tunables(RandomAccessIterator1, RandomAccessIterator1, RandomAccessIterator2, RandomAccessIterator2, RandomAccessIterator3, Compare comp)
+{
+  // determined by empirical testing on GTX 480
+  // ~4500 Mkeys/s on GTX 480
+  const unsigned int work_per_thread         = 5;
+  const unsigned int threads_per_block       = 128;
+  const unsigned int oversubscription_factor = 30;
+
+  return thrust::make_tuple(work_per_thread, threads_per_block, oversubscription_factor);
+}
+
+
+} // end merge_detail
+
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2, 
+	 typename RandomAccessIterator3,
+         typename Compare>
+RandomAccessIterator3 merge(execution_policy<DerivedPolicy> &exec,
+                            RandomAccessIterator1 first1,
+                            RandomAccessIterator1 last1,
+                            RandomAccessIterator2 first2,
+                            RandomAccessIterator2 last2,
+                            RandomAccessIterator3 result,
+                            Compare comp)
+{
+  typedef typename thrust::iterator_difference<RandomAccessIterator1>::type Size;
+  Size n1 = last1 - first1;
+  Size n2 = last2 - first2;
+  typename thrust::iterator_difference<RandomAccessIterator1>::type n = n1 + n2;
+
+  // empty result
+  if(n <= 0) return result;
+
+  unsigned int work_per_thread = 0, threads_per_block = 0, oversubscription_factor = 0;
+  thrust::tie(work_per_thread,threads_per_block,oversubscription_factor)
+    = merge_detail::tunables(first1, last1, first2, last2, result, comp);
+
+  const unsigned int work_per_block = work_per_thread * threads_per_block;
+
+  const unsigned int num_processors = device_properties().multiProcessorCount;
+  const unsigned int num_blocks = thrust::min<int>(oversubscription_factor * num_processors, thrust::detail::util::divide_ri(n, work_per_block));
+
+  typedef merge_detail::merge_n_closure<RandomAccessIterator1,Size,RandomAccessIterator2,RandomAccessIterator3,Compare> closure_type;
+  closure_type closure(first1, n1, first2, n2, result, comp, work_per_thread);
+
+  detail::launch_closure(closure, num_blocks, threads_per_block);
+
+  return result + n1 + n2;
+} // end merge()
+
+
+} // end namespace detail
+} // end namespace cuda
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/cuda/detail/mismatch.h b/compat/thrust/system/cuda/detail/mismatch.h
new file mode 100644
index 0000000000..a307fc5fc8
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/mismatch.h
@@ -0,0 +1,22 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system has no special version of this algorithm 
+
diff --git a/compat/thrust/system/cuda/detail/par.h b/compat/thrust/system/cuda/detail/par.h
new file mode 100644
index 0000000000..e56128c1d0
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/par.h
@@ -0,0 +1,66 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/cuda/detail/execution_policy.h>
+#include <thrust/detail/execute_with_allocator.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+
+
+struct par_t : thrust::system::cuda::detail::execution_policy<par_t>
+{
+  par_t() : thrust::system::cuda::detail::execution_policy<par_t>() {}
+
+  template<typename Allocator>
+    thrust::detail::execute_with_allocator<Allocator, thrust::system::cuda::detail::execution_policy>
+      operator()(Allocator &alloc) const
+  {
+    return thrust::detail::execute_with_allocator<Allocator, thrust::system::cuda::detail::execution_policy>(alloc);
+  }
+};
+
+
+} // end detail
+
+
+static const detail::par_t par;
+
+
+} // end cuda
+} // end system
+
+
+// alias par here
+namespace cuda
+{
+
+
+using thrust::system::cuda::par;
+
+
+} // end cuda
+} // end thrust
+
diff --git a/compat/thrust/system/cuda/detail/partition.h b/compat/thrust/system/cuda/detail/partition.h
new file mode 100644
index 0000000000..a307fc5fc8
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/partition.h
@@ -0,0 +1,22 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system has no special version of this algorithm 
+
diff --git a/compat/thrust/system/cuda/detail/reduce.h b/compat/thrust/system/cuda/detail/reduce.h
new file mode 100644
index 0000000000..d188f60f25
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/reduce.h
@@ -0,0 +1,54 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file reduce.h
+ *  \brief Reduce a sequence of elements with a given length.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/cuda/detail/execution_policy.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputType,
+         typename BinaryFunction>
+  OutputType reduce(execution_policy<DerivedPolicy> &exec,
+                    InputIterator first,
+                    InputIterator last,
+                    OutputType init,
+                    BinaryFunction binary_op);
+
+
+} // end namespace detail
+} // end namespace cuda
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/cuda/detail/reduce.inl>
+
diff --git a/compat/thrust/system/cuda/detail/reduce.inl b/compat/thrust/system/cuda/detail/reduce.inl
new file mode 100644
index 0000000000..66b4ac72c2
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/reduce.inl
@@ -0,0 +1,275 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file reduce.inl
+ *  \brief Inline file for reduce.h
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/distance.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/detail/minmax.h>
+#include <thrust/detail/temporary_array.h>
+#include <thrust/system/detail/generic/select_system.h>
+
+#include <thrust/system/cuda/detail/runtime_introspection.h>
+#include <thrust/system/cuda/detail/extern_shared_ptr.h>
+#include <thrust/system/cuda/detail/block/reduce.h>
+#include <thrust/system/cuda/detail/detail/launch_closure.h>
+#include <thrust/system/cuda/detail/detail/launch_calculator.h>
+#include <thrust/system/cuda/detail/execution_policy.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+
+namespace reduce_detail
+{
+
+/*
+ * Reduce a vector of n elements using binary_op()
+ *
+ * The order of reduction is not defined, so binary_op() should
+ * be a commutative (and associative) operator such as 
+ * (integer) addition.  Since floating point operations
+ * do not completely satisfy these criteria, the result is 
+ * generally not the same as a consecutive reduction of 
+ * the elements.
+ * 
+ * Uses the same pattern as reduce6() in the CUDA SDK
+ *
+ */
+template <typename InputIterator,
+          typename Size,
+          typename T,
+          typename OutputIterator,
+          typename BinaryFunction,
+          typename Context>
+struct unordered_reduce_closure
+{
+  InputIterator  input;
+  Size           n;
+  T              init;
+  OutputIterator output;
+  BinaryFunction binary_op;
+  unsigned int shared_array_size;
+
+  typedef Context context_type;
+  context_type context;
+
+  unordered_reduce_closure(InputIterator input, Size n, T init, OutputIterator output, BinaryFunction binary_op, unsigned int shared_array_size, Context context = Context())
+    : input(input), n(n), init(init), output(output), binary_op(binary_op), shared_array_size(shared_array_size), context(context) {}
+
+  __device__ __thrust_forceinline__
+  void operator()(void)
+  {
+    typedef typename thrust::iterator_value<OutputIterator>::type OutputType;
+    extern_shared_ptr<OutputType>  shared_array;
+
+    Size grid_size = context.block_dimension() * context.grid_dimension();
+
+    Size i = context.linear_index();
+      
+    input += i;
+
+    // compute reduction with all blockDim.x threads
+    OutputType sum = thrust::raw_reference_cast(*input);
+
+    i     += grid_size;
+    input += grid_size;
+
+    while (i < n)
+    {
+      OutputType val = thrust::raw_reference_cast(*input);
+
+      sum = binary_op(sum, val);
+
+      i      += grid_size;
+      input  += grid_size;
+    }
+
+    // write first shared_array_size values into shared memory
+    if (context.thread_index() < shared_array_size)
+      shared_array[context.thread_index()] = sum;  
+
+    // accumulate remaining values (if any) to shared memory in stages
+    if (context.block_dimension() > shared_array_size)
+    {
+      unsigned int lb = shared_array_size;
+      unsigned int ub = shared_array_size + lb;
+      
+      while (lb < context.block_dimension())
+      {
+        context.barrier();
+
+        if (lb <= context.thread_index() && context.thread_index() < ub)
+        {
+          OutputType tmp = shared_array[context.thread_index() - lb];
+          shared_array[context.thread_index() - lb] = binary_op(tmp, sum);
+        }
+
+        lb += shared_array_size;
+        ub += shared_array_size;
+      }
+    }
+    
+    context.barrier();
+
+    block::reduce_n(context, shared_array, thrust::min<unsigned int>(context.block_dimension(), shared_array_size), binary_op);
+  
+    if (context.thread_index() == 0)
+    {
+      OutputType tmp = shared_array[0];
+
+      if (context.grid_dimension() == 1)
+        tmp = binary_op(init, tmp);
+
+      output += context.block_index();
+      *output = tmp;
+    }
+  }
+};
+
+
+__THRUST_DISABLE_MSVC_POSSIBLE_LOSS_OF_DATA_WARNING_BEGIN
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputType,
+         typename BinaryFunction>
+  OutputType reduce(execution_policy<DerivedPolicy> &exec,
+                    InputIterator first,
+                    InputIterator last,
+                    OutputType init,
+                    BinaryFunction binary_op)
+{
+  // we're attempting to launch a kernel, assert we're compiling with nvcc
+  // ========================================================================
+  // X Note to the user: If you've found this line due to a compiler error, X
+  // X you need to compile your code using nvcc, rather than g++ or cl.exe  X
+  // ========================================================================
+  THRUST_STATIC_ASSERT( (thrust::detail::depend_on_instantiation<InputIterator, THRUST_DEVICE_COMPILER == THRUST_DEVICE_COMPILER_NVCC>::value) );
+
+  typedef typename thrust::iterator_difference<InputIterator>::type difference_type;
+
+  difference_type n = thrust::distance(first,last);
+
+  if (n == 0)
+    return init;
+
+  typedef thrust::detail::temporary_array<OutputType, DerivedPolicy> OutputArray;
+  typedef typename OutputArray::iterator OutputIterator;
+
+  typedef detail::blocked_thread_array Context;
+  typedef unordered_reduce_closure<InputIterator,difference_type,OutputType,OutputIterator,BinaryFunction,Context> Closure;
+    
+  function_attributes_t attributes = detail::closure_attributes<Closure>();
+  
+  // TODO chose this in a more principled manner
+  size_t threshold = thrust::max<size_t>(2 * attributes.maxThreadsPerBlock, 1024);
+
+  device_properties_t properties = device_properties();
+
+  // launch configuration
+  size_t num_blocks; 
+  size_t block_size; 
+  size_t array_size; 
+  size_t smem_bytes; 
+
+  // first level reduction
+  if (static_cast<size_t>(n) < threshold)
+  {
+    num_blocks = 1;
+    block_size = thrust::min(static_cast<size_t>(n), static_cast<size_t>(attributes.maxThreadsPerBlock));
+    array_size = thrust::min(block_size, (properties.sharedMemPerBlock - attributes.sharedSizeBytes) / sizeof(OutputType));
+    smem_bytes = sizeof(OutputType) * array_size;
+  }
+  else
+  {
+    detail::launch_calculator<Closure> calculator;
+    
+    thrust::tuple<size_t,size_t,size_t> config = calculator.with_variable_block_size_available_smem();
+
+    num_blocks = thrust::min(thrust::get<0>(config), static_cast<size_t>(n) / thrust::get<1>(config));
+    block_size = thrust::get<1>(config);
+    array_size = thrust::min(block_size, thrust::get<2>(config) / sizeof(OutputType));
+    smem_bytes = sizeof(OutputType) * array_size;
+  }
+ 
+  // TODO assert(n <= num_blocks * block_size);
+  // TODO if (shared_array_size < 1) throw cuda exception "insufficient shared memory"
+
+  OutputArray output(exec, num_blocks);
+
+  Closure closure(first, n, init, output.begin(), binary_op, array_size);
+  
+  //std::cout << "Launching " << num_blocks << " blocks of kernel with " << block_size << " threads and " << smem_bytes << " shared memory per block " << std::endl;
+
+  detail::launch_closure(closure, num_blocks, block_size, smem_bytes);
+
+  // second level reduction
+  if (num_blocks > 1)
+  {
+    typedef detail::blocked_thread_array Context;
+    typedef unordered_reduce_closure<OutputIterator,difference_type,OutputType,OutputIterator,BinaryFunction,Context> Closure;
+
+    function_attributes_t attributes = detail::closure_attributes<Closure>();
+
+    num_blocks = 1;
+    block_size = thrust::min(output.size(), static_cast<size_t>(attributes.maxThreadsPerBlock));
+    array_size = thrust::min(block_size, (properties.sharedMemPerBlock - attributes.sharedSizeBytes) / sizeof(OutputType));
+    smem_bytes = sizeof(OutputType) * array_size;
+  
+    // TODO if (shared_array_size < 1) throw cuda exception "insufficient shared memory"
+
+    Closure closure(output.begin(), output.size(), init, output.begin(), binary_op, array_size);
+
+    //std::cout << "Launching " << num_blocks << " blocks of kernel with " << block_size << " threads and " << smem_bytes << " shared memory per block " << std::endl;
+
+    detail::launch_closure(closure, num_blocks, block_size, smem_bytes);
+  }
+  
+  return output[0];
+} // end reduce
+
+} // end reduce_detail
+
+__THRUST_DISABLE_MSVC_POSSIBLE_LOSS_OF_DATA_WARNING_END
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputType,
+         typename BinaryFunction>
+  OutputType reduce(execution_policy<DerivedPolicy> &exec,
+                    InputIterator first,
+                    InputIterator last,
+                    OutputType init,
+                    BinaryFunction binary_op)
+{
+  return reduce_detail::reduce(exec, first, last, init, binary_op);
+} // end reduce()
+
+} // end namespace detail
+} // end namespace cuda
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/cuda/detail/reduce_by_key.h b/compat/thrust/system/cuda/detail/reduce_by_key.h
new file mode 100644
index 0000000000..9b8ec10936
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/reduce_by_key.h
@@ -0,0 +1,59 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file reduce_by_key.h
+ *  \brief CUDA implementation of reduce_by_key
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/cuda/detail/execution_policy.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+
+template <typename DerivedPolicy,
+          typename InputIterator1,
+          typename InputIterator2,
+          typename OutputIterator1,
+          typename OutputIterator2,
+          typename BinaryPredicate,
+          typename BinaryFunction>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    reduce_by_key(execution_policy<DerivedPolicy> &exec,
+                  InputIterator1 keys_first, 
+                  InputIterator1 keys_last,
+                  InputIterator2 values_first,
+                  OutputIterator1 keys_output,
+                  OutputIterator2 values_output,
+                  BinaryPredicate binary_pred,
+                  BinaryFunction binary_op);
+
+} // end namespace detail
+} // end namespace cuda
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/cuda/detail/reduce_by_key.inl>
+
diff --git a/compat/thrust/system/cuda/detail/reduce_by_key.inl b/compat/thrust/system/cuda/detail/reduce_by_key.inl
new file mode 100644
index 0000000000..18dc1e4994
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/reduce_by_key.inl
@@ -0,0 +1,705 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+#include <thrust/detail/config.h>
+
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/iterator/counting_iterator.h>
+#include <thrust/iterator/transform_iterator.h>
+#include <thrust/iterator/permutation_iterator.h>
+#include <thrust/iterator/zip_iterator.h>
+#include <thrust/system/detail/generic/select_system.h>
+
+#include <thrust/detail/type_traits.h>
+#include <thrust/detail/type_traits/function_traits.h>
+#include <thrust/detail/type_traits/iterator/is_output_iterator.h>
+#include <thrust/detail/type_traits/iterator/is_discard_iterator.h>
+
+#include <thrust/detail/minmax.h>
+#include <thrust/detail/internal_functional.h>
+#include <thrust/detail/temporary_array.h>
+
+#include <thrust/reduce.h>
+#include <thrust/scan.h>
+#include <thrust/system/cuda/detail/default_decomposition.h>
+#include <thrust/system/cuda/detail/block/inclusive_scan.h>
+#include <thrust/system/cuda/detail/execution_policy.h>
+#include <thrust/system/cuda/detail/detail/launch_closure.h>
+#include <thrust/system/cuda/detail/reduce_intervals.h>
+#include <thrust/system/cuda/detail/detail/uninitialized.h>
+
+__THRUST_DISABLE_MSVC_POSSIBLE_LOSS_OF_DATA_WARNING_BEGIN
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+namespace reduce_by_key_detail
+{
+
+template <typename FlagType, typename IndexType, typename KeyType, typename BinaryPredicate>
+struct tail_flag_functor
+{
+  BinaryPredicate binary_pred; // NB: this must be the first member for performance reasons
+  IndexType n;
+
+  typedef FlagType result_type;
+  
+  tail_flag_functor(IndexType n, BinaryPredicate binary_pred)
+    : n(n), binary_pred(binary_pred)
+  {}
+  
+  // XXX why is this noticably faster?  (it may read past the end of input)
+  //FlagType operator()(const thrust::tuple<IndexType,KeyType,KeyType>& t) const
+  
+  template <typename Tuple>
+  __host__ __device__ __thrust_forceinline__
+  FlagType operator()(const Tuple& t)
+  {
+    if (thrust::get<0>(t) == (n - 1) || !binary_pred(thrust::get<1>(t), thrust::get<2>(t)))
+      return 1;
+    else
+      return 0;
+  }
+};
+
+
+template <unsigned int CTA_SIZE,
+          unsigned int K,
+          bool FullBlock,
+          typename Context,
+          typename FlagIterator,
+          typename FlagType>
+__device__ __thrust_forceinline__
+FlagType load_flags(Context context,
+                    const unsigned int n,
+                    FlagIterator iflags,
+                    FlagType  (&sflag)[CTA_SIZE])
+{
+  FlagType flag_bits = 0;
+
+  // load flags in unordered fashion
+  for(unsigned int k = 0; k < K; k++)
+  {
+    const unsigned int offset = k*CTA_SIZE + context.thread_index();
+
+    if (FullBlock || offset < n)
+    {
+      FlagIterator temp = iflags + offset;
+      if (*temp)
+        flag_bits |= FlagType(1) << k;
+    }
+  }
+
+  sflag[context.thread_index()] = flag_bits;
+  
+  context.barrier();
+
+  flag_bits = 0;
+
+  // obtain flags for iflags[K * context.thread_index(), K * context.thread_index() + K)
+  for(unsigned int k = 0; k < K; k++)
+  {
+    const unsigned int offset = K * context.thread_index() + k;
+
+    if (FullBlock || offset < n)
+    {
+      flag_bits |= ((sflag[offset % CTA_SIZE] >> (offset / CTA_SIZE)) & FlagType(1)) << k;
+    }
+  }
+
+  context.barrier();
+  
+  sflag[context.thread_index()] = flag_bits;
+  
+  context.barrier();
+
+  return flag_bits;
+}
+
+template <unsigned int CTA_SIZE,
+          unsigned int K,
+          bool FullBlock,
+          typename Context,
+          typename InputIterator2,
+          typename ValueType>
+__device__ __thrust_forceinline__
+void load_values(Context context,
+                 const unsigned int n,
+                 InputIterator2 ivals,
+                 ValueType (&sdata)[K][CTA_SIZE + 1])
+{
+  for(unsigned int k = 0; k < K; k++)
+  {
+    const unsigned int offset = k*CTA_SIZE + context.thread_index();
+
+    if (FullBlock || offset < n)
+    {
+      InputIterator2 temp = ivals + offset;
+      sdata[offset % K][offset / K] = *temp;
+    }
+  }
+
+  context.barrier();
+}
+
+
+template <unsigned int CTA_SIZE,
+          unsigned int K,
+          bool FullBlock,
+          typename Context,
+          typename InputIterator1,
+          typename InputIterator2,
+          typename OutputIterator1,
+          typename OutputIterator2,
+          typename BinaryPredicate,
+          typename BinaryFunction,
+          typename FlagIterator,
+          typename FlagType,
+          typename IndexType,
+          typename ValueType>
+__device__ __thrust_forceinline__
+void reduce_by_key_body(Context context,
+                        const unsigned int n,
+                        InputIterator1   ikeys,
+                        InputIterator2   ivals,
+                        OutputIterator1  okeys,
+                        OutputIterator2  ovals,
+                        BinaryPredicate  binary_pred,
+                        BinaryFunction   binary_op,
+                        FlagIterator     iflags,
+                        FlagType  (&sflag)[CTA_SIZE],
+                        ValueType (&sdata)[K][CTA_SIZE + 1],
+                        bool&      carry_in,
+                        IndexType& carry_index,
+                        ValueType& carry_value)
+{
+  // load flags
+  const FlagType flag_bits  = load_flags<CTA_SIZE,K,FullBlock>(context, n, iflags, sflag);
+  const FlagType flag_count = __popc(flag_bits); // TODO hide this behind a template
+  const FlagType left_flag  = (context.thread_index() == 0) ? 0 : sflag[context.thread_index() - 1];
+  const FlagType head_flag  = (context.thread_index() == 0 || flag_bits & ((1 << (K - 1)) - 1) || left_flag & (1 << (K - 1))) ? 1 : 0;
+  
+  context.barrier();
+
+  // scan flag counts
+  sflag[context.thread_index()] = flag_count; context.barrier();
+
+  block::inclusive_scan(context, sflag, thrust::plus<FlagType>());
+
+  const FlagType output_position = (context.thread_index() == 0) ? 0 : sflag[context.thread_index() - 1];
+  const FlagType num_outputs     = sflag[CTA_SIZE - 1];
+
+  context.barrier();
+
+  // shuffle keys and write keys out
+  if (!thrust::detail::is_discard_iterator<OutputIterator1>::value)
+  {
+    // XXX this could be improved
+    for (unsigned int i = 0; i < num_outputs; i += CTA_SIZE)
+    {
+      FlagType position = output_position;
+
+      for(unsigned int k = 0; k < K; k++)
+      {
+        if (flag_bits & (FlagType(1) << k))
+        {
+          if (i <= position && position < i + CTA_SIZE)
+            sflag[position - i] = K * context.thread_index() + k;
+          position++;
+        }
+      }
+
+      context.barrier();
+
+      if (i + context.thread_index() < num_outputs)
+      {
+        InputIterator1  tmp1 = ikeys + sflag[context.thread_index()];
+        OutputIterator1 tmp2 = okeys + (i + context.thread_index());
+        *tmp2 = *tmp1; 
+      }
+      
+      context.barrier();
+    }
+  }
+
+  // load values
+  load_values<CTA_SIZE,K,FullBlock> (context, n, ivals, sdata);
+
+  ValueType ldata[K];
+  for (unsigned int k = 0; k < K; k++)
+      ldata[k] = sdata[k][context.thread_index()];
+
+  // carry in (if necessary)
+  if (context.thread_index() == 0 && carry_in)
+  {
+    // XXX WAR sm_10 issue
+    ValueType tmp1 = carry_value;
+    ldata[0] = binary_op(tmp1, ldata[0]);
+  }
+
+  context.barrier();
+
+  // sum local values
+  {
+    for(unsigned int k = 1; k < K; k++)
+    {
+      const unsigned int offset = K * context.thread_index() + k;
+
+      if (FullBlock || offset < n)
+      {
+        if (!(flag_bits & (FlagType(1) << (k - 1))))
+          ldata[k] = binary_op(ldata[k - 1], ldata[k]);
+      }
+    }
+  }
+
+  // second level segmented scan
+  {
+    // use head flags for segmented scan
+    sflag[context.thread_index()] = head_flag;  sdata[K - 1][context.thread_index()] = ldata[K - 1]; context.barrier();
+
+    if (FullBlock)
+      block::inclusive_scan_by_flag(context, sflag, sdata[K-1], binary_op);
+    else
+      block::inclusive_scan_by_flag_n(context, sflag, sdata[K-1], n, binary_op);
+  }
+
+  // update local values
+  if (context.thread_index() > 0)
+  {
+    unsigned int update_bits  = (flag_bits << 1) | (left_flag >> (K - 1));
+// TODO remove guard
+#if THRUST_DEVICE_COMPILER == THRUST_DEVICE_COMPILER_NVCC
+    unsigned int update_count = __ffs(update_bits) - 1u; // NB: this might wrap around to UINT_MAX
+#else
+    unsigned int update_count = 0;
+#endif // THRUST_DEVICE_COMPILER_NVCC
+
+    if (!FullBlock && (K + 1) * context.thread_index() > n)
+      update_count = thrust::min(n - K * context.thread_index(), update_count);
+
+    ValueType left = sdata[K - 1][context.thread_index() - 1];
+
+    for(unsigned int k = 0; k < K; k++)
+    {
+      if (k < update_count)
+        ldata[k] = binary_op(left, ldata[k]);
+    }
+  }
+  
+  context.barrier();
+
+  // store carry out
+  if (FullBlock)
+  {
+    if (context.thread_index() == CTA_SIZE - 1)
+    {
+      carry_value = ldata[K - 1];
+      carry_in    = (flag_bits & (FlagType(1) << (K - 1))) ? false : true;
+      carry_index = num_outputs;
+    }
+  }
+  else
+  {
+    if (context.thread_index() == (n - 1) / K)
+    {
+      for (unsigned int k = 0; k < K; k++)
+          if (k == (n - 1) % K)
+              carry_value = ldata[k];
+      carry_in    = (flag_bits & (FlagType(1) << ((n - 1) % K))) ? false : true;
+      carry_index = num_outputs;
+    }
+  }
+
+  // shuffle values
+  {
+    FlagType position = output_position;
+  
+    for(unsigned int k = 0; k < K; k++)
+    {
+      const unsigned int offset = K * context.thread_index() + k;
+  
+      if (FullBlock || offset < n)
+      {
+        if (flag_bits & (FlagType(1) << k))
+        {
+          sdata[position / CTA_SIZE][position % CTA_SIZE] = ldata[k];
+          position++;
+        }
+      }
+    }
+  }
+
+  context.barrier();
+
+
+  // write values out
+  for(unsigned int k = 0; k < K; k++)
+  {
+    const unsigned int offset = CTA_SIZE * k + context.thread_index();
+
+    if (offset < num_outputs)
+    {
+      OutputIterator2 tmp = ovals + offset;
+      *tmp = sdata[k][context.thread_index()];
+    }
+  }
+
+  context.barrier();
+}
+
+template <typename InputIterator1,
+          typename InputIterator2,
+          typename OutputIterator1,
+          typename OutputIterator2,
+          typename BinaryPredicate,
+          typename BinaryFunction,
+          typename FlagIterator,
+          typename IndexIterator,
+          typename ValueIterator,
+          typename BoolIterator,
+          typename Decomposition,
+          typename Context>
+struct reduce_by_key_closure
+{
+  InputIterator1   ikeys;
+  InputIterator2   ivals;
+  OutputIterator1  okeys;
+  OutputIterator2  ovals;
+  BinaryPredicate  binary_pred;
+  BinaryFunction   binary_op;
+  FlagIterator     iflags;
+  IndexIterator    interval_counts;
+  ValueIterator    interval_values;
+  BoolIterator     interval_carry;
+  Decomposition    decomp;
+  Context          context;
+
+  typedef Context context_type;
+
+  reduce_by_key_closure(InputIterator1   ikeys,
+                        InputIterator2   ivals,
+                        OutputIterator1  okeys,
+                        OutputIterator2  ovals,
+                        BinaryPredicate  binary_pred,
+                        BinaryFunction   binary_op,
+                        FlagIterator     iflags,
+                        IndexIterator    interval_counts,
+                        ValueIterator    interval_values,
+                        BoolIterator     interval_carry,
+                        Decomposition    decomp,
+                        Context          context = Context())
+    : ikeys(ikeys), ivals(ivals), okeys(okeys), ovals(ovals), binary_pred(binary_pred), binary_op(binary_op),
+      iflags(iflags), interval_counts(interval_counts), interval_values(interval_values), interval_carry(interval_carry),
+      decomp(decomp), context(context) {}
+
+  __device__ __thrust_forceinline__
+  void operator()(void)
+  {
+    typedef typename thrust::iterator_value<InputIterator1>::type KeyType;
+    typedef typename thrust::iterator_value<ValueIterator>::type  ValueType;
+    typedef typename Decomposition::index_type                    IndexType;
+    typedef typename thrust::iterator_value<FlagIterator>::type   FlagType;
+
+    const unsigned int CTA_SIZE = context_type::ThreadsPerBlock::value;
+
+// TODO centralize this mapping (__CUDA_ARCH__ -> smem bytes)
+#if __CUDA_ARCH__ >= 200
+    const unsigned int SMEM = (48 * 1024);
+#else
+    const unsigned int SMEM = (16 * 1024) - 256;
+#endif
+    const unsigned int SMEM_FIXED = CTA_SIZE * sizeof(FlagType) + sizeof(ValueType) + sizeof(IndexType) + sizeof(bool);
+    const unsigned int BOUND_1 = (SMEM - SMEM_FIXED) / ((CTA_SIZE + 1) * sizeof(ValueType));
+    const unsigned int BOUND_2 = 8 * sizeof(FlagType);
+    const unsigned int BOUND_3 = 6;
+  
+    // TODO replace this with a static_min<BOUND_1,BOUND_2,BOUND_3>::value
+    const unsigned int K = (BOUND_1 < BOUND_2) ? (BOUND_1 < BOUND_3 ? BOUND_1 : BOUND_3) : (BOUND_2 < BOUND_3 ? BOUND_2 : BOUND_3);
+  
+    __shared__ detail::uninitialized<FlagType[CTA_SIZE]>         sflag;
+    __shared__ detail::uninitialized<ValueType[K][CTA_SIZE + 1]> sdata;  // padded to avoid bank conflicts
+  
+    __shared__ detail::uninitialized<ValueType> carry_value; // storage for carry in and carry out
+    __shared__ detail::uninitialized<IndexType> carry_index;
+    __shared__ detail::uninitialized<bool>      carry_in; 
+
+    typename Decomposition::range_type interval = decomp[context.block_index()];
+    //thrust::system::detail::internal::index_range<IndexType> interval = decomp[context.block_index()];
+  
+
+    if (context.thread_index() == 0)
+    {
+      carry_in = false; // act as though the previous segment terminated just before us
+  
+      if (context.block_index() == 0)
+      {
+        carry_index = 0;
+      }
+      else
+      {
+        interval_counts += (context.block_index() - 1);
+        carry_index = *interval_counts;
+      }
+    }
+  
+    context.barrier();
+  
+    IndexType base = interval.begin();
+  
+    // advance input and output iterators
+    ikeys  += base;
+    ivals  += base;
+    iflags += base;
+    okeys  += carry_index;
+    ovals  += carry_index;
+  
+    const unsigned int unit_size = K * CTA_SIZE;
+  
+    // process full units
+    while (base + unit_size <= interval.end())
+    {
+      const unsigned int n = unit_size;
+      reduce_by_key_body<CTA_SIZE,K,true>(context, n, ikeys, ivals, okeys, ovals, binary_pred, binary_op, iflags, sflag.get(), sdata.get(), carry_in.get(), carry_index.get(), carry_value.get());
+      base   += unit_size;
+      ikeys  += unit_size;
+      ivals  += unit_size;
+      iflags += unit_size;
+      okeys  += carry_index;
+      ovals  += carry_index;
+    }
+  
+    // process partially full unit at end of input (if necessary)
+    if (base < interval.end())
+    {
+      const unsigned int n = interval.end() - base;
+      reduce_by_key_body<CTA_SIZE,K,false>(context, n, ikeys, ivals, okeys, ovals, binary_pred, binary_op, iflags, sflag.get(), sdata.get(), carry_in.get(), carry_index.get(), carry_value.get());
+    }
+  
+    if (context.thread_index() == 0)
+    {
+      interval_values += context.block_index();
+      interval_carry  += context.block_index();
+      *interval_values = carry_value;
+      *interval_carry  = carry_in;
+    }
+  }
+}; // end reduce_by_key_closure
+
+template <typename InputIterator1,
+          typename InputIterator2,
+          typename OutputIterator1,
+          typename OutputIterator2,
+          typename BinaryPredicate,
+          typename BinaryFunction>
+struct DefaultPolicy
+{
+  // typedefs
+  typedef unsigned int                                                       FlagType;
+  typedef typename thrust::iterator_traits<InputIterator1>::difference_type  IndexType;
+  typedef typename thrust::iterator_traits<InputIterator1>::value_type       KeyType;
+  typedef thrust::system::detail::internal::uniform_decomposition<IndexType> Decomposition;
+    
+  // the pseudocode for deducing the type of the temporary used below:
+  // 
+  // if BinaryFunction is AdaptableBinaryFunction
+  //   TemporaryType = AdaptableBinaryFunction::result_type
+  // else if OutputIterator2 is a "pure" output iterator
+  //   TemporaryType = InputIterator2::value_type
+  // else
+  //   TemporaryType = OutputIterator2::value_type
+  //
+  // XXX upon c++0x, TemporaryType needs to be:
+  // result_of<BinaryFunction>::type
+
+  typedef typename thrust::detail::eval_if<
+    thrust::detail::has_result_type<BinaryFunction>::value,
+    thrust::detail::result_type<BinaryFunction>,
+    thrust::detail::eval_if<
+      thrust::detail::is_output_iterator<OutputIterator2>::value,
+      thrust::iterator_value<InputIterator2>,
+      thrust::iterator_value<OutputIterator2>
+    >
+  >::type ValueType;
+ 
+  // XXX WAR problem on sm_11
+  // TODO tune this
+  const static unsigned int ThreadsPerBlock = (thrust::detail::is_pod<ValueType>::value) ? 256 : 192;
+
+  DefaultPolicy(InputIterator1 first1, InputIterator1 last1)
+    : decomp(default_decomposition<IndexType>(last1 - first1))
+  {}
+
+  // member variables
+  Decomposition decomp;
+};
+
+template <typename DerivedPolicy,
+          typename InputIterator1,
+          typename InputIterator2,
+          typename OutputIterator1,
+          typename OutputIterator2,
+          typename BinaryPredicate,
+          typename BinaryFunction,
+          typename Policy>
+  thrust::pair<OutputIterator1,OutputIterator2>
+  reduce_by_key(execution_policy<DerivedPolicy> &exec,
+                InputIterator1 keys_first, 
+                InputIterator1 keys_last,
+                InputIterator2 values_first,
+                OutputIterator1 keys_output,
+                OutputIterator2 values_output,
+                BinaryPredicate binary_pred,
+                BinaryFunction binary_op,
+                Policy policy)
+{
+    typedef typename Policy::FlagType       FlagType;
+    typedef typename Policy::Decomposition  Decomposition;
+    typedef typename Policy::IndexType      IndexType;
+    typedef typename Policy::KeyType        KeyType;
+    typedef typename Policy::ValueType      ValueType;
+
+    // temporary arrays
+    typedef thrust::detail::temporary_array<IndexType,DerivedPolicy> IndexArray;
+    typedef thrust::detail::temporary_array<KeyType,DerivedPolicy>   KeyArray;
+    typedef thrust::detail::temporary_array<ValueType,DerivedPolicy> ValueArray;
+    typedef thrust::detail::temporary_array<bool,DerivedPolicy>      BoolArray;
+
+    Decomposition decomp = policy.decomp;
+
+    // input size
+    IndexType n = keys_last - keys_first;
+
+    if (n == 0)
+      return thrust::make_pair(keys_output, values_output);
+
+    IndexArray interval_counts(exec, decomp.size());
+    ValueArray interval_values(exec, decomp.size());
+    BoolArray  interval_carry(exec, decomp.size());
+
+    // an ode to c++11 auto
+    typedef thrust::counting_iterator<IndexType> CountingIterator;
+    typedef thrust::transform_iterator<
+      tail_flag_functor<FlagType,IndexType,KeyType,BinaryPredicate>,
+      thrust::zip_iterator<
+        thrust::tuple<CountingIterator,InputIterator1,InputIterator1>
+      >
+    > FlagIterator;
+
+    FlagIterator iflag= thrust::make_transform_iterator
+       (thrust::make_zip_iterator(thrust::make_tuple(thrust::counting_iterator<IndexType>(0), keys_first, keys_first + 1)),
+        tail_flag_functor<FlagType,IndexType,KeyType,BinaryPredicate>(n, binary_pred));
+
+    // count number of tail flags per interval
+    thrust::system::cuda::detail::reduce_intervals(exec, iflag, interval_counts.begin(), thrust::plus<IndexType>(), decomp);
+
+    thrust::inclusive_scan(exec,
+                           interval_counts.begin(), interval_counts.end(),
+                           interval_counts.begin(),
+                           thrust::plus<IndexType>());
+ 
+    // determine output size
+    const IndexType N = interval_counts[interval_counts.size() - 1];
+   
+    const static unsigned int ThreadsPerBlock = Policy::ThreadsPerBlock;
+    typedef typename IndexArray::iterator IndexIterator;
+    typedef typename ValueArray::iterator ValueIterator; 
+    typedef typename BoolArray::iterator  BoolIterator;  
+    typedef detail::statically_blocked_thread_array<ThreadsPerBlock> Context;
+    typedef reduce_by_key_closure<InputIterator1,InputIterator2,OutputIterator1,OutputIterator2,BinaryPredicate,BinaryFunction,
+                                  FlagIterator,IndexIterator,ValueIterator,BoolIterator,Decomposition,Context> Closure;
+    Closure closure
+      (keys_first,  values_first,
+       keys_output, values_output,
+       binary_pred, binary_op,
+       iflag,
+       interval_counts.begin(),
+       interval_values.begin(),
+       interval_carry.begin(),
+       decomp);
+    detail::launch_closure(closure, decomp.size(), ThreadsPerBlock);
+   
+    if (decomp.size() > 1)
+    {
+      ValueArray interval_values2(exec, decomp.size());
+      IndexArray interval_counts2(exec, decomp.size());
+      BoolArray  interval_carry2(exec, decomp.size());
+
+      IndexType N2 = 
+      thrust::reduce_by_key
+        (exec,
+         thrust::make_zip_iterator(thrust::make_tuple(interval_counts.begin(), interval_carry.begin())),
+         thrust::make_zip_iterator(thrust::make_tuple(interval_counts.end(),   interval_carry.end())),
+         interval_values.begin(),
+         thrust::make_zip_iterator(thrust::make_tuple(interval_counts2.begin(), interval_carry2.begin())),
+         interval_values2.begin(),
+         thrust::equal_to< thrust::tuple<IndexType,bool> >(),
+         binary_op).first
+        -
+        thrust::make_zip_iterator(thrust::make_tuple(interval_counts2.begin(), interval_carry2.begin()));
+    
+      thrust::transform_if
+        (exec,
+         interval_values2.begin(), interval_values2.begin() + N2,
+         thrust::make_permutation_iterator(values_output, interval_counts2.begin()),
+         interval_carry2.begin(),
+         thrust::make_permutation_iterator(values_output, interval_counts2.begin()),
+         binary_op,
+         thrust::identity<bool>());
+    }
+  
+    return thrust::make_pair(keys_output + N, values_output + N); 
+}
+
+} // end namespace reduce_by_key_detail
+
+
+template <typename DerivedPolicy,
+          typename InputIterator1,
+          typename InputIterator2,
+          typename OutputIterator1,
+          typename OutputIterator2,
+          typename BinaryPredicate,
+          typename BinaryFunction>
+  thrust::pair<OutputIterator1,OutputIterator2>
+  reduce_by_key(execution_policy<DerivedPolicy> &exec,
+                InputIterator1 keys_first, 
+                InputIterator1 keys_last,
+                InputIterator2 values_first,
+                OutputIterator1 keys_output,
+                OutputIterator2 values_output,
+                BinaryPredicate binary_pred,
+                BinaryFunction binary_op)
+{
+  return reduce_by_key_detail::reduce_by_key
+    (exec, 
+     keys_first, keys_last, values_first, keys_output, values_output, binary_pred, binary_op,
+     reduce_by_key_detail::DefaultPolicy<InputIterator1,InputIterator2,OutputIterator1,OutputIterator2,BinaryPredicate,BinaryFunction>(keys_first, keys_last));
+} // end reduce_by_key()
+
+} // end namespace detail
+} // end namespace cuda
+} // end namespace system
+} // end namespace thrust
+
+__THRUST_DISABLE_MSVC_POSSIBLE_LOSS_OF_DATA_WARNING_END
+
diff --git a/compat/thrust/system/cuda/detail/reduce_intervals.h b/compat/thrust/system/cuda/detail/reduce_intervals.h
new file mode 100644
index 0000000000..505d13635c
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/reduce_intervals.h
@@ -0,0 +1,53 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file reduce_intervals.h
+ *  \brief CUDA implementations of reduce_intervals algorithms.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/cuda/detail/execution_policy.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+
+template <typename DerivedPolicy,
+          typename InputIterator,
+          typename OutputIterator,
+          typename BinaryFunction,
+          typename Decomposition>
+void reduce_intervals(execution_policy<DerivedPolicy> &exec,
+                      InputIterator input,
+                      OutputIterator output,
+                      BinaryFunction binary_op,
+                      Decomposition decomp);
+
+} // end namespace detail
+} // end namespace cuda
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/cuda/detail/reduce_intervals.inl>
+
diff --git a/compat/thrust/system/cuda/detail/reduce_intervals.inl b/compat/thrust/system/cuda/detail/reduce_intervals.inl
new file mode 100644
index 0000000000..2381769223
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/reduce_intervals.inl
@@ -0,0 +1,203 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/detail/config.h>
+
+#include <thrust/iterator/iterator_traits.h>
+
+#include <thrust/detail/minmax.h>
+#include <thrust/system/detail/internal/decompose.h>
+#include <thrust/system/cuda/detail/extern_shared_ptr.h>
+#include <thrust/system/cuda/detail/block/reduce.h>
+#include <thrust/system/cuda/detail/detail/launch_closure.h>
+#include <thrust/system/cuda/detail/detail/launch_calculator.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+
+template <typename InputIterator,
+          typename OutputIterator,
+          typename BinaryFunction,
+          typename Decomposition,
+          typename Context>
+struct commutative_reduce_intervals_closure
+{
+  InputIterator  input;
+  OutputIterator output;
+  BinaryFunction binary_op;
+  Decomposition  decomposition;
+  unsigned int shared_array_size;
+
+  typedef Context context_type;
+  context_type context;
+
+  commutative_reduce_intervals_closure(InputIterator input, OutputIterator output, BinaryFunction binary_op, Decomposition decomposition, unsigned int shared_array_size, Context context = Context())
+    : input(input), output(output), binary_op(binary_op), decomposition(decomposition), shared_array_size(shared_array_size), context(context) {}
+
+  __device__ __thrust_forceinline__
+  void operator()(void)
+  {
+    typedef typename thrust::iterator_value<OutputIterator>::type OutputType;
+    extern_shared_ptr<OutputType>  shared_array;
+
+    typedef typename Decomposition::index_type index_type;
+   
+    // this block processes results in [range.begin(), range.end())
+    thrust::system::detail::internal::index_range<index_type> range = decomposition[context.block_index()];
+
+    index_type i = range.begin() + context.thread_index();
+      
+    input += i;
+
+    if (range.size() < context.block_dimension())
+    {
+      // compute reduction with the first shared_array_size threads
+      if (context.thread_index() < thrust::min<index_type>(shared_array_size,range.size()))
+      {
+        OutputType sum = *input;
+
+        i     += shared_array_size;
+        input += shared_array_size;
+
+        while (i < range.end())
+        {
+          OutputType val = *input;
+
+          sum = binary_op(sum, val);
+
+          i      += shared_array_size;
+          input  += shared_array_size;
+        }
+
+        shared_array[context.thread_index()] = sum;  
+      }
+    }
+    else
+    {
+      // compute reduction with all blockDim.x threads
+      OutputType sum = *input;
+
+      i     += context.block_dimension();
+      input += context.block_dimension();
+
+      while (i < range.end())
+      {
+        OutputType val = *input;
+
+        sum = binary_op(sum, val);
+
+        i      += context.block_dimension();
+        input  += context.block_dimension();
+      }
+
+      // write first shared_array_size values into shared memory
+      if (context.thread_index() < shared_array_size)
+        shared_array[context.thread_index()] = sum;  
+
+      // accumulate remaining values (if any) to shared memory in stages
+      if (context.block_dimension() > shared_array_size)
+      {
+        unsigned int lb = shared_array_size;
+        unsigned int ub = shared_array_size + lb;
+        
+        while (lb < context.block_dimension())
+        {
+          context.barrier();
+
+          if (lb <= context.thread_index() && context.thread_index() < ub)
+          {
+            OutputType tmp = shared_array[context.thread_index() - lb];
+            shared_array[context.thread_index() - lb] = binary_op(tmp, sum);
+          }
+
+          lb += shared_array_size;
+          ub += shared_array_size;
+        }
+      }
+    }
+  
+    context.barrier();
+
+    block::reduce_n(context, shared_array, thrust::min<index_type>(range.size(), shared_array_size), binary_op);
+  
+    if (context.thread_index() == 0)
+    {
+      output += context.block_index();
+      *output = shared_array[0];
+    }
+  }
+};
+
+__THRUST_DISABLE_MSVC_POSSIBLE_LOSS_OF_DATA_WARNING_BEGIN
+
+template <typename ExecutionPolicy,
+          typename InputIterator,
+          typename OutputIterator,
+          typename BinaryFunction,
+          typename Decomposition>
+void reduce_intervals(execution_policy<ExecutionPolicy> &,
+                      InputIterator input,
+                      OutputIterator output,
+                      BinaryFunction binary_op,
+                      Decomposition decomp)
+{
+  // we're attempting to launch a kernel, assert we're compiling with nvcc
+  // ========================================================================
+  // X Note to the user: If you've found this line due to a compiler error, X
+  // X you need to compile your code using nvcc, rather than g++ or cl.exe  X
+  // ========================================================================
+  THRUST_STATIC_ASSERT( (thrust::detail::depend_on_instantiation<InputIterator, THRUST_DEVICE_COMPILER == THRUST_DEVICE_COMPILER_NVCC>::value) );
+
+  if (decomp.size() == 0)
+    return;
+  
+  // TODO if (decomp.size() > deviceProperties.maxGridSize[0]) throw cuda exception (or handle general case)
+
+  typedef detail::blocked_thread_array Context;
+  typedef commutative_reduce_intervals_closure<InputIterator,OutputIterator,BinaryFunction,Decomposition,Context> Closure;
+  typedef typename thrust::iterator_value<OutputIterator>::type OutputType;
+  
+  detail::launch_calculator<Closure> calculator;
+
+  thrust::tuple<size_t,size_t,size_t> config = calculator.with_variable_block_size_available_smem();
+
+  //size_t max_blocks = thrust::get<0>(config);
+  size_t block_size = thrust::get<1>(config);
+  size_t max_memory = thrust::get<2>(config);
+
+  // determine shared array size
+  size_t shared_array_size  = thrust::min(max_memory / sizeof(OutputType), block_size);
+  size_t shared_array_bytes = sizeof(OutputType) * shared_array_size;
+  
+  // TODO if (shared_array_size < 1) throw cuda exception "insufficient shared memory"
+
+  Closure closure(input, output, binary_op, decomp, shared_array_size);
+  detail::launch_closure(closure, decomp.size(), block_size, shared_array_bytes);
+}
+
+__THRUST_DISABLE_MSVC_POSSIBLE_LOSS_OF_DATA_WARNING_END
+
+} // end namespace detail
+} // end namespace cuda
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/cuda/detail/remove.h b/compat/thrust/system/cuda/detail/remove.h
new file mode 100644
index 0000000000..a307fc5fc8
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/remove.h
@@ -0,0 +1,22 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system has no special version of this algorithm 
+
diff --git a/compat/thrust/system/cuda/detail/replace.h b/compat/thrust/system/cuda/detail/replace.h
new file mode 100644
index 0000000000..a307fc5fc8
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/replace.h
@@ -0,0 +1,22 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system has no special version of this algorithm 
+
diff --git a/compat/thrust/system/cuda/detail/reverse.h b/compat/thrust/system/cuda/detail/reverse.h
new file mode 100644
index 0000000000..a307fc5fc8
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/reverse.h
@@ -0,0 +1,22 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system has no special version of this algorithm 
+
diff --git a/compat/thrust/system/cuda/detail/runtime_introspection.h b/compat/thrust/system/cuda/detail/runtime_introspection.h
new file mode 100644
index 0000000000..39f6c9fadc
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/runtime_introspection.h
@@ -0,0 +1,78 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file runtime_introspection.h
+ *  \brief Defines the interface to functions
+ *         providing introspection into the architecture
+ *         of CUDA devices.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// #include this for device_properties_t and function_attributes_t
+#include <thrust/system/cuda/detail/cuda_launch_config.h>
+
+// #include this for size_t
+#include <cstddef>
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+
+
+/*! Returns the current device ordinal.
+ */
+inline int current_device();
+
+/*! Returns a copy of the device_properties_t structure
+ *  that is associated with a given device.
+ */
+inline device_properties_t device_properties(int device_id);
+
+/*! Returns a copy of the device_properties_t structure
+ *  that is associated with the current device.
+ */
+inline device_properties_t device_properties(void);
+
+/*! Returns a copy of the function_attributes_t structure
+ *  that is associated with a given __global__ function
+ */
+template <typename KernelFunction>
+inline function_attributes_t function_attributes(KernelFunction kernel);
+
+/*! Returns the compute capability of a device in integer format.
+ *  For example, returns 10 for sm_10 and 21 for sm_21
+ *  \return The compute capability as an integer
+ */
+inline size_t compute_capability(const device_properties_t &properties);
+inline size_t compute_capability(void);
+
+
+} // end namespace detail
+} // end namespace cuda
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/cuda/detail/runtime_introspection.inl>
+
diff --git a/compat/thrust/system/cuda/detail/runtime_introspection.inl b/compat/thrust/system/cuda/detail/runtime_introspection.inl
new file mode 100644
index 0000000000..a5cc382964
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/runtime_introspection.inl
@@ -0,0 +1,169 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/system/cuda/detail/runtime_introspection.h>
+#include <thrust/system/cuda/detail/guarded_cuda_runtime_api.h>
+#include <thrust/detail/util/blocking.h>
+#include <thrust/detail/minmax.h>
+#include <thrust/system_error.h>
+#include <thrust/system/cuda/error.h>
+
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+namespace runtime_introspection_detail
+{
+
+
+inline void get_device_properties(device_properties_t &p, int device_id)
+{
+  cudaDeviceProp properties;
+  
+  cudaError_t error = cudaGetDeviceProperties(&properties, device_id);
+  
+  if(error)
+    throw thrust::system_error(error, thrust::cuda_category());
+
+  // be careful about how this is initialized!
+  device_properties_t temp = {
+    properties.major,
+    {
+      properties.maxGridSize[0],
+      properties.maxGridSize[1],
+      properties.maxGridSize[2]
+    },
+    properties.maxThreadsPerBlock,
+    properties.maxThreadsPerMultiProcessor,
+    properties.minor,
+    properties.multiProcessorCount,
+    properties.regsPerBlock,
+    properties.sharedMemPerBlock,
+    properties.warpSize
+  };
+
+  p = temp;
+} // end get_device_properties()
+
+
+} // end runtime_introspection_detail
+
+
+inline device_properties_t device_properties(int device_id)
+{
+  // cache the result of get_device_properties, because it is slow
+  // only cache the first few devices
+  static const int max_num_devices                              = 16;
+
+  static bool properties_exist[max_num_devices]                 = {0};
+  static device_properties_t device_properties[max_num_devices] = {};
+
+  if(device_id >= max_num_devices)
+  {
+    device_properties_t result;
+    runtime_introspection_detail::get_device_properties(result, device_id);
+    return result;
+  }
+
+  if(!properties_exist[device_id])
+  {
+    runtime_introspection_detail::get_device_properties(device_properties[device_id], device_id);
+
+    // disallow the compiler to move the write to properties_exist[device_id]
+    // before the initialization of device_properties[device_id]
+    __thrust_compiler_fence();
+    
+    properties_exist[device_id] = true;
+  }
+
+  return device_properties[device_id];
+}
+
+inline int current_device()
+{
+  int result = -1;
+
+  cudaError_t error = cudaGetDevice(&result);
+
+  if(error)
+    throw thrust::system_error(error, thrust::cuda_category());
+
+  if(result < 0)
+    throw thrust::system_error(cudaErrorNoDevice, thrust::cuda_category());
+
+  return result;
+}
+
+inline device_properties_t device_properties(void)
+{
+  return device_properties(current_device());
+}
+
+template <typename KernelFunction>
+inline function_attributes_t function_attributes(KernelFunction kernel)
+{
+// cudaFuncGetAttributes(), used below, only exists when __CUDACC__ is defined
+#ifdef __CUDACC__
+  typedef void (*fun_ptr_type)();
+
+  fun_ptr_type fun_ptr = reinterpret_cast<fun_ptr_type>(kernel);
+
+  cudaFuncAttributes attributes;
+  
+  cudaError_t error = cudaFuncGetAttributes(&attributes, fun_ptr);
+  
+  if(error)
+  {
+    throw thrust::system_error(error, thrust::cuda_category());
+  }
+
+  // be careful about how this is initialized!
+  function_attributes_t result = {
+    attributes.constSizeBytes,
+    attributes.localSizeBytes,
+    attributes.maxThreadsPerBlock,
+    attributes.numRegs,
+    attributes.sharedSizeBytes
+  };
+
+  return result;
+#else
+  return function_attributes_t();
+#endif // __CUDACC__
+}
+
+inline size_t compute_capability(const device_properties_t &properties)
+{
+  return 10 * properties.major + properties.minor;
+}
+
+inline size_t compute_capability(void)
+{
+  return compute_capability(device_properties());
+}
+
+
+} // end namespace detail
+} // end namespace cuda
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/cuda/detail/scan.h b/compat/thrust/system/cuda/detail/scan.h
new file mode 100644
index 0000000000..036c89a217
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/scan.h
@@ -0,0 +1,64 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file scan.h
+ *  \brief Scan operations (parallel prefix-sum) [cuda]
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/cuda/detail/execution_policy.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator,
+         typename AssociativeOperator>
+  OutputIterator inclusive_scan(execution_policy<DerivedPolicy> &exec,
+                                InputIterator first,
+                                InputIterator last,
+                                OutputIterator result,
+                                AssociativeOperator binary_op);
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator,
+         typename T,
+         typename AssociativeOperator>
+  OutputIterator exclusive_scan(execution_policy<DerivedPolicy> &exec,
+                                InputIterator first,
+                                InputIterator last,
+                                OutputIterator result,
+                                T init,
+                                AssociativeOperator binary_op);
+
+} // end namespace detail
+} // end namespace cuda
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/cuda/detail/scan.inl>
+
diff --git a/compat/thrust/system/cuda/detail/scan.inl b/compat/thrust/system/cuda/detail/scan.inl
new file mode 100644
index 0000000000..9d9c6d20ee
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/scan.inl
@@ -0,0 +1,82 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file scan.inl
+ *  \brief Inline file for scan.h.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/static_assert.h>
+
+#include <thrust/system/cuda/detail/detail/fast_scan.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator,
+         typename AssociativeOperator>
+  OutputIterator inclusive_scan(execution_policy<DerivedPolicy> &exec,
+                                InputIterator first,
+                                InputIterator last,
+                                OutputIterator result,
+                                AssociativeOperator binary_op)
+{
+    // we're attempting to launch a kernel, assert we're compiling with nvcc
+    // ========================================================================
+    // X Note to the user: If you've found this line due to a compiler error, X
+    // X you need to compile your code using nvcc, rather than g++ or cl.exe  X
+    // ========================================================================
+    THRUST_STATIC_ASSERT( (thrust::detail::depend_on_instantiation<InputIterator, THRUST_DEVICE_COMPILER == THRUST_DEVICE_COMPILER_NVCC>::value) );
+    
+    return thrust::system::cuda::detail::detail::fast_scan::inclusive_scan(exec, first, last, result, binary_op);
+}
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator,
+         typename T,
+         typename AssociativeOperator>
+  OutputIterator exclusive_scan(execution_policy<DerivedPolicy> &exec,
+                                InputIterator first,
+                                InputIterator last,
+                                OutputIterator result,
+                                T init,
+                                AssociativeOperator binary_op)
+{
+    // we're attempting to launch a kernel, assert we're compiling with nvcc
+    // ========================================================================
+    // X Note to the user: If you've found this line due to a compiler error, X
+    // X you need to compile your code using nvcc, rather than g++ or cl.exe  X
+    // ========================================================================
+    THRUST_STATIC_ASSERT( (thrust::detail::depend_on_instantiation<InputIterator, THRUST_DEVICE_COMPILER == THRUST_DEVICE_COMPILER_NVCC>::value) );
+
+    return thrust::system::cuda::detail::detail::fast_scan::exclusive_scan(exec, first, last, result, init, binary_op);
+}
+
+} // end namespace detail
+} // end namespace cuda
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/cuda/detail/scan_by_key.h b/compat/thrust/system/cuda/detail/scan_by_key.h
new file mode 100644
index 0000000000..a307fc5fc8
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/scan_by_key.h
@@ -0,0 +1,22 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system has no special version of this algorithm 
+
diff --git a/compat/thrust/system/cuda/detail/scatter.h b/compat/thrust/system/cuda/detail/scatter.h
new file mode 100644
index 0000000000..a307fc5fc8
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/scatter.h
@@ -0,0 +1,22 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system has no special version of this algorithm 
+
diff --git a/compat/thrust/system/cuda/detail/sequence.h b/compat/thrust/system/cuda/detail/sequence.h
new file mode 100644
index 0000000000..a307fc5fc8
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/sequence.h
@@ -0,0 +1,22 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system has no special version of this algorithm 
+
diff --git a/compat/thrust/system/cuda/detail/set_difference.inl b/compat/thrust/system/cuda/detail/set_difference.inl
new file mode 100644
index 0000000000..33d9884730
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/set_difference.inl
@@ -0,0 +1,138 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/cstdint.h>
+#include <thrust/system/cuda/detail/detail/set_operation.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+namespace set_difference_detail
+{
+
+
+struct serial_bounded_set_difference
+{
+  // max_input_size <= 32
+  template<typename Size, typename InputIterator1, typename InputIterator2, typename OutputIterator, typename Compare>
+  inline __device__
+    thrust::detail::uint32_t operator()(Size max_input_size,
+                                        InputIterator1 first1, InputIterator1 last1,
+                                        InputIterator2 first2, InputIterator2 last2,
+                                        OutputIterator result,
+                                        Compare comp)
+  {
+    thrust::detail::uint32_t active_mask = 0;
+    thrust::detail::uint32_t active_bit = 1;
+  
+    while(first1 != last1 && first2 != last2)
+    {
+      if(comp(*first1,*first2))
+      {
+        *result = *first1;
+        active_mask |= active_bit;
+        ++first1;
+      } // end if
+      else if(comp(*first2,*first1))
+      {
+        ++first2;
+      } // end else if
+      else
+      {
+        ++first1;
+        ++first2;
+      } // end else
+  
+      ++result;
+      active_bit <<= 1;
+    } // end while
+
+    while(first1 != last1)
+    {
+      *result = *first1;
+      ++first1;
+      ++result;
+      active_mask |= active_bit;
+      active_bit <<= 1;
+    }
+  
+    return active_mask;
+  }
+
+
+  template<typename Size, typename InputIterator1, typename InputIterator2, typename Compare>
+  inline __device__
+    Size count(Size max_input_size,
+               InputIterator1 first1, InputIterator1 last1,
+               InputIterator2 first2, InputIterator2 last2,
+               Compare comp)
+  {
+    Size result = 0;
+  
+    while(first1 != last1 && first2 != last2)
+    {
+      if(comp(*first1,*first2))
+      {
+        ++first1;
+        ++result;
+      } // end if
+      else if(comp(*first2,*first1))
+      {
+        ++first2;
+      } // end else if
+      else
+      {
+        ++first1;
+        ++first2;
+      } // end else
+    } // end while
+  
+    return result + last1 - first1;
+  }
+}; // end serial_bounded_set_difference
+
+
+} // end namespace set_difference_detail
+
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2, 
+	 typename RandomAccessIterator3,
+         typename Compare>
+RandomAccessIterator3 set_difference(execution_policy<DerivedPolicy> &exec,
+                                     RandomAccessIterator1 first1,
+                                     RandomAccessIterator1 last1,
+                                     RandomAccessIterator2 first2,
+                                     RandomAccessIterator2 last2,
+                                     RandomAccessIterator3 result,
+                                     Compare comp)
+{
+  return thrust::system::cuda::detail::detail::set_operation(exec, first1, last1, first2, last2, result, comp, set_difference_detail::serial_bounded_set_difference());
+} // end set_difference
+
+
+} // end namespace detail
+} // end namespace cuda
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/cuda/detail/set_intersection.inl b/compat/thrust/system/cuda/detail/set_intersection.inl
new file mode 100644
index 0000000000..e4810b6d1b
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/set_intersection.inl
@@ -0,0 +1,129 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/cstdint.h>
+#include <thrust/system/cuda/detail/detail/set_operation.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+namespace set_intersection_detail
+{
+
+
+struct serial_bounded_set_intersection
+{
+  // max_input_size <= 32
+  template<typename Size, typename InputIterator1, typename InputIterator2, typename OutputIterator, typename Compare>
+  inline __device__
+    thrust::detail::uint32_t operator()(Size max_input_size,
+                                        InputIterator1 first1, InputIterator1 last1,
+                                        InputIterator2 first2, InputIterator2 last2,
+                                        OutputIterator result,
+                                        Compare comp)
+  {
+    thrust::detail::uint32_t active_mask = 0;
+    thrust::detail::uint32_t active_bit = 1;
+  
+    while(first1 != last1 && first2 != last2)
+    {
+      if(comp(*first1,*first2))
+      {
+        ++first1;
+      } // end if
+      else if(comp(*first2,*first1))
+      {
+        ++first2;
+      } // end else if
+      else
+      {
+        *result = *first1;
+        ++first1;
+        ++first2;
+        active_mask |= active_bit;
+      } // end else
+  
+      ++result;
+      active_bit <<= 1;
+    } // end while
+  
+    return active_mask;
+  }
+
+
+  template<typename Size, typename InputIterator1, typename InputIterator2, typename Compare>
+  inline __device__
+    Size count(Size max_input_size,
+               InputIterator1 first1, InputIterator1 last1,
+               InputIterator2 first2, InputIterator2 last2,
+               Compare comp)
+  {
+    Size result = 0;
+  
+    while(first1 != last1 && first2 != last2)
+    {
+      if(comp(*first1,*first2))
+      {
+        ++first1;
+      } // end if
+      else if(comp(*first2,*first1))
+      {
+        ++first2;
+      } // end else if
+      else
+      {
+        ++result;
+        ++first1;
+        ++first2;
+      } // end else
+    } // end while
+  
+    return result;
+  }
+}; // end serial_bounded_set_intersection
+
+
+} // end namespace set_intersection_detail
+
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2, 
+	 typename RandomAccessIterator3,
+         typename Compare>
+RandomAccessIterator3 set_intersection(execution_policy<DerivedPolicy> &exec,
+                                       RandomAccessIterator1 first1,
+                                       RandomAccessIterator1 last1,
+                                       RandomAccessIterator2 first2,
+                                       RandomAccessIterator2 last2,
+                                       RandomAccessIterator3 result,
+                                       Compare comp)
+{
+  return thrust::system::cuda::detail::detail::set_operation(exec, first1, last1, first2, last2, result, comp, set_intersection_detail::serial_bounded_set_intersection());
+} // end set_intersection
+
+
+} // end namespace detail
+} // end namespace cuda
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/cuda/detail/set_operations.h b/compat/thrust/system/cuda/detail/set_operations.h
new file mode 100644
index 0000000000..040e3419ce
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/set_operations.h
@@ -0,0 +1,97 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/cuda/detail/execution_policy.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2, 
+	 typename RandomAccessIterator3,
+         typename Compare>
+RandomAccessIterator3 set_difference(execution_policy<DerivedPolicy> &exec,
+                                     RandomAccessIterator1 first1,
+                                     RandomAccessIterator1 last1,
+                                     RandomAccessIterator2 first2,
+                                     RandomAccessIterator2 last2,
+                                     RandomAccessIterator3 result,
+                                     Compare comp);
+
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2, 
+	 typename RandomAccessIterator3,
+         typename Compare>
+RandomAccessIterator3 set_intersection(execution_policy<DerivedPolicy> &exec,
+                                       RandomAccessIterator1 first1,
+                                       RandomAccessIterator1 last1,
+                                       RandomAccessIterator2 first2,
+                                       RandomAccessIterator2 last2,
+                                       RandomAccessIterator3 result,
+                                       Compare comp);
+
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2, 
+	 typename RandomAccessIterator3,
+         typename Compare>
+RandomAccessIterator3 set_symmetric_difference(execution_policy<DerivedPolicy> &exec,
+                                               RandomAccessIterator1 first1,
+                                               RandomAccessIterator1 last1,
+                                               RandomAccessIterator2 first2,
+                                               RandomAccessIterator2 last2,
+                                               RandomAccessIterator3 result,
+                                               Compare comp);
+
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2, 
+	 typename RandomAccessIterator3,
+         typename Compare>
+RandomAccessIterator3 set_union(execution_policy<DerivedPolicy> &exec,
+                                RandomAccessIterator1 first1,
+                                RandomAccessIterator1 last1,
+                                RandomAccessIterator2 first2,
+                                RandomAccessIterator2 last2,
+                                RandomAccessIterator3 result,
+                                Compare comp);
+
+
+} // end detail
+} // end cuda
+} // end system
+} // end thrust
+
+#include <thrust/system/cuda/detail/set_difference.inl>
+#include <thrust/system/cuda/detail/set_intersection.inl>
+#include <thrust/system/cuda/detail/set_symmetric_difference.inl>
+#include <thrust/system/cuda/detail/set_union.inl>
+
diff --git a/compat/thrust/system/cuda/detail/set_symmetric_difference.inl b/compat/thrust/system/cuda/detail/set_symmetric_difference.inl
new file mode 100644
index 0000000000..112c955bc0
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/set_symmetric_difference.inl
@@ -0,0 +1,150 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/cstdint.h>
+#include <thrust/system/cuda/detail/detail/set_operation.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+namespace set_symmetric_difference_detail
+{
+
+
+struct serial_bounded_set_symmetric_difference
+{
+  // max_input_size <= 32
+  template<typename Size, typename InputIterator1, typename InputIterator2, typename OutputIterator, typename Compare>
+  inline __device__
+    thrust::detail::uint32_t operator()(Size max_input_size,
+                                        InputIterator1 first1, InputIterator1 last1,
+                                        InputIterator2 first2, InputIterator2 last2,
+                                        OutputIterator result,
+                                        Compare comp)
+  {
+    thrust::detail::uint32_t active_mask = 0;
+    thrust::detail::uint32_t active_bit = 1;
+  
+    while(first1 != last1 && first2 != last2)
+    {
+      if(comp(*first1,*first2))
+      {
+        *result = *first1;
+        active_mask |= active_bit;
+        ++first1;
+      } // end if
+      else if(comp(*first2,*first1))
+      {
+        *result = *first2;
+        active_mask |= active_bit;
+        ++first2;
+      } // end else if
+      else
+      {
+        ++first1;
+        ++first2;
+      } // end else
+  
+      ++result;
+      active_bit <<= 1;
+    } // end while
+
+    while(first1 != last1)
+    {
+      *result = *first1;
+      ++first1;
+      ++result;
+      active_mask |= active_bit;
+      active_bit <<= 1;
+    }
+
+    while(first2 != last2)
+    {
+      *result = *first2;
+      ++first2;
+      ++result;
+      active_mask |= active_bit;
+      active_bit <<= 1;
+    }
+  
+    return active_mask;
+  }
+
+
+  template<typename Size, typename InputIterator1, typename InputIterator2, typename Compare>
+  inline __device__
+    Size count(Size max_input_size,
+               InputIterator1 first1, InputIterator1 last1,
+               InputIterator2 first2, InputIterator2 last2,
+               Compare comp)
+  {
+    Size result = 0;
+  
+    while(first1 != last1 && first2 != last2)
+    {
+      if(comp(*first1,*first2))
+      {
+        ++first1;
+        ++result;
+      } // end if
+      else if(comp(*first2,*first1))
+      {
+        ++first2;
+        ++result;
+      } // end else if
+      else
+      {
+        ++first1;
+        ++first2;
+      } // end else
+    } // end while
+  
+    return result + thrust::max(last1 - first1,last2 - first2);
+  }
+}; // end serial_bounded_set_symmetric_difference
+
+
+} // end namespace set_symmetric_difference_detail
+
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2, 
+	 typename RandomAccessIterator3,
+         typename Compare>
+RandomAccessIterator3 set_symmetric_difference(execution_policy<DerivedPolicy> &exec,
+                                               RandomAccessIterator1 first1,
+                                               RandomAccessIterator1 last1,
+                                               RandomAccessIterator2 first2,
+                                               RandomAccessIterator2 last2,
+                                               RandomAccessIterator3 result,
+                                               Compare comp)
+{
+  return thrust::system::cuda::detail::detail::set_operation(exec, first1, last1, first2, last2, result, comp, set_symmetric_difference_detail::serial_bounded_set_symmetric_difference());
+} // end set_symmetric_difference
+
+
+} // end namespace detail
+} // end namespace cuda
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/cuda/detail/set_union.inl b/compat/thrust/system/cuda/detail/set_union.inl
new file mode 100644
index 0000000000..66cccab39b
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/set_union.inl
@@ -0,0 +1,150 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/cstdint.h>
+#include <thrust/system/cuda/detail/detail/set_operation.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+namespace set_union_detail
+{
+
+
+struct serial_bounded_set_union
+{
+  // max_input_size <= 32
+  template<typename Size, typename InputIterator1, typename InputIterator2, typename OutputIterator, typename Compare>
+  inline __device__
+    thrust::detail::uint32_t operator()(Size max_input_size,
+                                        InputIterator1 first1, InputIterator1 last1,
+                                        InputIterator2 first2, InputIterator2 last2,
+                                        OutputIterator result,
+                                        Compare comp)
+  {
+    thrust::detail::uint32_t active_mask = 0;
+    thrust::detail::uint32_t active_bit = 1;
+  
+    while(first1 != last1 && first2 != last2)
+    {
+      if(comp(*first1,*first2))
+      {
+        *result = *first1;
+        ++first1;
+      } // end if
+      else if(comp(*first2,*first1))
+      {
+        *result = *first2;
+        ++first2;
+      } // end else if
+      else
+      {
+        *result = *first1;
+        ++first1;
+        ++first2;
+      } // end else
+  
+      ++result;
+      active_mask |= active_bit;
+      active_bit <<= 1;
+    } // end while
+
+    while(first1 != last1)
+    {
+      *result = *first1;
+      ++first1;
+      ++result;
+      active_mask |= active_bit;
+      active_bit <<= 1;
+    }
+
+    while(first2 != last2)
+    {
+      *result = *first2;
+      ++first2;
+      ++result;
+      active_mask |= active_bit;
+      active_bit <<= 1;
+    }
+  
+    return active_mask;
+  }
+
+
+  template<typename Size, typename InputIterator1, typename InputIterator2, typename Compare>
+  inline __device__
+    Size count(Size max_input_size,
+               InputIterator1 first1, InputIterator1 last1,
+               InputIterator2 first2, InputIterator2 last2,
+               Compare comp)
+  {
+    Size result = 0;
+  
+    while(first1 != last1 && first2 != last2)
+    {
+      if(comp(*first1,*first2))
+      {
+        ++first1;
+      } // end if
+      else if(comp(*first2,*first1))
+      {
+        ++first2;
+      } // end else if
+      else
+      {
+        ++first1;
+        ++first2;
+      } // end else
+
+      ++result;
+    } // end while
+  
+    return result + thrust::max(last1 - first1,last2 - first2);
+  }
+}; // end serial_bounded_set_union
+
+
+} // end namespace set_union_detail
+
+
+template<typename ExecutionPolicy,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2, 
+	 typename RandomAccessIterator3,
+         typename Compare>
+RandomAccessIterator3 set_union(execution_policy<ExecutionPolicy> &exec,
+                                RandomAccessIterator1 first1,
+                                RandomAccessIterator1 last1,
+                                RandomAccessIterator2 first2,
+                                RandomAccessIterator2 last2,
+                                RandomAccessIterator3 result,
+                                Compare comp)
+{
+  return thrust::system::cuda::detail::detail::set_operation(exec, first1, last1, first2, last2, result, comp, set_union_detail::serial_bounded_set_union());
+} // end set_union
+
+
+} // end namespace detail
+} // end namespace cuda
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/cuda/detail/sort.h b/compat/thrust/system/cuda/detail/sort.h
new file mode 100644
index 0000000000..e78d36a76b
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/sort.h
@@ -0,0 +1,55 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/cuda/detail/execution_policy.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator,
+         typename StrictWeakOrdering>
+  void stable_sort(execution_policy<DerivedPolicy> &exec,
+                   RandomAccessIterator first,
+                   RandomAccessIterator last,
+                   StrictWeakOrdering comp);
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2,
+         typename StrictWeakOrdering>
+  void stable_sort_by_key(execution_policy<DerivedPolicy> &exec,
+                          RandomAccessIterator1 keys_first,
+                          RandomAccessIterator1 keys_last,
+                          RandomAccessIterator2 values_first,
+                          StrictWeakOrdering comp);
+
+} // end namespace detail
+} // end namespace cuda
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/cuda/detail/sort.inl>
+
diff --git a/compat/thrust/system/cuda/detail/sort.inl b/compat/thrust/system/cuda/detail/sort.inl
new file mode 100644
index 0000000000..d7e0a60c45
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/sort.inl
@@ -0,0 +1,287 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file sort.inl
+ *  \brief Inline file for sort.h
+ */
+
+#include <thrust/system/cuda/detail/detail/stable_merge_sort.h>
+#include <thrust/system/cuda/detail/detail/stable_primitive_sort.h>
+
+#include <thrust/reverse.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/detail/type_traits.h>
+#include <thrust/system/cuda/detail/execution_policy.h>
+#include <thrust/system/cuda/detail/temporary_indirect_permutation.h>
+#include <thrust/detail/trivial_sequence.h>
+
+
+/*
+ *  This file implements the following dispatch procedure for cuda::stable_sort()
+ *  and cuda::stable_sort_by_key(). The first level inspects the KeyType
+ *  and StrictWeakOrdering to determine whether a sort assuming primitive-typed
+ *  data may be applied.
+ *
+ *  If a sort assuming primitive-typed data can be applied (i.e., a radix sort),
+ *  the input ranges are first trivialized (turned into simple contiguous ranges
+ *  if they are not already). To implement descending orderings, an ascending
+ *  sort will be reversed.
+ *
+ *  If a sort assuming primitive-typed data cannot be applied, a comparison-based
+ *  sort is used. Depending on the size of the key and value types, one level of
+ *  indirection may be applied to their input ranges. This transformation
+ *  may be applied to either range to convert an ill-suited problem (i.e. sorting with
+ *  large keys or large value) into a problem more amenable to the underlying
+ *  merge sort algorithm.
+ */
+
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+
+
+namespace stable_sort_detail
+{
+
+
+template<typename KeyType, typename StrictWeakCompare>
+  struct can_use_primitive_sort
+    : thrust::detail::and_<
+        thrust::detail::is_arithmetic<KeyType>,
+        thrust::detail::or_<
+          thrust::detail::is_same<StrictWeakCompare,thrust::less<KeyType> >,
+          thrust::detail::is_same<StrictWeakCompare,thrust::greater<KeyType> >
+        >
+      >
+{};
+
+
+template<typename RandomAccessIterator, typename StrictWeakCompare>
+  struct enable_if_primitive_sort
+    : thrust::detail::enable_if<
+        can_use_primitive_sort<
+          typename iterator_value<RandomAccessIterator>::type,
+          StrictWeakCompare
+        >::value
+      >
+{};
+
+
+template<typename RandomAccessIterator, typename StrictWeakCompare>
+  struct enable_if_comparison_sort
+    : thrust::detail::disable_if<
+        can_use_primitive_sort<
+          typename iterator_value<RandomAccessIterator>::type,
+          StrictWeakCompare
+        >::value
+      >
+{};
+
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator,
+         typename StrictWeakOrdering>
+  typename enable_if_primitive_sort<RandomAccessIterator,StrictWeakOrdering>::type
+    stable_sort(execution_policy<DerivedPolicy> &exec,
+                RandomAccessIterator first,
+                RandomAccessIterator last,
+                StrictWeakOrdering comp)
+{
+  // ensure sequence has trivial iterators
+  thrust::detail::trivial_sequence<RandomAccessIterator,DerivedPolicy> keys(exec, first, last);
+  
+  // CUDA path for thrust::stable_sort with primitive keys
+  // (e.g. int, float, short, etc.) and a less<T> or greater<T> comparison
+  // method is implemented with a primitive sort
+  thrust::system::cuda::detail::detail::stable_primitive_sort(exec, keys.begin(), keys.end());
+  
+  // copy results back, if necessary
+  if(!thrust::detail::is_trivial_iterator<RandomAccessIterator>::value)
+  {
+    thrust::copy(exec, keys.begin(), keys.end(), first);
+  }
+  
+  // if comp is greater<T> then reverse the keys
+  typedef typename thrust::iterator_traits<RandomAccessIterator>::value_type KeyType;
+  const static bool reverse = thrust::detail::is_same<StrictWeakOrdering, typename thrust::greater<KeyType> >::value;
+  
+  if(reverse)
+  {
+    thrust::reverse(first, last);
+  }
+}
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator,
+         typename StrictWeakOrdering>
+  typename enable_if_comparison_sort<RandomAccessIterator,StrictWeakOrdering>::type
+    stable_sort(execution_policy<DerivedPolicy> &exec,
+                RandomAccessIterator first,
+                RandomAccessIterator last,
+                StrictWeakOrdering comp)
+{
+  // decide whether to sort keys indirectly
+  typedef typename thrust::iterator_value<RandomAccessIterator>::type KeyType;
+  typedef thrust::detail::integral_constant<bool, (sizeof(KeyType) > 8)> use_key_indirection;
+  
+  conditional_temporary_indirect_ordering<use_key_indirection, DerivedPolicy, RandomAccessIterator, StrictWeakOrdering> potentially_indirect_keys(derived_cast(exec), first, last, comp);
+  
+  thrust::system::cuda::detail::detail::stable_merge_sort(exec,
+                                                          potentially_indirect_keys.begin(),
+                                                          potentially_indirect_keys.end(),
+                                                          potentially_indirect_keys.comp());
+}
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2,
+         typename StrictWeakOrdering>
+  typename enable_if_primitive_sort<RandomAccessIterator1,StrictWeakOrdering>::type
+    stable_sort_by_key(execution_policy<DerivedPolicy> &exec,
+                       RandomAccessIterator1 keys_first,
+                       RandomAccessIterator1 keys_last,
+                       RandomAccessIterator2 values_first,
+                       StrictWeakOrdering comp)
+{
+  // path for thrust::stable_sort_by_key with primitive keys
+  // (e.g. int, float, short, etc.) and a less<T> or greater<T> comparison
+  // method is implemented with stable_primitive_sort_by_key
+  
+  // if comp is greater<T> then reverse the keys and values
+  typedef typename thrust::iterator_traits<RandomAccessIterator1>::value_type KeyType;
+  const static bool reverse = thrust::detail::is_same<StrictWeakOrdering, typename thrust::greater<KeyType> >::value;
+  
+  // note, we also have to reverse the (unordered) input to preserve stability
+  if (reverse)
+  {
+    thrust::reverse(exec, keys_first,  keys_last);
+    thrust::reverse(exec, values_first, values_first + (keys_last - keys_first));
+  }
+  
+  // ensure sequences have trivial iterators
+  thrust::detail::trivial_sequence<RandomAccessIterator1,DerivedPolicy> keys(exec, keys_first, keys_last);
+  thrust::detail::trivial_sequence<RandomAccessIterator2,DerivedPolicy> values(exec, values_first, values_first + (keys_last - keys_first));
+  
+  thrust::system::cuda::detail::detail::stable_primitive_sort_by_key(exec, keys.begin(), keys.end(), values.begin());
+  
+  // copy results back, if necessary
+  if(!thrust::detail::is_trivial_iterator<RandomAccessIterator1>::value)
+      thrust::copy(exec, keys.begin(), keys.end(), keys_first);
+  if(!thrust::detail::is_trivial_iterator<RandomAccessIterator2>::value)
+      thrust::copy(exec, values.begin(), values.end(), values_first);
+  
+  if (reverse)
+  {
+    thrust::reverse(exec, keys_first,  keys_last);
+    thrust::reverse(exec, values_first, values_first + (keys_last - keys_first));
+  }
+}
+
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2,
+         typename StrictWeakOrdering>
+  typename enable_if_comparison_sort<RandomAccessIterator1,StrictWeakOrdering>::type
+    stable_sort_by_key(execution_policy<DerivedPolicy> &exec,
+                       RandomAccessIterator1 keys_first,
+                       RandomAccessIterator1 keys_last,
+                       RandomAccessIterator2 values_first,
+                       StrictWeakOrdering comp)
+{
+  // decide whether to apply indirection to either range
+  typedef typename thrust::iterator_value<RandomAccessIterator1>::type KeyType;
+  typedef typename thrust::iterator_value<RandomAccessIterator2>::type ValueType;
+  
+  typedef thrust::detail::integral_constant<bool, (sizeof(KeyType) > 8)> use_key_indirection;
+  typedef thrust::detail::integral_constant<bool, (sizeof(ValueType) > 4)> use_value_indirection;
+  
+  conditional_temporary_indirect_ordering<
+    use_key_indirection,
+    DerivedPolicy,
+    RandomAccessIterator1,
+    StrictWeakOrdering
+  > potentially_indirect_keys(derived_cast(exec), keys_first, keys_last, comp);
+  
+  conditional_temporary_indirect_permutation<
+    use_value_indirection,
+    DerivedPolicy,
+    RandomAccessIterator2
+  > potentially_indirect_values(derived_cast(exec), values_first, values_first + (keys_last - keys_first));
+  
+  thrust::system::cuda::detail::detail::stable_merge_sort_by_key(exec,
+                                                                 potentially_indirect_keys.begin(),
+                                                                 potentially_indirect_keys.end(),
+                                                                 potentially_indirect_values.begin(),
+                                                                 potentially_indirect_keys.comp());
+}
+
+
+} // end namespace stable_sort_detail
+
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator,
+         typename StrictWeakOrdering>
+  void stable_sort(execution_policy<DerivedPolicy> &exec,
+                   RandomAccessIterator first,
+                   RandomAccessIterator last,
+                   StrictWeakOrdering comp)
+{
+  // we're attempting to launch a kernel, assert we're compiling with nvcc
+  // ========================================================================
+  // X Note to the user: If you've found this line due to a compiler error, X
+  // X you need to compile your code using nvcc, rather than g++ or cl.exe  X
+  // ========================================================================
+  THRUST_STATIC_ASSERT( (thrust::detail::depend_on_instantiation<RandomAccessIterator, THRUST_DEVICE_COMPILER == THRUST_DEVICE_COMPILER_NVCC>::value) );
+  
+  stable_sort_detail::stable_sort(exec, first, last, comp);
+}
+
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2,
+         typename StrictWeakOrdering>
+  void stable_sort_by_key(execution_policy<DerivedPolicy> &exec,
+                          RandomAccessIterator1 keys_first,
+                          RandomAccessIterator1 keys_last,
+                          RandomAccessIterator2 values_first,
+                          StrictWeakOrdering comp)
+{
+  // we're attempting to launch a kernel, assert we're compiling with nvcc
+  // ========================================================================
+  // X Note to the user: If you've found this line due to a compiler error, X
+  // X you need to compile your code using nvcc, rather than g++ or cl.exe  X
+  // ========================================================================
+  THRUST_STATIC_ASSERT( (thrust::detail::depend_on_instantiation<RandomAccessIterator1, THRUST_DEVICE_COMPILER == THRUST_DEVICE_COMPILER_NVCC>::value) );
+  
+  stable_sort_detail::stable_sort_by_key(exec, keys_first, keys_last, values_first, comp);
+}
+
+
+} // end namespace detail
+} // end namespace cuda
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/cuda/detail/swap_ranges.h b/compat/thrust/system/cuda/detail/swap_ranges.h
new file mode 100644
index 0000000000..9b1949e60f
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/swap_ranges.h
@@ -0,0 +1,22 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// cuda has no special swap_ranges
+
diff --git a/compat/thrust/system/cuda/detail/synchronize.h b/compat/thrust/system/cuda/detail/synchronize.h
new file mode 100644
index 0000000000..762f4a39c4
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/synchronize.h
@@ -0,0 +1,41 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+
+inline void synchronize(const char *message = "");
+
+inline void synchronize_if_enabled(const char *message = "");
+
+} // end namespace detail
+} // end namespace cuda
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/cuda/detail/synchronize.inl>
+
diff --git a/compat/thrust/system/cuda/detail/synchronize.inl b/compat/thrust/system/cuda/detail/synchronize.inl
new file mode 100644
index 0000000000..5f70f799c4
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/synchronize.inl
@@ -0,0 +1,56 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/system/cuda/detail/synchronize.h>
+#include <thrust/system/cuda/detail/guarded_cuda_runtime_api.h>
+#include <thrust/system/cuda/error.h>
+#include <thrust/system_error.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+
+void synchronize(const char *message)
+{
+  cudaError_t error = cudaThreadSynchronize();
+  if(error)
+  {
+    throw thrust::system_error(error, thrust::cuda_category(), std::string("synchronize: ") + message);
+  } // end if
+} // end synchronize()
+
+void synchronize_if_enabled(const char *message)
+{
+// XXX this could potentially be a runtime decision
+#if __THRUST_SYNCHRONOUS
+  synchronize(message);
+#else
+  // WAR "unused parameter" warning
+  (void) message;
+#endif
+} // end synchronize_if_enabled()
+
+} // end namespace detail
+} // end namespace cuda
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/cuda/detail/tabulate.h b/compat/thrust/system/cuda/detail/tabulate.h
new file mode 100644
index 0000000000..a307fc5fc8
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/tabulate.h
@@ -0,0 +1,22 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system has no special version of this algorithm 
+
diff --git a/compat/thrust/system/cuda/detail/temporary_buffer.h b/compat/thrust/system/cuda/detail/temporary_buffer.h
new file mode 100644
index 0000000000..628bd75719
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/temporary_buffer.h
@@ -0,0 +1,22 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system has no special temporary buffer functions
+
diff --git a/compat/thrust/system/cuda/detail/temporary_indirect_permutation.h b/compat/thrust/system/cuda/detail/temporary_indirect_permutation.h
new file mode 100644
index 0000000000..3d05f44155
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/temporary_indirect_permutation.h
@@ -0,0 +1,217 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/temporary_array.h>
+#include <thrust/sequence.h>
+#include <thrust/gather.h>
+#include <thrust/detail/function.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+
+
+template<typename DerivedPolicy, typename RandomAccessIterator>
+  struct temporary_indirect_permutation
+{
+  private:
+    typedef unsigned int size_type;
+    typedef thrust::detail::temporary_array<size_type, DerivedPolicy> array_type;
+
+  public:
+    temporary_indirect_permutation(DerivedPolicy &exec, RandomAccessIterator first, RandomAccessIterator last)
+      : m_exec(exec),
+        m_src_first(first),
+        m_src_last(last),
+        m_permutation(0, m_exec, last - first)
+    {
+      // generate sorted index sequence
+      thrust::sequence(exec, m_permutation.begin(), m_permutation.end());
+    }
+
+    ~temporary_indirect_permutation()
+    {
+      // permute the source array using the indices
+      typedef typename thrust::iterator_value<RandomAccessIterator>::type value_type;
+      thrust::detail::temporary_array<value_type, DerivedPolicy> temp(m_exec, m_src_first, m_src_last);
+      thrust::gather(m_exec, m_permutation.begin(), m_permutation.end(), temp.begin(), m_src_first);
+    }
+
+    typedef typename array_type::iterator iterator;
+
+    iterator begin()
+    {
+      return m_permutation.begin();
+    }
+
+    iterator end()
+    {
+      return m_permutation.end();
+    }
+
+  private:
+    DerivedPolicy &m_exec;
+    RandomAccessIterator m_src_first, m_src_last;
+    thrust::detail::temporary_array<size_type, DerivedPolicy> m_permutation;
+};
+
+
+template<typename DerivedPolicy, typename RandomAccessIterator>
+  struct iterator_range_with_execution_policy
+{
+  iterator_range_with_execution_policy(DerivedPolicy &exec, RandomAccessIterator first, RandomAccessIterator last)
+    : m_exec(exec), m_first(first), m_last(last)
+  {}
+
+  typedef RandomAccessIterator iterator;
+
+  iterator begin()
+  {
+    return m_first;
+  }
+
+  iterator end()
+  {
+    return m_last;
+  }
+
+  DerivedPolicy &exec()
+  {
+    return m_exec;
+  }
+
+  DerivedPolicy &m_exec;
+  RandomAccessIterator m_first, m_last;
+};
+
+
+template<typename Condition, typename DerivedPolicy, typename RandomAccessIterator>
+  struct conditional_temporary_indirect_permutation
+    : thrust::detail::eval_if<
+        Condition::value,
+        thrust::detail::identity_<temporary_indirect_permutation<DerivedPolicy, RandomAccessIterator> >,
+        thrust::detail::identity_<iterator_range_with_execution_policy<DerivedPolicy, RandomAccessIterator> >
+      >::type
+{
+  typedef typename thrust::detail::eval_if<
+    Condition::value,
+    thrust::detail::identity_<temporary_indirect_permutation<DerivedPolicy, RandomAccessIterator> >,
+    thrust::detail::identity_<iterator_range_with_execution_policy<DerivedPolicy, RandomAccessIterator> >
+  >::type super_t;
+
+  conditional_temporary_indirect_permutation(DerivedPolicy &exec, RandomAccessIterator first, RandomAccessIterator last)
+    : super_t(exec, first, last)
+  {}
+};
+
+
+template<typename DerivedPolicy, typename RandomAccessIterator, typename Compare>
+  struct temporary_indirect_ordering
+    : temporary_indirect_permutation<DerivedPolicy,RandomAccessIterator>
+{
+  private:
+    typedef temporary_indirect_permutation<DerivedPolicy,RandomAccessIterator> super_t;
+
+  public:
+    temporary_indirect_ordering(DerivedPolicy &exec, RandomAccessIterator first, RandomAccessIterator last, Compare comp)
+      : super_t(exec, first, last),
+        m_comp(first, comp)
+    {}
+
+    struct compare
+    {
+      RandomAccessIterator first;
+
+      thrust::detail::host_device_function<
+        Compare,
+        bool
+      > comp;
+
+      compare(RandomAccessIterator first, Compare comp)
+        : first(first), comp(comp)
+      {}
+
+      template<typename Integral>
+      __host__ __device__
+      bool operator()(Integral a, Integral b)
+      {
+        return comp(first[a], first[b]);
+      }
+    };
+
+    compare comp() const
+    {
+      return m_comp;
+    }
+
+  private:
+    compare m_comp;
+};
+
+
+template<typename DerivedPolicy, typename RandomAccessIterator, typename Compare>
+  struct iterator_range_with_execution_policy_and_compare
+    : iterator_range_with_execution_policy<DerivedPolicy, RandomAccessIterator>
+{
+  typedef iterator_range_with_execution_policy<DerivedPolicy, RandomAccessIterator> super_t;
+
+  iterator_range_with_execution_policy_and_compare(DerivedPolicy &exec, RandomAccessIterator first, RandomAccessIterator last, Compare comp)
+    : super_t(exec, first, last), m_comp(comp)
+  {}
+
+  typedef Compare compare;
+
+  compare comp()
+  {
+    return m_comp;
+  }
+
+  Compare m_comp;
+};
+
+
+template<typename Condition, typename DerivedPolicy, typename RandomAccessIterator, typename Compare>
+  struct conditional_temporary_indirect_ordering
+    : thrust::detail::eval_if<
+        Condition::value,
+        thrust::detail::identity_<temporary_indirect_ordering<DerivedPolicy, RandomAccessIterator, Compare> >,
+        thrust::detail::identity_<iterator_range_with_execution_policy_and_compare<DerivedPolicy, RandomAccessIterator, Compare> >
+      >::type
+{
+  typedef typename thrust::detail::eval_if<
+    Condition::value,
+    thrust::detail::identity_<temporary_indirect_ordering<DerivedPolicy, RandomAccessIterator, Compare> >,
+    thrust::detail::identity_<iterator_range_with_execution_policy_and_compare<DerivedPolicy, RandomAccessIterator, Compare> >
+  >::type super_t;
+
+  conditional_temporary_indirect_ordering(DerivedPolicy &exec, RandomAccessIterator first, RandomAccessIterator last, Compare comp)
+    : super_t(exec, first, last, comp)
+  {}
+};
+
+
+} // end detail
+} // end cuda
+} // end system
+} // end thrust
+
diff --git a/compat/thrust/system/cuda/detail/transform.h b/compat/thrust/system/cuda/detail/transform.h
new file mode 100644
index 0000000000..0af87056e7
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/transform.h
@@ -0,0 +1,22 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// cuda has no special transform
+
diff --git a/compat/thrust/system/cuda/detail/transform_reduce.h b/compat/thrust/system/cuda/detail/transform_reduce.h
new file mode 100644
index 0000000000..a307fc5fc8
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/transform_reduce.h
@@ -0,0 +1,22 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system has no special version of this algorithm 
+
diff --git a/compat/thrust/system/cuda/detail/transform_scan.h b/compat/thrust/system/cuda/detail/transform_scan.h
new file mode 100644
index 0000000000..a307fc5fc8
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/transform_scan.h
@@ -0,0 +1,22 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system has no special version of this algorithm 
+
diff --git a/compat/thrust/system/cuda/detail/trivial_copy.h b/compat/thrust/system/cuda/detail/trivial_copy.h
new file mode 100644
index 0000000000..e0e898aad4
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/trivial_copy.h
@@ -0,0 +1,54 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/system/cuda/detail/execution_policy.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator1,
+         typename Size,
+         typename RandomAccessIterator2>
+  void trivial_copy_n(execution_policy<DerivedPolicy> &exec,
+                      RandomAccessIterator1 first,
+                      Size n,
+                      RandomAccessIterator2 result);
+
+template<typename System1,
+         typename System2,
+         typename RandomAccessIterator1,
+         typename Size,
+         typename RandomAccessIterator2>
+  void trivial_copy_n(cross_system<System1,System2> &exec,
+                      RandomAccessIterator1 first,
+                      Size n,
+                      RandomAccessIterator2 result);
+
+} // end detail
+} // end cuda
+} // end system
+} // end thrust
+
+#include <thrust/system/cuda/detail/trivial_copy.inl>
+
diff --git a/compat/thrust/system/cuda/detail/trivial_copy.inl b/compat/thrust/system/cuda/detail/trivial_copy.inl
new file mode 100644
index 0000000000..d23a4ef8c2
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/trivial_copy.inl
@@ -0,0 +1,114 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/cuda/detail/trivial_copy.h>
+#include <thrust/system/cuda/detail/guarded_cuda_runtime_api.h>
+#include <thrust/system_error.h>
+#include <thrust/system/cuda/error.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/system/cpp/detail/execution_policy.h>
+#include <thrust/detail/raw_pointer_cast.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+namespace detail
+{
+
+namespace trivial_copy_detail
+{
+
+inline void checked_cudaMemcpy(void *dst, const void *src, size_t count, enum cudaMemcpyKind kind)
+{
+  cudaError_t error = cudaMemcpy(dst,src,count,kind);
+  if(error)
+  {
+    throw thrust::system_error(error, thrust::cuda_category());
+  } // end error
+} // end checked_cudaMemcpy()
+
+
+template<typename System1,
+         typename System2>
+  cudaMemcpyKind cuda_memcpy_kind(const thrust::cuda::execution_policy<System1> &,
+                                  const thrust::cpp::execution_policy<System2> &)
+{
+  return cudaMemcpyDeviceToHost;
+} // end cuda_memcpy_kind()
+
+
+template<typename System1,
+         typename System2>
+  cudaMemcpyKind cuda_memcpy_kind(const thrust::cpp::execution_policy<System1> &,
+                                  const thrust::cuda::execution_policy<System2> &)
+{
+  return cudaMemcpyHostToDevice;
+} // end cuda_memcpy_kind()
+
+
+} // end namespace trivial_copy_detail
+
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator1,
+         typename Size,
+         typename RandomAccessIterator2>
+  void trivial_copy_n(execution_policy<DerivedPolicy> &exec,
+                      RandomAccessIterator1 first,
+                      Size n,
+                      RandomAccessIterator2 result)
+{
+  typedef typename thrust::iterator_value<RandomAccessIterator1>::type T;
+
+  void *dst = thrust::raw_pointer_cast(&*result);
+  const void *src = thrust::raw_pointer_cast(&*first);
+
+  trivial_copy_detail::checked_cudaMemcpy(dst, src, n * sizeof(T), cudaMemcpyDeviceToDevice);
+}
+
+
+template<typename System1,
+         typename System2,
+         typename RandomAccessIterator1,
+         typename Size,
+         typename RandomAccessIterator2>
+  void trivial_copy_n(cross_system<System1,System2> &systems,
+                      RandomAccessIterator1 first,
+                      Size n,
+                      RandomAccessIterator2 result)
+{
+  typedef typename thrust::iterator_value<RandomAccessIterator1>::type T;
+
+  void *dst = thrust::raw_pointer_cast(&*result);
+  const void *src = thrust::raw_pointer_cast(&*first);
+
+  cudaMemcpyKind kind = trivial_copy_detail::cuda_memcpy_kind(thrust::detail::derived_cast(systems.system1), thrust::detail::derived_cast(systems.system2));
+
+  trivial_copy_detail::checked_cudaMemcpy(dst, src, n * sizeof(T), kind);
+}
+
+
+} // end namespace detail
+} // end namespace cuda
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/cuda/detail/uninitialized_copy.h b/compat/thrust/system/cuda/detail/uninitialized_copy.h
new file mode 100644
index 0000000000..a307fc5fc8
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/uninitialized_copy.h
@@ -0,0 +1,22 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system has no special version of this algorithm 
+
diff --git a/compat/thrust/system/cuda/detail/uninitialized_fill.h b/compat/thrust/system/cuda/detail/uninitialized_fill.h
new file mode 100644
index 0000000000..a307fc5fc8
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/uninitialized_fill.h
@@ -0,0 +1,22 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system has no special version of this algorithm 
+
diff --git a/compat/thrust/system/cuda/detail/unique.h b/compat/thrust/system/cuda/detail/unique.h
new file mode 100644
index 0000000000..a307fc5fc8
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/unique.h
@@ -0,0 +1,22 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system has no special version of this algorithm 
+
diff --git a/compat/thrust/system/cuda/detail/unique_by_key.h b/compat/thrust/system/cuda/detail/unique_by_key.h
new file mode 100644
index 0000000000..a307fc5fc8
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/unique_by_key.h
@@ -0,0 +1,22 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system has no special version of this algorithm 
+
diff --git a/compat/thrust/system/cuda/detail/vector.inl b/compat/thrust/system/cuda/detail/vector.inl
new file mode 100644
index 0000000000..36598764b2
--- /dev/null
+++ b/compat/thrust/system/cuda/detail/vector.inl
@@ -0,0 +1,97 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in ccudaliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/cuda/vector.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+
+template<typename T, typename Allocator>
+  vector<T,Allocator>
+    ::vector()
+      : super_t()
+{}
+
+template<typename T, typename Allocator>
+  vector<T,Allocator>
+    ::vector(size_type n)
+      : super_t(n)
+{}
+
+template<typename T, typename Allocator>
+  vector<T,Allocator>
+    ::vector(size_type n, const value_type &value)
+      : super_t(n,value)
+{}
+
+template<typename T, typename Allocator>
+  vector<T,Allocator>
+    ::vector(const vector &x)
+      : super_t(x)
+{}
+
+template<typename T, typename Allocator>
+  template<typename OtherT, typename OtherAllocator>
+    vector<T,Allocator>
+      ::vector(const thrust::detail::vector_base<OtherT,OtherAllocator> &x)
+        : super_t(x)
+{}
+
+template<typename T, typename Allocator>
+  template<typename OtherT, typename OtherAllocator>
+    vector<T,Allocator>
+      ::vector(const std::vector<OtherT,OtherAllocator> &x)
+        : super_t(x)
+{}
+
+template<typename T, typename Allocator>
+  template<typename InputIterator>
+    vector<T,Allocator>
+      ::vector(InputIterator first, InputIterator last)
+        : super_t(first,last)
+{}
+
+template<typename T, typename Allocator>
+  template<typename OtherT, typename OtherAllocator>
+    vector<T,Allocator> &
+      vector<T,Allocator>
+        ::operator=(const std::vector<OtherT,OtherAllocator> &x)
+{
+  super_t::operator=(x);
+  return *this;
+}
+
+template<typename T, typename Allocator>
+  template<typename OtherT, typename OtherAllocator>
+    vector<T,Allocator> &
+      vector<T,Allocator>
+        ::operator=(const thrust::detail::vector_base<OtherT,OtherAllocator> &x)
+{
+  super_t::operator=(x);
+  return *this;
+}
+      
+} // end cuda
+} // end system
+} // end thrust
+
diff --git a/compat/thrust/system/cuda/error.h b/compat/thrust/system/cuda/error.h
new file mode 100644
index 0000000000..8d098538db
--- /dev/null
+++ b/compat/thrust/system/cuda/error.h
@@ -0,0 +1,186 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file thrust/system/cuda/error.h
+ *  \brief CUDA-specific error reporting
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/type_traits.h>
+#include <thrust/system/error_code.h>
+#include <driver_types.h>
+
+namespace thrust
+{
+
+namespace system
+{
+
+namespace cuda
+{
+
+/*! \addtogroup system
+ *  \{
+ */
+
+// To construct an error_code after a CUDA Runtime error:
+//
+//   error_code(::cudaGetLastError(), cuda_category())
+
+// XXX N3000 prefers enum class errc { ... }
+namespace errc
+{
+
+/*! \p errc_t enumerates the kinds of CUDA Runtime errors.
+ */
+enum errc_t
+{
+  // from cuda/include/driver_types.h
+  // mirror their order
+  success                            = cudaSuccess,
+  missing_configuration              = cudaErrorMissingConfiguration,
+  memory_allocation                  = cudaErrorMemoryAllocation,
+  initialization_error               = cudaErrorInitializationError,
+  launch_failure                     = cudaErrorLaunchFailure,
+  prior_launch_failure               = cudaErrorPriorLaunchFailure,
+  launch_timeout                     = cudaErrorLaunchTimeout,
+  launch_out_of_resources            = cudaErrorLaunchOutOfResources,
+  invalid_device_function            = cudaErrorInvalidDeviceFunction,
+  invalid_configuration              = cudaErrorInvalidConfiguration,
+  invalid_device                     = cudaErrorInvalidDevice,
+  invalid_value                      = cudaErrorInvalidValue,
+  invalid_pitch_value                = cudaErrorInvalidPitchValue,
+  invalid_symbol                     = cudaErrorInvalidSymbol,
+  map_buffer_object_failed           = cudaErrorMapBufferObjectFailed,
+  unmap_buffer_object_failed         = cudaErrorUnmapBufferObjectFailed,
+  invalid_host_pointer               = cudaErrorInvalidHostPointer,
+  invalid_device_pointer             = cudaErrorInvalidDevicePointer,
+  invalid_texture                    = cudaErrorInvalidTexture,
+  invalid_texture_binding            = cudaErrorInvalidTextureBinding,
+  invalid_channel_descriptor         = cudaErrorInvalidChannelDescriptor,
+  invalid_memcpy_direction           = cudaErrorInvalidMemcpyDirection,
+  address_of_constant_error          = cudaErrorAddressOfConstant,
+  texture_fetch_failed               = cudaErrorTextureFetchFailed,
+  texture_not_bound                  = cudaErrorTextureNotBound,
+  synchronization_error              = cudaErrorSynchronizationError,
+  invalid_filter_setting             = cudaErrorInvalidFilterSetting,
+  invalid_norm_setting               = cudaErrorInvalidNormSetting,
+  mixed_device_execution             = cudaErrorMixedDeviceExecution,
+  cuda_runtime_unloading             = cudaErrorCudartUnloading,
+  unknown                            = cudaErrorUnknown,
+  not_yet_implemented                = cudaErrorNotYetImplemented,
+  memory_value_too_large             = cudaErrorMemoryValueTooLarge,
+  invalid_resource_handle            = cudaErrorInvalidResourceHandle,
+  not_ready                          = cudaErrorNotReady,
+  insufficient_driver                = cudaErrorInsufficientDriver,
+  set_on_active_process_error        = cudaErrorSetOnActiveProcess,
+  no_device                          = cudaErrorNoDevice,
+  ecc_uncorrectable                  = cudaErrorECCUncorrectable,
+
+#if CUDART_VERSION >= 4020
+  shared_object_symbol_not_found     = cudaErrorSharedObjectSymbolNotFound,
+  shared_object_init_failed          = cudaErrorSharedObjectInitFailed,
+  unsupported_limit                  = cudaErrorUnsupportedLimit,
+  duplicate_variable_name            = cudaErrorDuplicateVariableName,
+  duplicate_texture_name             = cudaErrorDuplicateTextureName,
+  duplicate_surface_name             = cudaErrorDuplicateSurfaceName,
+  devices_unavailable                = cudaErrorDevicesUnavailable,
+  invalid_kernel_image               = cudaErrorInvalidKernelImage,
+  no_kernel_image_for_device         = cudaErrorNoKernelImageForDevice,
+  incompatible_driver_context        = cudaErrorIncompatibleDriverContext,
+  peer_access_already_enabled        = cudaErrorPeerAccessAlreadyEnabled,
+  peer_access_not_enabled            = cudaErrorPeerAccessNotEnabled,
+  device_already_in_use              = cudaErrorDeviceAlreadyInUse,
+  profiler_disabled                  = cudaErrorProfilerDisabled,
+  assert_triggered                   = cudaErrorAssert,
+  too_many_peers                     = cudaErrorTooManyPeers,
+  host_memory_already_registered     = cudaErrorHostMemoryAlreadyRegistered,
+  host_memory_not_registered         = cudaErrorHostMemoryNotRegistered,
+  operating_system_error             = cudaErrorOperatingSystem,
+#endif
+
+#if CUDART_VERSION >= 5000
+  peer_access_unsupported            = cudaErrorPeerAccessUnsupported,
+  launch_max_depth_exceeded          = cudaErrorLaunchMaxDepthExceeded,
+  launch_file_scoped_texture_used    = cudaErrorLaunchFileScopedTex,
+  launch_file_scoped_surface_used    = cudaErrorLaunchFileScopedSurf,
+  sync_depth_exceeded                = cudaErrorSyncDepthExceeded,
+  attempted_operation_not_permitted  = cudaErrorNotPermitted,
+  attempted_operation_not_supported  = cudaErrorNotSupported,
+#endif
+
+  startup_failure                    = cudaErrorStartupFailure
+}; // end errc_t
+
+
+} // end namespace errc
+
+} // end namespace cuda
+
+/*! \return A reference to an object of a type derived from class \p thrust::error_category.
+ *  \note The object's \p equivalent virtual functions shall behave as specified
+ *        for the class \p thrust::error_category. The object's \p name virtual function shall
+ *        return a pointer to the string <tt>"cuda"</tt>. The object's
+ *        \p default_error_condition virtual function shall behave as follows:
+ *
+ *        If the argument <tt>ev</tt> corresponds to a CUDA error value, the function
+ *        shall return <tt>error_condition(ev,cuda_category())</tt>.
+ *        Otherwise, the function shall return <tt>system_category.default_error_condition(ev)</tt>.
+ */
+inline const error_category &cuda_category(void);
+
+
+// XXX N3000 prefers is_error_code_enum<cuda::errc>
+
+/*! Specialization of \p is_error_code_enum for \p cuda::errc::errc_t
+ */
+template<> struct is_error_code_enum<cuda::errc::errc_t> : thrust::detail::true_type {};
+
+
+// XXX replace cuda::errc::errc_t with cuda::errc upon c++0x
+/*! \return <tt>error_code(static_cast<int>(e), cuda::error_category())</tt>
+ */
+inline error_code make_error_code(cuda::errc::errc_t e);
+
+
+// XXX replace cuda::errc::errc_t with cuda::errc upon c++0x
+/*! \return <tt>error_condition(static_cast<int>(e), cuda::error_category())</tt>.
+ */
+inline error_condition make_error_condition(cuda::errc::errc_t e);
+
+/*! \} // end system
+ */
+
+
+} // end system
+
+namespace cuda
+{
+
+// XXX replace with using system::cuda_errc upon c++0x
+namespace errc = system::cuda::errc;
+
+} // end cuda
+
+using system::cuda_category;
+
+} // end namespace thrust
+
+#include <thrust/system/cuda/detail/error.inl>
+
diff --git a/compat/thrust/system/cuda/execution_policy.h b/compat/thrust/system/cuda/execution_policy.h
new file mode 100644
index 0000000000..bbd33defd4
--- /dev/null
+++ b/compat/thrust/system/cuda/execution_policy.h
@@ -0,0 +1,165 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+/*! \file thrust/system/cuda/execution_policy.h
+ *  \brief Execution policies for Thrust's CUDA system.
+ */
+
+#include <thrust/detail/config.h>
+
+// get the execution policies definitions first
+#include <thrust/system/cuda/detail/execution_policy.h>
+
+// get the definition of par
+#include <thrust/system/cuda/detail/par.h>
+
+// now get all the algorithm defintitions
+
+// the order of the following #includes seems to matter, unfortunately
+
+// primitives come first, in order of increasing sophistication
+#include <thrust/system/cuda/detail/get_value.h>
+#include <thrust/system/cuda/detail/assign_value.h>
+#include <thrust/system/cuda/detail/iter_swap.h>
+
+#include <thrust/system/cuda/detail/for_each.h>
+#include <thrust/system/cuda/detail/copy.h>
+#include <thrust/system/cuda/detail/reduce.h>
+#include <thrust/system/cuda/detail/scan.h>
+#include <thrust/system/cuda/detail/sort.h>
+
+// these are alphabetical
+#include <thrust/system/cuda/detail/adjacent_difference.h>
+#include <thrust/system/cuda/detail/assign_value.h>
+#include <thrust/system/cuda/detail/binary_search.h>
+#include <thrust/system/cuda/detail/copy_if.h>
+#include <thrust/system/cuda/detail/count.h>
+#include <thrust/system/cuda/detail/equal.h>
+#include <thrust/system/cuda/detail/extrema.h>
+#include <thrust/system/cuda/detail/fill.h>
+#include <thrust/system/cuda/detail/find.h>
+#include <thrust/system/cuda/detail/gather.h>
+#include <thrust/system/cuda/detail/generate.h>
+#include <thrust/system/cuda/detail/inner_product.h>
+#include <thrust/system/cuda/detail/iter_swap.h>
+#include <thrust/system/cuda/detail/logical.h>
+#include <thrust/system/cuda/detail/malloc_and_free.h>
+#include <thrust/system/cuda/detail/merge.h>
+#include <thrust/system/cuda/detail/mismatch.h>
+#include <thrust/system/cuda/detail/partition.h>
+#include <thrust/system/cuda/detail/reduce_by_key.h>
+#include <thrust/system/cuda/detail/remove.h>
+#include <thrust/system/cuda/detail/replace.h>
+#include <thrust/system/cuda/detail/reverse.h>
+#include <thrust/system/cuda/detail/scan_by_key.h>
+#include <thrust/system/cuda/detail/scatter.h>
+#include <thrust/system/cuda/detail/sequence.h>
+#include <thrust/system/cuda/detail/set_operations.h>
+#include <thrust/system/cuda/detail/sort.h>
+#include <thrust/system/cuda/detail/swap_ranges.h>
+#include <thrust/system/cuda/detail/tabulate.h>
+#include <thrust/system/cuda/detail/transform.h>
+#include <thrust/system/cuda/detail/transform_reduce.h>
+#include <thrust/system/cuda/detail/transform_scan.h>
+#include <thrust/system/cuda/detail/uninitialized_copy.h>
+#include <thrust/system/cuda/detail/uninitialized_fill.h>
+#include <thrust/system/cuda/detail/unique.h>
+#include <thrust/system/cuda/detail/unique_by_key.h>
+
+
+// define these entities here for the purpose of Doxygenating them
+// they are actually defined elsewhere
+#if 0
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+
+
+/*! \addtogroup execution_policies
+ *  \{
+ */
+
+
+/*! \p thrust::cuda::execution_policy is the base class for all Thrust parallel execution
+ *  policies which are derived from Thrust's CUDA backend system.
+ */
+template<typename DerivedPolicy>
+struct execution_policy : thrust::execution_policy<DerivedPolicy>
+{};
+
+
+/*! \p cuda::tag is a type representing Thrust's CUDA backend system in C++'s type system.
+ *  Iterators "tagged" with a type which is convertible to \p cuda::tag assert that they may be
+ *  "dispatched" to algorithm implementations in the \p cuda system.
+ */
+struct tag : thrust::system::cuda::execution_policy<tag> { unspecified };
+
+
+/*! \p thrust::cuda::par is the parallel execution policy associated with Thrust's CUDA
+ *  backend system.
+ *
+ *  Instead of relying on implicit algorithm dispatch through iterator system tags, users may
+ *  directly target Thrust's CUDA backend system by providing \p thrust::cuda::par as an algorithm
+ *  parameter.
+ *
+ *  Explicit dispatch can be useful in avoiding the introduction of data copies into containers such
+ *  as \p thrust::cuda::vector.
+ *
+ *  The type of \p thrust::cuda::par is implementation-defined.
+ *
+ *  The following code snippet demonstrates how to use \p thrust::cuda::par to explicitly dispatch an
+ *  invocation of \p thrust::for_each to the CUDA backend system:
+ *
+ *  \code
+ *  #include <thrust/for_each.h>
+ *  #include <thrust/system/cuda/execution_policy.h>
+ *  #include <cstdio>
+ *
+ *  struct printf_functor
+ *  {
+ *    __host__ __device__
+ *    void operator()(int x)
+ *    {
+ *      printf("%d\n");
+ *    }
+ *  };
+ *  ...
+ *  int vec[3];
+ *  vec[0] = 0; vec[1] = 1; vec[2] = 2;
+ *
+ *  thrust::for_each(thrust::cuda::par, vec.begin(), vec.end(), printf_functor());
+ *
+ *  // 0 1 2 is printed to standard output in some unspecified order
+ *  \endcode
+ */
+static const unspecified par;
+
+
+/*! \}
+ */
+
+
+} // end cuda
+} // end system
+} // end thrust
+#endif
+
+
diff --git a/compat/thrust/system/cuda/experimental/pinned_allocator.h b/compat/thrust/system/cuda/experimental/pinned_allocator.h
new file mode 100644
index 0000000000..5294659e62
--- /dev/null
+++ b/compat/thrust/system/cuda/experimental/pinned_allocator.h
@@ -0,0 +1,239 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+/*! \file thrust/system/cuda/experimental/pinned_allocator.h
+ *  \brief An allocator which creates new elements in "pinned" memory with \p cudaMallocHost
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/cuda/detail/guarded_cuda_runtime_api.h>
+#include <stdexcept>
+#include <limits>
+#include <string>
+#include <thrust/system/system_error.h>
+#include <thrust/system/cuda/error.h>
+
+namespace thrust
+{
+
+namespace system
+{
+
+namespace cuda
+{
+
+namespace experimental
+{
+
+/*! \addtogroup memory_management Memory Management
+ *  \addtogroup memory_management_classes
+ *  \ingroup memory_management
+ *  \{
+ */
+
+/*! \p pinned_allocator is a CUDA-specific host memory allocator
+ *  that employs \c cudaMallocHost for allocation.
+ *
+ *  \see http://www.sgi.com/tech/stl/Allocators.html
+ */
+template<typename T> class pinned_allocator;
+
+template<>
+  class pinned_allocator<void>
+{
+  public:
+    typedef void           value_type;
+    typedef void       *   pointer;
+    typedef const void *   const_pointer;
+    typedef std::size_t    size_type;
+    typedef std::ptrdiff_t difference_type;
+
+    // convert a pinned_allocator<void> to pinned_allocator<U>
+    template<typename U>
+      struct rebind
+    {
+      typedef pinned_allocator<U> other;
+    }; // end rebind
+}; // end pinned_allocator
+
+
+template<typename T>
+  class pinned_allocator
+{
+  public:
+    typedef T              value_type;
+    typedef T*             pointer;
+    typedef const T*       const_pointer;
+    typedef T&             reference;
+    typedef const T&       const_reference;
+    typedef std::size_t    size_type;
+    typedef std::ptrdiff_t difference_type;
+
+    // convert a pinned_allocator<T> to pinned_allocator<U>
+    template<typename U>
+      struct rebind
+    {
+      typedef pinned_allocator<U> other;
+    }; // end rebind
+
+    /*! \p pinned_allocator's null constructor does nothing.
+     */
+    __host__ __device__
+    inline pinned_allocator() {}
+
+    /*! \p pinned_allocator's null destructor does nothing.
+     */
+    __host__ __device__
+    inline ~pinned_allocator() {}
+
+    /*! \p pinned_allocator's copy constructor does nothing.
+     */
+    __host__ __device__
+    inline pinned_allocator(pinned_allocator const &) {}
+
+    /*! This version of \p pinned_allocator's copy constructor
+     *  is templated on the \c value_type of the \p pinned_allocator
+     *  to copy from.  It is provided merely for convenience; it
+     *  does nothing.
+     */
+    template<typename U>
+    __host__ __device__
+    inline pinned_allocator(pinned_allocator<U> const &) {}
+
+    /*! This method returns the address of a \c reference of
+     *  interest.
+     *
+     *  \p r The \c reference of interest.
+     *  \return \c r's address.
+     */
+    __host__ __device__
+    inline pointer address(reference r) { return &r; }
+
+    /*! This method returns the address of a \c const_reference
+     *  of interest.
+     *
+     *  \p r The \c const_reference of interest.
+     *  \return \c r's address.
+     */
+    __host__ __device__
+    inline const_pointer address(const_reference r) { return &r; }
+
+    /*! This method allocates storage for objects in pinned host
+     *  memory.
+     *
+     *  \p cnt The number of objects to allocate.
+     *  \return a \c pointer to the newly allocated objects.
+     *  \note This method does not invoke \p value_type's constructor.
+     *        It is the responsibility of the caller to initialize the
+     *        objects at the returned \c pointer. 
+     */
+    __host__
+    inline pointer allocate(size_type cnt,
+                            const_pointer = 0)
+    {
+      if(cnt > this->max_size())
+      {
+        throw std::bad_alloc();
+      } // end if
+
+      pointer result(0);
+      cudaError_t error = cudaMallocHost(reinterpret_cast<void**>(&result), cnt * sizeof(value_type));
+
+      if(error)
+      {
+        throw std::bad_alloc();
+      } // end if
+
+      return result;
+    } // end allocate()
+
+    /*! This method deallocates pinned host memory previously allocated
+     *  with this \c pinned_allocator.
+     *
+     *  \p p A \c pointer to the previously allocated memory.
+     *  \p cnt The number of objects previously allocated at
+     *         \p p.
+     *  \note This method does not invoke \p value_type's destructor.
+     *        It is the responsibility of the caller to destroy
+     *        the objects stored at \p p.
+     */
+    __host__
+    inline void deallocate(pointer p, size_type cnt)
+    {
+      cudaError_t error = cudaFreeHost(p);
+      
+      if(error)
+      {
+        throw thrust::system_error(error, thrust::cuda_category());
+      } // end if
+    } // end deallocate()
+
+    /*! This method returns the maximum size of the \c cnt parameter
+     *  accepted by the \p allocate() method.
+     *
+     *  \return The maximum number of objects that may be allocated
+     *          by a single call to \p allocate().
+     */
+    inline size_type max_size() const
+    {
+      return (std::numeric_limits<size_type>::max)() / sizeof(T);
+    } // end max_size()
+
+    /*! This method tests this \p pinned_allocator for equality to
+     *  another.
+     *
+     *  \param x The other \p pinned_allocator of interest.
+     *  \return This method always returns \c true.
+     */
+    __host__ __device__
+    inline bool operator==(pinned_allocator const& x) { return true; }
+
+    /*! This method tests this \p pinned_allocator for inequality
+     *  to another.
+     *
+     *  \param x The other \p pinned_allocator of interest.
+     *  \return This method always returns \c false.
+     */
+    __host__ __device__
+    inline bool operator!=(pinned_allocator const &x) { return !operator==(x); }
+}; // end pinned_allocator
+
+/*! \}
+ */
+
+} // end experimental
+
+} // end cuda
+
+} // end system
+
+// alias cuda's members at top-level
+namespace cuda
+{
+
+namespace experimental
+{
+
+using thrust::system::cuda::experimental::pinned_allocator;
+
+} // end experimental
+
+} // end cuda
+
+} // end thrust
+
diff --git a/compat/thrust/system/cuda/memory.h b/compat/thrust/system/cuda/memory.h
new file mode 100644
index 0000000000..368eea265a
--- /dev/null
+++ b/compat/thrust/system/cuda/memory.h
@@ -0,0 +1,421 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in ccudaliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+/*! \file thrust/system/cuda/memory.h
+ *  \brief Managing memory associated with Thrust's CUDA system.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/cuda/detail/execution_policy.h>
+#include <thrust/memory.h>
+#include <thrust/detail/type_traits.h>
+#include <thrust/detail/allocator/malloc_allocator.h>
+#include <ostream>
+
+namespace thrust
+{
+namespace system
+{
+namespace cuda
+{
+
+template<typename> class pointer;
+
+} // end cuda
+} // end system
+} // end thrust
+
+
+/*! \cond
+ */
+
+// specialize std::iterator_traits to avoid problems with the name of
+// pointer's constructor shadowing its nested pointer type
+// do this before pointer is defined so the specialization is correctly
+// used inside the definition
+namespace std
+{
+
+template<typename Element>
+  struct iterator_traits<thrust::system::cuda::pointer<Element> >
+{
+  private:
+    typedef thrust::system::cuda::pointer<Element> ptr;
+
+  public:
+    typedef typename ptr::iterator_category       iterator_category;
+    typedef typename ptr::value_type              value_type;
+    typedef typename ptr::difference_type         difference_type;
+    typedef ptr                                   pointer;
+    typedef typename ptr::reference               reference;
+}; // end iterator_traits
+
+} // end std
+
+/*! \endcond
+ */
+
+
+namespace thrust
+{
+namespace system
+{
+
+/*! \addtogroup system_backends Systems
+ *  \ingroup system
+ *  \{
+ */
+
+/*! \namespace thrust::system::cuda
+ *  \brief \p thrust::system::cuda is the namespace containing functionality for allocating, manipulating,
+ *         and deallocating memory available to Thrust's CUDA backend system.
+ *         The identifiers are provided in a separate namespace underneath <tt>thrust::system</tt>
+ *         for import convenience but are also aliased in the top-level <tt>thrust::tbb</tt>
+ *         namespace for easy access.
+ *
+ */
+namespace cuda
+{
+
+// forward declaration of reference for pointer
+template<typename Element> class reference;
+
+/*! \cond
+ */
+
+// XXX nvcc + msvc have trouble instantiating reference below
+//     this is a workaround
+namespace detail
+{
+
+template<typename Element>
+  struct reference_msvc_workaround
+{
+  typedef thrust::system::cuda::reference<Element> type;
+}; // end reference_msvc_workaround
+
+} // end detail
+
+/*! \endcond
+ */
+
+#if 0
+/*! \p cuda::tag is type representing Thrust's CUDA backend system in C++'s type system.
+ *  Iterators "tagged" with a type which is convertible to \p cuda::tag assert that they may be
+ *  "dispatched" to algorithm implementations in the \p cuda system.
+ */
+struct tag { unspecified };
+#endif
+
+/*! \p pointer stores a pointer to an object allocated in memory available to the cuda system.
+ *  This type provides type safety when dispatching standard algorithms on ranges resident
+ *  in cuda memory.
+ *
+ *  \p pointer has pointer semantics: it may be dereferenced and manipulated with pointer arithmetic.
+ *
+ *  \p pointer can be created with the function \p cuda::malloc, or by explicitly calling its constructor
+ *  with a raw pointer.
+ *
+ *  The raw pointer encapsulated by a \p pointer may be obtained by eiter its <tt>get</tt> member function
+ *  or the \p raw_pointer_cast function.
+ *
+ *  \note \p pointer is not a "smart" pointer; it is the programmer's responsibility to deallocate memory
+ *  pointed to by \p pointer.
+ *
+ *  \tparam T specifies the type of the pointee.
+ *
+ *  \see cuda::malloc
+ *  \see cuda::free
+ *  \see raw_pointer_cast
+ */
+template<typename T>
+  class pointer
+    : public thrust::pointer<
+               T,
+               thrust::system::cuda::tag,
+               thrust::system::cuda::reference<T>,
+               thrust::system::cuda::pointer<T>
+             >
+{
+  /*! \cond
+   */
+
+  private:
+    typedef thrust::pointer<
+      T,
+      thrust::system::cuda::tag,
+      //thrust::system::cuda::reference<T>,
+      typename detail::reference_msvc_workaround<T>::type,
+      thrust::system::cuda::pointer<T>
+    > super_t;
+
+  /*! \endcond
+   */
+
+  public:
+
+    /*! \p pointer's no-argument constructor initializes its encapsulated pointer to \c 0.
+     */
+    __host__ __device__
+    pointer() : super_t() {}
+
+    /*! This constructor allows construction of a <tt>pointer<const T></tt> from a <tt>T*</tt>.
+     *
+     *  \param ptr A raw pointer to copy from, presumed to point to a location in memory
+     *         accessible by the \p tbb system.
+     *  \tparam OtherT \p OtherT shall be convertible to \p T.
+     */
+    template<typename OtherT>
+    __host__ __device__
+    explicit pointer(OtherT *ptr) : super_t(ptr) {}
+
+    /*! This constructor allows construction from another pointer-like object with related type.
+     *
+     *  \param other The \p OtherPointer to copy.
+     *  \tparam OtherPointer The system tag associated with \p OtherPointer shall be convertible
+     *          to \p thrust::system::cuda::tag and its element type shall be convertible to \p T.
+     */
+    template<typename OtherPointer>
+    __host__ __device__
+    pointer(const OtherPointer &other,
+            typename thrust::detail::enable_if_pointer_is_convertible<
+              OtherPointer,
+              pointer
+            >::type * = 0) : super_t(other) {}
+
+    /*! Assignment operator allows assigning from another pointer-like object with related type.
+     *
+     *  \param other The other pointer-like object to assign from.
+     *  \tparam OtherPointer The system tag associated with \p OtherPointer shall be convertible
+     *          to \p thrust::system::cuda::tag and its element type shall be convertible to \p T.
+     */
+    template<typename OtherPointer>
+    __host__ __device__
+    typename thrust::detail::enable_if_pointer_is_convertible<
+      OtherPointer,
+      pointer,
+      pointer &
+    >::type
+    operator=(const OtherPointer &other)
+    {
+      return super_t::operator=(other);
+    }
+}; // end pointer
+
+
+/*! \p reference is a wrapped reference to an object stored in memory available to the \p cuda system.
+ *  \p reference is the type of the result of dereferencing a \p cuda::pointer.
+ *
+ *  \tparam T Specifies the type of the referenced object.
+ */
+template<typename T>
+  class reference
+    : public thrust::reference<
+               T,
+               thrust::system::cuda::pointer<T>,
+               thrust::system::cuda::reference<T>
+             >
+{
+  /*! \cond
+   */
+
+  private:
+    typedef thrust::reference<
+      T,
+      thrust::system::cuda::pointer<T>,
+      thrust::system::cuda::reference<T>
+    > super_t;
+
+  /*! \endcond
+   */
+
+  public:
+    /*! \cond
+     */
+
+    typedef typename super_t::value_type value_type;
+    typedef typename super_t::pointer    pointer;
+
+    /*! \endcond
+     */
+
+    /*! This constructor initializes this \p reference to refer to an object
+     *  pointed to by the given \p pointer. After this \p reference is constructed,
+     *  it shall refer to the object pointed to by \p ptr.
+     *
+     *  \param ptr A \p pointer to copy from.
+     */
+    __host__ __device__
+    explicit reference(const pointer &ptr)
+      : super_t(ptr)
+    {}
+
+    /*! This constructor accepts a const reference to another \p reference of related type.
+     *  After this \p reference is constructed, it shall refer to the same object as \p other.
+     *
+     *  \param other A \p reference to copy from.
+     *  \tparam OtherT The element type of the other \p reference.
+     *
+     *  \note This constructor is templated primarily to allow initialization of <tt>reference<const T></tt>
+     *        from <tt>reference<T></tt>.
+     */
+    template<typename OtherT>
+    __host__ __device__
+    reference(const reference<OtherT> &other,
+              typename thrust::detail::enable_if_convertible<
+                typename reference<OtherT>::pointer,
+                pointer
+              >::type * = 0)
+      : super_t(other)
+    {}
+
+    /*! Copy assignment operator copy assigns from another \p reference of related type.
+     *
+     *  \param other The other \p reference to assign from.
+     *  \return <tt>*this</tt>
+     *  \tparam OtherT The element type of the other \p reference.
+     */
+    template<typename OtherT>
+    __host__ __device__
+    reference &operator=(const reference<OtherT> &other);
+
+    /*! Assignment operator assigns from a \p value_type.
+     *
+     *  \param x The \p value_type to assign from.
+     *  \return <tt>*this</tt>
+     */
+    __host__ __device__
+    reference &operator=(const value_type &x);
+}; // end reference
+
+/*! Exchanges the values of two objects referred to by \p reference.
+ *  \p x The first \p reference of interest.
+ *  \p y The second \p reference ot interest.
+ */
+template<typename T>
+__host__ __device__
+void swap(reference<T> x, reference<T> y);
+
+/*! Allocates an area of memory available to Thrust's <tt>cuda</tt> system.
+ *  \param n Number of bytes to allocate.
+ *  \return A <tt>cuda::pointer<void></tt> pointing to the beginning of the newly
+ *          allocated memory. A null <tt>cuda::pointer<void></tt> is returned if
+ *          an error occurs.
+ *  \note The <tt>cuda::pointer<void></tt> returned by this function must be
+ *        deallocated with \p cuda::free.
+ *  \see cuda::free
+ *  \see std::malloc
+ */
+inline pointer<void> malloc(std::size_t n);
+
+/*! Allocates a typed area of memory available to Thrust's <tt>cuda</tt> system.
+ *  \param n Number of elements to allocate.
+ *  \return A <tt>cuda::pointer<T></tt> pointing to the beginning of the newly
+ *          allocated memory. A null <tt>cuda::pointer<T></tt> is returned if
+ *          an error occurs.
+ *  \note The <tt>cuda::pointer<T></tt> returned by this function must be
+ *        deallocated with \p cuda::free.
+ *  \see cuda::free
+ *  \see std::malloc
+ */
+template<typename T>
+inline pointer<T> malloc(std::size_t n);
+
+/*! Deallocates an area of memory previously allocated by <tt>cuda::malloc</tt>.
+ *  \param ptr A <tt>cuda::pointer<void></tt> pointing to the beginning of an area
+ *         of memory previously allocated with <tt>cuda::malloc</tt>.
+ *  \see cuda::malloc
+ *  \see std::free
+ */
+inline void free(pointer<void> ptr);
+
+// XXX upon c++11
+// template<typename T> using allocator = thrust::detail::malloc_allocator<T,tag,pointer<T> >;
+
+/*! \p cuda::allocator is the default allocator used by the \p cuda system's containers such as
+ *  <tt>cuda::vector</tt> if no user-specified allocator is provided. \p cuda::allocator allocates
+ *  (deallocates) storage with \p cuda::malloc (\p cuda::free).
+ */
+template<typename T>
+  struct allocator
+    : thrust::detail::malloc_allocator<
+        T,
+        tag,
+        pointer<T>
+      >
+{
+  /*! The \p rebind metafunction provides the type of an \p allocator
+   *  instantiated with another type.
+   *
+   *  \tparam U The other type to use for instantiation.
+   */
+  template<typename U>
+    struct rebind
+  {
+    /*! The typedef \p other gives the type of the rebound \p allocator.
+     */
+    typedef allocator<U> other;
+  };
+
+  /*! No-argument constructor has no effect.
+   */
+  __host__ __device__
+  inline allocator() {}
+
+  /*! Copy constructor has no effect.
+   */
+  __host__ __device__
+  inline allocator(const allocator &) {}
+
+  /*! Constructor from other \p allocator has no effect.
+   */
+  template<typename U>
+  __host__ __device__
+  inline allocator(const allocator<U> &) {}
+
+  /*! Destructor has no effect.
+   */
+  __host__ __device__
+  inline ~allocator() {}
+}; // end allocator
+
+} // end cuda
+
+/*! \}
+ */
+
+} // end system
+
+/*! \namespace thrust::cuda
+ *  \brief \p thrust::cuda is a top-level alias for thrust::system::cuda.
+ */
+namespace cuda
+{
+
+using thrust::system::cuda::pointer;
+using thrust::system::cuda::reference;
+using thrust::system::cuda::malloc;
+using thrust::system::cuda::free;
+using thrust::system::cuda::allocator;
+
+} // end cuda
+
+} // end thrust
+
+#include <thrust/system/cuda/detail/memory.inl>
+
diff --git a/compat/thrust/system/cuda/vector.h b/compat/thrust/system/cuda/vector.h
new file mode 100644
index 0000000000..ac47a84e1d
--- /dev/null
+++ b/compat/thrust/system/cuda/vector.h
@@ -0,0 +1,148 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in ccudaliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+/*! \file thrust/system/cuda/vector.h
+ *  \brief A dynamically-sizable array of elements which reside in memory available to
+ *         Thrust's CUDA system.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/cuda/memory.h>
+#include <thrust/detail/vector_base.h>
+#include <vector>
+
+namespace thrust
+{
+
+// forward declaration of host_vector
+template<typename T, typename Allocator> class host_vector;
+
+namespace system
+{
+namespace cuda
+{
+
+// XXX upon c++11
+// template<typename T, typename Allocator = allocator<T> > using vector = thrust::detail::vector_base<T,Allocator>;
+
+/*! \p cuda::vector is a container that supports random access to elements,
+ *  constant time removal of elements at the end, and linear time insertion
+ *  and removal of elements at the beginning or in the middle. The number of
+ *  elements in a \p cuda::vector may vary dynamically; memory management is
+ *  automatic. The elements contained in a \p cuda::vector reside in memory
+ *  available to the \p cuda system.
+ *
+ *  \tparam T The element type of the \p cuda::vector.
+ *  \tparam Allocator The allocator type of the \p cuda::vector. Defaults to \p cuda::allocator.
+ *
+ *  \see http://www.sgi.com/tech/stl/Vector.html
+ *  \see host_vector For the documentation of the complete interface which is
+ *                   shared by \p cuda::vector
+ *  \see device_vector
+ */
+template<typename T, typename Allocator = allocator<T> >
+  class vector
+    : public thrust::detail::vector_base<T,Allocator>
+{
+  /*! \cond
+   */
+  private:
+    typedef thrust::detail::vector_base<T,Allocator> super_t;
+  /*! \endcond
+   */
+
+  public:
+
+  /*! \cond
+   */
+    typedef typename super_t::size_type  size_type;
+    typedef typename super_t::value_type value_type;
+  /*! \endcond
+   */
+
+    /*! This constructor creates an empty \p cuda::vector.
+     */
+    vector();
+
+    /*! This constructor creates a \p cuda::vector with \p n default-constructed elements.
+     *  \param n The size of the \p cuda::vector to create.
+     */
+    explicit vector(size_type n);
+
+    /*! This constructor creates a \p cuda::vector with \p n copies of \p value.
+     *  \param n The size of the \p cuda::vector to create.
+     *  \param value An element to copy.
+     */
+    explicit vector(size_type n, const value_type &value);
+
+    /*! Copy constructor copies from another \p cuda::vector.
+     *  \param x The other \p cuda::vector to copy.
+     */
+    vector(const vector &x);
+
+    /*! This constructor copies from another Thrust vector-like object.
+     *  \param x The other object to copy from.
+     */
+    template<typename OtherT, typename OtherAllocator>
+    vector(const thrust::detail::vector_base<OtherT,OtherAllocator> &x);
+
+    /*! This constructor copies from a \c std::vector.
+     *  \param x The \c std::vector to copy from.
+     */
+    template<typename OtherT, typename OtherAllocator>
+    vector(const std::vector<OtherT,OtherAllocator> &x);
+
+    /*! This constructor creates a \p cuda::vector by copying from a range.
+     *  \param first The beginning of the range.
+     *  \param last The end of the range.
+     */
+    template<typename InputIterator>
+    vector(InputIterator first, InputIterator last);
+
+    // XXX vector_base should take a Derived type so we don't have to define these superfluous assigns
+    //
+    /*! Assignment operator assigns from a \c std::vector.
+     *  \param x The \c std::vector to assign from.
+     *  \return <tt>*this</tt>
+     */
+    template<typename OtherT, typename OtherAllocator>
+    vector &operator=(const std::vector<OtherT,OtherAllocator> &x);
+
+    /*! Assignment operator assigns from another Thrust vector-like object.
+     *  \param x The other object to assign from.
+     *  \return <tt>*this</tt>
+     */
+    template<typename OtherT, typename OtherAllocator>
+    vector &operator=(const thrust::detail::vector_base<OtherT,OtherAllocator> &x);
+}; // end vector
+
+} // end cuda
+} // end system
+
+// alias system::cuda names at top-level
+namespace cuda
+{
+
+using thrust::system::cuda::vector;
+
+} // end cuda
+
+} // end thrust
+
+#include <thrust/system/cuda/detail/vector.inl>
+
diff --git a/compat/thrust/system/detail/adl/adjacent_difference.h b/compat/thrust/system/detail/adl/adjacent_difference.h
new file mode 100644
index 0000000000..246c1163bb
--- /dev/null
+++ b/compat/thrust/system/detail/adl/adjacent_difference.h
@@ -0,0 +1,32 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// the purpose of this header is to #include the adjacent_difference.h header
+// of the host and device systems. It should be #included in any
+// code which uses adl to dispatch adjacent_difference
+
+#define __THRUST_HOST_SYSTEM_ADJACENT_DIFFERENCE_HEADER <__THRUST_HOST_SYSTEM_ROOT/detail/adjacent_difference.h>
+#include __THRUST_HOST_SYSTEM_ADJACENT_DIFFERENCE_HEADER
+#undef __THRUST_HOST_SYSTEM_ADJACENT_DIFFERENCE_HEADER
+
+#define __THRUST_DEVICE_SYSTEM_ADJACENT_DIFFERENCE_HEADER <__THRUST_DEVICE_SYSTEM_ROOT/detail/adjacent_difference.h>
+#include __THRUST_DEVICE_SYSTEM_ADJACENT_DIFFERENCE_HEADER
+#undef __THRUST_DEVICE_SYSTEM_ADJACENT_DIFFERENCE_HEADER
+
diff --git a/compat/thrust/system/detail/adl/assign_value.h b/compat/thrust/system/detail/adl/assign_value.h
new file mode 100644
index 0000000000..b5c588ace7
--- /dev/null
+++ b/compat/thrust/system/detail/adl/assign_value.h
@@ -0,0 +1,32 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// the purpose of this header is to #include the assign_value.h header
+// of the host and device systems. It should be #included in any
+// code which uses adl to dispatch assign_value
+
+#define __THRUST_HOST_SYSTEM_ASSIGN_VALUE_HEADER <__THRUST_HOST_SYSTEM_ROOT/detail/assign_value.h>
+#include __THRUST_HOST_SYSTEM_ASSIGN_VALUE_HEADER
+#undef __THRUST_HOST_SYSTEM_ASSIGN_VALUE_HEADER
+
+#define __THRUST_DEVICE_SYSTEM_ASSIGN_VALUE_HEADER <__THRUST_DEVICE_SYSTEM_ROOT/detail/assign_value.h>
+#include __THRUST_DEVICE_SYSTEM_ASSIGN_VALUE_HEADER
+#undef __THRUST_DEVICE_SYSTEM_ASSIGN_VALUE_HEADER
+
diff --git a/compat/thrust/system/detail/adl/binary_search.h b/compat/thrust/system/detail/adl/binary_search.h
new file mode 100644
index 0000000000..7accfbc381
--- /dev/null
+++ b/compat/thrust/system/detail/adl/binary_search.h
@@ -0,0 +1,32 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// the purpose of this header is to #include the binary_search.h header
+// of the host and device systems. It should be #included in any
+// code which uses adl to dispatch binary_search
+
+#define __THRUST_HOST_SYSTEM_BINARY_SEARCH_HEADER <__THRUST_HOST_SYSTEM_ROOT/detail/binary_search.h>
+#include __THRUST_HOST_SYSTEM_BINARY_SEARCH_HEADER
+#undef __THRUST_HOST_SYSTEM_BINARY_SEARCH_HEADER
+
+#define __THRUST_DEVICE_SYSTEM_BINARY_SEARCH_HEADER <__THRUST_DEVICE_SYSTEM_ROOT/detail/binary_search.h>
+#include __THRUST_DEVICE_SYSTEM_BINARY_SEARCH_HEADER
+#undef __THRUST_DEVICE_SYSTEM_BINARY_SEARCH_HEADER
+
diff --git a/compat/thrust/system/detail/adl/copy.h b/compat/thrust/system/detail/adl/copy.h
new file mode 100644
index 0000000000..91a32cd345
--- /dev/null
+++ b/compat/thrust/system/detail/adl/copy.h
@@ -0,0 +1,32 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// the purpose of this header is to #include the copy.h header
+// of the host and device systems. It should be #included in any
+// code which uses adl to dispatch copy
+
+#define __THRUST_HOST_SYSTEM_COPY_HEADER <__THRUST_HOST_SYSTEM_ROOT/detail/copy.h>
+#include __THRUST_HOST_SYSTEM_COPY_HEADER
+#undef __THRUST_HOST_SYSTEM_COPY_HEADER
+
+#define __THRUST_DEVICE_SYSTEM_COPY_HEADER <__THRUST_DEVICE_SYSTEM_ROOT/detail/copy.h>
+#include __THRUST_DEVICE_SYSTEM_COPY_HEADER
+#undef __THRUST_DEVICE_SYSTEM_COPY_HEADER
+
diff --git a/compat/thrust/system/detail/adl/copy_if.h b/compat/thrust/system/detail/adl/copy_if.h
new file mode 100644
index 0000000000..fd1df977ab
--- /dev/null
+++ b/compat/thrust/system/detail/adl/copy_if.h
@@ -0,0 +1,32 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy_if.h of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// the purpose of this header is to #include the copy_if.h header
+// of the host and device systems. It should be #included in any
+// code which uses adl to dispatch copy_if
+
+#define __THRUST_HOST_SYSTEM_BINARY_SEARCH_HEADER <__THRUST_HOST_SYSTEM_ROOT/detail/copy_if.h>
+#include __THRUST_HOST_SYSTEM_BINARY_SEARCH_HEADER
+#undef __THRUST_HOST_SYSTEM_BINARY_SEARCH_HEADER
+
+#define __THRUST_DEVICE_SYSTEM_BINARY_SEARCH_HEADER <__THRUST_DEVICE_SYSTEM_ROOT/detail/copy_if.h>
+#include __THRUST_DEVICE_SYSTEM_BINARY_SEARCH_HEADER
+#undef __THRUST_DEVICE_SYSTEM_BINARY_SEARCH_HEADER
+
diff --git a/compat/thrust/system/detail/adl/count.h b/compat/thrust/system/detail/adl/count.h
new file mode 100644
index 0000000000..0dd9591a23
--- /dev/null
+++ b/compat/thrust/system/detail/adl/count.h
@@ -0,0 +1,32 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a count of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// the purpose of this header is to #include the count.h header
+// of the host and device systems. It should be #included in any
+// code which uses adl to dispatch count
+
+#define __THRUST_HOST_SYSTEM_COUNT_HEADER <__THRUST_HOST_SYSTEM_ROOT/detail/count.h>
+#include __THRUST_HOST_SYSTEM_COUNT_HEADER
+#undef __THRUST_HOST_SYSTEM_COUNT_HEADER
+
+#define __THRUST_DEVICE_SYSTEM_COUNT_HEADER <__THRUST_DEVICE_SYSTEM_ROOT/detail/count.h>
+#include __THRUST_DEVICE_SYSTEM_COUNT_HEADER
+#undef __THRUST_DEVICE_SYSTEM_COUNT_HEADER
+
diff --git a/compat/thrust/system/detail/adl/equal.h b/compat/thrust/system/detail/adl/equal.h
new file mode 100644
index 0000000000..f933d4f93a
--- /dev/null
+++ b/compat/thrust/system/detail/adl/equal.h
@@ -0,0 +1,32 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a equal of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// the purpose of this header is to #include the equal.h header
+// of the host and device systems. It should be #included in any
+// code which uses adl to dispatch equal
+
+#define __THRUST_HOST_SYSTEM_EQUAL_HEADER <__THRUST_HOST_SYSTEM_ROOT/detail/equal.h>
+#include __THRUST_HOST_SYSTEM_EQUAL_HEADER
+#undef __THRUST_HOST_SYSTEM_EQUAL_HEADER
+
+#define __THRUST_DEVICE_SYSTEM_EQUAL_HEADER <__THRUST_DEVICE_SYSTEM_ROOT/detail/equal.h>
+#include __THRUST_DEVICE_SYSTEM_EQUAL_HEADER
+#undef __THRUST_DEVICE_SYSTEM_EQUAL_HEADER
+
diff --git a/compat/thrust/system/detail/adl/extrema.h b/compat/thrust/system/detail/adl/extrema.h
new file mode 100644
index 0000000000..c766570fc3
--- /dev/null
+++ b/compat/thrust/system/detail/adl/extrema.h
@@ -0,0 +1,32 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a extrema of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// the purpose of this header is to #include the extrema.h header
+// of the host and device systems. It should be #included in any
+// code which uses adl to dispatch extrema
+
+#define __THRUST_HOST_SYSTEM_EXTREMA_HEADER <__THRUST_HOST_SYSTEM_ROOT/detail/extrema.h>
+#include __THRUST_HOST_SYSTEM_EXTREMA_HEADER
+#undef __THRUST_HOST_SYSTEM_EXTREMA_HEADER
+
+#define __THRUST_DEVICE_SYSTEM_EXTREMA_HEADER <__THRUST_DEVICE_SYSTEM_ROOT/detail/extrema.h>
+#include __THRUST_DEVICE_SYSTEM_EXTREMA_HEADER
+#undef __THRUST_DEVICE_SYSTEM_EXTREMA_HEADER
+
diff --git a/compat/thrust/system/detail/adl/fill.h b/compat/thrust/system/detail/adl/fill.h
new file mode 100644
index 0000000000..b241b8a3dc
--- /dev/null
+++ b/compat/thrust/system/detail/adl/fill.h
@@ -0,0 +1,32 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a fill of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// the purpose of this header is to #include the fill.h header
+// of the host and device systems. It should be #included in any
+// code which uses adl to dispatch fill
+
+#define __THRUST_HOST_SYSTEM_FILL_HEADER <__THRUST_HOST_SYSTEM_ROOT/detail/fill.h>
+#include __THRUST_HOST_SYSTEM_FILL_HEADER
+#undef __THRUST_HOST_SYSTEM_FILL_HEADER
+
+#define __THRUST_DEVICE_SYSTEM_FILL_HEADER <__THRUST_DEVICE_SYSTEM_ROOT/detail/fill.h>
+#include __THRUST_DEVICE_SYSTEM_FILL_HEADER
+#undef __THRUST_DEVICE_SYSTEM_FILL_HEADER
+
diff --git a/compat/thrust/system/detail/adl/find.h b/compat/thrust/system/detail/adl/find.h
new file mode 100644
index 0000000000..7c99f3e7b1
--- /dev/null
+++ b/compat/thrust/system/detail/adl/find.h
@@ -0,0 +1,32 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a fill of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// the purpose of this header is to #include the find.h header
+// of the host and device systems. It should be #included in any
+// code which uses adl to dispatch find
+
+#define __THRUST_HOST_SYSTEM_FIND_HEADER <__THRUST_HOST_SYSTEM_ROOT/detail/find.h>
+#include __THRUST_HOST_SYSTEM_FIND_HEADER
+#undef __THRUST_HOST_SYSTEM_FIND_HEADER
+
+#define __THRUST_DEVICE_SYSTEM_FIND_HEADER <__THRUST_DEVICE_SYSTEM_ROOT/detail/find.h>
+#include __THRUST_DEVICE_SYSTEM_FIND_HEADER
+#undef __THRUST_DEVICE_SYSTEM_FIND_HEADER
+
diff --git a/compat/thrust/system/detail/adl/for_each.h b/compat/thrust/system/detail/adl/for_each.h
new file mode 100644
index 0000000000..0b2717f34c
--- /dev/null
+++ b/compat/thrust/system/detail/adl/for_each.h
@@ -0,0 +1,32 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// the purpose of this header is to #include the for_each.h header
+// of the host and device systems. It should be #included in any
+// code which uses adl to dispatch for_each
+
+#define __THRUST_HOST_SYSTEM_FOR_EACH_HEADER <__THRUST_HOST_SYSTEM_ROOT/detail/for_each.h>
+#include __THRUST_HOST_SYSTEM_FOR_EACH_HEADER
+#undef __THRUST_HOST_SYSTEM_FOR_EACH_HEADER
+
+#define __THRUST_DEVICE_SYSTEM_FOR_EACH_HEADER <__THRUST_DEVICE_SYSTEM_ROOT/detail/for_each.h>
+#include __THRUST_DEVICE_SYSTEM_FOR_EACH_HEADER
+#undef __THRUST_DEVICE_SYSTEM_FOR_EACH_HEADER
+
diff --git a/compat/thrust/system/detail/adl/gather.h b/compat/thrust/system/detail/adl/gather.h
new file mode 100644
index 0000000000..da4c1d13d5
--- /dev/null
+++ b/compat/thrust/system/detail/adl/gather.h
@@ -0,0 +1,32 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a fill of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// the purpose of this header is to #include the gather.h header
+// of the host and device systems. It should be #included in any
+// code which uses adl to dispatch gather
+
+#define __THRUST_HOST_SYSTEM_FOR_EACH_HEADER <__THRUST_HOST_SYSTEM_ROOT/detail/gather.h>
+#include __THRUST_HOST_SYSTEM_FOR_EACH_HEADER
+#undef __THRUST_HOST_SYSTEM_FOR_EACH_HEADER
+
+#define __THRUST_DEVICE_SYSTEM_FOR_EACH_HEADER <__THRUST_DEVICE_SYSTEM_ROOT/detail/gather.h>
+#include __THRUST_DEVICE_SYSTEM_FOR_EACH_HEADER
+#undef __THRUST_DEVICE_SYSTEM_FOR_EACH_HEADER
+
diff --git a/compat/thrust/system/detail/adl/generate.h b/compat/thrust/system/detail/adl/generate.h
new file mode 100644
index 0000000000..3a988478f0
--- /dev/null
+++ b/compat/thrust/system/detail/adl/generate.h
@@ -0,0 +1,32 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a fill of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// the purpose of this header is to #include the generate.h header
+// of the host and device systems. It should be #included in any
+// code which uses adl to dispatch generate
+
+#define __THRUST_HOST_SYSTEM_GENERATE_HEADER <__THRUST_HOST_SYSTEM_ROOT/detail/generate.h>
+#include __THRUST_HOST_SYSTEM_GENERATE_HEADER
+#undef __THRUST_HOST_SYSTEM_GENERATE_HEADER
+
+#define __THRUST_DEVICE_SYSTEM_GENERATE_HEADER <__THRUST_DEVICE_SYSTEM_ROOT/detail/generate.h>
+#include __THRUST_DEVICE_SYSTEM_GENERATE_HEADER
+#undef __THRUST_DEVICE_SYSTEM_GENERATE_HEADER
+
diff --git a/compat/thrust/system/detail/adl/get_value.h b/compat/thrust/system/detail/adl/get_value.h
new file mode 100644
index 0000000000..ed4ef2cfef
--- /dev/null
+++ b/compat/thrust/system/detail/adl/get_value.h
@@ -0,0 +1,32 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// the purpose of this header is to #include the get_value.h header
+// of the host and device systems. It should be #included in any
+// code which uses adl to dispatch get_value
+
+#define __THRUST_HOST_SYSTEM_GET_VALUE_HEADER <__THRUST_HOST_SYSTEM_ROOT/detail/get_value.h>
+#include __THRUST_HOST_SYSTEM_GET_VALUE_HEADER
+#undef __THRUST_HOST_SYSTEM_GET_VALUE_HEADER
+
+#define __THRUST_DEVICE_SYSTEM_GET_VALUE_HEADER <__THRUST_DEVICE_SYSTEM_ROOT/detail/get_value.h>
+#include __THRUST_DEVICE_SYSTEM_GET_VALUE_HEADER
+#undef __THRUST_DEVICE_SYSTEM_GET_VALUE_HEADER
+
diff --git a/compat/thrust/system/detail/adl/inner_product.h b/compat/thrust/system/detail/adl/inner_product.h
new file mode 100644
index 0000000000..18cc65b9c1
--- /dev/null
+++ b/compat/thrust/system/detail/adl/inner_product.h
@@ -0,0 +1,32 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a fill of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// the purpose of this header is to #include the inner_product.h header
+// of the host and device systems. It should be #included in any
+// code which uses adl to dispatch inner_product
+
+#define __THRUST_HOST_SYSTEM_INNER_PRODUCT_HEADER <__THRUST_HOST_SYSTEM_ROOT/detail/inner_product.h>
+#include __THRUST_HOST_SYSTEM_INNER_PRODUCT_HEADER
+#undef __THRUST_HOST_SYSTEM_INNER_PRODUCT_HEADER
+
+#define __THRUST_DEVICE_SYSTEM_INNER_PRODUCT_HEADER <__THRUST_DEVICE_SYSTEM_ROOT/detail/inner_product.h>
+#include __THRUST_DEVICE_SYSTEM_INNER_PRODUCT_HEADER
+#undef __THRUST_DEVICE_SYSTEM_INNER_PRODUCT_HEADER
+
diff --git a/compat/thrust/system/detail/adl/iter_swap.h b/compat/thrust/system/detail/adl/iter_swap.h
new file mode 100644
index 0000000000..b302c25f7f
--- /dev/null
+++ b/compat/thrust/system/detail/adl/iter_swap.h
@@ -0,0 +1,32 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// the purpose of this header is to #include the iter_swap.h header
+// of the host and device systems. It should be #included in any
+// code which uses adl to dispatch iter_swap
+
+#define __THRUST_HOST_SYSTEM_ITER_SWAP_HEADER <__THRUST_HOST_SYSTEM_ROOT/detail/iter_swap.h>
+#include __THRUST_HOST_SYSTEM_ITER_SWAP_HEADER
+#undef __THRUST_HOST_SYSTEM_ITER_SWAP_HEADER
+
+#define __THRUST_DEVICE_SYSTEM_ITER_SWAP_HEADER <__THRUST_DEVICE_SYSTEM_ROOT/detail/iter_swap.h>
+#include __THRUST_DEVICE_SYSTEM_ITER_SWAP_HEADER
+#undef __THRUST_DEVICE_SYSTEM_ITER_SWAP_HEADER
+
diff --git a/compat/thrust/system/detail/adl/logical.h b/compat/thrust/system/detail/adl/logical.h
new file mode 100644
index 0000000000..585f71af36
--- /dev/null
+++ b/compat/thrust/system/detail/adl/logical.h
@@ -0,0 +1,32 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a fill of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// the purpose of this header is to #include the logical.h header
+// of the host and device systems. It should be #included in any
+// code which uses adl to dispatch logical
+
+#define __THRUST_HOST_SYSTEM_LOGICAL_HEADER <__THRUST_HOST_SYSTEM_ROOT/detail/logical.h>
+#include __THRUST_HOST_SYSTEM_LOGICAL_HEADER
+#undef __THRUST_HOST_SYSTEM_LOGICAL_HEADER
+
+#define __THRUST_DEVICE_SYSTEM_LOGICAL_HEADER <__THRUST_DEVICE_SYSTEM_ROOT/detail/logical.h>
+#include __THRUST_DEVICE_SYSTEM_LOGICAL_HEADER
+#undef __THRUST_DEVICE_SYSTEM_LOGICAL_HEADER
+
diff --git a/compat/thrust/system/detail/adl/malloc_and_free.h b/compat/thrust/system/detail/adl/malloc_and_free.h
new file mode 100644
index 0000000000..7d99a260df
--- /dev/null
+++ b/compat/thrust/system/detail/adl/malloc_and_free.h
@@ -0,0 +1,32 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// the purpose of this header is to #include the malloc_and_free.h header
+// of the host and device systems. It should be #included in any
+// code which uses adl to dispatch malloc_and_free
+
+#define __THRUST_HOST_SYSTEM_MALLOC_AND_FREE_HEADER <__THRUST_HOST_SYSTEM_ROOT/detail/malloc_and_free.h>
+#include __THRUST_HOST_SYSTEM_MALLOC_AND_FREE_HEADER
+#undef __THRUST_HOST_SYSTEM_MALLOC_AND_FREE_HEADER
+
+#define __THRUST_DEVICE_SYSTEM_MALLOC_AND_FREE_HEADER <__THRUST_DEVICE_SYSTEM_ROOT/detail/malloc_and_free.h>
+#include __THRUST_DEVICE_SYSTEM_MALLOC_AND_FREE_HEADER
+#undef __THRUST_DEVICE_SYSTEM_MALLOC_AND_FREE_HEADER
+
diff --git a/compat/thrust/system/detail/adl/merge.h b/compat/thrust/system/detail/adl/merge.h
new file mode 100644
index 0000000000..59d8aceb49
--- /dev/null
+++ b/compat/thrust/system/detail/adl/merge.h
@@ -0,0 +1,32 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a fill of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// the purpose of this header is to #include the merge.h header
+// of the host and device systems. It should be #included in any
+// code which uses adl to dispatch merge
+
+#define __THRUST_HOST_SYSTEM_MERGE_HEADER <__THRUST_HOST_SYSTEM_ROOT/detail/merge.h>
+#include __THRUST_HOST_SYSTEM_MERGE_HEADER
+#undef __THRUST_HOST_SYSTEM_MERGE_HEADER
+
+#define __THRUST_DEVICE_SYSTEM_MERGE_HEADER <__THRUST_DEVICE_SYSTEM_ROOT/detail/merge.h>
+#include __THRUST_DEVICE_SYSTEM_MERGE_HEADER
+#undef __THRUST_DEVICE_SYSTEM_MERGE_HEADER
+
diff --git a/compat/thrust/system/detail/adl/mismatch.h b/compat/thrust/system/detail/adl/mismatch.h
new file mode 100644
index 0000000000..d2d1831374
--- /dev/null
+++ b/compat/thrust/system/detail/adl/mismatch.h
@@ -0,0 +1,32 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a fill of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// the purpose of this header is to #include the mismatch.h header
+// of the host and device systems. It should be #included in any
+// code which uses adl to dispatch mismatch
+
+#define __THRUST_HOST_SYSTEM_MISMATCH_HEADER <__THRUST_HOST_SYSTEM_ROOT/detail/mismatch.h>
+#include __THRUST_HOST_SYSTEM_MISMATCH_HEADER
+#undef __THRUST_HOST_SYSTEM_MISMATCH_HEADER
+
+#define __THRUST_DEVICE_SYSTEM_MISMATCH_HEADER <__THRUST_DEVICE_SYSTEM_ROOT/detail/mismatch.h>
+#include __THRUST_DEVICE_SYSTEM_MISMATCH_HEADER
+#undef __THRUST_DEVICE_SYSTEM_MISMATCH_HEADER
+
diff --git a/compat/thrust/system/detail/adl/partition.h b/compat/thrust/system/detail/adl/partition.h
new file mode 100644
index 0000000000..efdc60555b
--- /dev/null
+++ b/compat/thrust/system/detail/adl/partition.h
@@ -0,0 +1,32 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a fill of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// the purpose of this header is to #include the partition.h header
+// of the host and device systems. It should be #included in any
+// code which uses adl to dispatch partition
+
+#define __THRUST_HOST_SYSTEM_PARTITION_HEADER <__THRUST_HOST_SYSTEM_ROOT/detail/partition.h>
+#include __THRUST_HOST_SYSTEM_PARTITION_HEADER
+#undef __THRUST_HOST_SYSTEM_PARTITION_HEADER
+
+#define __THRUST_DEVICE_SYSTEM_PARTITION_HEADER <__THRUST_DEVICE_SYSTEM_ROOT/detail/partition.h>
+#include __THRUST_DEVICE_SYSTEM_PARTITION_HEADER
+#undef __THRUST_DEVICE_SYSTEM_PARTITION_HEADER
+
diff --git a/compat/thrust/system/detail/adl/reduce.h b/compat/thrust/system/detail/adl/reduce.h
new file mode 100644
index 0000000000..afa00f9c60
--- /dev/null
+++ b/compat/thrust/system/detail/adl/reduce.h
@@ -0,0 +1,32 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// the purpose of this header is to #include the reduce.h header
+// of the host and device systems. It should be #included in any
+// code which uses adl to dispatch reduce
+
+#define __THRUST_HOST_SYSTEM_REDUCE_HEADER <__THRUST_HOST_SYSTEM_ROOT/detail/reduce.h>
+#include __THRUST_HOST_SYSTEM_REDUCE_HEADER
+#undef __THRUST_HOST_SYSTEM_REDUCE_HEADER
+
+#define __THRUST_DEVICE_SYSTEM_REDUCE_HEADER <__THRUST_DEVICE_SYSTEM_ROOT/detail/reduce.h>
+#include __THRUST_DEVICE_SYSTEM_REDUCE_HEADER
+#undef __THRUST_DEVICE_SYSTEM_REDUCE_HEADER
+
diff --git a/compat/thrust/system/detail/adl/reduce_by_key.h b/compat/thrust/system/detail/adl/reduce_by_key.h
new file mode 100644
index 0000000000..eac65b72d2
--- /dev/null
+++ b/compat/thrust/system/detail/adl/reduce_by_key.h
@@ -0,0 +1,32 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// the purpose of this header is to #include the reduce_by_key.h header
+// of the host and device systems. It should be #included in any
+// code which uses adl to dispatch reduce_by_key
+
+#define __THRUST_HOST_SYSTEM_REDUCE_BY_KEY_HEADER <__THRUST_HOST_SYSTEM_ROOT/detail/reduce_by_key.h>
+#include __THRUST_HOST_SYSTEM_REDUCE_BY_KEY_HEADER
+#undef __THRUST_HOST_SYSTEM_REDUCE_BY_KEY_HEADER
+
+#define __THRUST_DEVICE_SYSTEM_REDUCE_BY_KEY_HEADER <__THRUST_DEVICE_SYSTEM_ROOT/detail/reduce_by_key.h>
+#include __THRUST_DEVICE_SYSTEM_REDUCE_BY_KEY_HEADER
+#undef __THRUST_DEVICE_SYSTEM_REDUCE_BY_KEY_HEADER
+
diff --git a/compat/thrust/system/detail/adl/remove.h b/compat/thrust/system/detail/adl/remove.h
new file mode 100644
index 0000000000..9d64be8da3
--- /dev/null
+++ b/compat/thrust/system/detail/adl/remove.h
@@ -0,0 +1,32 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a fill of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// the purpose of this header is to #include the remove.h header
+// of the host and device systems. It should be #included in any
+// code which uses adl to dispatch remove
+
+#define __THRUST_HOST_SYSTEM_REMOVE_HEADER <__THRUST_HOST_SYSTEM_ROOT/detail/remove.h>
+#include __THRUST_HOST_SYSTEM_REMOVE_HEADER
+#undef __THRUST_HOST_SYSTEM_REMOVE_HEADER
+
+#define __THRUST_DEVICE_SYSTEM_REMOVE_HEADER <__THRUST_DEVICE_SYSTEM_ROOT/detail/remove.h>
+#include __THRUST_DEVICE_SYSTEM_REMOVE_HEADER
+#undef __THRUST_DEVICE_SYSTEM_REMOVE_HEADER
+
diff --git a/compat/thrust/system/detail/adl/replace.h b/compat/thrust/system/detail/adl/replace.h
new file mode 100644
index 0000000000..e4d8bd22ac
--- /dev/null
+++ b/compat/thrust/system/detail/adl/replace.h
@@ -0,0 +1,32 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a fill of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// the purpose of this header is to #include the replace.h header
+// of the host and device systems. It should be #included in any
+// code which uses adl to dispatch replace
+
+#define __THRUST_HOST_SYSTEM_REPLACE_HEADER <__THRUST_HOST_SYSTEM_ROOT/detail/replace.h>
+#include __THRUST_HOST_SYSTEM_REPLACE_HEADER
+#undef __THRUST_HOST_SYSTEM_REPLACE_HEADER
+
+#define __THRUST_DEVICE_SYSTEM_REPLACE_HEADER <__THRUST_DEVICE_SYSTEM_ROOT/detail/replace.h>
+#include __THRUST_DEVICE_SYSTEM_REPLACE_HEADER
+#undef __THRUST_DEVICE_SYSTEM_REPLACE_HEADER
+
diff --git a/compat/thrust/system/detail/adl/reverse.h b/compat/thrust/system/detail/adl/reverse.h
new file mode 100644
index 0000000000..8cbcfd833c
--- /dev/null
+++ b/compat/thrust/system/detail/adl/reverse.h
@@ -0,0 +1,32 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a fill of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// the purpose of this header is to #include the reverse.h header
+// of the host and device systems. It should be #included in any
+// code which uses adl to dispatch reverse
+
+#define __THRUST_HOST_SYSTEM_REVERSE_HEADER <__THRUST_HOST_SYSTEM_ROOT/detail/reverse.h>
+#include __THRUST_HOST_SYSTEM_REVERSE_HEADER
+#undef __THRUST_HOST_SYSTEM_REVERSE_HEADER
+
+#define __THRUST_DEVICE_SYSTEM_REVERSE_HEADER <__THRUST_DEVICE_SYSTEM_ROOT/detail/reverse.h>
+#include __THRUST_DEVICE_SYSTEM_REVERSE_HEADER
+#undef __THRUST_DEVICE_SYSTEM_REVERSE_HEADER
+
diff --git a/compat/thrust/system/detail/adl/scan.h b/compat/thrust/system/detail/adl/scan.h
new file mode 100644
index 0000000000..e70cd9fdda
--- /dev/null
+++ b/compat/thrust/system/detail/adl/scan.h
@@ -0,0 +1,32 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// the purpose of this header is to #include the scan.h header
+// of the host and device systems. It should be #included in any
+// code which uses adl to dispatch scan
+
+#define __THRUST_HOST_SYSTEM_SCAN_HEADER <__THRUST_HOST_SYSTEM_ROOT/detail/scan.h>
+#include __THRUST_HOST_SYSTEM_SCAN_HEADER
+#undef __THRUST_HOST_SYSTEM_SCAN_HEADER
+
+#define __THRUST_DEVICE_SYSTEM_SCAN_HEADER <__THRUST_DEVICE_SYSTEM_ROOT/detail/scan.h>
+#include __THRUST_DEVICE_SYSTEM_SCAN_HEADER
+#undef __THRUST_DEVICE_SYSTEM_SCAN_HEADER
+
diff --git a/compat/thrust/system/detail/adl/scan_by_key.h b/compat/thrust/system/detail/adl/scan_by_key.h
new file mode 100644
index 0000000000..02c4b84751
--- /dev/null
+++ b/compat/thrust/system/detail/adl/scan_by_key.h
@@ -0,0 +1,32 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// the purpose of this header is to #include the scan_by_key.h header
+// of the host and device systems. It should be #included in any
+// code which uses adl to dispatch scan_by_key
+
+#define __THRUST_HOST_SYSTEM_SCAN_BY_KEY_HEADER <__THRUST_HOST_SYSTEM_ROOT/detail/scan_by_key.h>
+#include __THRUST_HOST_SYSTEM_SCAN_BY_KEY_HEADER
+#undef __THRUST_HOST_SYSTEM_SCAN_BY_KEY_HEADER
+
+#define __THRUST_DEVICE_SYSTEM_SCAN_BY_KEY_HEADER <__THRUST_DEVICE_SYSTEM_ROOT/detail/scan_by_key.h>
+#include __THRUST_DEVICE_SYSTEM_SCAN_BY_KEY_HEADER
+#undef __THRUST_DEVICE_SYSTEM_SCAN_BY_KEY_HEADER
+
diff --git a/compat/thrust/system/detail/adl/scatter.h b/compat/thrust/system/detail/adl/scatter.h
new file mode 100644
index 0000000000..b94b0d9892
--- /dev/null
+++ b/compat/thrust/system/detail/adl/scatter.h
@@ -0,0 +1,32 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a fill of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// the purpose of this header is to #include the scatter.h header
+// of the host and device systems. It should be #included in any
+// code which uses adl to dispatch scatter
+
+#define __THRUST_HOST_SYSTEM_SCATTER_HEADER <__THRUST_HOST_SYSTEM_ROOT/detail/scatter.h>
+#include __THRUST_HOST_SYSTEM_SCATTER_HEADER
+#undef __THRUST_HOST_SYSTEM_SCATTER_HEADER
+
+#define __THRUST_DEVICE_SYSTEM_SCATTER_HEADER <__THRUST_DEVICE_SYSTEM_ROOT/detail/scatter.h>
+#include __THRUST_DEVICE_SYSTEM_SCATTER_HEADER
+#undef __THRUST_DEVICE_SYSTEM_SCATTER_HEADER
+
diff --git a/compat/thrust/system/detail/adl/sequence.h b/compat/thrust/system/detail/adl/sequence.h
new file mode 100644
index 0000000000..07dcc7b7c1
--- /dev/null
+++ b/compat/thrust/system/detail/adl/sequence.h
@@ -0,0 +1,32 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a fill of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// the purpose of this header is to #include the sequence.h header
+// of the host and device systems. It should be #included in any
+// code which uses adl to dispatch sequence
+
+#define __THRUST_HOST_SYSTEM_SEQUENCE_HEADER <__THRUST_HOST_SYSTEM_ROOT/detail/sequence.h>
+#include __THRUST_HOST_SYSTEM_SEQUENCE_HEADER
+#undef __THRUST_HOST_SYSTEM_SEQUENCE_HEADER
+
+#define __THRUST_DEVICE_SYSTEM_SEQUENCE_HEADER <__THRUST_DEVICE_SYSTEM_ROOT/detail/sequence.h>
+#include __THRUST_DEVICE_SYSTEM_SEQUENCE_HEADER
+#undef __THRUST_DEVICE_SYSTEM_SEQUENCE_HEADER
+
diff --git a/compat/thrust/system/detail/adl/set_operations.h b/compat/thrust/system/detail/adl/set_operations.h
new file mode 100644
index 0000000000..9901b46851
--- /dev/null
+++ b/compat/thrust/system/detail/adl/set_operations.h
@@ -0,0 +1,32 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a fill of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// the purpose of this header is to #include the set_operations.h header
+// of the host and device systems. It should be #included in any
+// code which uses adl to dispatch set_operations
+
+#define __THRUST_HOST_SYSTEM_SET_OPERATIONS_HEADER <__THRUST_HOST_SYSTEM_ROOT/detail/set_operations.h>
+#include __THRUST_HOST_SYSTEM_SET_OPERATIONS_HEADER
+#undef __THRUST_HOST_SYSTEM_SET_OPERATIONS_HEADER
+
+#define __THRUST_DEVICE_SYSTEM_SET_OPERATIONS_HEADER <__THRUST_DEVICE_SYSTEM_ROOT/detail/set_operations.h>
+#include __THRUST_DEVICE_SYSTEM_SET_OPERATIONS_HEADER
+#undef __THRUST_DEVICE_SYSTEM_SET_OPERATIONS_HEADER
+
diff --git a/compat/thrust/system/detail/adl/sort.h b/compat/thrust/system/detail/adl/sort.h
new file mode 100644
index 0000000000..afcb903f87
--- /dev/null
+++ b/compat/thrust/system/detail/adl/sort.h
@@ -0,0 +1,32 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// the purpose of this header is to #include the sort.h header
+// of the host and device systems. It should be #included in any
+// code which uses adl to dispatch sort
+
+#define __THRUST_HOST_SYSTEM_SORT_HEADER <__THRUST_HOST_SYSTEM_ROOT/detail/sort.h>
+#include __THRUST_HOST_SYSTEM_SORT_HEADER
+#undef __THRUST_HOST_SYSTEM_SORT_HEADER
+
+#define __THRUST_DEVICE_SYSTEM_SORT_HEADER <__THRUST_DEVICE_SYSTEM_ROOT/detail/sort.h>
+#include __THRUST_DEVICE_SYSTEM_SORT_HEADER
+#undef __THRUST_DEVICE_SYSTEM_SORT_HEADER
+
diff --git a/compat/thrust/system/detail/adl/swap_ranges.h b/compat/thrust/system/detail/adl/swap_ranges.h
new file mode 100644
index 0000000000..c0069369e0
--- /dev/null
+++ b/compat/thrust/system/detail/adl/swap_ranges.h
@@ -0,0 +1,32 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a fill of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// the purpose of this header is to #include the swap_ranges.h header
+// of the host and device systems. It should be #included in any
+// code which uses adl to dispatch swap_ranges
+
+#define __THRUST_HOST_SYSTEM_SWAP_RANGES_HEADER <__THRUST_HOST_SYSTEM_ROOT/detail/swap_ranges.h>
+#include __THRUST_HOST_SYSTEM_SWAP_RANGES_HEADER
+#undef __THRUST_HOST_SYSTEM_SWAP_RANGES_HEADER
+
+#define __THRUST_DEVICE_SYSTEM_SWAP_RANGES_HEADER <__THRUST_DEVICE_SYSTEM_ROOT/detail/swap_ranges.h>
+#include __THRUST_DEVICE_SYSTEM_SWAP_RANGES_HEADER
+#undef __THRUST_DEVICE_SYSTEM_SWAP_RANGES_HEADER
+
diff --git a/compat/thrust/system/detail/adl/tabulate.h b/compat/thrust/system/detail/adl/tabulate.h
new file mode 100644
index 0000000000..cb1fdebd11
--- /dev/null
+++ b/compat/thrust/system/detail/adl/tabulate.h
@@ -0,0 +1,32 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a fill of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// the purpose of this header is to #include the tabulate.h header
+// of the host and device systems. It should be #included in any
+// code which uses adl to dispatch tabulate
+
+#define __THRUST_HOST_SYSTEM_TABULATE_HEADER <__THRUST_HOST_SYSTEM_ROOT/detail/tabulate.h>
+#include __THRUST_HOST_SYSTEM_TABULATE_HEADER
+#undef __THRUST_HOST_SYSTEM_TABULATE_HEADER
+
+#define __THRUST_DEVICE_SYSTEM_TABULATE_HEADER <__THRUST_DEVICE_SYSTEM_ROOT/detail/tabulate.h>
+#include __THRUST_DEVICE_SYSTEM_TABULATE_HEADER
+#undef __THRUST_DEVICE_SYSTEM_TABULATE_HEADER
+
diff --git a/compat/thrust/system/detail/adl/temporary_buffer.h b/compat/thrust/system/detail/adl/temporary_buffer.h
new file mode 100644
index 0000000000..66df0ea85b
--- /dev/null
+++ b/compat/thrust/system/detail/adl/temporary_buffer.h
@@ -0,0 +1,32 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// the purpose of this header is to #include the temporary_buffer.h header
+// of the host and device systems. It should be #included in any
+// code which uses adl to dispatch get_temporary_buffer or return_temporary_buffer
+
+#define __THRUST_HOST_SYSTEM_TEMPORARY_BUFFER_HEADER <__THRUST_HOST_SYSTEM_ROOT/detail/temporary_buffer.h>
+#include __THRUST_HOST_SYSTEM_TEMPORARY_BUFFER_HEADER
+#undef __THRUST_HOST_SYSTEM_TEMPORARY_BUFFER_HEADER
+
+#define __THRUST_DEVICE_SYSTEM_TEMPORARY_BUFFER_HEADER <__THRUST_DEVICE_SYSTEM_ROOT/detail/temporary_buffer.h>
+#include __THRUST_DEVICE_SYSTEM_TEMPORARY_BUFFER_HEADER
+#undef __THRUST_DEVICE_SYSTEM_TEMPORARY_BUFFER_HEADER
+
diff --git a/compat/thrust/system/detail/adl/transform.h b/compat/thrust/system/detail/adl/transform.h
new file mode 100644
index 0000000000..c9e6a01ea1
--- /dev/null
+++ b/compat/thrust/system/detail/adl/transform.h
@@ -0,0 +1,32 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a fill of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// the purpose of this header is to #include the transform.h header
+// of the host and device systems. It should be #included in any
+// code which uses adl to dispatch transform
+
+#define __THRUST_HOST_SYSTEM_TRANSFORM_HEADER <__THRUST_HOST_SYSTEM_ROOT/detail/transform.h>
+#include __THRUST_HOST_SYSTEM_TRANSFORM_HEADER
+#undef __THRUST_HOST_SYSTEM_TRANSFORM_HEADER
+
+#define __THRUST_DEVICE_SYSTEM_TRANSFORM_HEADER <__THRUST_DEVICE_SYSTEM_ROOT/detail/transform.h>
+#include __THRUST_DEVICE_SYSTEM_TRANSFORM_HEADER
+#undef __THRUST_DEVICE_SYSTEM_TRANSFORM_HEADER
+
diff --git a/compat/thrust/system/detail/adl/transform_reduce.h b/compat/thrust/system/detail/adl/transform_reduce.h
new file mode 100644
index 0000000000..0a5d97749f
--- /dev/null
+++ b/compat/thrust/system/detail/adl/transform_reduce.h
@@ -0,0 +1,32 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a fill of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// the purpose of this header is to #include the transform_reduce.h header
+// of the host and device systems. It should be #included in any
+// code which uses adl to dispatch transform_reduce
+
+#define __THRUST_HOST_SYSTEM_TRANSFORM_REDUCE_HEADER <__THRUST_HOST_SYSTEM_ROOT/detail/transform_reduce.h>
+#include __THRUST_HOST_SYSTEM_TRANSFORM_REDUCE_HEADER
+#undef __THRUST_HOST_SYSTEM_TRANSFORM_REDUCE_HEADER
+
+#define __THRUST_DEVICE_SYSTEM_TRANSFORM_REDUCE_HEADER <__THRUST_DEVICE_SYSTEM_ROOT/detail/transform_reduce.h>
+#include __THRUST_DEVICE_SYSTEM_TRANSFORM_REDUCE_HEADER
+#undef __THRUST_DEVICE_SYSTEM_TRANSFORM_REDUCE_HEADER
+
diff --git a/compat/thrust/system/detail/adl/transform_scan.h b/compat/thrust/system/detail/adl/transform_scan.h
new file mode 100644
index 0000000000..47c1dc3ae9
--- /dev/null
+++ b/compat/thrust/system/detail/adl/transform_scan.h
@@ -0,0 +1,32 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a fill of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// the purpose of this header is to #include the transform_scan.h header
+// of the host and device systems. It should be #included in any
+// code which uses adl to dispatch transform_scan
+
+#define __THRUST_HOST_SYSTEM_TRANSFORM_SCAN_HEADER <__THRUST_HOST_SYSTEM_ROOT/detail/transform_scan.h>
+#include __THRUST_HOST_SYSTEM_TRANSFORM_SCAN_HEADER
+#undef __THRUST_HOST_SYSTEM_TRANSFORM_SCAN_HEADER
+
+#define __THRUST_DEVICE_SYSTEM_TRANSFORM_SCAN_HEADER <__THRUST_DEVICE_SYSTEM_ROOT/detail/transform_scan.h>
+#include __THRUST_DEVICE_SYSTEM_TRANSFORM_SCAN_HEADER
+#undef __THRUST_DEVICE_SYSTEM_TRANSFORM_SCAN_HEADER
+
diff --git a/compat/thrust/system/detail/adl/uninitialized_copy.h b/compat/thrust/system/detail/adl/uninitialized_copy.h
new file mode 100644
index 0000000000..7cb0b8e401
--- /dev/null
+++ b/compat/thrust/system/detail/adl/uninitialized_copy.h
@@ -0,0 +1,32 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a fill of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// the purpose of this header is to #include the uninitialized_copy.h header
+// of the host and device systems. It should be #included in any
+// code which uses adl to dispatch uninitialized_copy
+
+#define __THRUST_HOST_SYSTEM_UNINITIALIZED_COPY_HEADER <__THRUST_HOST_SYSTEM_ROOT/detail/uninitialized_copy.h>
+#include __THRUST_HOST_SYSTEM_UNINITIALIZED_COPY_HEADER
+#undef __THRUST_HOST_SYSTEM_UNINITIALIZED_COPY_HEADER
+
+#define __THRUST_DEVICE_SYSTEM_UNINITIALIZED_COPY_HEADER <__THRUST_DEVICE_SYSTEM_ROOT/detail/uninitialized_copy.h>
+#include __THRUST_DEVICE_SYSTEM_UNINITIALIZED_COPY_HEADER
+#undef __THRUST_DEVICE_SYSTEM_UNINITIALIZED_COPY_HEADER
+
diff --git a/compat/thrust/system/detail/adl/uninitialized_fill.h b/compat/thrust/system/detail/adl/uninitialized_fill.h
new file mode 100644
index 0000000000..9f00b51622
--- /dev/null
+++ b/compat/thrust/system/detail/adl/uninitialized_fill.h
@@ -0,0 +1,32 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a fill of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// the purpose of this header is to #include the uninitialized_fill.h header
+// of the host and device systems. It should be #included in any
+// code which uses adl to dispatch uninitialized_fill
+
+#define __THRUST_HOST_SYSTEM_UNINITIALIZED_FILL_HEADER <__THRUST_HOST_SYSTEM_ROOT/detail/uninitialized_fill.h>
+#include __THRUST_HOST_SYSTEM_UNINITIALIZED_FILL_HEADER
+#undef __THRUST_HOST_SYSTEM_UNINITIALIZED_FILL_HEADER
+
+#define __THRUST_DEVICE_SYSTEM_UNINITIALIZED_FILL_HEADER <__THRUST_DEVICE_SYSTEM_ROOT/detail/uninitialized_fill.h>
+#include __THRUST_DEVICE_SYSTEM_UNINITIALIZED_FILL_HEADER
+#undef __THRUST_DEVICE_SYSTEM_UNINITIALIZED_FILL_HEADER
+
diff --git a/compat/thrust/system/detail/adl/unique.h b/compat/thrust/system/detail/adl/unique.h
new file mode 100644
index 0000000000..932ff58e07
--- /dev/null
+++ b/compat/thrust/system/detail/adl/unique.h
@@ -0,0 +1,32 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a fill of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// the purpose of this header is to #include the unique.h header
+// of the host and device systems. It should be #included in any
+// code which uses adl to dispatch unique
+
+#define __THRUST_HOST_SYSTEM_UNIQUE_HEADER <__THRUST_HOST_SYSTEM_ROOT/detail/unique.h>
+#include __THRUST_HOST_SYSTEM_UNIQUE_HEADER
+#undef __THRUST_HOST_SYSTEM_UNIQUE_HEADER
+
+#define __THRUST_DEVICE_SYSTEM_UNIQUE_HEADER <__THRUST_DEVICE_SYSTEM_ROOT/detail/unique.h>
+#include __THRUST_DEVICE_SYSTEM_UNIQUE_HEADER
+#undef __THRUST_DEVICE_SYSTEM_UNIQUE_HEADER
+
diff --git a/compat/thrust/system/detail/adl/unique_by_key.h b/compat/thrust/system/detail/adl/unique_by_key.h
new file mode 100644
index 0000000000..30e6f2f2d1
--- /dev/null
+++ b/compat/thrust/system/detail/adl/unique_by_key.h
@@ -0,0 +1,32 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a fill of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// the purpose of this header is to #include the unique_by_key.h header
+// of the host and device systems. It should be #included in any
+// code which uses adl to dispatch unique_by_key
+
+#define __THRUST_HOST_SYSTEM_UNIQUE_BY_KEY_HEADER <__THRUST_HOST_SYSTEM_ROOT/detail/unique_by_key.h>
+#include __THRUST_HOST_SYSTEM_UNIQUE_BY_KEY_HEADER
+#undef __THRUST_HOST_SYSTEM_UNIQUE_BY_KEY_HEADER
+
+#define __THRUST_DEVICE_SYSTEM_UNIQUE_BY_KEY_HEADER <__THRUST_DEVICE_SYSTEM_ROOT/detail/unique_by_key.h>
+#include __THRUST_DEVICE_SYSTEM_UNIQUE_BY_KEY_HEADER
+#undef __THRUST_DEVICE_SYSTEM_UNIQUE_BY_KEY_HEADER
+
diff --git a/compat/thrust/system/detail/bad_alloc.h b/compat/thrust/system/detail/bad_alloc.h
new file mode 100644
index 0000000000..bb73d1f006
--- /dev/null
+++ b/compat/thrust/system/detail/bad_alloc.h
@@ -0,0 +1,57 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+#pragma once
+
+#include <new>
+#include <string>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+
+// define our own bad_alloc so we can set its .what()
+class bad_alloc
+  : public std::bad_alloc
+{
+  public:
+    inline bad_alloc(const std::string &w)
+      : std::bad_alloc(), m_what()
+    {
+      m_what = std::bad_alloc::what();
+      m_what += ": ";
+      m_what += w;
+    } // end bad_alloc()
+
+    inline virtual ~bad_alloc(void) throw () {};
+
+    inline virtual const char *what(void) const throw()
+    {
+      return m_what.c_str();
+    } // end what()
+
+  private:
+    std::string m_what;
+}; // end bad_alloc
+  
+} // end detail
+} // end system
+} // end thrust
+
diff --git a/compat/thrust/system/detail/errno.h b/compat/thrust/system/detail/errno.h
new file mode 100644
index 0000000000..34bc8cc568
--- /dev/null
+++ b/compat/thrust/system/detail/errno.h
@@ -0,0 +1,120 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// The rationale for the existence of these apparently redundant definitions is
+// to provide them portably and to avoid bringing in system headers which might
+// pollute the global namespace. These identifiers are in lowercase to avoid
+// colliding with the real macros in errno.h.
+
+namespace thrust
+{
+
+namespace system
+{
+
+namespace detail
+{
+
+static const int eafnosupport    = 9901;
+static const int eaddrinuse      = 9902;
+static const int eaddrnotavail   = 9903;
+static const int eisconn         = 9904;
+static const int ebadmsg         = 9905;
+static const int econnaborted    = 9906;
+static const int ealready        = 9907;
+static const int econnrefused    = 9908;
+static const int econnreset      = 9909;
+static const int edestaddrreq    = 9910;
+static const int ehostunreach    = 9911;
+static const int eidrm           = 9912;
+static const int emsgsize        = 9913;
+static const int enetdown        = 9914;
+static const int enetreset       = 9915;
+static const int enetunreach     = 9916;
+static const int enobufs         = 9917;
+static const int enolink         = 9918;
+static const int enodata         = 9919;
+static const int enomsg          = 9920;
+static const int enoprotoopt     = 9921;
+static const int enosr           = 9922;
+static const int enotsock        = 9923;
+static const int enostr          = 9924;
+static const int enotconn        = 9925;
+static const int enotsup         = 9926;
+static const int ecanceled       = 9927;
+static const int einprogress     = 9928;
+static const int eopnotsupp      = 9929;
+static const int ewouldblock     = 9930;
+static const int eownerdead      = 9931;
+static const int eproto          = 9932;
+static const int eprotonosupport = 9933;
+static const int enotrecoverable = 9934;
+static const int etime           = 9935;
+static const int etxtbsy         = 9936;
+static const int etimedout       = 9938;
+static const int eloop           = 9939;
+static const int eoverflow       = 9940;
+static const int eprototype      = 9941;
+static const int enosys          = 9942;
+static const int einval          = 9943;
+static const int erange          = 9944;
+static const int eilseq          = 9945;
+static const int e2big           = 9946;
+static const int edom            = 9947;
+static const int efault          = 9948;
+static const int ebadf           = 9949;
+static const int epipe           = 9950;
+static const int exdev           = 9951;
+static const int ebusy           = 9952;
+static const int enotempty       = 9953;
+static const int enoexec         = 9954;
+static const int eexist          = 9955;
+static const int efbig           = 9956;
+static const int enametoolong    = 9957;
+static const int enotty          = 9958;
+static const int eintr           = 9959;
+static const int espipe          = 9960;
+static const int eio             = 9961;
+static const int eisdir          = 9962;
+static const int echild          = 9963;
+static const int enolck          = 9964;
+static const int enospc          = 9965;
+static const int enxio           = 9966;
+static const int enodev          = 9967;
+static const int enoent          = 9968;
+static const int esrch           = 9969;
+static const int enotdir         = 9970;
+static const int enomem          = 9971;
+static const int eperm           = 9972;
+static const int eacces          = 9973;
+static const int erofs           = 9974;
+static const int edeadlk         = 9975;
+static const int eagain          = 9976;
+static const int enfile          = 9977;
+static const int emfile          = 9978;
+static const int emlink          = 9979;
+
+} // end detail
+
+} // end system
+
+} // end thrust
+
diff --git a/compat/thrust/system/detail/error_category.inl b/compat/thrust/system/detail/error_category.inl
new file mode 100644
index 0000000000..8e19c89db5
--- /dev/null
+++ b/compat/thrust/system/detail/error_category.inl
@@ -0,0 +1,234 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+#pragma once
+
+#include <thrust/system/error_code.h>
+#include <thrust/system/detail/errno.h>
+#include <thrust/functional.h>
+#include <cstring>
+
+namespace thrust
+{
+
+namespace system
+{
+
+error_category
+  ::~error_category(void)
+{
+  ;
+} // end error_category::~error_category()
+
+
+error_condition error_category
+  ::default_error_condition(int ev) const
+{
+  return error_condition(ev, *this);
+} // end error_category::default_error_condition()
+
+
+bool error_category
+  ::equivalent(int code, const error_condition &condition) const
+{
+  return default_error_condition(code) == condition;
+} // end error_condition::equivalent()
+
+
+bool error_category
+  ::equivalent(const error_code &code, int condition) const
+{
+  bool result = (this->operator==(code.category())) && (code.value() == condition);
+  return result;
+} // end error_code::equivalent()
+
+
+bool error_category
+  ::operator==(const error_category &rhs) const
+{
+  return this == &rhs;
+} // end error_category::operator==()
+
+
+bool error_category
+  ::operator!=(const error_category &rhs) const
+{
+  return !this->operator==(rhs);
+} // end error_category::operator!=()
+
+
+bool error_category
+  ::operator<(const error_category &rhs) const
+{
+  return thrust::less<const error_category*>()(this,&rhs);
+} // end error_category::operator<()
+
+
+namespace detail
+{
+
+
+class generic_error_category
+  : public error_category
+{
+  public:
+    inline generic_error_category(void) {}
+
+    inline virtual const char *name(void) const
+    {
+      return "generic";
+    }
+
+    inline virtual std::string message(int ev) const
+    {
+      static const std::string unknown_err("Unknown error");
+
+      // XXX strerror is not thread-safe:
+      //     prefer strerror_r (which is not provided on windows)
+      const char *c_str = std::strerror(ev);
+      return c_str ? std::string(c_str) : unknown_err;
+    }
+}; // end generic_category_result
+
+
+class system_error_category
+  : public error_category
+{
+  public:
+    inline system_error_category(void) {}
+
+    inline virtual const char *name(void) const
+    {
+      return "system";
+    }
+
+    inline virtual std::string message(int ev) const
+    {
+      return generic_category().message(ev);
+    }
+
+    inline virtual error_condition default_error_condition(int ev) const
+    {
+      using namespace errc;
+
+      switch(ev)
+      {
+        case eafnosupport:    return make_error_condition(address_family_not_supported);
+        case eaddrinuse:      return make_error_condition(address_in_use);
+        case eaddrnotavail:   return make_error_condition(address_not_available);
+        case eisconn:         return make_error_condition(already_connected);
+        case e2big:           return make_error_condition(argument_list_too_long);
+        case edom:            return make_error_condition(argument_out_of_domain);
+        case efault:          return make_error_condition(bad_address);
+        case ebadf:           return make_error_condition(bad_file_descriptor);
+        case ebadmsg:         return make_error_condition(bad_message);
+        case epipe:           return make_error_condition(broken_pipe);
+        case econnaborted:    return make_error_condition(connection_aborted);
+        case ealready:        return make_error_condition(connection_already_in_progress);
+        case econnrefused:    return make_error_condition(connection_refused);
+        case econnreset:      return make_error_condition(connection_reset);
+        case exdev:           return make_error_condition(cross_device_link);
+        case edestaddrreq:    return make_error_condition(destination_address_required);
+        case ebusy:           return make_error_condition(device_or_resource_busy);
+        case enotempty:       return make_error_condition(directory_not_empty);
+        case enoexec:         return make_error_condition(executable_format_error);
+        case eexist:          return make_error_condition(file_exists);
+        case efbig:           return make_error_condition(file_too_large);
+        case enametoolong:    return make_error_condition(filename_too_long);
+        case enosys:          return make_error_condition(function_not_supported);
+        case ehostunreach:    return make_error_condition(host_unreachable);
+        case eidrm:           return make_error_condition(identifier_removed);
+        case eilseq:          return make_error_condition(illegal_byte_sequence);
+        case enotty:          return make_error_condition(inappropriate_io_control_operation);
+        case eintr:           return make_error_condition(interrupted);
+        case einval:          return make_error_condition(invalid_argument);
+        case espipe:          return make_error_condition(invalid_seek);
+        case eio:             return make_error_condition(io_error);
+        case eisdir:          return make_error_condition(is_a_directory);
+        case emsgsize:        return make_error_condition(message_size);
+        case enetdown:        return make_error_condition(network_down);
+        case enetreset:       return make_error_condition(network_reset);
+        case enetunreach:     return make_error_condition(network_unreachable);
+        case enobufs:         return make_error_condition(no_buffer_space);
+        case echild:          return make_error_condition(no_child_process);
+        case enolink:         return make_error_condition(no_link);
+        case enolck:          return make_error_condition(no_lock_available);
+        case enodata:         return make_error_condition(no_message_available);
+        case enomsg:          return make_error_condition(no_message);
+        case enoprotoopt:     return make_error_condition(no_protocol_option);
+        case enospc:          return make_error_condition(no_space_on_device);
+        case enosr:           return make_error_condition(no_stream_resources);
+        case enxio:           return make_error_condition(no_such_device_or_address);
+        case enodev:          return make_error_condition(no_such_device);
+        case enoent:          return make_error_condition(no_such_file_or_directory);
+        case esrch:           return make_error_condition(no_such_process);
+        case enotdir:         return make_error_condition(not_a_directory);
+        case enotsock:        return make_error_condition(not_a_socket);
+        case enostr:          return make_error_condition(not_a_stream);
+        case enotconn:        return make_error_condition(not_connected);
+        case enomem:          return make_error_condition(not_enough_memory);
+        case enotsup:         return make_error_condition(not_supported);
+        case ecanceled:       return make_error_condition(operation_canceled);
+        case einprogress:     return make_error_condition(operation_in_progress);
+        case eperm:           return make_error_condition(operation_not_permitted);
+        case eopnotsupp:      return make_error_condition(operation_not_supported);
+        case ewouldblock:     return make_error_condition(operation_would_block);
+        case eownerdead:      return make_error_condition(owner_dead);
+        case eacces:          return make_error_condition(permission_denied);
+        case eproto:          return make_error_condition(protocol_error);
+        case eprotonosupport: return make_error_condition(protocol_not_supported);
+        case erofs:           return make_error_condition(read_only_file_system);
+        case edeadlk:         return make_error_condition(resource_deadlock_would_occur);
+        case eagain:          return make_error_condition(resource_unavailable_try_again);
+        case erange:          return make_error_condition(result_out_of_range);
+        case enotrecoverable: return make_error_condition(state_not_recoverable);
+        case etime:           return make_error_condition(stream_timeout);
+        case etxtbsy:         return make_error_condition(text_file_busy);
+        case etimedout:       return make_error_condition(timed_out);
+        case enfile:          return make_error_condition(too_many_files_open_in_system);
+        case emfile:          return make_error_condition(too_many_files_open);
+        case emlink:          return make_error_condition(too_many_links);
+        case eloop:           return make_error_condition(too_many_symbolic_link_levels);
+        case eoverflow:       return make_error_condition(value_too_large);
+        case eprototype:      return make_error_condition(wrong_protocol_type);
+        default:              return error_condition(ev,system_category());
+      }
+    }
+}; // end system_category_result
+
+
+} // end detail
+
+
+const error_category &generic_category(void)
+{
+  static const detail::generic_error_category result;
+  return result;
+}
+
+
+const error_category &system_category(void)
+{
+  static const detail::system_error_category result;
+  return result;
+}
+
+
+} // end system
+
+} // end thrust
+
diff --git a/compat/thrust/system/detail/error_code.inl b/compat/thrust/system/detail/error_code.inl
new file mode 100644
index 0000000000..0cf86b4821
--- /dev/null
+++ b/compat/thrust/system/detail/error_code.inl
@@ -0,0 +1,197 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+#pragma once
+
+#include <thrust/system/error_code.h>
+
+namespace thrust
+{
+
+namespace system
+{
+
+error_code
+  ::error_code(void)
+    :m_val(0),m_cat(&system_category())
+{
+  ;
+} // end error_code::error_code()
+
+
+error_code
+  ::error_code(int val, const error_category &cat)
+    :m_val(val),m_cat(&cat)
+{
+  ;
+} // end error_code::error_code()
+
+
+template <typename ErrorCodeEnum>
+  error_code
+    ::error_code(ErrorCodeEnum e
+// XXX WAR msvc's problem with enable_if
+#if THRUST_HOST_COMPILER != THRUST_HOST_COMPILER_MSVC
+                 , typename thrust::detail::enable_if<is_error_code_enum<ErrorCodeEnum>::value>::type *
+#endif // THRUST_HOST_COMPILER_MSVC
+                )
+{
+  *this = make_error_code(e);
+} // end error_code::error_code()
+
+
+void error_code
+  ::assign(int val, const error_category &cat)
+{
+  m_val = val;
+  m_cat = &cat;
+} // end error_code::assign()
+
+
+template <typename ErrorCodeEnum>
+// XXX WAR msvc's problem with enable_if
+#if THRUST_HOST_COMPILER != THRUST_HOST_COMPILER_MSVC
+  typename thrust::detail::enable_if<is_error_code_enum<ErrorCodeEnum>::value, error_code>::type &
+#else
+  error_code &
+#endif // THRUST_HOST_COMPILER_MSVC
+    error_code
+      ::operator=(ErrorCodeEnum e)
+{
+  *this = make_error_code(e);
+  return *this;
+} // end error_code::operator=()
+
+
+void error_code
+  ::clear(void)
+{
+  m_val = 0;
+  m_cat = &system_category();
+} // end error_code::clear()
+
+
+int error_code
+  ::value(void) const
+{
+  return m_val;
+} // end error_code::value()
+
+
+const error_category &error_code
+  ::category(void) const
+{
+  return *m_cat;
+} // end error_code::category()
+
+
+error_condition error_code
+  ::default_error_condition(void) const
+{
+  return category().default_error_condition(value());
+} // end error_code::default_error_condition()
+
+
+std::string error_code
+  ::message(void) const
+{
+  return category().message(value());
+} // end error_code::message()
+
+
+error_code
+  ::operator bool (void) const
+{
+  return value() != 0;
+} // end error_code::operator bool ()
+
+
+error_code make_error_code(errc::errc_t e)
+{
+  return error_code(static_cast<int>(e), generic_category());
+} // end make_error_code()
+
+
+bool operator<(const error_code &lhs, const error_code &rhs)
+{
+  bool result = lhs.category().operator<(rhs.category());
+  result = result || lhs.category().operator==(rhs.category());
+  result = result || lhs.value() < rhs.value();
+  return result;
+} // end operator==()
+
+
+template<typename charT, typename traits>
+  std::basic_ostream<charT,traits>&
+    operator<<(std::basic_ostream<charT,traits> &os, const error_code &ec)
+{
+  return os << ec.category().name() << ':' << ec.value();
+} // end operator<<()
+
+
+bool operator==(const error_code &lhs, const error_code &rhs)
+{
+  return lhs.category().operator==(rhs.category()) && lhs.value() == rhs.value();
+} // end operator==()
+
+
+bool operator==(const error_code &lhs, const error_condition &rhs)
+{
+  return lhs.category().equivalent(lhs.value(), rhs) || rhs.category().equivalent(lhs,rhs.value());
+} // end operator==()
+
+
+bool operator==(const error_condition &lhs, const error_code &rhs)
+{
+  return rhs.category().equivalent(lhs.value(), lhs) || lhs.category().equivalent(rhs, lhs.value());
+} // end operator==()
+
+
+bool operator==(const error_condition &lhs, const error_condition &rhs)
+{
+  return lhs.category().operator==(rhs.category()) && lhs.value() == rhs.value();
+} // end operator==()
+
+
+bool operator!=(const error_code &lhs, const error_code &rhs)
+{
+  return !(lhs == rhs);
+} // end operator!=()
+
+
+bool operator!=(const error_code &lhs, const error_condition &rhs)
+{
+  return !(lhs == rhs);
+} // end operator!=()
+
+
+bool operator!=(const error_condition &lhs, const error_code &rhs)
+{
+  return !(lhs == rhs);
+} // end operator!=()
+
+
+bool operator!=(const error_condition &lhs, const error_condition &rhs)
+{
+  return !(lhs == rhs);
+} // end operator!=()
+
+
+} // end system
+
+} // end thrust
+
diff --git a/compat/thrust/system/detail/error_condition.inl b/compat/thrust/system/detail/error_condition.inl
new file mode 100644
index 0000000000..00fbaf091d
--- /dev/null
+++ b/compat/thrust/system/detail/error_condition.inl
@@ -0,0 +1,133 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+#pragma once
+
+#include <thrust/system/detail/error_condition.inl>
+#include <thrust/functional.h>
+
+namespace thrust
+{
+
+namespace system
+{
+
+error_condition
+  ::error_condition(void)
+    :m_val(0),m_cat(&generic_category())
+{
+  ;
+} // end error_condition::error_condition()
+
+
+error_condition
+  ::error_condition(int val, const error_category &cat)
+    :m_val(val),m_cat(&cat)
+{
+  ;
+} // end error_condition::error_condition()
+
+
+template<typename ErrorConditionEnum>
+  error_condition
+    ::error_condition(ErrorConditionEnum e
+// XXX WAR msvc's problem with enable_if
+#if THRUST_HOST_COMPILER != THRUST_HOST_COMPILER_MSVC
+                      , typename thrust::detail::enable_if<is_error_condition_enum<ErrorConditionEnum>::value>::type *
+#endif // THRUST_HOST_COMPILER != THRUST_HOST_COMPILER_MSVC
+                     )
+{
+  *this = make_error_condition(e);
+} // end error_condition::error_condition()
+
+
+void error_condition
+  ::assign(int val, const error_category &cat)
+{
+  m_val = val;
+  m_cat = &cat;
+} // end error_category::assign()
+
+
+template<typename ErrorConditionEnum>
+// XXX WAR msvc's problem with enable_if
+#if THRUST_HOST_COMPILER != THRUST_HOST_COMPILER_MSVC
+  typename thrust::detail::enable_if<is_error_condition_enum<ErrorConditionEnum>::value, error_condition>::type &
+#else
+  error_condition &
+#endif // THRUST_HOST_COMPILER != THRUST_HOST_COMPILER_MSVC
+    error_condition
+      ::operator=(ErrorConditionEnum e)
+{
+  *this = make_error_condition(e);
+  return *this;
+} // end error_condition::operator=()
+
+
+void error_condition
+  ::clear(void)
+{
+  m_val = 0;
+  m_cat = &generic_category();
+} // end error_condition::clear()
+
+
+int error_condition
+  ::value(void) const
+{
+  return m_val;
+} // end error_condition::value()
+
+
+const error_category &error_condition
+  ::category(void) const
+{
+  return *m_cat;
+} // end error_condition::category()
+
+
+std::string error_condition
+  ::message(void) const
+{
+  return category().message(value());
+} // end error_condition::message()
+
+
+error_condition
+  ::operator bool (void) const
+{
+  return value() != 0;
+} // end error_condition::operator bool ()
+
+
+error_condition make_error_condition(errc::errc_t e)
+{
+  return error_condition(static_cast<int>(e), generic_category());
+} // end make_error_condition()
+
+
+bool operator<(const error_condition &lhs,
+               const error_condition &rhs)
+{
+  return lhs.category().operator<(rhs.category()) || (lhs.category().operator==(rhs.category()) && (lhs.value() < rhs.value()));
+} // end operator<()
+
+
+} // end system
+
+} // end thrust
+
diff --git a/compat/thrust/system/detail/generic/adjacent_difference.h b/compat/thrust/system/detail/generic/adjacent_difference.h
new file mode 100644
index 0000000000..bb340df490
--- /dev/null
+++ b/compat/thrust/system/detail/generic/adjacent_difference.h
@@ -0,0 +1,53 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file adjacent_difference.h
+ *  \brief Generic implementation of adjacent_difference.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/tag.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+template <typename DerivedPolicy, class InputIterator, class OutputIterator>
+OutputIterator adjacent_difference(thrust::execution_policy<DerivedPolicy> &exec,
+                                   InputIterator first, InputIterator last,
+                                   OutputIterator result);
+
+template <typename DerivedPolicy, class InputIterator, class OutputIterator, class BinaryFunction>
+OutputIterator adjacent_difference(thrust::execution_policy<DerivedPolicy> &exec,
+                                   InputIterator first, InputIterator last,
+                                   OutputIterator result,
+                                   BinaryFunction binary_op);
+
+} // end namespace generic
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/detail/generic/adjacent_difference.inl>
+
diff --git a/compat/thrust/system/detail/generic/adjacent_difference.inl b/compat/thrust/system/detail/generic/adjacent_difference.inl
new file mode 100644
index 0000000000..619b29f860
--- /dev/null
+++ b/compat/thrust/system/detail/generic/adjacent_difference.inl
@@ -0,0 +1,76 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/adjacent_difference.h>
+#include <thrust/adjacent_difference.h>
+#include <thrust/functional.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/detail/temporary_array.h>
+#include <thrust/transform.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+template <typename DerivedPolicy, class InputIterator, class OutputIterator>
+OutputIterator adjacent_difference(thrust::execution_policy<DerivedPolicy> &exec,
+                                   InputIterator first, InputIterator last,
+                                   OutputIterator result)
+{
+  typedef typename thrust::iterator_traits<InputIterator>::value_type InputType;
+  thrust::minus<InputType> binary_op;
+
+  return thrust::adjacent_difference(exec, first, last, result, binary_op);
+} // end adjacent_difference()
+
+template <typename DerivedPolicy, class InputIterator, class OutputIterator, class BinaryFunction>
+OutputIterator adjacent_difference(thrust::execution_policy<DerivedPolicy> &exec,
+                                   InputIterator first, InputIterator last,
+                                   OutputIterator result,
+                                   BinaryFunction binary_op)
+{
+  typedef typename thrust::iterator_traits<InputIterator>::value_type InputType;
+
+  if(first == last)
+  {
+    // empty range, nothing to do
+    return result; 
+  }
+  else 
+  {
+    // an in-place operation is requested, copy the input and call the entry point
+    // XXX a special-purpose kernel would be faster here since
+    // only block boundaries need to be copied
+    thrust::detail::temporary_array<InputType, DerivedPolicy> input_copy(exec, first, last);
+    
+    *result = *first;
+    thrust::transform(exec, input_copy.begin() + 1, input_copy.end(), input_copy.begin(), result + 1, binary_op); 
+  }
+
+  return result + (last - first);
+}
+
+} // end namespace generic
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/detail/generic/advance.h b/compat/thrust/system/detail/generic/advance.h
new file mode 100644
index 0000000000..249aac7e54
--- /dev/null
+++ b/compat/thrust/system/detail/generic/advance.h
@@ -0,0 +1,40 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+template<typename InputIterator, typename Distance>
+void advance(InputIterator& i, Distance n);
+
+} // end namespace generic
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/detail/generic/advance.inl>
+
diff --git a/compat/thrust/system/detail/generic/advance.inl b/compat/thrust/system/detail/generic/advance.inl
new file mode 100644
index 0000000000..b95737ad9e
--- /dev/null
+++ b/compat/thrust/system/detail/generic/advance.inl
@@ -0,0 +1,62 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/advance.h>
+#include <thrust/iterator/iterator_traits.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+namespace detail
+{
+
+template<typename InputIterator, typename Distance>
+void advance(InputIterator& i, Distance n, thrust::incrementable_traversal_tag)
+{
+  while(n)
+  {
+    ++i;
+    --n;
+  } // end while
+} // end advance()
+
+template<typename InputIterator, typename Distance>
+void advance(InputIterator& i, Distance n, thrust::random_access_traversal_tag)
+{
+  i += n;
+} // end advance()
+
+} // end detail
+
+template<typename InputIterator, typename Distance>
+void advance(InputIterator& i, Distance n)
+{
+  // dispatch on iterator traversal
+  thrust::system::detail::generic::detail::advance(i, n,
+    typename thrust::iterator_traversal<InputIterator>::type());
+} // end advance()
+
+} // end namespace detail
+} // end namespace generic
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/detail/generic/binary_search.h b/compat/thrust/system/detail/generic/binary_search.h
new file mode 100644
index 0000000000..7fd6c506ee
--- /dev/null
+++ b/compat/thrust/system/detail/generic/binary_search.h
@@ -0,0 +1,156 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file binary_search.h
+ *  \brief Generic implementations of binary search functions.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/tag.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+
+template <typename DerivedPolicy, typename ForwardIterator, typename T>
+ForwardIterator lower_bound(thrust::execution_policy<DerivedPolicy> &exec, 
+                            ForwardIterator begin,
+                            ForwardIterator end,
+                            const T& value);
+
+template <typename DerivedPolicy, typename ForwardIterator, typename T, typename StrictWeakOrdering>
+ForwardIterator lower_bound(thrust::execution_policy<DerivedPolicy> &exec,
+                            ForwardIterator begin,
+                            ForwardIterator end,
+                            const T& value, 
+                            StrictWeakOrdering comp);
+
+
+template <typename DerivedPolicy, typename ForwardIterator, typename T>
+ForwardIterator upper_bound(thrust::execution_policy<DerivedPolicy> &exec,
+                            ForwardIterator begin,
+                            ForwardIterator end,
+                            const T& value);
+
+template <typename DerivedPolicy, typename ForwardIterator, typename T, typename StrictWeakOrdering>
+ForwardIterator upper_bound(thrust::execution_policy<DerivedPolicy> &exec, 
+                            ForwardIterator begin,
+                            ForwardIterator end,
+                            const T& value, 
+                            StrictWeakOrdering comp);
+
+
+template <typename DerivedPolicy, typename ForwardIterator, typename T>
+bool binary_search(thrust::execution_policy<DerivedPolicy> &exec,
+                   ForwardIterator begin,
+                   ForwardIterator end,
+                   const T& value);
+
+template <typename DerivedPolicy, typename ForwardIterator, typename T, typename StrictWeakOrdering>
+bool binary_search(thrust::execution_policy<DerivedPolicy> &exec,
+                   ForwardIterator begin,
+                   ForwardIterator end,
+                   const T& value, 
+                   StrictWeakOrdering comp);
+
+
+template <typename DerivedPolicy, typename ForwardIterator, typename InputIterator, typename OutputIterator>
+OutputIterator lower_bound(thrust::execution_policy<DerivedPolicy> &exec,
+                           ForwardIterator begin, 
+                           ForwardIterator end,
+                           InputIterator values_begin, 
+                           InputIterator values_end,
+                           OutputIterator output);
+
+template <typename DerivedPolicy, typename ForwardIterator, typename InputIterator, typename OutputIterator, typename StrictWeakOrdering>
+OutputIterator lower_bound(thrust::execution_policy<DerivedPolicy> &exec,
+                           ForwardIterator begin, 
+                           ForwardIterator end,
+                           InputIterator values_begin, 
+                           InputIterator values_end,
+                           OutputIterator output,
+                           StrictWeakOrdering comp);
+
+
+template <typename DerivedPolicy, typename ForwardIterator, typename InputIterator, typename OutputIterator>
+OutputIterator upper_bound(thrust::execution_policy<DerivedPolicy> &exec,
+                           ForwardIterator begin, 
+                           ForwardIterator end,
+                           InputIterator values_begin, 
+                           InputIterator values_end,
+                           OutputIterator output);
+
+template <typename DerivedPolicy, typename ForwardIterator, typename InputIterator, typename OutputIterator, typename StrictWeakOrdering>
+OutputIterator upper_bound(thrust::execution_policy<DerivedPolicy> &exec,
+                           ForwardIterator begin, 
+                           ForwardIterator end,
+                           InputIterator values_begin, 
+                           InputIterator values_end,
+                           OutputIterator output,
+                           StrictWeakOrdering comp);
+
+
+template <typename DerivedPolicy, typename ForwardIterator, typename InputIterator, typename OutputIterator>
+OutputIterator binary_search(thrust::execution_policy<DerivedPolicy> &exec,
+                             ForwardIterator begin, 
+                             ForwardIterator end,
+                             InputIterator values_begin, 
+                             InputIterator values_end,
+                             OutputIterator output);
+
+template <typename DerivedPolicy, typename ForwardIterator, typename InputIterator, typename OutputIterator, typename StrictWeakOrdering>
+OutputIterator binary_search(thrust::execution_policy<DerivedPolicy> &exec,
+                             ForwardIterator begin, 
+                             ForwardIterator end,
+                             InputIterator values_begin, 
+                             InputIterator values_end,
+                             OutputIterator output,
+                             StrictWeakOrdering comp);
+
+
+template <typename DerivedPolicy, typename ForwardIterator, typename LessThanComparable>
+thrust::pair<ForwardIterator,ForwardIterator>
+equal_range(thrust::execution_policy<DerivedPolicy> &exec,
+            ForwardIterator first,
+            ForwardIterator last,
+            const LessThanComparable &value);
+
+template <typename DerivedPolicy, typename ForwardIterator, typename LessThanComparable, typename StrictWeakOrdering>
+thrust::pair<ForwardIterator,ForwardIterator>
+equal_range(thrust::execution_policy<DerivedPolicy> &exec,
+            ForwardIterator first,
+            ForwardIterator last,
+            const LessThanComparable &value,
+            StrictWeakOrdering comp);
+
+
+
+} // end namespace generic
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/detail/generic/binary_search.inl>
+
diff --git a/compat/thrust/system/detail/generic/binary_search.inl b/compat/thrust/system/detail/generic/binary_search.inl
new file mode 100644
index 0000000000..151ac0ea35
--- /dev/null
+++ b/compat/thrust/system/detail/generic/binary_search.inl
@@ -0,0 +1,342 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file binary_search.inl
+ *  \brief Inline file for binary_search.h
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/distance.h>
+#include <thrust/functional.h>
+#include <thrust/binary_search.h>
+#include <thrust/iterator/zip_iterator.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/binary_search.h>
+
+#include <thrust/for_each.h>
+#include <thrust/detail/function.h>
+#include <thrust/system/detail/generic/scalar/binary_search.h>
+
+#include <thrust/detail/temporary_array.h>
+#include <thrust/detail/type_traits.h>
+
+namespace thrust
+{
+namespace detail
+{
+
+// XXX WAR circular #inclusion with this forward declaration
+template<typename,typename> class temporary_array;
+
+} // end detail
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+namespace detail
+{
+
+
+// short names to avoid nvcc bug
+struct lbf
+{
+    template <typename RandomAccessIterator, typename T, typename StrictWeakOrdering>
+    __host__ __device__
+    typename thrust::iterator_traits<RandomAccessIterator>::difference_type
+    operator()(RandomAccessIterator begin, RandomAccessIterator end, const T& value, StrictWeakOrdering comp)
+    {
+        return thrust::system::detail::generic::scalar::lower_bound(begin, end, value, comp) - begin;
+    }
+};
+
+struct ubf
+{
+    template <typename RandomAccessIterator, typename T, typename StrictWeakOrdering>
+        __host__ __device__
+        typename thrust::iterator_traits<RandomAccessIterator>::difference_type
+     operator()(RandomAccessIterator begin, RandomAccessIterator end, const T& value, StrictWeakOrdering comp){
+         return thrust::system::detail::generic::scalar::upper_bound(begin, end, value, comp) - begin;
+     }
+};
+
+struct bsf
+{
+    template <typename RandomAccessIterator, typename T, typename StrictWeakOrdering>
+        __host__ __device__
+     bool operator()(RandomAccessIterator begin, RandomAccessIterator end, const T& value, StrictWeakOrdering comp){
+         RandomAccessIterator iter = thrust::system::detail::generic::scalar::lower_bound(begin, end, value, comp);
+
+         thrust::detail::host_device_function<StrictWeakOrdering,bool> wrapped_comp(comp);
+
+         return iter != end && !wrapped_comp(value, *iter);
+     }
+};
+
+
+template <typename ForwardIterator, typename StrictWeakOrdering, typename BinarySearchFunction>
+struct binary_search_functor
+{
+    ForwardIterator begin;
+    ForwardIterator end;
+    StrictWeakOrdering comp;
+    BinarySearchFunction func;
+
+    binary_search_functor(ForwardIterator begin, ForwardIterator end, StrictWeakOrdering comp, BinarySearchFunction func)
+        : begin(begin), end(end), comp(comp), func(func) {}
+
+    template <typename Tuple>
+        __host__ __device__
+        void operator()(Tuple t)
+        {
+            thrust::get<1>(t) = func(begin, end, thrust::get<0>(t), comp);
+        }
+}; // binary_search_functor
+
+
+// Vector Implementation
+template <typename DerivedPolicy, typename ForwardIterator, typename InputIterator, typename OutputIterator, typename StrictWeakOrdering, typename BinarySearchFunction>
+OutputIterator binary_search(thrust::execution_policy<DerivedPolicy> &exec,
+                             ForwardIterator begin, 
+                             ForwardIterator end,
+                             InputIterator values_begin, 
+                             InputIterator values_end,
+                             OutputIterator output,
+                             StrictWeakOrdering comp,
+                             BinarySearchFunction func)
+{
+    thrust::for_each(exec,
+                     thrust::make_zip_iterator(thrust::make_tuple(values_begin, output)),
+                     thrust::make_zip_iterator(thrust::make_tuple(values_end, output + thrust::distance(values_begin, values_end))),
+                     detail::binary_search_functor<ForwardIterator, StrictWeakOrdering, BinarySearchFunction>(begin, end, comp, func));
+
+    return output + thrust::distance(values_begin, values_end);
+}
+
+   
+
+// Scalar Implementation
+template <typename OutputType, typename DerivedPolicy, typename ForwardIterator, typename T, typename StrictWeakOrdering, typename BinarySearchFunction>
+OutputType binary_search(thrust::execution_policy<DerivedPolicy> &exec,
+                         ForwardIterator begin,
+                         ForwardIterator end,
+                         const T& value, 
+                         StrictWeakOrdering comp,
+                         BinarySearchFunction func)
+{
+    // use the vectorized path to implement the scalar version
+
+    // allocate device buffers for value and output
+    thrust::detail::temporary_array<T,DerivedPolicy>          d_value(exec,1);
+    thrust::detail::temporary_array<OutputType,DerivedPolicy> d_output(exec,1);
+
+    // copy value to device
+    d_value[0] = value;
+
+    // perform the query
+    thrust::system::detail::generic::detail::binary_search(exec, begin, end, d_value.begin(), d_value.end(), d_output.begin(), comp, func);
+
+    // copy result to host and return
+    return d_output[0];
+}
+   
+} // end namespace detail
+
+
+//////////////////////
+// Scalar Functions //
+//////////////////////
+
+template <typename DerivedPolicy, typename ForwardIterator, typename T>
+ForwardIterator lower_bound(thrust::execution_policy<DerivedPolicy> &exec,
+                            ForwardIterator begin,
+                            ForwardIterator end,
+                            const T& value)
+{
+  return thrust::lower_bound(exec, begin, end, value, thrust::less<T>());
+}
+
+template <typename DerivedPolicy, typename ForwardIterator, typename T, typename StrictWeakOrdering>
+ForwardIterator lower_bound(thrust::execution_policy<DerivedPolicy> &exec,
+                            ForwardIterator begin,
+                            ForwardIterator end,
+                            const T& value, 
+                            StrictWeakOrdering comp)
+{
+  typedef typename thrust::iterator_traits<ForwardIterator>::difference_type difference_type;
+  
+  return begin + detail::binary_search<difference_type>(exec, begin, end, value, comp, detail::lbf());
+}
+
+
+template <typename DerivedPolicy, typename ForwardIterator, typename T>
+ForwardIterator upper_bound(thrust::execution_policy<DerivedPolicy> &exec,
+                            ForwardIterator begin,
+                            ForwardIterator end,
+                            const T& value)
+{
+  return thrust::upper_bound(exec, begin, end, value, thrust::less<T>());
+}
+
+template <typename DerivedPolicy, typename ForwardIterator, typename T, typename StrictWeakOrdering>
+ForwardIterator upper_bound(thrust::execution_policy<DerivedPolicy> &exec,
+                            ForwardIterator begin,
+                            ForwardIterator end,
+                            const T& value, 
+                            StrictWeakOrdering comp)
+{
+  typedef typename thrust::iterator_traits<ForwardIterator>::difference_type difference_type;
+  
+  return begin + detail::binary_search<difference_type>(exec, begin, end, value, comp, detail::ubf());
+}
+
+
+template <typename DerivedPolicy, typename ForwardIterator, typename T>
+bool binary_search(thrust::execution_policy<DerivedPolicy> &exec,
+                   ForwardIterator begin,
+                   ForwardIterator end,
+                   const T& value)
+{
+  return thrust::binary_search(exec, begin, end, value, thrust::less<T>());
+}
+
+template <typename DerivedPolicy, typename ForwardIterator, typename T, typename StrictWeakOrdering>
+bool binary_search(thrust::execution_policy<DerivedPolicy> &exec,
+                   ForwardIterator begin,
+                   ForwardIterator end,
+                   const T& value, 
+                   StrictWeakOrdering comp)
+{
+  return detail::binary_search<bool>(exec, begin, end, value, comp, detail::bsf());
+}
+
+
+//////////////////////
+// Vector Functions //
+//////////////////////
+
+template <typename DerivedPolicy, typename ForwardIterator, typename InputIterator, typename OutputIterator>
+OutputIterator lower_bound(thrust::execution_policy<DerivedPolicy> &exec,
+                           ForwardIterator begin, 
+                           ForwardIterator end,
+                           InputIterator values_begin, 
+                           InputIterator values_end,
+                           OutputIterator output)
+{
+  typedef typename thrust::iterator_value<InputIterator>::type ValueType;
+
+  return thrust::lower_bound(exec, begin, end, values_begin, values_end, output, thrust::less<ValueType>());
+}
+
+template <typename DerivedPolicy, typename ForwardIterator, typename InputIterator, typename OutputIterator, typename StrictWeakOrdering>
+OutputIterator lower_bound(thrust::execution_policy<DerivedPolicy> &exec,
+                           ForwardIterator begin, 
+                           ForwardIterator end,
+                           InputIterator values_begin, 
+                           InputIterator values_end,
+                           OutputIterator output,
+                           StrictWeakOrdering comp)
+{
+  return detail::binary_search(exec, begin, end, values_begin, values_end, output, comp, detail::lbf());
+}
+
+
+template <typename DerivedPolicy, typename ForwardIterator, typename InputIterator, typename OutputIterator>
+OutputIterator upper_bound(thrust::execution_policy<DerivedPolicy> &exec,
+                           ForwardIterator begin, 
+                           ForwardIterator end,
+                           InputIterator values_begin, 
+                           InputIterator values_end,
+                           OutputIterator output)
+{
+  typedef typename thrust::iterator_value<InputIterator>::type ValueType;
+
+  return thrust::upper_bound(exec, begin, end, values_begin, values_end, output, thrust::less<ValueType>());
+}
+
+template <typename DerivedPolicy, typename ForwardIterator, typename InputIterator, typename OutputIterator, typename StrictWeakOrdering>
+OutputIterator upper_bound(thrust::execution_policy<DerivedPolicy> &exec,
+                           ForwardIterator begin, 
+                           ForwardIterator end,
+                           InputIterator values_begin, 
+                           InputIterator values_end,
+                           OutputIterator output,
+                           StrictWeakOrdering comp)
+{
+  return detail::binary_search(exec, begin, end, values_begin, values_end, output, comp, detail::ubf());
+}
+
+
+template <typename DerivedPolicy, typename ForwardIterator, typename InputIterator, typename OutputIterator>
+OutputIterator binary_search(thrust::execution_policy<DerivedPolicy> &exec,
+                             ForwardIterator begin, 
+                             ForwardIterator end,
+                             InputIterator values_begin, 
+                             InputIterator values_end,
+                             OutputIterator output)
+{
+  typedef typename thrust::iterator_value<InputIterator>::type ValueType;
+
+  return thrust::binary_search(exec, begin, end, values_begin, values_end, output, thrust::less<ValueType>());
+}
+
+template <typename DerivedPolicy, typename ForwardIterator, typename InputIterator, typename OutputIterator, typename StrictWeakOrdering>
+OutputIterator binary_search(thrust::execution_policy<DerivedPolicy> &exec,
+                             ForwardIterator begin, 
+                             ForwardIterator end,
+                             InputIterator values_begin, 
+                             InputIterator values_end,
+                             OutputIterator output,
+                             StrictWeakOrdering comp)
+{
+  return detail::binary_search(exec, begin, end, values_begin, values_end, output, comp, detail::bsf());
+}
+
+
+template <typename DerivedPolicy, typename ForwardIterator, typename LessThanComparable>
+thrust::pair<ForwardIterator,ForwardIterator>
+equal_range(thrust::execution_policy<DerivedPolicy> &exec,
+            ForwardIterator first,
+            ForwardIterator last,
+            const LessThanComparable &value)
+{
+  return thrust::equal_range(exec, first, last, value, thrust::less<LessThanComparable>());
+}
+
+
+template <typename DerivedPolicy, typename ForwardIterator, typename T, typename StrictWeakOrdering>
+thrust::pair<ForwardIterator,ForwardIterator>
+equal_range(thrust::execution_policy<DerivedPolicy> &exec,
+            ForwardIterator first,
+            ForwardIterator last,
+            const T &value,
+            StrictWeakOrdering comp)
+{
+  ForwardIterator lb = thrust::lower_bound(exec, first, last, value, comp);
+  ForwardIterator ub = thrust::upper_bound(exec, first, last, value, comp);
+  return thrust::make_pair(lb, ub);
+}
+
+
+} // end namespace generic
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/detail/generic/copy.h b/compat/thrust/system/detail/generic/copy.h
new file mode 100644
index 0000000000..8df98fe67f
--- /dev/null
+++ b/compat/thrust/system/detail/generic/copy.h
@@ -0,0 +1,57 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/tag.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator>
+  OutputIterator copy(thrust::execution_policy<DerivedPolicy> &exec,
+                      InputIterator  first,
+                      InputIterator  last,
+                      OutputIterator result);
+
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename Size,
+         typename OutputIterator>
+  OutputIterator copy_n(thrust::execution_policy<DerivedPolicy> &exec,
+                        InputIterator  first,
+                        Size           n,
+                        OutputIterator result);
+
+
+} // end generic
+} // end detail
+} // end system
+} // end thrust
+
+#include <thrust/system/detail/generic/copy.inl>
+
diff --git a/compat/thrust/system/detail/generic/copy.inl b/compat/thrust/system/detail/generic/copy.inl
new file mode 100644
index 0000000000..e081015f8e
--- /dev/null
+++ b/compat/thrust/system/detail/generic/copy.inl
@@ -0,0 +1,80 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/copy.h>
+#include <thrust/functional.h>
+#include <thrust/detail/internal_functional.h>
+#include <thrust/transform.h>
+#include <thrust/for_each.h>
+#include <thrust/tuple.h>
+#include <thrust/iterator/zip_iterator.h>
+#include <thrust/iterator/detail/minimum_system.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator>
+  OutputIterator copy(thrust::execution_policy<DerivedPolicy> &exec,
+                      InputIterator                            first,
+                      InputIterator                            last,
+                      OutputIterator                           result)
+{
+  typedef typename thrust::iterator_value<InputIterator>::type T;
+  return thrust::transform(exec, first, last, result, thrust::identity<T>());
+} // end copy()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename Size,
+         typename OutputIterator>
+  OutputIterator copy_n(thrust::execution_policy<DerivedPolicy> &exec,
+                        InputIterator                            first,
+                        Size                                     n,
+                        OutputIterator                           result)
+{
+  typedef typename thrust::iterator_value<InputIterator>::type value_type;
+  typedef thrust::identity<value_type>                         xfrm_type;
+
+  // XXX why do we need to do this? figure out why, and then see if we can do without
+  typedef typename thrust::detail::unary_transform_functor<DerivedPolicy,xfrm_type>::type functor_type;
+
+  typedef thrust::tuple<InputIterator,OutputIterator> iterator_tuple;
+  typedef thrust::zip_iterator<iterator_tuple>        zip_iter;
+
+  zip_iter zipped = thrust::make_zip_iterator(thrust::make_tuple(first,result));
+
+  return thrust::get<1>(thrust::for_each_n(exec, zipped, n, functor_type(xfrm_type())).get_iterator_tuple());
+} // end copy_n()
+
+
+} // end generic
+} // end detail
+} // end system
+} // end thrust
+
diff --git a/compat/thrust/system/detail/generic/copy_if.h b/compat/thrust/system/detail/generic/copy_if.h
new file mode 100644
index 0000000000..183f012a03
--- /dev/null
+++ b/compat/thrust/system/detail/generic/copy_if.h
@@ -0,0 +1,62 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/tag.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator,
+         typename Predicate>
+  OutputIterator copy_if(thrust::execution_policy<DerivedPolicy> &exec,
+                         InputIterator first,
+                         InputIterator last,
+                         OutputIterator result,
+                         Predicate pred);
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename Predicate>
+   OutputIterator copy_if(thrust::execution_policy<DerivedPolicy> &exec,
+                          InputIterator1 first,
+                          InputIterator1 last,
+                          InputIterator2 stencil,
+                          OutputIterator result,
+                          Predicate pred);
+
+
+} // end namespace generic
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/detail/generic/copy_if.inl>
+
diff --git a/compat/thrust/system/detail/generic/copy_if.inl b/compat/thrust/system/detail/generic/copy_if.inl
new file mode 100644
index 0000000000..145561c9bc
--- /dev/null
+++ b/compat/thrust/system/detail/generic/copy_if.inl
@@ -0,0 +1,155 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/copy_if.h>
+#include <thrust/detail/copy_if.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/iterator/detail/minimum_system.h>
+#include <thrust/functional.h>
+#include <thrust/distance.h>
+#include <thrust/transform.h>
+#include <thrust/detail/internal_functional.h>
+#include <thrust/detail/temporary_array.h>
+#include <thrust/detail/type_traits.h>
+#include <thrust/scan.h>
+#include <thrust/scatter.h>
+#include <limits>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+namespace detail
+{
+
+template<typename IndexType,
+         typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename Predicate>
+OutputIterator copy_if(thrust::execution_policy<DerivedPolicy> &exec,
+                       InputIterator1 first,
+                       InputIterator1 last,
+                       InputIterator2 stencil,
+                       OutputIterator result,
+                       Predicate pred)
+{
+    __THRUST_DISABLE_MSVC_POSSIBLE_LOSS_OF_DATA_WARNING(IndexType n = thrust::distance(first, last));
+
+    // compute {0,1} predicates
+    thrust::detail::temporary_array<IndexType, DerivedPolicy> predicates(exec, n);
+    thrust::transform(exec,
+                      stencil,
+                      stencil + n,
+                      predicates.begin(),
+                      thrust::detail::predicate_to_integral<Predicate,IndexType>(pred));
+
+    // scan {0,1} predicates
+    thrust::detail::temporary_array<IndexType, DerivedPolicy> scatter_indices(exec, n);
+    thrust::exclusive_scan(exec,
+                           predicates.begin(),
+                           predicates.end(),
+                           scatter_indices.begin(),
+                           static_cast<IndexType>(0),
+                           thrust::plus<IndexType>());
+
+    // scatter the true elements
+    thrust::scatter_if(exec,
+                       first,
+                       last,
+                       scatter_indices.begin(),
+                       predicates.begin(),
+                       result,
+                       thrust::identity<IndexType>());
+
+    // find the end of the new sequence
+    IndexType output_size = scatter_indices[n - 1] + predicates[n - 1];
+
+    return result + output_size;
+}
+
+} // end namespace detail
+
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator,
+         typename Predicate>
+  OutputIterator copy_if(thrust::execution_policy<DerivedPolicy> &exec,
+                         InputIterator first,
+                         InputIterator last,
+                         OutputIterator result,
+                         Predicate pred)
+{
+  // XXX it's potentially expensive to send [first,last) twice
+  //     we should probably specialize this case for POD
+  //     since we can safely keep the input in a temporary instead
+  //     of doing two loads
+  return thrust::copy_if(exec, first, last, first, result, pred);
+} // end copy_if()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename Predicate>
+   OutputIterator copy_if(thrust::execution_policy<DerivedPolicy> &exec,
+                          InputIterator1 first,
+                          InputIterator1 last,
+                          InputIterator2 stencil,
+                          OutputIterator result,
+                          Predicate pred)
+{
+  typedef typename thrust::iterator_traits<InputIterator1>::difference_type difference_type;
+  
+  // empty sequence
+  if(first == last)
+    return result;
+  
+  difference_type n = thrust::distance(first, last);
+  
+  // create an unsigned version of n (we know n is positive from the comparison above)
+  // to avoid a warning in the compare below
+  typename thrust::detail::make_unsigned<difference_type>::type unsigned_n(n);
+  
+  // use 32-bit indices when possible (almost always)
+  if(sizeof(difference_type) > sizeof(unsigned int) && unsigned_n > (std::numeric_limits<unsigned int>::max)())
+  {
+    result = detail::copy_if<difference_type>(exec, first, last, stencil, result, pred);
+  } // end if
+  else
+  {
+    result = detail::copy_if<unsigned int>(exec, first, last, stencil, result, pred);
+  } // end else
+
+  return result;
+} // end copy_if()
+
+
+} // end namespace generic
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/detail/generic/count.h b/compat/thrust/system/detail/generic/count.h
new file mode 100644
index 0000000000..bc4899e6d0
--- /dev/null
+++ b/compat/thrust/system/detail/generic/count.h
@@ -0,0 +1,46 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/tag.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+template <typename DerivedPolicy, typename InputIterator, typename EqualityComparable>
+typename thrust::iterator_traits<InputIterator>::difference_type
+count(thrust::execution_policy<DerivedPolicy> &exec, InputIterator first, InputIterator last, const EqualityComparable& value);
+
+template <typename DerivedPolicy, typename InputIterator, typename Predicate>
+typename thrust::iterator_traits<InputIterator>::difference_type
+count_if(thrust::execution_policy<DerivedPolicy> &exec, InputIterator first, InputIterator last, Predicate pred);
+
+} // end namespace generic
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/detail/generic/count.inl>
+
diff --git a/compat/thrust/system/detail/generic/count.inl b/compat/thrust/system/detail/generic/count.inl
new file mode 100644
index 0000000000..e3ab8714b7
--- /dev/null
+++ b/compat/thrust/system/detail/generic/count.inl
@@ -0,0 +1,75 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/count.h>
+#include <thrust/transform_reduce.h>
+#include <thrust/detail/internal_functional.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+template <typename InputType, typename Predicate, typename CountType>
+struct count_if_transform
+{
+  __host__ __device__ 
+  count_if_transform(Predicate _pred) : pred(_pred){}
+
+  __host__ __device__
+  CountType operator()(const InputType& val)
+  {
+    if(pred(val))
+      return 1;
+    else
+      return 0;
+  } // end operator()
+
+  Predicate pred;
+}; // end count_if_transform
+
+template <typename DerivedPolicy, typename InputIterator, typename EqualityComparable>
+typename thrust::iterator_traits<InputIterator>::difference_type
+count(thrust::execution_policy<DerivedPolicy> &exec, InputIterator first, InputIterator last, const EqualityComparable& value)
+{
+  typedef typename thrust::iterator_traits<InputIterator>::value_type InputType;
+  
+  // XXX use placeholder expression here
+  return thrust::count_if(exec, first, last, thrust::detail::equal_to_value<EqualityComparable>(value));
+} // end count()
+
+template <typename DerivedPolicy, typename InputIterator, typename Predicate>
+typename thrust::iterator_traits<InputIterator>::difference_type
+count_if(thrust::execution_policy<DerivedPolicy> &exec, InputIterator first, InputIterator last, Predicate pred)
+{
+  typedef typename thrust::iterator_traits<InputIterator>::value_type InputType;
+  typedef typename thrust::iterator_traits<InputIterator>::difference_type CountType;
+  
+  thrust::system::detail::generic::count_if_transform<InputType, Predicate, CountType> unary_op(pred);
+  thrust::plus<CountType> binary_op;
+  return thrust::transform_reduce(exec, first, last, unary_op, CountType(0), binary_op);
+} // end count_if()
+
+} // end generic
+} // end detail
+} // end system
+} // end thrust
+
diff --git a/compat/thrust/system/detail/generic/distance.h b/compat/thrust/system/detail/generic/distance.h
new file mode 100644
index 0000000000..80f051ca53
--- /dev/null
+++ b/compat/thrust/system/detail/generic/distance.h
@@ -0,0 +1,42 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/iterator/iterator_traits.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+template<typename InputIterator>
+  inline typename thrust::iterator_traits<InputIterator>::difference_type
+    distance(InputIterator first, InputIterator last);
+
+} // end namespace generic
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/detail/generic/distance.inl>
+
diff --git a/compat/thrust/system/detail/generic/distance.inl b/compat/thrust/system/detail/generic/distance.inl
new file mode 100644
index 0000000000..a1fdf1458b
--- /dev/null
+++ b/compat/thrust/system/detail/generic/distance.inl
@@ -0,0 +1,69 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/distance.h>
+#include <thrust/iterator/iterator_traits.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+namespace detail
+{
+
+template<typename InputIterator>
+  inline typename thrust::iterator_traits<InputIterator>::difference_type
+    distance(InputIterator first, InputIterator last, thrust::incrementable_traversal_tag)
+{
+  typename thrust::iterator_traits<InputIterator>::difference_type result(0);
+
+  while(first != last)
+  {
+    ++first;
+    ++result;
+  } // end while
+
+  return result;
+} // end advance()
+
+template<typename InputIterator>
+  inline typename thrust::iterator_traits<InputIterator>::difference_type
+    distance(InputIterator first, InputIterator last, thrust::random_access_traversal_tag)
+{
+  return last - first;
+} // end distance()
+
+} // end detail
+
+template<typename InputIterator>
+  inline typename thrust::iterator_traits<InputIterator>::difference_type
+    distance(InputIterator first, InputIterator last)
+{
+  // dispatch on iterator traversal
+  return thrust::system::detail::generic::detail::distance(first, last,
+    typename thrust::iterator_traversal<InputIterator>::type());
+} // end advance()
+
+} // end namespace generic
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/detail/generic/equal.h b/compat/thrust/system/detail/generic/equal.h
new file mode 100644
index 0000000000..da7d105825
--- /dev/null
+++ b/compat/thrust/system/detail/generic/equal.h
@@ -0,0 +1,43 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/tag.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+template<typename DerivedPolicy, typename InputIterator1, typename InputIterator2>
+bool equal(thrust::execution_policy<DerivedPolicy> &exec, InputIterator1 first1, InputIterator1 last1, InputIterator2 first2);
+
+template<typename DerivedPolicy, typename InputIterator1, typename InputIterator2, typename BinaryPredicate>
+bool equal(thrust::execution_policy<DerivedPolicy> &exec, InputIterator1 first1, InputIterator1 last1, InputIterator2 first2, BinaryPredicate binary_pred);
+
+} // end namespace generic
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/detail/generic/equal.inl>
+
diff --git a/compat/thrust/system/detail/generic/equal.inl b/compat/thrust/system/detail/generic/equal.inl
new file mode 100644
index 0000000000..12b8005a2a
--- /dev/null
+++ b/compat/thrust/system/detail/generic/equal.inl
@@ -0,0 +1,50 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/equal.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/detail/internal_functional.h>
+#include <thrust/mismatch.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+template<typename DerivedPolicy, typename InputIterator1, typename InputIterator2>
+bool equal(thrust::execution_policy<DerivedPolicy> &exec, InputIterator1 first1, InputIterator1 last1, InputIterator2 first2)
+{
+  typedef typename thrust::iterator_traits<InputIterator1>::value_type InputType1;
+  
+  return thrust::equal(exec, first1, last1, first2, thrust::detail::equal_to<InputType1>());
+}
+
+template<typename DerivedPolicy, typename InputIterator1, typename InputIterator2, typename BinaryPredicate>
+bool equal(thrust::execution_policy<DerivedPolicy> &exec, InputIterator1 first1, InputIterator1 last1, InputIterator2 first2, BinaryPredicate binary_pred)
+{
+  return thrust::mismatch(exec, first1, last1, first2, binary_pred).first == last1;
+}
+
+} // end generic
+} // end detail
+} // end system
+} // end thrust
+
diff --git a/compat/thrust/system/detail/generic/extrema.h b/compat/thrust/system/detail/generic/extrema.h
new file mode 100644
index 0000000000..abb4ddc210
--- /dev/null
+++ b/compat/thrust/system/detail/generic/extrema.h
@@ -0,0 +1,76 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file extrema.h
+ *  \brief Generic device implementations of extrema functions.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/pair.h>
+#include <thrust/system/detail/generic/tag.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+template <typename DerivedPolicy, typename ForwardIterator>
+ForwardIterator max_element(thrust::execution_policy<DerivedPolicy> &exec,
+                            ForwardIterator first,
+                            ForwardIterator last);
+
+template <typename DerivedPolicy, typename ForwardIterator, typename BinaryPredicate>
+ForwardIterator max_element(thrust::execution_policy<DerivedPolicy> &exec,
+                            ForwardIterator first,
+                            ForwardIterator last,
+                            BinaryPredicate comp);
+
+template <typename DerivedPolicy, typename ForwardIterator>
+ForwardIterator min_element(thrust::execution_policy<DerivedPolicy> &exec,
+                            ForwardIterator first,
+                            ForwardIterator last);
+
+template <typename DerivedPolicy, typename ForwardIterator, typename BinaryPredicate>
+ForwardIterator min_element(thrust::execution_policy<DerivedPolicy> &exec,
+                            ForwardIterator first,
+                            ForwardIterator last,
+                            BinaryPredicate comp);
+
+template <typename DerivedPolicy, typename ForwardIterator>
+thrust::pair<ForwardIterator,ForwardIterator> minmax_element(thrust::execution_policy<DerivedPolicy> &exec,
+                                                             ForwardIterator first, 
+                                                             ForwardIterator last);
+
+template <typename DerivedPolicy, typename ForwardIterator, typename BinaryPredicate>
+thrust::pair<ForwardIterator,ForwardIterator> minmax_element(thrust::execution_policy<DerivedPolicy> &exec,
+                                                             ForwardIterator first, 
+                                                             ForwardIterator last,
+                                                             BinaryPredicate comp);
+
+} // end namespace generic
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/detail/generic/extrema.inl>
+
diff --git a/compat/thrust/system/detail/generic/extrema.inl b/compat/thrust/system/detail/generic/extrema.inl
new file mode 100644
index 0000000000..b5f92c3935
--- /dev/null
+++ b/compat/thrust/system/detail/generic/extrema.inl
@@ -0,0 +1,244 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file distance.h
+ *  \brief Device implementations for distance.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/extrema.h>
+#include <thrust/functional.h>
+#include <thrust/pair.h>
+#include <thrust/reduce.h>
+#include <thrust/transform_reduce.h>
+
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/iterator/counting_iterator.h>
+#include <thrust/iterator/zip_iterator.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+namespace detail
+{
+
+//////////////
+// Functors //
+//////////////
+
+// return the smaller/larger element making sure to prefer the 
+// first occurance of the minimum/maximum element
+template <typename InputType, typename IndexType, typename BinaryPredicate>
+struct min_element_reduction
+{
+  BinaryPredicate comp;
+
+  __host__ __device__ 
+  min_element_reduction(BinaryPredicate comp) : comp(comp){}
+
+  __host__ __device__ 
+  thrust::tuple<InputType, IndexType>
+  operator()(const thrust::tuple<InputType, IndexType>& lhs, 
+             const thrust::tuple<InputType, IndexType>& rhs )
+  {
+    if(comp(thrust::get<0>(lhs), thrust::get<0>(rhs)))
+      return lhs;
+    if(comp(thrust::get<0>(rhs), thrust::get<0>(lhs)))
+      return rhs;
+
+    // values are equivalent, prefer value with smaller index
+    if(thrust::get<1>(lhs) < thrust::get<1>(rhs))
+      return lhs;
+    else
+      return rhs;
+  } // end operator()()
+
+}; // end min_element_reduction
+
+
+template <typename InputType, typename IndexType, typename BinaryPredicate>
+struct max_element_reduction
+{
+  BinaryPredicate comp;
+
+  __host__ __device__ 
+  max_element_reduction(BinaryPredicate comp) : comp(comp){}
+
+  __host__ __device__ 
+  thrust::tuple<InputType, IndexType>
+  operator()(const thrust::tuple<InputType, IndexType>& lhs, 
+             const thrust::tuple<InputType, IndexType>& rhs )
+  {
+    if(comp(thrust::get<0>(lhs), thrust::get<0>(rhs)))
+      return rhs;
+    if(comp(thrust::get<0>(rhs), thrust::get<0>(lhs)))
+      return lhs;
+
+    // values are equivalent, prefer value with smaller index
+    if(thrust::get<1>(lhs) < thrust::get<1>(rhs))
+      return lhs;
+    else
+      return rhs;
+  } // end operator()()
+
+}; // end max_element_reduction
+
+// return the smaller & larger element making sure to prefer the 
+// first occurance of the minimum/maximum element
+template <typename InputType, typename IndexType, typename BinaryPredicate>
+struct minmax_element_reduction
+{
+  BinaryPredicate comp;
+
+  minmax_element_reduction(BinaryPredicate comp) : comp(comp){}
+
+  __host__ __device__ 
+  thrust::tuple< thrust::tuple<InputType,IndexType>, thrust::tuple<InputType,IndexType> >
+  operator()(const thrust::tuple< thrust::tuple<InputType,IndexType>, thrust::tuple<InputType,IndexType> >& lhs, 
+             const thrust::tuple< thrust::tuple<InputType,IndexType>, thrust::tuple<InputType,IndexType> >& rhs )
+  {
+
+    return thrust::make_tuple(min_element_reduction<InputType, IndexType, BinaryPredicate>(comp)(thrust::get<0>(lhs), thrust::get<0>(rhs)),
+                              max_element_reduction<InputType, IndexType, BinaryPredicate>(comp)(thrust::get<1>(lhs), thrust::get<1>(rhs)));
+  } // end operator()()
+}; // end minmax_element_reduction
+
+template <typename InputType, typename IndexType>
+struct duplicate_tuple
+{
+  __host__ __device__ 
+  thrust::tuple< thrust::tuple<InputType,IndexType>, thrust::tuple<InputType,IndexType> >
+  operator()(const thrust::tuple<InputType,IndexType>& t)
+  {
+    return thrust::make_tuple(t, t);
+  }
+}; // end duplicate_tuple
+
+} // end namespace detail
+
+template <typename DerivedPolicy, typename ForwardIterator>
+ForwardIterator min_element(thrust::execution_policy<DerivedPolicy> &exec,
+                            ForwardIterator first,
+                            ForwardIterator last)
+{
+  typedef typename thrust::iterator_value<ForwardIterator>::type value_type;
+
+  return thrust::min_element(exec, first, last, thrust::less<value_type>());
+} // end min_element()
+
+template <typename DerivedPolicy, typename ForwardIterator, typename BinaryPredicate>
+ForwardIterator min_element(thrust::execution_policy<DerivedPolicy> &exec,
+                            ForwardIterator first,
+                            ForwardIterator last,
+                            BinaryPredicate comp)
+{
+  if (first == last)
+    return last;
+
+  typedef typename thrust::iterator_traits<ForwardIterator>::value_type      InputType;
+  typedef typename thrust::iterator_traits<ForwardIterator>::difference_type IndexType;
+
+  thrust::tuple<InputType, IndexType> result =
+    thrust::reduce
+      (exec,
+       thrust::make_zip_iterator(thrust::make_tuple(first, thrust::counting_iterator<IndexType>(0))),
+       thrust::make_zip_iterator(thrust::make_tuple(first, thrust::counting_iterator<IndexType>(0))) + (last - first),
+       thrust::tuple<InputType, IndexType>(*first, 0),
+       detail::min_element_reduction<InputType, IndexType, BinaryPredicate>(comp));
+
+  return first + thrust::get<1>(result);
+} // end min_element()
+
+template <typename DerivedPolicy, typename ForwardIterator>
+ForwardIterator max_element(thrust::execution_policy<DerivedPolicy> &exec,
+                            ForwardIterator first,
+                            ForwardIterator last)
+{
+  typedef typename thrust::iterator_value<ForwardIterator>::type value_type;
+
+  return thrust::max_element(exec, first, last, thrust::less<value_type>());
+} // end max_element()
+
+template <typename DerivedPolicy, typename ForwardIterator, typename BinaryPredicate>
+ForwardIterator max_element(thrust::execution_policy<DerivedPolicy> &exec,
+                            ForwardIterator first,
+                            ForwardIterator last,
+                            BinaryPredicate comp)
+{
+  if (first == last)
+    return last;
+
+  typedef typename thrust::iterator_traits<ForwardIterator>::value_type      InputType;
+  typedef typename thrust::iterator_traits<ForwardIterator>::difference_type IndexType;
+
+  thrust::tuple<InputType, IndexType> result =
+    thrust::reduce
+      (exec,
+       thrust::make_zip_iterator(thrust::make_tuple(first, thrust::counting_iterator<IndexType>(0))),
+       thrust::make_zip_iterator(thrust::make_tuple(first, thrust::counting_iterator<IndexType>(0))) + (last - first),
+       thrust::tuple<InputType, IndexType>(*first, 0),
+       detail::max_element_reduction<InputType, IndexType, BinaryPredicate>(comp));
+
+  return first + thrust::get<1>(result);
+} // end max_element()
+
+template <typename DerivedPolicy, typename ForwardIterator>
+thrust::pair<ForwardIterator,ForwardIterator> minmax_element(thrust::execution_policy<DerivedPolicy> &exec,
+                                                             ForwardIterator first, 
+                                                             ForwardIterator last)
+{
+  typedef typename thrust::iterator_value<ForwardIterator>::type value_type;
+
+  return thrust::minmax_element(exec, first, last, thrust::less<value_type>());
+} // end minmax_element()
+
+template <typename DerivedPolicy, typename ForwardIterator, typename BinaryPredicate>
+thrust::pair<ForwardIterator,ForwardIterator> minmax_element(thrust::execution_policy<DerivedPolicy> &exec,
+                                                             ForwardIterator first, 
+                                                             ForwardIterator last,
+                                                             BinaryPredicate comp)
+{
+  if (first == last)
+    return thrust::make_pair(last, last);
+
+  typedef typename thrust::iterator_traits<ForwardIterator>::value_type      InputType;
+  typedef typename thrust::iterator_traits<ForwardIterator>::difference_type IndexType;
+
+  thrust::tuple< thrust::tuple<InputType,IndexType>, thrust::tuple<InputType,IndexType> > result = 
+    thrust::transform_reduce
+      (exec,
+       thrust::make_zip_iterator(thrust::make_tuple(first, thrust::counting_iterator<IndexType>(0))),
+       thrust::make_zip_iterator(thrust::make_tuple(first, thrust::counting_iterator<IndexType>(0))) + (last - first),
+       detail::duplicate_tuple<InputType, IndexType>(),
+       detail::duplicate_tuple<InputType, IndexType>()(thrust::tuple<InputType, IndexType>(*first, 0)),
+       detail::minmax_element_reduction<InputType, IndexType, BinaryPredicate>(comp));
+
+  return thrust::make_pair(first + thrust::get<1>(thrust::get<0>(result)), first + thrust::get<1>(thrust::get<1>(result)));
+} // end minmax_element()
+
+} // end namespace generic
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/detail/generic/fill.h b/compat/thrust/system/detail/generic/fill.h
new file mode 100644
index 0000000000..9745b1cf57
--- /dev/null
+++ b/compat/thrust/system/detail/generic/fill.h
@@ -0,0 +1,63 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file fill.h
+ *  \brief Device implementation of fill.
+ */
+
+#pragma once
+
+#include <thrust/detail/internal_functional.h>
+#include <thrust/generate.h>
+#include <thrust/system/detail/generic/tag.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+
+template<typename DerivedPolicy, typename OutputIterator, typename Size, typename T>
+  OutputIterator fill_n(thrust::execution_policy<DerivedPolicy> &exec,
+                        OutputIterator first,
+                        Size n,
+                        const T &value)
+{
+  // XXX consider using the placeholder expression _1 = value
+  return thrust::generate_n(exec, first, n, thrust::detail::fill_functor<T>(value));
+}
+
+template<typename DerivedPolicy, typename ForwardIterator, typename T>
+  void fill(thrust::execution_policy<DerivedPolicy> &exec,
+            ForwardIterator first,
+            ForwardIterator last,
+            const T &value)
+{
+  // XXX consider using the placeholder expression _1 = value
+  thrust::generate(exec, first, last, thrust::detail::fill_functor<T>(value));
+}
+
+
+} // end namespace generic
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/detail/generic/find.h b/compat/thrust/system/detail/generic/find.h
new file mode 100644
index 0000000000..08888c5a7c
--- /dev/null
+++ b/compat/thrust/system/detail/generic/find.h
@@ -0,0 +1,56 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/tag.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+template<typename DerivedPolicy, typename InputIterator, typename T>
+InputIterator find(thrust::execution_policy<DerivedPolicy> &exec,
+                   InputIterator first,
+                   InputIterator last,
+                   const T& value);
+
+template<typename DerivedPolicy, typename InputIterator, typename Predicate>
+InputIterator find_if(thrust::execution_policy<DerivedPolicy> &exec,
+                      InputIterator first,
+                      InputIterator last,
+                      Predicate pred);
+
+template<typename DerivedPolicy, typename InputIterator, typename Predicate>
+InputIterator find_if_not(thrust::execution_policy<DerivedPolicy> &exec,
+                          InputIterator first,
+                          InputIterator last,
+                          Predicate pred);
+
+} // end namespace generic
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/detail/generic/find.inl>
+
diff --git a/compat/thrust/system/detail/generic/find.inl b/compat/thrust/system/detail/generic/find.inl
new file mode 100644
index 0000000000..a3414e1c28
--- /dev/null
+++ b/compat/thrust/system/detail/generic/find.inl
@@ -0,0 +1,141 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/find.h>
+#include <thrust/reduce.h>
+
+#include <thrust/tuple.h>
+#include <thrust/extrema.h>
+#include <thrust/iterator/counting_iterator.h>
+#include <thrust/iterator/transform_iterator.h>
+#include <thrust/iterator/zip_iterator.h>
+#include <thrust/detail/internal_functional.h>
+
+
+// Contributed by Erich Elsen
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+
+template<typename DerivedPolicy, typename InputIterator, typename T>
+InputIterator find(thrust::execution_policy<DerivedPolicy> &exec,
+                   InputIterator first,
+                   InputIterator last,
+                   const T& value)
+{
+  // XXX consider a placeholder expression here
+  return thrust::find_if(exec, first, last, thrust::detail::equal_to_value<T>(value));
+} // end find()
+
+
+template<typename TupleType>
+struct find_if_functor
+{
+    __host__ __device__
+    TupleType operator()(const TupleType& lhs, const TupleType& rhs) const
+    {
+        // select the smallest index among true results
+        if (thrust::get<0>(lhs) && thrust::get<0>(rhs))
+            return TupleType(true, (thrust::min)(thrust::get<1>(lhs), thrust::get<1>(rhs)));
+        else if (thrust::get<0>(lhs))
+            return lhs;
+        else
+            return rhs;
+    }
+};
+    
+
+template<typename DerivedPolicy, typename InputIterator, typename Predicate>
+InputIterator find_if(thrust::execution_policy<DerivedPolicy> &exec,
+                      InputIterator first,
+                      InputIterator last,
+                      Predicate pred)
+{
+    typedef typename thrust::iterator_traits<InputIterator>::difference_type difference_type;
+    typedef typename thrust::tuple<bool,difference_type> result_type;
+   
+    // empty sequence
+    if (first == last)
+        return last;
+
+    const difference_type n = thrust::distance(first, last);
+
+    // this implementation breaks up the sequence into separate intervals
+    // in an attempt to early-out as soon as a value is found
+
+    // TODO incorporate sizeof(InputType) into interval_threshold and round to multiple of 32
+    const difference_type interval_threshold = 1 << 20;
+    const difference_type interval_size = (std::min)(interval_threshold, n);
+
+    // force transform_iterator output to bool
+    typedef thrust::transform_iterator<Predicate, InputIterator, bool> XfrmIterator;
+    typedef thrust::tuple<XfrmIterator, thrust::counting_iterator<difference_type> > IteratorTuple;
+    typedef thrust::zip_iterator<IteratorTuple> ZipIterator;
+
+    IteratorTuple iter_tuple = thrust::make_tuple(XfrmIterator(first, pred),
+                                                  thrust::counting_iterator<difference_type>(0));
+
+    ZipIterator begin = thrust::make_zip_iterator(iter_tuple);
+    ZipIterator end   = begin + n;
+
+    for(ZipIterator interval_begin = begin; interval_begin < end; interval_begin += interval_size)
+    {
+        ZipIterator interval_end = interval_begin + interval_size;
+        if(end < interval_end)
+        {
+          interval_end = end;
+        } // end if
+
+        result_type result = thrust::reduce(exec,
+                                            interval_begin, interval_end,
+                                            result_type(false,interval_end - begin),
+                                            find_if_functor<result_type>());
+
+        // see if we found something
+        if (thrust::get<0>(result))
+        {
+            return first + thrust::get<1>(result);
+        }
+    }
+
+    //nothing was found if we reach here...
+    return first + n;
+}
+
+
+template<typename DerivedPolicy, typename InputIterator, typename Predicate>
+InputIterator find_if_not(thrust::execution_policy<DerivedPolicy> &exec,
+                          InputIterator first,
+                          InputIterator last,
+                          Predicate pred)
+{
+    return thrust::find_if(exec, first, last, thrust::detail::not1(pred));
+} // end find()
+
+
+} // end namespace generic
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/detail/generic/for_each.h b/compat/thrust/system/detail/generic/for_each.h
new file mode 100644
index 0000000000..61abe20b6f
--- /dev/null
+++ b/compat/thrust/system/detail/generic/for_each.h
@@ -0,0 +1,72 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+/*! \file for_each.h
+ *  \brief Generic implementation of for_each & for_each_n.
+ *         It is an error to call these functions; they have no implementation.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/execution_policy.h>
+#include <thrust/system/detail/generic/tag.h>
+#include <thrust/detail/static_assert.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename UnaryFunction>
+InputIterator for_each(thrust::execution_policy<DerivedPolicy> &exec,
+                       InputIterator first,
+                       InputIterator last,
+                       UnaryFunction f)
+{
+  // unimplemented
+  THRUST_STATIC_ASSERT( (thrust::detail::depend_on_instantiation<InputIterator, false>::value) );
+  return first;
+} // end for_each()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename Size,
+         typename UnaryFunction>
+InputIterator for_each_n(thrust::execution_policy<DerivedPolicy> &exec,
+                         InputIterator first,
+                         Size n,
+                         UnaryFunction f)
+{
+  // unimplemented
+  THRUST_STATIC_ASSERT( (thrust::detail::depend_on_instantiation<InputIterator, false>::value) );
+  return first;
+} // end for_each_n()
+
+
+} // end namespace generic
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/detail/generic/gather.h b/compat/thrust/system/detail/generic/gather.h
new file mode 100644
index 0000000000..cfb6f85ca5
--- /dev/null
+++ b/compat/thrust/system/detail/generic/gather.h
@@ -0,0 +1,78 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/tag.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+
+template<typename ExecutionPolicy,
+         typename InputIterator,
+         typename RandomAccessIterator,
+         typename OutputIterator>
+  OutputIterator gather(thrust::execution_policy<ExecutionPolicy> &exec,
+                        InputIterator                              map_first,
+                        InputIterator                              map_last,
+                        RandomAccessIterator                       input_first,
+                        OutputIterator                             result);
+
+
+template<typename ExecutionPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename RandomAccessIterator,
+         typename OutputIterator>
+  OutputIterator gather_if(thrust::execution_policy<ExecutionPolicy> &exec,
+                           InputIterator1                             map_first,
+                           InputIterator1                             map_last,
+                           InputIterator2                             stencil,
+                           RandomAccessIterator                       input_first,
+                           OutputIterator                             result);
+
+
+template<typename ExecutionPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename RandomAccessIterator,
+         typename OutputIterator,
+         typename Predicate>
+  OutputIterator gather_if(thrust::execution_policy<ExecutionPolicy> &exec,
+                           InputIterator1                             map_first,
+                           InputIterator1                             map_last,
+                           InputIterator2                             stencil,
+                           RandomAccessIterator                       input_first,
+                           OutputIterator                             result,
+                           Predicate                                  pred);
+
+
+} // end namespace generic
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/detail/generic/gather.inl>
+
diff --git a/compat/thrust/system/detail/generic/gather.inl b/compat/thrust/system/detail/generic/gather.inl
new file mode 100644
index 0000000000..ab2cdd8116
--- /dev/null
+++ b/compat/thrust/system/detail/generic/gather.inl
@@ -0,0 +1,102 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/gather.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/functional.h>
+#include <thrust/transform.h>
+#include <thrust/iterator/permutation_iterator.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename RandomAccessIterator,
+         typename OutputIterator>
+  OutputIterator gather(thrust::execution_policy<DerivedPolicy> &exec,
+                        InputIterator                            map_first,
+                        InputIterator                            map_last,
+                        RandomAccessIterator                     input_first,
+                        OutputIterator                           result)
+{
+  return thrust::transform(exec,
+                           thrust::make_permutation_iterator(input_first, map_first),
+                           thrust::make_permutation_iterator(input_first, map_last),
+                           result,
+                           thrust::identity<typename thrust::iterator_value<RandomAccessIterator>::type>());
+} // end gather()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename RandomAccessIterator,
+         typename OutputIterator>
+  OutputIterator gather_if(thrust::execution_policy<DerivedPolicy> &exec,
+                           InputIterator1                           map_first,
+                           InputIterator1                           map_last,
+                           InputIterator2                           stencil,
+                           RandomAccessIterator                     input_first,
+                           OutputIterator                           result)
+{
+  typedef typename thrust::iterator_value<InputIterator2>::type StencilType;
+  return thrust::gather_if(exec,
+                           map_first,
+                           map_last,
+                           stencil,
+                           input_first,
+                           result,
+                           thrust::identity<StencilType>());
+} // end gather_if()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename RandomAccessIterator,
+         typename OutputIterator,
+         typename Predicate>
+  OutputIterator gather_if(thrust::execution_policy<DerivedPolicy> &exec,
+                           InputIterator1                           map_first,
+                           InputIterator1                           map_last,
+                           InputIterator2                           stencil,
+                           RandomAccessIterator                     input_first,
+                           OutputIterator                           result,
+                           Predicate                                pred)
+{
+  typedef typename thrust::iterator_value<RandomAccessIterator>::type InputType;
+  return thrust::transform_if(exec,
+                              thrust::make_permutation_iterator(input_first, map_first),
+                              thrust::make_permutation_iterator(input_first, map_last),
+                              stencil,
+                              result,
+                              thrust::identity<InputType>(),
+                              pred);
+} // end gather_if()
+
+} // end namespace generic
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/detail/generic/generate.h b/compat/thrust/system/detail/generic/generate.h
new file mode 100644
index 0000000000..e7a8e00726
--- /dev/null
+++ b/compat/thrust/system/detail/generic/generate.h
@@ -0,0 +1,55 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/tag.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+template<typename ExecutionPolicy,
+         typename ForwardIterator,
+         typename Generator>
+  void generate(thrust::execution_policy<ExecutionPolicy> &exec,
+                ForwardIterator first,
+                ForwardIterator last,
+                Generator gen);
+
+template<typename ExecutionPolicy,
+         typename OutputIterator,
+         typename Size,
+         typename Generator>
+  OutputIterator generate_n(thrust::execution_policy<ExecutionPolicy> &exec,
+                            OutputIterator first,
+                            Size n,
+                            Generator gen);
+
+} // end namespace generic
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/detail/generic/generate.inl>
+
diff --git a/compat/thrust/system/detail/generic/generate.inl b/compat/thrust/system/detail/generic/generate.inl
new file mode 100644
index 0000000000..4da5763f9c
--- /dev/null
+++ b/compat/thrust/system/detail/generic/generate.inl
@@ -0,0 +1,59 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/generate.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/detail/internal_functional.h>
+#include <thrust/for_each.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+template<typename ExecutionPolicy,
+         typename ForwardIterator,
+         typename Generator>
+  void generate(thrust::execution_policy<ExecutionPolicy> &exec,
+                ForwardIterator first,
+                ForwardIterator last,
+                Generator gen)
+{
+  thrust::for_each(exec, first, last, typename thrust::detail::generate_functor<ExecutionPolicy,Generator>::type(gen));
+} // end generate()
+
+template<typename ExecutionPolicy,
+         typename OutputIterator,
+         typename Size,
+         typename Generator>
+  OutputIterator generate_n(thrust::execution_policy<ExecutionPolicy> &exec,
+                            OutputIterator first,
+                            Size n,
+                            Generator gen)
+{
+  return thrust::for_each_n(exec, first, n, typename thrust::detail::generate_functor<ExecutionPolicy,Generator>::type(gen));
+} // end generate()
+
+} // end namespace generic
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/detail/generic/inner_product.h b/compat/thrust/system/detail/generic/inner_product.h
new file mode 100644
index 0000000000..9ac5c69636
--- /dev/null
+++ b/compat/thrust/system/detail/generic/inner_product.h
@@ -0,0 +1,54 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/tag.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+template<typename DerivedPolicy, typename InputIterator1, typename InputIterator2, typename OutputType>
+  OutputType inner_product(thrust::execution_policy<DerivedPolicy> &exec,
+                           InputIterator1 first1,
+                           InputIterator1 last1,
+                           InputIterator2 first2,
+                           OutputType init);
+
+template<typename DerivedPolicy, typename InputIterator1, typename InputIterator2, typename OutputType, typename BinaryFunction1, typename BinaryFunction2>
+OutputType inner_product(thrust::execution_policy<DerivedPolicy> &exec,
+                         InputIterator1 first1,
+                         InputIterator1 last1,
+                         InputIterator2 first2,
+                         OutputType init, 
+                         BinaryFunction1 binary_op1,
+                         BinaryFunction2 binary_op2);
+
+} // end namespace generic
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/detail/generic/inner_product.inl>
+
diff --git a/compat/thrust/system/detail/generic/inner_product.inl b/compat/thrust/system/detail/generic/inner_product.inl
new file mode 100644
index 0000000000..b6a339ea15
--- /dev/null
+++ b/compat/thrust/system/detail/generic/inner_product.inl
@@ -0,0 +1,70 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/inner_product.h>
+#include <thrust/functional.h>
+#include <thrust/detail/internal_functional.h>
+#include <thrust/transform_reduce.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+
+template<typename DerivedPolicy, typename InputIterator1, typename InputIterator2, typename OutputType>
+OutputType inner_product(thrust::execution_policy<DerivedPolicy> &exec,
+                         InputIterator1 first1,
+                         InputIterator1 last1,
+                         InputIterator2 first2,
+                         OutputType init)
+{
+  thrust::plus<OutputType>       binary_op1;
+  thrust::multiplies<OutputType> binary_op2;
+  return thrust::inner_product(exec, first1, last1, first2, init, binary_op1, binary_op2);
+} // end inner_product()
+
+
+template<typename DerivedPolicy, typename InputIterator1, typename InputIterator2, typename OutputType, typename BinaryFunction1, typename BinaryFunction2>
+OutputType inner_product(thrust::execution_policy<DerivedPolicy> &exec,
+                         InputIterator1 first1,
+                         InputIterator1 last1,
+                         InputIterator2 first2,
+                         OutputType init, 
+                         BinaryFunction1 binary_op1,
+                         BinaryFunction2 binary_op2)
+{
+  typedef thrust::zip_iterator<thrust::tuple<InputIterator1,InputIterator2> > ZipIter;
+
+  ZipIter first = thrust::make_zip_iterator(thrust::make_tuple(first1,first2));
+
+  // only the first iterator in the tuple is relevant for the purposes of last
+  ZipIter last  = thrust::make_zip_iterator(thrust::make_tuple(last1, first2));
+
+  return thrust::transform_reduce(exec, first, last, thrust::detail::zipped_binary_op<OutputType,BinaryFunction2>(binary_op2), init, binary_op1);
+} // end inner_product()
+
+
+} // end generic
+} // end detail
+} // end system
+} // end thrust
+
diff --git a/compat/thrust/system/detail/generic/logical.h b/compat/thrust/system/detail/generic/logical.h
new file mode 100644
index 0000000000..e0d01e30a5
--- /dev/null
+++ b/compat/thrust/system/detail/generic/logical.h
@@ -0,0 +1,56 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/tag.h>
+#include <thrust/detail/internal_functional.h>
+#include <thrust/find.h>
+#include <thrust/logical.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+template <typename ExecutionPolicy, typename InputIterator, typename Predicate>
+bool all_of(thrust::execution_policy<ExecutionPolicy> &exec, InputIterator first, InputIterator last, Predicate pred)
+{
+  return thrust::find_if(exec, first, last, thrust::detail::not1(pred)) == last;
+}
+
+template <typename ExecutionPolicy, typename InputIterator, typename Predicate>
+bool any_of(thrust::execution_policy<ExecutionPolicy> &exec, InputIterator first, InputIterator last, Predicate pred)
+{
+  return thrust::find_if(exec, first, last, pred) != last;
+}
+
+template <typename ExecutionPolicy, typename InputIterator, typename Predicate>
+bool none_of(thrust::execution_policy<ExecutionPolicy> &exec, InputIterator first, InputIterator last, Predicate pred)
+{
+  return !thrust::any_of(exec, first, last, pred);
+}
+
+} // end generic
+} // end detail
+} // end system
+} // end thrust
+
diff --git a/compat/thrust/system/detail/generic/memory.h b/compat/thrust/system/detail/generic/memory.h
new file mode 100644
index 0000000000..c0fe623ac4
--- /dev/null
+++ b/compat/thrust/system/detail/generic/memory.h
@@ -0,0 +1,67 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file generic/memory.h
+ *  \brief Generic implementation of memory functions.
+ *         Calling some of these is an error. They have no implementation.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/execution_policy.h>
+#include <thrust/system/detail/generic/tag.h>
+#include <thrust/detail/type_traits.h>
+#include <thrust/detail/pointer.h>
+#include <thrust/pair.h>
+#include <thrust/system/detail/generic/type_traits.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+template<typename DerivedPolicy, typename Size> void malloc(thrust::execution_policy<DerivedPolicy> &, Size);
+
+template<typename T, typename DerivedPolicy>
+thrust::pointer<T,DerivedPolicy> malloc(thrust::execution_policy<DerivedPolicy> &s, std::size_t n);
+
+template<typename DerivedPolicy, typename Pointer> void free(thrust::execution_policy<DerivedPolicy> &, Pointer);
+
+template<typename Pointer1, typename Pointer2>
+__host__ __device__
+void assign_value(tag, Pointer1, Pointer2);
+
+template<typename DerivedPolicy, typename Pointer>
+__host__ __device__
+void get_value(thrust::execution_policy<DerivedPolicy> &, Pointer);
+
+template<typename Pointer1, typename Pointer2>
+__host__ __device__
+void iter_swap(tag, Pointer1, Pointer2);
+
+} // end generic
+} // end detail
+} // end system
+} // end thrust
+
+#include <thrust/system/detail/generic/memory.inl>
+
diff --git a/compat/thrust/system/detail/generic/memory.inl b/compat/thrust/system/detail/generic/memory.inl
new file mode 100644
index 0000000000..f89a763a62
--- /dev/null
+++ b/compat/thrust/system/detail/generic/memory.inl
@@ -0,0 +1,92 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/type_traits/pointer_traits.h>
+#include <thrust/system/detail/generic/memory.h>
+#include <thrust/system/detail/generic/select_system.h>
+#include <thrust/system/detail/adl/malloc_and_free.h>
+#include <thrust/detail/static_assert.h>
+#include <thrust/detail/malloc_and_free.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+
+template<typename DerivedPolicy, typename Size>
+  void malloc(thrust::execution_policy<DerivedPolicy> &, Size)
+{
+  // unimplemented
+  THRUST_STATIC_ASSERT( (thrust::detail::depend_on_instantiation<Size, false>::value) );
+}
+
+
+template<typename T, typename DerivedPolicy>
+  thrust::pointer<T,DerivedPolicy>
+    malloc(thrust::execution_policy<DerivedPolicy> &exec, std::size_t n)
+{
+  thrust::pointer<void,DerivedPolicy> void_ptr = thrust::malloc(exec, sizeof(T) * n);
+
+  return pointer<T,DerivedPolicy>(static_cast<T*>(void_ptr.get()));
+} // end malloc()
+
+
+template<typename DerivedPolicy, typename Pointer>
+  void free(thrust::execution_policy<DerivedPolicy> &, Pointer)
+{
+  // unimplemented
+  THRUST_STATIC_ASSERT( (thrust::detail::depend_on_instantiation<Pointer, false>::value) );
+}
+
+
+template<typename DerivedPolicy, typename Pointer1, typename Pointer2>
+__host__ __device__
+void assign_value(thrust::execution_policy<DerivedPolicy> &, Pointer1, Pointer2)
+{
+  // unimplemented
+  THRUST_STATIC_ASSERT( (thrust::detail::depend_on_instantiation<Pointer1, false>::value) );
+}
+
+
+template<typename DerivedPolicy, typename Pointer>
+__host__ __device__
+void get_value(thrust::execution_policy<DerivedPolicy> &, Pointer)
+{
+  // unimplemented
+  THRUST_STATIC_ASSERT( (thrust::detail::depend_on_instantiation<Pointer, false>::value) );
+}
+
+
+template<typename Pointer1, typename Pointer2>
+__host__ __device__
+void iter_swap(tag, Pointer1, Pointer2)
+{
+  // unimplemented
+  THRUST_STATIC_ASSERT( (thrust::detail::depend_on_instantiation<Pointer1, false>::value) );
+}
+
+
+} // end generic
+} // end detail
+} // end system
+} // end thrust
+
diff --git a/compat/thrust/system/detail/generic/merge.h b/compat/thrust/system/detail/generic/merge.h
new file mode 100644
index 0000000000..5f0b99640d
--- /dev/null
+++ b/compat/thrust/system/detail/generic/merge.h
@@ -0,0 +1,87 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/tag.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+
+// XXX calling this function is an error; there is no implementation
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename StrictWeakOrdering>
+  OutputIterator merge(thrust::execution_policy<DerivedPolicy> &exec,
+                       InputIterator1 first1,
+                       InputIterator1 last1,
+                       InputIterator2 first2,
+                       InputIterator2 last2,
+                       OutputIterator result,
+                       StrictWeakOrdering comp);
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator>
+  OutputIterator merge(thrust::execution_policy<DerivedPolicy> &exec,
+                       InputIterator1 first1,
+                       InputIterator1 last1,
+                       InputIterator2 first2,
+                       InputIterator2 last2,
+                       OutputIterator result);
+
+
+template<typename DerivedPolicy, typename InputIterator1, typename InputIterator2, typename InputIterator3, typename InputIterator4, typename OutputIterator1, typename OutputIterator2, typename Compare>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    merge_by_key(thrust::execution_policy<DerivedPolicy> &exec,
+                 InputIterator1 keys_first1, InputIterator1 keys_last1,
+                 InputIterator2 keys_first2, InputIterator2 keys_last2,
+                 InputIterator3 values_first1, InputIterator4 values_first2,
+                 OutputIterator1 keys_result,
+                 OutputIterator2 values_result,
+                 Compare comp);
+
+
+template<typename DerivedPolicy, typename InputIterator1, typename InputIterator2, typename InputIterator3, typename InputIterator4, typename OutputIterator1, typename OutputIterator2>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    merge_by_key(thrust::execution_policy<DerivedPolicy> &exec,
+                 InputIterator1 keys_first1, InputIterator1 keys_last1,
+                 InputIterator2 keys_first2, InputIterator2 keys_last2,
+                 InputIterator3 values_first1, InputIterator4 values_first2,
+                 OutputIterator1 keys_result,
+                 OutputIterator2 values_result);
+
+
+} // end namespace generic
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/detail/generic/merge.inl>
+
diff --git a/compat/thrust/system/detail/generic/merge.inl b/compat/thrust/system/detail/generic/merge.inl
new file mode 100644
index 0000000000..b913611168
--- /dev/null
+++ b/compat/thrust/system/detail/generic/merge.inl
@@ -0,0 +1,125 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/static_assert.h>
+#include <thrust/system/detail/generic/merge.h>
+#include <thrust/merge.h>
+#include <thrust/functional.h>
+#include <thrust/iterator/zip_iterator.h>
+#include <thrust/detail/internal_functional.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename StrictWeakOrdering>
+  OutputIterator merge(thrust::execution_policy<DerivedPolicy> &exec,
+                       InputIterator1 first1,
+                       InputIterator1 last1,
+                       InputIterator2 first2,
+                       InputIterator2 last2,
+                       OutputIterator result,
+                       StrictWeakOrdering comp)
+{
+  // unimplemented
+  THRUST_STATIC_ASSERT( (thrust::detail::depend_on_instantiation<InputIterator1, false>::value) );
+  return result;
+} // end merge()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator>
+  OutputIterator merge(thrust::execution_policy<DerivedPolicy> &exec,
+                       InputIterator1 first1,
+                       InputIterator1 last1,
+                       InputIterator2 first2,
+                       InputIterator2 last2,
+                       OutputIterator result)
+{
+  typedef typename thrust::iterator_value<InputIterator1>::type value_type;
+  return thrust::merge(exec,first1,last1,first2,last2,result,thrust::less<value_type>());
+} // end merge()
+
+
+template<typename DerivedPolicy, typename InputIterator1, typename InputIterator2, typename InputIterator3, typename InputIterator4, typename OutputIterator1, typename OutputIterator2, typename Compare>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    merge_by_key(thrust::execution_policy<DerivedPolicy> &exec,
+                 InputIterator1 keys_first1, InputIterator1 keys_last1,
+                 InputIterator2 keys_first2, InputIterator2 keys_last2,
+                 InputIterator3 values_first1, InputIterator4 values_first2,
+                 OutputIterator1 keys_result,
+                 OutputIterator2 values_result,
+                 Compare comp)
+{
+  typedef thrust::tuple<InputIterator1, InputIterator3>   iterator_tuple1;
+  typedef thrust::tuple<InputIterator2, InputIterator4>   iterator_tuple2;
+  typedef thrust::tuple<OutputIterator1, OutputIterator2> iterator_tuple3;
+
+  typedef thrust::zip_iterator<iterator_tuple1> zip_iterator1;
+  typedef thrust::zip_iterator<iterator_tuple2> zip_iterator2;
+  typedef thrust::zip_iterator<iterator_tuple3> zip_iterator3;
+
+  zip_iterator1 zipped_first1 = thrust::make_zip_iterator(thrust::make_tuple(keys_first1, values_first1));
+  zip_iterator1 zipped_last1  = thrust::make_zip_iterator(thrust::make_tuple(keys_last1, values_first1));
+
+  zip_iterator2 zipped_first2 = thrust::make_zip_iterator(thrust::make_tuple(keys_first2, values_first2));
+  zip_iterator2 zipped_last2  = thrust::make_zip_iterator(thrust::make_tuple(keys_last2, values_first2));
+
+  zip_iterator3 zipped_result = thrust::make_zip_iterator(thrust::make_tuple(keys_result, values_result));
+
+  thrust::detail::compare_first<Compare> comp_first(comp);
+
+  iterator_tuple3 result = thrust::merge(exec, zipped_first1, zipped_last1, zipped_first2, zipped_last2, zipped_result, comp_first).get_iterator_tuple();
+
+  return thrust::make_pair(thrust::get<0>(result), thrust::get<1>(result));
+} // end merge_by_key()
+
+
+template<typename DerivedPolicy, typename InputIterator1, typename InputIterator2, typename InputIterator3, typename InputIterator4, typename OutputIterator1, typename OutputIterator2>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    merge_by_key(thrust::execution_policy<DerivedPolicy> &exec,
+                 InputIterator1 keys_first1, InputIterator1 keys_last1,
+                 InputIterator2 keys_first2, InputIterator2 keys_last2,
+                 InputIterator3 values_first1, InputIterator4 values_first2,
+                 OutputIterator1 keys_result,
+                 OutputIterator2 values_result)
+{
+  typedef typename thrust::iterator_value<InputIterator1>::type value_type;
+  return thrust::merge_by_key(exec, keys_first1, keys_last1, keys_first2, keys_last2, values_first1, values_first2, keys_result, values_result, thrust::less<value_type>());
+} // end merge_by_key()
+
+
+} // end namespace generic
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/detail/generic/mismatch.h b/compat/thrust/system/detail/generic/mismatch.h
new file mode 100644
index 0000000000..dc581ffbee
--- /dev/null
+++ b/compat/thrust/system/detail/generic/mismatch.h
@@ -0,0 +1,56 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/tag.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+
+template<typename DerivedPolicy, typename InputIterator1, typename InputIterator2>
+  thrust::pair<InputIterator1, InputIterator2>
+    mismatch(thrust::execution_policy<DerivedPolicy> &exec,
+             InputIterator1 first1,
+             InputIterator1 last1,
+             InputIterator2 first2);
+
+
+template<typename DerivedPolicy, typename InputIterator1, typename InputIterator2, typename BinaryPredicate>
+  thrust::pair<InputIterator1, InputIterator2>
+    mismatch(thrust::execution_policy<DerivedPolicy> &exec,
+             InputIterator1 first1,
+             InputIterator1 last1,
+             InputIterator2 first2,
+             BinaryPredicate pred);
+
+
+} // end namespace generic
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/detail/generic/mismatch.inl>
+
diff --git a/compat/thrust/system/detail/generic/mismatch.inl b/compat/thrust/system/detail/generic/mismatch.inl
new file mode 100644
index 0000000000..923c27f71b
--- /dev/null
+++ b/compat/thrust/system/detail/generic/mismatch.inl
@@ -0,0 +1,70 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/mismatch.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/detail/internal_functional.h>
+#include <thrust/find.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+template<typename DerivedPolicy, typename InputIterator1, typename InputIterator2>
+  thrust::pair<InputIterator1, InputIterator2>
+    mismatch(thrust::execution_policy<DerivedPolicy> &exec,
+             InputIterator1 first1,
+             InputIterator1 last1,
+             InputIterator2 first2)
+{
+  typedef typename thrust::iterator_value<InputIterator1>::type InputType1;
+  
+  // XXX use a placeholder expression here
+  return thrust::mismatch(exec, first1, last1, first2, thrust::detail::equal_to<InputType1>());
+} // end mismatch()
+
+template<typename DerivedPolicy, typename InputIterator1, typename InputIterator2, typename BinaryPredicate>
+  thrust::pair<InputIterator1, InputIterator2>
+    mismatch(thrust::execution_policy<DerivedPolicy> &exec,
+             InputIterator1 first1,
+             InputIterator1 last1,
+             InputIterator2 first2,
+             BinaryPredicate pred)
+{
+  // Contributed by Erich Elsen
+  typedef thrust::tuple<InputIterator1,InputIterator2> IteratorTuple;
+  typedef thrust::zip_iterator<IteratorTuple>          ZipIterator;
+  
+  ZipIterator zipped_first = thrust::make_zip_iterator(thrust::make_tuple(first1,first2));
+  ZipIterator zipped_last  = thrust::make_zip_iterator(thrust::make_tuple(last1, first2));
+  
+  ZipIterator result = thrust::find_if_not(exec, zipped_first, zipped_last, thrust::detail::tuple_binary_predicate<BinaryPredicate>(pred));
+  
+  return thrust::make_pair(thrust::get<0>(result.get_iterator_tuple()),
+                           thrust::get<1>(result.get_iterator_tuple()));
+} // end mismatch()
+
+} // end generic
+} // end detail
+} // end system
+} // end thrust
+
diff --git a/compat/thrust/system/detail/generic/partition.h b/compat/thrust/system/detail/generic/partition.h
new file mode 100644
index 0000000000..63daa1d1c1
--- /dev/null
+++ b/compat/thrust/system/detail/generic/partition.h
@@ -0,0 +1,150 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file partition.h
+ *  \brief Generic implementations of partition functions.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/tag.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+template<typename ExecutionPolicy,
+         typename ForwardIterator,
+         typename Predicate>
+  ForwardIterator stable_partition(thrust::execution_policy<ExecutionPolicy> &exec,
+                                   ForwardIterator first,
+                                   ForwardIterator last,
+                                   Predicate pred);
+
+template<typename ExecutionPolicy,
+         typename ForwardIterator,
+         typename InputIterator,
+         typename Predicate>
+  ForwardIterator stable_partition(thrust::execution_policy<ExecutionPolicy> &exec,
+                                   ForwardIterator first,
+                                   ForwardIterator last,
+                                   InputIterator stencil,
+                                   Predicate pred);
+
+template<typename ExecutionPolicy,
+         typename InputIterator,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename Predicate>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    stable_partition_copy(thrust::execution_policy<ExecutionPolicy> &exec,
+                          InputIterator first,
+                          InputIterator last,
+                          OutputIterator1 out_true,
+                          OutputIterator2 out_false,
+                          Predicate pred);
+
+template<typename ExecutionPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename Predicate>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    stable_partition_copy(thrust::execution_policy<ExecutionPolicy> &exec,
+                          InputIterator1 first,
+                          InputIterator1 last,
+                          InputIterator2 stencil,
+                          OutputIterator1 out_true,
+                          OutputIterator2 out_false,
+                          Predicate pred);
+
+template<typename ExecutionPolicy,
+         typename ForwardIterator,
+         typename Predicate>
+  ForwardIterator partition(thrust::execution_policy<ExecutionPolicy> &exec,
+                            ForwardIterator first,
+                            ForwardIterator last,
+                            Predicate pred);
+
+template<typename ExecutionPolicy,
+         typename ForwardIterator,
+         typename InputIterator,
+         typename Predicate>
+  ForwardIterator partition(thrust::execution_policy<ExecutionPolicy> &exec,
+                            ForwardIterator first,
+                            ForwardIterator last,
+                            InputIterator stencil,
+                            Predicate pred);
+
+template<typename ExecutionPolicy,
+         typename InputIterator,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename Predicate>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    partition_copy(thrust::execution_policy<ExecutionPolicy> &exec,
+                   InputIterator first,
+                   InputIterator last,
+                   OutputIterator1 out_true,
+                   OutputIterator2 out_false,
+                   Predicate pred);
+
+template<typename ExecutionPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename Predicate>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    partition_copy(thrust::execution_policy<ExecutionPolicy> &exec,
+                   InputIterator1 first,
+                   InputIterator1 last,
+                   InputIterator2 stencil,
+                   OutputIterator1 out_true,
+                   OutputIterator2 out_false,
+                   Predicate pred);
+
+template<typename ExecutionPolicy,
+         typename ForwardIterator,
+         typename Predicate>
+  ForwardIterator partition_point(thrust::execution_policy<ExecutionPolicy> &exec,
+                                  ForwardIterator first,
+                                  ForwardIterator last,
+                                  Predicate pred);
+
+template<typename ExecutionPolicy,
+         typename InputIterator,
+         typename Predicate>
+  bool is_partitioned(thrust::execution_policy<ExecutionPolicy> &exec,
+                      InputIterator first,
+                      InputIterator last,
+                      Predicate pred);
+
+} // end namespace generic
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/detail/generic/partition.inl>
+
diff --git a/compat/thrust/system/detail/generic/partition.inl b/compat/thrust/system/detail/generic/partition.inl
new file mode 100644
index 0000000000..3298afc6f5
--- /dev/null
+++ b/compat/thrust/system/detail/generic/partition.inl
@@ -0,0 +1,238 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/partition.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/pair.h>
+
+#include <thrust/remove.h>
+#include <thrust/count.h>
+#include <thrust/advance.h>
+#include <thrust/partition.h>
+#include <thrust/sort.h>
+#include <thrust/iterator/transform_iterator.h>
+
+#include <thrust/detail/internal_functional.h>
+#include <thrust/detail/temporary_array.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+
+template<typename DerivedPolicy,
+         typename ForwardIterator,
+         typename Predicate>
+  ForwardIterator stable_partition(thrust::execution_policy<DerivedPolicy> &exec,
+                                   ForwardIterator first,
+                                   ForwardIterator last,
+                                   Predicate pred)
+{
+  typedef typename thrust::iterator_traits<ForwardIterator>::value_type InputType;
+
+  // copy input to temp buffer
+  thrust::detail::temporary_array<InputType,DerivedPolicy> temp(exec, first, last);
+
+  // count the size of the true partition
+  typename thrust::iterator_difference<ForwardIterator>::type num_true = thrust::count_if(exec, first,last,pred);
+
+  // point to the beginning of the false partition
+  ForwardIterator out_false = first;
+  thrust::advance(out_false, num_true);
+
+  return thrust::stable_partition_copy(exec, temp.begin(), temp.end(), first, out_false, pred).first;
+} // end stable_partition()
+
+
+template<typename DerivedPolicy,
+         typename ForwardIterator,
+         typename InputIterator,
+         typename Predicate>
+  ForwardIterator stable_partition(thrust::execution_policy<DerivedPolicy> &exec,
+                                   ForwardIterator first,
+                                   ForwardIterator last,
+                                   InputIterator stencil,
+                                   Predicate pred)
+{
+  typedef typename thrust::iterator_traits<ForwardIterator>::value_type InputType;
+
+  // copy input to temp buffer
+  thrust::detail::temporary_array<InputType,DerivedPolicy> temp(exec, first, last);
+
+  // count the size of the true partition
+  InputIterator stencil_last = stencil;
+  thrust::advance(stencil_last, temp.size());
+  typename thrust::iterator_difference<InputIterator>::type num_true = thrust::count_if(exec, stencil, stencil_last, pred);
+
+  // point to the beginning of the false partition
+  ForwardIterator out_false = first;
+  thrust::advance(out_false, num_true);
+
+  return thrust::stable_partition_copy(exec, temp.begin(), temp.end(), stencil, first, out_false, pred).first;
+} // end stable_partition()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename Predicate>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    stable_partition_copy(thrust::execution_policy<DerivedPolicy> &exec,
+                          InputIterator first,
+                          InputIterator last,
+                          OutputIterator1 out_true,
+                          OutputIterator2 out_false,
+                          Predicate pred)
+{
+  thrust::detail::unary_negate<Predicate> not_pred(pred);
+
+  // remove_copy_if the true partition to out_true
+  OutputIterator1 end_of_true_partition = thrust::remove_copy_if(exec, first, last, out_true, not_pred);
+
+  // remove_copy_if the false partition to out_false
+  OutputIterator2 end_of_false_partition = thrust::remove_copy_if(exec, first, last, out_false, pred);
+
+  return thrust::make_pair(end_of_true_partition, end_of_false_partition);
+} // end stable_partition_copy()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename Predicate>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    stable_partition_copy(thrust::execution_policy<DerivedPolicy> &exec,
+                          InputIterator1 first,
+                          InputIterator1 last,
+                          InputIterator2 stencil,
+                          OutputIterator1 out_true,
+                          OutputIterator2 out_false,
+                          Predicate pred)
+{
+  thrust::detail::unary_negate<Predicate> not_pred(pred);
+
+  // remove_copy_if the true partition to out_true
+  OutputIterator1 end_of_true_partition = thrust::remove_copy_if(exec, first, last, stencil, out_true, not_pred);
+
+  // remove_copy_if the false partition to out_false
+  OutputIterator2 end_of_false_partition = thrust::remove_copy_if(exec, first, last, stencil, out_false, pred);
+
+  return thrust::make_pair(end_of_true_partition, end_of_false_partition);
+} // end stable_partition_copy()
+
+
+template<typename DerivedPolicy,
+         typename ForwardIterator,
+         typename Predicate>
+  ForwardIterator partition(thrust::execution_policy<DerivedPolicy> &exec,
+                            ForwardIterator first,
+                            ForwardIterator last,
+                            Predicate pred)
+{
+  return thrust::stable_partition(exec, first, last, pred);
+} // end partition()
+
+
+template<typename DerivedPolicy,
+         typename ForwardIterator,
+         typename InputIterator,
+         typename Predicate>
+  ForwardIterator partition(thrust::execution_policy<DerivedPolicy> &exec,
+                            ForwardIterator first,
+                            ForwardIterator last,
+                            InputIterator stencil,
+                            Predicate pred)
+{
+  return thrust::stable_partition(exec, first, last, stencil, pred);
+} // end partition()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename Predicate>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    partition_copy(thrust::execution_policy<DerivedPolicy> &exec,
+                   InputIterator first,
+                   InputIterator last,
+                   OutputIterator1 out_true,
+                   OutputIterator2 out_false,
+                   Predicate pred)
+{
+  return thrust::stable_partition_copy(exec,first,last,out_true,out_false,pred);
+} // end partition_copy()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename Predicate>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    partition_copy(thrust::execution_policy<DerivedPolicy> &exec,
+                   InputIterator1 first,
+                   InputIterator1 last,
+                   InputIterator2 stencil,
+                   OutputIterator1 out_true,
+                   OutputIterator2 out_false,
+                   Predicate pred)
+{
+  return thrust::stable_partition_copy(exec,first,last,stencil,out_true,out_false,pred);
+} // end partition_copy()
+
+
+template<typename DerivedPolicy,
+         typename ForwardIterator,
+         typename Predicate>
+  ForwardIterator partition_point(thrust::execution_policy<DerivedPolicy> &exec,
+                                  ForwardIterator first,
+                                  ForwardIterator last,
+                                  Predicate pred)
+{
+  return thrust::find_if_not(exec, first, last, pred);
+} // end partition_point()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename Predicate>
+  bool is_partitioned(thrust::execution_policy<DerivedPolicy> &exec,
+                      InputIterator first,
+                      InputIterator last,
+                      Predicate pred)
+{
+  return thrust::is_sorted(exec,
+                           thrust::make_transform_iterator(first, thrust::detail::not1(pred)),
+                           thrust::make_transform_iterator(last,  thrust::detail::not1(pred)));
+} // end is_partitioned()
+
+
+} // end namespace generic
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/detail/generic/reduce.h b/compat/thrust/system/detail/generic/reduce.h
new file mode 100644
index 0000000000..2811df164f
--- /dev/null
+++ b/compat/thrust/system/detail/generic/reduce.h
@@ -0,0 +1,52 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/tag.h>
+#include <thrust/iterator/iterator_traits.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+template<typename DerivedPolicy, typename InputIterator>
+  typename thrust::iterator_traits<InputIterator>::value_type
+    reduce(thrust::execution_policy<DerivedPolicy> &exec, InputIterator first, InputIterator last);
+
+template<typename DerivedPolicy, typename InputIterator, typename T>
+  T reduce(thrust::execution_policy<DerivedPolicy> &exec, InputIterator first, InputIterator last, T init);
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename T,
+         typename BinaryFunction>
+  T reduce(thrust::execution_policy<DerivedPolicy> &exec, InputIterator first, InputIterator last, T init, BinaryFunction binary_op);
+
+} // end namespace generic
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/detail/generic/reduce.inl>
+
diff --git a/compat/thrust/system/detail/generic/reduce.inl b/compat/thrust/system/detail/generic/reduce.inl
new file mode 100644
index 0000000000..8f52385163
--- /dev/null
+++ b/compat/thrust/system/detail/generic/reduce.inl
@@ -0,0 +1,74 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/reduce.h>
+#include <thrust/system/detail/generic/reduce.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/functional.h>
+#include <thrust/detail/static_assert.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+
+template<typename ExecutionPolicy, typename InputIterator>
+  typename thrust::iterator_traits<InputIterator>::value_type
+    reduce(thrust::execution_policy<ExecutionPolicy> &exec, InputIterator first, InputIterator last)
+{
+  typedef typename thrust::iterator_value<InputIterator>::type InputType;
+
+  // use InputType(0) as init by default
+  return thrust::reduce(exec, first, last, InputType(0));
+} // end reduce()
+
+
+template<typename ExecutionPolicy, typename InputIterator, typename T>
+  T reduce(thrust::execution_policy<ExecutionPolicy> &exec, InputIterator first, InputIterator last, T init)
+{
+  // use plus<T> by default
+  return thrust::reduce(exec, first, last, init, thrust::plus<T>());
+} // end reduce()
+
+
+template<typename ExecutionPolicy,
+         typename RandomAccessIterator,
+         typename OutputType,
+         typename BinaryFunction>
+  OutputType reduce(thrust::execution_policy<ExecutionPolicy> &exec,
+                    RandomAccessIterator first,
+                    RandomAccessIterator last,
+                    OutputType init,
+                    BinaryFunction binary_op)
+{
+  // unimplemented
+  THRUST_STATIC_ASSERT( (thrust::detail::depend_on_instantiation<RandomAccessIterator, false>::value) );
+  return OutputType();
+} // end reduce()
+
+
+} // end namespace generic
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/detail/generic/reduce_by_key.h b/compat/thrust/system/detail/generic/reduce_by_key.h
new file mode 100644
index 0000000000..c6064ab53e
--- /dev/null
+++ b/compat/thrust/system/detail/generic/reduce_by_key.h
@@ -0,0 +1,86 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/tag.h>
+#include <thrust/iterator/iterator_traits.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator1,
+         typename OutputIterator2>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    reduce_by_key(thrust::execution_policy<DerivedPolicy> &exec,
+                  InputIterator1 keys_first, 
+                  InputIterator1 keys_last,
+                  InputIterator2 values_first,
+                  OutputIterator1 keys_output,
+                  OutputIterator2 values_output);
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename BinaryPredicate>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    reduce_by_key(thrust::execution_policy<DerivedPolicy> &exec,
+                  InputIterator1 keys_first, 
+                  InputIterator1 keys_last,
+                  InputIterator2 values_first,
+                  OutputIterator1 keys_output,
+                  OutputIterator2 values_output,
+                  BinaryPredicate binary_pred);
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename BinaryPredicate,
+         typename BinaryFunction>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    reduce_by_key(thrust::execution_policy<DerivedPolicy> &exec,
+                  InputIterator1 keys_first, 
+                  InputIterator1 keys_last,
+                  InputIterator2 values_first,
+                  OutputIterator1 keys_output,
+                  OutputIterator2 values_output,
+                  BinaryPredicate binary_pred,
+                  BinaryFunction binary_op);
+
+
+} // end namespace generic
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/detail/generic/reduce_by_key.inl>
+
diff --git a/compat/thrust/system/detail/generic/reduce_by_key.inl b/compat/thrust/system/detail/generic/reduce_by_key.inl
new file mode 100644
index 0000000000..2ca21a5aab
--- /dev/null
+++ b/compat/thrust/system/detail/generic/reduce_by_key.inl
@@ -0,0 +1,212 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file reduce_by_key.inl
+ *  \brief Inline file for reduce_by_key.h.
+ */
+
+#pragma once
+
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/iterator/detail/minimum_system.h>
+#include <thrust/detail/type_traits.h>
+#include <thrust/detail/type_traits/iterator/is_output_iterator.h>
+#include <thrust/detail/type_traits/function_traits.h>
+#include <thrust/transform.h>
+#include <thrust/scatter.h>
+#include <thrust/iterator/zip_iterator.h>
+#include <limits>
+
+#include <thrust/detail/internal_functional.h>
+#include <thrust/scan.h>
+#include <thrust/detail/temporary_array.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+namespace detail
+{
+
+template <typename ValueType, typename TailFlagType, typename AssociativeOperator>
+struct reduce_by_key_functor
+{
+    AssociativeOperator binary_op;
+
+    typedef typename thrust::tuple<ValueType, TailFlagType> result_type;
+
+    __host__ __device__
+    reduce_by_key_functor(AssociativeOperator _binary_op) : binary_op(_binary_op) {}
+
+    __host__ __device__
+    result_type operator()(result_type a, result_type b)
+    {
+        return result_type(thrust::get<1>(b) ? thrust::get<0>(b) : binary_op(thrust::get<0>(a), thrust::get<0>(b)),
+                           thrust::get<1>(a) | thrust::get<1>(b));
+    }
+};
+
+} // end namespace detail
+
+
+template<typename ExecutionPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename BinaryPredicate,
+         typename BinaryFunction>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    reduce_by_key(thrust::execution_policy<ExecutionPolicy> &exec,
+                  InputIterator1 keys_first, 
+                  InputIterator1 keys_last,
+                  InputIterator2 values_first,
+                  OutputIterator1 keys_output,
+                  OutputIterator2 values_output,
+                  BinaryPredicate binary_pred,
+                  BinaryFunction binary_op)
+{
+    typedef typename thrust::iterator_traits<InputIterator1>::difference_type difference_type;
+    typedef typename thrust::iterator_traits<InputIterator1>::value_type  KeyType;
+
+    typedef unsigned int FlagType;  // TODO use difference_type
+
+    // the pseudocode for deducing the type of the temporary used below:
+    // 
+    // if BinaryFunction is AdaptableBinaryFunction
+    //   TemporaryType = AdaptableBinaryFunction::result_type
+    // else if OutputIterator2 is a "pure" output iterator
+    //   TemporaryType = InputIterator2::value_type
+    // else
+    //   TemporaryType = OutputIterator2::value_type
+    //
+    // XXX upon c++0x, TemporaryType needs to be:
+    // result_of<BinaryFunction>::type
+
+    typedef typename thrust::detail::eval_if<
+      thrust::detail::has_result_type<BinaryFunction>::value,
+      thrust::detail::result_type<BinaryFunction>,
+      thrust::detail::eval_if<
+        thrust::detail::is_output_iterator<OutputIterator2>::value,
+        thrust::iterator_value<InputIterator2>,
+        thrust::iterator_value<OutputIterator2>
+      >
+    >::type ValueType;
+
+    if (keys_first == keys_last)
+        return thrust::make_pair(keys_output, values_output);
+
+    // input size
+    difference_type n = keys_last - keys_first;
+
+    InputIterator2 values_last = values_first + n;
+    
+    // compute head flags
+    thrust::detail::temporary_array<FlagType,ExecutionPolicy> head_flags(exec, n);
+    thrust::transform(exec, keys_first, keys_last - 1, keys_first + 1, head_flags.begin() + 1, thrust::detail::not2(binary_pred));
+    head_flags[0] = 1;
+
+    // compute tail flags
+    thrust::detail::temporary_array<FlagType,ExecutionPolicy> tail_flags(exec, n); //COPY INSTEAD OF TRANSFORM
+    thrust::transform(exec, keys_first, keys_last - 1, keys_first + 1, tail_flags.begin(), thrust::detail::not2(binary_pred));
+    tail_flags[n-1] = 1;
+
+    // scan the values by flag
+    thrust::detail::temporary_array<ValueType,ExecutionPolicy> scanned_values(exec, n);
+    thrust::detail::temporary_array<FlagType,ExecutionPolicy>  scanned_tail_flags(exec, n);
+    
+    thrust::inclusive_scan
+        (exec,
+         thrust::make_zip_iterator(thrust::make_tuple(values_first,           head_flags.begin())),
+         thrust::make_zip_iterator(thrust::make_tuple(values_last,            head_flags.end())),
+         thrust::make_zip_iterator(thrust::make_tuple(scanned_values.begin(), scanned_tail_flags.begin())),
+         detail::reduce_by_key_functor<ValueType, FlagType, BinaryFunction>(binary_op));
+
+    thrust::exclusive_scan(exec, tail_flags.begin(), tail_flags.end(), scanned_tail_flags.begin(), FlagType(0), thrust::plus<FlagType>());
+
+    // number of unique keys
+    FlagType N = scanned_tail_flags[n - 1] + 1;
+    
+    // scatter the keys and accumulated values    
+    thrust::scatter_if(exec, keys_first,            keys_last,             scanned_tail_flags.begin(), head_flags.begin(), keys_output);
+    thrust::scatter_if(exec, scanned_values.begin(), scanned_values.end(), scanned_tail_flags.begin(), tail_flags.begin(), values_output);
+
+    return thrust::make_pair(keys_output + N, values_output + N); 
+} // end reduce_by_key()
+
+
+template<typename ExecutionPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator1,
+         typename OutputIterator2>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    reduce_by_key(thrust::execution_policy<ExecutionPolicy> &exec,
+                  InputIterator1 keys_first, 
+                  InputIterator1 keys_last,
+                  InputIterator2 values_first,
+                  OutputIterator1 keys_output,
+                  OutputIterator2 values_output)
+{
+  typedef typename thrust::iterator_value<InputIterator1>::type KeyType;
+
+  // use equal_to<KeyType> as default BinaryPredicate
+  return thrust::reduce_by_key(exec, keys_first, keys_last, values_first, keys_output, values_output, thrust::equal_to<KeyType>());
+} // end reduce_by_key()
+
+
+template<typename ExecutionPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename BinaryPredicate>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    reduce_by_key(thrust::execution_policy<ExecutionPolicy> &exec,
+                  InputIterator1 keys_first, 
+                  InputIterator1 keys_last,
+                  InputIterator2 values_first,
+                  OutputIterator1 keys_output,
+                  OutputIterator2 values_output,
+                  BinaryPredicate binary_pred)
+{
+  typedef typename thrust::detail::eval_if<
+    thrust::detail::is_output_iterator<OutputIterator2>::value,
+    thrust::iterator_value<InputIterator2>,
+    thrust::iterator_value<OutputIterator2>
+  >::type T;
+
+  // use plus<T> as default BinaryFunction
+  return thrust::reduce_by_key(exec,
+                               keys_first, keys_last, 
+                               values_first,
+                               keys_output,
+                               values_output,
+                               binary_pred,
+                               thrust::plus<T>());
+} // end reduce_by_key()
+
+
+} // end namespace generic
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/detail/generic/remove.h b/compat/thrust/system/detail/generic/remove.h
new file mode 100644
index 0000000000..e23673574e
--- /dev/null
+++ b/compat/thrust/system/detail/generic/remove.h
@@ -0,0 +1,100 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file remove.h
+ *  \brief Generic implementations of remove functions.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/tag.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+template<typename DerivedPolicy,
+         typename ForwardIterator,
+         typename T>
+  ForwardIterator remove(thrust::execution_policy<DerivedPolicy> &exec,
+                         ForwardIterator first,
+                         ForwardIterator last,
+                         const T &value);
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator,
+         typename T>
+  OutputIterator remove_copy(thrust::execution_policy<DerivedPolicy> &exec,
+                             InputIterator first,
+                             InputIterator last,
+                             OutputIterator result,
+                             const T &value);
+
+template<typename DerivedPolicy,
+         typename ForwardIterator,
+         typename Predicate>
+  ForwardIterator remove_if(thrust::execution_policy<DerivedPolicy> &exec,
+                            ForwardIterator first,
+                            ForwardIterator last,
+                            Predicate pred);
+
+template<typename DerivedPolicy,
+         typename ForwardIterator,
+         typename InputIterator,
+         typename Predicate>
+  ForwardIterator remove_if(thrust::execution_policy<DerivedPolicy> &exec,
+                            ForwardIterator first,
+                            ForwardIterator last,
+                            InputIterator stencil,
+                            Predicate pred);
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator,
+         typename Predicate>
+  OutputIterator remove_copy_if(thrust::execution_policy<DerivedPolicy> &exec,
+                                InputIterator first,
+                                InputIterator last,
+                                OutputIterator result,
+                                Predicate pred);
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename Predicate>
+  OutputIterator remove_copy_if(thrust::execution_policy<DerivedPolicy> &exec,
+                                InputIterator1 first,
+                                InputIterator1 last,
+                                InputIterator2 stencil,
+                                OutputIterator result,
+                                Predicate pred);
+
+} // end namespace generic
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/detail/generic/remove.inl>
+
diff --git a/compat/thrust/system/detail/generic/remove.inl b/compat/thrust/system/detail/generic/remove.inl
new file mode 100644
index 0000000000..8a533e029a
--- /dev/null
+++ b/compat/thrust/system/detail/generic/remove.inl
@@ -0,0 +1,144 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file remove.inl
+ *  \brief Inline file for remove.h
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/remove.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/detail/copy_if.h>
+#include <thrust/detail/internal_functional.h>
+#include <thrust/detail/temporary_array.h>
+#include <thrust/remove.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+
+template<typename DerivedPolicy,
+         typename ForwardIterator,
+         typename T>
+  ForwardIterator remove(thrust::execution_policy<DerivedPolicy> &exec,
+                         ForwardIterator first,
+                         ForwardIterator last,
+                         const T &value)
+{
+  thrust::detail::equal_to_value<T> pred(value);
+
+  // XXX consider using a placeholder here
+  return thrust::remove_if(exec, first, last, pred);
+} // end remove()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator,
+         typename T>
+  OutputIterator remove_copy(thrust::execution_policy<DerivedPolicy> &exec,
+                             InputIterator first,
+                             InputIterator last,
+                             OutputIterator result,
+                             const T &value)
+{
+  thrust::detail::equal_to_value<T> pred(value);
+
+  // XXX consider using a placeholder here
+  return thrust::remove_copy_if(exec, first, last, result, pred);
+} // end remove_copy()
+
+
+template<typename DerivedPolicy,
+         typename ForwardIterator,
+         typename Predicate>
+  ForwardIterator remove_if(thrust::execution_policy<DerivedPolicy> &exec,
+                            ForwardIterator first,
+                            ForwardIterator last,
+                            Predicate pred)
+{
+  typedef typename thrust::iterator_traits<ForwardIterator>::value_type InputType;
+
+  // create temporary storage for an intermediate result
+  thrust::detail::temporary_array<InputType,DerivedPolicy> temp(exec, first, last);
+
+  // remove into temp
+  return thrust::remove_copy_if(exec, temp.begin(), temp.end(), temp.begin(), first, pred);
+} // end remove_if()
+
+
+template<typename DerivedPolicy,
+         typename ForwardIterator,
+         typename InputIterator,
+         typename Predicate>
+  ForwardIterator remove_if(thrust::execution_policy<DerivedPolicy> &exec,
+                            ForwardIterator first,
+                            ForwardIterator last,
+                            InputIterator stencil,
+                            Predicate pred)
+{
+  typedef typename thrust::iterator_traits<ForwardIterator>::value_type InputType;
+
+  // create temporary storage for an intermediate result
+  thrust::detail::temporary_array<InputType,DerivedPolicy> temp(exec, first, last);
+
+  // remove into temp
+  return thrust::remove_copy_if(exec, temp.begin(), temp.end(), stencil, first, pred);
+} // end remove_if() 
+
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator,
+         typename Predicate>
+  OutputIterator remove_copy_if(thrust::execution_policy<DerivedPolicy> &exec,
+                                InputIterator first,
+                                InputIterator last,
+                                OutputIterator result,
+                                Predicate pred)
+{
+  return thrust::remove_copy_if(exec, first, last, first, result, pred);
+} // end remove_copy_if()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename Predicate>
+  OutputIterator remove_copy_if(thrust::execution_policy<DerivedPolicy> &exec,
+                                InputIterator1 first,
+                                InputIterator1 last,
+                                InputIterator2 stencil,
+                                OutputIterator result,
+                                Predicate pred)
+{
+  return thrust::copy_if(exec, first, last, stencil, result, thrust::detail::not1(pred));
+} // end remove_copy_if()
+
+
+} // end namespace generic
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/detail/generic/replace.h b/compat/thrust/system/detail/generic/replace.h
new file mode 100644
index 0000000000..deb2e55bc8
--- /dev/null
+++ b/compat/thrust/system/detail/generic/replace.h
@@ -0,0 +1,92 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/tag.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+
+template<typename DerivedPolicy, typename InputIterator, typename OutputIterator, typename Predicate, typename T>
+  OutputIterator replace_copy_if(thrust::execution_policy<DerivedPolicy> &exec,
+                                 InputIterator first,
+                                 InputIterator last,
+                                 OutputIterator result,
+                                 Predicate pred,
+                                 const T &new_value);
+
+
+template<typename DerivedPolicy, typename InputIterator1, typename InputIterator2, typename OutputIterator, typename Predicate, typename T>
+  OutputIterator replace_copy_if(thrust::execution_policy<DerivedPolicy> &exec,
+                                 InputIterator1 first,
+                                 InputIterator1 last,
+                                 InputIterator2 stencil,
+                                 OutputIterator result,
+                                 Predicate pred,
+                                 const T &new_value);
+
+
+template<typename DerivedPolicy, typename InputIterator, typename OutputIterator, typename T>
+  OutputIterator replace_copy(thrust::execution_policy<DerivedPolicy> &exec,
+                              InputIterator first,
+                              InputIterator last,
+                              OutputIterator result,
+                              const T &old_value,
+                              const T &new_value);
+
+
+template<typename DerivedPolicy, typename ForwardIterator, typename Predicate, typename T>
+  void replace_if(thrust::execution_policy<DerivedPolicy> &exec,
+                  ForwardIterator first,
+                  ForwardIterator last,
+                  Predicate pred,
+                  const T &new_value);
+
+
+template<typename DerivedPolicy, typename ForwardIterator, typename InputIterator, typename Predicate, typename T>
+  void replace_if(thrust::execution_policy<DerivedPolicy> &exec,
+                  ForwardIterator first,
+                  ForwardIterator last,
+                  InputIterator stencil,
+                  Predicate pred,
+                  const T &new_value);
+
+
+template<typename DerivedPolicy, typename ForwardIterator, typename T>
+  void replace(thrust::execution_policy<DerivedPolicy> &exec,
+               ForwardIterator first,
+               ForwardIterator last,
+               const T &old_value,
+               const T &new_value);
+
+
+} // end namespace generic
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/detail/generic/replace.inl>
+
diff --git a/compat/thrust/system/detail/generic/replace.inl b/compat/thrust/system/detail/generic/replace.inl
new file mode 100644
index 0000000000..52e7118ecc
--- /dev/null
+++ b/compat/thrust/system/detail/generic/replace.inl
@@ -0,0 +1,168 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/replace.h>
+#include <thrust/transform.h>
+#include <thrust/replace.h>
+#include <thrust/detail/internal_functional.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+namespace detail
+{
+
+// this functor receives x, and returns a new_value if predicate(x) is true; otherwise,
+// it returns x
+template<typename Predicate, typename NewType, typename OutputType>
+  struct new_value_if
+{
+  new_value_if(Predicate p, NewType nv):pred(p),new_value(nv){}
+
+  template<typename InputType>
+  __host__ __device__
+  OutputType operator()(const InputType x) const
+  {
+    return pred(x) ? new_value : x;
+  } // end operator()()
+
+  // this version of operator()() works like the previous but
+  // feeds its second argument to pred
+  template<typename InputType, typename PredicateArgumentType>
+  __host__ __device__
+  OutputType operator()(const InputType x, const PredicateArgumentType y)
+  {
+    return pred(y) ? new_value : x;
+  } // end operator()()
+  
+  Predicate pred;
+  NewType new_value;
+}; // end new_value_if
+
+// this unary functor ignores its argument and returns a constant
+template<typename T>
+  struct constant_unary
+{
+  constant_unary(T _c):c(_c){}
+
+  template<typename U>
+  __host__ __device__
+  T operator()(U &x)
+  {
+    return c;
+  } // end operator()()
+
+  T c;
+}; // end constant_unary
+
+} // end detail
+
+template<typename DerivedPolicy, typename InputIterator, typename OutputIterator, typename Predicate, typename T>
+  OutputIterator replace_copy_if(thrust::execution_policy<DerivedPolicy> &exec,
+                                 InputIterator first,
+                                 InputIterator last,
+                                 OutputIterator result,
+                                 Predicate pred,
+                                 const T &new_value)
+{
+  typedef typename thrust::iterator_traits<InputIterator>::value_type InputType;
+  typedef typename thrust::iterator_traits<OutputIterator>::value_type OutputType;
+
+  detail::new_value_if<Predicate,T,OutputType> op(pred,new_value);
+  return thrust::transform(exec, first, last, result, op);
+} // end replace_copy_if()
+
+template<typename DerivedPolicy, typename InputIterator1, typename InputIterator2, typename OutputIterator, typename Predicate, typename T>
+  OutputIterator replace_copy_if(thrust::execution_policy<DerivedPolicy> &exec,
+                                 InputIterator1 first,
+                                 InputIterator1 last,
+                                 InputIterator2 stencil,
+                                 OutputIterator result,
+                                 Predicate pred,
+                                 const T &new_value)
+{
+  typedef typename thrust::iterator_traits<OutputIterator>::value_type OutputType;
+
+  detail::new_value_if<Predicate,T,OutputType> op(pred,new_value);
+  return thrust::transform(exec, first, last, stencil, result, op);
+} // end replace_copy_if()
+
+
+template<typename DerivedPolicy, typename InputIterator, typename OutputIterator, typename T>
+  OutputIterator replace_copy(thrust::execution_policy<DerivedPolicy> &exec,
+                              InputIterator first,
+                              InputIterator last,
+                              OutputIterator result,
+                              const T &old_value,
+                              const T &new_value)
+{
+  thrust::detail::equal_to_value<T> pred(old_value);
+  return thrust::replace_copy_if(exec, first, last, result, pred, new_value);
+} // end replace_copy()
+
+template<typename DerivedPolicy, typename ForwardIterator, typename Predicate, typename T>
+  void replace_if(thrust::execution_policy<DerivedPolicy> &exec,
+                  ForwardIterator first,
+                  ForwardIterator last,
+                  Predicate pred,
+                  const T &new_value)
+{
+  detail::constant_unary<T> f(new_value);
+
+  // XXX replace this with generate_if:
+  // constant_nullary<T> f(new_value);
+  // generate_if(first, last, first, f, pred);
+  thrust::transform_if(exec, first, last, first, first, f, pred);
+} // end replace_if()
+
+template<typename DerivedPolicy, typename ForwardIterator, typename InputIterator, typename Predicate, typename T>
+  void replace_if(thrust::execution_policy<DerivedPolicy> &exec,
+                  ForwardIterator first,
+                  ForwardIterator last,
+                  InputIterator stencil,
+                  Predicate pred,
+                  const T &new_value)
+{
+  detail::constant_unary<T> f(new_value);
+
+  // XXX replace this with generate_if:
+  // constant_nullary<T> f(new_value);
+  // generate_if(stencil, stencil + n, first, f, pred);
+  thrust::transform_if(exec, first, last, stencil, first, f, pred);
+} // end replace_if()
+
+template<typename DerivedPolicy, typename ForwardIterator, typename T>
+  void replace(thrust::execution_policy<DerivedPolicy> &exec,
+               ForwardIterator first,
+               ForwardIterator last,
+               const T &old_value,
+               const T &new_value)
+{
+  thrust::detail::equal_to_value<T> pred(old_value);
+  return thrust::replace_if(exec, first, last, pred, new_value);
+} // end replace()
+
+} // end namespace generic
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/detail/generic/reverse.h b/compat/thrust/system/detail/generic/reverse.h
new file mode 100644
index 0000000000..327bf221b1
--- /dev/null
+++ b/compat/thrust/system/detail/generic/reverse.h
@@ -0,0 +1,51 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/tag.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+template<typename DerivedPolicy, typename BidirectionalIterator>
+  void reverse(thrust::execution_policy<DerivedPolicy> &exec,
+               BidirectionalIterator first,
+               BidirectionalIterator last);
+
+template<typename DerivedPolicy,
+         typename BidirectionalIterator,
+         typename OutputIterator>
+  OutputIterator reverse_copy(thrust::execution_policy<DerivedPolicy> &exec,
+                              BidirectionalIterator first,
+                              BidirectionalIterator last,
+                              OutputIterator result);
+
+} // end namespace generic
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/detail/generic/reverse.inl>
+
diff --git a/compat/thrust/system/detail/generic/reverse.inl b/compat/thrust/system/detail/generic/reverse.inl
new file mode 100644
index 0000000000..27c1bbf2fa
--- /dev/null
+++ b/compat/thrust/system/detail/generic/reverse.inl
@@ -0,0 +1,70 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/reverse.h>
+#include <thrust/advance.h>
+#include <thrust/distance.h>
+#include <thrust/detail/copy.h>
+#include <thrust/swap.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/iterator/reverse_iterator.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+template<typename ExecutionPolicy, typename BidirectionalIterator>
+  void reverse(thrust::execution_policy<ExecutionPolicy> &exec,
+               BidirectionalIterator first,
+               BidirectionalIterator last)
+{
+  typedef typename thrust::iterator_difference<BidirectionalIterator>::type difference_type;
+
+  // find the midpoint of [first,last)
+  difference_type N = thrust::distance(first, last);
+  BidirectionalIterator mid(first);
+  thrust::advance(mid, N / 2);
+
+  // swap elements of [first,mid) with [last - 1, mid)
+  thrust::swap_ranges(exec, first, mid, thrust::make_reverse_iterator(last));
+} // end reverse()
+
+template<typename ExecutionPolicy,
+         typename BidirectionalIterator,
+         typename OutputIterator>
+  OutputIterator reverse_copy(thrust::execution_policy<ExecutionPolicy> &exec,
+                              BidirectionalIterator first,
+                              BidirectionalIterator last,
+                              OutputIterator result)
+{
+  return thrust::copy(exec,
+                      thrust::make_reverse_iterator(last),
+                      thrust::make_reverse_iterator(first),
+                      result);
+} // end reverse_copy()
+
+} // end namespace generic
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
+
diff --git a/compat/thrust/system/detail/generic/scalar/binary_search.h b/compat/thrust/system/detail/generic/scalar/binary_search.h
new file mode 100644
index 0000000000..6ed9e8d9ee
--- /dev/null
+++ b/compat/thrust/system/detail/generic/scalar/binary_search.h
@@ -0,0 +1,85 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/pair.h>
+
+namespace thrust
+{
+
+namespace system
+{
+
+namespace detail
+{
+
+namespace generic
+{
+
+namespace scalar
+{
+
+template<typename RandomAccessIterator, typename Size, typename T, typename BinaryPredicate>
+__host__ __device__
+RandomAccessIterator lower_bound_n(RandomAccessIterator first,
+                                   Size n,
+                                   const T &val,
+                                   BinaryPredicate comp);
+
+template<typename RandomAccessIterator, typename T, typename BinaryPredicate>
+__host__ __device__
+RandomAccessIterator lower_bound(RandomAccessIterator first, RandomAccessIterator last,
+                                 const T &val,
+                                 BinaryPredicate comp);
+
+template<typename RandomAccessIterator, typename Size, typename T, typename BinaryPredicate>
+__host__ __device__
+RandomAccessIterator upper_bound_n(RandomAccessIterator first,
+                                   Size n,
+                                   const T &val,
+                                   BinaryPredicate comp);
+
+template<typename RandomAccessIterator, typename T, typename BinaryPredicate>
+__host__ __device__
+RandomAccessIterator upper_bound(RandomAccessIterator first, RandomAccessIterator last,
+                                 const T &val,
+                                 BinaryPredicate comp);
+
+template<typename RandomAccessIterator, typename T, typename BinaryPredicate>
+__host__ __device__
+  pair<RandomAccessIterator,RandomAccessIterator>
+    equal_range(RandomAccessIterator first, RandomAccessIterator last,
+                const T &val,
+                BinaryPredicate comp);
+
+template<typename RandomAccessIterator, typename T, typename Compare>
+__host__ __device__
+bool binary_search(RandomAccessIterator first, RandomAccessIterator last, const T &value, Compare comp);
+
+} // end scalar
+
+} // end generic
+
+} // end detail
+
+} // end system
+
+} // end thrust
+
+#include <thrust/system/detail/generic/scalar/binary_search.inl>
+
diff --git a/compat/thrust/system/detail/generic/scalar/binary_search.inl b/compat/thrust/system/detail/generic/scalar/binary_search.inl
new file mode 100644
index 0000000000..5a9d379612
--- /dev/null
+++ b/compat/thrust/system/detail/generic/scalar/binary_search.inl
@@ -0,0 +1,159 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/pair.h>
+#include <thrust/detail/function.h>
+#include <thrust/iterator/iterator_traits.h>
+
+namespace thrust
+{
+
+namespace system
+{
+
+namespace detail
+{
+
+namespace generic
+{
+
+namespace scalar
+{
+
+template<typename RandomAccessIterator, typename Size, typename T, typename BinaryPredicate>
+__host__ __device__
+RandomAccessIterator lower_bound_n(RandomAccessIterator first,
+                                   Size n,
+                                   const T &val,
+                                   BinaryPredicate comp)
+{
+  // wrap comp
+  thrust::detail::host_device_function<
+    BinaryPredicate,
+    bool
+  > wrapped_comp(comp);
+
+  Size start = 0, i;
+  while(start < n)
+  {
+    i = (start + n) / 2;
+    if(wrapped_comp(first[i], val))
+    {
+      start = i + 1;
+    }
+    else
+    {
+      n = i;
+    }
+  } // end while
+  
+  return first + start;
+}
+
+// XXX generalize these upon implementation of scalar::distance & scalar::advance
+
+template<typename RandomAccessIterator, typename T, typename BinaryPredicate>
+__host__ __device__
+RandomAccessIterator lower_bound(RandomAccessIterator first, RandomAccessIterator last,
+                                 const T &val,
+                                 BinaryPredicate comp)
+{
+  typename thrust::iterator_difference<RandomAccessIterator>::type n = last - first;
+  return lower_bound_n(first, n, val, comp);
+}
+
+template<typename RandomAccessIterator, typename Size, typename T, typename BinaryPredicate>
+__host__ __device__
+RandomAccessIterator upper_bound_n(RandomAccessIterator first,
+                                   Size n,
+                                   const T &val,
+                                   BinaryPredicate comp)
+{
+  // wrap comp
+  thrust::detail::host_device_function<
+    BinaryPredicate,
+    bool
+  > wrapped_comp(comp);
+
+  Size start = 0, i;
+  while(start < n)
+  {
+    i = (start + n) / 2;
+    if(wrapped_comp(val, first[i]))
+    {
+      n = i;
+    }
+    else
+    {
+      start = i + 1;
+    }
+  } // end while
+  
+  return first + start;
+}
+
+template<typename RandomAccessIterator, typename T, typename BinaryPredicate>
+__host__ __device__
+RandomAccessIterator upper_bound(RandomAccessIterator first, RandomAccessIterator last,
+                                 const T &val,
+                                 BinaryPredicate comp)
+{
+  typename thrust::iterator_difference<RandomAccessIterator>::type n = last - first;
+  return upper_bound_n(first, n, val, comp);
+}
+
+template<typename RandomAccessIterator, typename T, typename BinaryPredicate>
+__host__ __device__
+  pair<RandomAccessIterator,RandomAccessIterator>
+    equal_range(RandomAccessIterator first, RandomAccessIterator last,
+                const T &val,
+                BinaryPredicate comp)
+{
+  RandomAccessIterator lb = thrust::system::detail::generic::scalar::lower_bound(first, last, val, comp);
+  return thrust::make_pair(lb, thrust::system::detail::generic::scalar::upper_bound(lb, last, val, comp));
+}
+
+
+template<typename RandomAccessIterator, typename T, typename Compare>
+__host__ __device__
+bool binary_search(RandomAccessIterator first, RandomAccessIterator last, const T &value, Compare comp)
+{
+  RandomAccessIterator iter = thrust::system::detail::generic::scalar::lower_bound(first, last, value, comp);
+
+  // wrap comp
+  thrust::detail::host_device_function<
+    Compare,
+    bool
+  > wrapped_comp(comp);
+
+  return iter != last && !wrapped_comp(value,*iter);
+}
+
+} // end scalar
+
+} // end generic
+
+} // end detail
+
+} // end system
+
+} // end thrust
+
+#include <thrust/system/detail/generic/scalar/binary_search.inl>
+
diff --git a/compat/thrust/system/detail/generic/scan.h b/compat/thrust/system/detail/generic/scan.h
new file mode 100644
index 0000000000..205f87ff9f
--- /dev/null
+++ b/compat/thrust/system/detail/generic/scan.h
@@ -0,0 +1,94 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/tag.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+
+template<typename ExecutionPolicy,
+         typename InputIterator,
+         typename OutputIterator>
+  OutputIterator inclusive_scan(thrust::execution_policy<ExecutionPolicy> &exec,
+                                InputIterator first,
+                                InputIterator last,
+                                OutputIterator result);
+
+
+// XXX it is an error to call this function; it has no implementation 
+template<typename ExecutionPolicy,
+         typename InputIterator,
+         typename OutputIterator,
+         typename BinaryFunction>
+  OutputIterator inclusive_scan(thrust::execution_policy<ExecutionPolicy> &exec,
+                                InputIterator first,
+                                InputIterator last,
+                                OutputIterator result,
+                                BinaryFunction binary_op);
+
+
+template<typename ExecutionPolicy,
+         typename InputIterator,
+         typename OutputIterator>
+  OutputIterator exclusive_scan(thrust::execution_policy<ExecutionPolicy> &exec,
+                                InputIterator first,
+                                InputIterator last,
+                                OutputIterator result);
+
+
+template<typename ExecutionPolicy,
+         typename InputIterator,
+         typename OutputIterator,
+         typename T>
+  OutputIterator exclusive_scan(thrust::execution_policy<ExecutionPolicy> &exec,
+                                InputIterator first,
+                                InputIterator last,
+                                OutputIterator result,
+                                T init);
+
+
+// XXX it is an error to call this function; it has no implementation 
+template<typename ExecutionPolicy,
+         typename InputIterator,
+         typename OutputIterator,
+         typename T,
+         typename BinaryFunction>
+  OutputIterator exclusive_scan(thrust::execution_policy<ExecutionPolicy> &exec,
+                                InputIterator first,
+                                InputIterator last,
+                                OutputIterator result,
+                                T init,
+                                BinaryFunction binary_op);
+
+
+} // end namespace generic
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/detail/generic/scan.inl>
+
diff --git a/compat/thrust/system/detail/generic/scan.inl b/compat/thrust/system/detail/generic/scan.inl
new file mode 100644
index 0000000000..33e0803c70
--- /dev/null
+++ b/compat/thrust/system/detail/generic/scan.inl
@@ -0,0 +1,144 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/static_assert.h>
+#include <thrust/system/detail/generic/scan.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/scan.h>
+#include <thrust/detail/type_traits.h>
+#include <thrust/detail/type_traits/iterator/is_output_iterator.h>
+#include <thrust/functional.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+
+template<typename ExecutionPolicy,
+         typename InputIterator,
+         typename OutputIterator>
+  OutputIterator inclusive_scan(thrust::execution_policy<ExecutionPolicy> &exec,
+                                InputIterator first,
+                                InputIterator last,
+                                OutputIterator result)
+{
+  // the pseudocode for deducing the type of the temporary used below:
+  // 
+  // if OutputIterator is a "pure" output iterator
+  //   TemporaryType = InputIterator::value_type
+  // else
+  //   TemporaryType = OutputIterator::value_type
+
+  typedef typename thrust::detail::eval_if<
+      thrust::detail::is_output_iterator<OutputIterator>::value,
+      thrust::iterator_value<InputIterator>,
+      thrust::iterator_value<OutputIterator>
+  >::type ValueType;
+
+  // assume plus as the associative operator
+  return thrust::inclusive_scan(exec, first, last, result, thrust::plus<ValueType>());
+} // end inclusive_scan()
+
+
+template<typename ExecutionPolicy,
+         typename InputIterator,
+         typename OutputIterator>
+  OutputIterator exclusive_scan(thrust::execution_policy<ExecutionPolicy> &exec,
+                                InputIterator first,
+                                InputIterator last,
+                                OutputIterator result)
+{
+  // the pseudocode for deducing the type of the temporary used below:
+  // 
+  // if OutputIterator is a "pure" output iterator
+  //   TemporaryType = InputIterator::value_type
+  // else
+  //   TemporaryType = OutputIterator::value_type
+
+  typedef typename thrust::detail::eval_if<
+      thrust::detail::is_output_iterator<OutputIterator>::value,
+      thrust::iterator_value<InputIterator>,
+      thrust::iterator_value<OutputIterator>
+  >::type ValueType;
+
+  // assume 0 as the initialization value
+  return thrust::exclusive_scan(exec, first, last, result, ValueType(0));
+} // end exclusive_scan()
+
+
+template<typename ExecutionPolicy,
+         typename InputIterator,
+         typename OutputIterator,
+         typename T>
+  OutputIterator exclusive_scan(thrust::execution_policy<ExecutionPolicy> &exec,
+                                InputIterator first,
+                                InputIterator last,
+                                OutputIterator result,
+                                T init)
+{
+  // assume plus as the associative operator
+  return thrust::exclusive_scan(exec, first, last, result, init, thrust::plus<T>());
+} // end exclusive_scan()
+
+
+template<typename ExecutionPolicy,
+         typename InputIterator,
+         typename OutputIterator,
+         typename BinaryFunction>
+  OutputIterator inclusive_scan(thrust::execution_policy<ExecutionPolicy> &exec,
+                                InputIterator first,
+                                InputIterator last,
+                                OutputIterator result,
+                                BinaryFunction binary_op)
+{
+  // unimplemented primitive
+  THRUST_STATIC_ASSERT( (thrust::detail::depend_on_instantiation<InputIterator, false>::value) );
+  return result;
+} // end inclusive_scan
+
+
+template<typename ExecutionPolicy,
+         typename InputIterator,
+         typename OutputIterator,
+         typename T,
+         typename BinaryFunction>
+  OutputIterator exclusive_scan(thrust::execution_policy<ExecutionPolicy> &exec,
+                                InputIterator first,
+                                InputIterator last,
+                                OutputIterator result,
+                                T init,
+                                BinaryFunction binary_op)
+{
+  // unimplemented primitive
+  THRUST_STATIC_ASSERT( (thrust::detail::depend_on_instantiation<InputIterator, false>::value) );
+  return result;
+} // end exclusive_scan()
+
+
+} // end namespace generic
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/detail/generic/scan_by_key.h b/compat/thrust/system/detail/generic/scan_by_key.h
new file mode 100644
index 0000000000..160121b58b
--- /dev/null
+++ b/compat/thrust/system/detail/generic/scan_by_key.h
@@ -0,0 +1,137 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file scan_by_key.h
+ *  \brief Generic implementations of key-value scans.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/tag.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator>
+  OutputIterator inclusive_scan_by_key(thrust::execution_policy<DerivedPolicy> &exec,
+                                       InputIterator1 first1,
+                                       InputIterator1 last1,
+                                       InputIterator2 first2,
+                                       OutputIterator result);
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename BinaryPredicate>
+  OutputIterator inclusive_scan_by_key(thrust::execution_policy<DerivedPolicy> &exec,
+                                       InputIterator1 first1,
+                                       InputIterator1 last1,
+                                       InputIterator2 first2,
+                                       OutputIterator result,
+                                       BinaryPredicate binary_pred);
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename BinaryPredicate,
+         typename AssociativeOperator>
+  OutputIterator inclusive_scan_by_key(thrust::execution_policy<DerivedPolicy> &exec,
+                                       InputIterator1 first1,
+                                       InputIterator1 last1,
+                                       InputIterator2 first2,
+                                       OutputIterator result,
+                                       BinaryPredicate binary_pred,
+                                       AssociativeOperator binary_op);
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator>
+  OutputIterator exclusive_scan_by_key(thrust::execution_policy<DerivedPolicy> &exec,
+                                       InputIterator1 first1,
+                                       InputIterator1 last1,
+                                       InputIterator2 first2,
+                                       OutputIterator result);
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename T>
+  OutputIterator exclusive_scan_by_key(thrust::execution_policy<DerivedPolicy> &exec,
+                                       InputIterator1 first1,
+                                       InputIterator1 last1,
+                                       InputIterator2 first2,
+                                       OutputIterator result,
+                                       T init);
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename T,
+         typename BinaryPredicate>
+  OutputIterator exclusive_scan_by_key(thrust::execution_policy<DerivedPolicy> &exec,
+                                       InputIterator1 first1,
+                                       InputIterator1 last1,
+                                       InputIterator2 first2,
+                                       OutputIterator result,
+                                       T init,
+                                       BinaryPredicate binary_pred);
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename T,
+         typename BinaryPredicate,
+         typename AssociativeOperator>
+  OutputIterator exclusive_scan_by_key(thrust::execution_policy<DerivedPolicy> &exec,
+                                       InputIterator1 first1,
+                                       InputIterator1 last1,
+                                       InputIterator2 first2,
+                                       OutputIterator result,
+                                       T init,
+                                       BinaryPredicate binary_pred,
+                                       AssociativeOperator binary_op);
+
+} // end namespace generic
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/detail/generic/scan_by_key.inl>
+
diff --git a/compat/thrust/system/detail/generic/scan_by_key.inl b/compat/thrust/system/detail/generic/scan_by_key.inl
new file mode 100644
index 0000000000..d866ddec15
--- /dev/null
+++ b/compat/thrust/system/detail/generic/scan_by_key.inl
@@ -0,0 +1,239 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/scan_by_key.h>
+#include <thrust/functional.h>
+#include <thrust/transform.h>
+#include <thrust/replace.h>
+#include <thrust/iterator/zip_iterator.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/detail/temporary_array.h>
+#include <thrust/detail/internal_functional.h>
+#include <thrust/scan.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+namespace detail
+{
+
+template <typename OutputType, typename HeadFlagType, typename AssociativeOperator>
+struct segmented_scan_functor
+{
+    AssociativeOperator binary_op;
+
+    typedef typename thrust::tuple<OutputType, HeadFlagType> result_type;
+
+    __host__ __device__
+    segmented_scan_functor(AssociativeOperator _binary_op) : binary_op(_binary_op) {}
+
+    __host__ __device__
+    result_type operator()(result_type a, result_type b)
+    {
+        return result_type(thrust::get<1>(b) ? thrust::get<0>(b) : binary_op(thrust::get<0>(a), thrust::get<0>(b)),
+                           thrust::get<1>(a) | thrust::get<1>(b));
+    }
+};
+
+} // end namespace detail
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator>
+  OutputIterator inclusive_scan_by_key(thrust::execution_policy<DerivedPolicy> &exec,
+                                       InputIterator1 first1,
+                                       InputIterator1 last1,
+                                       InputIterator2 first2,
+                                       OutputIterator result)
+{
+  typedef typename thrust::iterator_traits<InputIterator1>::value_type InputType1;
+  return thrust::inclusive_scan_by_key(exec, first1, last1, first2, result, thrust::equal_to<InputType1>());
+}
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename BinaryPredicate>
+  OutputIterator inclusive_scan_by_key(thrust::execution_policy<DerivedPolicy> &exec,
+                                       InputIterator1 first1,
+                                       InputIterator1 last1,
+                                       InputIterator2 first2,
+                                       OutputIterator result,
+                                       BinaryPredicate binary_pred)
+{
+  typedef typename thrust::iterator_traits<OutputIterator>::value_type OutputType;
+  return thrust::inclusive_scan_by_key(exec, first1, last1, first2, result, binary_pred, thrust::plus<OutputType>());
+}
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename BinaryPredicate,
+         typename AssociativeOperator>
+  OutputIterator inclusive_scan_by_key(thrust::execution_policy<DerivedPolicy> &exec,
+                                       InputIterator1 first1,
+                                       InputIterator1 last1,
+                                       InputIterator2 first2,
+                                       OutputIterator result,
+                                       BinaryPredicate binary_pred,
+                                       AssociativeOperator binary_op)
+{
+  typedef typename thrust::iterator_traits<OutputIterator>::value_type OutputType;
+  typedef unsigned int HeadFlagType;
+
+  const size_t n = last1 - first1;
+
+  if(n != 0)
+  {
+    // compute head flags
+    thrust::detail::temporary_array<HeadFlagType,DerivedPolicy> flags(exec, n);
+    flags[0] = 1; thrust::transform(exec, first1, last1 - 1, first1 + 1, flags.begin() + 1, thrust::detail::not2(binary_pred));
+
+    // scan key-flag tuples, 
+    // For additional details refer to Section 2 of the following paper
+    //    S. Sengupta, M. Harris, and M. Garland. "Efficient parallel scan algorithms for GPUs"
+    //    NVIDIA Technical Report NVR-2008-003, December 2008
+    //    http://mgarland.org/files/papers/nvr-2008-003.pdf
+    thrust::inclusive_scan
+        (exec,
+         thrust::make_zip_iterator(thrust::make_tuple(first2, flags.begin())),
+         thrust::make_zip_iterator(thrust::make_tuple(first2, flags.begin())) + n,
+         thrust::make_zip_iterator(thrust::make_tuple(result, flags.begin())),
+         detail::segmented_scan_functor<OutputType, HeadFlagType, AssociativeOperator>(binary_op));
+  }
+
+  return result + n;
+}
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator>
+  OutputIterator exclusive_scan_by_key(thrust::execution_policy<DerivedPolicy> &exec,
+                                       InputIterator1 first1,
+                                       InputIterator1 last1,
+                                       InputIterator2 first2,
+                                       OutputIterator result)
+{
+  typedef typename thrust::iterator_traits<OutputIterator>::value_type OutputType;
+  return thrust::exclusive_scan_by_key(exec, first1, last1, first2, result, OutputType(0));
+}
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename T>
+  OutputIterator exclusive_scan_by_key(thrust::execution_policy<DerivedPolicy> &exec,
+                                       InputIterator1 first1,
+                                       InputIterator1 last1,
+                                       InputIterator2 first2,
+                                       OutputIterator result,
+                                       T init)
+{
+  typedef typename thrust::iterator_traits<InputIterator1>::value_type InputType1;
+  return thrust::exclusive_scan_by_key(exec, first1, last1, first2, result, init, thrust::equal_to<InputType1>());
+}
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename T,
+         typename BinaryPredicate>
+  OutputIterator exclusive_scan_by_key(thrust::execution_policy<DerivedPolicy> &exec,
+                                       InputIterator1 first1,
+                                       InputIterator1 last1,
+                                       InputIterator2 first2,
+                                       OutputIterator result,
+                                       T init,
+                                       BinaryPredicate binary_pred)
+{
+  typedef typename thrust::iterator_traits<OutputIterator>::value_type OutputType;
+  return thrust::exclusive_scan_by_key(exec, first1, last1, first2, result, init, binary_pred, thrust::plus<OutputType>());
+}
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename T,
+         typename BinaryPredicate,
+         typename AssociativeOperator>
+  OutputIterator exclusive_scan_by_key(thrust::execution_policy<DerivedPolicy> &exec,
+                                       InputIterator1 first1,
+                                       InputIterator1 last1,
+                                       InputIterator2 first2,
+                                       OutputIterator result,
+                                       T init,
+                                       BinaryPredicate binary_pred,
+                                       AssociativeOperator binary_op)
+{
+  typedef typename thrust::iterator_traits<OutputIterator>::value_type OutputType;
+  typedef unsigned int HeadFlagType;
+
+  const size_t n = last1 - first1;
+
+  if(n != 0)
+  {
+    InputIterator2 last2 = first2 + n;
+
+    // compute head flags
+    thrust::detail::temporary_array<HeadFlagType,DerivedPolicy> flags(exec, n);
+    flags[0] = 1; thrust::transform(exec, first1, last1 - 1, first1 + 1, flags.begin() + 1, thrust::detail::not2(binary_pred));
+
+    // shift input one to the right and initialize segments with init
+    thrust::detail::temporary_array<OutputType,DerivedPolicy> temp(exec, n);
+    thrust::replace_copy_if(exec, first2, last2 - 1, flags.begin() + 1, temp.begin() + 1, thrust::negate<HeadFlagType>(), init);
+    temp[0] = init;
+
+    // scan key-flag tuples, 
+    // For additional details refer to Section 2 of the following paper
+    //    S. Sengupta, M. Harris, and M. Garland. "Efficient parallel scan algorithms for GPUs"
+    //    NVIDIA Technical Report NVR-2008-003, December 2008
+    //    http://mgarland.org/files/papers/nvr-2008-003.pdf
+    thrust::inclusive_scan(exec,
+                           thrust::make_zip_iterator(thrust::make_tuple(temp.begin(), flags.begin())),
+                           thrust::make_zip_iterator(thrust::make_tuple(temp.begin(), flags.begin())) + n,
+                           thrust::make_zip_iterator(thrust::make_tuple(result,       flags.begin())),
+                           detail::segmented_scan_functor<OutputType, HeadFlagType, AssociativeOperator>(binary_op));
+  }
+
+  return result + n;
+}
+
+} // end namespace generic
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/detail/generic/scatter.h b/compat/thrust/system/detail/generic/scatter.h
new file mode 100644
index 0000000000..858d11adc1
--- /dev/null
+++ b/compat/thrust/system/detail/generic/scatter.h
@@ -0,0 +1,76 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/tag.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename RandomAccessIterator>
+  void scatter(thrust::execution_policy<DerivedPolicy> &exec,
+               InputIterator1 first,
+               InputIterator1 last,
+               InputIterator2 map,
+               RandomAccessIterator output);
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename RandomAccessIterator>
+  void scatter_if(thrust::execution_policy<DerivedPolicy> &exec,
+                  InputIterator1 first,
+                  InputIterator1 last,
+                  InputIterator2 map,
+                  InputIterator3 stencil,
+                  RandomAccessIterator output);
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename RandomAccessIterator,
+         typename Predicate>
+  void scatter_if(thrust::execution_policy<DerivedPolicy> &exec,
+                  InputIterator1 first,
+                  InputIterator1 last,
+                  InputIterator2 map,
+                  InputIterator3 stencil,
+                  RandomAccessIterator output,
+                  Predicate pred);
+
+} // end namespace generic
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/detail/generic/scatter.inl>
+
diff --git a/compat/thrust/system/detail/generic/scatter.inl b/compat/thrust/system/detail/generic/scatter.inl
new file mode 100644
index 0000000000..8c40359844
--- /dev/null
+++ b/compat/thrust/system/detail/generic/scatter.inl
@@ -0,0 +1,93 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/scatter.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/functional.h>
+#include <thrust/transform.h>
+#include <thrust/iterator/permutation_iterator.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename RandomAccessIterator>
+  void scatter(thrust::execution_policy<DerivedPolicy> &exec,
+               InputIterator1 first,
+               InputIterator1 last,
+               InputIterator2 map,
+               RandomAccessIterator output)
+{
+  thrust::transform(exec,
+                    first,
+                    last,
+                    thrust::make_permutation_iterator(output, map),
+                    thrust::identity<typename thrust::iterator_value<InputIterator1>::type>());
+} // end scatter()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename RandomAccessIterator>
+  void scatter_if(thrust::execution_policy<DerivedPolicy> &exec,
+                  InputIterator1 first,
+                  InputIterator1 last,
+                  InputIterator2 map,
+                  InputIterator3 stencil,
+                  RandomAccessIterator output)
+{
+  // default predicate is identity
+  typedef typename thrust::iterator_value<InputIterator3>::type StencilType;
+  thrust::scatter_if(exec, first, last, map, stencil, output, thrust::identity<StencilType>());
+} // end scatter_if()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename RandomAccessIterator,
+         typename Predicate>
+  void scatter_if(thrust::execution_policy<DerivedPolicy> &exec,
+                  InputIterator1 first,
+                  InputIterator1 last,
+                  InputIterator2 map,
+                  InputIterator3 stencil,
+                  RandomAccessIterator output,
+                  Predicate pred)
+{
+  typedef typename thrust::iterator_value<InputIterator1>::type InputType;
+  thrust::transform_if(exec, first, last, stencil, thrust::make_permutation_iterator(output, map), thrust::identity<InputType>(), pred);
+} // end scatter_if()
+
+
+} // end namespace generic
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/detail/generic/select_system.h b/compat/thrust/system/detail/generic/select_system.h
new file mode 100644
index 0000000000..250a0bce44
--- /dev/null
+++ b/compat/thrust/system/detail/generic/select_system.h
@@ -0,0 +1,182 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/execution_policy.h>
+#include <thrust/detail/type_traits.h>
+#include <thrust/detail/type_traits/is_metafunction_defined.h>
+#include <thrust/iterator/detail/minimum_system.h>
+#include <thrust/system/detail/generic/type_traits.h>
+#include <thrust/iterator/iterator_traits.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+namespace select_system_detail
+{
+
+
+// min_system case 1: both systems have the same type, just return the first one
+template<typename System>
+__host__ __device__
+System &min_system(thrust::execution_policy<System> &system1,
+                   thrust::execution_policy<System> &)
+{
+  return thrust::detail::derived_cast(system1);
+} // end min_system()
+
+
+// min_system case 2: systems have differing type and the first type is considered the minimum
+template<typename System1, typename System2>
+__host__ __device__
+  typename thrust::detail::enable_if<
+    thrust::detail::is_same<
+      System1,
+      typename thrust::detail::minimum_system<System1,System2>::type
+    >::value,
+    System1 &
+  >::type
+    min_system(thrust::execution_policy<System1> &system1, thrust::execution_policy<System2> &)
+{
+  return thrust::detail::derived_cast(system1);
+} // end min_system()
+
+
+// min_system case 3: systems have differing type and the second type is considered the minimum
+template<typename System1, typename System2>
+__host__ __device__
+  typename thrust::detail::enable_if<
+    thrust::detail::is_same<
+      System2,
+      typename thrust::detail::minimum_system<System1,System2>::type
+    >::value,
+    System2 &
+  >::type
+    min_system(thrust::execution_policy<System1> &, thrust::execution_policy<System2> &system2)
+{
+  return thrust::detail::derived_cast(system2);
+} // end min_system()
+
+
+} // end select_system_detail
+
+
+template<typename System>
+__host__ __device__
+  typename thrust::detail::disable_if<
+    select_system1_exists<System>::value,
+    System &
+  >::type
+    select_system(thrust::execution_policy<System> &system)
+{
+  return thrust::detail::derived_cast(system);
+} // end select_system()
+
+
+template<typename System1, typename System2>
+__host__ __device__
+  typename thrust::detail::enable_if_defined<
+    thrust::detail::minimum_system<System1,System2>
+  >::type
+    &select_system(thrust::execution_policy<System1> &system1,
+                   thrust::execution_policy<System2> &system2)
+{
+  return select_system_detail::min_system(system1,system2);
+} // end select_system()
+
+
+template<typename System1, typename System2, typename System3>
+__host__ __device__
+  typename thrust::detail::lazy_disable_if<
+    select_system3_exists<System1,System2,System3>::value,
+    thrust::detail::minimum_system<System1,System2,System3>
+  >::type
+    &select_system(thrust::execution_policy<System1> &system1,
+                   thrust::execution_policy<System2> &system2,
+                   thrust::execution_policy<System3> &system3)
+{
+  return select_system(select_system(system1,system2), system3);
+} // end select_system()
+
+
+template<typename System1, typename System2, typename System3, typename System4>
+__host__ __device__
+  typename thrust::detail::lazy_disable_if<
+    select_system4_exists<System1,System2,System3,System4>::value,
+    thrust::detail::minimum_system<System1,System2,System3,System4>
+  >::type
+    &select_system(thrust::execution_policy<System1> &system1,
+                   thrust::execution_policy<System2> &system2,
+                   thrust::execution_policy<System3> &system3,
+                   thrust::execution_policy<System4> &system4)
+{
+  return select_system(select_system(system1,system2,system3), system4);
+} // end select_system()
+
+
+template<typename System1, typename System2, typename System3, typename System4, typename System5>
+__host__ __device__
+  typename thrust::detail::lazy_disable_if<
+    select_system5_exists<System1,System2,System3,System4,System5>::value,
+    thrust::detail::minimum_system<System1,System2,System3,System4,System5>
+  >::type
+    &select_system(thrust::execution_policy<System1> &system1,
+                   thrust::execution_policy<System2> &system2,
+                   thrust::execution_policy<System3> &system3,
+                   thrust::execution_policy<System4> &system4,
+                   thrust::execution_policy<System5> &system5)
+{
+  return select_system(select_system(system1,system2,system3,system4), system5);
+} // end select_system()
+
+
+template<typename System1, typename System2, typename System3, typename System4, typename System5, typename System6>
+__host__ __device__
+  typename thrust::detail::lazy_disable_if<
+    select_system6_exists<System1,System2,System3,System4,System5,System6>::value,
+    thrust::detail::minimum_system<System1,System2,System3,System4,System5,System6>
+  >::type
+    &select_system(thrust::execution_policy<System1> &system1,
+                   thrust::execution_policy<System2> &system2,
+                   thrust::execution_policy<System3> &system3,
+                   thrust::execution_policy<System4> &system4,
+                   thrust::execution_policy<System5> &system5,
+                   thrust::execution_policy<System6> &system6)
+{
+  return select_system(select_system(system1,system2,system3,system4,system5), system6);
+} // end select_system()
+
+
+// map a single any_system_tag to device_system_tag
+inline __host__ __device__
+thrust::device_system_tag select_system(thrust::any_system_tag)
+{
+  return thrust::device_system_tag();
+} // end select_system()
+
+
+} // end generic
+} // end detail
+} // end system
+} // end thrust
+
diff --git a/compat/thrust/system/detail/generic/sequence.h b/compat/thrust/system/detail/generic/sequence.h
new file mode 100644
index 0000000000..b23a7b5d22
--- /dev/null
+++ b/compat/thrust/system/detail/generic/sequence.h
@@ -0,0 +1,61 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/tag.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+
+template<typename DerivedPolicy,
+         typename ForwardIterator>
+  void sequence(thrust::execution_policy<DerivedPolicy> &exec,
+                ForwardIterator first,
+                ForwardIterator last);
+
+
+template<typename DerivedPolicy, typename ForwardIterator, typename T>
+  void sequence(thrust::execution_policy<DerivedPolicy> &exec,
+                ForwardIterator first,
+                ForwardIterator last,
+                T init);
+
+
+template<typename DerivedPolicy, typename ForwardIterator, typename T>
+  void sequence(thrust::execution_policy<DerivedPolicy> &exec,
+                ForwardIterator first,
+                ForwardIterator last,
+                T init,
+                T step);
+
+
+} // end namespace generic
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/detail/generic/sequence.inl>
+
diff --git a/compat/thrust/system/detail/generic/sequence.inl b/compat/thrust/system/detail/generic/sequence.inl
new file mode 100644
index 0000000000..45aec69829
--- /dev/null
+++ b/compat/thrust/system/detail/generic/sequence.inl
@@ -0,0 +1,69 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/sequence.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/functional.h>
+#include <thrust/tabulate.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+
+template<typename DerivedPolicy, typename ForwardIterator>
+  void sequence(thrust::execution_policy<DerivedPolicy> &exec,
+                ForwardIterator first,
+                ForwardIterator last)
+{
+  typedef typename thrust::iterator_traits<ForwardIterator>::value_type T;
+
+  thrust::sequence(exec, first, last, T(0));
+} // end sequence()
+
+
+template<typename DerivedPolicy, typename ForwardIterator, typename T>
+  void sequence(thrust::execution_policy<DerivedPolicy> &exec,
+                ForwardIterator first,
+                ForwardIterator last,
+                T init)
+{
+  thrust::sequence(exec, first, last, init, T(1));
+} // end sequence()
+
+
+template<typename DerivedPolicy, typename ForwardIterator, typename T>
+  void sequence(thrust::execution_policy<DerivedPolicy> &exec,
+                ForwardIterator first,
+                ForwardIterator last,
+                T init,
+                T step)
+{
+  thrust::tabulate(exec, first, last, init + step * thrust::placeholders::_1);
+} // end sequence()
+
+
+} // end namespace generic
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/detail/generic/set_operations.h b/compat/thrust/system/detail/generic/set_operations.h
new file mode 100644
index 0000000000..1ca8d391de
--- /dev/null
+++ b/compat/thrust/system/detail/generic/set_operations.h
@@ -0,0 +1,303 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/tag.h>
+#include <thrust/pair.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+
+template<typename ExecutionPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator>
+  OutputIterator set_difference(thrust::execution_policy<ExecutionPolicy> &exec,
+                                InputIterator1                             first1,
+                                InputIterator1                             last1,
+                                InputIterator2                             first2,
+                                InputIterator2                             last2,
+                                OutputIterator                             result);
+
+
+// XXX it is an error to call this function; it has no implementation
+template<typename ExecutionPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename StrictWeakOrdering>
+  OutputIterator set_difference(thrust::execution_policy<ExecutionPolicy> &exec,
+                                InputIterator1                             first1,
+                                InputIterator1                             last1,
+                                InputIterator2                             first2,
+                                InputIterator2                             last2,
+                                OutputIterator                             result,
+                                StrictWeakOrdering                         comp);
+
+
+template<typename ExecutionPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename InputIterator4,
+         typename OutputIterator1,
+         typename OutputIterator2>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    set_difference_by_key(thrust::execution_policy<ExecutionPolicy> &exec,
+                          InputIterator1                             keys_first1,
+                          InputIterator1                             keys_last1,
+                          InputIterator2                             keys_first2,
+                          InputIterator2                             keys_last2,
+                          InputIterator3                             values_first1,
+                          InputIterator4                             values_first2,
+                          OutputIterator1                            keys_result,
+                          OutputIterator2                            values_result);
+
+
+template<typename ExecutionPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename InputIterator4,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename StrictWeakOrdering>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    set_difference_by_key(thrust::execution_policy<ExecutionPolicy> &exec,
+                          InputIterator1                             keys_first1,
+                          InputIterator1                             keys_last1,
+                          InputIterator2                             keys_first2,
+                          InputIterator2                             keys_last2,
+                          InputIterator3                             values_first1,
+                          InputIterator4                             values_first2,
+                          OutputIterator1                            keys_result,
+                          OutputIterator2                            values_result,
+                          StrictWeakOrdering                         comp);
+
+
+template<typename ExecutionPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator>
+  OutputIterator set_intersection(thrust::execution_policy<ExecutionPolicy> &system,
+                                  InputIterator1                             first1,
+                                  InputIterator1                             last1,
+                                  InputIterator2                             first2,
+                                  InputIterator2                             last2,
+                                  OutputIterator                             result);
+
+
+// XXX it is an error to call this function; it has no implementation
+template<typename ExecutionPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename StrictWeakOrdering>
+  OutputIterator set_intersection(thrust::execution_policy<StrictWeakOrdering> &system,
+                                  InputIterator1                                first1,
+                                  InputIterator1                                last1,
+                                  InputIterator2                                first2,
+                                  InputIterator2                                last2,
+                                  OutputIterator                                result,
+                                  StrictWeakOrdering                            comp);
+
+
+template<typename ExecutionPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename OutputIterator1,
+         typename OutputIterator2>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    set_intersection_by_key(thrust::execution_policy<ExecutionPolicy> &system,
+                            InputIterator1                             keys_first1,
+                            InputIterator1                             keys_last1,
+                            InputIterator2                             keys_first2,
+                            InputIterator2                             keys_last2,
+                            InputIterator3                             values_first1,
+                            OutputIterator1                            keys_result,
+                            OutputIterator2                            values_result);
+
+
+template<typename ExecutionPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename StrictWeakOrdering>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    set_intersection_by_key(thrust::execution_policy<ExecutionPolicy> &system,
+                            InputIterator1                             keys_first1,
+                            InputIterator1                             keys_last1,
+                            InputIterator2                             keys_first2,
+                            InputIterator2                             keys_last2,
+                            InputIterator3                             values_first1,
+                            OutputIterator1                            keys_result,
+                            OutputIterator2                            values_result,
+                            StrictWeakOrdering                         comp);
+
+
+template<typename ExecutionPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator>
+  OutputIterator set_symmetric_difference(thrust::execution_policy<ExecutionPolicy> &system,
+                                          InputIterator1                             first1,
+                                          InputIterator1                             last1,
+                                          InputIterator2                             first2,
+                                          InputIterator2                             last2,
+                                          OutputIterator                             result);
+
+
+// XXX it is an error to call this function; it has no implementation
+template<typename ExecutionPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename StrictWeakOrdering>
+  OutputIterator set_symmetric_difference(thrust::execution_policy<ExecutionPolicy> &system,
+                                          InputIterator1                             first1,
+                                          InputIterator1                             last1,
+                                          InputIterator2                             first2,
+                                          InputIterator2                             last2,
+                                          OutputIterator                             result,
+                                          StrictWeakOrdering                         comp);
+
+
+template<typename ExecutionPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename InputIterator4,
+         typename OutputIterator1,
+         typename OutputIterator2>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    set_symmetric_difference_by_key(thrust::execution_policy<ExecutionPolicy> &system,
+                                    InputIterator1                             keys_first1,
+                                    InputIterator1                             keys_last1,
+                                    InputIterator2                             keys_first2,
+                                    InputIterator2                             keys_last2,
+                                    InputIterator3                             values_first1,
+                                    InputIterator4                             values_first2,
+                                    OutputIterator1                            keys_result,
+                                    OutputIterator2                            values_result);
+
+
+template<typename ExecutionPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename InputIterator4,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename StrictWeakOrdering>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    set_symmetric_difference_by_key(thrust::execution_policy<ExecutionPolicy> &system,
+                                    InputIterator1                             keys_first1,
+                                    InputIterator1                             keys_last1,
+                                    InputIterator2                             keys_first2,
+                                    InputIterator2                             keys_last2,
+                                    InputIterator3                             values_first1,
+                                    InputIterator4                             values_first2,
+                                    OutputIterator1                            keys_result,
+                                    OutputIterator2                            values_result,
+                                    StrictWeakOrdering                         comp);
+
+
+template<typename ExecutionPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator>
+  OutputIterator set_union(thrust::execution_policy<ExecutionPolicy> &system,
+                           InputIterator1                             first1,
+                           InputIterator1                             last1,
+                           InputIterator2                             first2,
+                           InputIterator2                             last2,
+                           OutputIterator                             result);
+
+
+// XXX it is an error to call this function; it has no implementation
+template<typename ExecutionPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename StrictWeakOrdering>
+  OutputIterator set_union(thrust::execution_policy<ExecutionPolicy> &system,
+                           InputIterator1                             first1,
+                           InputIterator1                             last1,
+                           InputIterator2                             first2,
+                           InputIterator2                             last2,
+                           OutputIterator                             result,
+                           StrictWeakOrdering                         comp);
+
+
+template<typename ExecutionPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename InputIterator4,
+         typename OutputIterator1,
+         typename OutputIterator2>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    set_union_by_key(thrust::execution_policy<ExecutionPolicy> &system,
+                     InputIterator1                             keys_first1,
+                     InputIterator1                             keys_last1,
+                     InputIterator2                             keys_first2,
+                     InputIterator2                             keys_last2,
+                     InputIterator3                             values_first1,
+                     InputIterator4                             values_first2,
+                     OutputIterator1                            keys_result,
+                     OutputIterator2                            values_result);
+
+
+template<typename ExecutionPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename InputIterator4,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename StrictWeakOrdering>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    set_union_by_key(thrust::execution_policy<ExecutionPolicy> &system,
+                     InputIterator1                             keys_first1,
+                     InputIterator1                             keys_last1,
+                     InputIterator2                             keys_first2,
+                     InputIterator2                             keys_last2,
+                     InputIterator3                             values_first1,
+                     InputIterator4                             values_first2,
+                     OutputIterator1                            keys_result,
+                     OutputIterator2                            values_result,
+                     StrictWeakOrdering                         comp);
+
+
+} // end namespace generic
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/detail/generic/set_operations.inl>
+
diff --git a/compat/thrust/system/detail/generic/set_operations.inl b/compat/thrust/system/detail/generic/set_operations.inl
new file mode 100644
index 0000000000..bac9ccd671
--- /dev/null
+++ b/compat/thrust/system/detail/generic/set_operations.inl
@@ -0,0 +1,449 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/static_assert.h>
+#include <thrust/system/detail/generic/set_operations.h>
+#include <thrust/functional.h>
+#include <thrust/detail/internal_functional.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/iterator/zip_iterator.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator>
+  OutputIterator set_difference(thrust::execution_policy<DerivedPolicy> &exec,
+                                InputIterator1                           first1,
+                                InputIterator1                           last1,
+                                InputIterator2                           first2,
+                                InputIterator2                           last2,
+                                OutputIterator                           result)
+{
+  typedef typename thrust::iterator_value<InputIterator1>::type value_type;
+  return thrust::set_difference(exec, first1, last1, first2, last2, result, thrust::less<value_type>());
+} // end set_difference()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename InputIterator4,
+         typename OutputIterator1,
+         typename OutputIterator2>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    set_difference_by_key(thrust::execution_policy<DerivedPolicy> &exec,
+                          InputIterator1                           keys_first1,
+                          InputIterator1                           keys_last1,
+                          InputIterator2                           keys_first2,
+                          InputIterator2                           keys_last2,
+                          InputIterator3                           values_first1,
+                          InputIterator4                           values_first2,
+                          OutputIterator1                          keys_result,
+                          OutputIterator2                          values_result)
+{
+  typedef typename thrust::iterator_value<InputIterator1>::type value_type;
+  return thrust::set_difference_by_key(exec, keys_first1, keys_last1, keys_first2, keys_last2, values_first1, values_first2, keys_result, values_result, thrust::less<value_type>());
+} // end set_difference_by_key()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename InputIterator4,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename StrictWeakOrdering>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    set_difference_by_key(thrust::execution_policy<DerivedPolicy> &exec,
+                          InputIterator1                           keys_first1,
+                          InputIterator1                           keys_last1,
+                          InputIterator2                           keys_first2,
+                          InputIterator2                           keys_last2,
+                          InputIterator3                           values_first1,
+                          InputIterator4                           values_first2,
+                          OutputIterator1                          keys_result,
+                          OutputIterator2                          values_result,
+                          StrictWeakOrdering                       comp)
+{
+  typedef thrust::tuple<InputIterator1, InputIterator3>   iterator_tuple1;
+  typedef thrust::tuple<InputIterator2, InputIterator4>   iterator_tuple2;
+  typedef thrust::tuple<OutputIterator1, OutputIterator2> iterator_tuple3;
+
+  typedef thrust::zip_iterator<iterator_tuple1> zip_iterator1;
+  typedef thrust::zip_iterator<iterator_tuple2> zip_iterator2;
+  typedef thrust::zip_iterator<iterator_tuple3> zip_iterator3;
+
+  zip_iterator1 zipped_first1 = thrust::make_zip_iterator(thrust::make_tuple(keys_first1, values_first1));
+  zip_iterator1 zipped_last1  = thrust::make_zip_iterator(thrust::make_tuple(keys_last1, values_first1));
+
+  zip_iterator2 zipped_first2 = thrust::make_zip_iterator(thrust::make_tuple(keys_first2, values_first2));
+  zip_iterator2 zipped_last2  = thrust::make_zip_iterator(thrust::make_tuple(keys_last2, values_first2));
+
+  zip_iterator3 zipped_result = thrust::make_zip_iterator(thrust::make_tuple(keys_result, values_result));
+
+  thrust::detail::compare_first<StrictWeakOrdering> comp_first(comp);
+
+  iterator_tuple3 result = thrust::set_difference(exec, zipped_first1, zipped_last1, zipped_first2, zipped_last2, zipped_result, comp_first).get_iterator_tuple();
+
+  return thrust::make_pair(thrust::get<0>(result), thrust::get<1>(result));
+} // end set_difference_by_key()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator>
+  OutputIterator set_intersection(thrust::execution_policy<DerivedPolicy> &exec,
+                                  InputIterator1                           first1,
+                                  InputIterator1                           last1,
+                                  InputIterator2                           first2,
+                                  InputIterator2                           last2,
+                                  OutputIterator                           result)
+{
+  typedef typename thrust::iterator_value<InputIterator1>::type value_type;
+  return thrust::set_intersection(exec, first1, last1, first2, last2, result, thrust::less<value_type>());
+} // end set_intersection()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename OutputIterator1,
+         typename OutputIterator2>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    set_intersection_by_key(thrust::execution_policy<DerivedPolicy> &exec,
+                            InputIterator1                           keys_first1,
+                            InputIterator1                           keys_last1,
+                            InputIterator2                           keys_first2,
+                            InputIterator2                           keys_last2,
+                            InputIterator3                           values_first1,
+                            OutputIterator1                          keys_result,
+                            OutputIterator2                          values_result)
+{
+  typedef typename thrust::iterator_value<InputIterator1>::type value_type;
+  return thrust::set_intersection_by_key(exec, keys_first1, keys_last1, keys_first2, keys_last2, values_first1, keys_result, values_result, thrust::less<value_type>());
+} // end set_intersection_by_key()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename StrictWeakOrdering>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    set_intersection_by_key(thrust::execution_policy<DerivedPolicy> &exec,
+                            InputIterator1                           keys_first1,
+                            InputIterator1                           keys_last1,
+                            InputIterator2                           keys_first2,
+                            InputIterator2                           keys_last2,
+                            InputIterator3                           values_first1,
+                            OutputIterator1                          keys_result,
+                            OutputIterator2                          values_result,
+                            StrictWeakOrdering                       comp)
+{
+  typedef thrust::tuple<InputIterator1, InputIterator3>   iterator_tuple1;
+  typedef thrust::tuple<InputIterator2, InputIterator2>   iterator_tuple2;
+  typedef thrust::tuple<OutputIterator1, OutputIterator2> iterator_tuple3;
+
+  typedef thrust::zip_iterator<iterator_tuple1> zip_iterator1;
+  typedef thrust::zip_iterator<iterator_tuple2> zip_iterator2;
+  typedef thrust::zip_iterator<iterator_tuple3> zip_iterator3;
+
+  // fabricate a values_first2 by "sending" keys twice
+  // it should never be dereferenced by set_intersection
+  InputIterator2 values_first2 = keys_first2;
+
+  zip_iterator1 zipped_first1 = thrust::make_zip_iterator(thrust::make_tuple(keys_first1, values_first1));
+  zip_iterator1 zipped_last1  = thrust::make_zip_iterator(thrust::make_tuple(keys_last1, values_first1));
+
+  zip_iterator2 zipped_first2 = thrust::make_zip_iterator(thrust::make_tuple(keys_first2, values_first2));
+  zip_iterator2 zipped_last2  = thrust::make_zip_iterator(thrust::make_tuple(keys_last2, values_first2));
+
+  zip_iterator3 zipped_result = thrust::make_zip_iterator(thrust::make_tuple(keys_result, values_result));
+
+  thrust::detail::compare_first<StrictWeakOrdering> comp_first(comp);
+
+  iterator_tuple3 result = thrust::set_intersection(exec, zipped_first1, zipped_last1, zipped_first2, zipped_last2, zipped_result, comp_first).get_iterator_tuple();
+
+  return thrust::make_pair(thrust::get<0>(result), thrust::get<1>(result));
+} // end set_intersection_by_key()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator>
+  OutputIterator set_symmetric_difference(thrust::execution_policy<DerivedPolicy> &exec,
+                                          InputIterator1                           first1,
+                                          InputIterator1                           last1,
+                                          InputIterator2                           first2,
+                                          InputIterator2                           last2,
+                                          OutputIterator                           result)
+{
+  typedef typename thrust::iterator_value<InputIterator1>::type value_type;
+  return thrust::set_symmetric_difference(exec, first1, last1, first2, last2, result, thrust::less<value_type>());
+} // end set_symmetric_difference()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename InputIterator4,
+         typename OutputIterator1,
+         typename OutputIterator2>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    set_symmetric_difference_by_key(thrust::execution_policy<DerivedPolicy> &exec,
+                                    InputIterator1                           keys_first1,
+                                    InputIterator1                           keys_last1,
+                                    InputIterator2                           keys_first2,
+                                    InputIterator2                           keys_last2,
+                                    InputIterator3                           values_first1,
+                                    InputIterator4                           values_first2,
+                                    OutputIterator1                          keys_result,
+                                    OutputIterator2                          values_result)
+{
+  typedef typename thrust::iterator_value<InputIterator1>::type value_type;
+  return thrust::set_symmetric_difference_by_key(exec, keys_first1, keys_last1, keys_first2, keys_last2, values_first1, values_first2, keys_result, values_result, thrust::less<value_type>());
+} // end set_symmetric_difference_by_key()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename InputIterator4,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename StrictWeakOrdering>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    set_symmetric_difference_by_key(thrust::execution_policy<DerivedPolicy> &exec,
+                                    InputIterator1                           keys_first1,
+                                    InputIterator1                           keys_last1,
+                                    InputIterator2                           keys_first2,
+                                    InputIterator2                           keys_last2,
+                                    InputIterator3                           values_first1,
+                                    InputIterator4                           values_first2,
+                                    OutputIterator1                          keys_result,
+                                    OutputIterator2                          values_result,
+                                    StrictWeakOrdering                       comp)
+{
+  typedef thrust::tuple<InputIterator1, InputIterator3>   iterator_tuple1;
+  typedef thrust::tuple<InputIterator2, InputIterator4>   iterator_tuple2;
+  typedef thrust::tuple<OutputIterator1, OutputIterator2> iterator_tuple3;
+
+  typedef thrust::zip_iterator<iterator_tuple1> zip_iterator1;
+  typedef thrust::zip_iterator<iterator_tuple2> zip_iterator2;
+  typedef thrust::zip_iterator<iterator_tuple3> zip_iterator3;
+
+  zip_iterator1 zipped_first1 = thrust::make_zip_iterator(thrust::make_tuple(keys_first1, values_first1));
+  zip_iterator1 zipped_last1  = thrust::make_zip_iterator(thrust::make_tuple(keys_last1, values_first1));
+
+  zip_iterator2 zipped_first2 = thrust::make_zip_iterator(thrust::make_tuple(keys_first2, values_first2));
+  zip_iterator2 zipped_last2  = thrust::make_zip_iterator(thrust::make_tuple(keys_last2, values_first2));
+
+  zip_iterator3 zipped_result = thrust::make_zip_iterator(thrust::make_tuple(keys_result, values_result));
+
+  thrust::detail::compare_first<StrictWeakOrdering> comp_first(comp);
+
+  iterator_tuple3 result = thrust::set_symmetric_difference(exec, zipped_first1, zipped_last1, zipped_first2, zipped_last2, zipped_result, comp_first).get_iterator_tuple();
+
+  return thrust::make_pair(thrust::get<0>(result), thrust::get<1>(result));
+} // end set_symmetric_difference_by_key()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator>
+  OutputIterator set_union(thrust::execution_policy<DerivedPolicy> &exec,
+                           InputIterator1                           first1,
+                           InputIterator1                           last1,
+                           InputIterator2                           first2,
+                           InputIterator2                           last2,
+                           OutputIterator                           result)
+{
+  typedef typename thrust::iterator_value<InputIterator1>::type value_type;
+  return thrust::set_union(exec, first1, last1, first2, last2, result, thrust::less<value_type>());
+} // end set_union()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename InputIterator4,
+         typename OutputIterator1,
+         typename OutputIterator2>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    set_union_by_key(thrust::execution_policy<DerivedPolicy> &exec,
+                     InputIterator1                           keys_first1,
+                     InputIterator1                           keys_last1,
+                     InputIterator2                           keys_first2,
+                     InputIterator2                           keys_last2,
+                     InputIterator3                           values_first1,
+                     InputIterator4                           values_first2,
+                     OutputIterator1                          keys_result,
+                     OutputIterator2                          values_result)
+{
+  typedef typename thrust::iterator_value<InputIterator1>::type value_type;
+  return thrust::set_union_by_key(exec, keys_first1, keys_last1, keys_first2, keys_last2, values_first1, values_first2, keys_result, values_result, thrust::less<value_type>());
+} // end set_union_by_key()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename InputIterator4,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename StrictWeakOrdering>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    set_union_by_key(thrust::execution_policy<DerivedPolicy> &exec,
+                     InputIterator1                           keys_first1,
+                     InputIterator1                           keys_last1,
+                     InputIterator2                           keys_first2,
+                     InputIterator2                           keys_last2,
+                     InputIterator3                           values_first1,
+                     InputIterator4                           values_first2,
+                     OutputIterator1                          keys_result,
+                     OutputIterator2                          values_result,
+                     StrictWeakOrdering                       comp)
+{
+  typedef thrust::tuple<InputIterator1, InputIterator3>   iterator_tuple1;
+  typedef thrust::tuple<InputIterator2, InputIterator4>   iterator_tuple2;
+  typedef thrust::tuple<OutputIterator1, OutputIterator2> iterator_tuple3;
+
+  typedef thrust::zip_iterator<iterator_tuple1> zip_iterator1;
+  typedef thrust::zip_iterator<iterator_tuple2> zip_iterator2;
+  typedef thrust::zip_iterator<iterator_tuple3> zip_iterator3;
+
+  zip_iterator1 zipped_first1 = thrust::make_zip_iterator(thrust::make_tuple(keys_first1, values_first1));
+  zip_iterator1 zipped_last1  = thrust::make_zip_iterator(thrust::make_tuple(keys_last1, values_first1));
+
+  zip_iterator2 zipped_first2 = thrust::make_zip_iterator(thrust::make_tuple(keys_first2, values_first2));
+  zip_iterator2 zipped_last2  = thrust::make_zip_iterator(thrust::make_tuple(keys_last2, values_first2));
+
+  zip_iterator3 zipped_result = thrust::make_zip_iterator(thrust::make_tuple(keys_result, values_result));
+
+  thrust::detail::compare_first<StrictWeakOrdering> comp_first(comp);
+
+  iterator_tuple3 result = thrust::set_union(exec, zipped_first1, zipped_last1, zipped_first2, zipped_last2, zipped_result, comp_first).get_iterator_tuple();
+
+  return thrust::make_pair(thrust::get<0>(result), thrust::get<1>(result));
+} // end set_union_by_key()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename StrictWeakOrdering>
+  OutputIterator set_difference(thrust::execution_policy<DerivedPolicy> &exec,
+                                InputIterator1                           first1,
+                                InputIterator1                           last1,
+                                InputIterator2                           first2,
+                                InputIterator2                           last2,
+                                OutputIterator                           result,
+                                StrictWeakOrdering                       comp)
+{
+  // unimplemented primitive
+  THRUST_STATIC_ASSERT( (thrust::detail::depend_on_instantiation<InputIterator1, false>::value) );
+  return result;
+} // end set_difference()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename StrictWeakOrdering>
+  OutputIterator set_intersection(thrust::execution_policy<DerivedPolicy> &exec,
+                                  InputIterator1                           first1,
+                                  InputIterator1                           last1,
+                                  InputIterator2                           first2,
+                                  InputIterator2                           last2,
+                                  OutputIterator                           result,
+                                  StrictWeakOrdering                       comp)
+{
+  // unimplemented primitive
+  THRUST_STATIC_ASSERT( (thrust::detail::depend_on_instantiation<InputIterator1, false>::value) );
+  return result;
+} // end set_intersection()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename StrictWeakOrdering>
+  OutputIterator set_symmetric_difference(thrust::execution_policy<DerivedPolicy> &exec,
+                                          InputIterator1                           first1,
+                                          InputIterator1                           last1,
+                                          InputIterator2                           first2,
+                                          InputIterator2                           last2,
+                                          OutputIterator                           result,
+                                          StrictWeakOrdering                       comp)
+{
+  // unimplemented primitive
+  THRUST_STATIC_ASSERT( (thrust::detail::depend_on_instantiation<InputIterator1, false>::value) );
+  return result;
+} // end set_symmetric_difference()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename StrictWeakOrdering>
+  OutputIterator set_union(thrust::execution_policy<DerivedPolicy> &exec,
+                           InputIterator1                           first1,
+                           InputIterator1                           last1,
+                           InputIterator2                           first2,
+                           InputIterator2                           last2,
+                           OutputIterator                           result,
+                           StrictWeakOrdering                       comp)
+{
+  // unimplemented primitive
+  THRUST_STATIC_ASSERT( (thrust::detail::depend_on_instantiation<InputIterator1, false>::value) );
+  return result;
+} // end set_union()
+
+
+} // end namespace generic
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/detail/generic/sort.h b/compat/thrust/system/detail/generic/sort.h
new file mode 100644
index 0000000000..5498708a3b
--- /dev/null
+++ b/compat/thrust/system/detail/generic/sort.h
@@ -0,0 +1,142 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/tag.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator>
+  void sort(thrust::execution_policy<DerivedPolicy> &exec,
+            RandomAccessIterator first,
+            RandomAccessIterator last);
+
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator,
+         typename StrictWeakOrdering>
+  void sort(thrust::execution_policy<DerivedPolicy> &exec,
+            RandomAccessIterator first,
+            RandomAccessIterator last,
+            StrictWeakOrdering comp);
+
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2>
+  void sort_by_key(thrust::execution_policy<DerivedPolicy> &exec,
+                   RandomAccessIterator1 keys_first,
+                   RandomAccessIterator1 keys_last,
+                   RandomAccessIterator2 values_first);
+
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2,
+         typename StrictWeakOrdering>
+  void sort_by_key(thrust::execution_policy<DerivedPolicy> &exec,
+                   RandomAccessIterator1 keys_first,
+                   RandomAccessIterator1 keys_last,
+                   RandomAccessIterator2 values_first,
+                   StrictWeakOrdering comp);
+
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator>
+  void stable_sort(thrust::execution_policy<DerivedPolicy> &exec,
+                   RandomAccessIterator first,
+                   RandomAccessIterator last);
+
+
+// XXX it is an error to call this function; it has no implementation
+template<typename DerivedPolicy,
+         typename RandomAccessIterator,
+         typename StrictWeakOrdering>
+  void stable_sort(thrust::execution_policy<DerivedPolicy> &exec,
+                   RandomAccessIterator first,
+                   RandomAccessIterator last,
+                   StrictWeakOrdering comp);
+
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2>
+  void stable_sort_by_key(thrust::execution_policy<DerivedPolicy> &exec,
+                          RandomAccessIterator1 keys_first,
+                          RandomAccessIterator1 keys_last,
+                          RandomAccessIterator2 values_first);
+
+
+// XXX it is an error to call this function; it has no implementation
+template<typename DerivedPolicy,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2,
+         typename StrictWeakOrdering>
+  void stable_sort_by_key(thrust::execution_policy<DerivedPolicy> &exec,
+                          RandomAccessIterator1 keys_first,
+                          RandomAccessIterator1 keys_last,
+                          RandomAccessIterator2 values_first,
+                          StrictWeakOrdering comp);
+
+
+template<typename DerivedPolicy, typename ForwardIterator>
+  bool is_sorted(thrust::execution_policy<DerivedPolicy> &exec,
+                 ForwardIterator first,
+                 ForwardIterator last);
+
+
+template<typename DerivedPolicy,
+         typename ForwardIterator,
+         typename Compare>
+  bool is_sorted(thrust::execution_policy<DerivedPolicy> &exec,
+                 ForwardIterator first,
+                 ForwardIterator last,
+                 Compare comp);
+
+
+template<typename DerivedPolicy, typename ForwardIterator>
+  ForwardIterator is_sorted_until(thrust::execution_policy<DerivedPolicy> &exec,
+                                  ForwardIterator first,
+                                  ForwardIterator last);
+
+
+template<typename DerivedPolicy,
+         typename ForwardIterator,
+         typename Compare>
+  ForwardIterator is_sorted_until(thrust::execution_policy<DerivedPolicy> &exec,
+                                  ForwardIterator first,
+                                  ForwardIterator last,
+                                  Compare comp);
+
+
+} // end generic
+} // end detail
+} // end system
+} // end thrust
+
+#include <thrust/system/detail/generic/sort.inl>
+
diff --git a/compat/thrust/system/detail/generic/sort.inl b/compat/thrust/system/detail/generic/sort.inl
new file mode 100644
index 0000000000..aabb2eed81
--- /dev/null
+++ b/compat/thrust/system/detail/generic/sort.inl
@@ -0,0 +1,202 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/sort.h>
+#include <thrust/functional.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/distance.h>
+#include <thrust/functional.h>
+#include <thrust/find.h>
+#include <thrust/iterator/zip_iterator.h>
+#include <thrust/tuple.h>
+#include <thrust/detail/internal_functional.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+
+template<typename ExecutionPolicy,
+         typename RandomAccessIterator>
+  void sort(thrust::execution_policy<ExecutionPolicy> &exec,
+            RandomAccessIterator first,
+            RandomAccessIterator last)
+{
+  typedef typename thrust::iterator_value<RandomAccessIterator>::type value_type; 
+  thrust::sort(exec, first, last, thrust::less<value_type>());
+} // end sort()
+
+
+template<typename ExecutionPolicy,
+         typename RandomAccessIterator,
+         typename StrictWeakOrdering>
+  void sort(thrust::execution_policy<ExecutionPolicy> &exec,
+            RandomAccessIterator first,
+            RandomAccessIterator last,
+            StrictWeakOrdering comp)
+{
+  // implement with stable_sort
+  thrust::stable_sort(exec, first, last, comp);
+} // end sort()
+
+
+template<typename ExecutionPolicy,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2>
+  void sort_by_key(thrust::execution_policy<ExecutionPolicy> &exec,
+                   RandomAccessIterator1 keys_first,
+                   RandomAccessIterator1 keys_last,
+                   RandomAccessIterator2 values_first)
+{
+  typedef typename thrust::iterator_value<RandomAccessIterator1>::type value_type;
+  thrust::sort_by_key(exec, keys_first, keys_last, values_first, thrust::less<value_type>());
+} // end sort_by_key()
+
+
+template<typename ExecutionPolicy,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2,
+         typename StrictWeakOrdering>
+  void sort_by_key(thrust::execution_policy<ExecutionPolicy> &exec,
+                   RandomAccessIterator1 keys_first,
+                   RandomAccessIterator1 keys_last,
+                   RandomAccessIterator2 values_first,
+                   StrictWeakOrdering comp)
+{
+  // implement with stable_sort_by_key
+  thrust::stable_sort_by_key(exec, keys_first, keys_last, values_first, comp);
+} // end sort_by_key()
+
+
+template<typename ExecutionPolicy,
+         typename RandomAccessIterator>
+  void stable_sort(thrust::execution_policy<ExecutionPolicy> &exec,
+                   RandomAccessIterator first,
+                   RandomAccessIterator last)
+{
+  typedef typename thrust::iterator_value<RandomAccessIterator>::type value_type;
+  thrust::stable_sort(exec, first, last, thrust::less<value_type>());
+} // end stable_sort()
+
+
+template<typename ExecutionPolicy,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2>
+  void stable_sort_by_key(thrust::execution_policy<ExecutionPolicy> &exec,
+                          RandomAccessIterator1 keys_first,
+                          RandomAccessIterator1 keys_last,
+                          RandomAccessIterator2 values_first)
+{
+  typedef typename iterator_value<RandomAccessIterator1>::type value_type;
+  thrust::stable_sort_by_key(exec, keys_first, keys_last, values_first, thrust::less<value_type>());
+} // end stable_sort_by_key()
+
+
+template<typename ExecutionPolicy, typename ForwardIterator>
+  bool is_sorted(thrust::execution_policy<ExecutionPolicy> &exec,
+                 ForwardIterator first,
+                 ForwardIterator last)
+{
+  return thrust::is_sorted_until(exec, first, last) == last;
+} // end is_sorted()
+
+
+template<typename ExecutionPolicy,
+         typename ForwardIterator,
+         typename Compare>
+  bool is_sorted(thrust::execution_policy<ExecutionPolicy> &exec,
+                 ForwardIterator first,
+                 ForwardIterator last,
+                 Compare comp)
+{
+  return thrust::is_sorted_until(exec, first, last, comp) == last;
+} // end is_sorted()
+
+
+template<typename ExecutionPolicy, typename ForwardIterator>
+  ForwardIterator is_sorted_until(thrust::execution_policy<ExecutionPolicy> &exec,
+                                  ForwardIterator first,
+                                  ForwardIterator last)
+{
+  typedef typename thrust::iterator_value<ForwardIterator>::type InputType;
+
+  return thrust::is_sorted_until(exec, first, last, thrust::less<InputType>());
+} // end is_sorted_until()
+
+
+template<typename ExecutionPolicy,
+         typename ForwardIterator,
+         typename Compare>
+  ForwardIterator is_sorted_until(thrust::execution_policy<ExecutionPolicy> &exec,
+                                  ForwardIterator first,
+                                  ForwardIterator last,
+                                  Compare comp)
+{
+  if(thrust::distance(first,last) < 2) return last;
+
+  typedef thrust::tuple<ForwardIterator,ForwardIterator> IteratorTuple;
+  typedef thrust::zip_iterator<IteratorTuple>            ZipIterator;
+
+  ForwardIterator first_plus_one = first;
+  thrust::advance(first_plus_one, 1);
+
+  ZipIterator zipped_first = thrust::make_zip_iterator(thrust::make_tuple(first_plus_one, first));
+  ZipIterator zipped_last  = thrust::make_zip_iterator(thrust::make_tuple(last, first));
+
+  return thrust::get<0>(thrust::find_if(exec, zipped_first, zipped_last, thrust::detail::tuple_binary_predicate<Compare>(comp)).get_iterator_tuple());
+} // end is_sorted_until()
+
+
+template<typename RandomAccessIterator,
+         typename StrictWeakOrdering>
+  void stable_sort(tag,
+                   RandomAccessIterator first,
+                   RandomAccessIterator last,
+                   StrictWeakOrdering comp)
+{
+  // unimplemented primitive
+  THRUST_STATIC_ASSERT( (thrust::detail::depend_on_instantiation<RandomAccessIterator, false>::value) );
+} // end stable_sort()
+
+
+template<typename RandomAccessIterator1,
+         typename RandomAccessIterator2,
+         typename StrictWeakOrdering>
+  void stable_sort_by_key(tag,
+                          RandomAccessIterator1 keys_first,
+                          RandomAccessIterator1 keys_last,
+                          RandomAccessIterator2 values_first,
+                          StrictWeakOrdering comp)
+{
+  // unimplemented primitive
+  THRUST_STATIC_ASSERT( (thrust::detail::depend_on_instantiation<RandomAccessIterator1, false>::value) );
+} // end stable_sort_by_key()
+
+
+} // end generic
+} // end detail
+} // end system
+} // end thrust
+
diff --git a/compat/thrust/system/detail/generic/swap_ranges.h b/compat/thrust/system/detail/generic/swap_ranges.h
new file mode 100644
index 0000000000..5d640d3feb
--- /dev/null
+++ b/compat/thrust/system/detail/generic/swap_ranges.h
@@ -0,0 +1,46 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/tag.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+template<typename DerivedPolicy,
+         typename ForwardIterator1,
+         typename ForwardIterator2>
+  ForwardIterator2 swap_ranges(thrust::execution_policy<DerivedPolicy> &exec,
+                               ForwardIterator1 first1,
+                               ForwardIterator1 last1,
+                               ForwardIterator2 first2);
+
+} // end namespace generic
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/detail/generic/swap_ranges.inl>
+
diff --git a/compat/thrust/system/detail/generic/swap_ranges.inl b/compat/thrust/system/detail/generic/swap_ranges.inl
new file mode 100644
index 0000000000..0e12d07627
--- /dev/null
+++ b/compat/thrust/system/detail/generic/swap_ranges.inl
@@ -0,0 +1,73 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/swap_ranges.h>
+#include <thrust/tuple.h>
+#include <thrust/iterator/zip_iterator.h>
+#include <thrust/detail/internal_functional.h>
+#include <thrust/for_each.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+namespace detail
+{
+
+// XXX define this here rather than in internal_functional.h
+// to avoid circular dependence between swap.h & internal_functional.h
+struct swap_pair_elements
+{
+  template <typename Tuple>
+  __host__ __device__
+  void operator()(Tuple t)
+  {
+    // use unqualified swap to allow ADL to catch any user-defined swap
+    using thrust::swap;
+    swap(thrust::get<0>(t), thrust::get<1>(t));
+  }
+}; // end swap_pair_elements
+
+} // end detail
+
+template<typename DerivedPolicy,
+         typename ForwardIterator1,
+         typename ForwardIterator2>
+  ForwardIterator2 swap_ranges(thrust::execution_policy<DerivedPolicy> &exec,
+                               ForwardIterator1 first1,
+                               ForwardIterator1 last1,
+                               ForwardIterator2 first2)
+{
+  typedef thrust::tuple<ForwardIterator1,ForwardIterator2> IteratorTuple;
+  typedef thrust::zip_iterator<IteratorTuple>              ZipIterator;
+
+  ZipIterator result = thrust::for_each(exec,
+                                        thrust::make_zip_iterator(thrust::make_tuple(first1, first2)),
+                                        thrust::make_zip_iterator(thrust::make_tuple(last1,  first2)),
+                                        detail::swap_pair_elements());
+  return thrust::get<1>(result.get_iterator_tuple());
+} // end swap_ranges()
+
+} // end generic
+} // end detail
+} // end system
+} // end thrust
+
diff --git a/compat/thrust/system/detail/generic/tabulate.h b/compat/thrust/system/detail/generic/tabulate.h
new file mode 100644
index 0000000000..e5911b14ec
--- /dev/null
+++ b/compat/thrust/system/detail/generic/tabulate.h
@@ -0,0 +1,55 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/tag.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+template<typename DerivedPolicy,
+         typename ForwardIterator,
+         typename UnaryOperation>
+  void tabulate(thrust::execution_policy<DerivedPolicy> &exec,
+                ForwardIterator first,
+                ForwardIterator last,
+                UnaryOperation unary_op);
+
+template<typename DerivedPolicy,
+         typename OutputIterator,
+         typename Size,
+         typename UnaryOperation>
+  OutputIterator tabulate_n(thrust::execution_policy<DerivedPolicy> &exec,
+                            OutputIterator first,
+                            Size n,
+                            UnaryOperation unary_op);
+
+} // end namespace generic
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/detail/generic/tabulate.inl>
+
diff --git a/compat/thrust/system/detail/generic/tabulate.inl b/compat/thrust/system/detail/generic/tabulate.inl
new file mode 100644
index 0000000000..d2ffc26df6
--- /dev/null
+++ b/compat/thrust/system/detail/generic/tabulate.inl
@@ -0,0 +1,59 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/tabulate.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/transform.h>
+#include <thrust/distance.h>
+#include <thrust/iterator/counting_iterator.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+
+template<typename DerivedPolicy,
+         typename ForwardIterator,
+         typename UnaryOperation>
+  void tabulate(thrust::execution_policy<DerivedPolicy> &exec,
+                ForwardIterator first,
+                ForwardIterator last,
+                UnaryOperation unary_op)
+{
+  typedef typename iterator_difference<ForwardIterator>::type difference_type;
+
+  // by default, counting_iterator uses a 64b difference_type on 32b platforms to avoid overflowing its counter.
+  // this causes problems when a zip_iterator is created in transform's implementation -- ForwardIterator is
+  // incremented by a 64b difference_type and some compilers warn
+  // to avoid this, specify the counting_iterator's difference_type to be the same as ForwardIterator's.
+  thrust::counting_iterator<difference_type, thrust::use_default, thrust::use_default, difference_type> iter(0);
+
+  thrust::transform(exec, iter, iter + thrust::distance(first, last), first, unary_op);
+} // end tabulate()
+
+
+} // end namespace generic
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
+
diff --git a/compat/thrust/system/detail/generic/tag.h b/compat/thrust/system/detail/generic/tag.h
new file mode 100644
index 0000000000..577d6a37a7
--- /dev/null
+++ b/compat/thrust/system/detail/generic/tag.h
@@ -0,0 +1,48 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file generic/tag.h
+ *  \brief Implementation of the generic backend's tag.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+// tag exists only to make the generic entry points the least priority match
+// during ADL. tag should not be derived from and is constructible from anything
+struct tag
+{
+  template<typename T>
+  __host__ __device__ inline
+  tag(const T &) {}
+};
+
+} // end generic
+} // end detail
+} // end system
+} // end thrust
+
diff --git a/compat/thrust/system/detail/generic/temporary_buffer.h b/compat/thrust/system/detail/generic/temporary_buffer.h
new file mode 100644
index 0000000000..8cb08b06ae
--- /dev/null
+++ b/compat/thrust/system/detail/generic/temporary_buffer.h
@@ -0,0 +1,49 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/tag.h>
+#include <thrust/pair.h>
+#include <thrust/detail/pointer.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+
+template<typename T, typename DerivedPolicy>
+  thrust::pair<thrust::pointer<T,DerivedPolicy>, typename thrust::pointer<T,DerivedPolicy>::difference_type>
+    get_temporary_buffer(thrust::execution_policy<DerivedPolicy> &exec, typename thrust::pointer<T,DerivedPolicy>::difference_type n);
+
+
+template<typename DerivedPolicy, typename Pointer>
+  void return_temporary_buffer(thrust::execution_policy<DerivedPolicy> &exec, Pointer p);
+
+
+} // end generic
+} // end detail
+} // end system
+} // end thrust
+
+#include <thrust/system/detail/generic/temporary_buffer.inl>
+
diff --git a/compat/thrust/system/detail/generic/temporary_buffer.inl b/compat/thrust/system/detail/generic/temporary_buffer.inl
new file mode 100644
index 0000000000..0a6be7ee08
--- /dev/null
+++ b/compat/thrust/system/detail/generic/temporary_buffer.inl
@@ -0,0 +1,60 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/temporary_buffer.h>
+#include <thrust/detail/pointer.h>
+#include <thrust/detail/malloc_and_free.h>
+#include <thrust/pair.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+
+template<typename T, typename DerivedPolicy>
+  thrust::pair<thrust::pointer<T,DerivedPolicy>, typename thrust::pointer<T,DerivedPolicy>::difference_type>
+    get_temporary_buffer(thrust::execution_policy<DerivedPolicy> &exec, typename thrust::pointer<T,DerivedPolicy>::difference_type n)
+{
+  thrust::pointer<T,DerivedPolicy> ptr = thrust::malloc<T>(exec, n);
+
+  // check for a failed malloc
+  if(!ptr.get())
+  {
+    n = 0;
+  } // end if
+
+  return thrust::make_pair(ptr, n);
+} // end get_temporary_buffer()
+
+
+template<typename DerivedPolicy, typename Pointer>
+  void return_temporary_buffer(thrust::execution_policy<DerivedPolicy> &exec, Pointer p)
+{
+  thrust::free(exec, p);
+} // end return_temporary_buffer()
+
+
+} // end generic
+} // end detail
+} // end system
+} // end thrust
+
diff --git a/compat/thrust/system/detail/generic/transform.h b/compat/thrust/system/detail/generic/transform.h
new file mode 100644
index 0000000000..e98d40291e
--- /dev/null
+++ b/compat/thrust/system/detail/generic/transform.h
@@ -0,0 +1,101 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/tag.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator,
+         typename UnaryFunction>
+  OutputIterator transform(thrust::execution_policy<DerivedPolicy> &exec,
+                           InputIterator first,
+                           InputIterator last,
+                           OutputIterator result,
+                           UnaryFunction op);
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename BinaryFunction>
+  OutputIterator transform(thrust::execution_policy<DerivedPolicy> &exec,
+                           InputIterator1 first1,
+                           InputIterator1 last1,
+                           InputIterator2 first2,
+                           OutputIterator result,
+                           BinaryFunction op);
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename ForwardIterator,
+         typename UnaryFunction,
+         typename Predicate>
+  ForwardIterator transform_if(thrust::execution_policy<DerivedPolicy> &exec,
+                               InputIterator first,
+                               InputIterator last,
+                               ForwardIterator result,
+                               UnaryFunction unary_op,
+                               Predicate pred);
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename ForwardIterator,
+         typename UnaryFunction,
+         typename Predicate>
+  ForwardIterator transform_if(thrust::execution_policy<DerivedPolicy> &exec,
+                               InputIterator1 first,
+                               InputIterator1 last,
+                               InputIterator2 stencil,
+                               ForwardIterator result,
+                               UnaryFunction unary_op,
+                               Predicate pred);
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename ForwardIterator,
+         typename BinaryFunction,
+         typename Predicate>
+  ForwardIterator transform_if(thrust::execution_policy<DerivedPolicy> &exec,
+                               InputIterator1 first1,
+                               InputIterator1 last1,
+                               InputIterator2 first2,
+                               InputIterator3 stencil,
+                               ForwardIterator result,
+                               BinaryFunction binary_op,
+                               Predicate pred);
+
+} // end namespace generic
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/detail/generic/transform.inl>
+
diff --git a/compat/thrust/system/detail/generic/transform.inl b/compat/thrust/system/detail/generic/transform.inl
new file mode 100644
index 0000000000..8f0995328c
--- /dev/null
+++ b/compat/thrust/system/detail/generic/transform.inl
@@ -0,0 +1,214 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/transform.h>
+#include <thrust/for_each.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/iterator/detail/minimum_system.h>
+#include <thrust/tuple.h>
+#include <thrust/iterator/zip_iterator.h>
+#include <thrust/detail/internal_functional.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator,
+         typename UnaryFunction>
+  OutputIterator transform(thrust::execution_policy<DerivedPolicy> &exec,
+                           InputIterator first,
+                           InputIterator last,
+                           OutputIterator result,
+                           UnaryFunction op)
+{
+  // XXX WAR the problem of a generic __host__ __device__ functor's inability to invoke
+  //     a function which is only __host__ or __device__ by selecting a generic functor
+  //     which is one or the other
+  //     when nvcc is able to deal with this, remove this WAR
+  
+  // given the minimal system, determine the unary transform functor we need
+  typedef typename thrust::detail::unary_transform_functor<DerivedPolicy,UnaryFunction>::type UnaryTransformFunctor;
+
+  // make an iterator tuple
+  typedef thrust::tuple<InputIterator,OutputIterator> IteratorTuple;
+  typedef thrust::zip_iterator<IteratorTuple> ZipIterator;
+
+  ZipIterator zipped_result =
+    thrust::for_each(exec,
+                     thrust::make_zip_iterator(thrust::make_tuple(first,result)),
+                     thrust::make_zip_iterator(thrust::make_tuple(last,result)),
+                     UnaryTransformFunctor(op));
+
+  return thrust::get<1>(zipped_result.get_iterator_tuple());
+} // end transform()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename BinaryFunction>
+  OutputIterator transform(thrust::execution_policy<DerivedPolicy> &exec,
+                           InputIterator1 first1,
+                           InputIterator1 last1,
+                           InputIterator2 first2,
+                           OutputIterator result,
+                           BinaryFunction op)
+{
+  // XXX WAR the problem of a generic __host__ __device__ functor's inability to invoke
+  //     a function which is only __host__ or __device__ by selecting a generic functor
+  //     which is one or the other
+  //     when nvcc is able to deal with this, remove this WAR
+  
+  // given the minimal system, determine the binary transform functor we need
+  typedef typename thrust::detail::binary_transform_functor<DerivedPolicy,BinaryFunction>::type BinaryTransformFunctor;
+
+  // make an iterator tuple
+  typedef thrust::tuple<InputIterator1,InputIterator2,OutputIterator> IteratorTuple;
+  typedef thrust::zip_iterator<IteratorTuple> ZipIterator;
+
+  ZipIterator zipped_result =
+    thrust::for_each(exec,
+                     thrust::make_zip_iterator(thrust::make_tuple(first1,first2,result)),
+                     thrust::make_zip_iterator(thrust::make_tuple(last1,first2,result)),
+                     BinaryTransformFunctor(op));
+
+  return thrust::get<2>(zipped_result.get_iterator_tuple());
+} // end transform()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename ForwardIterator,
+         typename UnaryFunction,
+         typename Predicate>
+  ForwardIterator transform_if(thrust::execution_policy<DerivedPolicy> &exec,
+                               InputIterator first,
+                               InputIterator last,
+                               ForwardIterator result,
+                               UnaryFunction unary_op,
+                               Predicate pred)
+{
+  // XXX WAR the problem of a generic __host__ __device__ functor's inability to invoke
+  //     a function which is only __host__ or __device__ by selecting a generic functor
+  //     which is one or the other
+  //     when nvcc is able to deal with this, remove this WAR
+  
+  // given the minimal system, determine the unary transform_if functor we need
+  typedef typename thrust::detail::unary_transform_if_functor<DerivedPolicy,UnaryFunction,Predicate>::type UnaryTransformIfFunctor;
+
+  // make an iterator tuple
+  typedef thrust::tuple<InputIterator,ForwardIterator> IteratorTuple;
+  typedef thrust::zip_iterator<IteratorTuple> ZipIterator;
+
+  ZipIterator zipped_result =
+    thrust::for_each(exec,
+                     thrust::make_zip_iterator(thrust::make_tuple(first,result)),
+                     thrust::make_zip_iterator(thrust::make_tuple(last,result)),
+                     UnaryTransformIfFunctor(unary_op,pred));
+
+  return thrust::get<1>(zipped_result.get_iterator_tuple());
+} // end transform_if()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename ForwardIterator,
+         typename UnaryFunction,
+         typename Predicate>
+  ForwardIterator transform_if(thrust::execution_policy<DerivedPolicy> &exec,
+                               InputIterator1 first,
+                               InputIterator1 last,
+                               InputIterator2 stencil,
+                               ForwardIterator result,
+                               UnaryFunction unary_op,
+                               Predicate pred)
+{
+  // XXX WAR the problem of a generic __host__ __device__ functor's inability to invoke
+  //     a function which is only __host__ or __device__ by selecting a generic functor
+  //     which is one or the other
+  //     when nvcc is able to deal with this, remove this WAR
+  
+  // given the minimal system, determine the unary transform_if functor we need
+  typedef typename thrust::detail::unary_transform_if_with_stencil_functor<DerivedPolicy,UnaryFunction,Predicate>::type UnaryTransformIfFunctor;
+
+  // make an iterator tuple
+  typedef thrust::tuple<InputIterator1,InputIterator2,ForwardIterator> IteratorTuple;
+  typedef thrust::zip_iterator<IteratorTuple> ZipIterator;
+
+  ZipIterator zipped_result =
+    thrust::for_each(exec,
+                     thrust::make_zip_iterator(thrust::make_tuple(first,stencil,result)),
+                     thrust::make_zip_iterator(thrust::make_tuple(last,stencil,result)),
+                     UnaryTransformIfFunctor(unary_op,pred));
+
+  return thrust::get<2>(zipped_result.get_iterator_tuple());
+} // end transform_if()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename ForwardIterator,
+         typename BinaryFunction,
+         typename Predicate>
+  ForwardIterator transform_if(thrust::execution_policy<DerivedPolicy> &exec,
+                               InputIterator1 first1,
+                               InputIterator1 last1,
+                               InputIterator2 first2,
+                               InputIterator3 stencil,
+                               ForwardIterator result,
+                               BinaryFunction binary_op,
+                               Predicate pred)
+{
+  // XXX WAR the problem of a generic __host__ __device__ functor's inability to invoke
+  //     a function which is only __host__ or __device__ by selecting a generic functor
+  //     which is one or the other
+  //     when nvcc is able to deal with this, remove this WAR
+  
+  // given the minimal system, determine the binary transform_if functor we need
+  typedef typename thrust::detail::binary_transform_if_functor<DerivedPolicy,BinaryFunction,Predicate>::type BinaryTransformIfFunctor;
+
+  // make an iterator tuple
+  typedef thrust::tuple<InputIterator1,InputIterator2,InputIterator3,ForwardIterator> IteratorTuple;
+  typedef thrust::zip_iterator<IteratorTuple> ZipIterator;
+
+  ZipIterator zipped_result =
+    thrust::for_each(exec,
+                     thrust::make_zip_iterator(thrust::make_tuple(first1,first2,stencil,result)),
+                     thrust::make_zip_iterator(thrust::make_tuple(last1,first2,stencil,result)),
+                     BinaryTransformIfFunctor(binary_op,pred));
+
+  return thrust::get<3>(zipped_result.get_iterator_tuple());
+} // end transform_if()
+
+
+} // end generic
+} // end detail
+} // end system
+} // end thrust
+
diff --git a/compat/thrust/system/detail/generic/transform_reduce.h b/compat/thrust/system/detail/generic/transform_reduce.h
new file mode 100644
index 0000000000..c1f098f50f
--- /dev/null
+++ b/compat/thrust/system/detail/generic/transform_reduce.h
@@ -0,0 +1,50 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/tag.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+template<typename ExecutionPolicy,
+         typename InputIterator, 
+         typename UnaryFunction, 
+         typename OutputType,
+         typename BinaryFunction>
+  OutputType transform_reduce(thrust::execution_policy<ExecutionPolicy> &exec,
+                              InputIterator first,
+                              InputIterator last,
+                              UnaryFunction unary_op,
+                              OutputType init,
+                              BinaryFunction binary_op);
+
+} // end namespace generic
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/detail/generic/transform_reduce.inl>
+
diff --git a/compat/thrust/system/detail/generic/transform_reduce.inl b/compat/thrust/system/detail/generic/transform_reduce.inl
new file mode 100644
index 0000000000..ce8b6a1213
--- /dev/null
+++ b/compat/thrust/system/detail/generic/transform_reduce.inl
@@ -0,0 +1,53 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/transform_reduce.h>
+#include <thrust/reduce.h>
+#include <thrust/iterator/transform_iterator.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+template<typename DerivedPolicy,
+         typename InputIterator, 
+         typename UnaryFunction, 
+         typename OutputType,
+         typename BinaryFunction>
+  OutputType transform_reduce(thrust::execution_policy<DerivedPolicy> &exec,
+                              InputIterator first,
+                              InputIterator last,
+                              UnaryFunction unary_op,
+                              OutputType init,
+                              BinaryFunction binary_op)
+{
+  thrust::transform_iterator<UnaryFunction, InputIterator, OutputType> xfrm_first(first, unary_op);
+  thrust::transform_iterator<UnaryFunction, InputIterator, OutputType> xfrm_last(last, unary_op);
+
+  return thrust::reduce(exec, xfrm_first, xfrm_last, init, binary_op);
+} // end transform_reduce()
+
+} // end generic
+} // end detail
+} // end system
+} // end thrust
+
diff --git a/compat/thrust/system/detail/generic/transform_scan.h b/compat/thrust/system/detail/generic/transform_scan.h
new file mode 100644
index 0000000000..99db86e4d9
--- /dev/null
+++ b/compat/thrust/system/detail/generic/transform_scan.h
@@ -0,0 +1,64 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/tag.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+template<typename ExecutionPolicy,
+         typename InputIterator,
+         typename OutputIterator,
+         typename UnaryFunction,
+         typename BinaryFunction>
+  OutputIterator transform_inclusive_scan(thrust::execution_policy<ExecutionPolicy> &exec,
+                                          InputIterator first,
+                                          InputIterator last,
+                                          OutputIterator result,
+                                          UnaryFunction unary_op,
+                                          BinaryFunction binary_op);
+
+template<typename ExecutionPolicy,
+         typename InputIterator,
+         typename OutputIterator,
+         typename UnaryFunction,
+         typename T,
+         typename AssociativeOperator>
+  OutputIterator transform_exclusive_scan(thrust::execution_policy<ExecutionPolicy> &exec,
+                                          InputIterator first,
+                                          InputIterator last,
+                                          OutputIterator result,
+                                          UnaryFunction unary_op,
+                                          T init,
+                                          AssociativeOperator binary_op);
+
+} // end namespace generic
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/detail/generic/transform_scan.inl>
+
diff --git a/compat/thrust/system/detail/generic/transform_scan.inl b/compat/thrust/system/detail/generic/transform_scan.inl
new file mode 100644
index 0000000000..a95ec20e67
--- /dev/null
+++ b/compat/thrust/system/detail/generic/transform_scan.inl
@@ -0,0 +1,124 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/transform_scan.h>
+#include <thrust/scan.h>
+#include <thrust/iterator/transform_iterator.h>
+#include <thrust/detail/type_traits.h>
+#include <thrust/detail/type_traits/function_traits.h>
+#include <thrust/detail/type_traits/iterator/is_output_iterator.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+template<typename ExecutionPolicy,
+         typename InputIterator,
+         typename OutputIterator,
+         typename UnaryFunction,
+         typename BinaryFunction>
+  OutputIterator transform_inclusive_scan(thrust::execution_policy<ExecutionPolicy> &exec,
+                                          InputIterator first,
+                                          InputIterator last,
+                                          OutputIterator result,
+                                          UnaryFunction unary_op,
+                                          BinaryFunction binary_op)
+{
+  // the pseudocode for deducing the type of the temporary used below:
+  // 
+  // if UnaryFunction is AdaptableUnaryFunction
+  //   TemporaryType = AdaptableUnaryFunction::result_type
+  // else if OutputIterator is a "pure" output iterator
+  //   TemporaryType = InputIterator::value_type
+  // else
+  //   TemporaryType = OutputIterator::value_type
+  //
+  // XXX upon c++0x, TemporaryType needs to be:
+  // result_of<UnaryFunction>::type
+
+  typedef typename thrust::detail::eval_if<
+    thrust::detail::has_result_type<UnaryFunction>::value,
+    thrust::detail::result_type<UnaryFunction>,
+    thrust::detail::eval_if<
+      thrust::detail::is_output_iterator<OutputIterator>::value,
+      thrust::iterator_value<InputIterator>,
+      thrust::iterator_value<OutputIterator>
+    >
+  >::type ValueType;
+
+  thrust::transform_iterator<UnaryFunction, InputIterator, ValueType> _first(first, unary_op);
+  thrust::transform_iterator<UnaryFunction, InputIterator, ValueType> _last(last, unary_op);
+
+  return thrust::inclusive_scan(exec, _first, _last, result, binary_op);
+} // end transform_inclusive_scan()
+
+template<typename ExecutionPolicy,
+         typename InputIterator,
+         typename OutputIterator,
+         typename UnaryFunction,
+         typename T,
+         typename AssociativeOperator>
+  OutputIterator transform_exclusive_scan(thrust::execution_policy<ExecutionPolicy> &exec,
+                                          InputIterator first,
+                                          InputIterator last,
+                                          OutputIterator result,
+                                          UnaryFunction unary_op,
+                                          T init,
+                                          AssociativeOperator binary_op)
+{
+  // the pseudocode for deducing the type of the temporary used below:
+  // 
+  // if UnaryFunction is AdaptableUnaryFunction
+  //   TemporaryType = AdaptableUnaryFunction::result_type
+  // else if OutputIterator is a "pure" output iterator
+  //   TemporaryType = InputIterator::value_type
+  // else
+  //   TemporaryType = OutputIterator::value_type
+  //
+  // XXX upon c++0x, TemporaryType needs to be:
+  // result_of<UnaryFunction>::type
+
+  typedef typename thrust::detail::eval_if<
+    thrust::detail::has_result_type<UnaryFunction>::value,
+    thrust::detail::result_type<UnaryFunction>,
+    thrust::detail::eval_if<
+      thrust::detail::is_output_iterator<OutputIterator>::value,
+      thrust::iterator_value<InputIterator>,
+      thrust::iterator_value<OutputIterator>
+    >
+  >::type ValueType;
+
+  thrust::transform_iterator<UnaryFunction, InputIterator, ValueType> _first(first, unary_op);
+  thrust::transform_iterator<UnaryFunction, InputIterator, ValueType> _last(last, unary_op);
+
+  return thrust::exclusive_scan(exec, _first, _last, result, init, binary_op);
+} // end transform_exclusive_scan()
+
+} // end namespace generic
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
+
diff --git a/compat/thrust/system/detail/generic/type_traits.h b/compat/thrust/system/detail/generic/type_traits.h
new file mode 100644
index 0000000000..40113525f3
--- /dev/null
+++ b/compat/thrust/system/detail/generic/type_traits.h
@@ -0,0 +1,168 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file generic/type_traits.h
+ *  \brief Introspection for free functions defined in generic.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+namespace thrust
+{
+
+// forward declaration of any_system_tag for any_conversion below
+struct any_system_tag;
+
+namespace system
+{
+namespace detail
+{
+
+// we must define these traits outside of generic's namespace
+namespace generic_type_traits_ns
+{
+
+typedef char yes;
+typedef char (&no)[2];
+
+struct any_conversion
+{
+  template<typename T> any_conversion(const T &);
+
+  // add this extra constructor to disambiguate conversion from any_system_tag
+  any_conversion(const any_system_tag &);
+};
+
+namespace select_system_exists_ns
+{
+  no select_system(const any_conversion &);
+  no select_system(const any_conversion &, const any_conversion &);
+  no select_system(const any_conversion &, const any_conversion &, const any_conversion &);
+  no select_system(const any_conversion &, const any_conversion &, const any_conversion &, const any_conversion &);
+  no select_system(const any_conversion &, const any_conversion &, const any_conversion &, const any_conversion &, const any_conversion &);
+  no select_system(const any_conversion &, const any_conversion &, const any_conversion &, const any_conversion &, const any_conversion &, const any_conversion &);
+
+  template<typename T> yes check(const T &);
+
+  no check(no);
+
+  template<typename Tag>
+    struct select_system1_exists
+  {
+    static Tag &tag;
+
+    static const bool value = sizeof(check(select_system(tag))) == sizeof(yes);
+  };
+
+  template<typename Tag1, typename Tag2>
+    struct select_system2_exists
+  {
+    static Tag1 &tag1;
+    static Tag2 &tag2;
+
+    static const bool value = sizeof(check(select_system(tag1,tag2))) == sizeof(yes);
+  };
+
+  template<typename Tag1, typename Tag2, typename Tag3>
+    struct select_system3_exists
+  {
+    static Tag1 &tag1;
+    static Tag2 &tag2;
+    static Tag3 &tag3;
+
+    static const bool value = sizeof(check(select_system(tag1,tag2,tag3))) == sizeof(yes);
+  };
+
+  template<typename Tag1, typename Tag2, typename Tag3, typename Tag4>
+    struct select_system4_exists
+  {
+    static Tag1 &tag1;
+    static Tag2 &tag2;
+    static Tag3 &tag3;
+    static Tag4 &tag4;
+
+    static const bool value = sizeof(check(select_system(tag1,tag2,tag3,tag4))) == sizeof(yes);
+  };
+
+  template<typename Tag1, typename Tag2, typename Tag3, typename Tag4, typename Tag5>
+    struct select_system5_exists
+  {
+    static Tag1 &tag1;
+    static Tag2 &tag2;
+    static Tag3 &tag3;
+    static Tag4 &tag4;
+    static Tag5 &tag5;
+
+    static const bool value = sizeof(check(select_system(tag1,tag2,tag3,tag4,tag5))) == sizeof(yes);
+  };
+
+  template<typename Tag1, typename Tag2, typename Tag3, typename Tag4, typename Tag5, typename Tag6>
+    struct select_system6_exists
+  {
+    static Tag1 &tag1;
+    static Tag2 &tag2;
+    static Tag3 &tag3;
+    static Tag4 &tag4;
+    static Tag5 &tag5;
+    static Tag6 &tag6;
+
+    static const bool value = sizeof(check(select_system(tag1,tag2,tag3,tag4,tag5,tag6))) == sizeof(yes);
+  };
+} // end select_system_exists_ns
+
+} // end generic_type_traits_ns
+
+namespace generic
+{
+
+template<typename Tag>
+  struct select_system1_exists
+    : generic_type_traits_ns::select_system_exists_ns::select_system1_exists<Tag>
+{};
+
+template<typename Tag1, typename Tag2>
+  struct select_system2_exists
+    : generic_type_traits_ns::select_system_exists_ns::select_system2_exists<Tag1,Tag2>
+{};
+
+template<typename Tag1, typename Tag2, typename Tag3>
+  struct select_system3_exists
+    : generic_type_traits_ns::select_system_exists_ns::select_system3_exists<Tag1,Tag2,Tag3>
+{};
+
+template<typename Tag1, typename Tag2, typename Tag3, typename Tag4>
+  struct select_system4_exists
+    : generic_type_traits_ns::select_system_exists_ns::select_system4_exists<Tag1,Tag2,Tag3,Tag4>
+{};
+
+template<typename Tag1, typename Tag2, typename Tag3, typename Tag4, typename Tag5>
+  struct select_system5_exists
+    : generic_type_traits_ns::select_system_exists_ns::select_system5_exists<Tag1,Tag2,Tag3,Tag4,Tag5>
+{};
+
+template<typename Tag1, typename Tag2, typename Tag3, typename Tag4, typename Tag5, typename Tag6>
+  struct select_system6_exists
+    : generic_type_traits_ns::select_system_exists_ns::select_system6_exists<Tag1,Tag2,Tag3,Tag4,Tag5,Tag6>
+{};
+
+} // end generic
+} // end detail
+} // end system
+} // end thrust
+
diff --git a/compat/thrust/system/detail/generic/uninitialized_copy.h b/compat/thrust/system/detail/generic/uninitialized_copy.h
new file mode 100644
index 0000000000..67e3e68328
--- /dev/null
+++ b/compat/thrust/system/detail/generic/uninitialized_copy.h
@@ -0,0 +1,55 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/tag.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+template<typename ExecutionPolicy,
+         typename InputIterator,
+         typename ForwardIterator>
+  ForwardIterator uninitialized_copy(thrust::execution_policy<ExecutionPolicy> &exec,
+                                     InputIterator first,
+                                     InputIterator last,
+                                     ForwardIterator result);
+
+template<typename ExecutionPolicy,
+         typename InputIterator,
+         typename Size,
+         typename ForwardIterator>
+  ForwardIterator uninitialized_copy_n(thrust::execution_policy<ExecutionPolicy> &exec,
+                                       InputIterator first,
+                                       Size n,
+                                       ForwardIterator result);
+
+} // end namespace generic
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/detail/generic/uninitialized_copy.inl>
+
diff --git a/compat/thrust/system/detail/generic/uninitialized_copy.inl b/compat/thrust/system/detail/generic/uninitialized_copy.inl
new file mode 100644
index 0000000000..414e6e48fe
--- /dev/null
+++ b/compat/thrust/system/detail/generic/uninitialized_copy.inl
@@ -0,0 +1,187 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/uninitialized_copy.h>
+#include <thrust/copy.h>
+#include <thrust/for_each.h>
+#include <thrust/detail/internal_functional.h>
+#include <thrust/detail/type_traits.h>
+#include <thrust/iterator/iterator_traits.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+namespace detail
+{
+
+template<typename InputType,
+         typename OutputType>
+  struct uninitialized_copy_functor
+{
+  template<typename Tuple>
+  __host__ __device__
+  void operator()(Tuple t)
+  {
+    const InputType &in = thrust::get<0>(t);
+    OutputType &out = thrust::get<1>(t);
+
+    ::new(static_cast<void*>(&out)) OutputType(in);
+  } // end operator()()
+}; // end uninitialized_copy_functor
+
+
+// non-trivial copy constructor path
+template<typename ExecutionPolicy,
+         typename InputIterator,
+         typename ForwardIterator>
+  ForwardIterator uninitialized_copy(thrust::execution_policy<ExecutionPolicy> &exec,
+                                     InputIterator first,
+                                     InputIterator last,
+                                     ForwardIterator result,
+                                     thrust::detail::false_type) // has_trivial_copy_constructor
+{
+  // zip up the iterators
+  typedef thrust::tuple<InputIterator,ForwardIterator> IteratorTuple;
+  typedef thrust::zip_iterator<IteratorTuple> ZipIterator;
+
+  ZipIterator begin = thrust::make_zip_iterator(thrust::make_tuple(first,result));
+  ZipIterator end = begin;
+
+  // get a zip_iterator pointing to the end
+  const typename thrust::iterator_difference<InputIterator>::type n = thrust::distance(first,last);
+  thrust::advance(end, n);
+
+  // create a functor
+  typedef typename iterator_traits<InputIterator>::value_type InputType;
+  typedef typename iterator_traits<ForwardIterator>::value_type OutputType;
+
+  detail::uninitialized_copy_functor<InputType, OutputType> f;
+
+  // do the for_each
+  thrust::for_each(exec, begin, end, f);
+
+  // return the end of the output range
+  return thrust::get<1>(end.get_iterator_tuple());
+} // end uninitialized_copy()
+
+
+// trivial copy constructor path
+template<typename ExecutionPolicy,
+         typename InputIterator,
+         typename ForwardIterator>
+  ForwardIterator uninitialized_copy(thrust::execution_policy<ExecutionPolicy> &exec,
+                                     InputIterator first,
+                                     InputIterator last,
+                                     ForwardIterator result,
+                                     thrust::detail::true_type) // has_trivial_copy_constructor
+{
+  return thrust::copy(exec, first, last, result);
+} // end uninitialized_copy()
+
+
+// non-trivial copy constructor path
+template<typename ExecutionPolicy,
+         typename InputIterator,
+         typename Size,
+         typename ForwardIterator>
+  ForwardIterator uninitialized_copy_n(thrust::execution_policy<ExecutionPolicy> &exec,
+                                       InputIterator first,
+                                       Size n,
+                                       ForwardIterator result,
+                                       thrust::detail::false_type) // has_trivial_copy_constructor
+{
+  // zip up the iterators
+  typedef thrust::tuple<InputIterator,ForwardIterator> IteratorTuple;
+  typedef thrust::zip_iterator<IteratorTuple> ZipIterator;
+
+  ZipIterator zipped_first = thrust::make_zip_iterator(thrust::make_tuple(first,result));
+
+  // create a functor
+  typedef typename iterator_traits<InputIterator>::value_type   InputType;
+  typedef typename iterator_traits<ForwardIterator>::value_type OutputType;
+
+  detail::uninitialized_copy_functor<InputType, OutputType> f;
+
+  // do the for_each_n
+  ZipIterator zipped_last = thrust::for_each_n(exec, zipped_first, n, f);
+
+  // return the end of the output range
+  return thrust::get<1>(zipped_last.get_iterator_tuple());
+} // end uninitialized_copy_n()
+
+
+// trivial copy constructor path
+template<typename ExecutionPolicy,
+         typename InputIterator,
+         typename Size,
+         typename ForwardIterator>
+  ForwardIterator uninitialized_copy_n(thrust::execution_policy<ExecutionPolicy> &exec,
+                                       InputIterator first,
+                                       Size n,
+                                       ForwardIterator result,
+                                       thrust::detail::true_type) // has_trivial_copy_constructor
+{
+  return thrust::copy_n(exec, first, n, result);
+} // end uninitialized_copy_n()
+
+
+} // end detail
+
+
+template<typename ExecutionPolicy,
+         typename InputIterator,
+         typename ForwardIterator>
+  ForwardIterator uninitialized_copy(thrust::execution_policy<ExecutionPolicy> &exec,
+                                     InputIterator first,
+                                     InputIterator last,
+                                     ForwardIterator result)
+{
+  typedef typename iterator_traits<ForwardIterator>::value_type ResultType;
+
+  typedef typename thrust::detail::has_trivial_copy_constructor<ResultType>::type ResultTypeHasTrivialCopyConstructor;
+
+  return thrust::system::detail::generic::detail::uninitialized_copy(exec, first, last, result, ResultTypeHasTrivialCopyConstructor());
+} // end uninitialized_copy()
+
+
+template<typename ExecutionPolicy,
+         typename InputIterator,
+         typename Size,
+         typename ForwardIterator>
+  ForwardIterator uninitialized_copy_n(thrust::execution_policy<ExecutionPolicy> &exec,
+                                       InputIterator first,
+                                       Size n,
+                                       ForwardIterator result)
+{
+  typedef typename iterator_traits<ForwardIterator>::value_type ResultType;
+
+  typedef typename thrust::detail::has_trivial_copy_constructor<ResultType>::type ResultTypeHasTrivialCopyConstructor;
+
+  return thrust::system::detail::generic::detail::uninitialized_copy_n(exec, first, n, result, ResultTypeHasTrivialCopyConstructor());
+} // end uninitialized_copy_n()
+
+
+} // end namespace generic
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/detail/generic/uninitialized_fill.h b/compat/thrust/system/detail/generic/uninitialized_fill.h
new file mode 100644
index 0000000000..c1df694c02
--- /dev/null
+++ b/compat/thrust/system/detail/generic/uninitialized_fill.h
@@ -0,0 +1,55 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/tag.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+template<typename DerivedPolicy,
+         typename ForwardIterator,
+         typename T>
+  void uninitialized_fill(thrust::execution_policy<DerivedPolicy> &exec,
+                          ForwardIterator first,
+                          ForwardIterator last,
+                          const T &x);
+
+template<typename DerivedPolicy,
+         typename ForwardIterator,
+         typename Size,
+         typename T>
+  ForwardIterator uninitialized_fill_n(thrust::execution_policy<DerivedPolicy> &exec,
+                                       ForwardIterator first,
+                                       Size n,
+                                       const T &x);
+
+} // end namespace generic
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/detail/generic/uninitialized_fill.inl>
+
diff --git a/compat/thrust/system/detail/generic/uninitialized_fill.inl b/compat/thrust/system/detail/generic/uninitialized_fill.inl
new file mode 100644
index 0000000000..bb30b24fd7
--- /dev/null
+++ b/compat/thrust/system/detail/generic/uninitialized_fill.inl
@@ -0,0 +1,128 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/uninitialized_fill.h>
+#include <thrust/fill.h>
+#include <thrust/detail/internal_functional.h>
+#include <thrust/detail/type_traits.h>
+#include <thrust/iterator/iterator_traits.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+namespace detail
+{
+
+template<typename DerivedPolicy,
+         typename ForwardIterator,
+         typename T>
+  void uninitialized_fill(thrust::execution_policy<DerivedPolicy> &exec,
+                          ForwardIterator first,
+                          ForwardIterator last,
+                          const T &x,
+                          thrust::detail::true_type) // has_trivial_copy_constructor
+{
+  thrust::fill(exec, first, last, x);
+} // end uninitialized_fill()
+
+template<typename DerivedPolicy,
+         typename ForwardIterator,
+         typename T>
+  void uninitialized_fill(thrust::execution_policy<DerivedPolicy> &exec,
+                          ForwardIterator first,
+                          ForwardIterator last,
+                          const T &x,
+                          thrust::detail::false_type) // has_trivial_copy_constructor
+{
+  typedef typename iterator_traits<ForwardIterator>::value_type ValueType;
+
+  thrust::for_each(exec, first, last, thrust::detail::uninitialized_fill_functor<ValueType>(x));
+} // end uninitialized_fill()
+
+template<typename DerivedPolicy,
+         typename ForwardIterator,
+         typename Size,
+         typename T>
+  ForwardIterator uninitialized_fill_n(thrust::execution_policy<DerivedPolicy> &exec,
+                                       ForwardIterator first,
+                                       Size n,
+                                       const T &x,
+                                       thrust::detail::true_type) // has_trivial_copy_constructor
+{
+  return thrust::fill_n(exec, first, n, x);
+} // end uninitialized_fill()
+
+template<typename DerivedPolicy,
+         typename ForwardIterator,
+         typename Size,
+         typename T>
+  ForwardIterator uninitialized_fill_n(thrust::execution_policy<DerivedPolicy> &exec,
+                                       ForwardIterator first,
+                                       Size n,
+                                       const T &x,
+                                       thrust::detail::false_type) // has_trivial_copy_constructor
+{
+  typedef typename iterator_traits<ForwardIterator>::value_type ValueType;
+
+  return thrust::for_each_n(exec, first, n, thrust::detail::uninitialized_fill_functor<ValueType>(x));
+} // end uninitialized_fill()
+
+} // end detail
+
+template<typename DerivedPolicy,
+         typename ForwardIterator,
+         typename T>
+  void uninitialized_fill(thrust::execution_policy<DerivedPolicy> &exec,
+                          ForwardIterator first,
+                          ForwardIterator last,
+                          const T &x)
+{
+  typedef typename iterator_traits<ForwardIterator>::value_type ValueType;
+
+  typedef thrust::detail::has_trivial_copy_constructor<ValueType> ValueTypeHasTrivialCopyConstructor;
+
+  thrust::system::detail::generic::detail::uninitialized_fill(exec, first, last, x,
+    ValueTypeHasTrivialCopyConstructor());
+} // end uninitialized_fill()
+
+template<typename DerivedPolicy,
+         typename ForwardIterator,
+         typename Size,
+         typename T>
+  ForwardIterator uninitialized_fill_n(thrust::execution_policy<DerivedPolicy> &exec,
+                                       ForwardIterator first,
+                                       Size n,
+                                       const T &x)
+{
+  typedef typename iterator_traits<ForwardIterator>::value_type ValueType;
+
+  typedef thrust::detail::has_trivial_copy_constructor<ValueType> ValueTypeHasTrivialCopyConstructor;
+
+  return thrust::system::detail::generic::detail::uninitialized_fill_n(exec, first, n, x,
+    ValueTypeHasTrivialCopyConstructor());
+} // end uninitialized_fill()
+
+} // end namespace generic
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/detail/generic/unique.h b/compat/thrust/system/detail/generic/unique.h
new file mode 100644
index 0000000000..57e17cafa4
--- /dev/null
+++ b/compat/thrust/system/detail/generic/unique.h
@@ -0,0 +1,74 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/tag.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+
+template<typename DerivedPolicy,
+         typename ForwardIterator>
+ForwardIterator unique(thrust::execution_policy<DerivedPolicy> &exec,
+                       ForwardIterator first,
+                       ForwardIterator last);
+
+
+template<typename DerivedPolicy,
+         typename ForwardIterator,
+         typename BinaryPredicate>
+ForwardIterator unique(thrust::execution_policy<DerivedPolicy> &exec,
+                       ForwardIterator first,
+                       ForwardIterator last,
+                       BinaryPredicate binary_pred);
+
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator>
+OutputIterator unique_copy(thrust::execution_policy<DerivedPolicy> &exec,
+                           InputIterator first,
+                           InputIterator last,
+                           OutputIterator output);
+
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator,
+         typename BinaryPredicate>
+OutputIterator unique_copy(thrust::execution_policy<DerivedPolicy> &exec,
+                           InputIterator first,
+                           InputIterator last,
+                           OutputIterator output,
+                           BinaryPredicate binary_pred);
+
+
+} // end namespace generic
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/detail/generic/unique.inl>
+
diff --git a/compat/thrust/system/detail/generic/unique.inl b/compat/thrust/system/detail/generic/unique.inl
new file mode 100644
index 0000000000..42d6b15e82
--- /dev/null
+++ b/compat/thrust/system/detail/generic/unique.inl
@@ -0,0 +1,114 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file unique.inl
+ *  \brief Inline file for unique.h.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/unique.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/transform.h>
+#include <thrust/unique.h>
+#include <thrust/detail/temporary_array.h>
+#include <thrust/detail/internal_functional.h>
+#include <thrust/detail/copy_if.h>
+#include <thrust/distance.h>
+#include <thrust/functional.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+
+template<typename DerivedPolicy,
+         typename ForwardIterator>
+  ForwardIterator unique(thrust::execution_policy<DerivedPolicy> &exec,
+                         ForwardIterator first,
+                         ForwardIterator last)
+{
+  typedef typename thrust::iterator_traits<ForwardIterator>::value_type InputType;
+
+  return thrust::unique(exec, first, last, thrust::equal_to<InputType>());
+} // end unique()
+
+
+template<typename DerivedPolicy,
+         typename ForwardIterator,
+         typename BinaryPredicate>
+  ForwardIterator unique(thrust::execution_policy<DerivedPolicy> &exec,
+                         ForwardIterator first,
+                         ForwardIterator last,
+                         BinaryPredicate binary_pred)
+{
+  typedef typename thrust::iterator_traits<ForwardIterator>::value_type InputType;
+  
+  thrust::detail::temporary_array<InputType,DerivedPolicy> input(exec, first, last);
+  
+  return thrust::unique_copy(exec, input.begin(), input.end(), first, binary_pred);
+} // end unique()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator>
+  OutputIterator unique_copy(thrust::execution_policy<DerivedPolicy> &exec,
+                             InputIterator first,
+                             InputIterator last,
+                             OutputIterator output)
+{
+  typedef typename thrust::iterator_value<InputIterator>::type value_type;
+  return thrust::unique_copy(exec, first,last,output,thrust::equal_to<value_type>());
+} // end unique_copy()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator,
+         typename BinaryPredicate>
+  OutputIterator unique_copy(thrust::execution_policy<DerivedPolicy> &exec,
+                             InputIterator first,
+                             InputIterator last,
+                             OutputIterator output,
+                             BinaryPredicate binary_pred)
+{
+  // empty sequence
+  if(first == last)
+    return output;
+  
+  thrust::detail::temporary_array<int,DerivedPolicy> stencil(exec, thrust::distance(first, last));
+  
+  // mark first element in each group
+  stencil[0] = 1; 
+  thrust::transform(exec, first, last - 1, first + 1, stencil.begin() + 1, thrust::detail::not2(binary_pred)); 
+  
+  return thrust::copy_if(exec, first, last, stencil.begin(), output, thrust::identity<int>());
+} // end unique_copy()
+
+
+} // end namespace generic
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/detail/generic/unique_by_key.h b/compat/thrust/system/detail/generic/unique_by_key.h
new file mode 100644
index 0000000000..aa62f73e51
--- /dev/null
+++ b/compat/thrust/system/detail/generic/unique_by_key.h
@@ -0,0 +1,91 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/tag.h>
+#include <thrust/pair.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+
+template<typename ExecutionPolicy,
+         typename ForwardIterator1,
+         typename ForwardIterator2>
+  thrust::pair<ForwardIterator1,ForwardIterator2>
+    unique_by_key(thrust::execution_policy<ExecutionPolicy> &exec,
+                  ForwardIterator1 keys_first, 
+                  ForwardIterator1 keys_last,
+                  ForwardIterator2 values_first);
+
+
+template<typename ExecutionPolicy,
+         typename ForwardIterator1,
+         typename ForwardIterator2,
+         typename BinaryPredicate>
+  thrust::pair<ForwardIterator1,ForwardIterator2>
+    unique_by_key(thrust::execution_policy<ExecutionPolicy> &exec,
+                  ForwardIterator1 keys_first, 
+                  ForwardIterator1 keys_last,
+                  ForwardIterator2 values_first,
+                  BinaryPredicate binary_pred);
+
+
+template<typename ExecutionPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator1,
+         typename OutputIterator2>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    unique_by_key_copy(thrust::execution_policy<ExecutionPolicy> &exec,
+                       InputIterator1 keys_first, 
+                       InputIterator1 keys_last,
+                       InputIterator2 values_first,
+                       OutputIterator1 keys_output,
+                       OutputIterator2 values_output);
+
+
+template<typename ExecutionPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename BinaryPredicate>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    unique_by_key_copy(thrust::execution_policy<ExecutionPolicy> &exec,
+                       InputIterator1 keys_first, 
+                       InputIterator1 keys_last,
+                       InputIterator2 values_first,
+                       OutputIterator1 keys_output,
+                       OutputIterator2 values_output,
+                       BinaryPredicate binary_pred);
+
+
+} // end namespace generic
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/detail/generic/unique_by_key.inl>
+
diff --git a/compat/thrust/system/detail/generic/unique_by_key.inl b/compat/thrust/system/detail/generic/unique_by_key.inl
new file mode 100644
index 0000000000..c780fa71d1
--- /dev/null
+++ b/compat/thrust/system/detail/generic/unique_by_key.inl
@@ -0,0 +1,142 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/unique_by_key.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/iterator/detail/minimum_system.h>
+#include <thrust/iterator/zip_iterator.h>
+#include <thrust/transform.h>
+#include <thrust/detail/temporary_array.h>
+#include <thrust/detail/internal_functional.h>
+#include <thrust/detail/copy_if.h>
+#include <thrust/unique.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace generic
+{
+
+
+template<typename ExecutionPolicy,
+         typename ForwardIterator1,
+         typename ForwardIterator2>
+  thrust::pair<ForwardIterator1,ForwardIterator2>
+    unique_by_key(thrust::execution_policy<ExecutionPolicy> &exec,
+                  ForwardIterator1 keys_first, 
+                  ForwardIterator1 keys_last,
+                  ForwardIterator2 values_first)
+{
+  typedef typename thrust::iterator_traits<ForwardIterator1>::value_type KeyType;
+  return thrust::unique_by_key(exec, keys_first, keys_last, values_first, thrust::equal_to<KeyType>());
+} // end unique_by_key()
+
+
+template<typename ExecutionPolicy,
+         typename ForwardIterator1,
+         typename ForwardIterator2,
+         typename BinaryPredicate>
+  thrust::pair<ForwardIterator1,ForwardIterator2>
+    unique_by_key(thrust::execution_policy<ExecutionPolicy> &exec,
+                  ForwardIterator1 keys_first, 
+                  ForwardIterator1 keys_last,
+                  ForwardIterator2 values_first,
+                  BinaryPredicate binary_pred)
+{
+  typedef typename thrust::iterator_traits<ForwardIterator1>::value_type InputType1;
+  typedef typename thrust::iterator_traits<ForwardIterator2>::value_type InputType2;
+  
+  ForwardIterator2 values_last = values_first + (keys_last - keys_first);
+  
+  thrust::detail::temporary_array<InputType1,ExecutionPolicy> keys(exec, keys_first, keys_last);
+  thrust::detail::temporary_array<InputType2,ExecutionPolicy> vals(exec, values_first, values_last);
+  
+  return thrust::unique_by_key_copy(exec, keys.begin(), keys.end(), vals.begin(), keys_first, values_first, binary_pred);
+} // end unique_by_key()
+
+
+template<typename ExecutionPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator1,
+         typename OutputIterator2>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    unique_by_key_copy(thrust::execution_policy<ExecutionPolicy> &exec,
+                       InputIterator1 keys_first, 
+                       InputIterator1 keys_last,
+                       InputIterator2 values_first,
+                       OutputIterator1 keys_output,
+                       OutputIterator2 values_output)
+{
+  typedef typename thrust::iterator_traits<InputIterator1>::value_type KeyType;
+  return thrust::unique_by_key_copy(exec, keys_first, keys_last, values_first, keys_output, values_output, thrust::equal_to<KeyType>());
+} // end unique_by_key_copy()
+
+
+template<typename ExecutionPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename BinaryPredicate>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    unique_by_key_copy(thrust::execution_policy<ExecutionPolicy> &exec,
+                       InputIterator1 keys_first, 
+                       InputIterator1 keys_last,
+                       InputIterator2 values_first,
+                       OutputIterator1 keys_output,
+                       OutputIterator2 values_output,
+                       BinaryPredicate binary_pred)
+{
+  typedef typename thrust::iterator_traits<InputIterator1>::difference_type difference_type;
+  
+  // empty sequence
+  if(keys_first == keys_last)
+    return thrust::make_pair(keys_output, values_output);
+  
+  difference_type n = thrust::distance(keys_first, keys_last);
+  
+  thrust::detail::temporary_array<int,ExecutionPolicy> stencil(exec,n);
+  
+  // mark first element in each group
+  stencil[0] = 1; 
+  thrust::transform(exec, keys_first, keys_last - 1, keys_first + 1, stencil.begin() + 1, thrust::detail::not2(binary_pred)); 
+  
+  thrust::zip_iterator< thrust::tuple<OutputIterator1, OutputIterator2> > result =
+    thrust::copy_if(exec,
+                    thrust::make_zip_iterator(thrust::make_tuple(keys_first, values_first)),
+                    thrust::make_zip_iterator(thrust::make_tuple(keys_first, values_first)) + n,
+                    stencil.begin(),
+                    thrust::make_zip_iterator(thrust::make_tuple(keys_output, values_output)),
+                    thrust::identity<int>());
+  
+  difference_type output_size = result - thrust::make_zip_iterator(thrust::make_tuple(keys_output, values_output));
+                                  
+  return thrust::make_pair(keys_output + output_size, values_output + output_size);
+} // end unique_by_key_copy()
+
+
+} // end namespace generic
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/detail/internal/decompose.h b/compat/thrust/system/detail/internal/decompose.h
new file mode 100644
index 0000000000..dea806d69c
--- /dev/null
+++ b/compat/thrust/system/detail/internal/decompose.h
@@ -0,0 +1,113 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace internal
+{
+
+  template <typename IndexType>
+    class index_range
+    {
+      public:
+        typedef IndexType index_type;
+
+        __host__ __device__
+          index_range(index_type begin, index_type end) : m_begin(begin), m_end(end) {}
+
+        __host__ __device__
+          index_type begin(void) const { return m_begin; }
+
+        __host__ __device__
+          index_type end(void)   const { return m_end; }
+
+        __host__ __device__
+          index_type size(void)  const { return m_end - m_begin; }
+
+      private:
+        index_type m_begin;
+        index_type m_end;
+    };
+
+  template <typename IndexType>
+    class uniform_decomposition
+    {
+      public:
+        typedef IndexType               index_type;
+        typedef index_range<index_type> range_type;
+
+        uniform_decomposition(index_type N, index_type granularity, index_type max_intervals)
+          : m_N(N),
+	    m_intervals((N + granularity - 1) / granularity),
+	    m_threshold(0),
+	    m_small_interval(granularity),
+	    m_large_interval(0)
+        {
+	  if(m_intervals > max_intervals)
+          {
+	    m_small_interval = granularity * (m_intervals / max_intervals);
+	    m_large_interval = m_small_interval + granularity;
+	    m_threshold      = m_intervals % max_intervals;
+	    m_intervals      = max_intervals;
+	  }
+        }
+
+        __host__ __device__
+          index_range<index_type> operator[](const index_type& i) const
+          {
+            if (i < m_threshold)
+            {
+              index_type begin = m_large_interval * i;
+              index_type end   = begin + m_large_interval;
+              return range_type(begin, end);
+            }
+            else
+            {
+              index_type begin = m_large_interval * m_threshold + m_small_interval * (i - m_threshold);
+              index_type end   = (begin + m_small_interval < m_N) ? begin + m_small_interval : m_N;
+              return range_type(begin, end);
+            }
+          }
+
+        __host__ __device__
+          index_type size(void) const
+          {
+            return m_intervals;
+          }
+
+      private:
+
+        index_type m_N;
+        index_type m_intervals;
+        index_type m_threshold;
+        index_type m_small_interval;
+        index_type m_large_interval;
+    };
+
+
+} // end namespace internal
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/detail/internal/scalar/adjacent_difference.h b/compat/thrust/system/detail/internal/scalar/adjacent_difference.h
new file mode 100644
index 0000000000..d1a95aeec4
--- /dev/null
+++ b/compat/thrust/system/detail/internal/scalar/adjacent_difference.h
@@ -0,0 +1,70 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file adjacent_difference.h
+ *  \brief Sequential implementation of adjacent_difference.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/iterator/iterator_traits.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace internal
+{
+namespace scalar
+{
+
+template <typename InputIterator,
+          typename OutputIterator,
+          typename BinaryFunction>
+OutputIterator adjacent_difference(InputIterator first,
+                                   InputIterator last,
+                                   OutputIterator result,
+                                   BinaryFunction binary_op)
+{
+  typedef typename thrust::iterator_traits<InputIterator>::value_type InputType;
+
+  if (first == last)
+    return result;
+
+  InputType curr = *first;
+
+  *result = curr;
+
+  while (++first != last)
+  {
+    InputType next = *first;
+    *(++result) = binary_op(next, curr);
+    curr = next;
+  }
+
+  return ++result;
+}
+
+} // end namespace scalar
+} // end namespace internal
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/detail/internal/scalar/binary_search.h b/compat/thrust/system/detail/internal/scalar/binary_search.h
new file mode 100644
index 0000000000..c3ac49fbd3
--- /dev/null
+++ b/compat/thrust/system/detail/internal/scalar/binary_search.h
@@ -0,0 +1,143 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file binary_search.h
+ *  \brief Sequential implementation of binary search algorithms.
+ */
+
+#pragma once
+
+#include <thrust/advance.h>
+#include <thrust/distance.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/detail/function.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace internal
+{
+namespace scalar
+{
+
+template <typename ForwardIterator,
+          typename T,
+          typename StrictWeakOrdering>
+ForwardIterator lower_bound(ForwardIterator first,
+                            ForwardIterator last,
+                            const T& val,
+                            StrictWeakOrdering comp)
+{
+  // wrap comp
+  thrust::detail::host_function<
+    StrictWeakOrdering,
+    bool
+  > wrapped_comp(comp);
+
+  typedef typename thrust::iterator_difference<ForwardIterator>::type difference_type;
+
+  difference_type len = thrust::distance(first, last);
+
+  while(len > 0)
+  {
+    difference_type half = len >> 1;
+    ForwardIterator middle = first;
+
+    thrust::advance(middle, half);
+
+    if(wrapped_comp(*middle, val))
+    {
+      first = middle;
+      ++first;
+      len = len - half - 1;
+    }
+    else
+    {
+      len = half;
+    }
+  }
+
+  return first;
+}
+
+
+template <typename ForwardIterator, typename T, typename StrictWeakOrdering>
+ForwardIterator upper_bound(ForwardIterator first,
+                            ForwardIterator last,
+                            const T& val, 
+                            StrictWeakOrdering comp)
+{
+  // wrap comp
+  thrust::detail::host_function<
+    StrictWeakOrdering,
+    bool
+  > wrapped_comp(comp);
+
+  typedef typename thrust::iterator_difference<ForwardIterator>::type difference_type;
+
+  difference_type len = thrust::distance(first, last);
+
+  while(len > 0)
+  {
+    difference_type half = len >> 1;
+    ForwardIterator middle = first;
+
+    thrust::advance(middle, half);
+
+    if(wrapped_comp(val, *middle))
+    {
+      len = half;
+    }
+    else
+    {
+      first = middle;
+      ++first;
+      len = len - half - 1;
+    }
+  }
+
+  return first;
+}
+
+template <typename ForwardIterator,
+          typename T,
+          typename StrictWeakOrdering>
+bool binary_search(ForwardIterator first,
+                   ForwardIterator last,
+                   const T& val, 
+                   StrictWeakOrdering comp)
+{
+  ForwardIterator iter = thrust::system::detail::internal::scalar::lower_bound(first, last, val, comp);
+
+  // wrap comp
+  thrust::detail::host_function<
+    StrictWeakOrdering,
+    bool
+  > wrapped_comp(comp);
+
+  return iter != last && !wrapped_comp(val,*iter);
+}
+
+} // end namespace scalar
+} // end namespace internal
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/detail/internal/scalar/copy.h b/compat/thrust/system/detail/internal/scalar/copy.h
new file mode 100644
index 0000000000..42cb385402
--- /dev/null
+++ b/compat/thrust/system/detail/internal/scalar/copy.h
@@ -0,0 +1,56 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+/*! \file copy.h
+ *  \brief Sequential implementations of copy algorithms.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace internal
+{
+namespace scalar
+{
+
+template<typename InputIterator,
+         typename OutputIterator>
+  OutputIterator copy(InputIterator first,
+                      InputIterator last,
+                      OutputIterator result);
+
+template<typename InputIterator,
+         typename Size,
+         typename OutputIterator>
+  OutputIterator copy_n(InputIterator first,
+                        Size n,
+                        OutputIterator result);
+
+} // end namespace scalar
+} // end namespace internal
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/detail/internal/scalar/copy.inl>
+
diff --git a/compat/thrust/system/detail/internal/scalar/copy.inl b/compat/thrust/system/detail/internal/scalar/copy.inl
new file mode 100644
index 0000000000..8c9f5c2825
--- /dev/null
+++ b/compat/thrust/system/detail/internal/scalar/copy.inl
@@ -0,0 +1,127 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/detail/dispatch/is_trivial_copy.h>
+#include <thrust/detail/type_traits.h>
+#include <thrust/system/detail/internal/scalar/general_copy.h>
+#include <thrust/system/detail/internal/scalar/trivial_copy.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/detail/type_traits/pointer_traits.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace internal
+{
+namespace scalar
+{
+namespace copy_detail
+{
+
+
+// returns the raw pointer associated with a Pointer-like thing
+template<typename Pointer>
+  typename thrust::detail::pointer_traits<Pointer>::raw_pointer
+    get(Pointer ptr)
+{
+  return thrust::detail::pointer_traits<Pointer>::get(ptr);
+}
+
+
+template<typename InputIterator,
+         typename OutputIterator>
+  OutputIterator copy(InputIterator first,
+                      InputIterator last,
+                      OutputIterator result,
+                      thrust::detail::true_type)  // is_trivial_copy
+{
+  typedef typename thrust::iterator_difference<InputIterator>::type Size;
+
+  const Size n = last - first;
+  thrust::system::detail::internal::scalar::trivial_copy_n(get(&*first), n, get(&*result));
+  return result + n;
+} // end copy()
+
+
+template<typename InputIterator,
+         typename OutputIterator>
+  OutputIterator copy(InputIterator first,
+                      InputIterator last,
+                      OutputIterator result,
+                      thrust::detail::false_type)  // is_trivial_copy
+{
+  return thrust::system::detail::internal::scalar::general_copy(first,last,result);
+} // end copy()
+
+
+template<typename InputIterator,
+         typename Size,
+         typename OutputIterator>
+  OutputIterator copy_n(InputIterator first,
+                        Size n,
+                        OutputIterator result,
+                        thrust::detail::true_type)  // is_trivial_copy
+{
+  thrust::system::detail::internal::scalar::trivial_copy_n(get(&*first), n, get(&*result));
+  return result + n;
+} // end copy_n()
+
+
+template<typename InputIterator,
+         typename Size,
+         typename OutputIterator>
+  OutputIterator copy_n(InputIterator first,
+                        Size n,
+                        OutputIterator result,
+                        thrust::detail::false_type)  // is_trivial_copy
+{
+  return thrust::system::detail::internal::scalar::general_copy_n(first,n,result);
+} // end copy_n()
+
+} // end namespace copy_detail
+
+
+template<typename InputIterator,
+         typename OutputIterator>
+  OutputIterator copy(InputIterator first,
+                      InputIterator last,
+                      OutputIterator result)
+{
+  return thrust::system::detail::internal::scalar::copy_detail::copy(first, last, result,
+    typename thrust::detail::dispatch::is_trivial_copy<InputIterator,OutputIterator>::type());
+} // end copy()
+
+
+template<typename InputIterator,
+         typename Size,
+         typename OutputIterator>
+  OutputIterator copy_n(InputIterator first,
+                        Size n,
+                        OutputIterator result)
+{
+  return thrust::system::detail::internal::scalar::copy_detail::copy_n(first, n, result,
+    typename thrust::detail::dispatch::is_trivial_copy<InputIterator,OutputIterator>::type());
+} // end copy_n()
+
+} // end namespace scalar
+} // end namespace internal
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/detail/internal/scalar/copy_backward.h b/compat/thrust/system/detail/internal/scalar/copy_backward.h
new file mode 100644
index 0000000000..36f8f66e44
--- /dev/null
+++ b/compat/thrust/system/detail/internal/scalar/copy_backward.h
@@ -0,0 +1,53 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace internal
+{
+namespace scalar
+{
+
+template <typename BidirectionalIterator1,
+          typename BidirectionalIterator2>
+BidirectionalIterator2 copy_backward(BidirectionalIterator1 first, 
+                                     BidirectionalIterator1 last, 
+                                     BidirectionalIterator2 result)
+{
+  while (first != last)
+  {
+    --last;
+    --result;
+    *result = *last;
+  }
+
+  return result;
+}
+
+} // end namespace scalar
+} // end namespace internal
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/detail/internal/scalar/copy_if.h b/compat/thrust/system/detail/internal/scalar/copy_if.h
new file mode 100644
index 0000000000..67f9402335
--- /dev/null
+++ b/compat/thrust/system/detail/internal/scalar/copy_if.h
@@ -0,0 +1,69 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+/*! \file copy_if.h
+ *  \brief Sequential implementation of copy_if.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/function.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace internal
+{
+namespace scalar
+{
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename Predicate>
+  OutputIterator copy_if(InputIterator1 first,
+                         InputIterator1 last,
+                         InputIterator2 stencil,
+                         OutputIterator result,
+                         Predicate pred)
+{
+  thrust::detail::host_function<Predicate,bool> wrapped_pred(pred);
+
+  while(first != last)
+  {
+    if(wrapped_pred(*stencil))
+    {
+      *result = *first;
+      ++result;
+    } // end if
+
+    ++first;
+    ++stencil;
+  } // end while
+
+  return result;
+} // end copy_if()
+
+} // end namespace scalar
+} // end namespace internal
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/detail/internal/scalar/extrema.h b/compat/thrust/system/detail/internal/scalar/extrema.h
new file mode 100644
index 0000000000..ebea756d8c
--- /dev/null
+++ b/compat/thrust/system/detail/internal/scalar/extrema.h
@@ -0,0 +1,127 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file extrema.h
+ *  \brief Sequential implementations of extrema functions.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/pair.h>
+#include <thrust/detail/function.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace internal
+{
+namespace scalar
+{
+
+template <typename ForwardIterator,
+          typename BinaryPredicate>
+ForwardIterator min_element(ForwardIterator first, 
+                            ForwardIterator last,
+                            BinaryPredicate comp)
+{
+  // wrap comp
+  thrust::detail::host_function<
+    BinaryPredicate,
+    bool
+  > wrapped_comp(comp);
+
+  ForwardIterator imin = first;
+
+  for (; first != last; first++)
+  {
+    if (wrapped_comp(*first, *imin))
+    {
+      imin = first;
+    }
+  }
+
+  return imin;
+}
+
+
+template <typename ForwardIterator,
+          typename BinaryPredicate>
+ForwardIterator max_element(ForwardIterator first, 
+                            ForwardIterator last,
+                            BinaryPredicate comp)
+{
+  // wrap comp
+  thrust::detail::host_function<
+    BinaryPredicate,
+    bool
+  > wrapped_comp(comp);
+
+  ForwardIterator imax = first;
+
+  for (; first != last; first++)
+  {
+    if (wrapped_comp(*imax, *first))
+    {
+      imax = first;
+    }
+  }
+
+  return imax;
+}
+
+
+template <typename ForwardIterator,
+          typename BinaryPredicate>
+thrust::pair<ForwardIterator,ForwardIterator> minmax_element(ForwardIterator first, 
+                                                             ForwardIterator last,
+                                                             BinaryPredicate comp)
+{
+  // wrap comp
+  thrust::detail::host_function<
+    BinaryPredicate,
+    bool
+  > wrapped_comp(comp);
+  
+  ForwardIterator imin = first;
+  ForwardIterator imax = first;
+
+  for (; first != last; first++)
+  {
+    if (wrapped_comp(*first, *imin))
+    {
+      imin = first;
+    }
+
+    if (wrapped_comp(*imax, *first))
+    {
+      imax = first;
+    }
+  }
+
+  return thrust::make_pair(imin, imax);
+}
+
+} // end namespace scalar
+} // end namespace internal
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/detail/internal/scalar/find.h b/compat/thrust/system/detail/internal/scalar/find.h
new file mode 100644
index 0000000000..6b2502199e
--- /dev/null
+++ b/compat/thrust/system/detail/internal/scalar/find.h
@@ -0,0 +1,67 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file find.h
+ *  \brief Sequential implementation of find_if. 
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/function.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace internal
+{
+namespace scalar
+{
+
+template <typename InputIterator,
+          typename Predicate>
+InputIterator find_if(InputIterator first,
+                      InputIterator last,
+                      Predicate pred)
+{
+  // wrap pred
+  thrust::detail::host_function<
+    Predicate,
+    bool
+  > wrapped_pred(pred);
+
+  while(first != last)
+  {
+    if (wrapped_pred(*first))
+      return first;
+
+    ++first;
+  }
+
+  // return first so zip_iterator works correctly
+  return first;
+}
+
+} // end namespace scalar
+} // end namespace internal
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/detail/internal/scalar/for_each.h b/compat/thrust/system/detail/internal/scalar/for_each.h
new file mode 100644
index 0000000000..4e31d9183b
--- /dev/null
+++ b/compat/thrust/system/detail/internal/scalar/for_each.h
@@ -0,0 +1,87 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file for_each.h
+ *  \brief Sequential implementations of for_each functions.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/function.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace internal
+{
+namespace scalar
+{
+
+template<typename InputIterator,
+         typename UnaryFunction>
+InputIterator for_each(InputIterator first,
+                       InputIterator last,
+                       UnaryFunction f)
+{
+  // wrap f
+  thrust::detail::host_function<
+    UnaryFunction,
+    void
+  > wrapped_f(f);
+
+  for(; first != last; ++first)
+  {
+    wrapped_f(*first);
+  }
+
+  return first;
+} // end for_each()
+
+template<typename InputIterator,
+         typename Size,
+         typename UnaryFunction>
+InputIterator for_each_n(InputIterator first,
+                         Size n,
+                         UnaryFunction f)
+{
+  // wrap f
+  thrust::detail::host_function<
+    UnaryFunction,
+    void
+  > wrapped_f(f);
+
+  for(Size i = 0; i != n; i++)
+  {
+    // we can dereference an OutputIterator if f does not
+    // try to use the reference for anything besides assignment
+    wrapped_f(*first);
+    ++first;
+  }
+
+  return first;
+} // end for_each_n()
+
+} // end namespace scalar
+} // end namespace internal
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/detail/internal/scalar/general_copy.h b/compat/thrust/system/detail/internal/scalar/general_copy.h
new file mode 100644
index 0000000000..aae061d31d
--- /dev/null
+++ b/compat/thrust/system/detail/internal/scalar/general_copy.h
@@ -0,0 +1,65 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+/*! \file general_copy.h
+ *  \brief Sequential copy algorithms for general iterators.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace internal
+{
+namespace scalar
+{
+
+template<typename InputIterator,
+         typename OutputIterator>
+  OutputIterator general_copy(InputIterator first,
+                              InputIterator last,
+                              OutputIterator result)
+{
+  for(; first != last; ++first, ++result)
+    *result = *first;
+  return result;
+} // end general_copy()
+
+
+template<typename InputIterator,
+         typename Size,
+         typename OutputIterator>
+  OutputIterator general_copy_n(InputIterator first,
+                                Size n,
+                                OutputIterator result)
+{
+  for(; n > Size(0); ++first, ++result, --n)
+    *result = *first;
+  return result;
+} // end general_copy_n()
+
+} // end namespace scalar
+} // end namespace internal
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/detail/internal/scalar/insertion_sort.h b/compat/thrust/system/detail/internal/scalar/insertion_sort.h
new file mode 100644
index 0000000000..5949ce7a65
--- /dev/null
+++ b/compat/thrust/system/detail/internal/scalar/insertion_sort.h
@@ -0,0 +1,149 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/detail/function.h>
+#include <thrust/system/detail/internal/scalar/copy_backward.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace internal
+{
+namespace scalar
+{
+
+template <typename RandomAccessIterator,
+          typename StrictWeakOrdering>
+void insertion_sort(RandomAccessIterator first,
+                    RandomAccessIterator last,
+                    StrictWeakOrdering comp)
+{
+  typedef typename thrust::iterator_value<RandomAccessIterator>::type value_type;
+
+  if (first == last) return;
+
+  // wrap comp
+  thrust::detail::host_function<
+    StrictWeakOrdering,
+    bool
+  > wrapped_comp(comp);
+
+  for(RandomAccessIterator i = first + 1; i != last; ++i)
+  {
+    value_type tmp = *i;
+
+    if (wrapped_comp(tmp, *first))
+    {
+      // tmp is the smallest value encountered so far
+      thrust::system::detail::internal::scalar::copy_backward(first, i, i + 1);
+
+      *first = tmp;
+    }
+    else
+    {
+      // tmp is not the smallest value, can avoid checking for j == first
+      RandomAccessIterator j = i;
+      RandomAccessIterator k = i - 1;
+
+      while(wrapped_comp(tmp, *k))
+      {
+        *j = *k;
+        j = k;
+        --k;
+      }
+
+      *j = tmp;
+    }
+  }
+}
+
+template <typename RandomAccessIterator1,
+          typename RandomAccessIterator2,
+          typename StrictWeakOrdering>
+void insertion_sort_by_key(RandomAccessIterator1 first1,
+                           RandomAccessIterator1 last1,
+                           RandomAccessIterator2 first2,
+                           StrictWeakOrdering comp)
+{
+  typedef typename thrust::iterator_value<RandomAccessIterator1>::type value_type1;
+  typedef typename thrust::iterator_value<RandomAccessIterator2>::type value_type2;
+
+  if (first1 == last1) return;
+
+  // wrap comp
+  thrust::detail::host_function<
+    StrictWeakOrdering,
+    bool
+  > wrapped_comp(comp);
+
+  RandomAccessIterator1 i1 = first1 + 1;
+  RandomAccessIterator2 i2 = first2 + 1;
+
+  for(; i1 != last1; ++i1, ++i2)
+  {
+    value_type1 tmp1 = *i1;
+    value_type2 tmp2 = *i2;
+
+    if (wrapped_comp(tmp1, *first1))
+    {
+      // tmp is the smallest value encountered so far
+      thrust::system::detail::internal::scalar::copy_backward(first1, i1, i1 + 1);
+      thrust::system::detail::internal::scalar::copy_backward(first2, i2, i2 + 1);
+
+      *first1 = tmp1;
+      *first2 = tmp2;
+    }
+    else
+    {
+      // tmp is not the smallest value, can avoid checking for j == first
+      RandomAccessIterator1 j1 = i1;
+      RandomAccessIterator1 k1 = i1 - 1;
+
+      RandomAccessIterator2 j2 = i2;
+      RandomAccessIterator2 k2 = i2 - 1;
+
+      while(wrapped_comp(tmp1, *k1))
+      {
+        *j1 = *k1;
+        *j2 = *k2;
+
+        j1 = k1;
+        j2 = k2;
+
+        --k1;
+        --k2;
+      }
+
+      *j1 = tmp1;
+      *j2 = tmp2;
+    }
+  }
+}
+
+} // end namespace scalar
+} // end namespace internal
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/detail/internal/scalar/merge.h b/compat/thrust/system/detail/internal/scalar/merge.h
new file mode 100644
index 0000000000..c02fca44b6
--- /dev/null
+++ b/compat/thrust/system/detail/internal/scalar/merge.h
@@ -0,0 +1,73 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file merge.h
+ *  \brief Sequential implementation of merge algorithms.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace internal
+{
+namespace scalar
+{
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename StrictWeakOrdering>
+OutputIterator merge(InputIterator1 first1,
+                     InputIterator1 last1,
+                     InputIterator2 first2,
+                     InputIterator2 last2,
+                     OutputIterator result,
+                     StrictWeakOrdering comp);
+
+template <typename InputIterator1,
+          typename InputIterator2,
+          typename InputIterator3,
+          typename InputIterator4,
+          typename OutputIterator1,
+          typename OutputIterator2,
+          typename StrictWeakOrdering>
+thrust::pair<OutputIterator1,OutputIterator2>
+  merge_by_key(InputIterator1 keys_first1,
+               InputIterator1 keys_last1,
+               InputIterator2 keys_first2,
+               InputIterator2 keys_last2,
+               InputIterator3 values_first1,
+               InputIterator4 values_first2,
+               OutputIterator1 keys_result,
+               OutputIterator2 values_result,
+               StrictWeakOrdering comp);
+
+} // end namespace scalar
+} // end namespace internal
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/detail/internal/scalar/merge.inl>
+
diff --git a/compat/thrust/system/detail/internal/scalar/merge.inl b/compat/thrust/system/detail/internal/scalar/merge.inl
new file mode 100644
index 0000000000..a7c2a393c4
--- /dev/null
+++ b/compat/thrust/system/detail/internal/scalar/merge.inl
@@ -0,0 +1,145 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/iterator/iterator_traits.h>
+
+#include <thrust/system/detail/internal/scalar/copy.h>
+#include <thrust/detail/function.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace internal
+{
+namespace scalar
+{
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename StrictWeakOrdering>
+OutputIterator merge(InputIterator1 first1,
+                     InputIterator1 last1,
+                     InputIterator2 first2,
+                     InputIterator2 last2,
+                     OutputIterator result,
+                     StrictWeakOrdering comp)
+{
+  // wrap comp
+  thrust::detail::host_function<
+    StrictWeakOrdering,
+    bool
+  > wrapped_comp(comp);
+
+  while(first1 != last1 && first2 != last2)
+  {
+    if(wrapped_comp(*first2, *first1))
+    {
+      *result = *first2;
+      ++first2;
+    } // end if
+    else
+    {
+      *result = *first1;
+      ++first1;
+    } // end else
+
+    ++result;
+  } // end while
+
+  return thrust::system::detail::internal::scalar::copy(first2, last2, thrust::system::detail::internal::scalar::copy(first1, last1, result));
+} // end merge()
+
+
+template <typename InputIterator1,
+          typename InputIterator2,
+          typename InputIterator3,
+          typename InputIterator4,
+          typename OutputIterator1,
+          typename OutputIterator2,
+          typename StrictWeakOrdering>
+thrust::pair<OutputIterator1,OutputIterator2>
+  merge_by_key(InputIterator1 keys_first1,
+               InputIterator1 keys_last1,
+               InputIterator2 keys_first2,
+               InputIterator2 keys_last2,
+               InputIterator3 values_first1,
+               InputIterator4 values_first2,
+               OutputIterator1 keys_result,
+               OutputIterator2 values_result,
+               StrictWeakOrdering comp)
+{
+  // wrap comp
+  thrust::detail::host_function<
+    StrictWeakOrdering,
+    bool
+  > wrapped_comp(comp);
+
+  while(keys_first1 != keys_last1 && keys_first2 != keys_last2)
+  {
+    if(!wrapped_comp(*keys_first2, *keys_first1))
+    {
+      // *keys_first1 <= *keys_first2
+      *keys_result   = *keys_first1;
+      *values_result = *values_first1;
+      ++keys_first1;
+      ++values_first1;
+    }
+    else
+    {
+      // *keys_first1 > keys_first2
+      *keys_result   = *keys_first2;
+      *values_result = *values_first2;
+      ++keys_first2;
+      ++values_first2;
+    }
+
+    ++keys_result;
+    ++values_result;
+  }
+
+  while(keys_first1 != keys_last1)
+  {
+    *keys_result   = *keys_first1;
+    *values_result = *values_first1;
+    ++keys_first1;
+    ++values_first1;
+    ++keys_result;
+    ++values_result;
+  }
+
+  while(keys_first2 != keys_last2)
+  {
+    *keys_result   = *keys_first2;
+    *values_result = *values_first2;
+    ++keys_first2;
+    ++values_first2;
+    ++keys_result;
+    ++values_result;
+  }
+
+  return thrust::make_pair(keys_result, values_result);
+}
+
+} // end namespace scalar
+} // end namespace internal
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/detail/internal/scalar/partition.h b/compat/thrust/system/detail/internal/scalar/partition.h
new file mode 100644
index 0000000000..7ba677ef22
--- /dev/null
+++ b/compat/thrust/system/detail/internal/scalar/partition.h
@@ -0,0 +1,262 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file partition.h
+ *  \brief Sequential implementations of partition functions.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/pair.h>
+#include <thrust/detail/temporary_array.h>
+#include <thrust/detail/function.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace internal
+{
+namespace scalar
+{
+
+template <typename ForwardIterator1,
+          typename ForwardIterator2>
+void iter_swap(ForwardIterator1 iter1, ForwardIterator2 iter2)
+{
+  // XXX this isn't correct because it doesn't use thrust::swap
+  using namespace thrust::detail;
+
+  typedef typename thrust::iterator_value<ForwardIterator1>::type T;
+
+  T temp = *iter1;
+  *iter1 = *iter2;
+  *iter2 = temp;
+}
+
+template<typename ForwardIterator,
+         typename Predicate>
+  ForwardIterator partition(ForwardIterator first,
+                            ForwardIterator last,
+                            Predicate pred)
+{
+  if (first == last)
+    return first;
+
+  // wrap pred
+  thrust::detail::host_function<
+    Predicate,
+    bool
+  > wrapped_pred(pred);
+
+  while (wrapped_pred(*first))
+  {
+    if (++first == last)
+      return first;
+  }
+
+  ForwardIterator next = first;
+
+  while (++next != last)
+  {
+    if (wrapped_pred(*next))
+    {
+      iter_swap(first, next);
+      ++first;
+    }
+  }
+
+  return first;
+}
+
+template<typename ForwardIterator,
+         typename Predicate>
+  ForwardIterator stable_partition(ForwardIterator first,
+                                   ForwardIterator last,
+                                   Predicate pred)
+{
+  // wrap pred
+  thrust::detail::host_function<
+    Predicate,
+    bool
+  > wrapped_pred(pred);
+
+  // XXX the type of exec should be:
+  //     typedef decltype(select_system(first, last)) system;
+  typedef typename thrust::iterator_system<ForwardIterator>::type ExecutionPolicy;
+  typedef typename thrust::iterator_value<ForwardIterator>::type T;
+
+  typedef thrust::detail::temporary_array<T,ExecutionPolicy> TempRange;
+  typedef typename TempRange::iterator                       TempIterator;
+
+  // XXX presumes ExecutionPolicy is default constructible
+  ExecutionPolicy exec;
+  TempRange temp(exec, first, last);
+
+  for(TempIterator iter = temp.begin(); iter != temp.end(); ++iter)
+  {
+    if (wrapped_pred(*iter))
+    {
+      *first = *iter;
+      ++first;
+    }
+  }
+
+  ForwardIterator middle = first;
+
+  for(TempIterator iter = temp.begin(); iter != temp.end(); ++iter)
+  {
+    if (!wrapped_pred(*iter))
+    {
+      *first = *iter;
+      ++first;
+    }
+  }
+
+  return middle;
+}
+
+template<typename ForwardIterator,
+         typename InputIterator,
+         typename Predicate>
+  ForwardIterator stable_partition(ForwardIterator first,
+                                   ForwardIterator last,
+                                   InputIterator stencil,
+                                   Predicate pred)
+{
+  // wrap pred
+  thrust::detail::host_function<
+    Predicate,
+    bool
+  > wrapped_pred(pred);
+
+  // XXX the type of exec should be:
+  //     typedef decltype(select_system(first, stencil)) system;
+  typedef typename thrust::iterator_system<ForwardIterator>::type ExecutionPolicy;
+  typedef typename thrust::iterator_value<ForwardIterator>::type T;
+
+  typedef thrust::detail::temporary_array<T,ExecutionPolicy> TempRange;
+  typedef typename TempRange::iterator                       TempIterator;
+
+  // XXX presumes ExecutionPolicy is default constructible
+  ExecutionPolicy exec;
+  TempRange temp(exec, first, last);
+
+  InputIterator stencil_iter = stencil;
+  for(TempIterator iter = temp.begin(); iter != temp.end(); ++iter, ++stencil_iter)
+  {
+    if (wrapped_pred(*stencil_iter))
+    {
+      *first = *iter;
+      ++first;
+    }
+  }
+
+  ForwardIterator middle = first;
+  stencil_iter = stencil;
+
+  for(TempIterator iter = temp.begin(); iter != temp.end(); ++iter, ++stencil_iter)
+  {
+    if (!wrapped_pred(*stencil_iter))
+    {
+      *first = *iter;
+      ++first;
+    }
+  }
+
+  return middle;
+}
+
+template<typename InputIterator,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename Predicate>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    stable_partition_copy(InputIterator first,
+                          InputIterator last,
+                          OutputIterator1 out_true,
+                          OutputIterator2 out_false,
+                          Predicate pred)
+{
+  // wrap pred
+  thrust::detail::host_function<
+    Predicate,
+    bool
+  > wrapped_pred(pred);
+
+  for(; first != last; ++first)
+  {
+    if(wrapped_pred(*first))
+    {
+      *out_true = *first;
+      ++out_true;
+    } // end if
+    else
+    {
+      *out_false = *first;
+      ++out_false;
+    } // end else
+  }
+
+  return thrust::make_pair(out_true, out_false);
+}
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename Predicate>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    stable_partition_copy(InputIterator1 first,
+                          InputIterator1 last,
+                          InputIterator2 stencil,
+                          OutputIterator1 out_true,
+                          OutputIterator2 out_false,
+                          Predicate pred)
+{
+  // wrap pred
+  thrust::detail::host_function<
+    Predicate,
+    bool
+  > wrapped_pred(pred);
+
+  for(; first != last; ++first, ++stencil)
+  {
+    if(wrapped_pred(*stencil))
+    {
+      *out_true = *first;
+      ++out_true;
+    } // end if
+    else
+    {
+      *out_false = *first;
+      ++out_false;
+    } // end else
+  }
+
+  return thrust::make_pair(out_true, out_false);
+}
+
+} // end namespace scalar
+} // end namespace internal
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/detail/internal/scalar/reduce.h b/compat/thrust/system/detail/internal/scalar/reduce.h
new file mode 100644
index 0000000000..7ad430ea50
--- /dev/null
+++ b/compat/thrust/system/detail/internal/scalar/reduce.h
@@ -0,0 +1,69 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file reduce.h
+ *  \brief Sequential implementation of reduce algorithm.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/function.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace internal
+{
+namespace scalar
+{
+
+template<typename InputIterator, 
+         typename OutputType,
+         typename BinaryFunction>
+  OutputType reduce(InputIterator begin,
+                    InputIterator end,
+                    OutputType init,
+                    BinaryFunction binary_op)
+{
+  // wrap binary_op
+  thrust::detail::host_function<
+    BinaryFunction,
+    OutputType
+  > wrapped_binary_op(binary_op);
+
+  // initialize the result
+  OutputType result = init;
+
+  while(begin != end)
+  {
+    result = wrapped_binary_op(result, *begin);
+    ++begin;
+  } // end while
+
+  return result;
+}
+
+} // end namespace scalar
+} // end namespace internal
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/detail/internal/scalar/reduce_by_key.h b/compat/thrust/system/detail/internal/scalar/reduce_by_key.h
new file mode 100644
index 0000000000..eeacb9dd49
--- /dev/null
+++ b/compat/thrust/system/detail/internal/scalar/reduce_by_key.h
@@ -0,0 +1,103 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/pair.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/detail/type_traits/algorithm/intermediate_type_from_function_and_iterators.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace internal
+{
+namespace scalar
+{
+
+template <typename InputIterator1,
+          typename InputIterator2,
+          typename OutputIterator1,
+          typename OutputIterator2,
+          typename BinaryPredicate,
+          typename BinaryFunction>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    reduce_by_key(InputIterator1 keys_first, 
+                  InputIterator1 keys_last,
+                  InputIterator2 values_first,
+                  OutputIterator1 keys_output,
+                  OutputIterator2 values_output,
+                  BinaryPredicate binary_pred,
+                  BinaryFunction binary_op)
+{
+  typedef typename thrust::iterator_traits<InputIterator1>::value_type  InputKeyType;
+  typedef typename thrust::iterator_traits<InputIterator2>::value_type  InputValueType;
+
+  typedef typename thrust::detail::intermediate_type_from_function_and_iterators<
+    InputIterator2,
+    OutputIterator2,
+    BinaryFunction
+  >::type TemporaryType;
+
+  if(keys_first != keys_last)
+  {
+    InputKeyType  temp_key   = *keys_first;
+    TemporaryType temp_value = *values_first;
+
+    for(++keys_first, ++values_first;
+        keys_first != keys_last;
+        ++keys_first, ++values_first)
+    {
+      InputKeyType    key  = *keys_first;
+      InputValueType value = *values_first;
+
+      if (binary_pred(temp_key, key))
+      {
+        temp_value = binary_op(temp_value, value);
+      }
+      else
+      {
+        *keys_output   = temp_key;
+        *values_output = temp_value;
+
+        ++keys_output;
+        ++values_output;
+
+        temp_key   = key;
+        temp_value = value;
+      }
+    }
+
+    *keys_output   = temp_key;
+    *values_output = temp_value;
+
+    ++keys_output;
+    ++values_output;
+  }
+
+  return thrust::make_pair(keys_output, values_output);
+}
+
+} // end namespace scalar
+} // end namespace internal
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/detail/internal/scalar/remove.h b/compat/thrust/system/detail/internal/scalar/remove.h
new file mode 100644
index 0000000000..2360019f11
--- /dev/null
+++ b/compat/thrust/system/detail/internal/scalar/remove.h
@@ -0,0 +1,185 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file remove.h
+ *  \brief Sequential implementations of remove functions.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/function.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace internal
+{
+namespace scalar
+{
+
+template<typename ForwardIterator,
+         typename Predicate>
+  ForwardIterator remove_if(ForwardIterator first,
+                            ForwardIterator last,
+                            Predicate pred)
+{
+  // wrap pred
+  thrust::detail::host_function<
+    Predicate,
+    bool
+  > wrapped_pred(pred);
+
+  // advance iterators until wrapped_pred(*first) is true or we reach the end of input
+  while(first != last && !wrapped_pred(*first))
+    ++first;
+
+  if(first == last)
+    return first;
+
+  // result always trails first 
+  ForwardIterator result = first;
+
+  ++first;
+
+  while(first != last)
+  {
+    if(!wrapped_pred(*first))
+    {
+      *result = *first;
+      ++result;
+    }
+    ++first;
+  }
+
+  return result;
+}
+
+
+template<typename ForwardIterator,
+         typename InputIterator,
+         typename Predicate>
+  ForwardIterator remove_if(ForwardIterator first,
+                            ForwardIterator last,
+                            InputIterator stencil,
+                            Predicate pred)
+{
+  // wrap pred
+  thrust::detail::host_function<
+    Predicate,
+    bool
+  > wrapped_pred(pred);
+
+  // advance iterators until wrapped_pred(*stencil) is true or we reach the end of input
+  while(first != last && !wrapped_pred(*stencil))
+  {
+    ++first;
+    ++stencil;
+  }
+
+  if(first == last)
+    return first;
+
+  // result always trails first 
+  ForwardIterator result = first;
+
+  ++first;
+  ++stencil;
+
+  while(first != last)
+  {
+    if(!wrapped_pred(*stencil))
+    {
+      *result = *first;
+      ++result;
+    }
+    ++first;
+    ++stencil;
+  }
+
+  return result;
+}
+
+
+template<typename InputIterator,
+         typename OutputIterator,
+         typename Predicate>
+  OutputIterator remove_copy_if(InputIterator first,
+                                InputIterator last,
+                                OutputIterator result,
+                                Predicate pred)
+{
+  // wrap pred
+  thrust::detail::host_function<
+    Predicate,
+    bool
+  > wrapped_pred(pred);
+
+  while (first != last)
+  {
+    if (!wrapped_pred(*first))
+    {
+      *result = *first;
+      ++result;
+    }
+
+    ++first;
+  }
+
+  return result;
+}
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename Predicate>
+  OutputIterator remove_copy_if(InputIterator1 first,
+                                InputIterator1 last,
+                                InputIterator2 stencil,
+                                OutputIterator result,
+                                Predicate pred)
+{
+  // wrap pred
+  thrust::detail::host_function<
+    Predicate,
+    bool
+  > wrapped_pred(pred);
+
+  while (first != last)
+  {
+    if (!wrapped_pred(*stencil))
+    {
+      *result = *first;
+      ++result;
+    }
+
+    ++first;
+    ++stencil;
+  }
+
+  return result;
+}
+
+} // end namespace scalar
+} // end namespace internal
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/detail/internal/scalar/scan.h b/compat/thrust/system/detail/internal/scalar/scan.h
new file mode 100644
index 0000000000..8f41150247
--- /dev/null
+++ b/compat/thrust/system/detail/internal/scalar/scan.h
@@ -0,0 +1,153 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file scan.h
+ *  \brief Sequential implementations of scan functions.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/cpp/detail/execution_policy.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/detail/type_traits.h>
+#include <thrust/detail/type_traits/function_traits.h>
+#include <thrust/detail/type_traits/iterator/is_output_iterator.h>
+#include <thrust/detail/function.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace internal
+{
+namespace scalar
+{
+
+template<typename InputIterator,
+         typename OutputIterator,
+         typename BinaryFunction>
+  OutputIterator inclusive_scan(InputIterator first,
+                                InputIterator last,
+                                OutputIterator result,
+                                BinaryFunction binary_op)
+{
+  // the pseudocode for deducing the type of the temporary used below:
+  // 
+  // if BinaryFunction is AdaptableBinaryFunction
+  //   TemporaryType = AdaptableBinaryFunction::result_type
+  // else if OutputIterator is a "pure" output iterator
+  //   TemporaryType = InputIterator::value_type
+  // else
+  //   TemporaryType = OutputIterator::value_type
+  //
+  // XXX upon c++0x, TemporaryType needs to be:
+  // result_of<BinaryFunction>::type
+  
+  using namespace thrust::detail;
+
+  typedef typename eval_if<
+    has_result_type<BinaryFunction>::value,
+    result_type<BinaryFunction>,
+    eval_if<
+      is_output_iterator<OutputIterator>::value,
+      thrust::iterator_value<InputIterator>,
+      thrust::iterator_value<OutputIterator>
+    >
+  >::type ValueType;
+
+  // wrap binary_op
+  thrust::detail::host_function<
+    BinaryFunction,
+    ValueType
+  > wrapped_binary_op(binary_op);
+
+  if(first != last)
+  {
+    ValueType sum = *first;
+
+    *result = sum;
+
+    for(++first, ++result; first != last; ++first, ++result)
+      *result = sum = wrapped_binary_op(sum,*first);
+  }
+
+  return result;
+}
+
+
+template<typename InputIterator,
+         typename OutputIterator,
+         typename T,
+         typename BinaryFunction>
+  OutputIterator exclusive_scan(InputIterator first,
+                                InputIterator last,
+                                OutputIterator result,
+                                T init,
+                                BinaryFunction binary_op)
+{
+  // the pseudocode for deducing the type of the temporary used below:
+  // 
+  // if BinaryFunction is AdaptableBinaryFunction
+  //   TemporaryType = AdaptableBinaryFunction::result_type
+  // else if OutputIterator is a "pure" output iterator
+  //   TemporaryType = InputIterator::value_type
+  // else
+  //   TemporaryType = OutputIterator::value_type
+  //
+  // XXX upon c++0x, TemporaryType needs to be:
+  // result_of<BinaryFunction>::type
+
+  using namespace thrust::detail;
+
+  typedef typename eval_if<
+    has_result_type<BinaryFunction>::value,
+    result_type<BinaryFunction>,
+    eval_if<
+      is_output_iterator<OutputIterator>::value,
+      thrust::iterator_value<InputIterator>,
+      thrust::iterator_value<OutputIterator>
+    >
+  >::type ValueType;
+
+  if(first != last)
+  {
+    ValueType tmp = *first;  // temporary value allows in-situ scan
+    ValueType sum = init;
+
+    *result = sum;
+    sum = binary_op(sum, tmp);
+
+    for(++first, ++result; first != last; ++first, ++result)
+    {
+      tmp = *first;
+      *result = sum;
+      sum = binary_op(sum, tmp);
+    }
+  }
+
+  return result;
+} 
+
+} // end namespace scalar
+} // end namespace internal
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/detail/internal/scalar/scan_by_key.h b/compat/thrust/system/detail/internal/scalar/scan_by_key.h
new file mode 100644
index 0000000000..a31fc60bab
--- /dev/null
+++ b/compat/thrust/system/detail/internal/scalar/scan_by_key.h
@@ -0,0 +1,147 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file scan_by_key.h
+ *  \brief Sequential implementation of scan_by_key functions.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/cpp/detail/execution_policy.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/detail/function.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace internal
+{
+namespace scalar
+{
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename BinaryPredicate,
+         typename BinaryFunction>
+  OutputIterator inclusive_scan_by_key(InputIterator1 first1,
+                                       InputIterator1 last1,
+                                       InputIterator2 first2,
+                                       OutputIterator result,
+                                       BinaryPredicate binary_pred,
+                                       BinaryFunction binary_op)
+{
+  using namespace thrust::detail;
+
+  typedef typename thrust::iterator_traits<InputIterator1>::value_type KeyType;
+  typedef typename thrust::iterator_traits<OutputIterator>::value_type ValueType;
+
+  // wrap binary_op
+  thrust::detail::host_function<
+    BinaryFunction,
+    ValueType
+  > wrapped_binary_op(binary_op);
+
+  if(first1 != last1)
+  {
+    KeyType   prev_key   = *first1;
+    ValueType prev_value = *first2;
+
+    *result = prev_value;
+
+    for(++first1, ++first2, ++result;
+        first1 != last1;
+        ++first1, ++first2, ++result)
+    {
+      KeyType key = *first1;
+
+      if (binary_pred(prev_key, key))
+        *result = prev_value = wrapped_binary_op(prev_value,*first2);
+      else
+        *result = prev_value = *first2;
+
+      prev_key = key;
+    }
+  }
+
+  return result;
+}
+
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename T,
+         typename BinaryPredicate,
+         typename BinaryFunction>
+  OutputIterator exclusive_scan_by_key(InputIterator1 first1,
+                                       InputIterator1 last1,
+                                       InputIterator2 first2,
+                                       OutputIterator result,
+                                       T init,
+                                       BinaryPredicate binary_pred,
+                                       BinaryFunction binary_op)
+{
+  using namespace thrust::detail;
+
+  typedef typename thrust::iterator_traits<InputIterator1>::value_type KeyType;
+  typedef typename thrust::iterator_traits<OutputIterator>::value_type ValueType;
+
+  if(first1 != last1)
+  {
+    KeyType   temp_key   = *first1;
+    ValueType temp_value = *first2;
+
+    ValueType next = init;
+
+    // first one is init
+    *result = next;
+
+    next = binary_op(next, temp_value);
+
+    for(++first1, ++first2, ++result;
+        first1 != last1;
+        ++first1, ++first2, ++result)
+    {
+      KeyType key = *first1;
+
+      // use temp to permit in-place scans
+      temp_value = *first2;
+
+      if (!binary_pred(temp_key, key))
+        next = init;  // reset sum
+
+      *result = next;  
+      next = binary_op(next, temp_value);
+
+      temp_key = key;
+    }
+  }
+
+  return result;
+}
+
+} // end namespace scalar
+} // end namespace internal
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/detail/internal/scalar/set_operations.h b/compat/thrust/system/detail/internal/scalar/set_operations.h
new file mode 100644
index 0000000000..f85b5108af
--- /dev/null
+++ b/compat/thrust/system/detail/internal/scalar/set_operations.h
@@ -0,0 +1,208 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file set_operations.h
+ *  \brief Sequential implementation of set operation functions.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/internal/scalar/copy.h>
+#include <thrust/detail/function.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace internal
+{
+namespace scalar
+{
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename StrictWeakOrdering>
+  OutputIterator set_difference(InputIterator1 first1,
+                                InputIterator1 last1,
+                                InputIterator2 first2,
+                                InputIterator2 last2,
+                                OutputIterator result,
+                                StrictWeakOrdering comp)
+{
+  // wrap comp
+  thrust::detail::host_function<
+    StrictWeakOrdering,
+    bool
+  > wrapped_comp(comp);
+
+  while(first1 != last1 && first2 != last2)
+  {
+    if(wrapped_comp(*first1,*first2))
+    {
+      *result = *first1;
+      ++first1;
+      ++result;
+    } // end if
+    else if(wrapped_comp(*first2,*first1))
+    {
+      ++first2;
+    } // end else if
+    else
+    {
+      ++first1;
+      ++first2;
+    } // end else
+  } // end while
+
+  return scalar::copy(first1, last1, result);
+} // end set_difference()
+
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename StrictWeakOrdering>
+  OutputIterator set_intersection(InputIterator1 first1,
+                                  InputIterator1 last1,
+                                  InputIterator2 first2,
+                                  InputIterator2 last2,
+                                  OutputIterator result,
+                                  StrictWeakOrdering comp)
+{
+  // wrap comp
+  thrust::detail::host_function<
+    StrictWeakOrdering,
+    bool
+  > wrapped_comp(comp);
+
+  while(first1 != last1 && first2 != last2)
+  {
+    if(wrapped_comp(*first1,*first2))
+    {
+      ++first1;
+    } // end if
+    else if(wrapped_comp(*first2,*first1))
+    {
+      ++first2;
+    } // end else if
+    else
+    {
+      *result = *first1;
+      ++first1;
+      ++first2;
+      ++result;
+    } // end else
+  } // end while
+
+  return result;
+} // end set_intersection()
+
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename StrictWeakOrdering>
+  OutputIterator set_symmetric_difference(InputIterator1 first1,
+                                          InputIterator1 last1,
+                                          InputIterator2 first2,
+                                          InputIterator2 last2,
+                                          OutputIterator result,
+                                          StrictWeakOrdering comp)
+{
+  // wrap comp
+  thrust::detail::host_function<
+    StrictWeakOrdering,
+    bool
+  > wrapped_comp(comp);
+
+  while(first1 != last1 && first2 != last2)
+  {
+    if(wrapped_comp(*first1,*first2))
+    {
+      *result = *first1;
+      ++first1;
+      ++result;
+    } // end if
+    else if(wrapped_comp(*first2,*first1))
+    {
+      *result = *first2;
+      ++first2;
+      ++result;
+    } // end else if
+    else
+    {
+      ++first1;
+      ++first2;
+    } // end else
+  } // end while
+
+  return scalar::copy(first2, last2, scalar::copy(first1, last1, result));
+} // end set_symmetric_difference()
+
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename StrictWeakOrdering>
+  OutputIterator set_union(InputIterator1 first1,
+                           InputIterator1 last1,
+                           InputIterator2 first2,
+                           InputIterator2 last2,
+                           OutputIterator result,
+                           StrictWeakOrdering comp)
+{
+  // wrap comp
+  thrust::detail::host_function<
+    StrictWeakOrdering,
+    bool
+  > wrapped_comp(comp);
+
+  while(first1 != last1 && first2 != last2)
+  {
+    if(wrapped_comp(*first1,*first2))
+    {
+      *result = *first1;
+      ++first1;
+    } // end if
+    else if(wrapped_comp(*first2,*first1))
+    {
+      *result = *first2;
+      ++first2;
+    } // end else if
+    else
+    {
+      *result = *first1;
+      ++first1;
+      ++first2;
+    } // end else
+
+    ++result;
+  } // end while
+
+  return scalar::copy(first2, last2, scalar::copy(first1, last1, result));
+} // end set_union()
+
+} // end namespace scalar
+} // end namespace internal
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/detail/internal/scalar/sort.h b/compat/thrust/system/detail/internal/scalar/sort.h
new file mode 100644
index 0000000000..9e465c8ca0
--- /dev/null
+++ b/compat/thrust/system/detail/internal/scalar/sort.h
@@ -0,0 +1,57 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+/*! \file sort.h
+ *  \brief Sequential implementations of sort algorithms.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace internal
+{
+namespace scalar
+{
+
+template<typename RandomAccessIterator,
+         typename StrictWeakOrdering>
+void stable_sort(RandomAccessIterator first,
+                 RandomAccessIterator last,
+                 StrictWeakOrdering comp);
+
+template<typename RandomAccessIterator1,
+         typename RandomAccessIterator2,
+         typename StrictWeakOrdering>
+void stable_sort_by_key(RandomAccessIterator1 first1,
+                        RandomAccessIterator1 last1,
+                        RandomAccessIterator2 first2,
+                        StrictWeakOrdering comp);
+
+} // end namespace scalar
+} // end namespace internal
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/detail/internal/scalar/sort.inl>
+
diff --git a/compat/thrust/system/detail/internal/scalar/sort.inl b/compat/thrust/system/detail/internal/scalar/sort.inl
new file mode 100644
index 0000000000..c6ed27324c
--- /dev/null
+++ b/compat/thrust/system/detail/internal/scalar/sort.inl
@@ -0,0 +1,161 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+#include <thrust/reverse.h>
+#include <thrust/detail/type_traits.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/system/detail/internal/scalar/stable_merge_sort.h>
+#include <thrust/system/detail/internal/scalar/stable_primitive_sort.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace internal
+{
+namespace scalar
+{
+namespace sort_detail
+{
+
+////////////////////
+// Primitive Sort //
+////////////////////
+
+template<typename RandomAccessIterator,
+         typename StrictWeakOrdering>
+void stable_sort(RandomAccessIterator first,
+                 RandomAccessIterator last,
+                 StrictWeakOrdering comp,
+                 thrust::detail::true_type)
+{
+  thrust::system::detail::internal::scalar::stable_primitive_sort(first, last);
+        
+  // if comp is greater<T> then reverse the keys
+  typedef typename thrust::iterator_traits<RandomAccessIterator>::value_type KeyType;
+  const static bool reverse = thrust::detail::is_same<StrictWeakOrdering, typename thrust::greater<KeyType> >::value;
+
+  if (reverse)
+    thrust::reverse(first, last);
+}
+
+template<typename RandomAccessIterator1,
+         typename RandomAccessIterator2,
+         typename StrictWeakOrdering>
+void stable_sort_by_key(RandomAccessIterator1 first1,
+                        RandomAccessIterator1 last1,
+                        RandomAccessIterator2 first2,
+                        StrictWeakOrdering comp,
+                        thrust::detail::true_type)
+{
+  // if comp is greater<T> then reverse the keys and values
+  typedef typename thrust::iterator_traits<RandomAccessIterator1>::value_type KeyType;
+  const static bool reverse = thrust::detail::is_same<StrictWeakOrdering, typename thrust::greater<KeyType> >::value;
+
+  // note, we also have to reverse the (unordered) input to preserve stability
+  if (reverse)
+  {
+    thrust::reverse(first1,  last1);
+    thrust::reverse(first2, first2 + (last1 - first1));
+  }
+
+  thrust::system::detail::internal::scalar::stable_primitive_sort_by_key(first1, last1, first2);
+
+  if (reverse)
+  {
+    thrust::reverse(first1,  last1);
+    thrust::reverse(first2, first2 + (last1 - first1));
+  }
+}
+
+////////////////
+// Merge Sort //
+////////////////
+
+template<typename RandomAccessIterator,
+         typename StrictWeakOrdering>
+void stable_sort(RandomAccessIterator first,
+                 RandomAccessIterator last,
+                 StrictWeakOrdering comp,
+                 thrust::detail::false_type)
+{
+  thrust::system::detail::internal::scalar::stable_merge_sort(first, last, comp);
+}
+
+template<typename RandomAccessIterator1,
+         typename RandomAccessIterator2,
+         typename StrictWeakOrdering>
+void stable_sort_by_key(RandomAccessIterator1 first1,
+                        RandomAccessIterator1 last1,
+                        RandomAccessIterator2 first2,
+                        StrictWeakOrdering comp,
+                        thrust::detail::false_type)
+{
+  thrust::system::detail::internal::scalar::stable_merge_sort_by_key(first1, last1, first2, comp);
+}
+
+
+} // end namespace sort_detail
+
+template<typename RandomAccessIterator,
+         typename StrictWeakOrdering>
+void stable_sort(RandomAccessIterator first,
+                 RandomAccessIterator last,
+                 StrictWeakOrdering comp)
+{
+  typedef typename thrust::iterator_traits<RandomAccessIterator>::value_type KeyType;
+  static const bool use_primitive_sort = thrust::detail::is_arithmetic<KeyType>::value &&
+                                         (thrust::detail::is_same<StrictWeakOrdering, typename thrust::less<KeyType> >::value ||
+                                          thrust::detail::is_same<StrictWeakOrdering, typename thrust::greater<KeyType> >::value);
+
+  // supress unused variable warning
+  (void) use_primitive_sort;
+
+  thrust::system::detail::internal::scalar::sort_detail::stable_sort
+    (first, last, comp, 
+      thrust::detail::integral_constant<bool, use_primitive_sort>());
+}
+
+template<typename RandomAccessIterator1,
+         typename RandomAccessIterator2,
+         typename StrictWeakOrdering>
+void stable_sort_by_key(RandomAccessIterator1 first1,
+                        RandomAccessIterator1 last1,
+                        RandomAccessIterator2 first2,
+                        StrictWeakOrdering comp)
+{
+  typedef typename thrust::iterator_traits<RandomAccessIterator1>::value_type KeyType;
+  static const bool use_primitive_sort = thrust::detail::is_arithmetic<KeyType>::value &&
+                                         (thrust::detail::is_same<StrictWeakOrdering, typename thrust::less<KeyType> >::value ||
+                                          thrust::detail::is_same<StrictWeakOrdering, typename thrust::greater<KeyType> >::value);
+
+  // supress unused variable warning
+  (void) use_primitive_sort;
+
+  thrust::system::detail::internal::scalar::sort_detail::stable_sort_by_key
+    (first1, last1, first2, comp, 
+      thrust::detail::integral_constant<bool, use_primitive_sort>());
+}
+
+} // end namespace scalar
+} // end namespace internal
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/detail/internal/scalar/stable_merge_sort.h b/compat/thrust/system/detail/internal/scalar/stable_merge_sort.h
new file mode 100644
index 0000000000..f68242cc5a
--- /dev/null
+++ b/compat/thrust/system/detail/internal/scalar/stable_merge_sort.h
@@ -0,0 +1,55 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file stable_merge_sort.h
+ *  \brief Sequential implementation of merge sort.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace cpp
+{
+namespace detail
+{
+
+template<typename RandomAccessIterator,
+         typename StrictWeakOrdering>
+void stable_merge_sort(RandomAccessIterator begin,
+                       RandomAccessIterator end,
+                       StrictWeakOrdering comp);
+
+template<typename RandomAccessIterator1,
+         typename RandomAccessIterator2,
+         typename StrictWeakOrdering>
+void stable_merge_sort_by_key(RandomAccessIterator1 keys_begin,
+                              RandomAccessIterator1 keys_end,
+                              RandomAccessIterator2 values_begin,
+                              StrictWeakOrdering comp);
+
+} // end namespace detail
+} // end namespace cpp
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/detail/internal/scalar/stable_merge_sort.inl>
+
diff --git a/compat/thrust/system/detail/internal/scalar/stable_merge_sort.inl b/compat/thrust/system/detail/internal/scalar/stable_merge_sort.inl
new file mode 100644
index 0000000000..41d320cb08
--- /dev/null
+++ b/compat/thrust/system/detail/internal/scalar/stable_merge_sort.inl
@@ -0,0 +1,150 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/detail/temporary_array.h>
+#include <thrust/system/detail/internal/scalar/merge.h>
+#include <thrust/system/detail/internal/scalar/insertion_sort.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace internal
+{
+namespace scalar
+{
+namespace detail
+{
+
+template <typename RandomAccessIterator,
+          typename StrictWeakOrdering>
+void inplace_merge(RandomAccessIterator first,
+                   RandomAccessIterator middle,
+                   RandomAccessIterator last,
+                   StrictWeakOrdering comp)
+{
+  // XXX the type of exec should be:
+  //     typedef decltype(select_system(first, middle, last)) DerivedPolicy;
+  typedef typename thrust::iterator_system<RandomAccessIterator>::type DerivedPolicy;
+  typedef typename thrust::iterator_value<RandomAccessIterator>::type value_type;
+
+  // XXX assumes DerivedPolicy is default constructible
+  // XXX find a way to get a stateful execution policy into this function
+  //     or simply pass scratch space
+  DerivedPolicy exec;
+  thrust::detail::temporary_array<value_type, DerivedPolicy> a(exec, first, middle);
+  thrust::detail::temporary_array<value_type, DerivedPolicy> b(exec, middle, last);
+
+  thrust::system::detail::internal::scalar::merge(a.begin(), a.end(), b.begin(), b.end(), first, comp);
+}
+
+template <typename RandomAccessIterator1,
+          typename RandomAccessIterator2,
+          typename StrictWeakOrdering>
+void inplace_merge_by_key(RandomAccessIterator1 first1,
+                          RandomAccessIterator1 middle1,
+                          RandomAccessIterator1 last1,
+                          RandomAccessIterator2 first2,
+                          StrictWeakOrdering comp)
+{
+  // XXX the type of exec should be:
+  //     typedef decltype(select_system(first1, middle1, last1, first2)) DerivedPolicy;
+  typedef typename thrust::iterator_system<RandomAccessIterator1>::type DerivedPolicy;
+  typedef typename thrust::iterator_value<RandomAccessIterator1>::type value_type1;
+  typedef typename thrust::iterator_value<RandomAccessIterator2>::type value_type2;
+
+  RandomAccessIterator2 middle2 = first2 + (middle1 - first1);
+  RandomAccessIterator2 last2   = first2 + (last1   - first1);
+
+  // XXX assumes DerivedPolicy is default constructible
+  // XXX find a way to get a stateful exec into this function
+  //     or simply pass scratch space
+  DerivedPolicy exec;
+  thrust::detail::temporary_array<value_type1, DerivedPolicy> lhs1(exec, first1, middle1);
+  thrust::detail::temporary_array<value_type1, DerivedPolicy> rhs1(exec, middle1, last1);
+  thrust::detail::temporary_array<value_type2, DerivedPolicy> lhs2(exec, first2, middle2);
+  thrust::detail::temporary_array<value_type2, DerivedPolicy> rhs2(exec, middle2, last2);
+
+  thrust::system::detail::internal::scalar::merge_by_key
+    (lhs1.begin(), lhs1.end(), rhs1.begin(), rhs1.end(),
+     lhs2.begin(), rhs2.begin(),
+     first1, first2, comp);
+}
+
+} // end namespace detail
+
+//////////////
+// Key Sort //
+//////////////
+
+template <typename RandomAccessIterator,
+          typename StrictWeakOrdering>
+void stable_merge_sort(RandomAccessIterator first,
+                       RandomAccessIterator last,
+                       StrictWeakOrdering comp)
+{
+  if (last - first < 32)
+  {
+    thrust::system::detail::internal::scalar::insertion_sort(first, last, comp);
+  }
+  else
+  {
+    RandomAccessIterator middle = first + (last - first) / 2;
+
+    thrust::system::detail::internal::scalar::stable_merge_sort(first, middle, comp);
+    thrust::system::detail::internal::scalar::stable_merge_sort(middle,  last, comp);
+    detail::inplace_merge(first, middle, last, comp);
+  }
+}
+
+
+////////////////////
+// Key-Value Sort //
+////////////////////
+
+template <typename RandomAccessIterator1,
+          typename RandomAccessIterator2,
+          typename StrictWeakOrdering>
+void stable_merge_sort_by_key(RandomAccessIterator1 first1,
+                              RandomAccessIterator1 last1,
+                              RandomAccessIterator2 first2,
+                              StrictWeakOrdering comp)
+{
+  if (last1 - first1 <= 32)
+  {
+    thrust::system::detail::internal::scalar::insertion_sort_by_key(first1, last1, first2, comp);
+  }
+  else
+  {
+    RandomAccessIterator1 middle1 = first1 + (last1 - first1) / 2;
+    RandomAccessIterator2 middle2 = first2 + (last1 - first1) / 2;
+
+    thrust::system::detail::internal::scalar::stable_merge_sort_by_key(first1, middle1, first2,  comp);
+    thrust::system::detail::internal::scalar::stable_merge_sort_by_key(middle1,  last1, middle2, comp);
+    detail::inplace_merge_by_key(first1, middle1, last1, first2, comp);
+  }
+}
+
+} // end namespace scalar
+} // end namespace internal
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/detail/internal/scalar/stable_primitive_sort.h b/compat/thrust/system/detail/internal/scalar/stable_primitive_sort.h
new file mode 100644
index 0000000000..f37bf27cd1
--- /dev/null
+++ b/compat/thrust/system/detail/internal/scalar/stable_primitive_sort.h
@@ -0,0 +1,49 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace internal
+{
+namespace scalar
+{
+
+template<typename RandomAccessIterator>
+void stable_primitive_sort(RandomAccessIterator first,
+                           RandomAccessIterator last);
+
+template<typename RandomAccessIterator1,
+         typename RandomAccessIterator2>
+void stable_primitive_sort_by_key(RandomAccessIterator1 keys_first,
+                                  RandomAccessIterator1 keys_last,
+                                  RandomAccessIterator2 values_first);
+
+} // end namespace scalar
+} // end namespace internal
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/detail/internal/scalar/stable_primitive_sort.inl>
+
diff --git a/compat/thrust/system/detail/internal/scalar/stable_primitive_sort.inl b/compat/thrust/system/detail/internal/scalar/stable_primitive_sort.inl
new file mode 100644
index 0000000000..c22b15c0e4
--- /dev/null
+++ b/compat/thrust/system/detail/internal/scalar/stable_primitive_sort.inl
@@ -0,0 +1,142 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/internal/scalar/stable_primitive_sort.h>
+#include <thrust/system/detail/internal/scalar/stable_radix_sort.h>
+#include <thrust/functional.h>
+#include <thrust/system/detail/internal/scalar/partition.h>
+#include <thrust/iterator/zip_iterator.h>
+#include <thrust/detail/type_traits.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace internal
+{
+namespace scalar
+{
+namespace stable_primitive_sort_detail
+{
+
+
+template<typename Iterator>
+  struct enable_if_bool_sort
+    : thrust::detail::enable_if<
+        thrust::detail::is_same<
+          bool,
+          typename thrust::iterator_value<Iterator>::type
+        >::value
+      >
+{};
+
+
+template<typename Iterator>
+  struct disable_if_bool_sort
+    : thrust::detail::disable_if<
+        thrust::detail::is_same<
+          bool,
+          typename thrust::iterator_value<Iterator>::type
+        >::value
+      >
+{};
+
+
+
+template<typename RandomAccessIterator>
+  typename enable_if_bool_sort<RandomAccessIterator>::type
+    stable_primitive_sort(RandomAccessIterator first, RandomAccessIterator last)
+{
+  // use stable_partition if we're sorting bool
+  // stable_partition puts true values first, so we need to logical_not
+  scalar::stable_partition(first, last, thrust::logical_not<bool>());
+}
+
+
+template<typename RandomAccessIterator>
+  typename disable_if_bool_sort<RandomAccessIterator>::type
+    stable_primitive_sort(RandomAccessIterator first, RandomAccessIterator last)
+{
+  // call stable_radix_sort
+  scalar::stable_radix_sort(first,last);
+}
+
+
+struct logical_not_first
+{
+  template<typename Tuple>
+  __host__ __device__
+  bool operator()(Tuple t)
+  {
+    return !thrust::get<0>(t);
+  }
+};
+
+
+template<typename RandomAccessIterator1, typename RandomAccessIterator2>
+  typename enable_if_bool_sort<RandomAccessIterator1>::type
+    stable_primitive_sort_by_key(RandomAccessIterator1 keys_first, RandomAccessIterator1 keys_last,
+                                 RandomAccessIterator2 values_first)
+{
+  // use stable_partition if we're sorting bool
+  // stable_partition puts true values first, so we need to logical_not
+  scalar::stable_partition(thrust::make_zip_iterator(thrust::make_tuple(keys_first, values_first)),
+                           thrust::make_zip_iterator(thrust::make_tuple(keys_last, values_first)),
+                           logical_not_first());
+}
+
+
+template<typename RandomAccessIterator1, typename RandomAccessIterator2>
+  typename disable_if_bool_sort<RandomAccessIterator1>::type
+    stable_primitive_sort_by_key(RandomAccessIterator1 keys_first, RandomAccessIterator1 keys_last,
+                                 RandomAccessIterator2 values_first)
+{
+  // call stable_radix_sort_by_key
+  scalar::stable_radix_sort_by_key(keys_first, keys_last, values_first);
+}
+
+
+}
+
+template<typename RandomAccessIterator>
+void stable_primitive_sort(RandomAccessIterator first,
+                           RandomAccessIterator last)
+{
+  scalar::stable_primitive_sort_detail::stable_primitive_sort(first,last);
+}
+
+template<typename RandomAccessIterator1,
+         typename RandomAccessIterator2>
+void stable_primitive_sort_by_key(RandomAccessIterator1 keys_first,
+                                  RandomAccessIterator1 keys_last,
+                                  RandomAccessIterator2 values_first)
+{
+  scalar::stable_primitive_sort_detail::stable_primitive_sort_by_key(keys_first, keys_last, values_first);
+}
+
+} // end namespace scalar
+} // end namespace internal
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/detail/internal/scalar/stable_primitive_sort.inl>
+
diff --git a/compat/thrust/system/detail/internal/scalar/stable_radix_sort.h b/compat/thrust/system/detail/internal/scalar/stable_radix_sort.h
new file mode 100644
index 0000000000..f2af22263a
--- /dev/null
+++ b/compat/thrust/system/detail/internal/scalar/stable_radix_sort.h
@@ -0,0 +1,54 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file stable_radix_sort.h
+ *  \brief Sequential implementation of radix sort.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace internal
+{
+namespace scalar
+{
+
+template<typename RandomAccessIterator>
+void stable_radix_sort(RandomAccessIterator begin,
+                       RandomAccessIterator end);
+
+template<typename RandomAccessIterator1,
+         typename RandomAccessIterator2>
+void stable_radix_sort_by_key(RandomAccessIterator1 keys_begin,
+                              RandomAccessIterator1 keys_end,
+                              RandomAccessIterator2 values_begin);
+
+} // end namespace scalar
+} // end namespace internal
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/detail/internal/scalar/stable_radix_sort.inl>
+
diff --git a/compat/thrust/system/detail/internal/scalar/stable_radix_sort.inl b/compat/thrust/system/detail/internal/scalar/stable_radix_sort.inl
new file mode 100644
index 0000000000..98846ab101
--- /dev/null
+++ b/compat/thrust/system/detail/internal/scalar/stable_radix_sort.inl
@@ -0,0 +1,434 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+#include <limits>
+
+#include <thrust/copy.h>
+#include <thrust/functional.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/detail/temporary_array.h>
+#include <thrust/detail/cstdint.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace internal
+{
+namespace scalar
+{
+namespace detail
+{
+
+template <typename T>
+struct RadixEncoder : public thrust::identity<T>
+{};
+
+template <>
+struct RadixEncoder<char> : public thrust::unary_function<char, unsigned char>
+{
+  unsigned char operator()(char x) const
+  {
+    if(std::numeric_limits<char>::is_signed)
+      return x ^ static_cast<unsigned char>(1) << (8 * sizeof(unsigned char) - 1);
+    else
+      return x;
+  }
+};
+
+template <>
+struct RadixEncoder<signed char> : public thrust::unary_function<signed char, unsigned char>
+{
+  unsigned char operator()(signed char x) const
+  {
+    return x ^ static_cast<unsigned char>(1) << (8 * sizeof(unsigned char) - 1);
+  }
+};
+
+template <>
+struct RadixEncoder<short> : public thrust::unary_function<short, unsigned short>
+{
+  unsigned short operator()(short x) const
+  {
+    return x ^ static_cast<unsigned short>(1) << (8 * sizeof(unsigned short) - 1);
+  }
+};
+
+template <>
+struct RadixEncoder<int> : public thrust::unary_function<int, unsigned int>
+{
+  unsigned long operator()(long x) const
+  {
+    return x ^ static_cast<unsigned int>(1) << (8 * sizeof(unsigned int) - 1);
+  }
+};
+
+template <>
+struct RadixEncoder<long> : public thrust::unary_function<long, unsigned long>
+{
+  unsigned long operator()(long x) const
+  {
+    return x ^ static_cast<unsigned long>(1) << (8 * sizeof(unsigned long) - 1);
+  }
+};
+
+template <>
+struct RadixEncoder<long long> : public thrust::unary_function<long long, unsigned long long>
+{
+  unsigned long long operator()(long long x) const
+  {
+    return x ^ static_cast<unsigned long long>(1) << (8 * sizeof(unsigned long long) - 1);
+  }
+};
+
+// ideally we'd use uint32 here and uint64 below
+template <>
+struct RadixEncoder<float> : public thrust::unary_function<float, thrust::detail::uint32_t>
+{
+  thrust::detail::uint32_t operator()(float x) const
+  {
+    union { float f; thrust::detail::uint32_t i; } u;
+    u.f = x;
+    thrust::detail::uint32_t mask = -static_cast<thrust::detail::int32_t>(u.i >> 31) | (static_cast<thrust::detail::uint32_t>(1) << 31);
+    return u.i ^ mask;
+  }
+};
+
+template <>
+struct RadixEncoder<double> : public thrust::unary_function<double, thrust::detail::uint64_t>
+{
+  thrust::detail::uint64_t operator()(double x) const
+  {
+    union { double f; thrust::detail::uint64_t i; } u;
+    u.f = x;
+    thrust::detail::uint64_t mask = -static_cast<thrust::detail::int64_t>(u.i >> 63) | (static_cast<thrust::detail::uint64_t>(1) << 63);
+    return u.i ^ mask;
+  }
+};
+
+
+template <unsigned int RadixBits,
+          bool HasValues,
+          typename RandomAccessIterator1,
+          typename RandomAccessIterator2,
+          typename RandomAccessIterator3,
+          typename RandomAccessIterator4>
+void radix_sort(RandomAccessIterator1 keys1,
+                RandomAccessIterator2 keys2,
+                RandomAccessIterator3 vals1,
+                RandomAccessIterator4 vals2,
+                const size_t N)
+{
+  typedef typename thrust::iterator_value<RandomAccessIterator1>::type KeyType;
+
+  typedef RadixEncoder<KeyType> Encoder;
+  typedef typename Encoder::result_type EncodedType;
+
+  static const unsigned int NumHistograms = (8 * sizeof(EncodedType) + (RadixBits - 1)) / RadixBits;
+  static const unsigned int HistogramSize =  1 << RadixBits;
+
+  static const EncodedType BitMask = static_cast<EncodedType>((1 << RadixBits) - 1);
+  
+  Encoder encode;
+
+  // storage for histograms
+  size_t histograms[NumHistograms][HistogramSize] = {{0}};
+
+  // see which passes can be eliminated
+  bool skip_shuffle[NumHistograms] = {false};
+  
+  // false if most recent data is stored in (keys1,vals1)
+  bool flip = false;
+    
+  // compute histograms
+  for (size_t i = 0; i < N; i++)
+  {
+    const EncodedType x = encode(keys1[i]);
+
+    for (unsigned int j = 0; j < NumHistograms; j++)
+    {
+      const EncodedType BitShift = RadixBits * j;
+      histograms[j][(x >> BitShift) & BitMask]++;
+    }
+  }
+
+  // scan histograms
+  for (unsigned int i = 0; i < NumHistograms; i++)
+  {
+    size_t sum = 0;
+
+    for (unsigned int j = 0; j < HistogramSize; j++)
+    {
+      size_t bin = histograms[i][j];
+
+      if (bin == N)
+        skip_shuffle[i] = true;
+
+      histograms[i][j] = sum;
+
+      sum = sum + bin;
+    }
+  }
+
+  // shuffle keys and (optionally) values 
+  for (unsigned int i = 0; i < NumHistograms; i++)
+  {
+    const EncodedType BitShift = static_cast<EncodedType>(RadixBits * i);
+
+    if (!skip_shuffle[i])
+    {
+      if (flip)
+      {
+        for (size_t j = 0; j < N; j++)
+        {
+          const EncodedType x = encode(keys2[j]);
+          size_t position = histograms[i][(x >> BitShift) & BitMask]++;
+
+          RandomAccessIterator1 temp_keys1 = keys1;
+          temp_keys1 += position;
+
+          RandomAccessIterator2 temp_keys2 = keys2;
+          temp_keys2 += j;
+
+          // keys1[position] = keys2[j]
+          *temp_keys1 = *temp_keys2;
+
+          if (HasValues)
+          {
+            RandomAccessIterator3 temp_vals1 = vals1;
+            temp_vals1 += position;
+
+            RandomAccessIterator4 temp_vals2 = vals2;
+            temp_vals2 += j;
+
+            // vals1[position] = vals2[j]
+            *temp_vals1 = *temp_vals2;
+          }
+        }
+      }
+      else
+      {
+        for (size_t j = 0; j < N; j++)
+        {
+          const EncodedType x = encode(keys1[j]);
+          size_t position = histograms[i][(x >> BitShift) & BitMask]++;
+
+          RandomAccessIterator1 temp_keys1 = keys1;
+          temp_keys1 += j;
+
+          RandomAccessIterator2 temp_keys2 = keys2;
+          temp_keys2 += position;
+
+          // keys2[position] = keys1[j];
+          *temp_keys2 = *temp_keys1;
+
+          if (HasValues)
+          {
+            RandomAccessIterator3 temp_vals1 = vals1;
+            temp_vals1 += j;
+
+            RandomAccessIterator4 temp_vals2 = vals2;
+            temp_vals2 += position;
+
+            // vals2[position] = vals1[j]
+            *temp_vals2 = *temp_vals1;
+          }
+        }
+      }
+        
+      flip = (flip) ? false : true;
+    }
+  }
+ 
+  // ensure final values are in (keys1,vals1)
+  if (flip)
+  {
+    thrust::copy(keys2, keys2 + N, keys1);
+    if (HasValues)
+      thrust::copy(vals2, vals2 + N, vals1);
+  }
+}
+
+
+// Select best radix sort parameters based on sizeof(T) and input size
+// These particular values were determined through empirical testing on a Core i7 950 CPU
+template <size_t KeySize>
+struct radix_sort_dispatcher
+{
+};
+
+template <>
+struct radix_sort_dispatcher<1>
+{
+  template <typename RandomAccessIterator1, typename RandomAccessIterator2>
+  void operator()(RandomAccessIterator1 keys1, RandomAccessIterator2 keys2, const size_t N)
+  {
+    detail::radix_sort<8,false>(keys1, keys2, static_cast<int *>(0), static_cast<int *>(0), N);
+  }
+  template <typename RandomAccessIterator1, typename RandomAccessIterator2, typename RandomAccessIterator3, typename RandomAccessIterator4>
+  void operator()(RandomAccessIterator1 keys1, RandomAccessIterator2 keys2, RandomAccessIterator3 vals1, RandomAccessIterator4 vals2, const size_t N)
+  {
+    detail::radix_sort<8,true>(keys1, keys2, vals1, vals2, N);
+  }
+};
+
+template <>
+struct radix_sort_dispatcher<2>
+{
+  template <typename RandomAccessIterator1, typename RandomAccessIterator2>
+  void operator()(RandomAccessIterator1 keys1, RandomAccessIterator2 keys2, const size_t N)
+  {
+    if (N < (1 << 16))
+      detail::radix_sort<8,false>(keys1, keys2, static_cast<int *>(0), static_cast<int *>(0), N);
+    else
+      detail::radix_sort<16,false>(keys1, keys2, static_cast<int *>(0), static_cast<int *>(0), N);
+  }
+  template <typename RandomAccessIterator1, typename RandomAccessIterator2, typename RandomAccessIterator3, typename RandomAccessIterator4>
+  void operator()(RandomAccessIterator1 keys1, RandomAccessIterator2 keys2, RandomAccessIterator3 vals1, RandomAccessIterator4 vals2, const size_t N)
+  {
+    if (N < (1 << 15))
+      detail::radix_sort<8,true>(keys1, keys2, vals1, vals2, N);
+    else
+      detail::radix_sort<16,true>(keys1, keys2, vals1, vals2, N);
+  }
+};
+
+template <>
+struct radix_sort_dispatcher<4>
+{
+  template <typename RandomAccessIterator1, typename RandomAccessIterator2>
+  void operator()(RandomAccessIterator1 keys1, RandomAccessIterator2 keys2, const size_t N)
+  {
+    if (N < (1 << 22))
+      detail::radix_sort<8,false>(keys1, keys2, static_cast<int *>(0), static_cast<int *>(0), N);
+    else
+      detail::radix_sort<4,false>(keys1, keys2, static_cast<int *>(0), static_cast<int *>(0), N);
+  }
+  template <typename RandomAccessIterator1, typename RandomAccessIterator2, typename RandomAccessIterator3, typename RandomAccessIterator4>
+  void operator()(RandomAccessIterator1 keys1, RandomAccessIterator2 keys2, RandomAccessIterator3 vals1, RandomAccessIterator4 vals2, const size_t N)
+  {
+    if (N < (1 << 22))
+      detail::radix_sort<8,true>(keys1, keys2, vals1, vals2, N);
+    else
+      detail::radix_sort<3,true>(keys1, keys2, vals1, vals2, N);
+  }
+};
+
+template <>
+struct radix_sort_dispatcher<8>
+{
+  template <typename RandomAccessIterator1, typename RandomAccessIterator2>
+  void operator()(RandomAccessIterator1 keys1, RandomAccessIterator2 keys2, const size_t N)
+  {
+    if (N < (1 << 21))
+      detail::radix_sort<8,false>(keys1, keys2, static_cast<int *>(0), static_cast<int *>(0), N);
+    else
+      detail::radix_sort<4,false>(keys1, keys2, static_cast<int *>(0), static_cast<int *>(0), N);
+  }
+  template <typename RandomAccessIterator1, typename RandomAccessIterator2, typename RandomAccessIterator3, typename RandomAccessIterator4>
+  void operator()(RandomAccessIterator1 keys1, RandomAccessIterator2 keys2, RandomAccessIterator3 vals1, RandomAccessIterator4 vals2, const size_t N)
+  {
+    if (N < (1 << 21))
+      detail::radix_sort<8,true>(keys1, keys2, vals1, vals2, N);
+    else
+      detail::radix_sort<3,true>(keys1, keys2, vals1, vals2, N);
+  }
+};
+
+template <typename RandomAccessIterator1,
+          typename RandomAccessIterator2>
+void radix_sort(RandomAccessIterator1 keys1,
+                RandomAccessIterator2 keys2,
+                const size_t N)
+{
+  typedef typename thrust::iterator_value<RandomAccessIterator1>::type KeyType;
+  radix_sort_dispatcher<sizeof(KeyType)>()(keys1, keys2, N);
+}
+
+template <typename RandomAccessIterator1,
+          typename RandomAccessIterator2,
+          typename RandomAccessIterator3,
+          typename RandomAccessIterator4>
+void radix_sort(RandomAccessIterator1 keys1,
+                RandomAccessIterator2 keys2,
+                RandomAccessIterator3 vals1,
+                RandomAccessIterator4 vals2,
+                const size_t N)
+{
+  typedef typename thrust::iterator_value<RandomAccessIterator1>::type KeyType;
+  radix_sort_dispatcher<sizeof(KeyType)>()(keys1, keys2, vals1, vals2, N);
+}
+
+} // namespace detail
+
+//////////////
+// Key Sort //
+//////////////
+
+template <typename RandomAccessIterator>
+void stable_radix_sort(RandomAccessIterator first,
+                       RandomAccessIterator last)
+{
+  typedef typename thrust::iterator_system<RandomAccessIterator>::type ExecutionPolicy;
+  typedef typename thrust::iterator_value<RandomAccessIterator>::type KeyType;
+
+  size_t N = last - first;
+  
+  // XXX assumes ExecutionPolicy is default constructible
+  // XXX consider how to get stateful systems into this function
+  ExecutionPolicy exec;
+  thrust::detail::temporary_array<KeyType, ExecutionPolicy> temp(exec, N);
+  
+  detail::radix_sort(first, temp.begin(), N);
+}
+
+
+////////////////////
+// Key-Value Sort //
+////////////////////
+
+template <typename RandomAccessIterator1,
+          typename RandomAccessIterator2>
+void stable_radix_sort_by_key(RandomAccessIterator1 first1,
+                              RandomAccessIterator1 last1,
+                              RandomAccessIterator2 first2)
+{
+  // XXX the type of exec should be
+  //     typedef decltype(select_system(first1,last1,first2)) system;
+  typedef typename thrust::iterator_system<RandomAccessIterator1>::type ExecutionPolicy;
+  typedef typename thrust::iterator_value<RandomAccessIterator1>::type KeyType;
+  typedef typename thrust::iterator_value<RandomAccessIterator2>::type ValueType;
+
+  size_t N = last1 - first1;
+  
+  // XXX assumes ExecutionPolicy is default constructible
+  // XXX consider how to get stateful systems into this function
+  ExecutionPolicy exec;
+  thrust::detail::temporary_array<KeyType, ExecutionPolicy>   temp1(exec, N);
+  thrust::detail::temporary_array<ValueType, ExecutionPolicy> temp2(exec, N);
+
+  detail::radix_sort(first1, temp1.begin(), first2, temp2.begin(), N);
+}
+
+} // end namespace scalar
+} // end namespace internal
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/detail/internal/scalar/trivial_copy.h b/compat/thrust/system/detail/internal/scalar/trivial_copy.h
new file mode 100644
index 0000000000..8f008b54e4
--- /dev/null
+++ b/compat/thrust/system/detail/internal/scalar/trivial_copy.h
@@ -0,0 +1,51 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+/*! \file trivial_copy.h
+ *  \brief Sequential copy algorithms for plain-old-data.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <cstring>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace internal
+{
+namespace scalar
+{
+
+template<typename T>
+  T *trivial_copy_n(const T *first,
+                    std::ptrdiff_t n,
+                    T *result)
+{
+  std::memmove(result, first, n * sizeof(T));
+  return result + n;
+} // end trivial_copy_n()
+
+} // end namespace scalar
+} // end namespace internal
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/detail/internal/scalar/unique.h b/compat/thrust/system/detail/internal/scalar/unique.h
new file mode 100644
index 0000000000..cfc60c948a
--- /dev/null
+++ b/compat/thrust/system/detail/internal/scalar/unique.h
@@ -0,0 +1,90 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file unique.h
+ *  \brief Sequential implementations of unique algorithms.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/pair.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace internal
+{
+namespace scalar
+{
+
+template<typename InputIterator,
+         typename OutputIterator,
+         typename BinaryPredicate>
+  OutputIterator unique_copy(InputIterator first,
+                             InputIterator last,
+                             OutputIterator output,
+                             BinaryPredicate binary_pred)
+{
+  typedef typename thrust::iterator_traits<InputIterator>::value_type T;
+
+  if(first != last)
+  {
+    T prev = *first;
+
+    for(++first; first != last; ++first)
+    {
+      T temp = *first;
+
+      if (!binary_pred(prev, temp))
+      {
+        *output = prev;
+
+        ++output;
+
+        prev = temp;
+      }
+    }
+
+    *output = prev;
+    ++output;
+  }
+
+  return output;
+} // end unique_copy()
+
+
+template<typename ForwardIterator,
+         typename BinaryPredicate>
+  ForwardIterator unique(ForwardIterator first,
+                         ForwardIterator last,
+                         BinaryPredicate binary_pred)
+{
+  // unique_copy() permits in-situ operation
+  return thrust::system::detail::internal::scalar::unique_copy(first, last, first, binary_pred);
+} // end unique()
+
+} // end namespace scalar
+} // end namespace internal
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/detail/internal/scalar/unique_by_key.h b/compat/thrust/system/detail/internal/scalar/unique_by_key.h
new file mode 100644
index 0000000000..b0be2663e7
--- /dev/null
+++ b/compat/thrust/system/detail/internal/scalar/unique_by_key.h
@@ -0,0 +1,109 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file unique_by_key.h
+ *  \brief Sequential implementations of unique_by_key algorithms.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/pair.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace detail
+{
+namespace internal
+{
+namespace scalar
+{
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename BinaryPredicate>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    unique_by_key_copy(InputIterator1 keys_first, 
+                       InputIterator1 keys_last,
+                       InputIterator2 values_first,
+                       OutputIterator1 keys_output,
+                       OutputIterator2 values_output,
+                       BinaryPredicate binary_pred)
+{
+  typedef typename thrust::iterator_traits<InputIterator1>::value_type  InputKeyType;
+  typedef typename thrust::iterator_traits<OutputIterator2>::value_type OutputValueType;
+
+  if(keys_first != keys_last)
+  {
+    InputKeyType    temp_key   = *keys_first;
+    OutputValueType temp_value = *values_first;
+
+    for(++keys_first, ++values_first;
+        keys_first != keys_last;
+        ++keys_first, ++values_first)
+    {
+      InputKeyType    key   = *keys_first;
+      OutputValueType value = *values_first;
+
+      if(!binary_pred(temp_key, key))
+      {
+        *keys_output   = temp_key;
+        *values_output = temp_value;
+
+        ++keys_output;
+        ++values_output;
+
+        temp_key   = key;
+        temp_value = value;
+      }
+    }
+
+    *keys_output   = temp_key;
+    *values_output = temp_value;
+
+    ++keys_output;
+    ++values_output;
+  }
+
+  return thrust::make_pair(keys_output, values_output);
+} // end unique_by_key_copy()
+
+
+template<typename ForwardIterator1,
+         typename ForwardIterator2,
+         typename BinaryPredicate>
+  thrust::pair<ForwardIterator1,ForwardIterator2>
+    unique_by_key(ForwardIterator1 keys_first, 
+                  ForwardIterator1 keys_last,
+                  ForwardIterator2 values_first,
+                  BinaryPredicate binary_pred)
+{
+  // unique_by_key_copy() permits in-situ operation
+  return thrust::system::detail::internal::scalar::unique_by_key_copy(keys_first, keys_last, values_first, keys_first, values_first, binary_pred);
+} // end unique_by_key()
+
+} // end namespace scalar
+} // end namespace internal
+} // end namespace detail
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/detail/system_error.inl b/compat/thrust/system/detail/system_error.inl
new file mode 100644
index 0000000000..74909bee8a
--- /dev/null
+++ b/compat/thrust/system/detail/system_error.inl
@@ -0,0 +1,111 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+#pragma once
+
+#include <thrust/system/system_error.h>
+
+namespace thrust
+{
+
+namespace system
+{
+
+
+system_error
+  ::system_error(error_code ec, const std::string &what_arg)
+    : std::runtime_error(what_arg), m_error_code(ec)
+{
+
+} // end system_error::system_error()
+
+
+system_error
+  ::system_error(error_code ec, const char *what_arg)
+    : std::runtime_error(what_arg), m_error_code(ec)
+{
+  ;
+} // end system_error::system_error()
+
+
+system_error
+  ::system_error(error_code ec)
+    : std::runtime_error(""), m_error_code(ec)
+{
+  ;
+} // end system_error::system_error()
+
+
+system_error
+  ::system_error(int ev, const error_category &ecat, const std::string &what_arg)
+    : std::runtime_error(what_arg), m_error_code(ev,ecat)
+{
+  ;
+} // end system_error::system_error()
+
+
+system_error
+  ::system_error(int ev, const error_category &ecat, const char *what_arg)
+    : std::runtime_error(what_arg), m_error_code(ev,ecat)
+{
+  ;
+} // end system_error::system_error()
+
+
+system_error
+  ::system_error(int ev, const error_category &ecat)
+    : std::runtime_error(""), m_error_code(ev,ecat)
+{
+  ;
+} // end system_error::system_error()
+
+
+const error_code &system_error
+  ::code(void) const throw()
+{
+  return m_error_code;
+} // end system_error::code()
+
+
+const char *system_error
+  ::what(void) const throw()
+{
+  if(m_what.empty())
+  {
+    try
+    {
+      m_what = this->std::runtime_error::what();
+      if(m_error_code)
+      {
+        if(!m_what.empty()) m_what += ": ";
+        m_what += m_error_code.message();
+      }
+    }
+    catch(...)
+    {
+      return std::runtime_error::what();
+    }
+  }
+
+  return m_what.c_str();
+} // end system_error::what()
+
+
+} // end system
+
+} // end thrust
+
diff --git a/compat/thrust/system/error_code.h b/compat/thrust/system/error_code.h
new file mode 100644
index 0000000000..2b6582c937
--- /dev/null
+++ b/compat/thrust/system/error_code.h
@@ -0,0 +1,521 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file error_code.h
+ *  \brief An object used to hold error values, such as those originating from the
+ *         operating system or other low-level application program interfaces.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/type_traits.h>
+#include <thrust/system/detail/errno.h>
+#include <iostream>
+
+namespace thrust
+{
+
+namespace system
+{
+
+
+/*! \addtogroup system_diagnostics
+ *  \{
+ */
+
+class error_condition;
+class error_code;
+
+/*! A metafunction returning whether or not the parameter is an \p error_code enum.
+ */
+template<typename T> struct is_error_code_enum : public thrust::detail::false_type {};
+
+/*! A metafunction returning whether or not the parameter is an \p error_condition enum.
+ */
+template<typename T> struct is_error_condition_enum : public thrust::detail::false_type {};
+
+
+// XXX N3092 prefers enum class errc { ... }
+namespace errc
+{
+
+enum errc_t
+{
+  address_family_not_supported       = detail::eafnosupport,
+  address_in_use                     = detail::eaddrinuse,
+  address_not_available              = detail::eaddrnotavail,
+  already_connected                  = detail::eisconn,
+  argument_list_too_long             = detail::e2big,
+  argument_out_of_domain             = detail::edom,
+  bad_address                        = detail::efault,
+  bad_file_descriptor                = detail::ebadf,
+  bad_message                        = detail::ebadmsg,
+  broken_pipe                        = detail::epipe,
+  connection_aborted                 = detail::econnaborted,
+  connection_already_in_progress     = detail::ealready,
+  connection_refused                 = detail::econnrefused,
+  connection_reset                   = detail::econnreset,
+  cross_device_link                  = detail::exdev,
+  destination_address_required       = detail::edestaddrreq,
+  device_or_resource_busy            = detail::ebusy,
+  directory_not_empty                = detail::enotempty,
+  executable_format_error            = detail::enoexec,
+  file_exists                        = detail::eexist,
+  file_too_large                     = detail::efbig,
+  filename_too_long                  = detail::enametoolong,
+  function_not_supported             = detail::enosys,
+  host_unreachable                   = detail::ehostunreach,
+  identifier_removed                 = detail::eidrm,
+  illegal_byte_sequence              = detail::eilseq,
+  inappropriate_io_control_operation = detail::enotty,
+  interrupted                        = detail::eintr,
+  invalid_argument                   = detail::einval,
+  invalid_seek                       = detail::espipe,
+  io_error                           = detail::eio,
+  is_a_directory                     = detail::eisdir,
+  message_size                       = detail::emsgsize,
+  network_down                       = detail::enetdown,
+  network_reset                      = detail::enetreset,
+  network_unreachable                = detail::enetunreach,
+  no_buffer_space                    = detail::enobufs,
+  no_child_process                   = detail::echild,
+  no_link                            = detail::enolink,
+  no_lock_available                  = detail::enolck,
+  no_message_available               = detail::enodata,
+  no_message                         = detail::enomsg,
+  no_protocol_option                 = detail::enoprotoopt,
+  no_space_on_device                 = detail::enospc,
+  no_stream_resources                = detail::enosr,
+  no_such_device_or_address          = detail::enxio,
+  no_such_device                     = detail::enodev,
+  no_such_file_or_directory          = detail::enoent,
+  no_such_process                    = detail::esrch,
+  not_a_directory                    = detail::enotdir,
+  not_a_socket                       = detail::enotsock,
+  not_a_stream                       = detail::enostr,
+  not_connected                      = detail::enotconn,
+  not_enough_memory                  = detail::enomem,
+  not_supported                      = detail::enotsup,
+  operation_canceled                 = detail::ecanceled,
+  operation_in_progress              = detail::einprogress,
+  operation_not_permitted            = detail::eperm,
+  operation_not_supported            = detail::eopnotsupp,
+  operation_would_block              = detail::ewouldblock,
+  owner_dead                         = detail::eownerdead,
+  permission_denied                  = detail::eacces,
+  protocol_error                     = detail::eproto,
+  protocol_not_supported             = detail::eprotonosupport,
+  read_only_file_system              = detail::erofs,
+  resource_deadlock_would_occur      = detail::edeadlk,
+  resource_unavailable_try_again     = detail::eagain,
+  result_out_of_range                = detail::erange,
+  state_not_recoverable              = detail::enotrecoverable,
+  stream_timeout                     = detail::etime,
+  text_file_busy                     = detail::etxtbsy,
+  timed_out                          = detail::etimedout,
+  too_many_files_open_in_system      = detail::enfile,
+  too_many_files_open                = detail::emfile,
+  too_many_links                     = detail::emlink,
+  too_many_symbolic_link_levels      = detail::eloop,
+  value_too_large                    = detail::eoverflow,
+  wrong_protocol_type                = detail::eprototype
+}; // end errc_t
+
+} // end namespace errc
+
+
+/*! Specialization of \p is_error_condition_enum for \p errc::errc_t
+ */
+template<> struct is_error_condition_enum<errc::errc_t> : public thrust::detail::true_type {};
+
+
+// [19.5.1.1] class error_category
+
+/*! \brief The class \p error_category serves as a base class for types used to identify the
+ *         source and encoding of a particular category of error code. Classes may be derived
+ *         from \p error_category to support categories of errors in addition to those defined
+ *         in the C++ International Standard.
+ */
+class error_category
+{
+  public:
+    /*! Destructor does nothing.
+     */
+    inline virtual ~error_category(void);
+
+    // XXX enable upon c++0x
+    // error_category(const error_category &) = delete;
+    // error_category &operator=(const error_category &) = delete;
+
+    /*! \return A string naming the error category.
+     */
+    inline virtual const char *name(void) const = 0;
+
+    /*! \return \p error_condition(ev, *this).
+     */
+    inline virtual error_condition default_error_condition(int ev) const;
+
+    /*! \return <tt>default_error_condition(code) == condition</tt>
+     */
+    inline virtual bool equivalent(int code, const error_condition &condition) const;
+
+    /*! \return <tt>*this == code.category() && code.value() == condition</tt>
+     */
+    inline virtual bool equivalent(const error_code &code, int condition) const;
+
+    /*! \return A string that describes the error condition denoted by \p ev.
+     */
+    virtual std::string message(int ev) const = 0;
+
+    /*! \return <tt>*this == &rhs</tt>
+     */
+    inline bool operator==(const error_category &rhs) const;
+
+    /*! \return <tt>!(*this == rhs)</tt>
+     */
+    inline bool operator!=(const error_category &rhs) const;
+
+    /*! \return <tt>less<const error_category*>()(this, &rhs)</tt>
+     *  \note \c less provides a total ordering for pointers.
+     */
+    inline bool operator<(const error_category &rhs) const;
+}; // end error_category
+
+
+// [19.5.1.5] error_category objects
+
+
+/*! \return A reference to an object of a type derived from class \p error_category.
+ *  \note The object's \p default_error_condition and \p equivalent virtual functions
+ *        shall behave as specified for the class \p error_category. The object's
+ *        \p name virtual function shall return a pointer to the string <tt>"generic"</tt>.
+ */
+inline const error_category &generic_category(void);
+
+
+/*! \return A reference to an object of a type derived from class \p error_category.
+ *  \note The object's \p equivalent virtual functions shall behave as specified for
+ *        class \p error_category. The object's \p name virtual function shall return
+ *        a pointer to the string <tt>"system"</tt>. The object's \p default_error_condition
+ *        virtual function shall behave as follows:
+ *
+ *        If the argument <tt>ev</tt> corresponds to a POSIX <tt>errno</tt> value
+ *        \c posv, the function shall return <tt>error_condition(ev,generic_category())</tt>.
+ *        Otherwise, the function shall return <tt>error_condition(ev,system_category())</tt>.
+ *        What constitutes correspondence for any given operating system is unspecified.
+ */
+inline const error_category &system_category(void);
+
+
+// [19.5.2] Class error_code
+
+
+/*! \brief The class \p error_code describes an object used to hold error code values, such as
+ *         those originating from the operating system or other low-level application program
+ *         interfaces.
+ */
+class error_code
+{
+  public:
+    // [19.5.2.2] constructors:
+
+    /*! Effects: Constructs an object of type \p error_code.
+     *  \post <tt>value() == 0</tt> and <tt>category() == &system_category()</tt>.
+     */
+    inline error_code(void);
+
+    /*! Effects: Constructs an object of type \p error_code.
+     *  \post <tt>value() == val</tt> and <tt>category() == &cat</tt>.
+     */
+    inline error_code(int val, const error_category &cat);
+
+    /*! Effects: Constructs an object of type \p error_code.
+     *  \post <tt>*this == make_error_code(e)</tt>.
+     */
+    template <typename ErrorCodeEnum>
+      error_code(ErrorCodeEnum e
+// XXX WAR msvc's problem with enable_if
+#if THRUST_HOST_COMPILER != THRUST_HOST_COMPILER_MSVC
+        , typename thrust::detail::enable_if<is_error_code_enum<ErrorCodeEnum>::value>::type * = 0
+#endif // THRUST_HOST_COMPILER_MSVC
+        );
+
+    // [19.5.2.3] modifiers:
+
+    /*! \post <tt>value() == val</tt> and <tt>category() == &cat</tt>.
+     */
+    inline void assign(int val, const error_category &cat);
+
+    /*! \post <tt>*this == make_error_code(e)</tt>.
+     */
+    template <typename ErrorCodeEnum>
+// XXX WAR msvc's problem with enable_if
+#if THRUST_HOST_COMPILER != THRUST_HOST_COMPILER_MSVC
+      typename thrust::detail::enable_if<is_error_code_enum<ErrorCodeEnum>::value, error_code>::type &
+#else
+      error_code &
+#endif // THRUST_HOST_COMPILER_MSVC
+        operator=(ErrorCodeEnum e);
+
+    /*! \post <tt>value() == 0</tt> and <tt>category() == system_category()</tt>.
+     */
+    inline void clear(void);
+
+    // [19.5.2.4] observers:
+
+    /*! \return An integral value of this \p error_code object.
+     */
+    inline int value(void) const;
+
+    /*! \return An \p error_category describing the category of this \p error_code object.
+     */
+    inline const error_category &category(void) const;
+
+    /*! \return <tt>category().default_error_condition()</tt>.
+     */
+    inline error_condition default_error_condition(void) const;
+
+    /*! \return <tt>category().message(value())</tt>.
+     */
+    inline std::string message(void) const;
+
+    // XXX replace the below upon c++0x
+    // inline explicit operator bool (void) const;
+
+    /*! \return <tt>value() != 0</tt>.
+     */
+    inline operator bool (void) const;
+
+    /*! \cond
+     */
+  private:
+    int m_val;
+    const error_category *m_cat;
+    /*! \endcond
+     */
+}; // end error_code
+
+
+// [19.5.2.5] Class error_code non-member functions
+
+
+// XXX replace errc::errc_t with errc upon c++0x
+/*! \return <tt>error_code(static_cast<int>(e), generic_category())</tt>
+ */
+inline error_code make_error_code(errc::errc_t e);
+
+
+/*! \return <tt>lhs.category() < rhs.category() || lhs.category() == rhs.category() && lhs.value() < rhs.value()</tt>.
+ */
+inline bool operator<(const error_code &lhs, const error_code &rhs);
+
+
+/*! Effects: <tt>os << ec.category().name() << ':' << ec.value()</tt>.
+ */
+template <typename charT, typename traits>
+  std::basic_ostream<charT,traits>&
+    operator<<(std::basic_ostream<charT,traits>& os, const error_code &ec);
+
+
+// [19.5.3] class error_condition
+
+
+/*! \brief The class \p error_condition describes an object used to hold values identifying
+ *  error conditions.
+ *
+ *  \note \p error_condition values are portable abstractions, while \p error_code values
+ *        are implementation specific.
+ */
+class error_condition
+{
+  public:
+    // [19.5.3.2] constructors
+
+    /*! Constructs an object of type \p error_condition.
+     *  \post <tt>value() == 0</tt>.
+     *  \post <tt>category() == generic_category()</tt>.
+     */
+    inline error_condition(void);
+
+    /*! Constructs an object of type \p error_condition.
+     *  \post <tt>value() == val</tt>.
+     *  \post <tt>category() == cat</tt>.
+     */
+    inline error_condition(int val, const error_category &cat);
+
+    /*! Constructs an object of type \p error_condition.
+     *  \post <tt>*this == make_error_condition(e)</tt>.
+     *  \note This constructor shall not participate in overload resolution unless
+     *        <tt>is_error_condition_enum<ErrorConditionEnum>::value</tt> is <tt>true</tt>.
+     */
+    template<typename ErrorConditionEnum>
+      error_condition(ErrorConditionEnum e
+// XXX WAR msvc's problem with enable_if
+#if THRUST_HOST_COMPILER != THRUST_HOST_COMPILER_MSVC
+        , typename thrust::detail::enable_if<is_error_condition_enum<ErrorConditionEnum>::value>::type * = 0
+#endif // THRUST_HOST_COMPILER != THRUST_HOST_COMPILER_MSVC
+                     );
+
+    // [19.5.3.3] modifiers
+
+    /*! Assigns to this \p error_code object from an error value and an \p error_category.
+     *  \param val The new value to return from <tt>value()</tt>.
+     *  \param cat The new \p error_category to return from <tt>category()</tt>.
+     *  \post <tt>value() == val</tt>.
+     *  \post <tt>category() == cat</tt>.
+     */
+    inline void assign(int val, const error_category &cat);
+
+    /*! Assigns to this \p error_code object from an error condition enumeration.
+     *  \return *this
+     *  \post <tt>*this == make_error_condition(e)</tt>.
+     *  \note This operator shall not participate in overload resolution unless
+     *        <tt>is_error_condition_enum<ErrorConditionEnum>::value</tt> is <tt>true</tt>.
+     */
+    template<typename ErrorConditionEnum>
+// XXX WAR msvc's problem with enable_if
+#if THRUST_HOST_COMPILER != THRUST_HOST_COMPILER_MSVC
+      typename thrust::detail::enable_if<is_error_condition_enum<ErrorConditionEnum>::value, error_condition>::type &
+#else
+      error_condition &
+#endif // THRUST_HOST_COMPILER != THRUST_HOST_COMPILER_MSVC
+        operator=(ErrorConditionEnum e);
+
+    /*! Clears this \p error_code object.
+     *  \post <tt>value == 0</tt>
+     *  \post <tt>category() == generic_category()</tt>.
+     */
+    inline void clear(void);
+
+    // [19.5.3.4] observers
+
+    /*! \return The value encoded by this \p error_condition.
+     */
+    inline int value(void) const;
+
+    /*! \return A <tt>const</tt> reference to the \p error_category encoded by this \p error_condition.
+     */
+    inline const error_category &category(void) const;
+
+    /*! \return <tt>category().message(value())</tt>.
+     */
+    inline std::string message(void) const;
+
+    // XXX replace below with this upon c++0x
+    //explicit operator bool (void) const;
+    
+    /*! \return <tt>value() != 0</tt>.
+     */
+    inline operator bool (void) const;
+
+    /*! \cond
+     */
+
+  private:
+    int m_val;
+    const error_category *m_cat;
+
+    /*! \endcond
+     */
+}; // end error_condition
+
+
+
+// [19.5.3.5] Class error_condition non-member functions
+
+// XXX replace errc::errc_t with errc upon c++0x
+/*! \return <tt>error_condition(static_cast<int>(e), generic_category())</tt>.
+ */
+inline error_condition make_error_condition(errc::errc_t e);
+
+
+/*! \return <tt>lhs.category() < rhs.category() || lhs.category() == rhs.category() && lhs.value() < rhs.value()</tt>.
+ */
+inline bool operator<(const error_condition &lhs, const error_condition &rhs);
+
+
+// [19.5.4] Comparison operators
+
+
+/*! \return <tt>lhs.category() == rhs.category() && lhs.value() == rhs.value()</tt>.
+ */
+inline bool operator==(const error_code &lhs, const error_code &rhs);
+
+
+/*! \return <tt>lhs.category().equivalent(lhs.value(), rhs) || rhs.category().equivalent(lhs,rhs.value())</tt>.
+ */
+inline bool operator==(const error_code &lhs, const error_condition &rhs);
+
+
+/*! \return <tt>rhs.category().equivalent(lhs.value(), lhs) || lhs.category().equivalent(rhs, lhs.value())</tt>.
+ */
+inline bool operator==(const error_condition &lhs, const error_code &rhs);
+
+
+/*! \return <tt>lhs.category() == rhs.category() && lhs.value() == rhs.value()</tt>
+ */
+inline bool operator==(const error_condition &lhs, const error_condition &rhs);
+
+
+/*! \return <tt>!(lhs == rhs)</tt>
+ */
+inline bool operator!=(const error_code &lhs, const error_code &rhs);
+
+
+/*! \return <tt>!(lhs == rhs)</tt>
+ */
+inline bool operator!=(const error_code &lhs, const error_condition &rhs);
+
+
+/*! \return <tt>!(lhs == rhs)</tt>
+ */
+inline bool operator!=(const error_condition &lhs, const error_code &rhs);
+
+
+/*! \return <tt>!(lhs == rhs)</tt>
+ */
+inline bool operator!=(const error_condition &lhs, const error_condition &rhs);
+
+/*! \} // end system_diagnostics
+ */
+
+
+} // end system
+
+
+// import names into thrust::
+using system::error_category;
+using system::error_code;
+using system::error_condition;
+using system::is_error_code_enum;
+using system::is_error_condition_enum;
+using system::make_error_code;
+using system::make_error_condition;
+
+// XXX replace with using system::errc upon c++0x
+namespace errc = system::errc;
+
+using system::generic_category;
+using system::system_category;
+
+} // end thrust
+
+#include <thrust/system/detail/error_category.inl>
+#include <thrust/system/detail/error_code.inl>
+#include <thrust/system/detail/error_condition.inl>
+
diff --git a/compat/thrust/system/omp/detail/adjacent_difference.h b/compat/thrust/system/omp/detail/adjacent_difference.h
new file mode 100644
index 0000000000..0bbc1884cc
--- /dev/null
+++ b/compat/thrust/system/omp/detail/adjacent_difference.h
@@ -0,0 +1,50 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/omp/detail/execution_policy.h>
+#include <thrust/system/detail/generic/adjacent_difference.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace omp
+{
+namespace detail
+{
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator,
+         typename BinaryFunction>
+  OutputIterator adjacent_difference(execution_policy<DerivedPolicy> &exec,
+                                     InputIterator first,
+                                     InputIterator last,
+                                     OutputIterator result,
+                                     BinaryFunction binary_op)
+{
+  // omp prefers generic::adjacent_difference to cpp::adjacent_difference
+  return thrust::system::detail::generic::adjacent_difference(exec, first, last, result, binary_op);
+} // end adjacent_difference()
+
+} // end detail
+} // end omp
+} // end system
+} // end thrust
+
diff --git a/compat/thrust/system/omp/detail/assign_value.h b/compat/thrust/system/omp/detail/assign_value.h
new file mode 100644
index 0000000000..eda3b977b8
--- /dev/null
+++ b/compat/thrust/system/omp/detail/assign_value.h
@@ -0,0 +1,23 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system inherits assign_value
+#include <thrust/system/cpp/detail/assign_value.h>
+
diff --git a/compat/thrust/system/omp/detail/binary_search.h b/compat/thrust/system/omp/detail/binary_search.h
new file mode 100644
index 0000000000..254e6fd445
--- /dev/null
+++ b/compat/thrust/system/omp/detail/binary_search.h
@@ -0,0 +1,73 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/omp/detail/execution_policy.h>
+#include <thrust/system/detail/generic/binary_search.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace omp
+{
+namespace detail
+{
+
+
+template <typename DerivedPolicy, typename ForwardIterator, typename T, typename StrictWeakOrdering>
+ForwardIterator lower_bound(execution_policy<DerivedPolicy> &exec,
+                            ForwardIterator begin,
+                            ForwardIterator end,
+                            const T& value, 
+                            StrictWeakOrdering comp)
+{
+    // omp prefers generic::lower_bound to cpp::lower_bound
+    return thrust::system::detail::generic::lower_bound(exec, begin, end, value, comp);
+}
+
+
+template <typename DerivedPolicy, typename ForwardIterator, typename T, typename StrictWeakOrdering, typename Backend>
+ForwardIterator upper_bound(execution_policy<DerivedPolicy> &exec,
+                            ForwardIterator begin,
+                            ForwardIterator end,
+                            const T& value, 
+                            StrictWeakOrdering comp)
+{
+    // omp prefers generic::upper_bound to cpp::upper_bound
+    return thrust::system::detail::generic::upper_bound(exec, begin, end, value, comp);
+}
+
+
+template <typename DerivedPolicy, typename ForwardIterator, typename T, typename StrictWeakOrdering>
+bool binary_search(execution_policy<DerivedPolicy> &exec,
+                   ForwardIterator begin,
+                   ForwardIterator end,
+                   const T& value, 
+                   StrictWeakOrdering comp)
+{
+    // omp prefers generic::binary_search to cpp::binary_search
+    return thrust::system::detail::generic::binary_search(exec, begin, end, value, comp);
+}
+
+
+} // end detail
+} // end omp
+} // end system
+} // end thrust
+
diff --git a/compat/thrust/system/omp/detail/copy.h b/compat/thrust/system/omp/detail/copy.h
new file mode 100644
index 0000000000..b23ac18801
--- /dev/null
+++ b/compat/thrust/system/omp/detail/copy.h
@@ -0,0 +1,57 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/omp/detail/execution_policy.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace omp
+{
+namespace detail
+{
+
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator>
+OutputIterator copy(execution_policy<DerivedPolicy> &exec,
+                    InputIterator first,
+                    InputIterator last,
+                    OutputIterator result);
+
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename Size,
+         typename OutputIterator>
+OutputIterator copy_n(execution_policy<DerivedPolicy> &exec,
+                      InputIterator first,
+                      Size n,
+                      OutputIterator result);
+
+
+} // end namespace detail
+} // end namespace omp
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/omp/detail/copy.inl>
+
diff --git a/compat/thrust/system/omp/detail/copy.inl b/compat/thrust/system/omp/detail/copy.inl
new file mode 100644
index 0000000000..915ff92d70
--- /dev/null
+++ b/compat/thrust/system/omp/detail/copy.inl
@@ -0,0 +1,147 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/omp/detail/copy.h>
+#include <thrust/system/detail/generic/copy.h>
+#include <thrust/detail/type_traits/minimum_type.h>
+#include <thrust/system/cpp/detail/copy.h>
+#include <thrust/iterator/detail/retag.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace omp
+{
+namespace detail
+{
+namespace dispatch
+{
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator>
+  OutputIterator copy(execution_policy<DerivedPolicy> &exec,
+                      InputIterator first,
+                      InputIterator last,
+                      OutputIterator result,
+                      thrust::incrementable_traversal_tag)
+{
+  return thrust::system::cpp::detail::copy(exec, first, last, result);
+} // end copy()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator>
+  OutputIterator copy(execution_policy<DerivedPolicy> &exec,
+                      InputIterator first,
+                      InputIterator last,
+                      OutputIterator result,
+                      thrust::random_access_traversal_tag)
+{
+  // XXX WAR problems reconciling unrelated types such as omp & tbb
+  // reinterpret iterators as the policy we were passed
+  // this ensures that generic::copy's implementation, which eventually results in
+  // zip_iterator works correctly
+  thrust::detail::tagged_iterator<OutputIterator,DerivedPolicy> retagged_result(result);
+
+  return thrust::system::detail::generic::copy(exec, thrust::reinterpret_tag<DerivedPolicy>(first), thrust::reinterpret_tag<DerivedPolicy>(last), retagged_result).base();
+} // end copy()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename Size,
+         typename OutputIterator>
+  OutputIterator copy_n(execution_policy<DerivedPolicy> &exec,
+                        InputIterator first,
+                        Size n,
+                        OutputIterator result,
+                        thrust::incrementable_traversal_tag)
+{
+  return thrust::system::cpp::detail::copy_n(exec, first, n, result);
+} // end copy_n()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename Size,
+         typename OutputIterator>
+  OutputIterator copy_n(execution_policy<DerivedPolicy> &exec,
+                        InputIterator first,
+                        Size n,
+                        OutputIterator result,
+                        thrust::random_access_traversal_tag)
+{
+  // XXX WAR problems reconciling unrelated types such as omp & tbb
+  // reinterpret iterators as the policy we were passed
+  // this ensures that generic::copy's implementation, which eventually results in
+  // zip_iterator works correctly
+  thrust::detail::tagged_iterator<OutputIterator,DerivedPolicy> retagged_result(result);
+
+  return thrust::system::detail::generic::copy_n(exec, thrust::reinterpret_tag<DerivedPolicy>(first), n, retagged_result).base();
+} // end copy_n()
+
+} // end dispatch
+
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator>
+OutputIterator copy(execution_policy<DerivedPolicy> &exec,
+                    InputIterator first,
+                    InputIterator last,
+                    OutputIterator result)
+{
+  typedef typename thrust::iterator_traversal<InputIterator>::type  traversal1;
+  typedef typename thrust::iterator_traversal<OutputIterator>::type traversal2;
+  
+  typedef typename thrust::detail::minimum_type<traversal1,traversal2>::type traversal;
+
+  // dispatch on minimum traversal
+  return thrust::system::omp::detail::dispatch::copy(exec, first,last,result,traversal());
+} // end copy()
+
+
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename Size,
+         typename OutputIterator>
+OutputIterator copy_n(execution_policy<DerivedPolicy> &exec,
+                      InputIterator first,
+                      Size n,
+                      OutputIterator result)
+{
+  typedef typename thrust::iterator_traversal<InputIterator>::type  traversal1;
+  typedef typename thrust::iterator_traversal<OutputIterator>::type traversal2;
+  
+  typedef typename thrust::detail::minimum_type<traversal1,traversal2>::type traversal;
+
+  // dispatch on minimum traversal
+  return thrust::system::omp::detail::dispatch::copy_n(exec,first,n,result,traversal());
+} // end copy_n()
+
+
+} // end namespace detail
+} // end namespace omp
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/omp/detail/copy_if.h b/compat/thrust/system/omp/detail/copy_if.h
new file mode 100644
index 0000000000..46754a9f21
--- /dev/null
+++ b/compat/thrust/system/omp/detail/copy_if.h
@@ -0,0 +1,51 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/omp/detail/execution_policy.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace omp
+{
+namespace detail
+{
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename Predicate>
+  OutputIterator copy_if(execution_policy<DerivedPolicy> &exec,
+                         InputIterator1 first,
+                         InputIterator1 last,
+                         InputIterator2 stencil,
+                         OutputIterator result,
+                         Predicate pred);
+
+
+} // end detail
+} // end omp
+} // end system
+} // end thrust
+
+#include <thrust/system/omp/detail/copy_if.inl>
+
diff --git a/compat/thrust/system/omp/detail/copy_if.inl b/compat/thrust/system/omp/detail/copy_if.inl
new file mode 100644
index 0000000000..1af6a215a6
--- /dev/null
+++ b/compat/thrust/system/omp/detail/copy_if.inl
@@ -0,0 +1,54 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/omp/detail/copy_if.h>
+#include <thrust/system/detail/generic/copy_if.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace omp
+{
+namespace detail
+{
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename Predicate>
+  OutputIterator copy_if(execution_policy<DerivedPolicy> &exec,
+                         InputIterator1 first,
+                         InputIterator1 last,
+                         InputIterator2 stencil,
+                         OutputIterator result,
+                         Predicate pred)
+{
+  // omp prefers generic::copy_if to cpp::copy_if
+  return thrust::system::detail::generic::copy_if(exec, first, last, stencil, result, pred);
+} // end copy_if()
+
+
+} // end detail
+} // end omp
+} // end system
+} // end thrust
+
diff --git a/compat/thrust/system/omp/detail/count.h b/compat/thrust/system/omp/detail/count.h
new file mode 100644
index 0000000000..da31ee8700
--- /dev/null
+++ b/compat/thrust/system/omp/detail/count.h
@@ -0,0 +1,23 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system inherits count
+#include <thrust/system/cpp/detail/count.h>
+
diff --git a/compat/thrust/system/omp/detail/default_decomposition.h b/compat/thrust/system/omp/detail/default_decomposition.h
new file mode 100644
index 0000000000..f1904c29c6
--- /dev/null
+++ b/compat/thrust/system/omp/detail/default_decomposition.h
@@ -0,0 +1,45 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file default_decomposition.h
+ *  \brief Return a decomposition that is appropriate for the OpenMP backend.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/internal/decompose.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace omp
+{
+namespace detail
+{
+
+template <typename IndexType>
+thrust::system::detail::internal::uniform_decomposition<IndexType> default_decomposition(IndexType n);
+
+} // end namespace detail
+} // end namespace omp
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/omp/detail/default_decomposition.inl>
+
diff --git a/compat/thrust/system/omp/detail/default_decomposition.inl b/compat/thrust/system/omp/detail/default_decomposition.inl
new file mode 100644
index 0000000000..366b4f56c2
--- /dev/null
+++ b/compat/thrust/system/omp/detail/default_decomposition.inl
@@ -0,0 +1,56 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/system/omp/detail/default_decomposition.h>
+
+// don't attempt to #include this file without omp support
+#if (THRUST_DEVICE_COMPILER_IS_OMP_CAPABLE == THRUST_TRUE)
+#include <omp.h>
+#endif // omp support
+
+namespace thrust
+{
+namespace system
+{
+namespace omp
+{
+namespace detail
+{
+
+template <typename IndexType>
+thrust::system::detail::internal::uniform_decomposition<IndexType> default_decomposition(IndexType n)
+{
+  // we're attempting to launch an omp kernel, assert we're compiling with omp support
+  // ========================================================================
+  // X Note to the user: If you've found this line due to a compiler error, X
+  // X you need to OpenMP support in your compiler.                         X
+  // ========================================================================
+  THRUST_STATIC_ASSERT( (thrust::detail::depend_on_instantiation<IndexType,
+                        (THRUST_DEVICE_COMPILER_IS_OMP_CAPABLE == THRUST_TRUE)>::value) );
+
+#if (THRUST_DEVICE_COMPILER_IS_OMP_CAPABLE == THRUST_TRUE)
+  return thrust::system::detail::internal::uniform_decomposition<IndexType>(n, 1, omp_get_num_procs());
+#else
+  return thrust::system::detail::internal::uniform_decomposition<IndexType>(n, 1, 1);
+#endif
+}
+
+} // end namespace detail
+} // end namespace omp
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/omp/detail/equal.h b/compat/thrust/system/omp/detail/equal.h
new file mode 100644
index 0000000000..74e55183d9
--- /dev/null
+++ b/compat/thrust/system/omp/detail/equal.h
@@ -0,0 +1,23 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system inherits equal
+#include <thrust/system/cpp/detail/equal.h>
+
diff --git a/compat/thrust/system/omp/detail/execution_policy.h b/compat/thrust/system/omp/detail/execution_policy.h
new file mode 100644
index 0000000000..1b06224217
--- /dev/null
+++ b/compat/thrust/system/omp/detail/execution_policy.h
@@ -0,0 +1,110 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/cpp/detail/execution_policy.h>
+#include <thrust/system/tbb/detail/execution_policy.h>
+#include <thrust/iterator/detail/any_system_tag.h>
+#include <thrust/detail/type_traits.h>
+
+namespace thrust
+{
+namespace system
+{
+// put the canonical tag in the same ns as the backend's entry points
+namespace omp
+{
+namespace detail
+{
+
+// this awkward sequence of definitions arise
+// from the desire both for tag to derive
+// from execution_policy and for execution_policy
+// to convert to tag (when execution_policy is not
+// an ancestor of tag)
+
+// forward declaration of tag
+struct tag;
+
+// forward declaration of execution_policy
+template<typename> struct execution_policy;
+
+// specialize execution_policy for tag
+template<>
+  struct execution_policy<tag>
+    : thrust::system::cpp::detail::execution_policy<tag>
+{};
+
+// tag's definition comes before the
+// generic definition of execution_policy
+struct tag : execution_policy<tag> {};
+
+// allow conversion to tag when it is not a successor
+template<typename Derived>
+  struct execution_policy
+    : thrust::system::cpp::detail::execution_policy<Derived>
+{
+  // allow conversion to tag
+  inline operator tag () const
+  {
+    return tag();
+  }
+};
+
+
+// overloads of select_system
+
+// XXX select_system(tbb, omp) & select_system(omp, tbb) are ambiguous
+//     because both convert to cpp without these overloads, which we
+//     arbitrarily define in the omp backend
+
+template<typename System1, typename System2>
+inline __host__ __device__
+  System1 select_system(execution_policy<System1> s, thrust::system::tbb::detail::execution_policy<System2>)
+{
+  return thrust::detail::derived_cast(s);
+} // end select_system()
+
+
+template<typename System1, typename System2>
+inline __host__ __device__
+  System2 select_system(thrust::system::tbb::detail::execution_policy<System1>, execution_policy<System2> s)
+{
+  return thrust::detail::derived_cast(s);
+} // end select_system()
+
+
+} // end detail
+
+// alias execution_policy and tag here
+using thrust::system::omp::detail::execution_policy;
+using thrust::system::omp::detail::tag;
+
+} // end omp
+} // end system
+
+// alias items at top-level
+namespace omp
+{
+
+using thrust::system::omp::execution_policy;
+using thrust::system::omp::tag;
+
+} // end omp
+} // end thrust
+
diff --git a/compat/thrust/system/omp/detail/extrema.h b/compat/thrust/system/omp/detail/extrema.h
new file mode 100644
index 0000000000..fb96770b97
--- /dev/null
+++ b/compat/thrust/system/omp/detail/extrema.h
@@ -0,0 +1,67 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/omp/detail/execution_policy.h>
+#include <thrust/system/detail/generic/extrema.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace omp
+{
+namespace detail
+{
+
+template <typename DerivedPolicy, typename ForwardIterator, typename BinaryPredicate>
+ForwardIterator max_element(execution_policy<DerivedPolicy> &exec,
+                            ForwardIterator first, 
+                            ForwardIterator last,
+                            BinaryPredicate comp)
+{
+  // omp prefers generic::max_element to cpp::max_element
+  return thrust::system::detail::generic::max_element(exec, first, last, comp);
+} // end max_element()
+
+template <typename DerivedPolicy, typename ForwardIterator, typename BinaryPredicate>
+ForwardIterator min_element(execution_policy<DerivedPolicy> &exec,
+                            ForwardIterator first, 
+                            ForwardIterator last,
+                            BinaryPredicate comp)
+{
+  // omp prefers generic::min_element to cpp::min_element
+  return thrust::system::detail::generic::min_element(exec, first, last, comp);
+} // end min_element()
+
+template <typename DerivedPolicy, typename ForwardIterator, typename BinaryPredicate>
+thrust::pair<ForwardIterator,ForwardIterator> minmax_element(execution_policy<DerivedPolicy> &exec,
+                                                             ForwardIterator first, 
+                                                             ForwardIterator last,
+                                                             BinaryPredicate comp)
+{
+  // omp prefers generic::minmax_element to cpp::minmax_element
+  return thrust::system::detail::generic::minmax_element(exec, first, last, comp);
+} // end minmax_element()
+
+} // end detail
+} // end omp
+} // end system
+} // end thrust
+
+
diff --git a/compat/thrust/system/omp/detail/fill.h b/compat/thrust/system/omp/detail/fill.h
new file mode 100644
index 0000000000..5219e1c7c5
--- /dev/null
+++ b/compat/thrust/system/omp/detail/fill.h
@@ -0,0 +1,23 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system inherits fill
+#include <thrust/system/cpp/detail/fill.h>
+
diff --git a/compat/thrust/system/omp/detail/find.h b/compat/thrust/system/omp/detail/find.h
new file mode 100644
index 0000000000..a8dca5ad4d
--- /dev/null
+++ b/compat/thrust/system/omp/detail/find.h
@@ -0,0 +1,51 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file find.h
+ *  \brief OpenMP implementation of find_if. 
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/find.h>
+#include <thrust/system/omp/detail/execution_policy.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace omp
+{
+namespace detail
+{
+
+template <typename DerivedPolicy, typename InputIterator, typename Predicate>
+InputIterator find_if(execution_policy<DerivedPolicy> &exec,
+                      InputIterator first,
+                      InputIterator last,
+                      Predicate pred)
+{
+  // omp prefers generic::find_if to cpp::find_if
+  return thrust::system::detail::generic::find_if(exec, first, last, pred);
+}
+
+} // end namespace detail
+} // end namespace omp
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/omp/detail/for_each.h b/compat/thrust/system/omp/detail/for_each.h
new file mode 100644
index 0000000000..1030623ec0
--- /dev/null
+++ b/compat/thrust/system/omp/detail/for_each.h
@@ -0,0 +1,60 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file for_each.h
+ *  \brief Defines the interface for a function that executes a 
+ *  function or functional for each value in a given range.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/omp/detail/execution_policy.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace omp
+{
+namespace detail
+{
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator,
+         typename UnaryFunction>
+  RandomAccessIterator for_each(execution_policy<DerivedPolicy> &exec,
+                                RandomAccessIterator first,
+                                RandomAccessIterator last,
+                                UnaryFunction f);
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator,
+         typename Size,
+         typename UnaryFunction>
+  RandomAccessIterator for_each_n(execution_policy<DerivedPolicy> &exec,
+                                  RandomAccessIterator first,
+                                  Size n,
+                                  UnaryFunction f);
+
+} // end namespace detail
+} // end namespace omp
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/omp/detail/for_each.inl>
+
diff --git a/compat/thrust/system/omp/detail/for_each.inl b/compat/thrust/system/omp/detail/for_each.inl
new file mode 100644
index 0000000000..c6ab8277ba
--- /dev/null
+++ b/compat/thrust/system/omp/detail/for_each.inl
@@ -0,0 +1,97 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file for_each.inl
+ *  \brief Inline file for for_each.h.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/static_assert.h>
+#include <thrust/distance.h>
+#include <thrust/detail/function.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/distance.h>
+#include <thrust/for_each.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace omp
+{
+namespace detail
+{
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator,
+         typename Size,
+         typename UnaryFunction>
+RandomAccessIterator for_each_n(execution_policy<DerivedPolicy> &,
+                                RandomAccessIterator first,
+                                Size n,
+                                UnaryFunction f)
+{
+  // we're attempting to launch an omp kernel, assert we're compiling with omp support
+  // ========================================================================
+  // X Note to the user: If you've found this line due to a compiler error, X
+  // X you need to enable OpenMP support in your compiler.                  X
+  // ========================================================================
+  THRUST_STATIC_ASSERT( (thrust::detail::depend_on_instantiation<RandomAccessIterator,
+                        (THRUST_DEVICE_COMPILER_IS_OMP_CAPABLE == THRUST_TRUE)>::value) );
+
+  if (n <= 0) return first;  //empty range
+
+  // create a wrapped function for f
+  typedef typename thrust::iterator_reference<RandomAccessIterator>::type reference;
+  thrust::detail::host_function<UnaryFunction,void> wrapped_f(f);
+
+// do not attempt to compile the body of this function, which depends on #pragma omp,
+// without support from the compiler
+// XXX implement the body of this function in another file to eliminate this ugliness
+#if (THRUST_DEVICE_COMPILER_IS_OMP_CAPABLE == THRUST_TRUE)
+  // use a signed type for the iteration variable or suffer the consequences of warnings
+  typedef typename thrust::iterator_difference<RandomAccessIterator>::type DifferenceType;
+  DifferenceType signed_n = n;
+#pragma omp parallel for
+  for(DifferenceType i = 0;
+      i < signed_n;
+      ++i)
+  {
+    RandomAccessIterator temp = first + i;
+    wrapped_f(*temp);
+  }
+#endif // THRUST_DEVICE_COMPILER_IS_OMP_CAPABLE
+
+  return first + n;
+} // end for_each_n() 
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator,
+         typename UnaryFunction>
+  RandomAccessIterator for_each(execution_policy<DerivedPolicy> &s,
+                                RandomAccessIterator first,
+                                RandomAccessIterator last,
+                                UnaryFunction f)
+{
+  return omp::detail::for_each_n(s, first, thrust::distance(first,last), f);
+} // end for_each()
+
+} // end namespace detail
+} // end namespace omp
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/omp/detail/gather.h b/compat/thrust/system/omp/detail/gather.h
new file mode 100644
index 0000000000..dfb7d7fc2d
--- /dev/null
+++ b/compat/thrust/system/omp/detail/gather.h
@@ -0,0 +1,23 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system inherits gather
+#include <thrust/system/cpp/detail/gather.h>
+
diff --git a/compat/thrust/system/omp/detail/generate.h b/compat/thrust/system/omp/detail/generate.h
new file mode 100644
index 0000000000..0cb33b9336
--- /dev/null
+++ b/compat/thrust/system/omp/detail/generate.h
@@ -0,0 +1,23 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system inherits generate
+#include <thrust/system/cpp/detail/generate.h>
+
diff --git a/compat/thrust/system/omp/detail/get_value.h b/compat/thrust/system/omp/detail/get_value.h
new file mode 100644
index 0000000000..e376e65749
--- /dev/null
+++ b/compat/thrust/system/omp/detail/get_value.h
@@ -0,0 +1,23 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system inherits get_value
+#include <thrust/system/cpp/detail/get_value.h>
+
diff --git a/compat/thrust/system/omp/detail/inner_product.h b/compat/thrust/system/omp/detail/inner_product.h
new file mode 100644
index 0000000000..351421a577
--- /dev/null
+++ b/compat/thrust/system/omp/detail/inner_product.h
@@ -0,0 +1,23 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system inherits inner_product
+#include <thrust/system/cpp/detail/inner_product.h>
+
diff --git a/compat/thrust/system/omp/detail/iter_swap.h b/compat/thrust/system/omp/detail/iter_swap.h
new file mode 100644
index 0000000000..16176ec69b
--- /dev/null
+++ b/compat/thrust/system/omp/detail/iter_swap.h
@@ -0,0 +1,23 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system inherits iter_swap
+#include <thrust/system/cpp/detail/iter_swap.h>
+
diff --git a/compat/thrust/system/omp/detail/logical.h b/compat/thrust/system/omp/detail/logical.h
new file mode 100644
index 0000000000..b2a80de70f
--- /dev/null
+++ b/compat/thrust/system/omp/detail/logical.h
@@ -0,0 +1,23 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system inherits logical
+#include <thrust/system/cpp/detail/logical.h>
+
diff --git a/compat/thrust/system/omp/detail/malloc_and_free.h b/compat/thrust/system/omp/detail/malloc_and_free.h
new file mode 100644
index 0000000000..811a552a4f
--- /dev/null
+++ b/compat/thrust/system/omp/detail/malloc_and_free.h
@@ -0,0 +1,23 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system inherits malloc and free
+#include <thrust/system/cpp/detail/malloc_and_free.h>
+
diff --git a/compat/thrust/system/omp/detail/memory.inl b/compat/thrust/system/omp/detail/memory.inl
new file mode 100644
index 0000000000..7d53de60a1
--- /dev/null
+++ b/compat/thrust/system/omp/detail/memory.inl
@@ -0,0 +1,110 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/system/cpp/detail/execution_policy.h>
+#include <thrust/system/omp/memory.h>
+#include <thrust/system/cpp/memory.h>
+#include <limits>
+
+namespace thrust
+{
+namespace system
+{
+namespace omp
+{
+
+
+template<typename T>
+  template<typename OtherT>
+    reference<T> &
+      reference<T>
+        ::operator=(const reference<OtherT> &other)
+{
+  return super_t::operator=(other);
+} // end reference::operator=()
+
+template<typename T>
+  reference<T> &
+    reference<T>
+      ::operator=(const value_type &x)
+{
+  return super_t::operator=(x);
+} // end reference::operator=()
+
+template<typename T>
+__host__ __device__
+void swap(reference<T> a, reference<T> b)
+{
+  a.swap(b);
+} // end swap()
+
+namespace detail
+{
+
+// XXX circular #inclusion problems cause the compiler to believe that cpp::malloc
+//     is not defined
+//     WAR the problem by using adl to call cpp::malloc, which requires it to depend
+//     on a template parameter
+template<typename Tag>
+  pointer<void> malloc_workaround(Tag t, std::size_t n)
+{
+  return pointer<void>(malloc(t, n));
+} // end malloc_workaround()
+
+// XXX circular #inclusion problems cause the compiler to believe that cpp::free
+//     is not defined
+//     WAR the problem by using adl to call cpp::free, which requires it to depend
+//     on a template parameter
+template<typename Tag>
+  void free_workaround(Tag t, pointer<void> ptr)
+{
+  free(t, ptr.get());
+} // end free_workaround()
+
+} // end detail
+
+inline pointer<void> malloc(std::size_t n)
+{
+  // XXX this is how we'd like to implement this function,
+  //     if not for circular #inclusion problems:
+  //
+  // return pointer<void>(thrust::system::cpp::malloc(n))
+  //
+  return detail::malloc_workaround(cpp::tag(), n);
+} // end malloc()
+
+template<typename T>
+pointer<T> malloc(std::size_t n)
+{
+  pointer<void> raw_ptr = thrust::system::omp::malloc(sizeof(T) * n);
+  return pointer<T>(reinterpret_cast<T*>(raw_ptr.get()));
+} // end malloc()
+
+inline void free(pointer<void> ptr)
+{
+  // XXX this is how we'd like to implement this function,
+  //     if not for circular #inclusion problems:
+  //
+  // thrust::system::cpp::free(ptr)
+  //
+  detail::free_workaround(cpp::tag(), ptr);
+} // end free()
+
+} // end omp
+} // end system
+} // end thrust
+
diff --git a/compat/thrust/system/omp/detail/merge.h b/compat/thrust/system/omp/detail/merge.h
new file mode 100644
index 0000000000..a7047aa028
--- /dev/null
+++ b/compat/thrust/system/omp/detail/merge.h
@@ -0,0 +1,23 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system inherits merge
+#include <thrust/system/cpp/detail/merge.h>
+
diff --git a/compat/thrust/system/omp/detail/mismatch.h b/compat/thrust/system/omp/detail/mismatch.h
new file mode 100644
index 0000000000..03980cfcd5
--- /dev/null
+++ b/compat/thrust/system/omp/detail/mismatch.h
@@ -0,0 +1,23 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system inherits mismatch
+#include <thrust/system/cpp/detail/mismatch.h>
+
diff --git a/compat/thrust/system/omp/detail/par.h b/compat/thrust/system/omp/detail/par.h
new file mode 100644
index 0000000000..fa6d18e64c
--- /dev/null
+++ b/compat/thrust/system/omp/detail/par.h
@@ -0,0 +1,66 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/omp/detail/execution_policy.h>
+#include <thrust/detail/execute_with_allocator.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace omp
+{
+namespace detail
+{
+
+
+struct par_t : thrust::system::omp::detail::execution_policy<par_t>
+{
+  par_t() : thrust::system::omp::detail::execution_policy<par_t>() {}
+
+  template<typename Allocator>
+    thrust::detail::execute_with_allocator<Allocator, thrust::system::omp::detail::execution_policy>
+      operator()(Allocator &alloc) const
+  {
+    return thrust::detail::execute_with_allocator<Allocator, thrust::system::omp::detail::execution_policy>(alloc);
+  }
+};
+
+
+} // end detail
+
+
+static const detail::par_t par;
+
+
+} // end omp
+} // end system
+
+
+// alias par here
+namespace omp
+{
+
+
+using thrust::system::omp::par;
+
+
+} // end omp
+} // end thrust
+
diff --git a/compat/thrust/system/omp/detail/partition.h b/compat/thrust/system/omp/detail/partition.h
new file mode 100644
index 0000000000..edcbc30834
--- /dev/null
+++ b/compat/thrust/system/omp/detail/partition.h
@@ -0,0 +1,91 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file reduce.h
+ *  \brief OpenMP implementation of reduce algorithms.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/omp/detail/execution_policy.h>
+#include <thrust/pair.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace omp
+{
+namespace detail
+{
+
+
+template<typename DerivedPolicy,
+         typename ForwardIterator,
+         typename Predicate>
+  ForwardIterator stable_partition(execution_policy<DerivedPolicy> &exec,
+                                   ForwardIterator first,
+                                   ForwardIterator last,
+                                   Predicate pred);
+
+template<typename DerivedPolicy,
+         typename ForwardIterator,
+         typename InputIterator,
+         typename Predicate>
+  ForwardIterator stable_partition(execution_policy<DerivedPolicy> &exec,
+                                   ForwardIterator first,
+                                   ForwardIterator last,
+                                   InputIterator stencil,
+                                   Predicate pred);
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename Predicate>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    stable_partition_copy(execution_policy<DerivedPolicy> &exec,
+                          InputIterator first,
+                          InputIterator last,
+                          OutputIterator1 out_true,
+                          OutputIterator2 out_false,
+                          Predicate pred);
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename Predicate>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    stable_partition_copy(execution_policy<DerivedPolicy> &exec,
+                          InputIterator1 first,
+                          InputIterator1 last,
+                          InputIterator2 stencil,
+                          OutputIterator1 out_true,
+                          OutputIterator2 out_false,
+                          Predicate pred);
+
+
+} // end namespace detail
+} // end namespace omp
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/omp/detail/partition.inl>
+
diff --git a/compat/thrust/system/omp/detail/partition.inl b/compat/thrust/system/omp/detail/partition.inl
new file mode 100644
index 0000000000..da629e5c69
--- /dev/null
+++ b/compat/thrust/system/omp/detail/partition.inl
@@ -0,0 +1,108 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file reduce.h
+ *  \brief OpenMP implementation of reduce algorithms.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/omp/detail/partition.h>
+#include <thrust/system/detail/generic/partition.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace omp
+{
+namespace detail
+{
+
+
+template<typename DerivedPolicy,
+         typename ForwardIterator,
+         typename Predicate>
+  ForwardIterator stable_partition(execution_policy<DerivedPolicy> &exec,
+                                   ForwardIterator first,
+                                   ForwardIterator last,
+                                   Predicate pred)
+{
+  // omp prefers generic::stable_partition to cpp::stable_partition
+  return thrust::system::detail::generic::stable_partition(exec, first, last, pred);
+} // end stable_partition()
+
+
+template<typename DerivedPolicy,
+         typename ForwardIterator,
+         typename InputIterator,
+         typename Predicate>
+  ForwardIterator stable_partition(execution_policy<DerivedPolicy> &exec,
+                                   ForwardIterator first,
+                                   ForwardIterator last,
+                                   InputIterator stencil,
+                                   Predicate pred)
+{
+  // omp prefers generic::stable_partition to cpp::stable_partition
+  return thrust::system::detail::generic::stable_partition(exec, first, last, stencil, pred);
+} // end stable_partition()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename Predicate>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    stable_partition_copy(execution_policy<DerivedPolicy> &exec,
+                          InputIterator first,
+                          InputIterator last,
+                          OutputIterator1 out_true,
+                          OutputIterator2 out_false,
+                          Predicate pred)
+{
+  // omp prefers generic::stable_partition_copy to cpp::stable_partition_copy
+  return thrust::system::detail::generic::stable_partition_copy(exec, first, last, out_true, out_false, pred);
+} // end stable_partition_copy()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename Predicate>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    stable_partition_copy(execution_policy<DerivedPolicy> &exec,
+                          InputIterator1 first,
+                          InputIterator1 last,
+                          InputIterator2 stencil,
+                          OutputIterator1 out_true,
+                          OutputIterator2 out_false,
+                          Predicate pred)
+{
+  // omp prefers generic::stable_partition_copy to cpp::stable_partition_copy
+  return thrust::system::detail::generic::stable_partition_copy(exec, first, last, stencil, out_true, out_false, pred);
+} // end stable_partition_copy()
+
+
+} // end namespace detail
+} // end namespace omp
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/omp/detail/reduce.h b/compat/thrust/system/omp/detail/reduce.h
new file mode 100644
index 0000000000..0cc5cebc2f
--- /dev/null
+++ b/compat/thrust/system/omp/detail/reduce.h
@@ -0,0 +1,54 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file reduce.h
+ *  \brief OpenMP implementation of reduce algorithms.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/omp/detail/execution_policy.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace omp
+{
+namespace detail
+{
+
+
+template<typename DerivedPolicy,
+         typename InputIterator, 
+         typename OutputType,
+         typename BinaryFunction>
+  OutputType reduce(execution_policy<DerivedPolicy> &exec,
+                    InputIterator first,
+                    InputIterator last,
+                    OutputType init,
+                    BinaryFunction binary_op);
+
+
+} // end namespace detail
+} // end namespace omp
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/omp/detail/reduce.inl>
+
diff --git a/compat/thrust/system/omp/detail/reduce.inl b/compat/thrust/system/omp/detail/reduce.inl
new file mode 100644
index 0000000000..1347bfd004
--- /dev/null
+++ b/compat/thrust/system/omp/detail/reduce.inl
@@ -0,0 +1,72 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/system/omp/detail/reduce.h>
+#include <thrust/system/omp/detail/default_decomposition.h>
+#include <thrust/system/omp/detail/reduce_intervals.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace omp
+{
+namespace detail
+{
+
+
+template<typename DerivedPolicy,
+         typename InputIterator, 
+         typename OutputType,
+         typename BinaryFunction>
+  OutputType reduce(execution_policy<DerivedPolicy> &exec,
+                    InputIterator first,
+                    InputIterator last,
+                    OutputType init,
+                    BinaryFunction binary_op)
+{
+  typedef typename thrust::iterator_difference<InputIterator>::type difference_type;
+
+  const difference_type n = thrust::distance(first,last);
+
+  // determine first and second level decomposition
+  thrust::system::detail::internal::uniform_decomposition<difference_type> decomp1 = thrust::system::omp::detail::default_decomposition(n);
+  thrust::system::detail::internal::uniform_decomposition<difference_type> decomp2(decomp1.size() + 1, 1, 1);
+
+  // allocate storage for the initializer and partial sums
+  // XXX use select_system for Tag
+  thrust::detail::temporary_array<OutputType,DerivedPolicy> partial_sums(exec, decomp1.size() + 1);
+  
+  // set first element of temp array to init
+  partial_sums[0] = init;
+  
+  // accumulate partial sums (first level reduction)
+  thrust::system::omp::detail::reduce_intervals(exec, first, partial_sums.begin() + 1, binary_op, decomp1);
+
+  // reduce partial sums (second level reduction)
+  thrust::system::omp::detail::reduce_intervals(exec, partial_sums.begin(), partial_sums.begin(), binary_op, decomp2);
+
+  return partial_sums[0];
+} // end reduce()
+
+
+} // end detail
+} // end omp
+} // end system
+} // end thrust
+
diff --git a/compat/thrust/system/omp/detail/reduce_by_key.h b/compat/thrust/system/omp/detail/reduce_by_key.h
new file mode 100644
index 0000000000..d7243ee0ca
--- /dev/null
+++ b/compat/thrust/system/omp/detail/reduce_by_key.h
@@ -0,0 +1,61 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file reduce.h
+ *  \brief OpenMP implementation of reduce algorithms.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/omp/detail/execution_policy.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace omp
+{
+namespace detail
+{
+
+
+template <typename DerivedPolicy,
+          typename InputIterator1,
+          typename InputIterator2,
+          typename OutputIterator1,
+          typename OutputIterator2,
+          typename BinaryPredicate,
+          typename BinaryFunction>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    reduce_by_key(execution_policy<DerivedPolicy> &exec,
+                  InputIterator1 keys_first, 
+                  InputIterator1 keys_last,
+                  InputIterator2 values_first,
+                  OutputIterator1 keys_output,
+                  OutputIterator2 values_output,
+                  BinaryPredicate binary_pred,
+                  BinaryFunction binary_op);
+
+
+} // end namespace detail
+} // end namespace omp
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/omp/detail/reduce_by_key.inl>
+
diff --git a/compat/thrust/system/omp/detail/reduce_by_key.inl b/compat/thrust/system/omp/detail/reduce_by_key.inl
new file mode 100644
index 0000000000..91402d8280
--- /dev/null
+++ b/compat/thrust/system/omp/detail/reduce_by_key.inl
@@ -0,0 +1,57 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/system/omp/detail/reduce_by_key.h>
+#include <thrust/system/detail/generic/reduce_by_key.h>
+#include <thrust/distance.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace omp
+{
+namespace detail
+{
+
+template <typename DerivedPolicy,
+          typename InputIterator1,
+          typename InputIterator2,
+          typename OutputIterator1,
+          typename OutputIterator2,
+          typename BinaryPredicate,
+          typename BinaryFunction>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    reduce_by_key(execution_policy<DerivedPolicy> &exec,
+                  InputIterator1 keys_first, 
+                  InputIterator1 keys_last,
+                  InputIterator2 values_first,
+                  OutputIterator1 keys_output,
+                  OutputIterator2 values_output,
+                  BinaryPredicate binary_pred,
+                  BinaryFunction binary_op)
+{
+  // omp prefers generic::reduce_by_key to cpp::reduce_by_key
+  return thrust::system::detail::generic::reduce_by_key(exec, keys_first, keys_last, values_first, keys_output, values_output, binary_pred, binary_op);
+} // end reduce_by_key()
+
+
+} // end detail
+} // end omp
+} // end system
+} // end thrust
+
diff --git a/compat/thrust/system/omp/detail/reduce_intervals.h b/compat/thrust/system/omp/detail/reduce_intervals.h
new file mode 100644
index 0000000000..7bce2074a3
--- /dev/null
+++ b/compat/thrust/system/omp/detail/reduce_intervals.h
@@ -0,0 +1,53 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file reduce_intervals.h
+ *  \brief OpenMP implementations of reduce_intervals algorithms.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/omp/detail/execution_policy.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace omp
+{
+namespace detail
+{
+
+template <typename DerivedPolicy,
+          typename InputIterator,
+          typename OutputIterator,
+          typename BinaryFunction,
+          typename Decomposition>
+void reduce_intervals(execution_policy<DerivedPolicy> &exec,
+                      InputIterator input,
+                      OutputIterator output,
+                      BinaryFunction binary_op,
+                      Decomposition decomp);
+
+} // end namespace detail
+} // end namespace omp
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/omp/detail/reduce_intervals.inl>
+
diff --git a/compat/thrust/system/omp/detail/reduce_intervals.inl b/compat/thrust/system/omp/detail/reduce_intervals.inl
new file mode 100644
index 0000000000..0752b8aab9
--- /dev/null
+++ b/compat/thrust/system/omp/detail/reduce_intervals.inl
@@ -0,0 +1,93 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+#include <thrust/detail/config.h>
+#include <thrust/system/omp/detail/reduce_intervals.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/detail/function.h>
+#include <thrust/detail/cstdint.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace omp
+{
+namespace detail
+{
+
+template <typename DerivedPolicy,
+          typename InputIterator,
+          typename OutputIterator,
+          typename BinaryFunction,
+          typename Decomposition>
+void reduce_intervals(execution_policy<DerivedPolicy> &,
+                      InputIterator input,
+                      OutputIterator output,
+                      BinaryFunction binary_op,
+                      Decomposition decomp)
+{
+  // we're attempting to launch an omp kernel, assert we're compiling with omp support
+  // ========================================================================
+  // X Note to the user: If you've found this line due to a compiler error, X
+  // X you need to enable OpenMP support in your compiler.                  X
+  // ========================================================================
+  THRUST_STATIC_ASSERT( (thrust::detail::depend_on_instantiation<InputIterator,
+                        (THRUST_DEVICE_COMPILER_IS_OMP_CAPABLE == THRUST_TRUE)>::value) );
+
+#if (THRUST_DEVICE_COMPILER_IS_OMP_CAPABLE == THRUST_TRUE)
+  typedef typename thrust::iterator_value<OutputIterator>::type OutputType;
+
+  // wrap binary_op
+  thrust::detail::host_function<BinaryFunction,OutputType> wrapped_binary_op(binary_op);
+
+  typedef thrust::detail::intptr_t index_type;
+
+  index_type n = static_cast<index_type>(decomp.size());
+
+#if (THRUST_DEVICE_COMPILER_IS_OMP_CAPABLE == THRUST_TRUE)
+# pragma omp parallel for
+#endif // THRUST_DEVICE_COMPILER_IS_OMP_CAPABLE
+  for(index_type i = 0; i < n; i++)
+  {
+    InputIterator begin = input + decomp[i].begin();
+    InputIterator end   = input + decomp[i].end();
+
+    if (begin != end)
+    {
+      OutputType sum = thrust::raw_reference_cast(*begin);
+
+      ++begin;
+
+      while (begin != end)
+      {
+        sum = wrapped_binary_op(sum, *begin);
+        ++begin;
+      }
+
+      OutputIterator tmp = output + i;
+      *tmp = sum;
+    }
+  }
+#endif // THRUST_DEVICE_COMPILER_IS_OMP_CAPABLE
+}
+
+} // end namespace detail
+} // end namespace omp
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/omp/detail/remove.h b/compat/thrust/system/omp/detail/remove.h
new file mode 100644
index 0000000000..ebcb49613c
--- /dev/null
+++ b/compat/thrust/system/omp/detail/remove.h
@@ -0,0 +1,81 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/omp/detail/execution_policy.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace omp
+{
+namespace detail
+{
+
+template<typename DerivedPolicy,
+         typename ForwardIterator,
+         typename Predicate>
+  ForwardIterator remove_if(execution_policy<DerivedPolicy> &exec,
+                            ForwardIterator first,
+                            ForwardIterator last,
+                            Predicate pred);
+
+
+template<typename DerivedPolicy,
+         typename ForwardIterator,
+         typename InputIterator,
+         typename Predicate>
+  ForwardIterator remove_if(execution_policy<DerivedPolicy> &exec,
+                            ForwardIterator first,
+                            ForwardIterator last,
+                            InputIterator stencil,
+                            Predicate pred);
+
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator,
+         typename Predicate>
+  OutputIterator remove_copy_if(execution_policy<DerivedPolicy> &exec,
+                                InputIterator first,
+                                InputIterator last,
+                                OutputIterator result,
+                                Predicate pred);
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename Predicate>
+  OutputIterator remove_copy_if(execution_policy<DerivedPolicy> &exec,
+                                InputIterator1 first,
+                                InputIterator1 last,
+                                InputIterator2 stencil,
+                                OutputIterator result,
+                                Predicate pred);
+
+
+} // end namespace detail
+} // end namespace omp
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/omp/detail/remove.inl>
+
diff --git a/compat/thrust/system/omp/detail/remove.inl b/compat/thrust/system/omp/detail/remove.inl
new file mode 100644
index 0000000000..c056f967e4
--- /dev/null
+++ b/compat/thrust/system/omp/detail/remove.inl
@@ -0,0 +1,94 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/omp/detail/remove.h>
+#include <thrust/system/detail/generic/remove.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace omp
+{
+namespace detail
+{
+
+template<typename DerivedPolicy,
+         typename ForwardIterator,
+         typename Predicate>
+  ForwardIterator remove_if(execution_policy<DerivedPolicy> &exec,
+                            ForwardIterator first,
+                            ForwardIterator last,
+                            Predicate pred)
+{
+  // omp prefers generic::remove_if to cpp::remove_if
+  return thrust::system::detail::generic::remove_if(exec, first, last, pred);
+}
+
+
+template<typename DerivedPolicy,
+         typename ForwardIterator,
+         typename InputIterator,
+         typename Predicate>
+  ForwardIterator remove_if(execution_policy<DerivedPolicy> &exec,
+                            ForwardIterator first,
+                            ForwardIterator last,
+                            InputIterator stencil,
+                            Predicate pred)
+{
+  // omp prefers generic::remove_if to cpp::remove_if
+  return thrust::system::detail::generic::remove_if(exec, first, last, stencil, pred);
+}
+
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator,
+         typename Predicate>
+  OutputIterator remove_copy_if(execution_policy<DerivedPolicy> &exec,
+                                InputIterator first,
+                                InputIterator last,
+                                OutputIterator result,
+                                Predicate pred)
+{
+  // omp prefers generic::remove_copy_if to cpp::remove_copy_if
+  return thrust::system::detail::generic::remove_copy_if(exec, first, last, result, pred);
+}
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename Predicate>
+  OutputIterator remove_copy_if(execution_policy<DerivedPolicy> &exec,
+                                InputIterator1 first,
+                                InputIterator1 last,
+                                InputIterator2 stencil,
+                                OutputIterator result,
+                                Predicate pred)
+{
+  // omp prefers generic::remove_copy_if to cpp::remove_copy_if
+  return thrust::system::detail::generic::remove_copy_if(exec, first, last, stencil, result, pred);
+}
+
+} // end namespace detail
+} // end namespace omp
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/omp/detail/replace.h b/compat/thrust/system/omp/detail/replace.h
new file mode 100644
index 0000000000..c48555d0e1
--- /dev/null
+++ b/compat/thrust/system/omp/detail/replace.h
@@ -0,0 +1,23 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system inherits this algorithm
+#include <thrust/system/cpp/detail/scatter.h>
+
diff --git a/compat/thrust/system/omp/detail/reverse.h b/compat/thrust/system/omp/detail/reverse.h
new file mode 100644
index 0000000000..04923d1f6a
--- /dev/null
+++ b/compat/thrust/system/omp/detail/reverse.h
@@ -0,0 +1,23 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system inherits reverse
+#include <thrust/system/cpp/detail/reverse.h>
+
diff --git a/compat/thrust/system/omp/detail/scan.h b/compat/thrust/system/omp/detail/scan.h
new file mode 100644
index 0000000000..c105951fff
--- /dev/null
+++ b/compat/thrust/system/omp/detail/scan.h
@@ -0,0 +1,23 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system inherits scan
+#include <thrust/system/cpp/detail/scan.h>
+
diff --git a/compat/thrust/system/omp/detail/scan_by_key.h b/compat/thrust/system/omp/detail/scan_by_key.h
new file mode 100644
index 0000000000..bfbd5d69bd
--- /dev/null
+++ b/compat/thrust/system/omp/detail/scan_by_key.h
@@ -0,0 +1,23 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system inherits this algorithm
+#include <thrust/system/cpp/detail/scan_by_key.h>
+
diff --git a/compat/thrust/system/omp/detail/scatter.h b/compat/thrust/system/omp/detail/scatter.h
new file mode 100644
index 0000000000..c48555d0e1
--- /dev/null
+++ b/compat/thrust/system/omp/detail/scatter.h
@@ -0,0 +1,23 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system inherits this algorithm
+#include <thrust/system/cpp/detail/scatter.h>
+
diff --git a/compat/thrust/system/omp/detail/sequence.h b/compat/thrust/system/omp/detail/sequence.h
new file mode 100644
index 0000000000..811d8f5fbb
--- /dev/null
+++ b/compat/thrust/system/omp/detail/sequence.h
@@ -0,0 +1,23 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system inherits sequence
+#include <thrust/system/cpp/detail/sequence.h>
+
diff --git a/compat/thrust/system/omp/detail/set_operations.h b/compat/thrust/system/omp/detail/set_operations.h
new file mode 100644
index 0000000000..687edb2e7d
--- /dev/null
+++ b/compat/thrust/system/omp/detail/set_operations.h
@@ -0,0 +1,23 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system inherits set_operations
+#include <thrust/system/cpp/detail/set_operations.h>
+
diff --git a/compat/thrust/system/omp/detail/sort.h b/compat/thrust/system/omp/detail/sort.h
new file mode 100644
index 0000000000..9a480f2799
--- /dev/null
+++ b/compat/thrust/system/omp/detail/sort.h
@@ -0,0 +1,55 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/omp/detail/execution_policy.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace omp
+{
+namespace detail
+{
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator,
+         typename StrictWeakOrdering>
+void stable_sort(execution_policy<DerivedPolicy> &exec,
+                 RandomAccessIterator first,
+                 RandomAccessIterator last,
+                 StrictWeakOrdering comp);
+    
+template<typename DerivedPolicy,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2,
+         typename StrictWeakOrdering>
+void stable_sort_by_key(execution_policy<DerivedPolicy> &exec,
+                        RandomAccessIterator1 keys_first,
+                        RandomAccessIterator1 keys_last,
+                        RandomAccessIterator2 values_first,
+                        StrictWeakOrdering comp);
+
+} // end namespace detail
+} // end namespace omp
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/omp/detail/sort.inl>
+
diff --git a/compat/thrust/system/omp/detail/sort.inl b/compat/thrust/system/omp/detail/sort.inl
new file mode 100644
index 0000000000..ab4f4a1bbe
--- /dev/null
+++ b/compat/thrust/system/omp/detail/sort.inl
@@ -0,0 +1,249 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+#include <thrust/detail/config.h>
+
+// don't attempt to #include this file without omp support
+#if (THRUST_DEVICE_COMPILER_IS_OMP_CAPABLE == THRUST_TRUE)
+#include <omp.h>
+#endif // omp support
+
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/system/detail/generic/select_system.h>
+#include <thrust/system/cpp/detail/sort.h>
+#include <thrust/system/cpp/detail/merge.h>
+#include <thrust/system/cpp/detail/execution_policy.h>
+#include <thrust/detail/temporary_array.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace omp
+{
+namespace detail
+{
+namespace sort_detail
+{
+
+
+template <typename DerivedPolicy,
+          typename RandomAccessIterator,
+          typename StrictWeakOrdering>
+void inplace_merge(execution_policy<DerivedPolicy> &exec,
+                   RandomAccessIterator first,
+                   RandomAccessIterator middle,
+                   RandomAccessIterator last,
+                   StrictWeakOrdering comp)
+{
+  typedef typename thrust::iterator_value<RandomAccessIterator>::type value_type;
+
+  thrust::detail::temporary_array<value_type,DerivedPolicy> a(exec, first, middle);
+  thrust::detail::temporary_array<value_type,DerivedPolicy> b(exec, middle, last);
+
+  thrust::system::cpp::detail::merge(exec, a.begin(), a.end(), b.begin(), b.end(), first, comp);
+}
+
+
+template <typename DerivedPolicy,
+          typename RandomAccessIterator1,
+          typename RandomAccessIterator2,
+          typename StrictWeakOrdering>
+void inplace_merge_by_key(execution_policy<DerivedPolicy> &exec,
+                          RandomAccessIterator1 first1,
+                          RandomAccessIterator1 middle1,
+                          RandomAccessIterator1 last1,
+                          RandomAccessIterator2 first2,
+                          StrictWeakOrdering comp)
+{
+  typedef typename thrust::iterator_value<RandomAccessIterator1>::type value_type1;
+  typedef typename thrust::iterator_value<RandomAccessIterator2>::type value_type2;
+
+  RandomAccessIterator2 middle2 = first2 + (middle1 - first1);
+  RandomAccessIterator2 last2   = first2 + (last1   - first1);
+
+  thrust::detail::temporary_array<value_type1,DerivedPolicy> lhs1(exec, first1, middle1);
+  thrust::detail::temporary_array<value_type1,DerivedPolicy> rhs1(exec, middle1, last1);
+  thrust::detail::temporary_array<value_type2,DerivedPolicy> lhs2(exec, first2, middle2);
+  thrust::detail::temporary_array<value_type2,DerivedPolicy> rhs2(exec, middle2, last2);
+
+  thrust::system::cpp::detail::merge_by_key
+    (exec,
+     lhs1.begin(), lhs1.end(), rhs1.begin(), rhs1.end(),
+     lhs2.begin(), rhs2.begin(),
+     first1, first2, comp);
+}
+
+
+} // end sort_detail
+
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator,
+         typename StrictWeakOrdering>
+void stable_sort(execution_policy<DerivedPolicy> &exec,
+                 RandomAccessIterator first,
+                 RandomAccessIterator last,
+                 StrictWeakOrdering comp)
+{
+  // we're attempting to launch an omp kernel, assert we're compiling with omp support
+  // ========================================================================
+  // X Note to the user: If you've found this line due to a compiler error, X
+  // X you need to enable OpenMP support in your compiler.                  X
+  // ========================================================================
+  THRUST_STATIC_ASSERT( (thrust::detail::depend_on_instantiation<RandomAccessIterator,
+                        (THRUST_DEVICE_COMPILER_IS_OMP_CAPABLE == THRUST_TRUE)>::value) );
+
+#if (THRUST_DEVICE_COMPILER_IS_OMP_CAPABLE == THRUST_TRUE)
+  typedef typename thrust::iterator_difference<RandomAccessIterator>::type IndexType;
+  
+  if (first == last)
+    return;
+
+  #pragma omp parallel
+  {
+    thrust::system::detail::internal::uniform_decomposition<IndexType> decomp(last - first, 1, omp_get_num_threads());
+
+    // process id
+    IndexType p_i = omp_get_thread_num();
+
+    // every thread sorts its own tile
+    if (p_i < decomp.size())
+    {
+      thrust::system::cpp::detail::stable_sort(exec,
+                                               first + decomp[p_i].begin(),
+                                               first + decomp[p_i].end(),
+                                               comp);
+    }
+
+    #pragma omp barrier
+
+    IndexType nseg = decomp.size();
+    IndexType h = 2;
+
+    // keep track of which sub-range we're processing
+    IndexType a=p_i, b=p_i, c=p_i+1;
+
+    while( nseg>1 )
+    {
+        if(c >= decomp.size())
+          c = decomp.size() - 1;
+
+        if((p_i % h) == 0 && c > b)
+        {
+          thrust::system::omp::detail::sort_detail::inplace_merge
+            (exec,
+             first + decomp[a].begin(),
+             first + decomp[b].end(),
+             first + decomp[c].end(),
+             comp);
+            b = c;
+            c += h;
+        }
+
+        nseg = (nseg + 1) / 2;
+        h *= 2;
+
+        #pragma omp barrier
+    }
+  }
+#endif // THRUST_DEVICE_COMPILER_IS_OMP_CAPABLE
+}
+
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2,
+         typename StrictWeakOrdering>
+void stable_sort_by_key(execution_policy<DerivedPolicy> &exec,
+                        RandomAccessIterator1 keys_first,
+                        RandomAccessIterator1 keys_last,
+                        RandomAccessIterator2 values_first,
+                        StrictWeakOrdering comp)
+{
+  // we're attempting to launch an omp kernel, assert we're compiling with omp support
+  // ========================================================================
+  // X Note to the user: If you've found this line due to a compiler error, X
+  // X you need to enable OpenMP support in your compiler.                  X
+  // ========================================================================
+  THRUST_STATIC_ASSERT( (thrust::detail::depend_on_instantiation<RandomAccessIterator1,
+                        (THRUST_DEVICE_COMPILER_IS_OMP_CAPABLE == THRUST_TRUE)>::value) );
+
+#if (THRUST_DEVICE_COMPILER_IS_OMP_CAPABLE == THRUST_TRUE)
+  typedef typename thrust::iterator_difference<RandomAccessIterator1>::type IndexType;
+  
+  if (keys_first == keys_last)
+    return;
+
+  #pragma omp parallel
+  {
+    thrust::system::detail::internal::uniform_decomposition<IndexType> decomp(keys_last - keys_first, 1, omp_get_num_threads());
+
+    // process id
+    IndexType p_i = omp_get_thread_num();
+
+    // every thread sorts its own tile
+    if (p_i < decomp.size())
+    {
+      thrust::system::cpp::detail::stable_sort_by_key(exec,
+                                                      keys_first + decomp[p_i].begin(),
+                                                      keys_first + decomp[p_i].end(),
+                                                      values_first + decomp[p_i].begin(),
+                                                      comp);
+    }
+
+    #pragma omp barrier
+
+    IndexType nseg = decomp.size();
+    IndexType h = 2;
+
+    // keep track of which sub-range we're processing
+    IndexType a=p_i, b=p_i, c=p_i+1;
+
+    while( nseg>1 )
+    {
+        if(c >= decomp.size())
+          c = decomp.size() - 1;
+
+        if((p_i % h) == 0 && c > b)
+        {
+          thrust::system::omp::detail::sort_detail::inplace_merge_by_key
+            (exec,
+             keys_first + decomp[a].begin(),
+             keys_first + decomp[b].end(),
+             keys_first + decomp[c].end(),
+             values_first + decomp[a].begin(),
+             comp);
+            b = c;
+            c += h;
+        }
+
+        nseg = (nseg + 1) / 2;
+        h *= 2;
+
+        #pragma omp barrier
+    }
+  }
+#endif // THRUST_DEVICE_COMPILER_IS_OMP_CAPABLE
+}
+
+
+} // end namespace detail
+} // end namespace omp
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/omp/detail/swap_ranges.h b/compat/thrust/system/omp/detail/swap_ranges.h
new file mode 100644
index 0000000000..e683aaaa6e
--- /dev/null
+++ b/compat/thrust/system/omp/detail/swap_ranges.h
@@ -0,0 +1,23 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// omp inherits swap_ranges
+#include <thrust/system/cpp/detail/swap_ranges.h>
+
diff --git a/compat/thrust/system/omp/detail/tabulate.h b/compat/thrust/system/omp/detail/tabulate.h
new file mode 100644
index 0000000000..da65d8e44d
--- /dev/null
+++ b/compat/thrust/system/omp/detail/tabulate.h
@@ -0,0 +1,23 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system inherits tabulate
+#include <thrust/system/cpp/detail/tabulate.h>
+
diff --git a/compat/thrust/system/omp/detail/temporary_buffer.h b/compat/thrust/system/omp/detail/temporary_buffer.h
new file mode 100644
index 0000000000..628bd75719
--- /dev/null
+++ b/compat/thrust/system/omp/detail/temporary_buffer.h
@@ -0,0 +1,22 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system has no special temporary buffer functions
+
diff --git a/compat/thrust/system/omp/detail/transform.h b/compat/thrust/system/omp/detail/transform.h
new file mode 100644
index 0000000000..70ce1f41b6
--- /dev/null
+++ b/compat/thrust/system/omp/detail/transform.h
@@ -0,0 +1,23 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// omp inherits transform
+#include <thrust/system/cpp/detail/transform.h>
+
diff --git a/compat/thrust/system/omp/detail/transform_reduce.h b/compat/thrust/system/omp/detail/transform_reduce.h
new file mode 100644
index 0000000000..23ed07054a
--- /dev/null
+++ b/compat/thrust/system/omp/detail/transform_reduce.h
@@ -0,0 +1,23 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system inherits transform_reduce
+#include <thrust/system/cpp/detail/transform_reduce.h>
+
diff --git a/compat/thrust/system/omp/detail/transform_scan.h b/compat/thrust/system/omp/detail/transform_scan.h
new file mode 100644
index 0000000000..fc2e55d0c0
--- /dev/null
+++ b/compat/thrust/system/omp/detail/transform_scan.h
@@ -0,0 +1,23 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system inherits transform_scan
+#include <thrust/system/cpp/detail/transform_scan.h>
+
diff --git a/compat/thrust/system/omp/detail/uninitialized_copy.h b/compat/thrust/system/omp/detail/uninitialized_copy.h
new file mode 100644
index 0000000000..944f4baf0e
--- /dev/null
+++ b/compat/thrust/system/omp/detail/uninitialized_copy.h
@@ -0,0 +1,23 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system inherits uninitialized_copy
+#include <thrust/system/cpp/detail/uninitialized_copy.h>
+
diff --git a/compat/thrust/system/omp/detail/uninitialized_fill.h b/compat/thrust/system/omp/detail/uninitialized_fill.h
new file mode 100644
index 0000000000..b9d6de20fa
--- /dev/null
+++ b/compat/thrust/system/omp/detail/uninitialized_fill.h
@@ -0,0 +1,23 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system inherits uninitialized_fill
+#include <thrust/system/cpp/detail/uninitialized_fill.h>
+
diff --git a/compat/thrust/system/omp/detail/unique.h b/compat/thrust/system/omp/detail/unique.h
new file mode 100644
index 0000000000..60c617bee9
--- /dev/null
+++ b/compat/thrust/system/omp/detail/unique.h
@@ -0,0 +1,59 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/omp/detail/execution_policy.h>
+#include <thrust/pair.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace omp
+{
+namespace detail
+{
+
+
+template<typename DerivedPolicy,
+         typename ForwardIterator,
+         typename BinaryPredicate>
+  ForwardIterator unique(execution_policy<DerivedPolicy> &exec,
+                         ForwardIterator first,
+                         ForwardIterator last,
+                         BinaryPredicate binary_pred);
+
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator,
+         typename BinaryPredicate>
+  OutputIterator unique_copy(execution_policy<DerivedPolicy> &exec,
+                             InputIterator first,
+                             InputIterator last,
+                             OutputIterator output,
+                             BinaryPredicate binary_pred);
+
+
+} // end namespace detail
+} // end namespace omp 
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/omp/detail/unique.inl>
+
diff --git a/compat/thrust/system/omp/detail/unique.inl b/compat/thrust/system/omp/detail/unique.inl
new file mode 100644
index 0000000000..d66ac3bf4f
--- /dev/null
+++ b/compat/thrust/system/omp/detail/unique.inl
@@ -0,0 +1,66 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/omp/detail/unique.h>
+#include <thrust/system/detail/generic/unique.h>
+#include <thrust/pair.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace omp
+{
+namespace detail
+{
+
+
+template<typename DerivedPolicy,
+         typename ForwardIterator,
+         typename BinaryPredicate>
+  ForwardIterator unique(execution_policy<DerivedPolicy> &exec,
+                         ForwardIterator first,
+                         ForwardIterator last,
+                         BinaryPredicate binary_pred)
+{
+  // omp prefers generic::unique to cpp::unique
+  return thrust::system::detail::generic::unique(exec,first,last,binary_pred);
+} // end unique()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator,
+         typename BinaryPredicate>
+  OutputIterator unique_copy(execution_policy<DerivedPolicy> &exec,
+                             InputIterator first,
+                             InputIterator last,
+                             OutputIterator output,
+                             BinaryPredicate binary_pred)
+{
+  // omp prefers generic::unique_copy to cpp::unique_copy
+  return thrust::system::detail::generic::unique_copy(exec,first,last,output,binary_pred);
+} // end unique_copy()
+
+
+} // end namespace detail
+} // end namespace omp 
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/omp/detail/unique_by_key.h b/compat/thrust/system/omp/detail/unique_by_key.h
new file mode 100644
index 0000000000..8fdde66edf
--- /dev/null
+++ b/compat/thrust/system/omp/detail/unique_by_key.h
@@ -0,0 +1,67 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/omp/detail/execution_policy.h>
+#include <thrust/pair.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace omp
+{
+namespace detail
+{
+
+
+template<typename DerivedPolicy,
+         typename ForwardIterator1,
+         typename ForwardIterator2,
+         typename BinaryPredicate>
+  thrust::pair<ForwardIterator1,ForwardIterator2>
+    unique_by_key(execution_policy<DerivedPolicy> &exec,
+                  ForwardIterator1 keys_first, 
+                  ForwardIterator1 keys_last,
+                  ForwardIterator2 values_first,
+                  BinaryPredicate binary_pred);
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename BinaryPredicate>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    unique_by_key_copy(execution_policy<DerivedPolicy> &exec,
+                       InputIterator1 keys_first, 
+                       InputIterator1 keys_last,
+                       InputIterator2 values_first,
+                       OutputIterator1 keys_output,
+                       OutputIterator2 values_output,
+                       BinaryPredicate binary_pred);
+
+
+} // end namespace detail
+} // end namespace omp 
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/omp/detail/unique_by_key.inl>
+
diff --git a/compat/thrust/system/omp/detail/unique_by_key.inl b/compat/thrust/system/omp/detail/unique_by_key.inl
new file mode 100644
index 0000000000..644b5ed6aa
--- /dev/null
+++ b/compat/thrust/system/omp/detail/unique_by_key.inl
@@ -0,0 +1,74 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/omp/detail/unique_by_key.h>
+#include <thrust/system/detail/generic/unique_by_key.h>
+#include <thrust/pair.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace omp
+{
+namespace detail
+{
+
+
+template<typename DerivedPolicy,
+         typename ForwardIterator1,
+         typename ForwardIterator2,
+         typename BinaryPredicate>
+  thrust::pair<ForwardIterator1,ForwardIterator2>
+    unique_by_key(execution_policy<DerivedPolicy> &exec,
+                  ForwardIterator1 keys_first, 
+                  ForwardIterator1 keys_last,
+                  ForwardIterator2 values_first,
+                  BinaryPredicate binary_pred)
+{
+  // omp prefers generic::unique_by_key to cpp::unique_by_key
+  return thrust::system::detail::generic::unique_by_key(exec,keys_first,keys_last,values_first,binary_pred);
+} // end unique_by_key()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename BinaryPredicate>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    unique_by_key_copy(execution_policy<DerivedPolicy> &exec,
+                       InputIterator1 keys_first, 
+                       InputIterator1 keys_last,
+                       InputIterator2 values_first,
+                       OutputIterator1 keys_output,
+                       OutputIterator2 values_output,
+                       BinaryPredicate binary_pred)
+{
+  // omp prefers generic::unique_by_key_copy to cpp::unique_by_key_copy
+  return thrust::system::detail::generic::unique_by_key_copy(exec,keys_first,keys_last,values_first,keys_output,values_output,binary_pred);
+} // end unique_by_key_copy()
+
+
+} // end namespace detail
+} // end namespace omp
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/omp/detail/vector.inl b/compat/thrust/system/omp/detail/vector.inl
new file mode 100644
index 0000000000..32c845c4ad
--- /dev/null
+++ b/compat/thrust/system/omp/detail/vector.inl
@@ -0,0 +1,97 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/omp/vector.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace omp
+{
+
+template<typename T, typename Allocator>
+  vector<T,Allocator>
+    ::vector()
+      : super_t()
+{}
+
+template<typename T, typename Allocator>
+  vector<T,Allocator>
+    ::vector(size_type n)
+      : super_t(n)
+{}
+
+template<typename T, typename Allocator>
+  vector<T,Allocator>
+    ::vector(size_type n, const value_type &value)
+      : super_t(n,value)
+{}
+
+template<typename T, typename Allocator>
+  vector<T,Allocator>
+    ::vector(const vector &x)
+      : super_t(x)
+{}
+
+template<typename T, typename Allocator>
+  template<typename OtherT, typename OtherAllocator>
+    vector<T,Allocator>
+      ::vector(const thrust::detail::vector_base<OtherT,OtherAllocator> &x)
+        : super_t(x)
+{}
+
+template<typename T, typename Allocator>
+  template<typename OtherT, typename OtherAllocator>
+    vector<T,Allocator>
+      ::vector(const std::vector<OtherT,OtherAllocator> &x)
+        : super_t(x)
+{}
+
+template<typename T, typename Allocator>
+  template<typename InputIterator>
+    vector<T,Allocator>
+      ::vector(InputIterator first, InputIterator last)
+        : super_t(first,last)
+{}
+
+template<typename T, typename Allocator>
+  template<typename OtherT, typename OtherAllocator>
+    vector<T,Allocator> &
+      vector<T,Allocator>
+        ::operator=(const std::vector<OtherT,OtherAllocator> &x)
+{
+  super_t::operator=(x);
+  return *this;
+}
+
+template<typename T, typename Allocator>
+  template<typename OtherT, typename OtherAllocator>
+    vector<T,Allocator> &
+      vector<T,Allocator>
+        ::operator=(const thrust::detail::vector_base<OtherT,OtherAllocator> &x)
+{
+  super_t::operator=(x);
+  return *this;
+}
+      
+} // end omp
+} // end system
+} // end thrust
+
diff --git a/compat/thrust/system/omp/execution_policy.h b/compat/thrust/system/omp/execution_policy.h
new file mode 100644
index 0000000000..7d5d1d80d3
--- /dev/null
+++ b/compat/thrust/system/omp/execution_policy.h
@@ -0,0 +1,156 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+/*! \file thrust/system/omp/execution_policy.h
+ *  \brief Execution policies for Thrust's OpenMP system.
+ */
+
+#include <thrust/detail/config.h>
+
+// get the execution policies definitions first
+#include <thrust/system/omp/detail/execution_policy.h>
+
+// get the definition of par
+#include <thrust/system/omp/detail/par.h>
+
+// now get all the algorithm definitions
+
+#include <thrust/system/omp/detail/adjacent_difference.h>
+#include <thrust/system/omp/detail/assign_value.h>
+#include <thrust/system/omp/detail/binary_search.h>
+#include <thrust/system/omp/detail/copy.h>
+#include <thrust/system/omp/detail/copy_if.h>
+#include <thrust/system/omp/detail/count.h>
+#include <thrust/system/omp/detail/equal.h>
+#include <thrust/system/omp/detail/extrema.h>
+#include <thrust/system/omp/detail/fill.h>
+#include <thrust/system/omp/detail/find.h>
+#include <thrust/system/omp/detail/for_each.h>
+#include <thrust/system/omp/detail/gather.h>
+#include <thrust/system/omp/detail/generate.h>
+#include <thrust/system/omp/detail/get_value.h>
+#include <thrust/system/omp/detail/inner_product.h>
+#include <thrust/system/omp/detail/iter_swap.h>
+#include <thrust/system/omp/detail/logical.h>
+#include <thrust/system/omp/detail/malloc_and_free.h>
+#include <thrust/system/omp/detail/merge.h>
+#include <thrust/system/omp/detail/mismatch.h>
+#include <thrust/system/omp/detail/partition.h>
+#include <thrust/system/omp/detail/reduce.h>
+#include <thrust/system/omp/detail/reduce_by_key.h>
+#include <thrust/system/omp/detail/remove.h>
+#include <thrust/system/omp/detail/replace.h>
+#include <thrust/system/omp/detail/reverse.h>
+#include <thrust/system/omp/detail/scan.h>
+#include <thrust/system/omp/detail/scan_by_key.h>
+#include <thrust/system/omp/detail/scatter.h>
+#include <thrust/system/omp/detail/sequence.h>
+#include <thrust/system/omp/detail/set_operations.h>
+#include <thrust/system/omp/detail/sort.h>
+#include <thrust/system/omp/detail/swap_ranges.h>
+#include <thrust/system/omp/detail/tabulate.h>
+#include <thrust/system/omp/detail/transform.h>
+#include <thrust/system/omp/detail/transform_reduce.h>
+#include <thrust/system/omp/detail/transform_scan.h>
+#include <thrust/system/omp/detail/uninitialized_copy.h>
+#include <thrust/system/omp/detail/uninitialized_fill.h>
+#include <thrust/system/omp/detail/unique.h>
+#include <thrust/system/omp/detail/unique_by_key.h>
+
+
+// define these entities here for the purpose of Doxygenating them
+// they are actually defined elsewhere
+#if 0
+namespace thrust
+{
+namespace system
+{
+namespace omp
+{
+
+
+/*! \addtogroup execution_policies
+ *  \{
+ */
+
+
+/*! \p thrust::omp::execution_policy is the base class for all Thrust parallel execution
+ *  policies which are derived from Thrust's OpenMP backend system.
+ */
+template<typename DerivedPolicy>
+struct execution_policy : thrust::execution_policy<DerivedPolicy>
+{};
+
+
+/*! \p omp::tag is a type representing Thrust's standard C++ backend system in C++'s type system.
+ *  Iterators "tagged" with a type which is convertible to \p omp::tag assert that they may be
+ *  "dispatched" to algorithm implementations in the \p omp system.
+ */
+struct tag : thrust::system::omp::execution_policy<tag> { unspecified };
+
+
+/*! \p thrust::omp::par is the parallel execution policy associated with Thrust's OpenMP
+ *  backend system.
+ *
+ *  Instead of relying on implicit algorithm dispatch through iterator system tags, users may
+ *  directly target Thrust's OpenMP backend system by providing \p thrust::omp::par as an algorithm
+ *  parameter.
+ *
+ *  Explicit dispatch can be useful in avoiding the introduction of data copies into containers such
+ *  as \p thrust::omp::vector.
+ *
+ *  The type of \p thrust::omp::par is implementation-defined.
+ *
+ *  The following code snippet demonstrates how to use \p thrust::omp::par to explicitly dispatch an
+ *  invocation of \p thrust::for_each to the OpenMP backend system:
+ *
+ *  \code
+ *  #include <thrust/for_each.h>
+ *  #include <thrust/system/omp/execution_policy.h>
+ *  #include <cstdio>
+ *
+ *  struct printf_functor
+ *  {
+ *    __host__ __device__
+ *    void operator()(int x)
+ *    {
+ *      printf("%d\n");
+ *    }
+ *  };
+ *  ...
+ *  int vec[3];
+ *  vec[0] = 0; vec[1] = 1; vec[2] = 2;
+ *
+ *  thrust::for_each(thrust::omp::par, vec.begin(), vec.end(), printf_functor());
+ *
+ *  // 0 1 2 is printed to standard output in some unspecified order
+ *  \endcode
+ */
+static const unspecified par;
+
+
+/*! \}
+ */
+
+
+} // end cpp
+} // end system
+} // end thrust
+#endif
+
+
diff --git a/compat/thrust/system/omp/memory.h b/compat/thrust/system/omp/memory.h
new file mode 100644
index 0000000000..0a23434833
--- /dev/null
+++ b/compat/thrust/system/omp/memory.h
@@ -0,0 +1,414 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+/*! \file thrust/system/omp/memory.h
+ *  \brief Managing memory associated with Thrust's OpenMP system.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/omp/detail/execution_policy.h>
+#include <thrust/memory.h>
+#include <thrust/detail/type_traits.h>
+#include <thrust/detail/allocator/malloc_allocator.h>
+#include <ostream>
+
+namespace thrust
+{
+namespace system
+{
+namespace omp
+{
+
+template<typename> class pointer;
+
+} // end omp
+} // end system
+} // end thrust
+
+
+/*! \cond
+ */
+
+// specialize std::iterator_traits to avoid problems with the name of
+// pointer's constructor shadowing its nested pointer type
+// do this before pointer is defined so the specialization is correctly
+// used inside the definition
+namespace std
+{
+
+template<typename Element>
+  struct iterator_traits<thrust::system::omp::pointer<Element> >
+{
+  private:
+    typedef thrust::system::omp::pointer<Element> ptr;
+
+  public:
+    typedef typename ptr::iterator_category       iterator_category;
+    typedef typename ptr::value_type              value_type;
+    typedef typename ptr::difference_type         difference_type;
+    typedef ptr                                   pointer;
+    typedef typename ptr::reference               reference;
+}; // end iterator_traits
+
+} // end std
+
+/*! \endcond
+ */
+
+
+namespace thrust
+{
+namespace system
+{
+
+/*! \addtogroup system_backends Systems
+ *  \ingroup system
+ *  \{
+ */
+
+/*! \namespace thrust::system::omp
+ *  \brief \p thrust::system::omp is the namespace containing functionality for allocating, manipulating,
+ *         and deallocating memory available to Thrust's OpenMP backend system.
+ *         The identifiers are provided in a separate namespace underneath <tt>thrust::system</tt>
+ *         for import convenience but are also aliased in the top-level <tt>thrust::omp</tt>
+ *         namespace for easy access.
+ *
+ */
+namespace omp
+{
+
+// forward declaration of reference for pointer
+template<typename Element> class reference;
+
+/*! \cond
+ */
+
+// XXX nvcc + msvc have trouble instantiating reference below
+//     this is a workaround
+namespace detail
+{
+
+template<typename Element>
+  struct reference_msvc_workaround
+{
+  typedef thrust::system::omp::reference<Element> type;
+}; // end reference_msvc_workaround
+
+} // end detail
+
+/*! \endcond
+ */
+
+
+/*! \p pointer stores a pointer to an object allocated in memory available to the omp system.
+ *  This type provides type safety when dispatching standard algorithms on ranges resident
+ *  in omp memory.
+ *
+ *  \p pointer has pointer semantics: it may be dereferenced and manipulated with pointer arithmetic.
+ *
+ *  \p pointer can be created with the function \p omp::malloc, or by explicitly calling its constructor
+ *  with a raw pointer.
+ *
+ *  The raw pointer encapsulated by a \p pointer may be obtained by eiter its <tt>get</tt> member function
+ *  or the \p raw_pointer_cast function.
+ *
+ *  \note \p pointer is not a "smart" pointer; it is the programmer's responsibility to deallocate memory
+ *  pointed to by \p pointer.
+ *
+ *  \tparam T specifies the type of the pointee.
+ *
+ *  \see omp::malloc
+ *  \see omp::free
+ *  \see raw_pointer_cast
+ */
+template<typename T>
+  class pointer
+    : public thrust::pointer<
+               T,
+               thrust::system::omp::tag,
+               thrust::system::omp::reference<T>,
+               thrust::system::omp::pointer<T>
+             >
+{
+  /*! \cond
+   */
+
+  private:
+    typedef thrust::pointer<
+      T,
+      thrust::system::omp::tag,
+      //thrust::system::omp::reference<T>,
+      typename detail::reference_msvc_workaround<T>::type,
+      thrust::system::omp::pointer<T>
+    > super_t;
+
+  /*! \endcond
+   */
+
+  public:
+    // note that omp::pointer's member functions need __host__ __device__
+    // to interoperate with nvcc + iterators' dereference member function
+
+    /*! \p pointer's no-argument constructor initializes its encapsulated pointer to \c 0.
+     */
+    __host__ __device__
+    pointer() : super_t() {}
+
+    /*! This constructor allows construction of a <tt>pointer<const T></tt> from a <tt>T*</tt>.
+     *
+     *  \param ptr A raw pointer to copy from, presumed to point to a location in memory
+     *         accessible by the \p omp system.
+     *  \tparam OtherT \p OtherT shall be convertible to \p T.
+     */
+    template<typename OtherT>
+    __host__ __device__
+    explicit pointer(OtherT *ptr) : super_t(ptr) {}
+
+    /*! This constructor allows construction from another pointer-like object with related type.
+     *
+     *  \param other The \p OtherPointer to copy.
+     *  \tparam OtherPointer The system tag associated with \p OtherPointer shall be convertible
+     *          to \p thrust::system::omp::tag and its element type shall be convertible to \p T.
+     */
+    template<typename OtherPointer>
+    __host__ __device__
+    pointer(const OtherPointer &other,
+            typename thrust::detail::enable_if_pointer_is_convertible<
+              OtherPointer,
+              pointer
+            >::type * = 0) : super_t(other) {}
+
+    /*! Assignment operator allows assigning from another pointer-like object with related type.
+     *
+     *  \param other The other pointer-like object to assign from.
+     *  \tparam OtherPointer The system tag associated with \p OtherPointer shall be convertible
+     *          to \p thrust::system::omp::tag and its element type shall be convertible to \p T.
+     */
+    template<typename OtherPointer>
+    __host__ __device__
+    typename thrust::detail::enable_if_pointer_is_convertible<
+      OtherPointer,
+      pointer,
+      pointer &
+    >::type
+    operator=(const OtherPointer &other)
+    {
+      return super_t::operator=(other);
+    }
+}; // end pointer
+
+
+/*! \p reference is a wrapped reference to an object stored in memory available to the \p omp system.
+ *  \p reference is the type of the result of dereferencing a \p omp::pointer.
+ *
+ *  \tparam T Specifies the type of the referenced object.
+ */
+template<typename T>
+  class reference
+    : public thrust::reference<
+               T,
+               thrust::system::omp::pointer<T>,
+               thrust::system::omp::reference<T>
+             >
+{
+  /*! \cond
+   */
+
+  private:
+    typedef thrust::reference<
+      T,
+      thrust::system::omp::pointer<T>,
+      thrust::system::omp::reference<T>
+    > super_t;
+
+  /*! \endcond
+   */
+
+  public:
+    /*! \cond
+     */
+
+    typedef typename super_t::value_type value_type;
+    typedef typename super_t::pointer    pointer;
+
+    /*! \endcond
+     */
+
+    /*! This constructor initializes this \p reference to refer to an object
+     *  pointed to by the given \p pointer. After this \p reference is constructed,
+     *  it shall refer to the object pointed to by \p ptr.
+     *
+     *  \param ptr A \p pointer to copy from.
+     */
+    __host__ __device__
+    explicit reference(const pointer &ptr)
+      : super_t(ptr)
+    {}
+
+    /*! This constructor accepts a const reference to another \p reference of related type.
+     *  After this \p reference is constructed, it shall refer to the same object as \p other.
+     *
+     *  \param other A \p reference to copy from.
+     *  \tparam OtherT The element type of the other \p reference.
+     *
+     *  \note This constructor is templated primarily to allow initialization of <tt>reference<const T></tt>
+     *        from <tt>reference<T></tt>.
+     */
+    template<typename OtherT>
+    __host__ __device__
+    reference(const reference<OtherT> &other,
+              typename thrust::detail::enable_if_convertible<
+                typename reference<OtherT>::pointer,
+                pointer
+              >::type * = 0)
+      : super_t(other)
+    {}
+
+    /*! Copy assignment operator copy assigns from another \p reference of related type.
+     *
+     *  \param other The other \p reference to assign from.
+     *  \return <tt>*this</tt>
+     *  \tparam OtherT The element type of the other \p reference.
+     */
+    template<typename OtherT>
+    reference &operator=(const reference<OtherT> &other);
+
+    /*! Assignment operator assigns from a \p value_type.
+     *
+     *  \param x The \p value_type to assign from.
+     *  \return <tt>*this</tt>
+     */
+    reference &operator=(const value_type &x);
+}; // end reference
+
+/*! Exchanges the values of two objects referred to by \p reference.
+ *  \p x The first \p reference of interest.
+ *  \p y The second \p reference ot interest.
+ */
+template<typename T>
+__host__ __device__
+void swap(reference<T> x, reference<T> y);
+
+/*! Allocates an area of memory available to Thrust's <tt>omp</tt> system.
+ *  \param n Number of bytes to allocate.
+ *  \return A <tt>omp::pointer<void></tt> pointing to the beginning of the newly
+ *          allocated memory. A null <tt>omp::pointer<void></tt> is returned if
+ *          an error occurs.
+ *  \note The <tt>omp::pointer<void></tt> returned by this function must be
+ *        deallocated with \p omp::free.
+ *  \see omp::free
+ *  \see std::malloc
+ */
+inline pointer<void> malloc(std::size_t n);
+
+/*! Allocates a typed area of memory available to Thrust's <tt>omp</tt> system.
+ *  \param n Number of elements to allocate.
+ *  \return A <tt>omp::pointer<T></tt> pointing to the beginning of the newly
+ *          allocated memory. A null <tt>omp::pointer<T></tt> is returned if
+ *          an error occurs.
+ *  \note The <tt>omp::pointer<T></tt> returned by this function must be
+ *        deallocated with \p omp::free.
+ *  \see omp::free
+ *  \see std::malloc
+ */
+template<typename T>
+inline pointer<T> malloc(std::size_t n);
+
+/*! Deallocates an area of memory previously allocated by <tt>omp::malloc</tt>.
+ *  \param ptr A <tt>omp::pointer<void></tt> pointing to the beginning of an area
+ *         of memory previously allocated with <tt>omp::malloc</tt>.
+ *  \see omp::malloc
+ *  \see std::free
+ */
+inline void free(pointer<void> ptr);
+
+// XXX upon c++11
+// template<typename T> using allocator = thrust::detail::malloc_allocator<T,tag,pointer<T> >;
+
+/*! \p omp::allocator is the default allocator used by the \p omp system's containers such as
+ *  <tt>omp::vector</tt> if no user-specified allocator is provided. \p omp::allocator allocates
+ *  (deallocates) storage with \p omp::malloc (\p omp::free).
+ */
+template<typename T>
+  struct allocator
+    : thrust::detail::malloc_allocator<
+        T,
+        tag,
+        pointer<T>
+      >
+{
+  /*! The \p rebind metafunction provides the type of an \p allocator
+   *  instantiated with another type.
+   *
+   *  \tparam U The other type to use for instantiation.
+   */
+  template<typename U>
+    struct rebind
+  {
+    /*! The typedef \p other gives the type of the rebound \p allocator.
+     */
+    typedef allocator<U> other;
+  };
+
+  /*! No-argument constructor has no effect.
+   */
+  __host__ __device__
+  inline allocator() {}
+
+  /*! Copy constructor has no effect.
+   */
+  __host__ __device__
+  inline allocator(const allocator &) {}
+
+  /*! Constructor from other \p allocator has no effect.
+   */
+  template<typename U>
+  __host__ __device__
+  inline allocator(const allocator<U> &) {}
+
+  /*! Destructor has no effect.
+   */
+  __host__ __device__
+  inline ~allocator() {}
+}; // end allocator
+
+} // end omp
+
+/*! \}
+ */
+
+} // end system
+
+/*! \namespace thrust::omp
+ *  \brief \p thrust::omp is a top-level alias for thrust::system::omp.
+ */
+namespace omp
+{
+
+using thrust::system::omp::pointer;
+using thrust::system::omp::reference;
+using thrust::system::omp::malloc;
+using thrust::system::omp::free;
+using thrust::system::omp::allocator;
+
+} // end omp
+
+} // end thrust
+
+#include <thrust/system/omp/detail/memory.inl>
+
diff --git a/compat/thrust/system/omp/vector.h b/compat/thrust/system/omp/vector.h
new file mode 100644
index 0000000000..5f45a9169b
--- /dev/null
+++ b/compat/thrust/system/omp/vector.h
@@ -0,0 +1,149 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+/*! \file thrust/system/omp/vector.h
+ *  \brief A dynamically-sizable array of elements which reside in memory available to
+ *         Thrust's OpenMP system.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/omp/memory.h>
+#include <thrust/detail/vector_base.h>
+#include <vector>
+
+namespace thrust
+{
+
+// forward declaration of host_vector
+// XXX why is this here? it doesn't seem necessary for anything below
+template<typename T, typename Allocator> class host_vector;
+
+namespace system
+{
+namespace omp
+{
+
+// XXX upon c++11
+// template<typename T, typename Allocator = allocator<T> > using vector = thrust::detail::vector_base<T,Allocator>;
+
+/*! \p omp::vector is a container that supports random access to elements,
+ *  constant time removal of elements at the end, and linear time insertion
+ *  and removal of elements at the beginning or in the middle. The number of
+ *  elements in a \p omp::vector may vary dynamically; memory management is
+ *  automatic. The elements contained in an \p omp::vector reside in memory
+ *  available to the \p omp system.
+ *
+ *  \tparam T The element type of the \p omp::vector.
+ *  \tparam Allocator The allocator type of the \p omp::vector. Defaults to \p omp::allocator.
+ *
+ *  \see http://www.sgi.com/tech/stl/Vector.html
+ *  \see host_vector For the documentation of the complete interface which is
+ *                   shared by \p omp::vector
+ *  \see device_vector
+ */
+template<typename T, typename Allocator = allocator<T> >
+  class vector
+    : public thrust::detail::vector_base<T,Allocator>
+{
+  /*! \cond
+   */
+  private:
+    typedef thrust::detail::vector_base<T,Allocator> super_t;
+  /*! \endcond
+   */
+
+  public:
+    
+  /*! \cond
+   */
+    typedef typename super_t::size_type  size_type;
+    typedef typename super_t::value_type value_type;
+  /*! \endcond
+   */
+
+    /*! This constructor creates an empty \p omp::vector.
+     */
+    vector();
+
+    /*! This constructor creates a \p omp::vector with \p n default-constructed elements.
+     *  \param n The size of the \p omp::vector to create.
+     */
+    explicit vector(size_type n);
+
+    /*! This constructor creates a \p omp::vector with \p n copies of \p value.
+     *  \param n The size of the \p omp::vector to create.
+     *  \param value An element to copy.
+     */
+    explicit vector(size_type n, const value_type &value);
+
+    /*! Copy constructor copies from another \p omp::vector.
+     *  \param x The other \p omp::vector to copy.
+     */
+    vector(const vector &x);
+
+    /*! This constructor copies from another Thrust vector-like object.
+     *  \param x The other object to copy from.
+     */
+    template<typename OtherT, typename OtherAllocator>
+    vector(const thrust::detail::vector_base<OtherT,OtherAllocator> &x);
+
+    /*! This constructor copies from a \c std::vector.
+     *  \param x The \c std::vector to copy from.
+     */
+    template<typename OtherT, typename OtherAllocator>
+    vector(const std::vector<OtherT,OtherAllocator> &x);
+
+    /*! This constructor creates an \p omp::vector by copying from a range.
+     *  \param first The beginning of the range.
+     *  \param last The end of the range.
+     */
+    template<typename InputIterator>
+    vector(InputIterator first, InputIterator last);
+
+    // XXX vector_base should take a Derived type so we don't have to define these superfluous assigns
+
+    /*! Assignment operator assigns from a \c std::vector.
+     *  \param x The \c std::vector to assign from.
+     *  \return <tt>*this</tt>
+     */
+    template<typename OtherT, typename OtherAllocator>
+    vector &operator=(const std::vector<OtherT,OtherAllocator> &x);
+
+    /*! Assignment operator assigns from another Thrust vector-like object.
+     *  \param x The other object to assign from.
+     *  \return <tt>*this</tt>
+     */
+    template<typename OtherT, typename OtherAllocator>
+    vector &operator=(const thrust::detail::vector_base<OtherT,OtherAllocator> &x);
+}; // end vector
+
+} // end omp
+} // end system
+
+// alias system::omp names at top-level
+namespace omp
+{
+
+using thrust::system::omp::vector;
+
+} // end omp
+
+} // end thrust
+
+#include <thrust/system/omp/detail/vector.inl>
+
diff --git a/compat/thrust/system/system_error.h b/compat/thrust/system/system_error.h
new file mode 100644
index 0000000000..6f94b61e0a
--- /dev/null
+++ b/compat/thrust/system/system_error.h
@@ -0,0 +1,179 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file system/system_error.h
+ *  \brief An exception object used to report error conditions that have an
+ *         associated error code
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <stdexcept>
+#include <string>
+
+#include <thrust/system/error_code.h>
+
+namespace thrust
+{
+
+namespace system
+{
+
+// [19.5.5] Class system_error
+
+// [19.5.5.1] Class system_error overview
+
+/*! \addtogroup system_diagnostics System Diagnostics
+ *  \ingroup system
+ *  \{
+ */
+
+/*! \brief The class \p system_error describes an exception object used to report error
+ *  conditions that have an associated \p error_code. Such error conditions typically
+ *  originate from the operating system or other low-level application program interfaces.
+ *
+ *  Thrust uses \p system_error to report the error codes returned from device backends
+ *  such as the CUDA runtime.
+ *
+ *  The following code listing demonstrates how to catch a \p system_error to recover
+ *  from an error.
+ *
+ *  \code
+ *
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/system.h>
+ *  #include <thrust/sort.h>
+ *
+ *  void terminate_gracefully(void)
+ *  {
+ *    // application-specific termination code here
+ *    ...
+ *  }
+ *
+ *  int main(void)
+ *  {
+ *    try
+ *    {
+ *      thrust::device_vector<float> vec;
+ *      thrust::sort(vec.begin(), vec.end());
+ *    }
+ *    catch(thrust::system_error e)
+ *    {
+ *      std::cerr << "Error inside sort: " << e.what() << std::endl;
+ *      terminate_gracefully();
+ *    }
+ *
+ *    return 0;
+ *  }
+ *
+ *  \endcode
+ *
+ *  \note If an error represents an out-of-memory condition, implementations are encouraged
+ *  to throw an exception object of type \p std::bad_alloc rather than \p system_error.
+ */
+class system_error
+  : public std::runtime_error
+{
+  public:
+    // [19.5.5.2] Class system_error members
+    
+    /*! Constructs an object of class \p system_error.
+     *  \param ec The value returned by \p code().
+     *  \param what_arg A string to include in the result returned by \p what().
+     *  \post <tt>code() == ec</tt>.
+     *  \post <tt>std::string(what()).find(what_arg) != string::npos</tt>.
+     */
+    inline system_error(error_code ec, const std::string &what_arg);
+
+    /*! Constructs an object of class \p system_error.
+     *  \param ec The value returned by \p code().
+     *  \param what_arg A string to include in the result returned by \p what().
+     *  \post <tt>code() == ec</tt>.
+     *  \post <tt>std::string(what()).find(what_arg) != string::npos</tt>.
+     */
+    inline system_error(error_code ec, const char *what_arg);
+
+    /*! Constructs an object of class \p system_error.
+     *  \param ec The value returned by \p code().
+     *  \post <tt>code() == ec</tt>.
+     */
+    inline system_error(error_code ec);
+
+    /*! Constructs an object of class \p system_error.
+     *  \param ev The error value used to create an \p error_code.
+     *  \param ecat The \p error_category used to create an \p error_code.
+     *  \param what_arg A string to include in the result returned by \p what().
+     *  \post <tt>code() == error_code(ev, ecat)</tt>.
+     *  \post <tt>std::string(what()).find(what_arg) != string::npos</tt>.
+     */
+    inline system_error(int ev, const error_category &ecat, const std::string &what_arg);
+
+    /*! Constructs an object of class \p system_error.
+     *  \param ev The error value used to create an \p error_code.
+     *  \param ecat The \p error_category used to create an \p error_code.
+     *  \param what_arg A string to include in the result returned by \p what().
+     *  \post <tt>code() == error_code(ev, ecat)</tt>.
+     *  \post <tt>std::string(what()).find(what_arg) != string::npos</tt>.
+     */
+    inline system_error(int ev, const error_category &ecat, const char *what_arg);
+
+    /*! Constructs an object of class \p system_error.
+     *  \param ev The error value used to create an \p error_code.
+     *  \param ecat The \p error_category used to create an \p error_code.
+     *  \post <tt>code() == error_code(ev, ecat)</tt>.
+     */
+    inline system_error(int ev, const error_category &ecat);
+
+    /*! Destructor does not throw.
+     */
+    inline virtual ~system_error(void) throw () {};
+    
+    /*! Returns an object encoding the error.
+     *  \return <tt>ec</tt> or <tt>error_code(ev, ecat)</tt>, from the
+     *          constructor, as appropriate.
+     */
+    inline const error_code &code(void) const throw();
+
+    /*! Returns a human-readable string indicating the nature of the error.
+     *  \return a string incorporating <tt>code().message()</tt> and the
+     *          arguments supplied in the constructor.
+     */
+    inline const char *what(void) const throw();
+
+    /*! \cond
+     */
+  private:
+    error_code          m_error_code;
+    mutable std::string m_what;
+
+    /*! \endcond
+     */
+}; // end system_error
+
+} // end system
+
+/*! \} // end system_diagnostics
+ */
+
+// import names into thrust::
+using system::system_error;
+
+} // end thrust
+
+#include <thrust/system/detail/system_error.inl>
+
diff --git a/compat/thrust/system/tbb/detail/adjacent_difference.h b/compat/thrust/system/tbb/detail/adjacent_difference.h
new file mode 100644
index 0000000000..37c9adc2c9
--- /dev/null
+++ b/compat/thrust/system/tbb/detail/adjacent_difference.h
@@ -0,0 +1,50 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/tbb/detail/execution_policy.h>
+#include <thrust/system/detail/generic/adjacent_difference.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace tbb
+{
+namespace detail
+{
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator,
+         typename BinaryFunction>
+  OutputIterator adjacent_difference(execution_policy<DerivedPolicy> &exec,
+                                     InputIterator first,
+                                     InputIterator last,
+                                     OutputIterator result,
+                                     BinaryFunction binary_op)
+{
+  // tbb prefers generic::adjacent_difference to cpp::adjacent_difference
+  return thrust::system::detail::generic::adjacent_difference(exec, first, last, result, binary_op);
+} // end adjacent_difference()
+
+} // end detail
+} // end tbb
+} // end system
+} // end thrust
+
diff --git a/compat/thrust/system/tbb/detail/assign_value.h b/compat/thrust/system/tbb/detail/assign_value.h
new file mode 100644
index 0000000000..eda3b977b8
--- /dev/null
+++ b/compat/thrust/system/tbb/detail/assign_value.h
@@ -0,0 +1,23 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system inherits assign_value
+#include <thrust/system/cpp/detail/assign_value.h>
+
diff --git a/compat/thrust/system/tbb/detail/binary_search.h b/compat/thrust/system/tbb/detail/binary_search.h
new file mode 100644
index 0000000000..8dec989771
--- /dev/null
+++ b/compat/thrust/system/tbb/detail/binary_search.h
@@ -0,0 +1,23 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system inherits binary_search
+#include <thrust/system/cpp/detail/binary_search.h>
+
diff --git a/compat/thrust/system/tbb/detail/copy.h b/compat/thrust/system/tbb/detail/copy.h
new file mode 100644
index 0000000000..7604e6f4ad
--- /dev/null
+++ b/compat/thrust/system/tbb/detail/copy.h
@@ -0,0 +1,57 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in ctbbliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/tbb/detail/execution_policy.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace tbb
+{
+namespace detail
+{
+
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator>
+OutputIterator copy(execution_policy<DerivedPolicy> &exec,
+                    InputIterator first,
+                    InputIterator last,
+                    OutputIterator result);
+
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename Size,
+         typename OutputIterator>
+OutputIterator copy_n(execution_policy<DerivedPolicy> &exec,
+                      InputIterator first,
+                      Size n,
+                      OutputIterator result);
+
+
+} // end namespace detail
+} // end namespace tbb
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/tbb/detail/copy.inl>
+
diff --git a/compat/thrust/system/tbb/detail/copy.inl b/compat/thrust/system/tbb/detail/copy.inl
new file mode 100644
index 0000000000..6d354d0b18
--- /dev/null
+++ b/compat/thrust/system/tbb/detail/copy.inl
@@ -0,0 +1,134 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in ctbbliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/tbb/detail/copy.h>
+#include <thrust/system/detail/generic/copy.h>
+#include <thrust/detail/type_traits/minimum_type.h>
+#include <thrust/system/cpp/detail/copy.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace tbb
+{
+namespace detail
+{
+namespace dispatch
+{
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator>
+  OutputIterator copy(execution_policy<DerivedPolicy> &exec,
+                      InputIterator first,
+                      InputIterator last,
+                      OutputIterator result,
+                      thrust::incrementable_traversal_tag)
+{
+  return thrust::system::cpp::detail::copy(exec, first, last, result);
+} // end copy()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator>
+  OutputIterator copy(execution_policy<DerivedPolicy> &exec,
+                      InputIterator first,
+                      InputIterator last,
+                      OutputIterator result,
+                      thrust::random_access_traversal_tag)
+{
+  return thrust::system::detail::generic::copy(exec, first, last, result);
+} // end copy()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename Size,
+         typename OutputIterator>
+  OutputIterator copy_n(execution_policy<DerivedPolicy> &exec,
+                        InputIterator first,
+                        Size n,
+                        OutputIterator result,
+                        thrust::incrementable_traversal_tag)
+{
+  return thrust::system::cpp::detail::copy_n(exec, first, n, result);
+} // end copy_n()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename Size,
+         typename OutputIterator>
+  OutputIterator copy_n(execution_policy<DerivedPolicy> &exec,
+                        InputIterator first,
+                        Size n,
+                        OutputIterator result,
+                        thrust::random_access_traversal_tag)
+{
+  return thrust::system::detail::generic::copy_n(exec, first, n, result);
+} // end copy_n()
+
+} // end dispatch
+
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator>
+OutputIterator copy(execution_policy<DerivedPolicy> &exec,
+                    InputIterator first,
+                    InputIterator last,
+                    OutputIterator result)
+{
+  typedef typename thrust::iterator_traversal<InputIterator>::type  traversal1;
+  typedef typename thrust::iterator_traversal<OutputIterator>::type traversal2;
+  
+  typedef typename thrust::detail::minimum_type<traversal1,traversal2>::type traversal;
+
+  // dispatch on minimum traversal
+  return thrust::system::tbb::detail::dispatch::copy(exec,first,last,result,traversal());
+} // end copy()
+
+
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename Size,
+         typename OutputIterator>
+OutputIterator copy_n(execution_policy<DerivedPolicy> &exec,
+                      InputIterator first,
+                      Size n,
+                      OutputIterator result)
+{
+  typedef typename thrust::iterator_traversal<InputIterator>::type  traversal1;
+  typedef typename thrust::iterator_traversal<OutputIterator>::type traversal2;
+  
+  typedef typename thrust::detail::minimum_type<traversal1,traversal2>::type traversal;
+
+  // dispatch on minimum traversal
+  return thrust::system::tbb::detail::dispatch::copy_n(exec,first,n,result,traversal());
+} // end copy_n()
+
+
+} // end namespace detail
+} // end namespace tbb
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/tbb/detail/copy_if.h b/compat/thrust/system/tbb/detail/copy_if.h
new file mode 100644
index 0000000000..ffbd4f8f46
--- /dev/null
+++ b/compat/thrust/system/tbb/detail/copy_if.h
@@ -0,0 +1,50 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/tbb/detail/execution_policy.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace tbb
+{
+namespace detail
+{
+
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename Predicate>
+  OutputIterator copy_if(tag,
+                         InputIterator1 first,
+                         InputIterator1 last,
+                         InputIterator2 stencil,
+                         OutputIterator result,
+                         Predicate pred);
+
+
+} // end detail
+} // end tbb
+} // end system
+} // end thrust
+
+#include <thrust/system/tbb/detail/copy_if.inl>
+
diff --git a/compat/thrust/system/tbb/detail/copy_if.inl b/compat/thrust/system/tbb/detail/copy_if.inl
new file mode 100644
index 0000000000..4353b3b2f7
--- /dev/null
+++ b/compat/thrust/system/tbb/detail/copy_if.inl
@@ -0,0 +1,131 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/function.h>
+#include <thrust/system/tbb/detail/copy_if.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/distance.h>
+#include <tbb/blocked_range.h>
+#include <tbb/parallel_scan.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace tbb
+{
+namespace detail
+{
+namespace copy_if_detail
+{
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename Predicate,
+         typename Size>
+struct body
+{
+
+  InputIterator1 first;
+  InputIterator2 stencil;
+  OutputIterator result;
+  thrust::detail::host_function<Predicate,bool> pred;
+  Size sum;
+
+  body(InputIterator1 first, InputIterator2 stencil, OutputIterator result, Predicate pred)
+    : first(first), stencil(stencil), result(result), pred(pred), sum(0)
+  {}
+
+  body(body& b, ::tbb::split)
+    : first(b.first), stencil(b.stencil), result(b.result), pred(b.pred), sum(0)
+  {}
+
+  void operator()(const ::tbb::blocked_range<Size>& r, ::tbb::pre_scan_tag)
+  {
+    InputIterator2 iter = stencil + r.begin();
+
+    for (Size i = r.begin(); i != r.end(); ++i, ++iter)
+    {
+      if (pred(*iter))
+        ++sum;
+    }
+  }
+  
+  void operator()(const ::tbb::blocked_range<Size>& r, ::tbb::final_scan_tag)
+  {
+    InputIterator1  iter1 = first   + r.begin();
+    InputIterator2  iter2 = stencil + r.begin();
+    OutputIterator  iter3 = result  + sum;
+      
+    for (Size i = r.begin(); i != r.end(); ++i, ++iter1, ++iter2)
+    {
+      if (pred(*iter2))
+      {
+        *iter3 = *iter1;
+        ++sum;
+        ++iter3;
+      }
+    }
+  }
+
+  void reverse_join(body& b)
+  {
+    sum = b.sum + sum;
+  } 
+
+  void assign(body& b)
+  {
+    sum = b.sum;
+  } 
+}; // end body
+
+} // end copy_if_detail
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename Predicate>
+  OutputIterator copy_if(tag,
+                         InputIterator1 first,
+                         InputIterator1 last,
+                         InputIterator2 stencil,
+                         OutputIterator result,
+                         Predicate pred)
+{
+  typedef typename thrust::iterator_difference<InputIterator1>::type Size; 
+  typedef typename copy_if_detail::body<InputIterator1,InputIterator2,OutputIterator,Predicate,Size> Body;
+  
+  Size n = thrust::distance(first, last);
+
+  if (n != 0)
+  {
+    Body body(first, stencil, result, pred);
+    ::tbb::parallel_scan(::tbb::blocked_range<Size>(0,n), body);
+    thrust::advance(result, body.sum);
+  }
+
+  return result;
+} // end copy_if()
+
+} // end detail
+} // end tbb
+} // end system
+} // end thrust
+
diff --git a/compat/thrust/system/tbb/detail/count.h b/compat/thrust/system/tbb/detail/count.h
new file mode 100644
index 0000000000..da31ee8700
--- /dev/null
+++ b/compat/thrust/system/tbb/detail/count.h
@@ -0,0 +1,23 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system inherits count
+#include <thrust/system/cpp/detail/count.h>
+
diff --git a/compat/thrust/system/tbb/detail/equal.h b/compat/thrust/system/tbb/detail/equal.h
new file mode 100644
index 0000000000..74e55183d9
--- /dev/null
+++ b/compat/thrust/system/tbb/detail/equal.h
@@ -0,0 +1,23 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system inherits equal
+#include <thrust/system/cpp/detail/equal.h>
+
diff --git a/compat/thrust/system/tbb/detail/execution_policy.h b/compat/thrust/system/tbb/detail/execution_policy.h
new file mode 100644
index 0000000000..167d1dc4ce
--- /dev/null
+++ b/compat/thrust/system/tbb/detail/execution_policy.h
@@ -0,0 +1,86 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in ctbbliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/cpp/detail/execution_policy.h>
+#include <thrust/iterator/detail/any_system_tag.h>
+#include <thrust/detail/type_traits.h>
+
+namespace thrust
+{
+namespace system
+{
+// put the canonical tag in the same ns as the backend's entry points
+namespace tbb
+{
+namespace detail
+{
+
+// this awkward sequence of definitions arise
+// from the desire both for tag to derive
+// from execution_policy and for execution_policy
+// to convert to tag (when execution_policy is not
+// an ancestor of tag)
+
+// forward declaration of tag
+struct tag;
+
+// forward declaration of execution_policy
+template<typename> struct execution_policy;
+
+// specialize execution_policy for tag
+template<>
+  struct execution_policy<tag>
+    : thrust::system::cpp::detail::execution_policy<tag>
+{};
+
+// tag's definition comes before the
+// generic definition of execution_policy
+struct tag : execution_policy<tag> {};
+
+// allow conversion to tag when it is not a successor
+template<typename Derived>
+  struct execution_policy
+    : thrust::system::cpp::detail::execution_policy<Derived>
+{
+  // allow conversion to tag
+  inline operator tag () const
+  {
+    return tag();
+  }
+};
+
+} // end detail
+
+// alias execution_policy and tag here
+using thrust::system::tbb::detail::execution_policy;
+using thrust::system::tbb::detail::tag;
+
+} // end tbb
+} // end system
+
+// alias items at top-level
+namespace tbb
+{
+
+using thrust::system::tbb::execution_policy;
+using thrust::system::tbb::tag;
+
+} // end tbb
+} // end thrust
+
diff --git a/compat/thrust/system/tbb/detail/extrema.h b/compat/thrust/system/tbb/detail/extrema.h
new file mode 100644
index 0000000000..4715a8948b
--- /dev/null
+++ b/compat/thrust/system/tbb/detail/extrema.h
@@ -0,0 +1,67 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in ctbbliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/tbb/detail/execution_policy.h>
+#include <thrust/system/detail/generic/extrema.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace tbb
+{
+namespace detail
+{
+
+template <typename DerivedPolicy, typename ForwardIterator, typename BinaryPredicate>
+ForwardIterator max_element(execution_policy<DerivedPolicy> &exec,
+                            ForwardIterator first, 
+                            ForwardIterator last,
+                            BinaryPredicate comp)
+{
+  // tbb prefers generic::max_element to cpp::max_element
+  return thrust::system::detail::generic::max_element(exec, first, last, comp);
+} // end max_element()
+
+template <typename DerivedPolicy, typename ForwardIterator, typename BinaryPredicate>
+ForwardIterator min_element(execution_policy<DerivedPolicy> &exec,
+                            ForwardIterator first, 
+                            ForwardIterator last,
+                            BinaryPredicate comp)
+{
+  // tbb prefers generic::min_element to cpp::min_element
+  return thrust::system::detail::generic::min_element(exec, first, last, comp);
+} // end min_element()
+
+template <typename DerivedPolicy, typename ForwardIterator, typename BinaryPredicate>
+thrust::pair<ForwardIterator,ForwardIterator> minmax_element(execution_policy<DerivedPolicy> &exec,
+                                                             ForwardIterator first, 
+                                                             ForwardIterator last,
+                                                             BinaryPredicate comp)
+{
+  // tbb prefers generic::minmax_element to cpp::minmax_element
+  return thrust::system::detail::generic::minmax_element(exec, first, last, comp);
+} // end minmax_element()
+
+} // end detail
+} // end tbb
+} // end system
+} // end thrust
+
+
diff --git a/compat/thrust/system/tbb/detail/fill.h b/compat/thrust/system/tbb/detail/fill.h
new file mode 100644
index 0000000000..5219e1c7c5
--- /dev/null
+++ b/compat/thrust/system/tbb/detail/fill.h
@@ -0,0 +1,23 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system inherits fill
+#include <thrust/system/cpp/detail/fill.h>
+
diff --git a/compat/thrust/system/tbb/detail/find.h b/compat/thrust/system/tbb/detail/find.h
new file mode 100644
index 0000000000..d351454c16
--- /dev/null
+++ b/compat/thrust/system/tbb/detail/find.h
@@ -0,0 +1,46 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/detail/generic/find.h>
+#include <thrust/system/tbb/detail/execution_policy.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace tbb
+{
+namespace detail
+{
+
+template <typename DerivedPolicy, typename InputIterator, typename Predicate>
+InputIterator find_if(execution_policy<DerivedPolicy> &exec,
+                      InputIterator first,
+                      InputIterator last,
+                      Predicate pred)
+{
+  // tbb prefers generic::find_if to cpp::find_if
+  return thrust::system::detail::generic::find_if(exec, first, last, pred);
+}
+
+} // end namespace detail
+} // end namespace tbb
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/tbb/detail/for_each.h b/compat/thrust/system/tbb/detail/for_each.h
new file mode 100644
index 0000000000..573bb819a2
--- /dev/null
+++ b/compat/thrust/system/tbb/detail/for_each.h
@@ -0,0 +1,54 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in ctbbliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/tbb/detail/execution_policy.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace tbb
+{
+namespace detail
+{
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator,
+         typename UnaryFunction>
+  RandomAccessIterator for_each(execution_policy<DerivedPolicy> &exec,
+                                RandomAccessIterator first,
+                                RandomAccessIterator last,
+                                UnaryFunction f);
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator,
+         typename Size,
+         typename UnaryFunction>
+  RandomAccessIterator for_each_n(execution_policy<DerivedPolicy> &exec,
+                                  RandomAccessIterator first,
+                                  Size n,
+                                  UnaryFunction f);
+
+} // end namespace detail
+} // end namespace tbb
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/tbb/detail/for_each.inl>
+
diff --git a/compat/thrust/system/tbb/detail/for_each.inl b/compat/thrust/system/tbb/detail/for_each.inl
new file mode 100644
index 0000000000..b09c7be881
--- /dev/null
+++ b/compat/thrust/system/tbb/detail/for_each.inl
@@ -0,0 +1,100 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in ctbbliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/static_assert.h>
+#include <thrust/distance.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/distance.h>
+#include <thrust/system/detail/internal/scalar/for_each.h>
+#include <tbb/blocked_range.h>
+#include <tbb/parallel_for.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace tbb
+{
+namespace detail
+{
+namespace for_each_detail
+{
+
+template<typename RandomAccessIterator,
+         typename Size,
+         typename UnaryFunction>
+  struct body
+{
+  RandomAccessIterator m_first;
+  UnaryFunction m_f;
+
+  body(RandomAccessIterator first, UnaryFunction f)
+    : m_first(first), m_f(f)
+  {}
+
+  void operator()(const ::tbb::blocked_range<Size> &r) const
+  {
+    // we assume that blocked_range specifies a contiguous range of integers
+    thrust::system::detail::internal::scalar::for_each_n(m_first + r.begin(), r.size(), m_f);
+  } // end operator()()
+}; // end body
+
+
+template<typename Size, typename RandomAccessIterator, typename UnaryFunction>
+  body<RandomAccessIterator,Size,UnaryFunction>
+    make_body(RandomAccessIterator first, UnaryFunction f)
+{
+  return body<RandomAccessIterator,Size,UnaryFunction>(first, f);
+} // end make_body()
+
+
+} // end for_each_detail
+
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator,
+         typename Size,
+         typename UnaryFunction>
+RandomAccessIterator for_each_n(execution_policy<DerivedPolicy> &,
+                                RandomAccessIterator first,
+                                Size n,
+                                UnaryFunction f)
+{
+  ::tbb::parallel_for(::tbb::blocked_range<Size>(0,n), for_each_detail::make_body<Size>(first,f));
+
+  // return the end of the range
+  return first + n;
+} // end for_each_n 
+
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator,
+         typename UnaryFunction>
+  RandomAccessIterator for_each(execution_policy<DerivedPolicy> &s,
+                                RandomAccessIterator first,
+                                RandomAccessIterator last,
+                                UnaryFunction f)
+{
+  return tbb::detail::for_each_n(s, first, thrust::distance(first,last), f);
+} // end for_each()
+
+
+} // end namespace detail
+} // end namespace tbb
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/tbb/detail/gather.h b/compat/thrust/system/tbb/detail/gather.h
new file mode 100644
index 0000000000..dfb7d7fc2d
--- /dev/null
+++ b/compat/thrust/system/tbb/detail/gather.h
@@ -0,0 +1,23 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system inherits gather
+#include <thrust/system/cpp/detail/gather.h>
+
diff --git a/compat/thrust/system/tbb/detail/generate.h b/compat/thrust/system/tbb/detail/generate.h
new file mode 100644
index 0000000000..0cb33b9336
--- /dev/null
+++ b/compat/thrust/system/tbb/detail/generate.h
@@ -0,0 +1,23 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system inherits generate
+#include <thrust/system/cpp/detail/generate.h>
+
diff --git a/compat/thrust/system/tbb/detail/get_value.h b/compat/thrust/system/tbb/detail/get_value.h
new file mode 100644
index 0000000000..e376e65749
--- /dev/null
+++ b/compat/thrust/system/tbb/detail/get_value.h
@@ -0,0 +1,23 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system inherits get_value
+#include <thrust/system/cpp/detail/get_value.h>
+
diff --git a/compat/thrust/system/tbb/detail/inner_product.h b/compat/thrust/system/tbb/detail/inner_product.h
new file mode 100644
index 0000000000..351421a577
--- /dev/null
+++ b/compat/thrust/system/tbb/detail/inner_product.h
@@ -0,0 +1,23 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system inherits inner_product
+#include <thrust/system/cpp/detail/inner_product.h>
+
diff --git a/compat/thrust/system/tbb/detail/iter_swap.h b/compat/thrust/system/tbb/detail/iter_swap.h
new file mode 100644
index 0000000000..16176ec69b
--- /dev/null
+++ b/compat/thrust/system/tbb/detail/iter_swap.h
@@ -0,0 +1,23 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system inherits iter_swap
+#include <thrust/system/cpp/detail/iter_swap.h>
+
diff --git a/compat/thrust/system/tbb/detail/logical.h b/compat/thrust/system/tbb/detail/logical.h
new file mode 100644
index 0000000000..b2a80de70f
--- /dev/null
+++ b/compat/thrust/system/tbb/detail/logical.h
@@ -0,0 +1,23 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system inherits logical
+#include <thrust/system/cpp/detail/logical.h>
+
diff --git a/compat/thrust/system/tbb/detail/malloc_and_free.h b/compat/thrust/system/tbb/detail/malloc_and_free.h
new file mode 100644
index 0000000000..811a552a4f
--- /dev/null
+++ b/compat/thrust/system/tbb/detail/malloc_and_free.h
@@ -0,0 +1,23 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system inherits malloc and free
+#include <thrust/system/cpp/detail/malloc_and_free.h>
+
diff --git a/compat/thrust/system/tbb/detail/memory.inl b/compat/thrust/system/tbb/detail/memory.inl
new file mode 100644
index 0000000000..420a8a14b4
--- /dev/null
+++ b/compat/thrust/system/tbb/detail/memory.inl
@@ -0,0 +1,110 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in ctbbliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/system/cpp/detail/execution_policy.h>
+#include <thrust/system/tbb/memory.h>
+#include <thrust/system/cpp/memory.h>
+#include <limits>
+
+namespace thrust
+{
+namespace system
+{
+namespace tbb
+{
+
+
+template<typename T>
+  template<typename OtherT>
+    reference<T> &
+      reference<T>
+        ::operator=(const reference<OtherT> &other)
+{
+  return super_t::operator=(other);
+} // end reference::operator=()
+
+template<typename T>
+  reference<T> &
+    reference<T>
+      ::operator=(const value_type &x)
+{
+  return super_t::operator=(x);
+} // end reference::operator=()
+
+template<typename T>
+__host__ __device__
+void swap(reference<T> a, reference<T> b)
+{
+  a.swap(b);
+} // end swap()
+
+namespace detail
+{
+
+// XXX circular #inclusion problems cause the compiler to believe that cpp::malloc
+//     is not defined
+//     WAR the problem by using adl to call cpp::malloc, which requires it to depend
+//     on a template parameter
+template<typename Tag>
+  pointer<void> malloc_workaround(Tag t, std::size_t n)
+{
+  return pointer<void>(malloc(t, n));
+} // end malloc_workaround()
+
+// XXX circular #inclusion problems cause the compiler to believe that cpp::free
+//     is not defined
+//     WAR the problem by using adl to call cpp::free, which requires it to depend
+//     on a template parameter
+template<typename Tag>
+  void free_workaround(Tag t, pointer<void> ptr)
+{
+  free(t, ptr.get());
+} // end free_workaround()
+
+} // end detail
+
+inline pointer<void> malloc(std::size_t n)
+{
+  // XXX this is how we'd like to implement this function,
+  //     if not for circular #inclusion problems:
+  //
+  // return pointer<void>(thrust::system::cpp::malloc(n))
+  //
+  return detail::malloc_workaround(cpp::tag(), n);
+} // end malloc()
+
+template<typename T>
+pointer<T> malloc(std::size_t n)
+{
+  pointer<void> raw_ptr = thrust::system::tbb::malloc(sizeof(T) * n);
+  return pointer<T>(reinterpret_cast<T*>(raw_ptr.get()));
+} // end malloc()
+
+inline void free(pointer<void> ptr)
+{
+  // XXX this is how we'd like to implement this function,
+  //     if not for circular #inclusion problems:
+  //
+  // thrust::system::cpp::free(ptr)
+  //
+  detail::free_workaround(cpp::tag(), ptr);
+} // end free()
+
+} // end tbb
+} // end system
+} // end thrust
+
diff --git a/compat/thrust/system/tbb/detail/merge.h b/compat/thrust/system/tbb/detail/merge.h
new file mode 100644
index 0000000000..7b203ec327
--- /dev/null
+++ b/compat/thrust/system/tbb/detail/merge.h
@@ -0,0 +1,70 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/tbb/detail/execution_policy.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace tbb
+{
+namespace detail
+{
+
+template<typename ExecutionPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename StrictWeakOrdering>
+OutputIterator merge(execution_policy<ExecutionPolicy> &exec,
+                     InputIterator1 first1,
+                     InputIterator1 last1,
+                     InputIterator2 first2,
+                     InputIterator2 last2,
+                     OutputIterator result,
+                     StrictWeakOrdering comp);
+
+template <typename ExecutionPolicy,
+          typename InputIterator1,
+          typename InputIterator2,
+          typename InputIterator3,
+          typename InputIterator4,
+          typename OutputIterator1,
+          typename OutputIterator2,
+          typename StrictWeakOrdering>
+thrust::pair<OutputIterator1,OutputIterator2>
+  merge_by_key(execution_policy<ExecutionPolicy> &exec,
+               InputIterator1 keys_first1,
+               InputIterator1 keys_last1,
+               InputIterator2 keys_first2,
+               InputIterator2 keys_last2,
+               InputIterator3 values_first3,
+               InputIterator4 values_first4,
+               OutputIterator1 keys_result,
+               OutputIterator2 values_result,
+               StrictWeakOrdering comp);
+
+} // end detail
+} // end tbb
+} // end system
+} // end thrust
+
+#include <thrust/system/tbb/detail/merge.inl>
+
diff --git a/compat/thrust/system/tbb/detail/merge.inl b/compat/thrust/system/tbb/detail/merge.inl
new file mode 100644
index 0000000000..cc902af85b
--- /dev/null
+++ b/compat/thrust/system/tbb/detail/merge.inl
@@ -0,0 +1,285 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/detail/temporary_array.h>
+#include <thrust/system/tbb/detail/execution_policy.h>
+#include <thrust/system/detail/internal/scalar/merge.h>
+#include <thrust/system/detail/internal/scalar/binary_search.h>
+#include <tbb/parallel_for.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace tbb
+{
+namespace detail
+{
+namespace merge_detail
+{
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename StrictWeakOrdering>
+struct range
+{
+  InputIterator1 first1, last1;
+  InputIterator2 first2, last2;
+  OutputIterator result;
+  StrictWeakOrdering comp;
+  size_t grain_size;
+
+  range(InputIterator1 first1, InputIterator1 last1,
+        InputIterator2 first2, InputIterator2 last2,
+        OutputIterator result,
+        StrictWeakOrdering comp,
+        size_t grain_size = 1024)
+    : first1(first1), last1(last1),
+      first2(first2), last2(last2),
+      result(result), comp(comp), grain_size(grain_size)
+  {}
+  
+  range(range& r, ::tbb::split)
+    : first1(r.first1), last1(r.last1),
+      first2(r.first2), last2(r.last2),
+      result(r.result), comp(r.comp), grain_size(r.grain_size)
+  {
+    // we can assume n1 and n2 are not both 0
+    size_t n1 = thrust::distance(first1, last1);
+    size_t n2 = thrust::distance(first2, last2);
+
+    InputIterator1 mid1 = first1;
+    InputIterator2 mid2 = first2;
+
+    if (n1 > n2)
+    {
+      mid1 += n1 / 2;
+      mid2 = thrust::system::detail::internal::scalar::lower_bound(first2, last2, raw_reference_cast(*mid1), comp);
+    }
+    else
+    {
+      mid2 += n2 / 2;
+      mid1 = thrust::system::detail::internal::scalar::upper_bound(first1, last1, raw_reference_cast(*mid2), comp);
+    }
+    
+    // set first range to [first1, mid1), [first2, mid2), result
+    r.last1 = mid1;
+    r.last2 = mid2;
+
+    // set second range to [mid1, last1), [mid2, last2), result + (mid1 - first1) + (mid2 - first2)
+    first1 = mid1;
+    first2 = mid2;
+    result += thrust::distance(r.first1, mid1) + thrust::distance(r.first2, mid2);
+  }
+
+  bool empty(void) const
+  {
+    return (first1 == last1) && (first2 == last2);
+  }
+
+  bool is_divisible(void) const
+  {
+    return static_cast<size_t>(thrust::distance(first1, last1) + thrust::distance(first2, last2)) > grain_size;
+  }
+};
+
+struct body
+{
+  template <typename Range>
+  void operator()(Range& r) const
+  {
+    thrust::system::detail::internal::scalar::merge
+      (r.first1, r.last1,
+       r.first2, r.last2,
+       r.result,
+       r.comp);
+  }
+};
+
+} // end namespace merge_detail
+
+namespace merge_by_key_detail
+{
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename InputIterator4,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename StrictWeakOrdering>
+struct range
+{
+  InputIterator1 keys_first1, keys_last1;
+  InputIterator2 keys_first2, keys_last2;
+  InputIterator3 values_first1;
+  InputIterator4 values_first2;
+  OutputIterator1 keys_result;
+  OutputIterator2 values_result;
+  StrictWeakOrdering comp;
+  size_t grain_size;
+
+  range(InputIterator1 keys_first1, InputIterator1 keys_last1,
+        InputIterator2 keys_first2, InputIterator2 keys_last2,
+        InputIterator3 values_first1,
+        InputIterator4 values_first2,
+        OutputIterator1 keys_result,
+        OutputIterator2 values_result,
+        StrictWeakOrdering comp,
+        size_t grain_size = 1024)
+    : keys_first1(keys_first1), keys_last1(keys_last1),
+      keys_first2(keys_first2), keys_last2(keys_last2),
+      values_first1(values_first1),
+      values_first2(values_first2),
+      keys_result(keys_result), values_result(values_result),
+      comp(comp), grain_size(grain_size)
+  {}
+  
+  range(range& r, ::tbb::split)
+    : keys_first1(r.keys_first1), keys_last1(r.keys_last1),
+      keys_first2(r.keys_first2), keys_last2(r.keys_last2),
+      values_first1(r.values_first1),
+      values_first2(r.values_first2),
+      keys_result(r.keys_result), values_result(r.values_result),
+      comp(r.comp), grain_size(r.grain_size)
+  {
+    // we can assume n1 and n2 are not both 0
+    size_t n1 = thrust::distance(keys_first1, keys_last1);
+    size_t n2 = thrust::distance(keys_first2, keys_last2);
+
+    InputIterator1 mid1 = keys_first1;
+    InputIterator2 mid2 = keys_first2;
+
+    if (n1 > n2)
+    {
+      mid1 += n1 / 2;
+      mid2 = thrust::system::detail::internal::scalar::lower_bound(keys_first2, keys_last2, raw_reference_cast(*mid1), comp);
+    }
+    else
+    {
+      mid2 += n2 / 2;
+      mid1 = thrust::system::detail::internal::scalar::upper_bound(keys_first1, keys_last1, raw_reference_cast(*mid2), comp);
+    }
+    
+    // set first range to [keys_first1, mid1), [keys_first2, mid2), keys_result, values_result
+    r.keys_last1 = mid1;
+    r.keys_last2 = mid2;
+
+    // set second range to [mid1, keys_last1), [mid2, keys_last2), keys_result + (mid1 - keys_first1) + (mid2 - keys_first2), values_result + (mid1 - keys_first1) + (mid2 - keys_first2) 
+    keys_first1 = mid1;
+    keys_first2 = mid2;
+    values_first1 += thrust::distance(r.keys_first1, mid1);
+    values_first2 += thrust::distance(r.keys_first2, mid2);
+    keys_result += thrust::distance(r.keys_first1, mid1) + thrust::distance(r.keys_first2, mid2);
+    values_result += thrust::distance(r.keys_first1, mid1) + thrust::distance(r.keys_first2, mid2);
+  }
+
+  bool empty(void) const
+  {
+    return (keys_first1 == keys_last1) && (keys_first2 == keys_last2);
+  }
+
+  bool is_divisible(void) const
+  {
+    return static_cast<size_t>(thrust::distance(keys_first1, keys_last1) + thrust::distance(keys_first2, keys_last2)) > grain_size;
+  }
+};
+
+struct body
+{
+  template <typename Range>
+  void operator()(Range& r) const
+  {
+    thrust::system::detail::internal::scalar::merge_by_key
+      (r.keys_first1, r.keys_last1,
+       r.keys_first2, r.keys_last2,
+       r.values_first1,
+       r.values_first2,
+       r.keys_result,
+       r.values_result,
+       r.comp);
+  }
+};
+
+} // end namespace merge_by_key_detail
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename StrictWeakOrdering>
+OutputIterator merge(execution_policy<DerivedPolicy> &exec,
+                     InputIterator1 first1,
+                     InputIterator1 last1,
+                     InputIterator2 first2,
+                     InputIterator2 last2,
+                     OutputIterator result,
+                     StrictWeakOrdering comp)
+{
+  typedef typename merge_detail::range<InputIterator1,InputIterator2,OutputIterator,StrictWeakOrdering> Range;
+  typedef          merge_detail::body                                                                   Body;
+  Range range(first1, last1, first2, last2, result, comp);
+  Body  body;
+
+  ::tbb::parallel_for(range, body);
+
+  thrust::advance(result, thrust::distance(first1, last1) + thrust::distance(first2, last2));
+
+  return result;
+} // end merge()
+
+template <typename DerivedPolicy,
+          typename InputIterator1,
+          typename InputIterator2,
+          typename InputIterator3,
+          typename InputIterator4,
+          typename OutputIterator1,
+          typename OutputIterator2,
+          typename StrictWeakOrdering>
+thrust::pair<OutputIterator1,OutputIterator2>
+  merge_by_key(execution_policy<DerivedPolicy> &exec,
+               InputIterator1 keys_first1,
+               InputIterator1 keys_last1,
+               InputIterator2 keys_first2,
+               InputIterator2 keys_last2,
+               InputIterator3 values_first3,
+               InputIterator4 values_first4,
+               OutputIterator1 keys_result,
+               OutputIterator2 values_result,
+               StrictWeakOrdering comp)
+{
+  typedef typename merge_by_key_detail::range<InputIterator1,InputIterator2,InputIterator3,InputIterator4,OutputIterator1,OutputIterator2,StrictWeakOrdering> Range;
+  typedef          merge_by_key_detail::body                                                                                                                  Body;
+
+  Range range(keys_first1, keys_last1, keys_first2, keys_last2, values_first3, values_first4, keys_result, values_result, comp);
+  Body  body;
+
+  ::tbb::parallel_for(range, body);
+
+  thrust::advance(keys_result,   thrust::distance(keys_first1, keys_last1) + thrust::distance(keys_first2, keys_last2));
+  thrust::advance(values_result, thrust::distance(keys_first1, keys_last1) + thrust::distance(keys_first2, keys_last2));
+
+  return thrust::make_pair(keys_result,values_result);
+}
+
+} // end namespace detail
+} // end namespace tbb
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/tbb/detail/mismatch.h b/compat/thrust/system/tbb/detail/mismatch.h
new file mode 100644
index 0000000000..03980cfcd5
--- /dev/null
+++ b/compat/thrust/system/tbb/detail/mismatch.h
@@ -0,0 +1,23 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system inherits mismatch
+#include <thrust/system/cpp/detail/mismatch.h>
+
diff --git a/compat/thrust/system/tbb/detail/par.h b/compat/thrust/system/tbb/detail/par.h
new file mode 100644
index 0000000000..74801ab914
--- /dev/null
+++ b/compat/thrust/system/tbb/detail/par.h
@@ -0,0 +1,66 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/tbb/detail/execution_policy.h>
+#include <thrust/detail/execute_with_allocator.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace tbb
+{
+namespace detail
+{
+
+
+struct par_t : thrust::system::tbb::detail::execution_policy<par_t>
+{
+  par_t() : thrust::system::tbb::detail::execution_policy<par_t>() {}
+
+  template<typename Allocator>
+    thrust::detail::execute_with_allocator<Allocator, thrust::system::tbb::detail::execution_policy>
+      operator()(Allocator &alloc) const
+  {
+    return thrust::detail::execute_with_allocator<Allocator, thrust::system::tbb::detail::execution_policy>(alloc);
+  }
+};
+
+
+} // end detail
+
+
+static const detail::par_t par;
+
+
+} // end tbb
+} // end system
+
+
+// alias par here
+namespace tbb
+{
+
+
+using thrust::system::tbb::par;
+
+
+} // end tbb
+} // end thrust
+
diff --git a/compat/thrust/system/tbb/detail/partition.h b/compat/thrust/system/tbb/detail/partition.h
new file mode 100644
index 0000000000..af37121888
--- /dev/null
+++ b/compat/thrust/system/tbb/detail/partition.h
@@ -0,0 +1,87 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/tbb/detail/execution_policy.h>
+#include <thrust/pair.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace tbb
+{
+namespace detail
+{
+
+
+template<typename DerivedPolicy,
+         typename ForwardIterator,
+         typename Predicate>
+  ForwardIterator stable_partition(execution_policy<DerivedPolicy> &exec,
+                                   ForwardIterator first,
+                                   ForwardIterator last,
+                                   Predicate pred);
+
+template<typename DerivedPolicy,
+         typename ForwardIterator,
+         typename InputIterator,
+         typename Predicate>
+  ForwardIterator stable_partition(execution_policy<DerivedPolicy> &exec,
+                                   ForwardIterator first,
+                                   ForwardIterator last,
+                                   InputIterator stencil,
+                                   Predicate pred);
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename Predicate>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    stable_partition_copy(execution_policy<DerivedPolicy> &exec,
+                          InputIterator first,
+                          InputIterator last,
+                          OutputIterator1 out_true,
+                          OutputIterator2 out_false,
+                          Predicate pred);
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename Predicate>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    stable_partition_copy(execution_policy<DerivedPolicy> &exec,
+                          InputIterator1 first,
+                          InputIterator1 last,
+                          InputIterator2 stencil,
+                          OutputIterator1 out_true,
+                          OutputIterator2 out_false,
+                          Predicate pred);
+
+
+} // end namespace detail
+} // end namespace tbb
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/tbb/detail/partition.inl>
+
diff --git a/compat/thrust/system/tbb/detail/partition.inl b/compat/thrust/system/tbb/detail/partition.inl
new file mode 100644
index 0000000000..1e421e10f3
--- /dev/null
+++ b/compat/thrust/system/tbb/detail/partition.inl
@@ -0,0 +1,102 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/tbb/detail/partition.h>
+#include <thrust/system/detail/generic/partition.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace tbb
+{
+namespace detail
+{
+
+
+template<typename DerivedPolicy,
+         typename ForwardIterator,
+         typename Predicate>
+  ForwardIterator stable_partition(execution_policy<DerivedPolicy> &exec,
+                                   ForwardIterator first,
+                                   ForwardIterator last,
+                                   Predicate pred)
+{
+  // tbb prefers generic::stable_partition to cpp::stable_partition
+  return thrust::system::detail::generic::stable_partition(exec, first, last, pred);
+} // end stable_partition()
+
+
+template<typename DerivedPolicy,
+         typename ForwardIterator,
+         typename InputIterator,
+         typename Predicate>
+  ForwardIterator stable_partition(execution_policy<DerivedPolicy> &exec,
+                                   ForwardIterator first,
+                                   ForwardIterator last,
+                                   InputIterator stencil,
+                                   Predicate pred)
+{
+  // tbb prefers generic::stable_partition to cpp::stable_partition
+  return thrust::system::detail::generic::stable_partition(exec, first, last, stencil, pred);
+} // end stable_partition()
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename Predicate>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    stable_partition_copy(execution_policy<DerivedPolicy> &exec,
+                          InputIterator first,
+                          InputIterator last,
+                          OutputIterator1 out_true,
+                          OutputIterator2 out_false,
+                          Predicate pred)
+{
+  // tbb prefers generic::stable_partition_copy to cpp::stable_partition_copy
+  return thrust::system::detail::generic::stable_partition_copy(exec, first, last, out_true, out_false, pred);
+} // end stable_partition_copy()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename Predicate>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    stable_partition_copy(execution_policy<DerivedPolicy> &exec,
+                          InputIterator1 first,
+                          InputIterator1 last,
+                          InputIterator2 stencil,
+                          OutputIterator1 out_true,
+                          OutputIterator2 out_false,
+                          Predicate pred)
+{
+  // tbb prefers generic::stable_partition_copy to cpp::stable_partition_copy
+  return thrust::system::detail::generic::stable_partition_copy(exec, first, last, stencil, out_true, out_false, pred);
+} // end stable_partition_copy()
+
+
+} // end namespace detail
+} // end namespace tbb
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/tbb/detail/reduce.h b/compat/thrust/system/tbb/detail/reduce.h
new file mode 100644
index 0000000000..83a7cc3214
--- /dev/null
+++ b/compat/thrust/system/tbb/detail/reduce.h
@@ -0,0 +1,54 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file reduce.h
+ *  \brief TBB implementation of reduce.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/tbb/detail/execution_policy.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace tbb
+{
+namespace detail
+{
+
+
+template<typename DerivedPolicy,
+         typename InputIterator, 
+         typename OutputType,
+         typename BinaryFunction>
+  OutputType reduce(execution_policy<DerivedPolicy> &exec,
+                    InputIterator begin,
+                    InputIterator end,
+                    OutputType init,
+                    BinaryFunction binary_op);
+
+
+} // end namespace detail
+} // end namespace tbb
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/tbb/detail/reduce.inl>
+
diff --git a/compat/thrust/system/tbb/detail/reduce.inl b/compat/thrust/system/tbb/detail/reduce.inl
new file mode 100644
index 0000000000..c249852769
--- /dev/null
+++ b/compat/thrust/system/tbb/detail/reduce.inl
@@ -0,0 +1,131 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/function.h>
+#include <thrust/detail/static_assert.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/distance.h>
+#include <thrust/reduce.h>
+#include <tbb/blocked_range.h>
+#include <tbb/parallel_reduce.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace tbb
+{
+namespace detail
+{
+namespace reduce_detail
+{
+
+template<typename RandomAccessIterator,
+         typename OutputType,
+         typename BinaryFunction>
+struct body
+{
+  RandomAccessIterator first;
+  OutputType sum;
+  bool first_call;  // TBB can invoke operator() multiple times on the same body
+  thrust::detail::host_function<BinaryFunction,OutputType> binary_op;
+
+  // note: we only initalize sum with init to avoid calling OutputType's default constructor
+  body(RandomAccessIterator first, OutputType init, BinaryFunction binary_op)
+    : first(first), sum(init), first_call(true), binary_op(binary_op)
+  {}
+
+  // note: we only initalize sum with b.sum to avoid calling OutputType's default constructor
+  body(body& b, ::tbb::split)
+    : first(b.first), sum(b.sum), first_call(true), binary_op(b.binary_op)
+  {}
+
+  template <typename Size>
+  void operator()(const ::tbb::blocked_range<Size> &r)
+  {
+    // we assume that blocked_range specifies a contiguous range of integers
+    
+    if (r.empty()) return; // nothing to do
+
+    RandomAccessIterator iter = first + r.begin();
+
+    OutputType temp = thrust::raw_reference_cast(*iter);
+
+    ++iter;
+
+    for (Size i = r.begin() + 1; i != r.end(); ++i, ++iter)
+      temp = binary_op(temp, *iter);
+
+
+    if (first_call)
+    {
+      // first time body has been invoked
+      first_call = false;
+      sum = temp;
+    }
+    else
+    {
+      // body has been previously invoked, accumulate temp into sum
+      sum = binary_op(sum, temp);
+    }
+  } // end operator()()
+  
+  void join(body& b)
+  {
+    sum = binary_op(sum, b.sum);
+  }
+}; // end body
+
+} // end reduce_detail
+
+
+template<typename DerivedPolicy,
+         typename InputIterator, 
+         typename OutputType,
+         typename BinaryFunction>
+  OutputType reduce(execution_policy<DerivedPolicy> &exec,
+                    InputIterator begin,
+                    InputIterator end,
+                    OutputType init,
+                    BinaryFunction binary_op)
+{
+  typedef typename thrust::iterator_difference<InputIterator>::type Size; 
+
+  Size n = thrust::distance(begin, end);
+
+  if (n == 0)
+  {
+    return init;
+  }
+  else
+  {
+    typedef typename reduce_detail::body<InputIterator,OutputType,BinaryFunction> Body;
+    Body reduce_body(begin, init, binary_op);
+    ::tbb::parallel_reduce(::tbb::blocked_range<Size>(0,n), reduce_body);
+    return binary_op(init, reduce_body.sum);
+  }
+}
+
+
+} // end namespace detail
+} // end namespace tbb
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/tbb/detail/reduce_by_key.h b/compat/thrust/system/tbb/detail/reduce_by_key.h
new file mode 100644
index 0000000000..0149a763f7
--- /dev/null
+++ b/compat/thrust/system/tbb/detail/reduce_by_key.h
@@ -0,0 +1,57 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/tbb/detail/execution_policy.h>
+#include <thrust/pair.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace tbb
+{
+namespace detail
+{
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename BinaryPredicate,
+         typename BinaryFunction>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    reduce_by_key(execution_policy<DerivedPolicy> &exec,
+                  InputIterator1 keys_first, 
+                  InputIterator1 keys_last,
+                  InputIterator2 values_first,
+                  OutputIterator1 keys_output,
+                  OutputIterator2 values_output,
+                  BinaryPredicate binary_pred,
+                  BinaryFunction binary_op);
+
+
+} // end namespace detail
+} // end namespace tbb
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/tbb/detail/reduce_by_key.inl>
+
diff --git a/compat/thrust/system/tbb/detail/reduce_by_key.inl b/compat/thrust/system/tbb/detail/reduce_by_key.inl
new file mode 100644
index 0000000000..10d2d8b4a8
--- /dev/null
+++ b/compat/thrust/system/tbb/detail/reduce_by_key.inl
@@ -0,0 +1,344 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/tbb/detail/reduce_by_key.h>
+#include <thrust/iterator/reverse_iterator.h>
+#include <thrust/system/cpp/execution_policy.h>
+#include <thrust/system/tbb/detail/execution_policy.h>
+#include <thrust/system/tbb/detail/reduce_intervals.h>
+#include <thrust/detail/minmax.h>
+#include <thrust/detail/temporary_array.h>
+#include <thrust/detail/range/tail_flags.h>
+#include <tbb/blocked_range.h>
+#include <tbb/parallel_for.h>
+#include <tbb/tbb_thread.h>
+#include <cassert>
+
+
+namespace thrust
+{
+namespace system
+{
+namespace tbb
+{
+namespace detail
+{
+namespace reduce_by_key_detail
+{
+
+
+template<typename L, typename R>
+  inline L divide_ri(const L x, const R y)
+{
+  return (x + (y - 1)) / y;
+}
+
+
+template<typename InputIterator, typename BinaryFunction, typename OutputIterator = void>
+  struct partial_sum_type
+    : thrust::detail::eval_if<
+        thrust::detail::has_result_type<BinaryFunction>::value,
+        thrust::detail::result_type<BinaryFunction>,
+        thrust::detail::eval_if<
+          thrust::detail::is_output_iterator<OutputIterator>::value,
+          thrust::iterator_value<InputIterator>,
+          thrust::iterator_value<OutputIterator>
+        >
+      >
+{};
+
+
+template<typename InputIterator, typename BinaryFunction>
+  struct partial_sum_type<InputIterator,BinaryFunction,void>
+    : thrust::detail::eval_if<
+        thrust::detail::has_result_type<BinaryFunction>::value,
+        thrust::detail::result_type<BinaryFunction>,
+        thrust::iterator_value<InputIterator>
+      >
+{};
+
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename BinaryPredicate,
+         typename BinaryFunction>
+  thrust::pair<
+    InputIterator1,
+    thrust::pair<
+      typename InputIterator1::value_type,
+      typename partial_sum_type<InputIterator2,BinaryFunction>::type
+    >
+  >
+    reduce_last_segment_backward(InputIterator1 keys_first,
+                                 InputIterator1 keys_last,
+                                 InputIterator2 values_first,
+                                 BinaryPredicate binary_pred,
+                                 BinaryFunction binary_op)
+{
+  typename thrust::iterator_difference<InputIterator1>::type n = keys_last - keys_first;
+
+  // reverse the ranges and consume from the end
+  thrust::reverse_iterator<InputIterator1> keys_first_r(keys_last);
+  thrust::reverse_iterator<InputIterator1> keys_last_r(keys_first);
+  thrust::reverse_iterator<InputIterator2> values_first_r(values_first + n);
+
+  typename InputIterator1::value_type result_key = *keys_first_r;
+  typename partial_sum_type<InputIterator2,BinaryFunction>::type result_value = *values_first_r;
+
+  // consume the entirety of the first key's sequence
+  for(++keys_first_r, ++values_first_r;
+      (keys_first_r != keys_last_r) && binary_pred(*keys_first_r, result_key);
+      ++keys_first_r, ++values_first_r)
+  {
+    result_value = binary_op(result_value, *values_first_r);
+  }
+
+  return thrust::make_pair(keys_first_r.base(), thrust::make_pair(result_key, result_value));
+}
+
+
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename BinaryPredicate,
+         typename BinaryFunction>
+  thrust::tuple<
+    OutputIterator1,
+    OutputIterator2,
+    typename InputIterator1::value_type,
+    typename partial_sum_type<InputIterator2,BinaryFunction>::type
+  >
+    reduce_by_key_with_carry(InputIterator1 keys_first, 
+                             InputIterator1 keys_last,
+                             InputIterator2 values_first,
+                             OutputIterator1 keys_output,
+                             OutputIterator2 values_output,
+                             BinaryPredicate binary_pred,
+                             BinaryFunction binary_op)
+{
+  // first, consume the last sequence to produce the carry
+  // XXX is there an elegant way to pose this such that we don't need to default construct carry?
+  thrust::pair<
+    typename InputIterator1::value_type,
+    typename partial_sum_type<InputIterator2,BinaryFunction>::type
+  > carry;
+
+  thrust::tie(keys_last, carry) = reduce_last_segment_backward(keys_first, keys_last, values_first, binary_pred, binary_op);
+
+  // finish with sequential reduce_by_key
+  thrust::cpp::tag seq;
+  thrust::tie(keys_output, values_output) =
+    thrust::reduce_by_key(seq, keys_first, keys_last, values_first, keys_output, values_output, binary_pred, binary_op);
+  
+  return thrust::make_tuple(keys_output, values_output, carry.first, carry.second);
+}
+
+
+template<typename Iterator>
+  bool interval_has_carry(size_t interval_idx, size_t interval_size, size_t num_intervals, Iterator tail_flags)
+{
+  // to discover whether the interval has a carry, look at the tail_flag corresponding to its last element 
+  // the final interval never has a carry by definition
+  return (interval_idx + 1 < num_intervals) ? !tail_flags[(interval_idx + 1) * interval_size - 1] : false;
+}
+
+
+template<typename Iterator1, typename Iterator2, typename Iterator3, typename Iterator4, typename Iterator5, typename Iterator6, typename BinaryPredicate, typename BinaryFunction>
+  struct serial_reduce_by_key_body
+{
+  typedef typename thrust::iterator_difference<Iterator1>::type size_type;
+
+  Iterator1 keys_first;
+  Iterator2 values_first;
+  Iterator3 result_offset;
+  Iterator4 keys_result;
+  Iterator5 values_result;
+  Iterator6 carry_result;
+
+  size_type n;
+  size_type interval_size;
+  size_type num_intervals;
+
+  BinaryPredicate binary_pred;
+  BinaryFunction binary_op;
+
+  serial_reduce_by_key_body(Iterator1 keys_first, Iterator2 values_first, Iterator3 result_offset, Iterator4 keys_result, Iterator5 values_result, Iterator6 carry_result, size_type n, size_type interval_size, size_type num_intervals, BinaryPredicate binary_pred, BinaryFunction binary_op)
+    : keys_first(keys_first), values_first(values_first),
+      result_offset(result_offset),
+      keys_result(keys_result),
+      values_result(values_result),
+      carry_result(carry_result),
+      n(n),
+      interval_size(interval_size),
+      num_intervals(num_intervals),
+      binary_pred(binary_pred),
+      binary_op(binary_op)
+  {}
+
+  void operator()(const ::tbb::blocked_range<size_type> &r) const
+  {
+    assert(r.size() == 1);
+
+    const size_type interval_idx = r.begin();
+
+    const size_type offset_to_first = interval_size * interval_idx;
+    const size_type offset_to_last = thrust::min(n, offset_to_first + interval_size);
+
+    Iterator1 my_keys_first     = keys_first    + offset_to_first;
+    Iterator1 my_keys_last      = keys_first    + offset_to_last;
+    Iterator2 my_values_first   = values_first  + offset_to_first;
+    Iterator3 my_result_offset  = result_offset + interval_idx;
+    Iterator4 my_keys_result    = keys_result   + *my_result_offset;
+    Iterator5 my_values_result  = values_result + *my_result_offset;
+    Iterator6 my_carry_result   = carry_result  + interval_idx;
+
+    // consume the rest of the interval with reduce_by_key
+    typedef typename thrust::iterator_value<Iterator1>::type key_type;
+    typedef typename partial_sum_type<Iterator2,BinaryFunction>::type value_type;
+
+    // XXX is there a way to pose this so that we don't require default construction of carry?
+    thrust::pair<key_type, value_type> carry;
+
+    thrust::tie(my_keys_result, my_values_result, carry.first, carry.second) =
+      reduce_by_key_with_carry(my_keys_first,
+                               my_keys_last,
+                               my_values_first,
+                               my_keys_result,
+                               my_values_result,
+                               binary_pred,
+                               binary_op);
+
+    // store to carry only when we actually have a carry
+    // store to my_keys_result & my_values_result otherwise
+    
+    // create tail_flags so we can check for a carry
+    thrust::detail::tail_flags<Iterator1,BinaryPredicate> flags = thrust::detail::make_tail_flags(keys_first, keys_first + n, binary_pred);
+
+    if(interval_has_carry(interval_idx, interval_size, num_intervals, flags.begin()))
+    {
+      // we can ignore the carry's key
+      // XXX because the carry result is uninitialized, we should copy construct
+      *my_carry_result = carry.second;
+    }
+    else
+    {
+      *my_keys_result = carry.first;
+      *my_values_result = carry.second;
+    }
+  }
+};
+
+
+template<typename Iterator1, typename Iterator2, typename Iterator3, typename Iterator4, typename Iterator5, typename Iterator6, typename BinaryPredicate, typename BinaryFunction>
+  serial_reduce_by_key_body<Iterator1,Iterator2,Iterator3,Iterator4,Iterator5,Iterator6,BinaryPredicate,BinaryFunction>
+    make_serial_reduce_by_key_body(Iterator1 keys_first, Iterator2 values_first, Iterator3 result_offset, Iterator4 keys_result, Iterator5 values_result, Iterator6 carry_result, typename thrust::iterator_difference<Iterator1>::type n, size_t interval_size, size_t num_intervals, BinaryPredicate binary_pred, BinaryFunction binary_op)
+{
+  return serial_reduce_by_key_body<Iterator1,Iterator2,Iterator3,Iterator4,Iterator5,Iterator6,BinaryPredicate,BinaryFunction>(keys_first, values_first, result_offset, keys_result, values_result, carry_result, n, interval_size, num_intervals, binary_pred, binary_op);
+}
+
+
+} // end reduce_by_key_detail
+
+
+template<typename DerivedPolicy, typename Iterator1, typename Iterator2, typename Iterator3, typename Iterator4, typename BinaryPredicate, typename BinaryFunction>
+  thrust::pair<Iterator3,Iterator4>
+    reduce_by_key(thrust::tbb::execution_policy<DerivedPolicy> &exec,
+                  Iterator1 keys_first, Iterator1 keys_last, 
+                  Iterator2 values_first,
+                  Iterator3 keys_result,
+                  Iterator4 values_result,
+                  BinaryPredicate binary_pred,
+                  BinaryFunction binary_op)
+{
+
+  typedef typename thrust::iterator_difference<Iterator1>::type difference_type;
+  difference_type n = keys_last - keys_first;
+  if(n == 0) return thrust::make_pair(keys_result, values_result);
+
+  // XXX this value is a tuning opportunity
+  const difference_type parallelism_threshold = 10000;
+
+  if(n < parallelism_threshold)
+  {
+    // don't bother parallelizing for small n
+    thrust::cpp::tag seq;
+    return thrust::reduce_by_key(seq, keys_first, keys_last, values_first, keys_result, values_result, binary_pred, binary_op);
+  }
+
+  // count the number of processors
+  const unsigned int p = thrust::max<unsigned int>(1u, ::tbb::tbb_thread::hardware_concurrency());
+
+  // generate O(P) intervals of sequential work
+  // XXX oversubscribing is a tuning opportunity
+  const unsigned int subscription_rate = 1;
+  difference_type interval_size = thrust::min<difference_type>(parallelism_threshold, thrust::max<difference_type>(n, n / (subscription_rate * p)));
+  difference_type num_intervals = reduce_by_key_detail::divide_ri(n, interval_size);
+
+  // decompose the input into intervals of size N / num_intervals
+  // add one extra element to this vector to store the size of the entire result
+  thrust::detail::temporary_array<difference_type, DerivedPolicy> interval_output_offsets(0, exec, num_intervals + 1);
+
+  // first count the number of tail flags in each interval
+  thrust::detail::tail_flags<Iterator1,BinaryPredicate> tail_flags = thrust::detail::make_tail_flags(keys_first, keys_last, binary_pred);
+  thrust::system::tbb::detail::reduce_intervals(exec, tail_flags.begin(), tail_flags.end(), interval_size, interval_output_offsets.begin() + 1, thrust::plus<size_t>());
+  interval_output_offsets[0] = 0;
+
+  // scan the counts to get each body's output offset
+  thrust::cpp::tag seq;
+  thrust::inclusive_scan(seq,
+                         interval_output_offsets.begin() + 1, interval_output_offsets.end(), 
+                         interval_output_offsets.begin() + 1);
+
+  // do a reduce_by_key serially in each thread
+  // the final interval never has a carry by definition, so don't reserve space for it
+  typedef typename reduce_by_key_detail::partial_sum_type<Iterator2,BinaryFunction>::type carry_type;
+  thrust::detail::temporary_array<carry_type, DerivedPolicy> carries(0, exec, num_intervals - 1);
+
+  // force grainsize == 1 with simple_partioner()
+  ::tbb::parallel_for(::tbb::blocked_range<difference_type>(0, num_intervals, 1),
+    reduce_by_key_detail::make_serial_reduce_by_key_body(keys_first, values_first, interval_output_offsets.begin(), keys_result, values_result, carries.begin(), n, interval_size, num_intervals, binary_pred, binary_op),
+    ::tbb::simple_partitioner());
+
+  difference_type size_of_result = interval_output_offsets[num_intervals];
+
+  // sequentially accumulate the carries
+  // note that the last interval does not have a carry
+  // XXX find a way to express this loop via a sequential algorithm, perhaps reduce_by_key
+  for(typename thrust::detail::temporary_array<carry_type,DerivedPolicy>::size_type i = 0; i < carries.size(); ++i)
+  {
+    // if our interval has a carry, then we need to sum the carry to the next interval's output offset
+    // if it does not have a carry, then we need to ignore carry_value[i]
+    if(reduce_by_key_detail::interval_has_carry(i, interval_size, num_intervals, tail_flags.begin()))
+    {
+      difference_type output_idx = interval_output_offsets[i+1];
+
+      values_result[output_idx] = binary_op(values_result[output_idx], carries[i]);
+    }
+  }
+
+  return thrust::make_pair(keys_result + size_of_result, values_result + size_of_result);
+}
+
+
+} // end detail
+} // end tbb
+} // end system
+} // end thrust
+
diff --git a/compat/thrust/system/tbb/detail/reduce_intervals.h b/compat/thrust/system/tbb/detail/reduce_intervals.h
new file mode 100644
index 0000000000..0647ffd464
--- /dev/null
+++ b/compat/thrust/system/tbb/detail/reduce_intervals.h
@@ -0,0 +1,126 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/tbb/detail/execution_policy.h>
+
+#include <tbb/parallel_for.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/detail/minmax.h>
+#include <thrust/system/cpp/memory.h>
+#include <thrust/reduce.h>
+#include <cassert>
+
+namespace thrust
+{
+namespace system
+{
+namespace tbb
+{
+namespace detail
+{
+namespace reduce_intervals_detail
+{
+
+
+template<typename L, typename R>
+  inline L divide_ri(const L x, const R y)
+{
+  return (x + (y - 1)) / y;
+}
+
+
+template<typename RandomAccessIterator1, typename RandomAccessIterator2, typename Size, typename BinaryFunction>
+  struct body
+{
+  RandomAccessIterator1 first;
+  RandomAccessIterator2 result;
+  Size n, interval_size;
+  BinaryFunction binary_op;
+
+  body(RandomAccessIterator1 first, RandomAccessIterator2 result, Size n, Size interval_size, BinaryFunction binary_op)
+    : first(first), result(result), n(n), interval_size(interval_size), binary_op(binary_op)
+  {}
+
+  void operator()(const ::tbb::blocked_range<Size> &r) const
+  {
+    assert(r.size() == 1);
+
+    Size interval_idx = r.begin();
+
+    Size offset_to_first = interval_size * interval_idx;
+    Size offset_to_last = thrust::min(n, offset_to_first + interval_size);
+
+    RandomAccessIterator1 my_first = first + offset_to_first;
+    RandomAccessIterator1 my_last  = first + offset_to_last;
+
+    thrust::cpp::tag seq;
+
+    // carefully pass the init value for the interval with raw_reference_cast
+    typedef typename BinaryFunction::result_type sum_type;
+    result[interval_idx] =
+      thrust::reduce(seq, my_first + 1, my_last, sum_type(thrust::raw_reference_cast(*my_first)), binary_op);
+  }
+};
+
+
+template<typename RandomAccessIterator1, typename RandomAccessIterator2, typename Size, typename BinaryFunction>
+  body<RandomAccessIterator1,RandomAccessIterator2,Size,BinaryFunction>
+    make_body(RandomAccessIterator1 first, RandomAccessIterator2 result, Size n, Size interval_size, BinaryFunction binary_op)
+{
+  return body<RandomAccessIterator1,RandomAccessIterator2,Size,BinaryFunction>(first, result, n, interval_size, binary_op);
+}
+
+
+} // end reduce_intervals_detail
+
+
+template<typename DerivedPolicy, typename RandomAccessIterator1, typename Size, typename RandomAccessIterator2, typename BinaryFunction>
+  void reduce_intervals(thrust::tbb::execution_policy<DerivedPolicy> &,
+                        RandomAccessIterator1 first,
+                        RandomAccessIterator1 last,
+                        Size interval_size,
+                        RandomAccessIterator2 result,
+                        BinaryFunction binary_op)
+{
+  typename thrust::iterator_difference<RandomAccessIterator1>::type n = last - first;
+
+  Size num_intervals = reduce_intervals_detail::divide_ri(n, interval_size);
+
+  ::tbb::parallel_for(::tbb::blocked_range<Size>(0, num_intervals, 1), reduce_intervals_detail::make_body(first, result, Size(n), interval_size, binary_op), ::tbb::simple_partitioner());
+}
+
+
+template<typename DerivedPolicy, typename RandomAccessIterator1, typename Size, typename RandomAccessIterator2>
+  void reduce_intervals(thrust::tbb::execution_policy<DerivedPolicy> &exec,
+                        RandomAccessIterator1 first,
+                        RandomAccessIterator1 last,
+                        Size interval_size,
+                        RandomAccessIterator2 result)
+{
+  typedef typename thrust::iterator_value<RandomAccessIterator1>::type value_type;
+
+  return thrust::system::tbb::detail::reduce_intervals(exec, first, last, interval_size, result, thrust::plus<value_type>());
+}
+
+
+} // end detail
+} // end tbb
+} // end system
+} // end thrust
+
diff --git a/compat/thrust/system/tbb/detail/remove.h b/compat/thrust/system/tbb/detail/remove.h
new file mode 100644
index 0000000000..48cbb5c322
--- /dev/null
+++ b/compat/thrust/system/tbb/detail/remove.h
@@ -0,0 +1,81 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/omp/detail/execution_policy.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace omp
+{
+namespace detail
+{
+
+template<typename ExecutionPolicy,
+         typename ForwardIterator,
+         typename Predicate>
+  ForwardIterator remove_if(execution_policy<ExecutionPolicy> &exec,
+                            ForwardIterator first,
+                            ForwardIterator last,
+                            Predicate pred);
+
+
+template<typename ExecutionPolicy,
+         typename ForwardIterator,
+         typename InputIterator,
+         typename Predicate>
+  ForwardIterator remove_if(execution_policy<ExecutionPolicy> &exec,
+                            ForwardIterator first,
+                            ForwardIterator last,
+                            InputIterator stencil,
+                            Predicate pred);
+
+
+template<typename ExecutionPolicy,
+         typename InputIterator,
+         typename OutputIterator,
+         typename Predicate>
+  OutputIterator remove_copy_if(execution_policy<ExecutionPolicy> &exec,
+                                InputIterator first,
+                                InputIterator last,
+                                OutputIterator result,
+                                Predicate pred);
+
+
+template<typename ExecutionPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename Predicate>
+  OutputIterator remove_copy_if(execution_policy<ExecutionPolicy> &exec,
+                                InputIterator1 first,
+                                InputIterator1 last,
+                                InputIterator2 stencil,
+                                OutputIterator result,
+                                Predicate pred);
+
+
+} // end namespace detail
+} // end namespace omp
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/tbb/detail/remove.inl>
+
diff --git a/compat/thrust/system/tbb/detail/remove.inl b/compat/thrust/system/tbb/detail/remove.inl
new file mode 100644
index 0000000000..01916c52c8
--- /dev/null
+++ b/compat/thrust/system/tbb/detail/remove.inl
@@ -0,0 +1,94 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/tbb/detail/remove.h>
+#include <thrust/system/detail/generic/remove.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace tbb
+{
+namespace detail
+{
+
+template<typename DerivedPolicy,
+         typename ForwardIterator,
+         typename Predicate>
+  ForwardIterator remove_if(execution_policy<DerivedPolicy> &exec,
+                            ForwardIterator first,
+                            ForwardIterator last,
+                            Predicate pred)
+{
+  // tbb prefers generic::remove_if to cpp::remove_if
+  return thrust::system::detail::generic::remove_if(exec, first, last, pred);
+}
+
+
+template<typename DerivedPolicy,
+         typename ForwardIterator,
+         typename InputIterator,
+         typename Predicate>
+  ForwardIterator remove_if(execution_policy<DerivedPolicy> &exec,
+                            ForwardIterator first,
+                            ForwardIterator last,
+                            InputIterator stencil,
+                            Predicate pred)
+{
+  // tbb prefers generic::remove_if to cpp::remove_if
+  return thrust::system::detail::generic::remove_if(exec, first, last, stencil, pred);
+}
+
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator,
+         typename Predicate>
+  OutputIterator remove_copy_if(execution_policy<DerivedPolicy> &exec,
+                                InputIterator first,
+                                InputIterator last,
+                                OutputIterator result,
+                                Predicate pred)
+{
+  // tbb prefers generic::remove_copy_if to cpp::remove_copy_if
+  return thrust::system::detail::generic::remove_copy_if(exec, first, last, result, pred);
+}
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename Predicate>
+  OutputIterator remove_copy_if(execution_policy<DerivedPolicy> &exec,
+                                InputIterator1 first,
+                                InputIterator1 last,
+                                InputIterator2 stencil,
+                                OutputIterator result,
+                                Predicate pred)
+{
+  // tbb prefers generic::remove_copy_if to cpp::remove_copy_if
+  return thrust::system::detail::generic::remove_copy_if(exec, first, last, stencil, result, pred);
+}
+
+} // end namespace detail
+} // end namespace tbb
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/tbb/detail/replace.h b/compat/thrust/system/tbb/detail/replace.h
new file mode 100644
index 0000000000..c48555d0e1
--- /dev/null
+++ b/compat/thrust/system/tbb/detail/replace.h
@@ -0,0 +1,23 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system inherits this algorithm
+#include <thrust/system/cpp/detail/scatter.h>
+
diff --git a/compat/thrust/system/tbb/detail/reverse.h b/compat/thrust/system/tbb/detail/reverse.h
new file mode 100644
index 0000000000..04923d1f6a
--- /dev/null
+++ b/compat/thrust/system/tbb/detail/reverse.h
@@ -0,0 +1,23 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system inherits reverse
+#include <thrust/system/cpp/detail/reverse.h>
+
diff --git a/compat/thrust/system/tbb/detail/scan.h b/compat/thrust/system/tbb/detail/scan.h
new file mode 100644
index 0000000000..ed5cacd7e9
--- /dev/null
+++ b/compat/thrust/system/tbb/detail/scan.h
@@ -0,0 +1,64 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file scan.h
+ *  \brief TBB implementations of scan functions.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/tbb/detail/execution_policy.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace tbb
+{
+namespace detail
+{
+
+template<typename InputIterator,
+         typename OutputIterator,
+         typename BinaryFunction>
+  OutputIterator inclusive_scan(tag,
+                                InputIterator first,
+                                InputIterator last,
+                                OutputIterator result,
+                                BinaryFunction binary_op);
+
+
+template<typename InputIterator,
+         typename OutputIterator,
+         typename T,
+         typename BinaryFunction>
+  OutputIterator exclusive_scan(tag,
+                                InputIterator first,
+                                InputIterator last,
+                                OutputIterator result,
+                                T init,
+                                BinaryFunction binary_op);
+
+
+} // end namespace detail
+} // end namespace tbb
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/tbb/detail/scan.inl>
+
diff --git a/compat/thrust/system/tbb/detail/scan.inl b/compat/thrust/system/tbb/detail/scan.inl
new file mode 100644
index 0000000000..48878241c9
--- /dev/null
+++ b/compat/thrust/system/tbb/detail/scan.inl
@@ -0,0 +1,293 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/tbb/detail/scan.h>
+#include <thrust/distance.h>
+#include <thrust/advance.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/detail/function.h>
+#include <thrust/detail/type_traits.h>
+#include <thrust/detail/type_traits/function_traits.h>
+#include <thrust/detail/type_traits/iterator/is_output_iterator.h>
+#include <tbb/blocked_range.h>
+#include <tbb/parallel_scan.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace tbb
+{
+namespace detail
+{
+namespace scan_detail
+{
+
+template<typename InputIterator,
+         typename OutputIterator,
+         typename BinaryFunction,
+         typename ValueType>
+struct inclusive_body
+{
+  InputIterator input;
+  OutputIterator output;
+  thrust::detail::host_function<BinaryFunction,ValueType> binary_op;
+  ValueType sum;
+  bool first_call;
+
+  inclusive_body(InputIterator input, OutputIterator output, BinaryFunction binary_op, ValueType dummy)
+    : input(input), output(output), binary_op(binary_op), sum(dummy), first_call(true)
+  {}
+    
+  inclusive_body(inclusive_body& b, ::tbb::split)
+    : input(b.input), output(b.output), binary_op(b.binary_op), sum(b.sum), first_call(true)
+  {}
+
+  template<typename Size> 
+  void operator()(const ::tbb::blocked_range<Size>& r, ::tbb::pre_scan_tag)
+  {
+    InputIterator iter = input + r.begin();
+ 
+    ValueType temp = *iter;
+
+    ++iter;
+
+    for (Size i = r.begin() + 1; i != r.end(); ++i, ++iter)
+      temp = binary_op(temp, *iter);
+
+    if (first_call)
+      sum = temp;
+    else
+      sum = binary_op(sum, temp);
+      
+    first_call = false;
+  }
+  
+  template<typename Size> 
+  void operator()(const ::tbb::blocked_range<Size>& r, ::tbb::final_scan_tag)
+  {
+    InputIterator  iter1 = input  + r.begin();
+    OutputIterator iter2 = output + r.begin();
+
+    if (first_call)
+    {
+      *iter2 = sum = *iter1;
+      ++iter1;
+      ++iter2;
+      for (Size i = r.begin() + 1; i != r.end(); ++i, ++iter1, ++iter2)
+        *iter2 = sum = binary_op(sum, *iter1);
+    }
+    else
+    {
+      for (Size i = r.begin(); i != r.end(); ++i, ++iter1, ++iter2)
+        *iter2 = sum = binary_op(sum, *iter1);
+    }
+
+    first_call = false;
+  }
+
+  void reverse_join(inclusive_body& b)
+  {
+    sum = binary_op(b.sum, sum);
+  } 
+
+  void assign(inclusive_body& b)
+  {
+    sum = b.sum;
+  } 
+};
+
+
+template<typename InputIterator,
+         typename OutputIterator,
+         typename BinaryFunction,
+         typename ValueType>
+struct exclusive_body
+{
+  InputIterator input;
+  OutputIterator output;
+  thrust::detail::host_function<BinaryFunction,ValueType> binary_op;
+  ValueType sum;
+  bool first_call;
+
+  exclusive_body(InputIterator input, OutputIterator output, BinaryFunction binary_op, ValueType init)
+    : input(input), output(output), binary_op(binary_op), sum(init), first_call(true)
+  {}
+    
+  exclusive_body(exclusive_body& b, ::tbb::split)
+    : input(b.input), output(b.output), binary_op(b.binary_op), sum(b.sum), first_call(true)
+  {}
+
+  template<typename Size> 
+  void operator()(const ::tbb::blocked_range<Size>& r, ::tbb::pre_scan_tag)
+  {
+    InputIterator iter = input + r.begin();
+ 
+    ValueType temp = *iter;
+
+    ++iter;
+
+    for (Size i = r.begin() + 1; i != r.end(); ++i, ++iter)
+      temp = binary_op(temp, *iter);
+
+    if (first_call && r.begin() > 0)
+      sum = temp;
+    else
+      sum = binary_op(sum, temp);
+      
+    first_call = false;
+  }
+  
+  template<typename Size> 
+  void operator()(const ::tbb::blocked_range<Size>& r, ::tbb::final_scan_tag)
+  {
+    InputIterator  iter1 = input  + r.begin();
+    OutputIterator iter2 = output + r.begin();
+
+    for (Size i = r.begin(); i != r.end(); ++i, ++iter1, ++iter2)
+    {
+      ValueType temp = binary_op(sum, *iter1);
+      *iter2 = sum;
+      sum = temp;
+    }
+    
+    first_call = false;
+  }
+
+  void reverse_join(exclusive_body& b)
+  {
+    sum = binary_op(b.sum, sum);
+  } 
+
+  void assign(exclusive_body& b)
+  {
+    sum = b.sum;
+  } 
+};
+
+} // end scan_detail
+
+
+
+template<typename InputIterator,
+         typename OutputIterator,
+         typename BinaryFunction>
+  OutputIterator inclusive_scan(tag,
+                                InputIterator first,
+                                InputIterator last,
+                                OutputIterator result,
+                                BinaryFunction binary_op)
+{
+  // the pseudocode for deducing the type of the temporary used below:
+  // 
+  // if BinaryFunction is AdaptableBinaryFunction
+  //   TemporaryType = AdaptableBinaryFunction::result_type
+  // else if OutputIterator is a "pure" output iterator
+  //   TemporaryType = InputIterator::value_type
+  // else
+  //   TemporaryType = OutputIterator::value_type
+  //
+  // XXX upon c++0x, TemporaryType needs to be:
+  // result_of<BinaryFunction>::type
+  
+  using namespace thrust::detail;
+
+  typedef typename eval_if<
+    has_result_type<BinaryFunction>::value,
+    result_type<BinaryFunction>,
+    eval_if<
+      is_output_iterator<OutputIterator>::value,
+      thrust::iterator_value<InputIterator>,
+      thrust::iterator_value<OutputIterator>
+    >
+  >::type ValueType;
+  
+  typedef typename thrust::iterator_difference<InputIterator>::type Size; 
+  
+  Size n = thrust::distance(first, last);
+
+  if (n != 0)
+  {
+    typedef typename scan_detail::inclusive_body<InputIterator,OutputIterator,BinaryFunction,ValueType> Body;
+    Body scan_body(first, result, binary_op, *first);
+    ::tbb::parallel_scan(::tbb::blocked_range<Size>(0,n), scan_body);
+  }
+ 
+  thrust::advance(result, n);
+
+  return result;
+}
+
+
+template<typename InputIterator,
+         typename OutputIterator,
+         typename T,
+         typename BinaryFunction>
+  OutputIterator exclusive_scan(tag,
+                                InputIterator first,
+                                InputIterator last,
+                                OutputIterator result,
+                                T init,
+                                BinaryFunction binary_op)
+{
+  // the pseudocode for deducing the type of the temporary used below:
+  // 
+  // if BinaryFunction is AdaptableBinaryFunction
+  //   TemporaryType = AdaptableBinaryFunction::result_type
+  // else if OutputIterator is a "pure" output iterator
+  //   TemporaryType = InputIterator::value_type
+  // else
+  //   TemporaryType = OutputIterator::value_type
+  //
+  // XXX upon c++0x, TemporaryType needs to be:
+  // result_of<BinaryFunction>::type
+
+  using namespace thrust::detail;
+
+  typedef typename eval_if<
+    has_result_type<BinaryFunction>::value,
+    result_type<BinaryFunction>,
+    eval_if<
+      is_output_iterator<OutputIterator>::value,
+      thrust::iterator_value<InputIterator>,
+      thrust::iterator_value<OutputIterator>
+    >
+  >::type ValueType;
+
+  typedef typename thrust::iterator_difference<InputIterator>::type Size; 
+  
+  Size n = thrust::distance(first, last);
+
+  if (n != 0)
+  {
+    typedef typename scan_detail::exclusive_body<InputIterator,OutputIterator,BinaryFunction,ValueType> Body;
+    Body scan_body(first, result, binary_op, init);
+    ::tbb::parallel_scan(::tbb::blocked_range<Size>(0,n), scan_body);
+  }
+ 
+  thrust::advance(result, n);
+
+  return result;
+} 
+
+} // end namespace detail
+} // end namespace tbb
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/tbb/detail/scan_by_key.h b/compat/thrust/system/tbb/detail/scan_by_key.h
new file mode 100644
index 0000000000..cad4fc1454
--- /dev/null
+++ b/compat/thrust/system/tbb/detail/scan_by_key.h
@@ -0,0 +1,23 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system inherits scan_by_key
+#include <thrust/system/cpp/detail/scan_by_key.h>
+
diff --git a/compat/thrust/system/tbb/detail/scatter.h b/compat/thrust/system/tbb/detail/scatter.h
new file mode 100644
index 0000000000..c48555d0e1
--- /dev/null
+++ b/compat/thrust/system/tbb/detail/scatter.h
@@ -0,0 +1,23 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system inherits this algorithm
+#include <thrust/system/cpp/detail/scatter.h>
+
diff --git a/compat/thrust/system/tbb/detail/sequence.h b/compat/thrust/system/tbb/detail/sequence.h
new file mode 100644
index 0000000000..811d8f5fbb
--- /dev/null
+++ b/compat/thrust/system/tbb/detail/sequence.h
@@ -0,0 +1,23 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system inherits sequence
+#include <thrust/system/cpp/detail/sequence.h>
+
diff --git a/compat/thrust/system/tbb/detail/set_operations.h b/compat/thrust/system/tbb/detail/set_operations.h
new file mode 100644
index 0000000000..687edb2e7d
--- /dev/null
+++ b/compat/thrust/system/tbb/detail/set_operations.h
@@ -0,0 +1,23 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system inherits set_operations
+#include <thrust/system/cpp/detail/set_operations.h>
+
diff --git a/compat/thrust/system/tbb/detail/sort.h b/compat/thrust/system/tbb/detail/sort.h
new file mode 100644
index 0000000000..3b6f63075e
--- /dev/null
+++ b/compat/thrust/system/tbb/detail/sort.h
@@ -0,0 +1,55 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/tbb/detail/execution_policy.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace tbb
+{
+namespace detail
+{
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator,
+         typename StrictWeakOrdering>
+  void stable_sort(execution_policy<DerivedPolicy> &exec,
+                   RandomAccessIterator first,
+                   RandomAccessIterator last,
+                   StrictWeakOrdering comp);
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2,
+         typename StrictWeakOrdering>
+  void stable_sort_by_key(execution_policy<DerivedPolicy> &exec,
+                          RandomAccessIterator1 keys_first,
+                          RandomAccessIterator1 keys_last,
+                          RandomAccessIterator2 values_first,
+                          StrictWeakOrdering comp);
+
+} // end namespace detail
+} // end namespace tbb
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/tbb/detail/sort.inl>
+
diff --git a/compat/thrust/system/tbb/detail/sort.inl b/compat/thrust/system/tbb/detail/sort.inl
new file mode 100644
index 0000000000..f292789067
--- /dev/null
+++ b/compat/thrust/system/tbb/detail/sort.inl
@@ -0,0 +1,251 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/temporary_array.h>
+#include <thrust/detail/copy.h>
+#include <thrust/system/detail/internal/scalar/sort.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/distance.h>
+#include <thrust/merge.h>
+#include <tbb/parallel_invoke.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace tbb
+{
+namespace detail
+{
+namespace sort_detail
+{
+
+// TODO tune this based on data type and comp
+const static int threshold = 128 * 1024;
+  
+template <typename DerivedPolicy, typename Iterator1, typename Iterator2, typename StrictWeakOrdering>
+void merge_sort(execution_policy<DerivedPolicy> &exec, Iterator1 first1, Iterator1 last1, Iterator2 first2, StrictWeakOrdering comp, bool inplace);
+
+template <typename DerivedPolicy, typename Iterator1, typename Iterator2, typename StrictWeakOrdering>
+struct merge_sort_closure
+{
+  execution_policy<DerivedPolicy> &exec;
+  Iterator1 first1, last1;
+  Iterator2 first2;
+  StrictWeakOrdering comp;
+  bool inplace;
+
+  merge_sort_closure(execution_policy<DerivedPolicy> &exec, Iterator1 first1, Iterator1 last1, Iterator2 first2, StrictWeakOrdering comp, bool inplace)
+    : exec(exec), first1(first1), last1(last1), first2(first2), comp(comp), inplace(inplace)
+  {}
+
+  void operator()(void) const
+  {
+    merge_sort(exec, first1, last1, first2, comp, inplace);
+  }
+};
+
+
+template <typename DerivedPolicy, typename Iterator1, typename Iterator2, typename StrictWeakOrdering>
+void merge_sort(execution_policy<DerivedPolicy> &exec, Iterator1 first1, Iterator1 last1, Iterator2 first2, StrictWeakOrdering comp, bool inplace)
+{
+  typedef typename thrust::iterator_difference<Iterator1>::type difference_type;
+
+  difference_type n = thrust::distance(first1, last1);
+
+  if (n < threshold)
+  {
+    thrust::system::detail::internal::scalar::stable_sort(first1, last1, comp);
+    
+    if (!inplace)
+      thrust::system::detail::internal::scalar::copy(first1, last1, first2);
+
+    return;
+  }
+
+  Iterator1 mid1  = first1 + (n / 2);
+  Iterator2 mid2  = first2 + (n / 2);
+  Iterator2 last2 = first2 + n;
+
+  typedef merge_sort_closure<DerivedPolicy,Iterator1,Iterator2,StrictWeakOrdering> Closure;
+  
+  Closure left (exec, first1, mid1,  first2, comp, !inplace);
+  Closure right(exec, mid1,   last1, mid2,   comp, !inplace);
+
+  ::tbb::parallel_invoke(left, right);
+
+  if (inplace) thrust::merge(exec, first2, mid2, mid2, last2, first1, comp);
+  else	       thrust::merge(exec, first1, mid1, mid1, last1, first2, comp);
+}
+
+} // end namespace sort_detail
+
+
+namespace sort_by_key_detail
+{
+
+// TODO tune this based on data type and comp
+const static int threshold = 128 * 1024;
+  
+template <typename DerivedPolicy,
+          typename Iterator1,
+          typename Iterator2,
+          typename Iterator3,
+          typename Iterator4,
+          typename StrictWeakOrdering>
+void merge_sort_by_key(execution_policy<DerivedPolicy> &exec,
+                       Iterator1 first1,
+                       Iterator1 last1,
+                       Iterator2 first2,
+                       Iterator3 first3,
+                       Iterator4 first4,
+                       StrictWeakOrdering comp,
+                       bool inplace);
+
+template <typename DerivedPolicy,
+          typename Iterator1,
+          typename Iterator2,
+          typename Iterator3,
+          typename Iterator4,
+          typename StrictWeakOrdering>
+struct merge_sort_by_key_closure
+{
+  execution_policy<DerivedPolicy> &exec;
+  Iterator1 first1, last1;
+  Iterator2 first2;
+  Iterator3 first3;
+  Iterator4 first4;
+  StrictWeakOrdering comp;
+  bool inplace;
+
+  merge_sort_by_key_closure(execution_policy<DerivedPolicy> &exec,
+                            Iterator1 first1,
+                            Iterator1 last1,
+                            Iterator2 first2,
+                            Iterator3 first3,
+                            Iterator4 first4,
+                            StrictWeakOrdering comp,
+                            bool inplace)
+    : exec(exec), first1(first1), last1(last1), first2(first2), first3(first3), first4(first4), comp(comp), inplace(inplace)
+  {}
+
+  void operator()(void) const
+  {
+    merge_sort_by_key(exec, first1, last1, first2, first3, first4, comp, inplace);
+  }
+};
+
+
+template <typename DerivedPolicy,
+          typename Iterator1,
+          typename Iterator2,
+          typename Iterator3,
+          typename Iterator4,
+          typename StrictWeakOrdering>
+void merge_sort_by_key(execution_policy<DerivedPolicy> &exec,
+                       Iterator1 first1,
+                       Iterator1 last1,
+                       Iterator2 first2,
+                       Iterator3 first3,
+                       Iterator4 first4,
+                       StrictWeakOrdering comp,
+                       bool inplace)
+{
+  typedef typename thrust::iterator_difference<Iterator1>::type difference_type;
+
+  difference_type n = thrust::distance(first1, last1);
+  
+  Iterator1 mid1  = first1 + (n / 2);
+  Iterator2 mid2  = first2 + (n / 2);
+  Iterator3 mid3  = first3 + (n / 2);
+  Iterator4 mid4  = first4 + (n / 2);
+  Iterator2 last2 = first2 + n;
+  Iterator3 last3 = first3 + n;
+
+  if (n < threshold)
+  {
+    thrust::system::detail::internal::scalar::stable_sort_by_key(first1, last1, first2, comp);
+    
+    if (!inplace)
+    {
+      thrust::system::detail::internal::scalar::copy(first1, last1, first3);
+      thrust::system::detail::internal::scalar::copy(first2, last2, first4);
+    }
+
+    return;
+  }
+
+  typedef merge_sort_by_key_closure<DerivedPolicy,Iterator1,Iterator2,Iterator3,Iterator4,StrictWeakOrdering> Closure;
+  
+  Closure left (exec, first1, mid1,  first2, first3, first4, comp, !inplace);
+  Closure right(exec, mid1,   last1, mid2,   mid3,   mid4,   comp, !inplace);
+
+  ::tbb::parallel_invoke(left, right);
+
+  if(inplace)
+  {
+    thrust::merge_by_key(exec, first3, mid3, mid3, last3, first4, mid4, first1, first2, comp);
+  }
+  else
+  {
+    thrust::merge_by_key(exec, first1, mid1, mid1, last1, first2, mid2, first3, first4, comp);
+  }
+}
+
+} // end namespace sort_detail
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator,
+         typename StrictWeakOrdering>
+void stable_sort(execution_policy<DerivedPolicy> &exec,
+                 RandomAccessIterator first,
+                 RandomAccessIterator last,
+                 StrictWeakOrdering comp)
+{
+  typedef typename thrust::iterator_value<RandomAccessIterator>::type key_type;
+
+  thrust::detail::temporary_array<key_type, DerivedPolicy> temp(exec, first, last);
+
+  sort_detail::merge_sort(exec, first, last, temp.begin(), comp, true);
+}
+
+template<typename DerivedPolicy,
+         typename RandomAccessIterator1,
+         typename RandomAccessIterator2,
+         typename StrictWeakOrdering>
+  void stable_sort_by_key(execution_policy<DerivedPolicy> &exec,
+                          RandomAccessIterator1 first1,
+                          RandomAccessIterator1 last1,
+                          RandomAccessIterator2 first2,
+                          StrictWeakOrdering comp)
+{
+  typedef typename thrust::iterator_value<RandomAccessIterator1>::type key_type;
+  typedef typename thrust::iterator_value<RandomAccessIterator2>::type val_type;
+
+  RandomAccessIterator2 last2 = first2 + thrust::distance(first1, last1);
+
+  thrust::detail::temporary_array<key_type, DerivedPolicy> temp1(exec, first1, last1);
+  thrust::detail::temporary_array<val_type, DerivedPolicy> temp2(exec, first2, last2);
+
+  sort_by_key_detail::merge_sort_by_key(exec, first1, last1, first2, temp1.begin(), temp2.begin(), comp, true);
+}
+
+} // end namespace detail
+} // end namespace tbb
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/tbb/detail/swap_ranges.h b/compat/thrust/system/tbb/detail/swap_ranges.h
new file mode 100644
index 0000000000..15f8f55310
--- /dev/null
+++ b/compat/thrust/system/tbb/detail/swap_ranges.h
@@ -0,0 +1,23 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// tbb inherits swap_ranges
+#include <thrust/system/cpp/detail/swap_ranges.h>
+
diff --git a/compat/thrust/system/tbb/detail/tabulate.h b/compat/thrust/system/tbb/detail/tabulate.h
new file mode 100644
index 0000000000..da65d8e44d
--- /dev/null
+++ b/compat/thrust/system/tbb/detail/tabulate.h
@@ -0,0 +1,23 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system inherits tabulate
+#include <thrust/system/cpp/detail/tabulate.h>
+
diff --git a/compat/thrust/system/tbb/detail/temporary_buffer.h b/compat/thrust/system/tbb/detail/temporary_buffer.h
new file mode 100644
index 0000000000..628bd75719
--- /dev/null
+++ b/compat/thrust/system/tbb/detail/temporary_buffer.h
@@ -0,0 +1,22 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system has no special temporary buffer functions
+
diff --git a/compat/thrust/system/tbb/detail/transform.h b/compat/thrust/system/tbb/detail/transform.h
new file mode 100644
index 0000000000..70ce1f41b6
--- /dev/null
+++ b/compat/thrust/system/tbb/detail/transform.h
@@ -0,0 +1,23 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// omp inherits transform
+#include <thrust/system/cpp/detail/transform.h>
+
diff --git a/compat/thrust/system/tbb/detail/transform_reduce.h b/compat/thrust/system/tbb/detail/transform_reduce.h
new file mode 100644
index 0000000000..23ed07054a
--- /dev/null
+++ b/compat/thrust/system/tbb/detail/transform_reduce.h
@@ -0,0 +1,23 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system inherits transform_reduce
+#include <thrust/system/cpp/detail/transform_reduce.h>
+
diff --git a/compat/thrust/system/tbb/detail/transform_scan.h b/compat/thrust/system/tbb/detail/transform_scan.h
new file mode 100644
index 0000000000..fc2e55d0c0
--- /dev/null
+++ b/compat/thrust/system/tbb/detail/transform_scan.h
@@ -0,0 +1,23 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system inherits transform_scan
+#include <thrust/system/cpp/detail/transform_scan.h>
+
diff --git a/compat/thrust/system/tbb/detail/uninitialized_copy.h b/compat/thrust/system/tbb/detail/uninitialized_copy.h
new file mode 100644
index 0000000000..944f4baf0e
--- /dev/null
+++ b/compat/thrust/system/tbb/detail/uninitialized_copy.h
@@ -0,0 +1,23 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system inherits uninitialized_copy
+#include <thrust/system/cpp/detail/uninitialized_copy.h>
+
diff --git a/compat/thrust/system/tbb/detail/uninitialized_fill.h b/compat/thrust/system/tbb/detail/uninitialized_fill.h
new file mode 100644
index 0000000000..b9d6de20fa
--- /dev/null
+++ b/compat/thrust/system/tbb/detail/uninitialized_fill.h
@@ -0,0 +1,23 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+// this system inherits uninitialized_fill
+#include <thrust/system/cpp/detail/uninitialized_fill.h>
+
diff --git a/compat/thrust/system/tbb/detail/unique.h b/compat/thrust/system/tbb/detail/unique.h
new file mode 100644
index 0000000000..34538cac7b
--- /dev/null
+++ b/compat/thrust/system/tbb/detail/unique.h
@@ -0,0 +1,59 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in ctbbliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/tbb/detail/execution_policy.h>
+#include <thrust/pair.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace tbb
+{
+namespace detail
+{
+
+
+template<typename ExecutionPolicy,
+         typename ForwardIterator,
+         typename BinaryPredicate>
+  ForwardIterator unique(execution_policy<ExecutionPolicy> &exec,
+                         ForwardIterator first,
+                         ForwardIterator last,
+                         BinaryPredicate binary_pred);
+
+
+template<typename ExecutionPolicy,
+         typename InputIterator,
+         typename OutputIterator,
+         typename BinaryPredicate>
+  OutputIterator unique_copy(execution_policy<ExecutionPolicy> &exec,
+                             InputIterator first,
+                             InputIterator last,
+                             OutputIterator output,
+                             BinaryPredicate binary_pred);
+
+
+} // end namespace detail
+} // end namespace tbb 
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/tbb/detail/unique.inl>
+
diff --git a/compat/thrust/system/tbb/detail/unique.inl b/compat/thrust/system/tbb/detail/unique.inl
new file mode 100644
index 0000000000..06e6a30bb1
--- /dev/null
+++ b/compat/thrust/system/tbb/detail/unique.inl
@@ -0,0 +1,66 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in ctbbliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/tbb/detail/unique.h>
+#include <thrust/system/detail/generic/unique.h>
+#include <thrust/pair.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace tbb
+{
+namespace detail
+{
+
+
+template<typename DerivedPolicy,
+         typename ForwardIterator,
+         typename BinaryPredicate>
+  ForwardIterator unique(execution_policy<DerivedPolicy> &exec,
+                         ForwardIterator first,
+                         ForwardIterator last,
+                         BinaryPredicate binary_pred)
+{
+  // tbb prefers generic::unique to cpp::unique
+  return thrust::system::detail::generic::unique(exec,first,last,binary_pred);
+} // end unique()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator,
+         typename BinaryPredicate>
+  OutputIterator unique_copy(execution_policy<DerivedPolicy> &exec,
+                             InputIterator first,
+                             InputIterator last,
+                             OutputIterator output,
+                             BinaryPredicate binary_pred)
+{
+  // tbb prefers generic::unique_copy to cpp::unique_copy
+  return thrust::system::detail::generic::unique_copy(exec,first,last,output,binary_pred);
+} // end unique_copy()
+
+
+} // end namespace detail
+} // end namespace tbb 
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/tbb/detail/unique_by_key.h b/compat/thrust/system/tbb/detail/unique_by_key.h
new file mode 100644
index 0000000000..c6d053243f
--- /dev/null
+++ b/compat/thrust/system/tbb/detail/unique_by_key.h
@@ -0,0 +1,67 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in ctbbliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/tbb/detail/execution_policy.h>
+#include <thrust/pair.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace tbb
+{
+namespace detail
+{
+
+
+template<typename DerivedPolicy,
+         typename ForwardIterator1,
+         typename ForwardIterator2,
+         typename BinaryPredicate>
+  thrust::pair<ForwardIterator1,ForwardIterator2>
+    unique_by_key(execution_policy<DerivedPolicy> &exec,
+                  ForwardIterator1 keys_first, 
+                  ForwardIterator1 keys_last,
+                  ForwardIterator2 values_first,
+                  BinaryPredicate binary_pred);
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename BinaryPredicate>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    unique_by_key_copy(execution_policy<DerivedPolicy> &exec,
+                       InputIterator1 keys_first, 
+                       InputIterator1 keys_last,
+                       InputIterator2 values_first,
+                       OutputIterator1 keys_output,
+                       OutputIterator2 values_output,
+                       BinaryPredicate binary_pred);
+
+
+} // end namespace detail
+} // end namespace tbb 
+} // end namespace system
+} // end namespace thrust
+
+#include <thrust/system/tbb/detail/unique_by_key.inl>
+
diff --git a/compat/thrust/system/tbb/detail/unique_by_key.inl b/compat/thrust/system/tbb/detail/unique_by_key.inl
new file mode 100644
index 0000000000..7747ca4c2e
--- /dev/null
+++ b/compat/thrust/system/tbb/detail/unique_by_key.inl
@@ -0,0 +1,74 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in ctbbliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/tbb/detail/unique_by_key.h>
+#include <thrust/system/detail/generic/unique_by_key.h>
+#include <thrust/pair.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace tbb
+{
+namespace detail
+{
+
+
+template<typename DerivedPolicy,
+         typename ForwardIterator1,
+         typename ForwardIterator2,
+         typename BinaryPredicate>
+  thrust::pair<ForwardIterator1,ForwardIterator2>
+    unique_by_key(execution_policy<DerivedPolicy> &exec,
+                  ForwardIterator1 keys_first, 
+                  ForwardIterator1 keys_last,
+                  ForwardIterator2 values_first,
+                  BinaryPredicate binary_pred)
+{
+  // tbb prefers generic::unique_by_key to cpp::unique_by_key
+  return thrust::system::detail::generic::unique_by_key(exec,keys_first,keys_last,values_first,binary_pred);
+} // end unique_by_key()
+
+
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename BinaryPredicate>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    unique_by_key_copy(execution_policy<DerivedPolicy> &exec,
+                       InputIterator1 keys_first, 
+                       InputIterator1 keys_last,
+                       InputIterator2 values_first,
+                       OutputIterator1 keys_output,
+                       OutputIterator2 values_output,
+                       BinaryPredicate binary_pred)
+{
+  // tbb prefers generic::unique_by_key_copy to cpp::unique_by_key_copy
+  return thrust::system::detail::generic::unique_by_key_copy(exec,keys_first,keys_last,values_first,keys_output,values_output,binary_pred);
+} // end unique_by_key_copy()
+
+
+} // end namespace detail
+} // end namespace tbb
+} // end namespace system
+} // end namespace thrust
+
diff --git a/compat/thrust/system/tbb/detail/vector.inl b/compat/thrust/system/tbb/detail/vector.inl
new file mode 100644
index 0000000000..d87e670fd6
--- /dev/null
+++ b/compat/thrust/system/tbb/detail/vector.inl
@@ -0,0 +1,97 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in ctbbliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/tbb/vector.h>
+
+namespace thrust
+{
+namespace system
+{
+namespace tbb
+{
+
+template<typename T, typename Allocator>
+  vector<T,Allocator>
+    ::vector()
+      : super_t()
+{}
+
+template<typename T, typename Allocator>
+  vector<T,Allocator>
+    ::vector(size_type n)
+      : super_t(n)
+{}
+
+template<typename T, typename Allocator>
+  vector<T,Allocator>
+    ::vector(size_type n, const value_type &value)
+      : super_t(n,value)
+{}
+
+template<typename T, typename Allocator>
+  vector<T,Allocator>
+    ::vector(const vector &x)
+      : super_t(x)
+{}
+
+template<typename T, typename Allocator>
+  template<typename OtherT, typename OtherAllocator>
+    vector<T,Allocator>
+      ::vector(const thrust::detail::vector_base<OtherT,OtherAllocator> &x)
+        : super_t(x)
+{}
+
+template<typename T, typename Allocator>
+  template<typename OtherT, typename OtherAllocator>
+    vector<T,Allocator>
+      ::vector(const std::vector<OtherT,OtherAllocator> &x)
+        : super_t(x)
+{}
+
+template<typename T, typename Allocator>
+  template<typename InputIterator>
+    vector<T,Allocator>
+      ::vector(InputIterator first, InputIterator last)
+        : super_t(first,last)
+{}
+
+template<typename T, typename Allocator>
+  template<typename OtherT, typename OtherAllocator>
+    vector<T,Allocator> &
+      vector<T,Allocator>
+        ::operator=(const std::vector<OtherT,OtherAllocator> &x)
+{
+  super_t::operator=(x);
+  return *this;
+}
+
+template<typename T, typename Allocator>
+  template<typename OtherT, typename OtherAllocator>
+    vector<T,Allocator> &
+      vector<T,Allocator>
+        ::operator=(const thrust::detail::vector_base<OtherT,OtherAllocator> &x)
+{
+  super_t::operator=(x);
+  return *this;
+}
+      
+} // end tbb
+} // end system
+} // end thrust
+
diff --git a/compat/thrust/system/tbb/execution_policy.h b/compat/thrust/system/tbb/execution_policy.h
new file mode 100644
index 0000000000..c462586f82
--- /dev/null
+++ b/compat/thrust/system/tbb/execution_policy.h
@@ -0,0 +1,156 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+/*! \file thrust/system/tbb/execution_policy.h
+ *  \brief Execution policies for Thrust's TBB system.
+ */
+
+#include <thrust/detail/config.h>
+
+// get the execution policies definitions first
+#include <thrust/system/tbb/detail/execution_policy.h>
+
+// get the definition of par
+#include <thrust/system/tbb/detail/par.h>
+
+// now get all the algorithm definitions
+
+#include <thrust/system/tbb/detail/adjacent_difference.h>
+#include <thrust/system/tbb/detail/assign_value.h>
+#include <thrust/system/tbb/detail/binary_search.h>
+#include <thrust/system/tbb/detail/copy.h>
+#include <thrust/system/tbb/detail/copy_if.h>
+#include <thrust/system/tbb/detail/count.h>
+#include <thrust/system/tbb/detail/equal.h>
+#include <thrust/system/tbb/detail/extrema.h>
+#include <thrust/system/tbb/detail/fill.h>
+#include <thrust/system/tbb/detail/find.h>
+#include <thrust/system/tbb/detail/for_each.h>
+#include <thrust/system/tbb/detail/gather.h>
+#include <thrust/system/tbb/detail/generate.h>
+#include <thrust/system/tbb/detail/get_value.h>
+#include <thrust/system/tbb/detail/inner_product.h>
+#include <thrust/system/tbb/detail/iter_swap.h>
+#include <thrust/system/tbb/detail/logical.h>
+#include <thrust/system/tbb/detail/malloc_and_free.h>
+#include <thrust/system/tbb/detail/merge.h>
+#include <thrust/system/tbb/detail/mismatch.h>
+#include <thrust/system/tbb/detail/partition.h>
+#include <thrust/system/tbb/detail/reduce.h>
+#include <thrust/system/tbb/detail/reduce_by_key.h>
+#include <thrust/system/tbb/detail/remove.h>
+#include <thrust/system/tbb/detail/replace.h>
+#include <thrust/system/tbb/detail/reverse.h>
+#include <thrust/system/tbb/detail/scan.h>
+#include <thrust/system/tbb/detail/scan_by_key.h>
+#include <thrust/system/tbb/detail/scatter.h>
+#include <thrust/system/tbb/detail/sequence.h>
+#include <thrust/system/tbb/detail/set_operations.h>
+#include <thrust/system/tbb/detail/sort.h>
+#include <thrust/system/tbb/detail/swap_ranges.h>
+#include <thrust/system/tbb/detail/tabulate.h>
+#include <thrust/system/tbb/detail/transform.h>
+#include <thrust/system/tbb/detail/transform_reduce.h>
+#include <thrust/system/tbb/detail/transform_scan.h>
+#include <thrust/system/tbb/detail/uninitialized_copy.h>
+#include <thrust/system/tbb/detail/uninitialized_fill.h>
+#include <thrust/system/tbb/detail/unique.h>
+#include <thrust/system/tbb/detail/unique_by_key.h>
+
+
+// define these entities here for the purpose of Doxygenating them
+// they are actually defined elsewhere
+#if 0
+namespace thrust
+{
+namespace system
+{
+namespace tbb
+{
+
+
+/*! \addtogroup execution_policies
+ *  \{
+ */
+
+
+/*! \p thrust::tbb::execution_policy is the base class for all Thrust parallel execution
+ *  policies which are derived from Thrust's TBB backend system.
+ */
+template<typename DerivedPolicy>
+struct execution_policy : thrust::execution_policy<DerivedPolicy>
+{};
+
+
+/*! \p tbb::tag is a type representing Thrust's TBB backend system in C++'s type system.
+ *  Iterators "tagged" with a type which is convertible to \p tbb::tag assert that they may be
+ *  "dispatched" to algorithm implementations in the \p tbb system.
+ */
+struct tag : thrust::system::tbb::execution_policy<tag> { unspecified };
+
+
+/*! \p thrust::tbb::par is the parallel execution policy associated with Thrust's TBB
+ *  backend system.
+ *
+ *  Instead of relying on implicit algorithm dispatch through iterator system tags, users may
+ *  directly target Thrust's TBB backend system by providing \p thrust::tbb::par as an algorithm
+ *  parameter.
+ *
+ *  Explicit dispatch can be useful in avoiding the introduction of data copies into containers such
+ *  as \p thrust::tbb::vector.
+ *
+ *  The type of \p thrust::tbb::par is implementation-defined.
+ *
+ *  The following code snippet demonstrates how to use \p thrust::tbb::par to explicitly dispatch an
+ *  invocation of \p thrust::for_each to the TBB backend system:
+ *
+ *  \code
+ *  #include <thrust/for_each.h>
+ *  #include <thrust/system/tbb/execution_policy.h>
+ *  #include <cstdio>
+ *
+ *  struct printf_functor
+ *  {
+ *    __host__ __device__
+ *    void operator()(int x)
+ *    {
+ *      printf("%d\n");
+ *    }
+ *  };
+ *  ...
+ *  int vec[3];
+ *  vec[0] = 0; vec[1] = 1; vec[2] = 2;
+ *
+ *  thrust::for_each(thrust::tbb::par, vec.begin(), vec.end(), printf_functor());
+ *
+ *  // 0 1 2 is printed to standard output in some unspecified order
+ *  \endcode
+ */
+static const unspecified par;
+
+
+/*! \}
+ */
+
+
+} // end tbb
+} // end system
+} // end thrust
+#endif
+
+
diff --git a/compat/thrust/system/tbb/memory.h b/compat/thrust/system/tbb/memory.h
new file mode 100644
index 0000000000..deea7eed7d
--- /dev/null
+++ b/compat/thrust/system/tbb/memory.h
@@ -0,0 +1,414 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in ctbbliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+/*! \file thrust/system/tbb/memory.h
+ *  \brief Managing memory associated with Thrust's TBB system.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/tbb/detail/execution_policy.h>
+#include <thrust/memory.h>
+#include <thrust/detail/type_traits.h>
+#include <thrust/detail/allocator/malloc_allocator.h>
+#include <ostream>
+
+namespace thrust
+{
+namespace system
+{
+namespace tbb
+{
+
+template<typename> class pointer;
+
+} // end tbb
+} // end system
+} // end thrust
+
+
+/*! \cond
+ */
+
+// specialize std::iterator_traits to avoid problems with the name of
+// pointer's constructor shadowing its nested pointer type
+// do this before pointer is defined so the specialization is correctly
+// used inside the definition
+namespace std
+{
+
+template<typename Element>
+  struct iterator_traits<thrust::system::tbb::pointer<Element> >
+{
+  private:
+    typedef thrust::system::tbb::pointer<Element> ptr;
+
+  public:
+    typedef typename ptr::iterator_category       iterator_category;
+    typedef typename ptr::value_type              value_type;
+    typedef typename ptr::difference_type         difference_type;
+    typedef ptr                                   pointer;
+    typedef typename ptr::reference               reference;
+}; // end iterator_traits
+
+} // end std
+
+/*! \endcond
+ */
+
+
+namespace thrust
+{
+namespace system
+{
+
+/*! \addtogroup system_backends Systems
+ *  \ingroup system
+ *  \{
+ */
+
+/*! \namespace thrust::system::tbb
+ *  \brief \p thrust::system::tbb is the namespace containing functionality for allocating, manipulating,
+ *         and deallocating memory available to Thrust's TBB backend system.
+ *         The identifiers are provided in a separate namespace underneath <tt>thrust::system</tt>
+ *         for import convenience but are also aliased in the top-level <tt>thrust::tbb</tt>
+ *         namespace for easy access.
+ *
+ */
+namespace tbb
+{
+
+// forward declaration of reference for pointer
+template<typename Element> class reference;
+
+/*! \cond
+ */
+
+// XXX nvcc + msvc have trouble instantiating reference below
+//     this is a workaround
+namespace detail
+{
+
+template<typename Element>
+  struct reference_msvc_workaround
+{
+  typedef thrust::system::tbb::reference<Element> type;
+}; // end reference_msvc_workaround
+
+} // end detail
+
+/*! \endcond
+ */
+
+
+/*! \p pointer stores a pointer to an object allocated in memory available to the tbb system.
+ *  This type provides type safety when dispatching standard algorithms on ranges resident
+ *  in tbb memory.
+ *
+ *  \p pointer has pointer semantics: it may be dereferenced and manipulated with pointer arithmetic.
+ *
+ *  \p pointer can be created with the function \p tbb::malloc, or by explicitly calling its constructor
+ *  with a raw pointer.
+ *
+ *  The raw pointer encapsulated by a \p pointer may be obtained by eiter its <tt>get</tt> member function
+ *  or the \p raw_pointer_cast function.
+ *
+ *  \note \p pointer is not a "smart" pointer; it is the programmer's responsibility to deallocate memory
+ *  pointed to by \p pointer.
+ *
+ *  \tparam T specifies the type of the pointee.
+ *
+ *  \see tbb::malloc
+ *  \see tbb::free
+ *  \see raw_pointer_cast
+ */
+template<typename T>
+  class pointer
+    : public thrust::pointer<
+               T,
+               thrust::system::tbb::tag,
+               thrust::system::tbb::reference<T>,
+               thrust::system::tbb::pointer<T>
+             >
+{
+  /*! \cond
+   */
+
+  private:
+    typedef thrust::pointer<
+      T,
+      thrust::system::tbb::tag,
+      //thrust::system::tbb::reference<T>,
+      typename detail::reference_msvc_workaround<T>::type,
+      thrust::system::tbb::pointer<T>
+    > super_t;
+
+  /*! \endcond
+   */
+
+  public:
+    // note that tbb::pointer's member functions need __host__ __device__
+    // to interoperate with nvcc + iterators' dereference member function
+
+    /*! \p pointer's no-argument constructor initializes its encapsulated pointer to \c 0.
+     */
+    __host__ __device__
+    pointer() : super_t() {}
+
+    /*! This constructor allows construction of a <tt>pointer<const T></tt> from a <tt>T*</tt>.
+     *
+     *  \param ptr A raw pointer to copy from, presumed to point to a location in memory
+     *         accessible by the \p tbb system.
+     *  \tparam OtherT \p OtherT shall be convertible to \p T.
+     */
+    template<typename OtherT>
+    __host__ __device__
+    explicit pointer(OtherT *ptr) : super_t(ptr) {}
+
+    /*! This constructor allows construction from another pointer-like object with related type.
+     *
+     *  \param other The \p OtherPointer to copy.
+     *  \tparam OtherPointer The system tag associated with \p OtherPointer shall be convertible
+     *          to \p thrust::system::tbb::tag and its element type shall be convertible to \p T.
+     */
+    template<typename OtherPointer>
+    __host__ __device__
+    pointer(const OtherPointer &other,
+            typename thrust::detail::enable_if_pointer_is_convertible<
+              OtherPointer,
+              pointer
+            >::type * = 0) : super_t(other) {}
+
+    /*! Assignment operator allows assigning from another pointer-like object with related type.
+     *
+     *  \param other The other pointer-like object to assign from.
+     *  \tparam OtherPointer The system tag associated with \p OtherPointer shall be convertible
+     *          to \p thrust::system::tbb::tag and its element type shall be convertible to \p T.
+     */
+    template<typename OtherPointer>
+    __host__ __device__
+    typename thrust::detail::enable_if_pointer_is_convertible<
+      OtherPointer,
+      pointer,
+      pointer &
+    >::type
+    operator=(const OtherPointer &other)
+    {
+      return super_t::operator=(other);
+    }
+}; // end pointer
+
+
+/*! \p reference is a wrapped reference to an object stored in memory available to the \p tbb system.
+ *  \p reference is the type of the result of dereferencing a \p tbb::pointer.
+ *
+ *  \tparam T Specifies the type of the referenced object.
+ */
+template<typename T>
+  class reference
+    : public thrust::reference<
+               T,
+               thrust::system::tbb::pointer<T>,
+               thrust::system::tbb::reference<T>
+             >
+{
+  /*! \cond
+   */
+
+  private:
+    typedef thrust::reference<
+      T,
+      thrust::system::tbb::pointer<T>,
+      thrust::system::tbb::reference<T>
+    > super_t;
+
+  /*! \endcond
+   */
+
+  public:
+    /*! \cond
+     */
+
+    typedef typename super_t::value_type value_type;
+    typedef typename super_t::pointer    pointer;
+
+    /*! \endcond
+     */
+
+    /*! This constructor initializes this \p reference to refer to an object
+     *  pointed to by the given \p pointer. After this \p reference is constructed,
+     *  it shall refer to the object pointed to by \p ptr.
+     *
+     *  \param ptr A \p pointer to copy from.
+     */
+    __host__ __device__
+    explicit reference(const pointer &ptr)
+      : super_t(ptr)
+    {}
+
+    /*! This constructor accepts a const reference to another \p reference of related type.
+     *  After this \p reference is constructed, it shall refer to the same object as \p other.
+     *
+     *  \param other A \p reference to copy from.
+     *  \tparam OtherT The element type of the other \p reference.
+     *
+     *  \note This constructor is templated primarily to allow initialization of <tt>reference<const T></tt>
+     *        from <tt>reference<T></tt>.
+     */
+    template<typename OtherT>
+    __host__ __device__
+    reference(const reference<OtherT> &other,
+              typename thrust::detail::enable_if_convertible<
+                typename reference<OtherT>::pointer,
+                pointer
+              >::type * = 0)
+      : super_t(other)
+    {}
+
+    /*! Copy assignment operator copy assigns from another \p reference of related type.
+     *
+     *  \param other The other \p reference to assign from.
+     *  \return <tt>*this</tt>
+     *  \tparam OtherT The element type of the other \p reference.
+     */
+    template<typename OtherT>
+    reference &operator=(const reference<OtherT> &other);
+
+    /*! Assignment operator assigns from a \p value_type.
+     *
+     *  \param x The \p value_type to assign from.
+     *  \return <tt>*this</tt>
+     */
+    reference &operator=(const value_type &x);
+}; // end reference
+
+/*! Exchanges the values of two objects referred to by \p reference.
+ *  \p x The first \p reference of interest.
+ *  \p y The second \p reference ot interest.
+ */
+template<typename T>
+__host__ __device__
+void swap(reference<T> x, reference<T> y);
+
+/*! Allocates an area of memory available to Thrust's <tt>tbb</tt> system.
+ *  \param n Number of bytes to allocate.
+ *  \return A <tt>tbb::pointer<void></tt> pointing to the beginning of the newly
+ *          allocated memory. A null <tt>tbb::pointer<void></tt> is returned if
+ *          an error occurs.
+ *  \note The <tt>tbb::pointer<void></tt> returned by this function must be
+ *        deallocated with \p tbb::free.
+ *  \see tbb::free
+ *  \see std::malloc
+ */
+inline pointer<void> malloc(std::size_t n);
+
+/*! Allocates a typed area of memory available to Thrust's <tt>tbb</tt> system.
+ *  \param n Number of elements to allocate.
+ *  \return A <tt>tbb::pointer<T></tt> pointing to the beginning of the newly
+ *          allocated memory. A null <tt>tbb::pointer<T></tt> is returned if
+ *          an error occurs.
+ *  \note The <tt>tbb::pointer<T></tt> returned by this function must be
+ *        deallocated with \p tbb::free.
+ *  \see tbb::free
+ *  \see std::malloc
+ */
+template<typename T>
+inline pointer<T> malloc(std::size_t n);
+
+/*! Deallocates an area of memory previously allocated by <tt>tbb::malloc</tt>.
+ *  \param ptr A <tt>tbb::pointer<void></tt> pointing to the beginning of an area
+ *         of memory previously allocated with <tt>tbb::malloc</tt>.
+ *  \see tbb::malloc
+ *  \see std::free
+ */
+inline void free(pointer<void> ptr);
+
+// XXX upon c++11
+// template<typename T> using allocator = thrust::detail::malloc_allocator<T,tag,pointer<T> >;
+
+/*! \p tbb::allocator is the default allocator used by the \p tbb system's containers such as
+ *  <tt>tbb::vector</tt> if no user-specified allocator is provided. \p tbb::allocator allocates
+ *  (deallocates) storage with \p tbb::malloc (\p tbb::free).
+ */
+template<typename T>
+  struct allocator
+    : thrust::detail::malloc_allocator<
+        T,
+        tag,
+        pointer<T>
+      >
+{
+  /*! The \p rebind metafunction provides the type of an \p allocator
+   *  instantiated with another type.
+   *
+   *  \tparam U The other type to use for instantiation.
+   */
+  template<typename U>
+    struct rebind
+  {
+    /*! The typedef \p other gives the type of the rebound \p allocator.
+     */
+    typedef allocator<U> other;
+  };
+
+  /*! No-argument constructor has no effect.
+   */
+  __host__ __device__
+  inline allocator() {}
+
+  /*! Copy constructor has no effect.
+   */
+  __host__ __device__
+  inline allocator(const allocator &) {}
+
+  /*! Constructor from other \p allocator has no effect.
+   */
+  template<typename U>
+  __host__ __device__
+  inline allocator(const allocator<U> &) {}
+
+  /*! Destructor has no effect.
+   */
+  __host__ __device__
+  inline ~allocator() {}
+}; // end allocator
+
+} // end tbb
+
+/*! \}
+ */
+
+} // end system
+
+/*! \namespace thrust::tbb
+ *  \brief \p thrust::tbb is a top-level alias for thrust::system::tbb.
+ */
+namespace tbb
+{
+
+using thrust::system::tbb::pointer;
+using thrust::system::tbb::reference;
+using thrust::system::tbb::malloc;
+using thrust::system::tbb::free;
+using thrust::system::tbb::allocator;
+
+} // end tbb
+
+} // end thrust
+
+#include <thrust/system/tbb/detail/memory.inl>
+
diff --git a/compat/thrust/system/tbb/vector.h b/compat/thrust/system/tbb/vector.h
new file mode 100644
index 0000000000..1c49c3f9e0
--- /dev/null
+++ b/compat/thrust/system/tbb/vector.h
@@ -0,0 +1,144 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in ctbbliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+/*! \file thrust/system/tbb/vector.h
+ *  \brief A dynamically-sizable array of elements which reside in memory available to
+ *         Thrust's TBB system.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/system/tbb/memory.h>
+#include <thrust/detail/vector_base.h>
+#include <vector>
+
+namespace thrust
+{
+namespace system
+{
+namespace tbb
+{
+
+// XXX upon c++11
+// template<typename T, typename Allocator = allocator<T> > using vector = thrust::detail::vector_base<T,Allocator>;
+
+/*! \p tbb::vector is a container that supports random access to elements,
+ *  constant time removal of elements at the end, and linear time insertion
+ *  and removal of elements at the beginning or in the middle. The number of
+ *  elements in a \p tbb::vector may vary dynamically; memory management is
+ *  automatic. The elements contained in a \p tbb::vector reside in memory
+ *  available to the \p tbb system.
+ *
+ *  \tparam T The element type of the \p tbb::vector.
+ *  \tparam Allocator The allocator type of the \p tbb::vector. Defaults to \p tbb::allocator.
+ *
+ *  \see http://www.sgi.com/tech/stl/Vector.html
+ *  \see host_vector For the documentation of the complete interface which is
+ *                   shared by \p tbb::vector
+ *  \see device_vector
+ */
+template<typename T, typename Allocator = allocator<T> >
+  class vector
+    : public thrust::detail::vector_base<T,Allocator>
+{
+  /*! \cond
+   */
+  private:
+    typedef thrust::detail::vector_base<T,Allocator> super_t;
+  /*! \endcond
+   */
+
+  public:
+
+  /*! \cond
+   */
+    typedef typename super_t::size_type  size_type;
+    typedef typename super_t::value_type value_type;
+  /*! \endcond
+   */
+
+    /*! This constructor creates an empty \p tbb::vector.
+     */
+    vector();
+
+    /*! This constructor creates a \p tbb::vector with \p n default-constructed elements.
+     *  \param n The size of the \p tbb::vector to create.
+     */
+    explicit vector(size_type n);
+
+    /*! This constructor creates a \p tbb::vector with \p n copies of \p value.
+     *  \param n The size of the \p tbb::vector to create.
+     *  \param value An element to copy.
+     */
+    explicit vector(size_type n, const value_type &value);
+
+    /*! Copy constructor copies from another \p tbb::vector.
+     *  \param x The other \p tbb::vector to copy.
+     */
+    vector(const vector &x);
+
+    /*! This constructor copies from another Thrust vector-like object.
+     *  \param x The other object to copy from.
+     */
+    template<typename OtherT, typename OtherAllocator>
+    vector(const thrust::detail::vector_base<OtherT,OtherAllocator> &x);
+
+    /*! This constructor copies from a \c std::vector.
+     *  \param x The \c std::vector to copy from.
+     */
+    template<typename OtherT, typename OtherAllocator>
+    vector(const std::vector<OtherT,OtherAllocator> &x);
+
+    /*! This constructor creates a \p tbb::vector by copying from a range.
+     *  \param first The beginning of the range.
+     *  \param last The end of the range.
+     */
+    template<typename InputIterator>
+    vector(InputIterator first, InputIterator last);
+
+    // XXX vector_base should take a Derived type so we don't have to define these superfluous assigns
+
+    /*! Assignment operator assigns from a \c std::vector.
+     *  \param x The \c std::vector to assign from.
+     *  \return <tt>*this</tt>
+     */
+    template<typename OtherT, typename OtherAllocator>
+    vector &operator=(const std::vector<OtherT,OtherAllocator> &x);
+
+    /*! Assignment operator assigns from another Thrust vector-like object.
+     *  \param x The other object to assign from.
+     *  \return <tt>*this</tt>
+     */
+    template<typename OtherT, typename OtherAllocator>
+    vector &operator=(const thrust::detail::vector_base<OtherT,OtherAllocator> &x);
+}; // end vector
+
+} // end tbb
+} // end system
+
+// alias system::tbb names at top-level
+namespace tbb
+{
+
+using thrust::system::tbb::vector;
+
+} // end tbb
+
+} // end thrust
+
+#include <thrust/system/tbb/detail/vector.inl>
+
diff --git a/compat/thrust/system_error.h b/compat/thrust/system_error.h
new file mode 100644
index 0000000000..ce88fe6bd8
--- /dev/null
+++ b/compat/thrust/system_error.h
@@ -0,0 +1,51 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+/*! \file thrust/system_error.h
+ *  \brief System diagnostics
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+namespace thrust
+{
+
+/*! \addtogroup system System Access
+ *  \{
+ */
+
+/*! \namespace thrust::system
+ *  \brief \p thrust::system is the namespace which contains functionality for manipulating
+ *         memory specific to one of Thrust's backend systems. It also contains functionality
+ *         for reporting error conditions originating from the operating system or other
+ *         low-level application program interfaces such as the CUDA runtime.
+ *         They are provided in a separate namespace for import convenience but are
+ *         also aliased in the top-level \p thrust namespace for easy access.
+ */
+namespace system
+{
+} // end system
+
+/*! \} // end system
+ */
+
+} // end thrust
+
+#include <thrust/system/error_code.h>
+#include <thrust/system/system_error.h>
+
diff --git a/compat/thrust/tabulate.h b/compat/thrust/tabulate.h
new file mode 100644
index 0000000000..c87edf01ab
--- /dev/null
+++ b/compat/thrust/tabulate.h
@@ -0,0 +1,128 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file tabulate.h
+ *  \brief Fills a range with the tabulation of a function
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/execution_policy.h>
+
+namespace thrust
+{
+
+
+/*! \addtogroup transformations
+ *  \{
+ */
+
+
+/*! \p tabulate fills the range <tt>[first, last)</tt> with the value of a function applied to each
+ *     element's index.
+ *
+ *  For each iterator \c i in the range <tt>[first, last)</tt>, \p tabulate performs the assignment
+ *  <tt>*i = unary_op(i - first)</tt>.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the range.
+ *  \param last The end of the range.
+ *  \param unary_op The unary operation to apply.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>,
+ *          and \p ForwardIterator is mutable,
+ *          and if \c x and \c y are objects of \c ForwardIterator's \c value_type, then <tt>x + y</tt> is defined,
+ *          and if \c T is \p ForwardIterator's \c value_type, then <tt>T(0)</tt> is defined.
+ *  \tparam UnaryOperation is a model of <a href="http://www.sgi.com/tech/stl/UnaryFunction.html">Unary Function</a>
+ *                         and \c UnaryFunction's \c result_type is convertible to \c OutputIterator's \c value_type.
+ *
+ *  The following code snippet demonstrates how to use \p tabulate to generate the first \c n non-positive integers
+ *  using the \p thrust::host execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/tabulate.h>
+ *  #include <thrust/functional.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  const int N = 10;
+ *  int A[N];
+ *  thrust::tabulate(thrust::host, A, A + 10, thrust::negate<int>());
+ *  // A is now {0, -1, -2, -3, -4, -5, -6, -7, -8, -9}
+ *  \endcode
+ *
+ *  \see thrust::fill
+ *  \see thrust::generate
+ *  \see thrust::sequence
+ */
+template<typename DerivedPolicy, typename ForwardIterator, typename UnaryOperation>
+  void tabulate(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                ForwardIterator first,
+                ForwardIterator last,
+                UnaryOperation unary_op);
+
+
+/*! \p tabulate fills the range <tt>[first, last)</tt> with the value of a function applied to each
+ *     element's index.
+ *
+ *  For each iterator \c i in the range <tt>[first, last)</tt>, \p tabulate performs the assignment
+ *  <tt>*i = unary_op(i - first)</tt>.
+ *
+ *  \param first The beginning of the range.
+ *  \param last The end of the range.
+ *  \param unary_op The unary operation to apply.
+ *
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>,
+ *          and \p ForwardIterator is mutable,
+ *          and if \c x and \c y are objects of \c ForwardIterator's \c value_type, then <tt>x + y</tt> is defined,
+ *          and if \c T is \p ForwardIterator's \c value_type, then <tt>T(0)</tt> is defined.
+ *  \tparam UnaryOperation is a model of <a href="http://www.sgi.com/tech/stl/UnaryFunction.html">Unary Function</a>
+ *                         and \c UnaryFunction's \c result_type is convertible to \c OutputIterator's \c value_type.
+ *
+ *  The following code snippet demonstrates how to use \p tabulate to generate the first \c n non-positive integers:
+ *
+ *  \code
+ *  #include <thrust/tabulate.h>
+ *  #include <thrust/functional.h>
+ *  ...
+ *  const int N = 10;
+ *  int A[N];
+ *  thrust::tabulate(A, A + 10, thrust::negate<int>());
+ *  // A is now {0, -1, -2, -3, -4, -5, -6, -7, -8, -9}
+ *  \endcode
+ *
+ *  \see thrust::fill
+ *  \see thrust::generate
+ *  \see thrust::sequence
+ */
+template<typename ForwardIterator, typename UnaryOperation>
+  void tabulate(ForwardIterator first,
+                ForwardIterator last,
+                UnaryOperation unary_op);
+
+
+/*! \} // end transformations
+ */
+
+
+} // end namespace thrust
+
+#include <thrust/detail/tabulate.inl>
+
diff --git a/compat/thrust/transform.h b/compat/thrust/transform.h
new file mode 100644
index 0000000000..1ada105828
--- /dev/null
+++ b/compat/thrust/transform.h
@@ -0,0 +1,720 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file transform.h
+ *  \brief Transforms input ranges using a function object
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/execution_policy.h>
+
+namespace thrust
+{
+
+
+/*! \addtogroup algorithms
+ */
+
+/*! \addtogroup transformations
+ *  \ingroup algorithms
+ *  \{
+ */
+
+
+/*! This version of \p transform applies a unary function to each element
+ *  of an input sequence and stores the result in the corresponding 
+ *  position in an output sequence.  Specifically, for each iterator 
+ *  <tt>i</tt> in the range [\p first, \p last) the operation 
+ *  <tt>op(*i)</tt> is performed and the result is assigned to <tt>*o</tt>,
+ *  where <tt>o</tt> is the corresponding output iterator in the range
+ *  [\p result, \p result + (\p last - \p first) ).  The input and
+ *  output sequences may coincide, resulting in an in-place transformation.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *    
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the input sequence.
+ *  \param last The end of the input sequence.
+ *  \param result The beginning of the output sequence.
+ *  \param op The tranformation operation.
+ *  \return The end of the output sequence.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>
+ *                        and \c InputIterator's \c value_type is convertible to \c UnaryFunction's \c argument_type.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam UnaryFunction is a model of <a href="http://www.sgi.com/tech/stl/UnaryFunction.html">Unary Function</a>
+ *                              and \c UnaryFunction's \c result_type is convertible to \c OutputIterator's \c value_type.
+ *
+ *  \pre \p first may equal \p result, but the range <tt>[first, last)</tt> shall not overlap the range <tt>[result, result + (last - first))</tt> otherwise.
+ *
+ *  The following code snippet demonstrates how to use \p transform to negate a range in-place
+ *  using the \p thrust::host execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/transform.h>
+ *  #include <thrust/functional.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  
+ *  int data[10] = {-5, 0, 2, -3, 2, 4, 0, -1, 2, 8};
+ * 
+ *  thrust::negate<int> op;
+ *
+ *  thrust::transform(thrust::host, data, data + 10, data, op); // in-place transformation
+ *
+ *  // data is now {5, 0, -2, 3, -2, -4, 0, 1, -2, -8};
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/transform.html
+ */
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator,
+         typename UnaryFunction>
+  OutputIterator transform(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                           InputIterator first, InputIterator last,
+                           OutputIterator result,
+                           UnaryFunction op);
+
+	
+/*! This version of \p transform applies a unary function to each element
+ *  of an input sequence and stores the result in the corresponding 
+ *  position in an output sequence.  Specifically, for each iterator 
+ *  <tt>i</tt> in the range [\p first, \p last) the operation 
+ *  <tt>op(*i)</tt> is performed and the result is assigned to <tt>*o</tt>,
+ *  where <tt>o</tt> is the corresponding output iterator in the range
+ *  [\p result, \p result + (\p last - \p first) ).  The input and
+ *  output sequences may coincide, resulting in an in-place transformation.
+ *    
+ *  \param first The beginning of the input sequence.
+ *  \param last The end of the input sequence.
+ *  \param result The beginning of the output sequence.
+ *  \param op The tranformation operation.
+ *  \return The end of the output sequence.
+ *
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>
+ *                        and \c InputIterator's \c value_type is convertible to \c UnaryFunction's \c argument_type.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam UnaryFunction is a model of <a href="http://www.sgi.com/tech/stl/UnaryFunction.html">Unary Function</a>
+ *                              and \c UnaryFunction's \c result_type is convertible to \c OutputIterator's \c value_type.
+ *
+ *  \pre \p first may equal \p result, but the range <tt>[first, last)</tt> shall not overlap the range <tt>[result, result + (last - first))</tt> otherwise.
+ *
+ *  The following code snippet demonstrates how to use \p transform
+ *
+ *  \code
+ *  #include <thrust/transform.h>
+ *  #include <thrust/functional.h>
+ *  
+ *  int data[10] = {-5, 0, 2, -3, 2, 4, 0, -1, 2, 8};
+ * 
+ *  thrust::negate<int> op;
+ *
+ *  thrust::transform(data, data + 10, data, op); // in-place transformation
+ *
+ *  // data is now {5, 0, -2, 3, -2, -4, 0, 1, -2, -8};
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/transform.html
+ */
+template<typename InputIterator,
+         typename OutputIterator,
+         typename UnaryFunction>
+  OutputIterator transform(InputIterator first, InputIterator last,
+                           OutputIterator result,
+                           UnaryFunction op);
+
+
+/*! This version of \p transform applies a binary function to each pair
+ *  of elements from two input sequences and stores the result in the
+ *  corresponding position in an output sequence.  Specifically, for
+ *  each iterator <tt>i</tt> in the range [\p first1, \p last1) and 
+ *  <tt>j = first + (i - first1)</tt> in the range [\p first2, \p last2)
+ *  the operation <tt>op(*i,*j)</tt> is performed and the result is 
+ *  assigned to <tt>*o</tt>,  where <tt>o</tt> is the corresponding
+ *  output iterator in the range [\p result, \p result + (\p last - \p first) ).
+ *  The input and output sequences may coincide, resulting in an 
+ *  in-place transformation.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *    
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first1 The beginning of the first input sequence.
+ *  \param last1 The end of the first input sequence.
+ *  \param first2 The beginning of the second input sequence.
+ *  \param result The beginning of the output sequence.
+ *  \param op The tranformation operation.
+ *  \return The end of the output sequence.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>
+ *                        and \c InputIterator1's \c value_type is convertible to \c BinaryFunction's \c first_argument_type.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>
+ *                        and \c InputIterator2's \c value_type is convertible to \c BinaryFunction's \c second_argument_type.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam BinaryFunction is a model of <a href="http://www.sgi.com/tech/stl/BinaryFunction.html">Binary Function</a>
+ *                              and \c BinaryFunction's \c result_type is convertible to \c OutputIterator's \c value_type.
+ *
+ *  \pre \p first1 may equal \p result, but the range <tt>[first1, last1)</tt> shall not overlap the range <tt>[result, result + (last1 - first1))</tt> otherwise.
+ *  \pre \p first2 may equal \p result, but the range <tt>[first2, first2 + (last1 - first1))</tt> shall not overlap the range <tt>[result, result + (last1 - first1))</tt> otherwise.
+ *
+ *  The following code snippet demonstrates how to use \p transform to compute the sum of two
+ *  ranges using the \p thrust::host execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/transform.h>
+ *  #include <thrust/functional.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  
+ *  int input1[6] = {-5,  0,  2,  3,  2,  4};
+ *  int input2[6] = { 3,  6, -2,  1,  2,  3};
+ *  int output[6];
+ * 
+ *  thrust::plus<int> op;
+ *
+ *  thrust::transform(thrust::host, input1, input1 + 6, input2, output, op);
+ *
+ *  // output is now {-2,  6,  0,  4,  4,  7};
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/transform.html
+ */
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename BinaryFunction>
+  OutputIterator transform(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                           InputIterator1 first1, InputIterator1 last1,
+                           InputIterator2 first2,
+                           OutputIterator result,
+                           BinaryFunction op);
+
+
+/*! This version of \p transform applies a binary function to each pair
+ *  of elements from two input sequences and stores the result in the
+ *  corresponding position in an output sequence.  Specifically, for
+ *  each iterator <tt>i</tt> in the range [\p first1, \p last1) and 
+ *  <tt>j = first + (i - first1)</tt> in the range [\p first2, \p last2)
+ *  the operation <tt>op(*i,*j)</tt> is performed and the result is 
+ *  assigned to <tt>*o</tt>,  where <tt>o</tt> is the corresponding
+ *  output iterator in the range [\p result, \p result + (\p last - \p first) ).
+ *  The input and output sequences may coincide, resulting in an 
+ *  in-place transformation.
+ *    
+ *  \param first1 The beginning of the first input sequence.
+ *  \param last1 The end of the first input sequence.
+ *  \param first2 The beginning of the second input sequence.
+ *  \param result The beginning of the output sequence.
+ *  \param op The tranformation operation.
+ *  \return The end of the output sequence.
+ *
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>
+ *                        and \c InputIterator1's \c value_type is convertible to \c BinaryFunction's \c first_argument_type.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>
+ *                        and \c InputIterator2's \c value_type is convertible to \c BinaryFunction's \c second_argument_type.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam BinaryFunction is a model of <a href="http://www.sgi.com/tech/stl/BinaryFunction.html">Binary Function</a>
+ *                              and \c BinaryFunction's \c result_type is convertible to \c OutputIterator's \c value_type.
+ *
+ *  \pre \p first1 may equal \p result, but the range <tt>[first1, last1)</tt> shall not overlap the range <tt>[result, result + (last1 - first1))</tt> otherwise.
+ *  \pre \p first2 may equal \p result, but the range <tt>[first2, first2 + (last1 - first1))</tt> shall not overlap the range <tt>[result, result + (last1 - first1))</tt> otherwise.
+ *
+ *  The following code snippet demonstrates how to use \p transform
+ *
+ *  \code
+ *  #include <thrust/transform.h>
+ *  #include <thrust/functional.h>
+ *  
+ *  int input1[6] = {-5,  0,  2,  3,  2,  4};
+ *  int input2[6] = { 3,  6, -2,  1,  2,  3};
+ *  int output[6];
+ * 
+ *  thrust::plus<int> op;
+ *
+ *  thrust::transform(input1, input1 + 6, input2, output, op);
+ *
+ *  // output is now {-2,  6,  0,  4,  4,  7};
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/transform.html
+ */
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator,
+         typename BinaryFunction>
+  OutputIterator transform(InputIterator1 first1, InputIterator1 last1,
+                           InputIterator2 first2,
+                           OutputIterator result,
+                           BinaryFunction op);
+
+
+/*! This version of \p transform_if conditionally applies a unary function
+ *  to each element of an input sequence and stores the result in the corresponding 
+ *  position in an output sequence if the corresponding position in the input sequence
+ *  satifies a predicate. Otherwise, the corresponding position in the
+ *  output sequence is not modified.
+ *
+ *  Specifically, for each iterator <tt>i</tt> in the range <tt>[first, last)</tt> the
+ *  predicate <tt>pred(*i)</tt> is evaluated. If this predicate
+ *  evaluates to \c true, the result of <tt>op(*i)</tt> is assigned to <tt>*o</tt>,
+ *  where <tt>o</tt> is the corresponding output iterator in the range
+ *  <tt>[result, result + (last - first) )</tt>. Otherwise, <tt>op(*i)</tt> is
+ *  not evaluated and no assignment occurs. The input and output sequences may coincide,
+ *  resulting in an in-place transformation.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *    
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the input sequence.
+ *  \param last The end of the input sequence.
+ *  \param result The beginning of the output sequence.
+ *  \param op The tranformation operation.
+ *  \param pred The predicate operation.
+ *  \return The end of the output sequence.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *                        and \c InputIterator's \c value_type is convertible to \c Predicate's \c argument_type,
+ *                        and \c InputIterator's \c value_type is convertible to \c UnaryFunction's \c argument_type.
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>.
+ *  \tparam UnaryFunction is a model of <a href="http://www.sgi.com/tech/stl/UnaryFunction.html">Unary Function</a>
+ *                        and \c UnaryFunction's \c result_type is convertible to \c OutputIterator's \c value_type.
+ *  \tparam Predicate is a model of <a href="http://www.sgi.com/tech/stl/Predicate.html">Predicate</a>.
+ *
+ *  \pre \p first may equal \p result, but the range <tt>[first, last)</tt> shall not overlap the range <tt>[result, result + (last - first))</tt> otherwise.
+ *
+ *  The following code snippet demonstrates how to use \p transform_if to negate the odd-valued
+ *  elements of a range using the \p thrust::host execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/transform.h>
+ *  #include <thrust/functional.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  
+ *  int data[10]    = {-5, 0, 2, -3, 2, 4, 0, -1, 2, 8};
+ *
+ *  struct is_odd
+ *  {
+ *    __host__ __device__
+ *    bool operator()(int x)
+ *    {
+ *      return x % 2;
+ *    }
+ *  };
+ * 
+ *  thrust::negate<int> op;
+ *  thrust::identity<int> identity;
+ *
+ *  // negate odd elements
+ *  thrust::transform_if(thrust::host, data, data + 10, data, op, is_odd()); // in-place transformation
+ *
+ *  // data is now {5, 0, 2, 3, 2, 4, 0, 1, 2, 8};
+ *  \endcode
+ *
+ *  \see thrust::transform
+ */
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename ForwardIterator,
+         typename UnaryFunction,
+         typename Predicate>
+  ForwardIterator transform_if(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                               InputIterator first, InputIterator last,
+                               ForwardIterator result,
+                               UnaryFunction op,
+                               Predicate pred);
+
+
+/*! This version of \p transform_if conditionally applies a unary function
+ *  to each element of an input sequence and stores the result in the corresponding 
+ *  position in an output sequence if the corresponding position in the input sequence
+ *  satifies a predicate. Otherwise, the corresponding position in the
+ *  output sequence is not modified.
+ *
+ *  Specifically, for each iterator <tt>i</tt> in the range <tt>[first, last)</tt> the
+ *  predicate <tt>pred(*i)</tt> is evaluated. If this predicate
+ *  evaluates to \c true, the result of <tt>op(*i)</tt> is assigned to <tt>*o</tt>,
+ *  where <tt>o</tt> is the corresponding output iterator in the range
+ *  <tt>[result, result + (last - first) )</tt>. Otherwise, <tt>op(*i)</tt> is
+ *  not evaluated and no assignment occurs. The input and output sequences may coincide,
+ *  resulting in an in-place transformation.
+ *    
+ *  \param first The beginning of the input sequence.
+ *  \param last The end of the input sequence.
+ *  \param result The beginning of the output sequence.
+ *  \param op The tranformation operation.
+ *  \param pred The predicate operation.
+ *  \return The end of the output sequence.
+ *
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *                        and \c InputIterator's \c value_type is convertible to \c Predicate's \c argument_type,
+ *                        and \c InputIterator's \c value_type is convertible to \c UnaryFunction's \c argument_type.
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>.
+ *  \tparam UnaryFunction is a model of <a href="http://www.sgi.com/tech/stl/UnaryFunction.html">Unary Function</a>
+ *                        and \c UnaryFunction's \c result_type is convertible to \c OutputIterator's \c value_type.
+ *  \tparam Predicate is a model of <a href="http://www.sgi.com/tech/stl/Predicate.html">Predicate</a>.
+ *
+ *  \pre \p first may equal \p result, but the range <tt>[first, last)</tt> shall not overlap the range <tt>[result, result + (last - first))</tt> otherwise.
+ *
+ *  The following code snippet demonstrates how to use \p transform_if:
+ *
+ *  \code
+ *  #include <thrust/transform.h>
+ *  #include <thrust/functional.h>
+ *  
+ *  int data[10]    = {-5, 0, 2, -3, 2, 4, 0, -1, 2, 8};
+ *
+ *  struct is_odd
+ *  {
+ *    __host__ __device__
+ *    bool operator()(int x)
+ *    {
+ *      return x % 2;
+ *    }
+ *  };
+ * 
+ *  thrust::negate<int> op;
+ *  thrust::identity<int> identity;
+ *
+ *  // negate odd elements
+ *  thrust::transform_if(data, data + 10, data, op, is_odd()); // in-place transformation
+ *
+ *  // data is now {5, 0, 2, 3, 2, 4, 0, 1, 2, 8};
+ *  \endcode
+ *
+ *  \see thrust::transform
+ */
+template<typename InputIterator,
+         typename ForwardIterator,
+         typename UnaryFunction,
+         typename Predicate>
+  ForwardIterator transform_if(InputIterator first, InputIterator last,
+                               ForwardIterator result,
+                               UnaryFunction op,
+                               Predicate pred);
+
+
+/*! This version of \p transform_if conditionally applies a unary function
+ *  to each element of an input sequence and stores the result in the corresponding 
+ *  position in an output sequence if the corresponding position in a stencil sequence
+ *  satisfies a predicate. Otherwise, the corresponding position in the
+ *  output sequence is not modified.
+ *
+ *  Specifically, for each iterator <tt>i</tt> in the range <tt>[first, last)</tt> the
+ *  predicate <tt>pred(*s)</tt> is evaluated, where <tt>s</tt> is the corresponding input
+ *  iterator in the range <tt>[stencil, stencil + (last - first) )</tt>. If this predicate
+ *  evaluates to \c true, the result of <tt>op(*i)</tt> is assigned to <tt>*o</tt>,
+ *  where <tt>o</tt> is the corresponding output iterator in the range
+ *  <tt>[result, result + (last - first) )</tt>. Otherwise, <tt>op(*i)</tt> is
+ *  not evaluated and no assignment occurs. The input and output sequences may coincide,
+ *  resulting in an in-place transformation.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *    
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the input sequence.
+ *  \param last The end of the input sequence.
+ *  \param stencil The beginning of the stencil sequence.
+ *  \param result The beginning of the output sequence.
+ *  \param op The tranformation operation.
+ *  \param pred The predicate operation.
+ *  \return The end of the output sequence.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>
+ *                         and \c InputIterator1's \c value_type is convertible to \c UnaryFunction's \c argument_type.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>
+ *                         and \c InputIterator2's \c value_type is convertible to \c Predicate's \c argument_type.
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>.
+ *  \tparam UnaryFunction is a model of <a href="http://www.sgi.com/tech/stl/UnaryFunction.html">Unary Function</a>
+ *                        and \c UnaryFunction's \c result_type is convertible to \c OutputIterator's \c value_type.
+ *  \tparam Predicate is a model of <a href="http://www.sgi.com/tech/stl/Predicate.html">Predicate</a>.
+ *
+ *  \pre \p first may equal \p result, but the range <tt>[first, last)</tt> shall not overlap the range <tt>[result, result + (last - first))</tt> otherwise.
+ *  \pre \p stencil may equal \p result, but the range <tt>[stencil, stencil + (last - first))</tt> shall not overlap the range <tt>[result, result + (last - first))</tt> otherwise.
+ *
+ *  The following code snippet demonstrates how to use \p transform_if using the \p thrust::host
+ *  execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/transform.h>
+ *  #include <thrust/functional.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  
+ *  int data[10]    = {-5, 0, 2, -3, 2, 4, 0, -1, 2, 8};
+ *  int stencil[10] = { 1, 0, 1,  0, 1, 0, 1,  0, 1, 0};
+ * 
+ *  thrust::negate<int> op;
+ *  thrust::identity<int> identity;
+ *
+ *  thrust::transform_if(thrust::host, data, data + 10, stencil, data, op, identity); // in-place transformation
+ *
+ *  // data is now {5, 0, -2, -3, -2,  4, 0, -1, -2,  8};
+ *  \endcode
+ *
+ *  \see thrust::transform
+ */
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename ForwardIterator,
+         typename UnaryFunction,
+         typename Predicate>
+  ForwardIterator transform_if(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                               InputIterator1 first, InputIterator1 last,
+                               InputIterator2 stencil,
+                               ForwardIterator result,
+                               UnaryFunction op,
+                               Predicate pred);
+
+
+/*! This version of \p transform_if conditionally applies a unary function
+ *  to each element of an input sequence and stores the result in the corresponding 
+ *  position in an output sequence if the corresponding position in a stencil sequence
+ *  satisfies a predicate. Otherwise, the corresponding position in the
+ *  output sequence is not modified.
+ *
+ *  Specifically, for each iterator <tt>i</tt> in the range <tt>[first, last)</tt> the
+ *  predicate <tt>pred(*s)</tt> is evaluated, where <tt>s</tt> is the corresponding input
+ *  iterator in the range <tt>[stencil, stencil + (last - first) )</tt>. If this predicate
+ *  evaluates to \c true, the result of <tt>op(*i)</tt> is assigned to <tt>*o</tt>,
+ *  where <tt>o</tt> is the corresponding output iterator in the range
+ *  <tt>[result, result + (last - first) )</tt>. Otherwise, <tt>op(*i)</tt> is
+ *  not evaluated and no assignment occurs. The input and output sequences may coincide,
+ *  resulting in an in-place transformation.
+ *    
+ *  \param first The beginning of the input sequence.
+ *  \param last The end of the input sequence.
+ *  \param stencil The beginning of the stencil sequence.
+ *  \param result The beginning of the output sequence.
+ *  \param op The tranformation operation.
+ *  \param pred The predicate operation.
+ *  \return The end of the output sequence.
+ *
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>
+ *                         and \c InputIterator1's \c value_type is convertible to \c UnaryFunction's \c argument_type.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>
+ *                         and \c InputIterator2's \c value_type is convertible to \c Predicate's \c argument_type.
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>.
+ *  \tparam UnaryFunction is a model of <a href="http://www.sgi.com/tech/stl/UnaryFunction.html">Unary Function</a>
+ *                        and \c UnaryFunction's \c result_type is convertible to \c OutputIterator's \c value_type.
+ *  \tparam Predicate is a model of <a href="http://www.sgi.com/tech/stl/Predicate.html">Predicate</a>.
+ *
+ *  \pre \p first may equal \p result, but the range <tt>[first, last)</tt> shall not overlap the range <tt>[result, result + (last - first))</tt> otherwise.
+ *  \pre \p stencil may equal \p result, but the range <tt>[stencil, stencil + (last - first))</tt> shall not overlap the range <tt>[result, result + (last - first))</tt> otherwise.
+ *
+ *  The following code snippet demonstrates how to use \p transform_if:
+ *
+ *  \code
+ *  #include <thrust/transform.h>
+ *  #include <thrust/functional.h>
+ *  
+ *  int data[10]    = {-5, 0, 2, -3, 2, 4, 0, -1, 2, 8};
+ *  int stencil[10] = { 1, 0, 1,  0, 1, 0, 1,  0, 1, 0};
+ * 
+ *  thrust::negate<int> op;
+ *  thrust::identity<int> identity;
+ *
+ *  thrust::transform_if(data, data + 10, stencil, data, op, identity); // in-place transformation
+ *
+ *  // data is now {5, 0, -2, -3, -2,  4, 0, -1, -2,  8};
+ *  \endcode
+ *
+ *  \see thrust::transform
+ */
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename ForwardIterator,
+         typename UnaryFunction,
+         typename Predicate>
+  ForwardIterator transform_if(InputIterator1 first, InputIterator1 last,
+                               InputIterator2 stencil,
+                               ForwardIterator result,
+                               UnaryFunction op,
+                               Predicate pred);
+
+
+/*! This version of \p transform_if conditionally applies a binary function
+ *  to each pair of elements from two input sequences and stores the result in the corresponding 
+ *  position in an output sequence if the corresponding position in a stencil sequence
+ *  satifies a predicate. Otherwise, the corresponding position in the
+ *  output sequence is not modified.
+ *
+ *  Specifically, for each iterator <tt>i</tt> in the range <tt>[first1, last1)</tt> and 
+ *  <tt>j = first2 + (i - first1)</tt> in the range <tt>[first2, first2 + (last1 - first1) )</tt>,
+ *  the predicate <tt>pred(*s)</tt> is evaluated, where <tt>s</tt> is the corresponding input
+ *  iterator in the range <tt>[stencil, stencil + (last1 - first1) )</tt>. If this predicate
+ *  evaluates to \c true, the result of <tt>binary_op(*i,*j)</tt> is assigned to <tt>*o</tt>,
+ *  where <tt>o</tt> is the corresponding output iterator in the range
+ *  <tt>[result, result + (last1 - first1) )</tt>. Otherwise, <tt>binary_op(*i,*j)</tt> is
+ *  not evaluated and no assignment occurs. The input and output sequences may coincide,
+ *  resulting in an in-place transformation.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *    
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first1 The beginning of the first input sequence.
+ *  \param last1 The end of the first input sequence.
+ *  \param first2 The beginning of the second input sequence.
+ *  \param stencil The beginning of the stencil sequence.
+ *  \param result The beginning of the output sequence.
+ *  \param binary_op The transformation operation.
+ *  \param pred The predicate operation.
+ *  \return The end of the output sequence.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>
+ *                         and \c InputIterator1's \c value_type is convertible to \c BinaryFunction's \c first_argument_type.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>
+ *                         and \c InputIterator2's \c value_type is convertible to \c BinaryFunction's \c second_argument_type.
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>.
+ *  \tparam BinaryFunction is a model of <a href="http://www.sgi.com/tech/stl/BinaryFunction.html">Binary Function</a>
+ *                         and \c BinaryFunction's \c result_type is convertible to \c OutputIterator's \c value_type.
+ *  \tparam Predicate is a model of <a href="http://www.sgi.com/tech/stl/Predicate.html">Predicate</a>.
+ *
+ *  \pre \p first1 may equal \p result, but the range <tt>[first1, last1)</tt> shall not overlap the range <tt>[result, result + (last1 - first1))</tt> otherwise.
+ *  \pre \p first2 may equal \p result, but the range <tt>[first2, first2 + (last1 - first1))</tt> shall not overlap the range <tt>[result, result + (last1 - first1))</tt> otherwise.
+ *  \pre \p stencil may equal \p result, but the range <tt>[stencil, stencil + (last1 - first1))</tt> shall not overlap the range <tt>[result, result + (last1 - first1))</tt> otherwise.
+ *
+ *  The following code snippet demonstrates how to use \p transform_if using the \p thrust::host
+ *  execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/transform.h>
+ *  #include <thrust/functional.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  
+ *  int input1[6]  = {-5,  0,  2,  3,  2,  4};
+ *  int input2[6]  = { 3,  6, -2,  1,  2,  3};
+ *  int stencil[8] = { 1,  0,  1,  0,  1,  0};
+ *  int output[6];
+ * 
+ *  thrust::plus<int> op;
+ *  thrust::identity<int> identity;
+ *
+ *  thrust::transform_if(thrust::host, input1, input1 + 6, input2, stencil, output, op, identity);
+ *
+ *  // output is now {-2,  0,  0,  3,  4,  4};
+ *  \endcode
+ *
+ *  \see thrust::transform
+ */
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename ForwardIterator,
+         typename BinaryFunction,
+         typename Predicate>
+  ForwardIterator transform_if(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                               InputIterator1 first1, InputIterator1 last1,
+                               InputIterator2 first2,
+                               InputIterator3 stencil,
+                               ForwardIterator result,
+                               BinaryFunction binary_op,
+                               Predicate pred);
+
+
+/*! This version of \p transform_if conditionally applies a binary function
+ *  to each pair of elements from two input sequences and stores the result in the corresponding 
+ *  position in an output sequence if the corresponding position in a stencil sequence
+ *  satifies a predicate. Otherwise, the corresponding position in the
+ *  output sequence is not modified.
+ *
+ *  Specifically, for each iterator <tt>i</tt> in the range <tt>[first1, last1)</tt> and 
+ *  <tt>j = first2 + (i - first1)</tt> in the range <tt>[first2, first2 + (last1 - first1) )</tt>,
+ *  the predicate <tt>pred(*s)</tt> is evaluated, where <tt>s</tt> is the corresponding input
+ *  iterator in the range <tt>[stencil, stencil + (last1 - first1) )</tt>. If this predicate
+ *  evaluates to \c true, the result of <tt>binary_op(*i,*j)</tt> is assigned to <tt>*o</tt>,
+ *  where <tt>o</tt> is the corresponding output iterator in the range
+ *  <tt>[result, result + (last1 - first1) )</tt>. Otherwise, <tt>binary_op(*i,*j)</tt> is
+ *  not evaluated and no assignment occurs. The input and output sequences may coincide,
+ *  resulting in an in-place transformation.
+ *    
+ *  \param first1 The beginning of the first input sequence.
+ *  \param last1 The end of the first input sequence.
+ *  \param first2 The beginning of the second input sequence.
+ *  \param stencil The beginning of the stencil sequence.
+ *  \param result The beginning of the output sequence.
+ *  \param binary_op The transformation operation.
+ *  \param pred The predicate operation.
+ *  \return The end of the output sequence.
+ *
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>
+ *                         and \c InputIterator1's \c value_type is convertible to \c BinaryFunction's \c first_argument_type.
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>
+ *                         and \c InputIterator2's \c value_type is convertible to \c BinaryFunction's \c second_argument_type.
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>.
+ *  \tparam BinaryFunction is a model of <a href="http://www.sgi.com/tech/stl/BinaryFunction.html">Binary Function</a>
+ *                         and \c BinaryFunction's \c result_type is convertible to \c OutputIterator's \c value_type.
+ *  \tparam Predicate is a model of <a href="http://www.sgi.com/tech/stl/Predicate.html">Predicate</a>.
+ *
+ *  \pre \p first1 may equal \p result, but the range <tt>[first1, last1)</tt> shall not overlap the range <tt>[result, result + (last1 - first1))</tt> otherwise.
+ *  \pre \p first2 may equal \p result, but the range <tt>[first2, first2 + (last1 - first1))</tt> shall not overlap the range <tt>[result, result + (last1 - first1))</tt> otherwise.
+ *  \pre \p stencil may equal \p result, but the range <tt>[stencil, stencil + (last1 - first1))</tt> shall not overlap the range <tt>[result, result + (last1 - first1))</tt> otherwise.
+ *
+ *  The following code snippet demonstrates how to use \p transform_if:
+ *
+ *  \code
+ *  #include <thrust/transform.h>
+ *  #include <thrust/functional.h>
+ *  
+ *  int input1[6]  = {-5,  0,  2,  3,  2,  4};
+ *  int input2[6]  = { 3,  6, -2,  1,  2,  3};
+ *  int stencil[8] = { 1,  0,  1,  0,  1,  0};
+ *  int output[6];
+ * 
+ *  thrust::plus<int> op;
+ *  thrust::identity<int> identity;
+ *
+ *  thrust::transform_if(input1, input1 + 6, input2, stencil, output, op, identity);
+ *
+ *  // output is now {-2,  0,  0,  3,  4,  4};
+ *  \endcode
+ *
+ *  \see thrust::transform
+ */
+template<typename InputIterator1,
+         typename InputIterator2,
+         typename InputIterator3,
+         typename ForwardIterator,
+         typename BinaryFunction,
+         typename Predicate>
+  ForwardIterator transform_if(InputIterator1 first1, InputIterator1 last1,
+                               InputIterator2 first2,
+                               InputIterator3 stencil,
+                               ForwardIterator result,
+                               BinaryFunction binary_op,
+                               Predicate pred);
+
+
+/*! \} // end transformations
+ */
+
+
+} // end namespace thrust
+
+#include <thrust/detail/transform.inl>
+
diff --git a/compat/thrust/transform_reduce.h b/compat/thrust/transform_reduce.h
new file mode 100644
index 0000000000..3ef5efd6aa
--- /dev/null
+++ b/compat/thrust/transform_reduce.h
@@ -0,0 +1,197 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file transform_reduce.h
+ *  \brief Fused transform / reduction
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/execution_policy.h>
+
+namespace thrust
+{
+
+
+/*! \addtogroup reductions
+ *  \{
+ *  \addtogroup transformed_reductions Transformed Reductions
+ *  \ingroup reductions
+ *  \{
+ */
+
+
+/*! \p transform_reduce fuses the \p transform and \p reduce operations.
+ *  \p transform_reduce is equivalent to performing a transformation defined by
+ *  \p unary_op into a temporary sequence and then performing \p reduce on the
+ *  transformed sequence. In most cases, fusing these two operations together is
+ *  more efficient, since fewer memory reads and writes are required.
+ *
+ *  \p transform_reduce performs a reduction on the transformation of the
+ *  sequence <tt>[first, last)</tt> according to \p unary_op. Specifically,
+ *  \p unary_op is applied to each element of the sequence and then the result
+ *  is reduced to a single value with \p binary_op using the initial value 
+ *  \p init.  Note that the transformation \p unary_op is not applied to 
+ *  the initial value \p init.  The order of reduction is not specified, 
+ *  so \p binary_op must be both commutative and associative. 
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the sequence.
+ *  \param last The end of the sequence.
+ *  \param unary_op The function to apply to each element of the input sequence.
+ *  \param init The result is initialized to this value.
+ *  \param binary_op The reduction operation.
+ *  \return The result of the transformed reduction.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator's \c value_type is convertible to \p UnaryFunction's \c argument_type.
+ *  \tparam UnaryFunction is a model of <a href="http://www.sgi.com/tech/stl/UnaryFunction.html">Unary Function</a>,
+ *          and \p UnaryFunction's \c result_type is convertible to \c OutputType.
+ *  \tparam OutputType is a model of <a href="http://www.sgi.com/tech/stl/Assignable.html">Assignable</a>,
+ *          and is convertible to \p BinaryFunction's \c first_argument_type and \c second_argument_type.
+ *  \tparam BinaryFunction is a model of <a href="http://www.sgi.com/tech/stl/BinaryFunction.html">Binary Function</a>,
+ *          and \p BinaryFunction's \c result_type is convertible to \p OutputType.
+ *
+ *  The following code snippet demonstrates how to use \p transform_reduce
+ *  to compute the maximum value of the absolute value of the elements
+ *  of a range using the \p thrust::host execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/transform_reduce.h>
+ *  #include <thrust/functional.h>
+ *  #include <thrust/execution_policy.h>
+ *
+ *  template<typename T>
+ *  struct absolute_value : public unary_function<T,T>
+ *  {
+ *    __host__ __device__ T operator()(const T &x) const
+ *    {
+ *      return x < T(0) ? -x : x;
+ *    }
+ *  };
+ *
+ *  ...
+ *
+ *  int data[6] = {-1, 0, -2, -2, 1, -3};
+ *  int result = thrust::transform_reduce(thrust::host,
+ *                                        data, data + 6,
+ *                                        absolute_value<int>(),
+ *                                        0,
+ *                                        thrust::maximum<int>());
+ *  // result == 3
+ *  \endcode
+ *
+ *  \see \c transform
+ *  \see \c reduce
+ */
+template<typename DerivedPolicy,
+         typename InputIterator, 
+         typename UnaryFunction, 
+         typename OutputType,
+         typename BinaryFunction>
+  OutputType transform_reduce(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                              InputIterator first,
+                              InputIterator last,
+                              UnaryFunction unary_op,
+                              OutputType init,
+                              BinaryFunction binary_op);
+
+
+/*! \p transform_reduce fuses the \p transform and \p reduce operations.
+ *  \p transform_reduce is equivalent to performing a transformation defined by
+ *  \p unary_op into a temporary sequence and then performing \p reduce on the
+ *  transformed sequence. In most cases, fusing these two operations together is
+ *  more efficient, since fewer memory reads and writes are required.
+ *
+ *  \p transform_reduce performs a reduction on the transformation of the
+ *  sequence <tt>[first, last)</tt> according to \p unary_op. Specifically,
+ *  \p unary_op is applied to each element of the sequence and then the result
+ *  is reduced to a single value with \p binary_op using the initial value 
+ *  \p init.  Note that the transformation \p unary_op is not applied to 
+ *  the initial value \p init.  The order of reduction is not specified, 
+ *  so \p binary_op must be both commutative and associative. 
+ *
+ *  \param first The beginning of the sequence.
+ *  \param last The end of the sequence.
+ *  \param unary_op The function to apply to each element of the input sequence.
+ *  \param init The result is initialized to this value.
+ *  \param binary_op The reduction operation.
+ *  \return The result of the transformed reduction.
+ *
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator's \c value_type is convertible to \p UnaryFunction's \c argument_type.
+ *  \tparam UnaryFunction is a model of <a href="http://www.sgi.com/tech/stl/UnaryFunction.html">Unary Function</a>,
+ *          and \p UnaryFunction's \c result_type is convertible to \c OutputType.
+ *  \tparam OutputType is a model of <a href="http://www.sgi.com/tech/stl/Assignable.html">Assignable</a>,
+ *          and is convertible to \p BinaryFunction's \c first_argument_type and \c second_argument_type.
+ *  \tparam BinaryFunction is a model of <a href="http://www.sgi.com/tech/stl/BinaryFunction.html">Binary Function</a>,
+ *          and \p BinaryFunction's \c result_type is convertible to \p OutputType.
+ *
+ *  The following code snippet demonstrates how to use \p transform_reduce
+ *  to compute the maximum value of the absolute value of the elements
+ *  of a range.
+ *
+ *  \code
+ *  #include <thrust/transform_reduce.h>
+ *  #include <thrust/functional.h>
+ *
+ *  template<typename T>
+ *  struct absolute_value : public unary_function<T,T>
+ *  {
+ *    __host__ __device__ T operator()(const T &x) const
+ *    {
+ *      return x < T(0) ? -x : x;
+ *    }
+ *  };
+ *
+ *  ...
+ *
+ *  int data[6] = {-1, 0, -2, -2, 1, -3};
+ *  int result = thrust::transform_reduce(data, data + 6,
+ *                                        absolute_value<int>(),
+ *                                        0,
+ *                                        thrust::maximum<int>());
+ *  // result == 3
+ *  \endcode
+ *
+ *  \see \c transform
+ *  \see \c reduce
+ */
+template<typename InputIterator, 
+         typename UnaryFunction, 
+         typename OutputType,
+         typename BinaryFunction>
+  OutputType transform_reduce(InputIterator first,
+                              InputIterator last,
+                              UnaryFunction unary_op,
+                              OutputType init,
+                              BinaryFunction binary_op);
+
+
+/*! \} // end transformed_reductions
+ *  \} // end reductions
+ */
+
+
+} // end namespace thrust
+
+#include <thrust/detail/transform_reduce.inl>
+
diff --git a/compat/thrust/transform_scan.h b/compat/thrust/transform_scan.h
new file mode 100644
index 0000000000..e9943e401f
--- /dev/null
+++ b/compat/thrust/transform_scan.h
@@ -0,0 +1,322 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file transform_scan.h
+ *  \brief Fused transform / prefix-sum
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/execution_policy.h>
+
+namespace thrust
+{
+
+
+/*! \addtogroup algorithms
+ */
+
+/*! \addtogroup prefixsums Prefix Sums
+ *  \ingroup algorithms
+ *  \{
+ */
+	
+/*! \addtogroup transformed_prefixsums Transformed Prefix Sums
+ *  \ingroup prefixsums
+ *  \{
+ */
+
+
+/*! \p transform_inclusive_scan fuses the \p transform and \p inclusive_scan
+ *  operations.  \p transform_inclusive_scan is equivalent to performing a
+ *  tranformation defined by \p unary_op into a temporary sequence and then
+ *  performing an \p inclusive_scan on the tranformed sequence.  In most
+ *  cases, fusing these two operations together is more efficient, since
+ *  fewer memory reads and writes are required. In \p transform_inclusive_scan,
+ *  <tt>unary_op(\*first)</tt> is assigned to <tt>\*result</tt> and the result
+ *  of <tt>binary_op(unary_op(\*first), unary_op(\*(first + 1)))</tt> is
+ *  assigned to <tt>\*(result + 1)</tt>, and so on.  The transform scan
+ *  operation is permitted to be in-place.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the input sequence.
+ *  \param last The end of the input sequence.
+ *  \param result The beginning of the output sequence.
+ *  \param unary_op The function used to tranform the input sequence.
+ *  \param binary_op The associatve operator used to 'sum' transformed values.
+ *  \return The end of the output sequence.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>
+ *                               and \c InputIterator's \c value_type is convertible to \c unary_op's input type.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam UnaryFunction is a model of <a href="http://www.sgi.com/tech/stl/UnaryFunction.html">Unary Function</a>
+ *                               and accepts inputs of \c InputIterator's \c value_type.  \c UnaryFunction's result_type
+ *                               is convertable to \c OutputIterator's \c value_type.
+ *  \tparam AssociativeOperator is a model of <a href="http://www.sgi.com/tech/stl/BinaryFunction.html">Binary Function</a>
+ *                              and \c AssociativeOperator's \c result_type is
+ *                              convertible to \c OutputIterator's \c value_type.
+ *
+ *  \pre \p first may equal \p result, but the range <tt>[first, last)</tt> and the range <tt>[result, result + (last - first))</tt> shall not overlap otherwise.
+ *
+ *  The following code snippet demonstrates how to use \p transform_inclusive_scan using the
+ *  \p thrust::host execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/transform_scan.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  
+ *  int data[6] = {1, 0, 2, 2, 1, 3};
+ *
+ *  thrust::negate<int> unary_op;
+ *  thrust::plus<int> binary_op;
+ *
+ *  thrust::transform_inclusive_scan(thrust::host, data, data + 6, data, unary_op, binary_op); // in-place scan
+ *
+ *  // data is now {-1, -1, -3, -5, -6, -9}
+ *  \endcode
+ *
+ *  \see \p transform
+ *  \see \p inclusive_scan
+ *
+ */
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator,
+         typename UnaryFunction,
+         typename AssociativeOperator>
+  OutputIterator transform_inclusive_scan(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                          InputIterator first,
+                                          InputIterator last,
+                                          OutputIterator result,
+                                          UnaryFunction unary_op,
+                                          AssociativeOperator binary_op);
+
+
+/*! \p transform_inclusive_scan fuses the \p transform and \p inclusive_scan
+ *  operations.  \p transform_inclusive_scan is equivalent to performing a
+ *  tranformation defined by \p unary_op into a temporary sequence and then
+ *  performing an \p inclusive_scan on the tranformed sequence.  In most
+ *  cases, fusing these two operations together is more efficient, since
+ *  fewer memory reads and writes are required. In \p transform_inclusive_scan,
+ *  <tt>unary_op(\*first)</tt> is assigned to <tt>\*result</tt> and the result
+ *  of <tt>binary_op(unary_op(\*first), unary_op(\*(first + 1)))</tt> is
+ *  assigned to <tt>\*(result + 1)</tt>, and so on.  The transform scan
+ *  operation is permitted to be in-place.
+ *
+ *  \param first The beginning of the input sequence.
+ *  \param last The end of the input sequence.
+ *  \param result The beginning of the output sequence.
+ *  \param unary_op The function used to tranform the input sequence.
+ *  \param binary_op The associatve operator used to 'sum' transformed values.
+ *  \return The end of the output sequence.
+ *
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>
+ *                               and \c InputIterator's \c value_type is convertible to \c unary_op's input type.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam UnaryFunction is a model of <a href="http://www.sgi.com/tech/stl/UnaryFunction.html">Unary Function</a>
+ *                               and accepts inputs of \c InputIterator's \c value_type.  \c UnaryFunction's result_type
+ *                               is convertable to \c OutputIterator's \c value_type.
+ *  \tparam AssociativeOperator is a model of <a href="http://www.sgi.com/tech/stl/BinaryFunction.html">Binary Function</a>
+ *                              and \c AssociativeOperator's \c result_type is
+ *                              convertible to \c OutputIterator's \c value_type.
+ *
+ *  \pre \p first may equal \p result, but the range <tt>[first, last)</tt> and the range <tt>[result, result + (last - first))</tt> shall not overlap otherwise.
+ *
+ *  The following code snippet demonstrates how to use \p transform_inclusive_scan
+ *
+ *  \code
+ *  #include <thrust/transform_scan.h>
+ *  
+ *  int data[6] = {1, 0, 2, 2, 1, 3};
+ *
+ *  thrust::negate<int> unary_op;
+ *  thrust::plus<int> binary_op;
+ *
+ *  thrust::transform_inclusive_scan(data, data + 6, data, unary_op, binary_op); // in-place scan
+ *
+ *  // data is now {-1, -1, -3, -5, -6, -9}
+ *  \endcode
+ *
+ *  \see \p transform
+ *  \see \p inclusive_scan
+ *
+ */
+template<typename InputIterator,
+         typename OutputIterator,
+         typename UnaryFunction,
+         typename AssociativeOperator>
+  OutputIterator transform_inclusive_scan(InputIterator first,
+                                          InputIterator last,
+                                          OutputIterator result,
+                                          UnaryFunction unary_op,
+                                          AssociativeOperator binary_op);
+
+
+/*! \p transform_exclusive_scan fuses the \p transform and \p exclusive_scan
+ *  operations.  \p transform_exclusive_scan is equivalent to performing a
+ *  tranformation defined by \p unary_op into a temporary sequence and then
+ *  performing an \p exclusive_scan on the tranformed sequence.  In most
+ *  cases, fusing these two operations together is more efficient, since
+ *  fewer memory reads and writes are required. In 
+ *  \p transform_exclusive_scan, \p init is assigned to <tt>\*result</tt> 
+ *  and the result of <tt>binary_op(init, unary_op(\*first))</tt> is assigned
+ *  to <tt>\*(result + 1)</tt>, and so on.  The transform scan operation is 
+ *  permitted to be in-place.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the input sequence.
+ *  \param last The end of the input sequence.
+ *  \param result The beginning of the output sequence.
+ *  \param unary_op The function used to tranform the input sequence.
+ *  \param init The initial value of the \p exclusive_scan
+ *  \param binary_op The associatve operator used to 'sum' transformed values.
+ *  \return The end of the output sequence.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>
+ *                               and \c InputIterator's \c value_type is convertible to \c unary_op's input type.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam UnaryFunction is a model of <a href="http://www.sgi.com/tech/stl/UnaryFunction.html">Unary Function</a>
+ *                               and accepts inputs of \c InputIterator's \c value_type.  \c UnaryFunction's result_type
+ *                               is convertable to \c OutputIterator's \c value_type.
+ *  \tparam T is convertible to \c OutputIterator's \c value_type.
+ *  \tparam AssociativeOperator is a model of <a href="http://www.sgi.com/tech/stl/BinaryFunction.html">Binary Function</a>
+ *                              and \c AssociativeOperator's \c result_type is
+ *                              convertible to \c OutputIterator's \c value_type.
+ *
+ *  \pre \p first may equal \p result, but the range <tt>[first, last)</tt> and the range <tt>[result, result + (last - first))</tt> shall not overlap otherwise.
+ *
+ *  The following code snippet demonstrates how to use \p transform_exclusive_scan using the
+ *  \p thrust::host execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/transform_scan.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  
+ *  int data[6] = {1, 0, 2, 2, 1, 3};
+ *
+ *  thrust::negate<int> unary_op;
+ *  thrust::plus<int> binary_op;
+ *
+ *  thrust::transform_exclusive_scan(thrust::host, data, data + 6, data, unary_op, 4, binary_op); // in-place scan
+ *
+ *  // data is now {4, 3, 3, 1, -1, -2}
+ *  \endcode
+ *
+ *  \see \p transform
+ *  \see \p exclusive_scan
+ *
+ */
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator,
+         typename UnaryFunction,
+         typename T,
+         typename AssociativeOperator>
+  OutputIterator transform_exclusive_scan(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                          InputIterator first,
+                                          InputIterator last,
+                                          OutputIterator result,
+                                          UnaryFunction unary_op,
+                                          T init,
+                                          AssociativeOperator binary_op);
+
+
+/*! \p transform_exclusive_scan fuses the \p transform and \p exclusive_scan
+ *  operations.  \p transform_exclusive_scan is equivalent to performing a
+ *  tranformation defined by \p unary_op into a temporary sequence and then
+ *  performing an \p exclusive_scan on the tranformed sequence.  In most
+ *  cases, fusing these two operations together is more efficient, since
+ *  fewer memory reads and writes are required. In 
+ *  \p transform_exclusive_scan, \p init is assigned to <tt>\*result</tt> 
+ *  and the result of <tt>binary_op(init, unary_op(\*first))</tt> is assigned
+ *  to <tt>\*(result + 1)</tt>, and so on.  The transform scan operation is 
+ *  permitted to be in-place.
+ *
+ *  \param first The beginning of the input sequence.
+ *  \param last The end of the input sequence.
+ *  \param result The beginning of the output sequence.
+ *  \param unary_op The function used to tranform the input sequence.
+ *  \param init The initial value of the \p exclusive_scan
+ *  \param binary_op The associatve operator used to 'sum' transformed values.
+ *  \return The end of the output sequence.
+ *
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>
+ *                               and \c InputIterator's \c value_type is convertible to \c unary_op's input type.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>.
+ *  \tparam UnaryFunction is a model of <a href="http://www.sgi.com/tech/stl/UnaryFunction.html">Unary Function</a>
+ *                               and accepts inputs of \c InputIterator's \c value_type.  \c UnaryFunction's result_type
+ *                               is convertable to \c OutputIterator's \c value_type.
+ *  \tparam T is convertible to \c OutputIterator's \c value_type.
+ *  \tparam AssociativeOperator is a model of <a href="http://www.sgi.com/tech/stl/BinaryFunction.html">Binary Function</a>
+ *                              and \c AssociativeOperator's \c result_type is
+ *                              convertible to \c OutputIterator's \c value_type.
+ *
+ *  \pre \p first may equal \p result, but the range <tt>[first, last)</tt> and the range <tt>[result, result + (last - first))</tt> shall not overlap otherwise.
+ *
+ *  The following code snippet demonstrates how to use \p transform_exclusive_scan
+ *
+ *  \code
+ *  #include <thrust/transform_scan.h>
+ *  
+ *  int data[6] = {1, 0, 2, 2, 1, 3};
+ *
+ *  thrust::negate<int> unary_op;
+ *  thrust::plus<int> binary_op;
+ *
+ *  thrust::transform_exclusive_scan(data, data + 6, data, unary_op, 4, binary_op); // in-place scan
+ *
+ *  // data is now {4, 3, 3, 1, -1, -2}
+ *  \endcode
+ *
+ *  \see \p transform
+ *  \see \p exclusive_scan
+ *
+ */
+template<typename InputIterator,
+         typename OutputIterator,
+         typename UnaryFunction,
+         typename T,
+         typename AssociativeOperator>
+  OutputIterator transform_exclusive_scan(InputIterator first,
+                                          InputIterator last,
+                                          OutputIterator result,
+                                          UnaryFunction unary_op,
+                                          T init,
+                                          AssociativeOperator binary_op);
+
+
+/*! \} // end transformed_prefixsums
+ */
+
+
+/*! \} // end prefixsums
+ */
+
+	
+} // end namespace thrust
+
+#include <thrust/detail/transform_scan.inl>
+
diff --git a/compat/thrust/tuple.h b/compat/thrust/tuple.h
new file mode 100644
index 0000000000..3961d982fa
--- /dev/null
+++ b/compat/thrust/tuple.h
@@ -0,0 +1,583 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file tuple.h
+ *  \brief A type encapsulating a heterogeneous collection of elements
+ */
+
+/*
+ * Copyright (C) 1999, 2000 Jaakko Järvi (jaakko.jarvi@cs.utu.fi)
+ * 
+ * Distributed under the Boost Software License, Version 1.0.
+ * (See accompanying NOTICE file for the complete license)
+ *
+ * For more information, see http://www.boost.org
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/tuple.inl>
+#include <thrust/pair.h>
+
+namespace thrust
+{
+
+/*! \addtogroup utility
+ *  \{
+ */
+
+/*! \addtogroup tuple
+ *  \{
+ */
+
+/*! \cond
+ */
+
+struct null_type;
+
+/*! \endcond
+ */
+
+/*! This metafunction returns the type of a
+ *  \p tuple's <tt>N</tt>th element.
+ *
+ *  \tparam N This parameter selects the element of interest.
+ *  \tparam T A \c tuple type of interest.
+ *
+ *  \see pair
+ *  \see tuple
+ */
+template<int N, class T>
+  struct tuple_element
+{
+  private:
+    typedef typename T::tail_type Next;
+
+  public:
+    /*! The result of this metafunction is returned in \c type.
+     */
+    typedef typename tuple_element<N-1, Next>::type type;
+}; // end tuple_element
+
+/*! This metafunction returns the number of elements
+ *  of a \p tuple type of interest.
+ *
+ *  \tparam T A \c tuple type of interest.
+ *
+ *  \see pair
+ *  \see tuple
+ */
+template<class T>
+  struct tuple_size
+{
+  /*! The result of this metafunction is returned in \c value.
+   */
+  static const int value = 1 + tuple_size<typename T::tail_type>::value;
+}; // end tuple_size
+
+// get function for non-const cons-lists, returns a reference to the element
+
+/*! The \p get function returns a reference to a \p tuple element of
+ *  interest.
+ *
+ *  \param t A reference to a \p tuple of interest.
+ *  \return A reference to \p t's <tt>N</tt>th element.
+ *
+ *  \tparam N The index of the element of interest.
+ *
+ *  The following code snippet demonstrates how to use \p get to print
+ *  the value of a \p tuple element.
+ *
+ *  \code
+ *  #include <thrust/tuple.h>
+ *  #include <iostream>
+ *  ...
+ *  thrust::tuple<int, const char *> t(13, "thrust");
+ *
+ *  std::cout << "The 1st value of t is " << thrust::get<1>(t) << std::endl;
+ *  \endcode
+ *
+ *  \see pair
+ *  \see tuple
+ */
+template<int N, class HT, class TT>
+__host__ __device__
+inline typename access_traits<
+                  typename tuple_element<N, detail::cons<HT, TT> >::type
+                >::non_const_type
+get(detail::cons<HT, TT>& t);
+
+
+/*! The \p get function returns a \c const reference to a \p tuple element of
+ *  interest.
+ *
+ *  \param t A reference to a \p tuple of interest.
+ *  \return A \c const reference to \p t's <tt>N</tt>th element.
+ *
+ *  \tparam N The index of the element of interest.
+ *
+ *  The following code snippet demonstrates how to use \p get to print
+ *  the value of a \p tuple element.
+ *
+ *  \code
+ *  #include <thrust/tuple.h>
+ *  #include <iostream>
+ *  ...
+ *  thrust::tuple<int, const char *> t(13, "thrust");
+ *
+ *  std::cout << "The 1st value of t is " << thrust::get<1>(t) << std::endl;
+ *  \endcode
+ *
+ *  \see pair
+ *  \see tuple
+ */
+template<int N, class HT, class TT>
+__host__ __device__
+inline typename access_traits<
+                  typename tuple_element<N, detail::cons<HT, TT> >::type
+                >::const_type
+get(const detail::cons<HT, TT>& t);
+
+
+
+/*! \p tuple is a class template that can be instantiated with up to ten arguments.
+ *  Each template argument specifies the type of element in the \p tuple.
+ *  Consequently, tuples are heterogeneous, fixed-size collections of values. An
+ *  instantiation of \p tuple with two arguments is similar to an instantiation
+ *  of \p pair with the same two arguments. Individual elements of a \p tuple may
+ *  be accessed with the \p get function.
+ *
+ *  \tparam TN The type of the <tt>N</tt> \c tuple element. Thrust's \p tuple
+ *          type currently supports up to ten elements.
+ *
+ *  The following code snippet demonstrates how to create a new \p tuple object
+ *  and inspect and modify the value of its elements.
+ *
+ *  \code
+ *  #include <thrust/tuple.h>
+ *  #include <iostream>
+ *  ...
+ *  // create a tuple containing an int, a float, and a string
+ *  thrust::tuple<int, float, const char*> t(13, 0.1f, "thrust");
+ *
+ *  // individual members are accessed with the free function get
+ *  std::cout << "The first element's value is " << thrust::get<0>(t) << std::endl; 
+ *
+ *  // or the member function get
+ *  std::cout << "The second element's value is " << t.get<1>() << std::endl;
+ *
+ *  // we can also modify elements with the same function
+ *  thrust::get<0>(t) += 10;
+ *  \endcode
+ *
+ *  \see pair
+ *  \see get
+ *  \see make_tuple
+ *  \see tuple_element
+ *  \see tuple_size
+ *  \see tie
+ */
+template <class T0, class T1, class T2, class T3, class T4,
+          class T5, class T6, class T7, class T8, class T9>
+  class tuple :
+    public detail::map_tuple_to_cons<T0, T1, T2, T3, T4, T5, T6, T7, T8, T9>::type
+{
+  /*! \cond
+   */
+
+  private:
+  typedef typename detail::map_tuple_to_cons<T0, T1, T2, T3, T4, T5, T6, T7, T8, T9>::type inherited;
+
+  /*! \endcond
+   */
+
+  public:
+  /*! \p tuple's no-argument constructor initializes each element.
+   */
+  inline __host__ __device__
+  tuple(void) {}
+
+  /*! \p tuple's one-argument constructor copy constructs the first element from the given parameter
+   *     and intializes all other elements.
+   *  \param t0 The value to assign to this \p tuple's first element.
+   */
+  inline __host__ __device__ 
+  tuple(typename access_traits<T0>::parameter_type t0)
+    : inherited(t0,
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type())) {}
+
+  /*! \p tuple's one-argument constructor copy constructs the first two elements from the given parameters
+   *     and intializes all other elements.
+   *  \param t0 The value to assign to this \p tuple's first element.
+   *  \param t1 The value to assign to this \p tuple's second element.
+   *  \note \p tuple's constructor has ten variants of this form, the rest of which are ommitted here for brevity.
+   */
+  inline __host__ __device__ 
+  tuple(typename access_traits<T0>::parameter_type t0,
+        typename access_traits<T1>::parameter_type t1)
+    : inherited(t0, t1,
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type())) {}
+
+  /*! \cond
+   */
+
+  inline __host__ __device__ 
+  tuple(typename access_traits<T0>::parameter_type t0,
+        typename access_traits<T1>::parameter_type t1,
+        typename access_traits<T2>::parameter_type t2)
+    : inherited(t0, t1, t2,
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type())) {}
+
+  inline __host__ __device__ 
+  tuple(typename access_traits<T0>::parameter_type t0,
+        typename access_traits<T1>::parameter_type t1,
+        typename access_traits<T2>::parameter_type t2,
+        typename access_traits<T3>::parameter_type t3)
+    : inherited(t0, t1, t2, t3,
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type())) {}
+
+  inline __host__ __device__ 
+  tuple(typename access_traits<T0>::parameter_type t0,
+        typename access_traits<T1>::parameter_type t1,
+        typename access_traits<T2>::parameter_type t2,
+        typename access_traits<T3>::parameter_type t3,
+        typename access_traits<T4>::parameter_type t4)
+    : inherited(t0, t1, t2, t3, t4,
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type())) {}
+
+  inline __host__ __device__ 
+  tuple(typename access_traits<T0>::parameter_type t0,
+        typename access_traits<T1>::parameter_type t1,
+        typename access_traits<T2>::parameter_type t2,
+        typename access_traits<T3>::parameter_type t3,
+        typename access_traits<T4>::parameter_type t4,
+        typename access_traits<T5>::parameter_type t5)
+    : inherited(t0, t1, t2, t3, t4, t5,
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type())) {}
+
+  inline __host__ __device__ 
+  tuple(typename access_traits<T0>::parameter_type t0,
+        typename access_traits<T1>::parameter_type t1,
+        typename access_traits<T2>::parameter_type t2,
+        typename access_traits<T3>::parameter_type t3,
+        typename access_traits<T4>::parameter_type t4,
+        typename access_traits<T5>::parameter_type t5,
+        typename access_traits<T6>::parameter_type t6)
+    : inherited(t0, t1, t2, t3, t4, t5, t6,
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type())) {}
+
+  inline __host__ __device__ 
+  tuple(typename access_traits<T0>::parameter_type t0,
+        typename access_traits<T1>::parameter_type t1,
+        typename access_traits<T2>::parameter_type t2,
+        typename access_traits<T3>::parameter_type t3,
+        typename access_traits<T4>::parameter_type t4,
+        typename access_traits<T5>::parameter_type t5,
+        typename access_traits<T6>::parameter_type t6,
+        typename access_traits<T7>::parameter_type t7)
+    : inherited(t0, t1, t2, t3, t4, t5, t6, t7,
+                static_cast<const null_type&>(null_type()),
+                static_cast<const null_type&>(null_type())) {}
+
+  inline __host__ __device__ 
+  tuple(typename access_traits<T0>::parameter_type t0,
+        typename access_traits<T1>::parameter_type t1,
+        typename access_traits<T2>::parameter_type t2,
+        typename access_traits<T3>::parameter_type t3,
+        typename access_traits<T4>::parameter_type t4,
+        typename access_traits<T5>::parameter_type t5,
+        typename access_traits<T6>::parameter_type t6,
+        typename access_traits<T7>::parameter_type t7,
+        typename access_traits<T8>::parameter_type t8)
+    : inherited(t0, t1, t2, t3, t4, t5, t6, t7, t8,
+                static_cast<const null_type&>(null_type())) {}
+
+  inline __host__ __device__ 
+  tuple(typename access_traits<T0>::parameter_type t0,
+        typename access_traits<T1>::parameter_type t1,
+        typename access_traits<T2>::parameter_type t2,
+        typename access_traits<T3>::parameter_type t3,
+        typename access_traits<T4>::parameter_type t4,
+        typename access_traits<T5>::parameter_type t5,
+        typename access_traits<T6>::parameter_type t6,
+        typename access_traits<T7>::parameter_type t7,
+        typename access_traits<T8>::parameter_type t8,
+        typename access_traits<T9>::parameter_type t9)
+    : inherited(t0, t1, t2, t3, t4, t5, t6, t7, t8, t9) {}
+
+
+  template<class U1, class U2>
+  inline __host__ __device__ 
+  tuple(const detail::cons<U1, U2>& p) : inherited(p) {}
+
+  template <class U1, class U2>
+  inline __host__ __device__ 
+  tuple& operator=(const detail::cons<U1, U2>& k)
+  {
+    inherited::operator=(k);
+    return *this;
+  }
+
+  /*! \endcond
+   */
+
+  /*! This assignment operator allows assigning the first two elements of this \p tuple from a \p pair.
+   *  \param k A \p pair to assign from.
+   */
+  template <class U1, class U2>
+  __host__ __device__ inline
+  tuple& operator=(const thrust::pair<U1, U2>& k) {
+    //BOOST_STATIC_ASSERT(length<tuple>::value == 2);// check_length = 2
+    this->head = k.first;
+    this->tail.head = k.second;
+    return *this;
+  }
+
+  /*! \p swap swaps the elements of two <tt>tuple</tt>s.
+   *
+   *  \param t The other <tt>tuple</tt> with which to swap.
+   */
+  inline __host__ __device__
+  void swap(tuple &t)
+  {
+    inherited::swap(t);
+  }
+};
+
+/*! \cond
+ */
+
+template <>
+class tuple<null_type, null_type, null_type, null_type, null_type, null_type, null_type, null_type, null_type, null_type>  :
+  public null_type
+{
+public:
+  typedef null_type inherited;
+};
+
+/*! \endcond
+ */
+
+
+/*! This version of \p make_tuple creates a new \c tuple object from a
+ *  single object.
+ *
+ *  \param t0 The object to copy from.
+ *  \return A \p tuple object with a single member which is a copy of \p t0.
+ */
+template<class T0>
+__host__ __device__ inline
+  typename detail::make_tuple_mapper<T0>::type
+    make_tuple(const T0& t0);
+
+/*! This version of \p make_tuple creates a new \c tuple object from two
+ *  objects.
+ *
+ *  \param t0 The first object to copy from.
+ *  \param t1 The second object to copy from.
+ *  \return A \p tuple object with two members which are copies of \p t0
+ *          and \p t1.
+ *
+ *  \note \p make_tuple has ten variants, the rest of which are omitted here
+ *        for brevity.
+ */
+template<class T0, class T1>
+__host__ __device__ inline
+  typename detail::make_tuple_mapper<T0, T1>::type
+    make_tuple(const T0& t0, const T1& t1);
+
+/*! This version of \p tie creates a new \c tuple whose single element is
+ *  a reference which refers to this function's argument.
+ *
+ *  \param t0 The object to reference.
+ *  \return A \p tuple object with one member which is a reference to \p t0.
+ */
+template<typename T0>
+__host__ __device__ inline
+tuple<T0&> tie(T0& t0);
+
+/*! This version of \p tie creates a new \c tuple of references object which
+ *  refers to this function's arguments.
+ *
+ *  \param t0 The first object to reference.
+ *  \param t1 The second object to reference.
+ *  \return A \p tuple object with two members which are references to \p t0
+ *          and \p t1.
+ *
+ *  \note \p tie has ten variants, the rest of which are omitted here for
+ *           brevity.
+ */
+template<typename T0, typename T1>
+__host__ __device__ inline
+tuple<T0&,T1&> tie(T0& t0, T1& t1);
+
+/*! \p swap swaps the contents of two <tt>tuple</tt>s.
+ *
+ *  \param x The first \p tuple to swap.
+ *  \param y The second \p tuple to swap.
+ */
+template<
+  typename T0, typename T1, typename T2, typename T3, typename T4, typename T5, typename T6, typename T7, typename T8, typename T9,
+  typename U0, typename U1, typename U2, typename U3, typename U4, typename U5, typename U6, typename U7, typename U8, typename U9
+>
+inline __host__ __device__
+void swap(tuple<T0,T1,T2,T3,T4,T5,T6,T7,T8,T9> &x,
+          tuple<U0,U1,U2,U3,U4,U5,U6,U7,U8,U9> &y);
+
+
+
+/*! \cond
+ */
+
+template<class T0, class T1, class T2>
+__host__ __device__ inline
+  typename detail::make_tuple_mapper<T0, T1, T2>::type
+    make_tuple(const T0& t0, const T1& t1, const T2& t2);
+
+template<class T0, class T1, class T2, class T3>
+__host__ __device__ inline
+  typename detail::make_tuple_mapper<T0, T1, T2, T3>::type
+    make_tuple(const T0& t0, const T1& t1, const T2& t2, const T3& t3);
+
+template<class T0, class T1, class T2, class T3, class T4>
+__host__ __device__ inline
+  typename detail::make_tuple_mapper<T0, T1, T2, T3, T4>::type
+    make_tuple(const T0& t0, const T1& t1, const T2& t2, const T3& t3, const T4& t4);
+
+template<class T0, class T1, class T2, class T3, class T4, class T5>
+__host__ __device__ inline
+  typename detail::make_tuple_mapper<T0, T1, T2, T3, T4, T5>::type
+    make_tuple(const T0& t0, const T1& t1, const T2& t2, const T3& t3, const T4& t4, const T5& t5);
+
+template<class T0, class T1, class T2, class T3, class T4, class T5, class T6>
+__host__ __device__ inline
+  typename detail::make_tuple_mapper<T0, T1, T2, T3, T4, T5, T6>::type
+    make_tuple(const T0& t0, const T1& t1, const T2& t2, const T3& t3, const T4& t4, const T5& t5, const T6& t6);
+
+template<class T0, class T1, class T2, class T3, class T4, class T5, class T6, class T7>
+__host__ __device__ inline
+  typename detail::make_tuple_mapper<T0, T1, T2, T3, T4, T5, T6, T7>::type
+    make_tuple(const T0& t0, const T1& t1, const T2& t2, const T3& t3, const T4& t4, const T5& t5, const T6& t6, const T7& t7);
+
+template<class T0, class T1, class T2, class T3, class T4, class T5, class T6, class T7, class T8>
+__host__ __device__ inline
+  typename detail::make_tuple_mapper<T0, T1, T2, T3, T4, T5, T6, T7, T8>::type
+    make_tuple(const T0& t0, const T1& t1, const T2& t2, const T3& t3, const T4& t4, const T5& t5, const T6& t6, const T7& t7, const T8& t8);
+
+template<class T0, class T1, class T2, class T3, class T4, class T5, class T6, class T7, class T8, class T9>
+__host__ __device__ inline
+  typename detail::make_tuple_mapper<T0, T1, T2, T3, T4, T5, T6, T7, T8, T9>::type
+    make_tuple(const T0& t0, const T1& t1, const T2& t2, const T3& t3, const T4& t4, const T5& t5, const T6& t6, const T7& t7, const T8& t8, const T9& t9);
+
+template<typename T0, typename T1, typename T2>
+__host__ __device__ inline
+tuple<T0&,T1&,T2&> tie(T0 &t0, T1 &t1, T2 &t2);
+
+template<typename T0, typename T1, typename T2, typename T3>
+__host__ __device__ inline
+tuple<T0&,T1&,T2&,T3&> tie(T0 &t0, T1 &t1, T2 &t2, T3 &t3);
+
+template<typename T0, typename T1, typename T2, typename T3, typename T4>
+__host__ __device__ inline
+tuple<T0&,T1&,T2&,T3&,T4&> tie(T0 &t0, T1 &t1, T2 &t2, T3 &t3, T4 &t4);
+
+template<typename T0, typename T1, typename T2, typename T3, typename T4, typename T5>
+__host__ __device__ inline
+tuple<T0&,T1&,T2&,T3&,T4&,T5&> tie(T0 &t0, T1 &t1, T2 &t2, T3 &t3, T4 &t4, T5 &t5);
+
+template<typename T0, typename T1, typename T2, typename T3, typename T4, typename T5, typename T6>
+__host__ __device__ inline
+tuple<T0&,T1&,T2&,T3&,T4&,T5&,T6&> tie(T0 &t0, T1 &t1, T2 &t2, T3 &t3, T4 &t4, T5 &t5, T6 &t6);
+
+template<typename T0, typename T1, typename T2, typename T3, typename T4, typename T5, typename T6, typename T7>
+__host__ __device__ inline
+tuple<T0&,T1&,T2&,T3&,T4&,T5&,T6&,T7&> tie(T0 &t0, T1 &t1, T2 &t2, T3 &t3, T4 &t4, T5 &t5, T6 &t6, T7 &t7);
+
+template<typename T0, typename T1, typename T2, typename T3, typename T4, typename T5, typename T6, typename T7, typename T8>
+__host__ __device__ inline
+tuple<T0&,T1&,T2&,T3&,T4&,T5&,T6&,T7&,T8&> tie(T0 &t0, T1 &t1, T2 &t2, T3 &t3, T4 &t4, T5 &t5, T6 &t6, T7 &t7, T8 &t8);
+
+template<typename T0, typename T1, typename T2, typename T3, typename T4, typename T5, typename T6, typename T7, typename T8, typename T9>
+__host__ __device__ inline
+tuple<T0&,T1&,T2&,T3&,T4&,T5&,T6&,T7&,T8&,T9&> tie(T0 &t0, T1 &t1, T2 &t2, T3 &t3, T4 &t4, T5 &t5, T6 &t6, T7 &t7, T8 &t8, T9 &t9);
+
+
+__host__ __device__ inline
+bool operator==(const null_type&, const null_type&);
+
+__host__ __device__ inline
+bool operator>=(const null_type&, const null_type&);
+
+__host__ __device__ inline
+bool operator<=(const null_type&, const null_type&);
+
+__host__ __device__ inline
+bool operator!=(const null_type&, const null_type&);
+
+__host__ __device__ inline
+bool operator<(const null_type&, const null_type&);
+
+__host__ __device__ inline
+bool operator>(const null_type&, const null_type&);
+
+/*! \endcond
+ */
+
+/*! \} // tuple
+ */
+
+/*! \} // utility
+ */
+
+} // end thrust
+
diff --git a/compat/thrust/uninitialized_copy.h b/compat/thrust/uninitialized_copy.h
new file mode 100644
index 0000000000..77b673c425
--- /dev/null
+++ b/compat/thrust/uninitialized_copy.h
@@ -0,0 +1,301 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file uninitialized_copy.h
+ *  \brief Copy construction into a range of uninitialized elements from a source range
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/execution_policy.h>
+
+namespace thrust
+{
+
+
+/*! \addtogroup copying
+ *  \{
+ */
+
+
+/*! In \c thrust, the function \c thrust::device_new allocates memory for
+ *  an object and then creates an object at that location by calling a constructor.
+ *  Occasionally, however, it is useful to separate those two operations.
+ *  If each iterator in the range <tt>[result, result + (last - first))</tt> points
+ *  to uninitialized memory, then \p uninitialized_copy creates a copy of
+ *  <tt>[first, last)</tt> in that range. That is, for each iterator \c i in
+ *  the input, \p uninitialized_copy creates a copy of \c *i in the location pointed
+ *  to by the corresponding iterator in the output range by \p ForwardIterator's
+ *  \c value_type's copy constructor with *i as its argument.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The first element of the input range to copy from.
+ *  \param last The last element of the input range to copy from.
+ *  \param result The first element of the output range to copy to.
+ *  \return An iterator pointing to the last element of the output range.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>.
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>,
+ *          \p ForwardIterator is mutable, and \p ForwardIterator's \c value_type has a constructor that takes
+ *          a single argument whose type is \p InputIterator's \c value_type.
+ *
+ *  \pre \p first may equal \p result, but the range <tt>[first, last)</tt> and the range <tt>[result, result + (last - first))</tt> shall not overlap otherwise.
+ *
+ *  The following code snippet demonstrates how to use \p uninitialized_copy to initialize
+ *  a range of uninitialized memory using the \p thrust::device execution policy for
+ *  parallelization:
+ *
+ *  \code
+ *  #include <thrust/uninitialized_copy.h>
+ *  #include <thrust/device_malloc.h>
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/execution_policy.h>
+ *  
+ *  struct Int
+ *  {
+ *    __host__ __device__
+ *    Int(int x) : val(x) {}
+ *    int val;
+ *  };  
+ *  ...
+ *  const int N = 137;
+ *
+ *  Int val(46);
+ *  thrust::device_vector<Int> input(N, val);
+ *  thrust::device_ptr<Int> array = thrust::device_malloc<Int>(N);
+ *  thrust::uninitialized_copy(thrust::device, input.begin(), input.end(), array);
+ *
+ *  // Int x = array[i];
+ *  // x.val == 46 for all 0 <= i < N
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/uninitialized_copy.html
+ *  \see \c copy
+ *  \see \c uninitialized_fill
+ *  \see \c device_new
+ *  \see \c device_malloc
+ */
+template<typename DerivedPolicy, typename InputIterator, typename ForwardIterator>
+  ForwardIterator uninitialized_copy(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                     InputIterator first,
+                                     InputIterator last,
+                                     ForwardIterator result);
+
+
+/*! In \c thrust, the function \c thrust::device_new allocates memory for
+ *  an object and then creates an object at that location by calling a constructor.
+ *  Occasionally, however, it is useful to separate those two operations.
+ *  If each iterator in the range <tt>[result, result + (last - first))</tt> points
+ *  to uninitialized memory, then \p uninitialized_copy creates a copy of
+ *  <tt>[first, last)</tt> in that range. That is, for each iterator \c i in
+ *  the input, \p uninitialized_copy creates a copy of \c *i in the location pointed
+ *  to by the corresponding iterator in the output range by \p ForwardIterator's
+ *  \c value_type's copy constructor with *i as its argument.
+ *
+ *  \param first The first element of the input range to copy from.
+ *  \param last The last element of the input range to copy from.
+ *  \param result The first element of the output range to copy to.
+ *  \return An iterator pointing to the last element of the output range.
+ *
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>.
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>,
+ *          \p ForwardIterator is mutable, and \p ForwardIterator's \c value_type has a constructor that takes
+ *          a single argument whose type is \p InputIterator's \c value_type.
+ *
+ *  \pre \p first may equal \p result, but the range <tt>[first, last)</tt> and the range <tt>[result, result + (last - first))</tt> shall not overlap otherwise.
+ *
+ *  The following code snippet demonstrates how to use \p uninitialized_copy to initialize
+ *  a range of uninitialized memory.
+ *
+ *  \code
+ *  #include <thrust/uninitialized_copy.h>
+ *  #include <thrust/device_malloc.h>
+ *  #include <thrust/device_vector.h>
+ *  
+ *  struct Int
+ *  {
+ *    __host__ __device__
+ *    Int(int x) : val(x) {}
+ *    int val;
+ *  };  
+ *  ...
+ *  const int N = 137;
+ *
+ *  Int val(46);
+ *  thrust::device_vector<Int> input(N, val);
+ *  thrust::device_ptr<Int> array = thrust::device_malloc<Int>(N);
+ *  thrust::uninitialized_copy(input.begin(), input.end(), array);
+ *
+ *  // Int x = array[i];
+ *  // x.val == 46 for all 0 <= i < N
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/uninitialized_copy.html
+ *  \see \c copy
+ *  \see \c uninitialized_fill
+ *  \see \c device_new
+ *  \see \c device_malloc
+ */
+template<typename InputIterator, typename ForwardIterator>
+  ForwardIterator uninitialized_copy(InputIterator first,
+                                     InputIterator last,
+                                     ForwardIterator result);
+
+
+/*! In \c thrust, the function \c thrust::device_new allocates memory for
+ *  an object and then creates an object at that location by calling a constructor.
+ *  Occasionally, however, it is useful to separate those two operations.
+ *  If each iterator in the range <tt>[result, result + n)</tt> points
+ *  to uninitialized memory, then \p uninitialized_copy_n creates a copy of
+ *  <tt>[first, first + n)</tt> in that range. That is, for each iterator \c i in
+ *  the input, \p uninitialized_copy_n creates a copy of \c *i in the location pointed
+ *  to by the corresponding iterator in the output range by \p InputIterator's
+ *  \c value_type's copy constructor with *i as its argument.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The first element of the input range to copy from.
+ *  \param n The number of elements to copy.
+ *  \param result The first element of the output range to copy to.
+ *  \return An iterator pointing to the last element of the output range.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>.
+ *  \tparam Size is an integral type.
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>,
+ *          \p ForwardIterator is mutable, and \p ForwardIterator's \c value_type has a constructor that takes
+ *          a single argument whose type is \p InputIterator's \c value_type.
+ *
+ *  \pre \p first may equal \p result, but the range <tt>[first, first + n)</tt> and the range <tt>[result, result + n)</tt> shall not overlap otherwise.
+ *
+ *  The following code snippet demonstrates how to use \p uninitialized_copy to initialize
+ *  a range of uninitialized memory using the \p thrust::device execution policy for
+ *  parallelization:
+ *
+ *  \code
+ *  #include <thrust/uninitialized_copy.h>
+ *  #include <thrust/device_malloc.h>
+ *  #include <thrust/device_vector.h>
+ *  #include <thrust/execution_policy.h>
+ *  
+ *  struct Int
+ *  {
+ *    __host__ __device__
+ *    Int(int x) : val(x) {}
+ *    int val;
+ *  };  
+ *  ...
+ *  const int N = 137;
+ *
+ *  Int val(46);
+ *  thrust::device_vector<Int> input(N, val);
+ *  thrust::device_ptr<Int> array = thrust::device_malloc<Int>(N);
+ *  thrust::uninitialized_copy_n(thrust::device, input.begin(), N, array);
+ *
+ *  // Int x = array[i];
+ *  // x.val == 46 for all 0 <= i < N
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/uninitialized_copy.html
+ *  \see \c uninitialized_copy
+ *  \see \c copy
+ *  \see \c uninitialized_fill
+ *  \see \c device_new
+ *  \see \c device_malloc
+ */
+template<typename DerivedPolicy, typename InputIterator, typename Size, typename ForwardIterator>
+  ForwardIterator uninitialized_copy_n(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                       InputIterator first,
+                                       Size n,
+                                       ForwardIterator result);
+
+
+/*! In \c thrust, the function \c thrust::device_new allocates memory for
+ *  an object and then creates an object at that location by calling a constructor.
+ *  Occasionally, however, it is useful to separate those two operations.
+ *  If each iterator in the range <tt>[result, result + n)</tt> points
+ *  to uninitialized memory, then \p uninitialized_copy_n creates a copy of
+ *  <tt>[first, first + n)</tt> in that range. That is, for each iterator \c i in
+ *  the input, \p uninitialized_copy_n creates a copy of \c *i in the location pointed
+ *  to by the corresponding iterator in the output range by \p InputIterator's
+ *  \c value_type's copy constructor with *i as its argument.
+ *
+ *  \param first The first element of the input range to copy from.
+ *  \param n The number of elements to copy.
+ *  \param result The first element of the output range to copy to.
+ *  \return An iterator pointing to the last element of the output range.
+ *
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>.
+ *  \tparam Size is an integral type.
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>,
+ *          \p ForwardIterator is mutable, and \p ForwardIterator's \c value_type has a constructor that takes
+ *          a single argument whose type is \p InputIterator's \c value_type.
+ *
+ *  \pre \p first may equal \p result, but the range <tt>[first, first + n)</tt> and the range <tt>[result, result + n)</tt> shall not overlap otherwise.
+ *
+ *  The following code snippet demonstrates how to use \p uninitialized_copy to initialize
+ *  a range of uninitialized memory.
+ *
+ *  \code
+ *  #include <thrust/uninitialized_copy.h>
+ *  #include <thrust/device_malloc.h>
+ *  #include <thrust/device_vector.h>
+ *  
+ *  struct Int
+ *  {
+ *    __host__ __device__
+ *    Int(int x) : val(x) {}
+ *    int val;
+ *  };  
+ *  ...
+ *  const int N = 137;
+ *
+ *  Int val(46);
+ *  thrust::device_vector<Int> input(N, val);
+ *  thrust::device_ptr<Int> array = thrust::device_malloc<Int>(N);
+ *  thrust::uninitialized_copy_n(input.begin(), N, array);
+ *
+ *  // Int x = array[i];
+ *  // x.val == 46 for all 0 <= i < N
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/uninitialized_copy.html
+ *  \see \c uninitialized_copy
+ *  \see \c copy
+ *  \see \c uninitialized_fill
+ *  \see \c device_new
+ *  \see \c device_malloc
+ */
+template<typename InputIterator, typename Size, typename ForwardIterator>
+  ForwardIterator uninitialized_copy_n(InputIterator first,
+                                       Size n,
+                                       ForwardIterator result);
+
+
+/*! \} // copying
+ */
+
+
+} // end thrust
+
+#include <thrust/detail/uninitialized_copy.inl>
+
diff --git a/compat/thrust/uninitialized_fill.h b/compat/thrust/uninitialized_fill.h
new file mode 100644
index 0000000000..c726241ddd
--- /dev/null
+++ b/compat/thrust/uninitialized_fill.h
@@ -0,0 +1,273 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file uninitialized_fill.h
+ *  \brief Copy construction into a range of uninitialized elements from a source value
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/execution_policy.h>
+
+namespace thrust
+{
+
+
+/*! \addtogroup filling
+ *  \ingroup transformations
+ *  \{
+ */
+
+
+/*! In \c thrust, the function \c thrust::device_new allocates memory for
+ *  an object and then creates an object at that location by calling a
+ *  constructor. Occasionally, however, it is useful to separate those two
+ *  operations. If each iterator in the range <tt>[first, last)</tt> points
+ *  to unitialized memory, then \p unitialized_fill creates copies of \c x
+ *  in that range. That is, for each iterator \c i in the range <tt>[first, last)</tt>,
+ *  \p uninitialized_fill creates a copy of \c x in the location pointed to \c i by
+ *  calling \p ForwardIterator's \c value_type's copy constructor.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *  
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The first element of the range of interest.
+ *  \param last The last element of the range of interest.
+ *  \param x The value to use as the exemplar of the copy constructor.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator">Forward Iterator</a>,
+ *          \p ForwardIterator is mutable, and \p ForwardIterator's \c value_type has a constructor that
+ *          takes a single argument of type \p T.
+ *
+ *  The following code snippet demonstrates how to use \p uninitialized_fill to initialize a range of
+ *  uninitialized memory using the \p thrust::device execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/uninitialized_fill.h>
+ *  #include <thrust/device_malloc.h>
+ *  #include <thrust/execution_policy.h>
+ *  
+ *  struct Int
+ *  {
+ *    __host__ __device__
+ *    Int(int x) : val(x) {}
+ *    int val;
+ *  };  
+ *  ...
+ *  const int N = 137;
+ *
+ *  Int val(46);
+ *  thrust::device_ptr<Int> array = thrust::device_malloc<Int>(N);
+ *  thrust::uninitialized_fill(thrust::device, array, array + N, val);
+ *
+ *  // Int x = array[i];
+ *  // x.val == 46 for all 0 <= i < N
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/uninitialized_fill.html
+ *  \see \c uninitialized_fill_n
+ *  \see \c fill
+ *  \see \c uninitialized_copy
+ *  \see \c device_new
+ *  \see \c device_malloc
+ */
+template<typename DerivedPolicy, typename ForwardIterator, typename T>
+  void uninitialized_fill(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                          ForwardIterator first,
+                          ForwardIterator last,
+                          const T &x);
+
+
+/*! In \c thrust, the function \c thrust::device_new allocates memory for
+ *  an object and then creates an object at that location by calling a
+ *  constructor. Occasionally, however, it is useful to separate those two
+ *  operations. If each iterator in the range <tt>[first, last)</tt> points
+ *  to unitialized memory, then \p unitialized_fill creates copies of \c x
+ *  in that range. That is, for each iterator \c i in the range <tt>[first, last)</tt>,
+ *  \p uninitialized_fill creates a copy of \c x in the location pointed to \c i by
+ *  calling \p ForwardIterator's \c value_type's copy constructor.
+ *  
+ *  \param first The first element of the range of interest.
+ *  \param last The last element of the range of interest.
+ *  \param x The value to use as the exemplar of the copy constructor.
+ *
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator">Forward Iterator</a>,
+ *          \p ForwardIterator is mutable, and \p ForwardIterator's \c value_type has a constructor that
+ *          takes a single argument of type \p T.
+ *
+ *  The following code snippet demonstrates how to use \p uninitialized_fill to initialize a range of
+ *  uninitialized memory.
+ *
+ *  \code
+ *  #include <thrust/uninitialized_fill.h>
+ *  #include <thrust/device_malloc.h>
+ *  
+ *  struct Int
+ *  {
+ *    __host__ __device__
+ *    Int(int x) : val(x) {}
+ *    int val;
+ *  };  
+ *  ...
+ *  const int N = 137;
+ *
+ *  Int val(46);
+ *  thrust::device_ptr<Int> array = thrust::device_malloc<Int>(N);
+ *  thrust::uninitialized_fill(array, array + N, val);
+ *
+ *  // Int x = array[i];
+ *  // x.val == 46 for all 0 <= i < N
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/uninitialized_fill.html
+ *  \see \c uninitialized_fill_n
+ *  \see \c fill
+ *  \see \c uninitialized_copy
+ *  \see \c device_new
+ *  \see \c device_malloc
+ */
+template<typename ForwardIterator, typename T>
+  void uninitialized_fill(ForwardIterator first,
+                          ForwardIterator last,
+                          const T &x);
+
+
+/*! In \c thrust, the function \c thrust::device_new allocates memory for
+ *  an object and then creates an object at that location by calling a
+ *  constructor. Occasionally, however, it is useful to separate those two
+ *  operations. If each iterator in the range <tt>[first, first+n)</tt> points
+ *  to unitialized memory, then \p unitialized_fill creates copies of \c x
+ *  in that range. That is, for each iterator \c i in the range <tt>[first, first+n)</tt>,
+ *  \p uninitialized_fill creates a copy of \c x in the location pointed to \c i by
+ *  calling \p ForwardIterator's \c value_type's copy constructor.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *  
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The first element of the range of interest.
+ *  \param n The size of the range of interest.
+ *  \param x The value to use as the exemplar of the copy constructor.
+ *  \return <tt>first+n</tt>
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator">Forward Iterator</a>,
+ *          \p ForwardIterator is mutable, and \p ForwardIterator's \c value_type has a constructor that
+ *          takes a single argument of type \p T.
+ *
+ *  The following code snippet demonstrates how to use \p uninitialized_fill to initialize a range of
+ *  uninitialized memory using the \p thrust::device execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/uninitialized_fill.h>
+ *  #include <thrust/device_malloc.h>
+ *  #include <thrust/execution_policy.h>
+ *  
+ *  struct Int
+ *  {
+ *    __host__ __device__
+ *    Int(int x) : val(x) {}
+ *    int val;
+ *  };  
+ *  ...
+ *  const int N = 137;
+ *
+ *  Int val(46);
+ *  thrust::device_ptr<Int> array = thrust::device_malloc<Int>(N);
+ *  thrust::uninitialized_fill_n(thrust::device, array, N, val);
+ *
+ *  // Int x = array[i];
+ *  // x.val == 46 for all 0 <= i < N
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/uninitialized_fill.html
+ *  \see \c uninitialized_fill
+ *  \see \c fill
+ *  \see \c uninitialized_copy_n
+ *  \see \c device_new
+ *  \see \c device_malloc
+ */
+template<typename DerivedPolicy, typename ForwardIterator, typename Size, typename T>
+  ForwardIterator uninitialized_fill_n(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                                       ForwardIterator first,
+                                       Size n,
+                                       const T &x);
+
+
+/*! In \c thrust, the function \c thrust::device_new allocates memory for
+ *  an object and then creates an object at that location by calling a
+ *  constructor. Occasionally, however, it is useful to separate those two
+ *  operations. If each iterator in the range <tt>[first, first+n)</tt> points
+ *  to unitialized memory, then \p unitialized_fill creates copies of \c x
+ *  in that range. That is, for each iterator \c i in the range <tt>[first, first+n)</tt>,
+ *  \p uninitialized_fill creates a copy of \c x in the location pointed to \c i by
+ *  calling \p ForwardIterator's \c value_type's copy constructor.
+ *  
+ *  \param first The first element of the range of interest.
+ *  \param n The size of the range of interest.
+ *  \param x The value to use as the exemplar of the copy constructor.
+ *  \return <tt>first+n</tt>
+ *
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator">Forward Iterator</a>,
+ *          \p ForwardIterator is mutable, and \p ForwardIterator's \c value_type has a constructor that
+ *          takes a single argument of type \p T.
+ *
+ *  The following code snippet demonstrates how to use \p uninitialized_fill to initialize a range of
+ *  uninitialized memory.
+ *
+ *  \code
+ *  #include <thrust/uninitialized_fill.h>
+ *  #include <thrust/device_malloc.h>
+ *  
+ *  struct Int
+ *  {
+ *    __host__ __device__
+ *    Int(int x) : val(x) {}
+ *    int val;
+ *  };  
+ *  ...
+ *  const int N = 137;
+ *
+ *  Int val(46);
+ *  thrust::device_ptr<Int> array = thrust::device_malloc<Int>(N);
+ *  thrust::uninitialized_fill_n(array, N, val);
+ *
+ *  // Int x = array[i];
+ *  // x.val == 46 for all 0 <= i < N
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/uninitialized_fill.html
+ *  \see \c uninitialized_fill
+ *  \see \c fill
+ *  \see \c uninitialized_copy_n
+ *  \see \c device_new
+ *  \see \c device_malloc
+ */
+template<typename ForwardIterator, typename Size, typename T>
+  ForwardIterator uninitialized_fill_n(ForwardIterator first,
+                                       Size n,
+                                       const T &x);
+
+/*! \} // end filling
+ *  \} // transformations
+ */
+
+} // end thrust
+
+#include <thrust/detail/uninitialized_fill.inl>
+
diff --git a/compat/thrust/unique.h b/compat/thrust/unique.h
new file mode 100644
index 0000000000..98150f36c9
--- /dev/null
+++ b/compat/thrust/unique.h
@@ -0,0 +1,960 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+
+/*! \file unique.h
+ *  \brief Move unique elements to the front of a range
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+#include <thrust/detail/execution_policy.h>
+#include <thrust/pair.h>
+
+namespace thrust
+{
+
+
+/*! \addtogroup stream_compaction
+ *  \{
+ */
+
+
+/*! For each group of consecutive elements in the range <tt>[first, last)</tt>
+ *  with the same value, \p unique removes all but the first element of 
+ *  the group. The return value is an iterator \c new_last such that 
+ *  no two consecutive elements in the range <tt>[first, new_last)</tt> are
+ *  equal. The iterators in the range <tt>[new_last, last)</tt> are all still
+ *  dereferenceable, but the elements that they point to are unspecified.
+ *  \p unique is stable, meaning that the relative order of elements that are
+ *  not removed is unchanged.
+ *
+ *  This version of \p unique uses \c operator== to test for equality.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the input range.
+ *  \param last  The end of the input range.
+ *  \return The end of the unique range <tt>[first, new_last)</tt>.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>,
+ *          and \p ForwardIterator is mutable,
+ *          and \p ForwardIterator's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/EqualityComparable.html">Equality Comparable</a>.
+ *
+ *  The following code snippet demonstrates how to use \p unique to
+ *  compact a sequence of numbers to remove consecutive duplicates using the \p thrust::host execution policy
+ *  for parallelization:
+ *
+ *  \code
+ *  #include <thrust/unique.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  const int N = 7;
+ *  int A[N] = {1, 3, 3, 3, 2, 2, 1};
+ *  int *new_end = thrust::unique(thrust::host, A, A + N);
+ *  // The first four values of A are now {1, 3, 2, 1}
+ *  // Values beyond new_end are unspecified.
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/unique.html
+ *  \see unique_copy
+ */
+template<typename DerivedPolicy,
+         typename ForwardIterator>
+ForwardIterator unique(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                       ForwardIterator first,
+                       ForwardIterator last);
+
+
+/*! For each group of consecutive elements in the range <tt>[first, last)</tt>
+ *  with the same value, \p unique removes all but the first element of 
+ *  the group. The return value is an iterator \c new_last such that 
+ *  no two consecutive elements in the range <tt>[first, new_last)</tt> are
+ *  equal. The iterators in the range <tt>[new_last, last)</tt> are all still
+ *  dereferenceable, but the elements that they point to are unspecified.
+ *  \p unique is stable, meaning that the relative order of elements that are
+ *  not removed is unchanged.
+ *
+ *  This version of \p unique uses \c operator== to test for equality.
+ *
+ *  \param first The beginning of the input range.
+ *  \param last  The end of the input range.
+ *  \return The end of the unique range <tt>[first, new_last)</tt>.
+ *
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>,
+ *          and \p ForwardIterator is mutable,
+ *          and \p ForwardIterator's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/EqualityComparable.html">Equality Comparable</a>.
+ *
+ *  The following code snippet demonstrates how to use \p unique to
+ *  compact a sequence of numbers to remove consecutive duplicates.
+ *
+ *  \code
+ *  #include <thrust/unique.h>
+ *  ...
+ *  const int N = 7;
+ *  int A[N] = {1, 3, 3, 3, 2, 2, 1};
+ *  int *new_end = thrust::unique(A, A + N);
+ *  // The first four values of A are now {1, 3, 2, 1}
+ *  // Values beyond new_end are unspecified.
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/unique.html
+ *  \see unique_copy
+ */
+template <typename ForwardIterator>
+ForwardIterator unique(ForwardIterator first,
+                       ForwardIterator last);
+
+
+/*! For each group of consecutive elements in the range <tt>[first, last)</tt>
+ *  with the same value, \p unique removes all but the first element of 
+ *  the group. The return value is an iterator \c new_last such that 
+ *  no two consecutive elements in the range <tt>[first, new_last)</tt> are
+ *  equal. The iterators in the range <tt>[new_last, last)</tt> are all still
+ *  dereferenceable, but the elements that they point to are unspecified.
+ *  \p unique is stable, meaning that the relative order of elements that are
+ *  not removed is unchanged.
+ *
+ *  This version of \p unique uses the function object \p binary_pred to test
+ *  for equality.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the input range.
+ *  \param last  The end of the input range.
+ *  \param binary_pred  The binary predicate used to determine equality.
+ *  \return The end of the unique range <tt>[first, new_last)</tt>
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>,
+ *          and \p ForwardIterator is mutable,
+ *          and \p ForwardIterator's \c value_type is convertible to \p BinaryPredicate's \c first_argument_type and to \p BinaryPredicate's \c second_argument_type.
+ *  \tparam BinaryPredicate is a model of <a href="http://www.sgi.com/tech/stl/BinaryPredicate.html">Binary Predicate</a>.
+ *
+ *  The following code snippet demonstrates how to use \p unique to
+ *  compact a sequence of numbers to remove consecutive duplicates using the \p thrust::host execution policy
+ *  for parallelization:
+ *
+ *  \code
+ *  #include <thrust/unique.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  const int N = 7;
+ *  int A[N] = {1, 3, 3, 3, 2, 2, 1};
+ *  int *new_end = thrust::unique(thrust::host, A, A + N, thrust::equal_to<int>());
+ *  // The first four values of A are now {1, 3, 2, 1}
+ *  // Values beyond new_end are unspecified.
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/unique.html
+ *  \see unique_copy
+ */
+template<typename DerivedPolicy,
+         typename ForwardIterator,
+         typename BinaryPredicate>
+ForwardIterator unique(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                       ForwardIterator first,
+                       ForwardIterator last,
+                       BinaryPredicate binary_pred);
+
+
+/*! For each group of consecutive elements in the range <tt>[first, last)</tt>
+ *  with the same value, \p unique removes all but the first element of 
+ *  the group. The return value is an iterator \c new_last such that 
+ *  no two consecutive elements in the range <tt>[first, new_last)</tt> are
+ *  equal. The iterators in the range <tt>[new_last, last)</tt> are all still
+ *  dereferenceable, but the elements that they point to are unspecified.
+ *  \p unique is stable, meaning that the relative order of elements that are
+ *  not removed is unchanged.
+ *
+ *  This version of \p unique uses the function object \p binary_pred to test
+ *  for equality.
+ *
+ *  \param first The beginning of the input range.
+ *  \param last  The end of the input range.
+ *  \param binary_pred  The binary predicate used to determine equality.
+ *  \return The end of the unique range <tt>[first, new_last)</tt>
+ *
+ *  \tparam ForwardIterator is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>,
+ *          and \p ForwardIterator is mutable,
+ *          and \p ForwardIterator's \c value_type is convertible to \p BinaryPredicate's \c first_argument_type and to \p BinaryPredicate's \c second_argument_type.
+ *  \tparam BinaryPredicate is a model of <a href="http://www.sgi.com/tech/stl/BinaryPredicate.html">Binary Predicate</a>.
+ *
+ *  The following code snippet demonstrates how to use \p unique to
+ *  compact a sequence of numbers to remove consecutive duplicates.
+ *
+ *  \code
+ *  #include <thrust/unique.h>
+ *  ...
+ *  const int N = 7;
+ *  int A[N] = {1, 3, 3, 3, 2, 2, 1};
+ *  int *new_end = thrust::unique(A, A + N, thrust::equal_to<int>());
+ *  // The first four values of A are now {1, 3, 2, 1}
+ *  // Values beyond new_end are unspecified.
+ *  \endcode
+ *
+ *  \see http://www.sgi.com/tech/stl/unique.html
+ *  \see unique_copy
+ */
+template <typename ForwardIterator,
+          typename BinaryPredicate>
+ForwardIterator unique(ForwardIterator first,
+                       ForwardIterator last,
+                       BinaryPredicate binary_pred);
+
+
+/*! \p unique_copy copies elements from the range <tt>[first, last)</tt>
+ * to a range beginning with \p result, except that in a consecutive group
+ * of duplicate elements only the first one is copied. The return value
+ * is the end of the range to which the elements are copied. 
+ *
+ * The reason there are two different versions of unique_copy is that there
+ * are two different definitions of what it means for a consecutive group of
+ * elements to be duplicates. In the first version, the test is simple
+ * equality: the elements in a range <tt>[f, l)</tt> are duplicates if,
+ * for every iterator \p i in the range, either <tt>i == f</tt> or else 
+ * <tt>*i == *(i-1)</tt>. In the second, the test is an arbitrary 
+ * \p BinaryPredicate \p binary_pred: the elements in <tt>[f, l)</tt> are
+ * duplicates if, for every iterator \p i in the range, either <tt>i == f</tt>
+ * or else <tt>binary_pred(*i, *(i-1))</tt> is \p true.
+ *
+ * This version of \p unique_copy uses \c operator== to test for equality.
+ *
+ * The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the input range.
+ *  \param last  The end of the input range.
+ *  \param result The beginning of the output range.
+ *  \return The end of the unique range <tt>[result, result_end)</tt>.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/EqualityComparable.html">Equality Comparable</a>.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a> and
+ *          and \p InputIterator's \c value_type is convertible to \c OutputIterator's \c value_type.
+ *
+ *  \pre The range <tt>[first,last)</tt> and the range <tt>[result, result + (last - first))</tt> shall not overlap.
+ *
+ *  The following code snippet demonstrates how to use \p unique_copy to
+ *  compact a sequence of numbers to remove consecutive duplicates using the \p thrust::host execution
+ *  policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/unique.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  const int N = 7;
+ *  int A[N] = {1, 3, 3, 3, 2, 2, 1};
+ *  int B[N];
+ *  int *result_end = thrust::unique_copy(thrust::host, A, A + N, B);
+ *  // The first four values of B are now {1, 3, 2, 1} and (result_end - B) is 4
+ *  // Values beyond result_end are unspecified
+ *  \endcode
+ *
+ *  \see unique
+ *  \see http://www.sgi.com/tech/stl/unique_copy.html
+ */
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator>
+OutputIterator unique_copy(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                           InputIterator first,
+                           InputIterator last,
+                           OutputIterator result);
+
+
+/*! \p unique_copy copies elements from the range <tt>[first, last)</tt>
+ * to a range beginning with \p result, except that in a consecutive group
+ * of duplicate elements only the first one is copied. The return value
+ * is the end of the range to which the elements are copied. 
+ *
+ * The reason there are two different versions of unique_copy is that there
+ * are two different definitions of what it means for a consecutive group of
+ * elements to be duplicates. In the first version, the test is simple
+ * equality: the elements in a range <tt>[f, l)</tt> are duplicates if,
+ * for every iterator \p i in the range, either <tt>i == f</tt> or else 
+ * <tt>*i == *(i-1)</tt>. In the second, the test is an arbitrary 
+ * \p BinaryPredicate \p binary_pred: the elements in <tt>[f, l)</tt> are
+ * duplicates if, for every iterator \p i in the range, either <tt>i == f</tt>
+ * or else <tt>binary_pred(*i, *(i-1))</tt> is \p true.
+ *
+ * This version of \p unique_copy uses \c operator== to test for equality.
+ *
+ *  \param first The beginning of the input range.
+ *  \param last  The end of the input range.
+ *  \param result The beginning of the output range.
+ *  \return The end of the unique range <tt>[result, result_end)</tt>.
+ *
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/EqualityComparable.html">Equality Comparable</a>.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a> and
+ *          and \p InputIterator's \c value_type is convertible to \c OutputIterator's \c value_type.
+ *
+ *  \pre The range <tt>[first,last)</tt> and the range <tt>[result, result + (last - first))</tt> shall not overlap.
+ *
+ *  The following code snippet demonstrates how to use \p unique_copy to
+ *  compact a sequence of numbers to remove consecutive duplicates.
+ *
+ *  \code
+ *  #include <thrust/unique.h>
+ *  ...
+ *  const int N = 7;
+ *  int A[N] = {1, 3, 3, 3, 2, 2, 1};
+ *  int B[N];
+ *  int *result_end = thrust::unique_copy(A, A + N, B);
+ *  // The first four values of B are now {1, 3, 2, 1} and (result_end - B) is 4
+ *  // Values beyond result_end are unspecified
+ *  \endcode
+ *
+ *  \see unique
+ *  \see http://www.sgi.com/tech/stl/unique_copy.html
+ */
+template <typename InputIterator,
+          typename OutputIterator>
+OutputIterator unique_copy(InputIterator first,
+                           InputIterator last,
+                           OutputIterator result);
+
+
+/*! \p unique_copy copies elements from the range <tt>[first, last)</tt>
+ * to a range beginning with \p result, except that in a consecutive group
+ * of duplicate elements only the first one is copied. The return value
+ * is the end of the range to which the elements are copied. 
+ *
+ * This version of \p unique_copy uses the function object \c binary_pred 
+ * to test for equality.
+ *
+ * The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param first The beginning of the input range.
+ *  \param last  The end of the input range.
+ *  \param result The beginning of the output range.
+ *  \param binary_pred  The binary predicate used to determine equality.
+ *  \return The end of the unique range <tt>[result, result_end)</tt>.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/EqualityComparable.html">Equality Comparable</a>.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a> and
+ *          and \p InputIterator's \c value_type is convertible to \c OutputIterator's \c value_type.
+ *  \tparam BinaryPredicate is a model of <a href="http://www.sgi.com/tech/stl/BinaryPredicate.html">Binary Predicate</a>.
+ *
+ *  \pre The range <tt>[first,last)</tt> and the range <tt>[result, result + (last - first))</tt> shall not overlap.
+ *
+ *  The following code snippet demonstrates how to use \p unique_copy to
+ *  compact a sequence of numbers to remove consecutive duplicates using the \p thrust::host execution
+ *  policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/unique.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  const int N = 7;
+ *  int A[N] = {1, 3, 3, 3, 2, 2, 1};
+ *  int B[N];
+ *  int *result_end = thrust::unique_copy(thrust::host, A, A + N, B, thrust::equal_to<int>());
+ *  // The first four values of B are now {1, 3, 2, 1} and (result_end - B) is 4
+ *  // Values beyond result_end are unspecified.
+ *  \endcode
+ *
+ *  \see unique
+ *  \see http://www.sgi.com/tech/stl/unique_copy.html
+ */
+template<typename DerivedPolicy,
+         typename InputIterator,
+         typename OutputIterator,
+         typename BinaryPredicate>
+OutputIterator unique_copy(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                           InputIterator first,
+                           InputIterator last,
+                           OutputIterator result,
+                           BinaryPredicate binary_pred);
+                       
+
+/*! \p unique_copy copies elements from the range <tt>[first, last)</tt>
+ * to a range beginning with \p result, except that in a consecutive group
+ * of duplicate elements only the first one is copied. The return value
+ * is the end of the range to which the elements are copied. 
+ *
+ * This version of \p unique_copy uses the function object \c binary_pred 
+ * to test for equality.
+ *
+ *  \param first The beginning of the input range.
+ *  \param last  The end of the input range.
+ *  \param result The beginning of the output range.
+ *  \param binary_pred  The binary predicate used to determine equality.
+ *  \return The end of the unique range <tt>[result, result_end)</tt>.
+ *
+ *  \tparam InputIterator is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *          and \p InputIterator's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/EqualityComparable.html">Equality Comparable</a>.
+ *  \tparam OutputIterator is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a> and
+ *          and \p InputIterator's \c value_type is convertible to \c OutputIterator's \c value_type.
+ *  \tparam BinaryPredicate is a model of <a href="http://www.sgi.com/tech/stl/BinaryPredicate.html">Binary Predicate</a>.
+ *
+ *  \pre The range <tt>[first,last)</tt> and the range <tt>[result, result + (last - first))</tt> shall not overlap.
+ *
+ *  The following code snippet demonstrates how to use \p unique_copy to
+ *  compact a sequence of numbers to remove consecutive duplicates.
+ *
+ *  \code
+ *  #include <thrust/unique.h>
+ *  ...
+ *  const int N = 7;
+ *  int A[N] = {1, 3, 3, 3, 2, 2, 1};
+ *  int B[N];
+ *  int *result_end = thrust::unique_copy(A, A + N, B, thrust::equal_to<int>());
+ *  // The first four values of B are now {1, 3, 2, 1} and (result_end - B) is 4
+ *  // Values beyond result_end are unspecified.
+ *  \endcode
+ *
+ *  \see unique
+ *  \see http://www.sgi.com/tech/stl/unique_copy.html
+ */
+template <typename InputIterator,
+          typename OutputIterator,
+          typename BinaryPredicate>
+OutputIterator unique_copy(InputIterator first,
+                           InputIterator last,
+                           OutputIterator result,
+                           BinaryPredicate binary_pred);
+
+
+/*! \p unique_by_key is a generalization of \p unique to key-value pairs.
+ *  For each group of consecutive keys in the range <tt>[keys_first, keys_last)</tt>
+ *  that are equal, \p unique_by_key removes all but the first element of 
+ *  the group.  Similarly, the corresponding values in the range
+ *  <tt>[values_first, values_first + (keys_last - keys_first))</tt> 
+ *  are also removed.
+ *
+ *  The return value is a \p pair of iterators <tt>(new_keys_last,new_values_last)</tt>
+ *  such that no two consecutive elements in the range <tt>[keys_first, new_keys_last)</tt>
+ *  are equal.
+ *
+ *  This version of \p unique_by_key uses \c operator== to test for equality and 
+ *  \c project1st to reduce values with equal keys.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param keys_first The beginning of the key range.
+ *  \param keys_last  The end of the key range.
+ *  \param values_first The beginning of the value range.
+ *  \return A pair of iterators at end of the ranges <tt>[key_first, keys_new_last)</tt> and <tt>[values_first, values_new_last)</tt>.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam ForwardIterator1 is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>,
+ *          and \p ForwardIterator1 is mutable,
+ *          and \p ForwardIterator's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/EqualityComparable.html">Equality Comparable</a>.
+ *  \tparam ForwardIterator2 is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>,
+ *          and \p ForwardIterator2 is mutable.
+ *
+ *  \pre The range <tt>[keys_first, keys_last)</tt> and the range <tt>[values_first, values_first + (keys_last - keys_first))</tt> shall not overlap.
+ *
+ *  The following code snippet demonstrates how to use \p unique_by_key to
+ *  compact a sequence of key/value pairs to remove consecutive duplicates using the \p thrust::host
+ *  execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/unique.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  const int N = 7;
+ *  int A[N] = {1, 3, 3, 3, 2, 2, 1}; // keys
+ *  int B[N] = {9, 8, 7, 6, 5, 4, 3}; // values
+ *
+ *  thrust::pair<int*,int*> new_end;
+ *  new_end = thrust::unique_by_key(thrust::host, A, A + N, B);
+ *
+ *  // The first four keys in A are now {1, 3, 2, 1} and new_end.first - A is 4.
+ *  // The first four values in B are now {9, 8, 5, 3} and new_end.second - B is 4.
+ *  \endcode
+ *
+ *  \see unique
+ *  \see unique_by_key_copy
+ *  \see reduce_by_key
+ */
+template<typename DerivedPolicy,
+         typename ForwardIterator1,
+         typename ForwardIterator2>
+  thrust::pair<ForwardIterator1,ForwardIterator2>
+  unique_by_key(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                ForwardIterator1 keys_first, 
+                ForwardIterator1 keys_last,
+                ForwardIterator2 values_first);
+
+
+/*! \p unique_by_key is a generalization of \p unique to key-value pairs.
+ *  For each group of consecutive keys in the range <tt>[keys_first, keys_last)</tt>
+ *  that are equal, \p unique_by_key removes all but the first element of 
+ *  the group.  Similarly, the corresponding values in the range
+ *  <tt>[values_first, values_first + (keys_last - keys_first))</tt> 
+ *  are also removed.
+ *
+ *  The return value is a \p pair of iterators <tt>(new_keys_last,new_values_last)</tt>
+ *  such that no two consecutive elements in the range <tt>[keys_first, new_keys_last)</tt>
+ *  are equal.
+ *
+ *  This version of \p unique_by_key uses \c operator== to test for equality and 
+ *  \c project1st to reduce values with equal keys.
+ *
+ *  \param keys_first The beginning of the key range.
+ *  \param keys_last  The end of the key range.
+ *  \param values_first The beginning of the value range.
+ *  \return A pair of iterators at end of the ranges <tt>[key_first, keys_new_last)</tt> and <tt>[values_first, values_new_last)</tt>.
+ *
+ *  \tparam ForwardIterator1 is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>,
+ *          and \p ForwardIterator1 is mutable,
+ *          and \p ForwardIterator's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/EqualityComparable.html">Equality Comparable</a>.
+ *  \tparam ForwardIterator2 is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>,
+ *          and \p ForwardIterator2 is mutable.
+ *
+ *  \pre The range <tt>[keys_first, keys_last)</tt> and the range <tt>[values_first, values_first + (keys_last - keys_first))</tt> shall not overlap.
+ *
+ *  The following code snippet demonstrates how to use \p unique_by_key to
+ *  compact a sequence of key/value pairs to remove consecutive duplicates.
+ *
+ *  \code
+ *  #include <thrust/unique.h>
+ *  ...
+ *  const int N = 7;
+ *  int A[N] = {1, 3, 3, 3, 2, 2, 1}; // keys
+ *  int B[N] = {9, 8, 7, 6, 5, 4, 3}; // values
+ *
+ *  thrust::pair<int*,int*> new_end;
+ *  new_end = thrust::unique_by_key(A, A + N, B);
+ *
+ *  // The first four keys in A are now {1, 3, 2, 1} and new_end.first - A is 4.
+ *  // The first four values in B are now {9, 8, 5, 3} and new_end.second - B is 4.
+ *  \endcode
+ *
+ *  \see unique
+ *  \see unique_by_key_copy
+ *  \see reduce_by_key
+ */
+template <typename ForwardIterator1,
+          typename ForwardIterator2>
+  thrust::pair<ForwardIterator1,ForwardIterator2>
+  unique_by_key(ForwardIterator1 keys_first, 
+                ForwardIterator1 keys_last,
+                ForwardIterator2 values_first);
+
+
+/*! \p unique_by_key is a generalization of \p unique to key-value pairs.
+ *  For each group of consecutive keys in the range <tt>[keys_first, keys_last)</tt>
+ *  that are equal, \p unique_by_key removes all but the first element of 
+ *  the group.  Similarly, the corresponding values in the range
+ *  <tt>[values_first, values_first + (keys_last - keys_first))</tt> 
+ *  are also removed.
+ *
+ *  This version of \p unique_by_key uses the function object \c binary_pred
+ *  to test for equality and \c project1st to reduce values with equal keys.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param keys_first The beginning of the key range.
+ *  \param keys_last  The end of the key range.
+ *  \param values_first The beginning of the value range.
+ *  \param binary_pred  The binary predicate used to determine equality.
+ *  \return The end of the unique range <tt>[first, new_last)</tt>.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam ForwardIterator1 is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>,
+ *          and \p ForwardIterator1 is mutable,
+ *          and \p ForwardIterator's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/EqualityComparable.html">Equality Comparable</a>.
+ *  \tparam ForwardIterator2 is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>,
+ *          and \p ForwardIterator2 is mutable.
+ *  \tparam BinaryPredicate is a model of <a href="http://www.sgi.com/tech/stl/BinaryPredicate.html">Binary Predicate</a>.
+ *
+ *  \pre The range <tt>[keys_first, keys_last)</tt> and the range <tt>[values_first, values_first + (keys_last - keys_first))</tt> shall not overlap.
+ *
+ *  The following code snippet demonstrates how to use \p unique_by_key to
+ *  compact a sequence of key/value pairs to remove consecutive duplicates using the \p thrust::host
+ *  execution policy for parallelization:
+ *
+ *  \code
+ *  #include <thrust/unique.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  const int N = 7;
+ *  int A[N] = {1, 3, 3, 3, 2, 2, 1}; // keys
+ *  int B[N] = {9, 8, 7, 6, 5, 4, 3}; // values
+ *
+ *  thrust::pair<int*,int*> new_end;
+ *  thrust::equal_to<int> binary_pred;
+ *  new_end = thrust::unique_by_key(thrust::host, keys, keys + N, values, binary_pred);
+ *
+ *  // The first four keys in A are now {1, 3, 2, 1} and new_end.first - A is 4.
+ *  // The first four values in B are now {9, 8, 5, 3} and new_end.second - B is 4.
+ *  \endcode
+ *
+ *  \see unique
+ *  \see unique_by_key_copy
+ *  \see reduce_by_key
+ */
+template<typename DerivedPolicy,
+         typename ForwardIterator1,
+         typename ForwardIterator2,
+         typename BinaryPredicate>
+  thrust::pair<ForwardIterator1,ForwardIterator2>
+    unique_by_key(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                  ForwardIterator1 keys_first, 
+                  ForwardIterator1 keys_last,
+                  ForwardIterator2 values_first,
+                  BinaryPredicate binary_pred);
+
+
+/*! \p unique_by_key is a generalization of \p unique to key-value pairs.
+ *  For each group of consecutive keys in the range <tt>[keys_first, keys_last)</tt>
+ *  that are equal, \p unique_by_key removes all but the first element of 
+ *  the group.  Similarly, the corresponding values in the range
+ *  <tt>[values_first, values_first + (keys_last - keys_first))</tt> 
+ *  are also removed.
+ *
+ *  This version of \p unique_by_key uses the function object \c binary_pred
+ *  to test for equality and \c project1st to reduce values with equal keys.
+ *
+ *  \param keys_first The beginning of the key range.
+ *  \param keys_last  The end of the key range.
+ *  \param values_first The beginning of the value range.
+ *  \param binary_pred  The binary predicate used to determine equality.
+ *  \return The end of the unique range <tt>[first, new_last)</tt>.
+ *
+ *  \tparam ForwardIterator1 is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>,
+ *          and \p ForwardIterator1 is mutable,
+ *          and \p ForwardIterator's \c value_type is a model of <a href="http://www.sgi.com/tech/stl/EqualityComparable.html">Equality Comparable</a>.
+ *  \tparam ForwardIterator2 is a model of <a href="http://www.sgi.com/tech/stl/ForwardIterator.html">Forward Iterator</a>,
+ *          and \p ForwardIterator2 is mutable.
+ *  \tparam BinaryPredicate is a model of <a href="http://www.sgi.com/tech/stl/BinaryPredicate.html">Binary Predicate</a>.
+ *
+ *  \pre The range <tt>[keys_first, keys_last)</tt> and the range <tt>[values_first, values_first + (keys_last - keys_first))</tt> shall not overlap.
+ *
+ *  The following code snippet demonstrates how to use \p unique_by_key to
+ *  compact a sequence of key/value pairs to remove consecutive duplicates.
+ *
+ *  \code
+ *  #include <thrust/unique.h>
+ *  ...
+ *  const int N = 7;
+ *  int A[N] = {1, 3, 3, 3, 2, 2, 1}; // keys
+ *  int B[N] = {9, 8, 7, 6, 5, 4, 3}; // values
+ *
+ *  thrust::pair<int*,int*> new_end;
+ *  thrust::equal_to<int> binary_pred;
+ *  new_end = thrust::unique_by_key(keys, keys + N, values, binary_pred);
+ *
+ *  // The first four keys in A are now {1, 3, 2, 1} and new_end.first - A is 4.
+ *  // The first four values in B are now {9, 8, 5, 3} and new_end.second - B is 4.
+ *  \endcode
+ *
+ *  \see unique
+ *  \see unique_by_key_copy
+ *  \see reduce_by_key
+ */
+template <typename ForwardIterator1,
+          typename ForwardIterator2,
+          typename BinaryPredicate>
+  thrust::pair<ForwardIterator1,ForwardIterator2>
+  unique_by_key(ForwardIterator1 keys_first, 
+                ForwardIterator1 keys_last,
+                ForwardIterator2 values_first,
+                BinaryPredicate binary_pred);
+
+
+/*! \p unique_by_key_copy is a generalization of \p unique_copy to key-value pairs.
+ *  For each group of consecutive keys in the range <tt>[keys_first, keys_last)</tt>
+ *  that are equal, \p unique_by_key_copy copies the first element of the group to
+ *  a range beginning with \c keys_result and the corresponding values from the range
+ *  <tt>[values_first, values_first + (keys_last - keys_first))</tt> are copied to a range
+ *  beginning with \c values_result.
+ *
+ *  This version of \p unique_by_key_copy uses \c operator== to test for equality and
+ *  \c project1st to reduce values with equal keys.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param keys_first The beginning of the input key range.
+ *  \param keys_last  The end of the input key range.
+ *  \param values_first The beginning of the input value range.
+ *  \param keys_result The beginning of the output key range.
+ *  \param values_result The beginning of the output value range.
+ *  \return A pair of iterators at end of the ranges <tt>[keys_result, keys_result_last)</tt> and <tt>[values_result, values_result_last)</tt>.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *  \tparam OutputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a> and
+ *          and \p InputIterator1's \c value_type is convertible to \c OutputIterator1's \c value_type.
+ *  \tparam OutputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a> and
+ *          and \p InputIterator2's \c value_type is convertible to \c OutputIterator2's \c value_type.
+ *
+ *  \pre The input ranges shall not overlap either output range.
+ *
+ *  The following code snippet demonstrates how to use \p unique_by_key_copy to
+ *  compact a sequence of key/value pairs and with equal keys using the \p thrust::host execution policy
+ *  for parallelization:
+ *
+ *  \code
+ *  #include <thrust/unique.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  const int N = 7;
+ *  int A[N] = {1, 3, 3, 3, 2, 2, 1}; // input keys
+ *  int B[N] = {9, 8, 7, 6, 5, 4, 3}; // input values
+ *  int C[N];                         // output keys
+ *  int D[N];                         // output values
+ *
+ *  thrust::pair<int*,int*> new_end;
+ *  new_end = thrust::unique_by_key_copy(thrust::host, A, A + N, B, C, D);
+ *
+ *  // The first four keys in C are now {1, 3, 2, 1} and new_end.first - C is 4.
+ *  // The first four values in D are now {9, 8, 5, 3} and new_end.second - D is 4.
+ *  \endcode
+ *
+ *  \see unique_copy
+ *  \see unique_by_key
+ *  \see reduce_by_key
+ */
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator1,
+         typename OutputIterator2>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    unique_by_key_copy(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                       InputIterator1 keys_first, 
+                       InputIterator1 keys_last,
+                       InputIterator2 values_first,
+                       OutputIterator1 keys_result,
+                       OutputIterator2 values_result);
+
+
+/*! \p unique_by_key_copy is a generalization of \p unique_copy to key-value pairs.
+ *  For each group of consecutive keys in the range <tt>[keys_first, keys_last)</tt>
+ *  that are equal, \p unique_by_key_copy copies the first element of the group to
+ *  a range beginning with \c keys_result and the corresponding values from the range
+ *  <tt>[values_first, values_first + (keys_last - keys_first))</tt> are copied to a range
+ *  beginning with \c values_result.
+ *
+ *  This version of \p unique_by_key_copy uses \c operator== to test for equality and
+ *  \c project1st to reduce values with equal keys.
+ *
+ *  \param keys_first The beginning of the input key range.
+ *  \param keys_last  The end of the input key range.
+ *  \param values_first The beginning of the input value range.
+ *  \param keys_result The beginning of the output key range.
+ *  \param values_result The beginning of the output value range.
+ *  \return A pair of iterators at end of the ranges <tt>[keys_result, keys_result_last)</tt> and <tt>[values_result, values_result_last)</tt>.
+ *
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *  \tparam OutputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a> and
+ *          and \p InputIterator1's \c value_type is convertible to \c OutputIterator1's \c value_type.
+ *  \tparam OutputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a> and
+ *          and \p InputIterator2's \c value_type is convertible to \c OutputIterator2's \c value_type.
+ *
+ *  \pre The input ranges shall not overlap either output range.
+ *
+ *  The following code snippet demonstrates how to use \p unique_by_key_copy to
+ *  compact a sequence of key/value pairs and with equal keys.
+ *
+ *  \code
+ *  #include <thrust/unique.h>
+ *  ...
+ *  const int N = 7;
+ *  int A[N] = {1, 3, 3, 3, 2, 2, 1}; // input keys
+ *  int B[N] = {9, 8, 7, 6, 5, 4, 3}; // input values
+ *  int C[N];                         // output keys
+ *  int D[N];                         // output values
+ *
+ *  thrust::pair<int*,int*> new_end;
+ *  new_end = thrust::unique_by_key_copy(A, A + N, B, C, D);
+ *
+ *  // The first four keys in C are now {1, 3, 2, 1} and new_end.first - C is 4.
+ *  // The first four values in D are now {9, 8, 5, 3} and new_end.second - D is 4.
+ *  \endcode
+ *
+ *  \see unique_copy
+ *  \see unique_by_key
+ *  \see reduce_by_key
+ */
+template <typename InputIterator1,
+          typename InputIterator2,
+          typename OutputIterator1,
+          typename OutputIterator2>
+  thrust::pair<OutputIterator1,OutputIterator2>
+  unique_by_key_copy(InputIterator1 keys_first, 
+                     InputIterator1 keys_last,
+                     InputIterator2 values_first,
+                     OutputIterator1 keys_result,
+                     OutputIterator2 values_result);
+
+
+/*! \p unique_by_key_copy is a generalization of \p unique_copy to key-value pairs.
+ *  For each group of consecutive keys in the range <tt>[keys_first, keys_last)</tt>
+ *  that are equal, \p unique_by_key_copy copies the first element of the group to
+ *  a range beginning with \c keys_result and the corresponding values from the range
+ *  <tt>[values_first, values_first + (keys_last - keys_first))</tt> are copied to a range
+ *  beginning with \c values_result.
+ *
+ *  This version of \p unique_by_key_copy uses the function object \c binary_pred
+ *  to test for equality and \c project1st to reduce values with equal keys.
+ *
+ *  The algorithm's execution is parallelized as determined by \p exec.
+ *
+ *  \param exec The execution policy to use for parallelization.
+ *  \param keys_first The beginning of the input key range.
+ *  \param keys_last  The end of the input key range.
+ *  \param values_first The beginning of the input value range.
+ *  \param keys_result The beginning of the output key range.
+ *  \param values_result The beginning of the output value range.
+ *  \param binary_pred  The binary predicate used to determine equality.
+ *  \return A pair of iterators at end of the ranges <tt>[keys_result, keys_result_last)</tt> and <tt>[values_result, values_result_last)</tt>.
+ *
+ *  \tparam DerivedPolicy The name of the derived execution policy.
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *  \tparam OutputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a> and
+ *          and \p InputIterator1's \c value_type is convertible to \c OutputIterator1's \c value_type.
+ *  \tparam OutputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a> and
+ *          and \p InputIterator2's \c value_type is convertible to \c OutputIterator2's \c value_type.
+ *  \tparam BinaryPredicate is a model of <a href="http://www.sgi.com/tech/stl/BinaryPredicate.html">Binary Predicate</a>.
+ *
+ *  \pre The input ranges shall not overlap either output range.
+ *
+ *  The following code snippet demonstrates how to use \p unique_by_key_copy to
+ *  compact a sequence of key/value pairs and with equal keys using the \p thrust::host execution policy for
+ *  parallelization:
+ *
+ *  \code
+ *  #include <thrust/unique.h>
+ *  #include <thrust/execution_policy.h>
+ *  ...
+ *  const int N = 7;
+ *  int A[N] = {1, 3, 3, 3, 2, 2, 1}; // input keys
+ *  int B[N] = {9, 8, 7, 6, 5, 4, 3}; // input values
+ *  int C[N];                         // output keys
+ *  int D[N];                         // output values
+ *
+ *  thrust::pair<int*,int*> new_end;
+ *  thrust::equal_to<int> binary_pred;
+ *  new_end = thrust::unique_by_key_copy(thrust::host, A, A + N, B, C, D, binary_pred);
+ *
+ *  // The first four keys in C are now {1, 3, 2, 1} and new_end.first - C is 4.
+ *  // The first four values in D are now {9, 8, 5, 3} and new_end.second - D is 4.
+ *  \endcode
+ *
+ *  \see unique_copy
+ *  \see unique_by_key
+ *  \see reduce_by_key
+ */
+template<typename DerivedPolicy,
+         typename InputIterator1,
+         typename InputIterator2,
+         typename OutputIterator1,
+         typename OutputIterator2,
+         typename BinaryPredicate>
+  thrust::pair<OutputIterator1,OutputIterator2>
+    unique_by_key_copy(const thrust::detail::execution_policy_base<DerivedPolicy> &exec,
+                       InputIterator1 keys_first, 
+                       InputIterator1 keys_last,
+                       InputIterator2 values_first,
+                       OutputIterator1 keys_result,
+                       OutputIterator2 values_result,
+                       BinaryPredicate binary_pred);
+
+
+/*! \p unique_by_key_copy is a generalization of \p unique_copy to key-value pairs.
+ *  For each group of consecutive keys in the range <tt>[keys_first, keys_last)</tt>
+ *  that are equal, \p unique_by_key_copy copies the first element of the group to
+ *  a range beginning with \c keys_result and the corresponding values from the range
+ *  <tt>[values_first, values_first + (keys_last - keys_first))</tt> are copied to a range
+ *  beginning with \c values_result.
+ *
+ *  This version of \p unique_by_key_copy uses the function object \c binary_pred
+ *  to test for equality and \c project1st to reduce values with equal keys.
+ *
+ *  \param keys_first The beginning of the input key range.
+ *  \param keys_last  The end of the input key range.
+ *  \param values_first The beginning of the input value range.
+ *  \param keys_result The beginning of the output key range.
+ *  \param values_result The beginning of the output value range.
+ *  \param binary_pred  The binary predicate used to determine equality.
+ *  \return A pair of iterators at end of the ranges <tt>[keys_result, keys_result_last)</tt> and <tt>[values_result, values_result_last)</tt>.
+ *
+ *  \tparam InputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *  \tparam InputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>,
+ *  \tparam OutputIterator1 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a> and
+ *          and \p InputIterator1's \c value_type is convertible to \c OutputIterator1's \c value_type.
+ *  \tparam OutputIterator2 is a model of <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a> and
+ *          and \p InputIterator2's \c value_type is convertible to \c OutputIterator2's \c value_type.
+ *  \tparam BinaryPredicate is a model of <a href="http://www.sgi.com/tech/stl/BinaryPredicate.html">Binary Predicate</a>.
+ *
+ *  \pre The input ranges shall not overlap either output range.
+ *
+ *  The following code snippet demonstrates how to use \p unique_by_key_copy to
+ *  compact a sequence of key/value pairs and with equal keys.
+ *
+ *  \code
+ *  #include <thrust/unique.h>
+ *  ...
+ *  const int N = 7;
+ *  int A[N] = {1, 3, 3, 3, 2, 2, 1}; // input keys
+ *  int B[N] = {9, 8, 7, 6, 5, 4, 3}; // input values
+ *  int C[N];                         // output keys
+ *  int D[N];                         // output values
+ *
+ *  thrust::pair<int*,int*> new_end;
+ *  thrust::equal_to<int> binary_pred;
+ *  new_end = thrust::unique_by_key_copy(A, A + N, B, C, D, binary_pred);
+ *
+ *  // The first four keys in C are now {1, 3, 2, 1} and new_end.first - C is 4.
+ *  // The first four values in D are now {9, 8, 5, 3} and new_end.second - D is 4.
+ *  \endcode
+ *
+ *  \see unique_copy
+ *  \see unique_by_key
+ *  \see reduce_by_key
+ */
+template <typename InputIterator1,
+          typename InputIterator2,
+          typename OutputIterator1,
+          typename OutputIterator2,
+          typename BinaryPredicate>
+  thrust::pair<OutputIterator1,OutputIterator2>
+  unique_by_key_copy(InputIterator1 keys_first, 
+                     InputIterator1 keys_last,
+                     InputIterator2 values_first,
+                     OutputIterator1 keys_result,
+                     OutputIterator2 values_result,
+                     BinaryPredicate binary_pred);
+
+
+/*! \} // end stream_compaction
+ */
+
+
+} // end namespace thrust
+
+#include <thrust/detail/unique.inl>
+
diff --git a/compat/thrust/version.h b/compat/thrust/version.h
new file mode 100644
index 0000000000..730997eecc
--- /dev/null
+++ b/compat/thrust/version.h
@@ -0,0 +1,73 @@
+/*
+ *  Copyright 2008-2012 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+/*! \file version.h
+ *  \brief Compile-time macros encoding Thrust release version
+ *
+ *         <thrust/version.h> is the only Thrust header that is guaranteed to
+ *         change with every thrust release.
+ */
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+//  This is the only thrust header that is guaranteed to 
+//  change with every thrust release.
+//
+//  THRUST_VERSION % 100 is the sub-minor version
+//  THRUST_VERSION / 100 % 1000 is the minor version
+//  THRUST_VERSION / 100000 is the major version
+
+/*! \def THRUST_VERSION
+ *  \brief The preprocessor macro \p THRUST_VERSION encodes the version
+ *         number of the Thrust library.
+ *
+ *         <tt>THRUST_VERSION % 100</tt> is the sub-minor version.
+ *         <tt>THRUST_VERSION / 100 % 1000</tt> is the minor version.
+ *         <tt>THRUST_VERSION / 100000</tt> is the major version.
+ */
+#define THRUST_VERSION 100700
+
+/*! \def THRUST_MAJOR_VERSION
+ *  \brief The preprocessor macro \p THRUST_MAJOR_VERSION encodes the
+ *         major version number of the Thrust library.
+ */
+#define THRUST_MAJOR_VERSION     (THRUST_VERSION / 100000)
+
+/*! \def THRUST_MINOR_VERSION
+ *  \brief The preprocessor macro \p THRUST_MINOR_VERSION encodes the
+ *         minor version number of the Thrust library.
+ */
+#define THRUST_MINOR_VERSION     (THRUST_VERSION / 100 % 1000)
+
+/*! \def THRUST_SUBMINOR_VERSION
+ *  \brief The preprocessor macro \p THRUST_SUBMINOR_VERSION encodes the
+ *         sub-minor version number of the Thrust library.
+ */
+#define THRUST_SUBMINOR_VERSION  (THRUST_VERSION % 100)
+
+// Declare these namespaces here for the purpose of Doxygenating them
+
+/*! \namespace thrust
+ *  \brief \p thrust is the top-level namespace which contains all Thrust
+ *         functions and types.
+ */
+namespace thrust
+{
+
+}
+
diff --git a/compat/unistd.h b/compat/unistd.h
new file mode 100644
index 0000000000..193da661a1
--- /dev/null
+++ b/compat/unistd.h
@@ -0,0 +1,2 @@
+#pragma once
+#include "getopt/getopt.h"
\ No newline at end of file
diff --git a/compile b/compile
new file mode 100644
index 0000000000..a49b6d0905
--- /dev/null
+++ b/compile
@@ -0,0 +1,310 @@
+#! /bin/sh
+# Wrapper for compilers which do not understand '-c -o'.
+
+scriptversion=2012-01-04.17; # UTC
+
+# Copyright (C) 1999, 2000, 2003, 2004, 2005, 2009, 2010, 2012 Free
+# Software Foundation, Inc.
+# Written by Tom Tromey <tromey@cygnus.com>.
+#
+# This program is free software; you can redistribute it and/or modify
+# it under the terms of the GNU General Public License as published by
+# the Free Software Foundation; either version 2, or (at your option)
+# any later version.
+#
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU General Public License
+# along with this program.  If not, see <http://www.gnu.org/licenses/>.
+
+# As a special exception to the GNU General Public License, if you
+# distribute this file as part of a program that contains a
+# configuration script generated by Autoconf, you may include it under
+# the same distribution terms that you use for the rest of that program.
+
+# This file is maintained in Automake, please report
+# bugs to <bug-automake@gnu.org> or send patches to
+# <automake-patches@gnu.org>.
+
+nl='
+'
+
+# We need space, tab and new line, in precisely that order.  Quoting is
+# there to prevent tools from complaining about whitespace usage.
+IFS=" ""	$nl"
+
+file_conv=
+
+# func_file_conv build_file lazy
+# Convert a $build file to $host form and store it in $file
+# Currently only supports Windows hosts. If the determined conversion
+# type is listed in (the comma separated) LAZY, no conversion will
+# take place.
+func_file_conv ()
+{
+  file=$1
+  case $file in
+    / | /[!/]*) # absolute file, and not a UNC file
+      if test -z "$file_conv"; then
+	# lazily determine how to convert abs files
+	case `uname -s` in
+	  MINGW*)
+	    file_conv=mingw
+	    ;;
+	  CYGWIN*)
+	    file_conv=cygwin
+	    ;;
+	  *)
+	    file_conv=wine
+	    ;;
+	esac
+      fi
+      case $file_conv/,$2, in
+	*,$file_conv,*)
+	  ;;
+	mingw/*)
+	  file=`cmd //C echo "$file " | sed -e 's/"\(.*\) " *$/\1/'`
+	  ;;
+	cygwin/*)
+	  file=`cygpath -m "$file" || echo "$file"`
+	  ;;
+	wine/*)
+	  file=`winepath -w "$file" || echo "$file"`
+	  ;;
+      esac
+      ;;
+  esac
+}
+
+# func_cl_wrapper cl arg...
+# Adjust compile command to suit cl
+func_cl_wrapper ()
+{
+  # Assume a capable shell
+  lib_path=
+  shared=:
+  linker_opts=
+  for arg
+  do
+    if test -n "$eat"; then
+      eat=
+    else
+      case $1 in
+	-o)
+	  # configure might choose to run compile as 'compile cc -o foo foo.c'.
+	  eat=1
+	  case $2 in
+	    *.o | *.[oO][bB][jJ])
+	      func_file_conv "$2"
+	      set x "$@" -Fo"$file"
+	      shift
+	      ;;
+	    *)
+	      func_file_conv "$2"
+	      set x "$@" -Fe"$file"
+	      shift
+	      ;;
+	  esac
+	  ;;
+	-I*)
+	  func_file_conv "${1#-I}" mingw
+	  set x "$@" -I"$file"
+	  shift
+	  ;;
+	-l*)
+	  lib=${1#-l}
+	  found=no
+	  save_IFS=$IFS
+	  IFS=';'
+	  for dir in $lib_path $LIB
+	  do
+	    IFS=$save_IFS
+	    if $shared && test -f "$dir/$lib.dll.lib"; then
+	      found=yes
+	      set x "$@" "$dir/$lib.dll.lib"
+	      break
+	    fi
+	    if test -f "$dir/$lib.lib"; then
+	      found=yes
+	      set x "$@" "$dir/$lib.lib"
+	      break
+	    fi
+	  done
+	  IFS=$save_IFS
+
+	  test "$found" != yes && set x "$@" "$lib.lib"
+	  shift
+	  ;;
+	-L*)
+	  func_file_conv "${1#-L}"
+	  if test -z "$lib_path"; then
+	    lib_path=$file
+	  else
+	    lib_path="$lib_path;$file"
+	  fi
+	  linker_opts="$linker_opts -LIBPATH:$file"
+	  ;;
+	-static)
+	  shared=false
+	  ;;
+	-Wl,*)
+	  arg=${1#-Wl,}
+	  save_ifs="$IFS"; IFS=','
+	  for flag in $arg; do
+	    IFS="$save_ifs"
+	    linker_opts="$linker_opts $flag"
+	  done
+	  IFS="$save_ifs"
+	  ;;
+	-Xlinker)
+	  eat=1
+	  linker_opts="$linker_opts $2"
+	  ;;
+	-*)
+	  set x "$@" "$1"
+	  shift
+	  ;;
+	*.cc | *.CC | *.cxx | *.CXX | *.[cC]++)
+	  func_file_conv "$1"
+	  set x "$@" -Tp"$file"
+	  shift
+	  ;;
+	*.c | *.cpp | *.CPP | *.lib | *.LIB | *.Lib | *.OBJ | *.obj | *.[oO])
+	  func_file_conv "$1" mingw
+	  set x "$@" "$file"
+	  shift
+	  ;;
+	*)
+	  set x "$@" "$1"
+	  shift
+	  ;;
+      esac
+    fi
+    shift
+  done
+  if test -n "$linker_opts"; then
+    linker_opts="-link$linker_opts"
+  fi
+  exec "$@" $linker_opts
+  exit 1
+}
+
+eat=
+
+case $1 in
+  '')
+     echo "$0: No command.  Try '$0 --help' for more information." 1>&2
+     exit 1;
+     ;;
+  -h | --h*)
+    cat <<\EOF
+Usage: compile [--help] [--version] PROGRAM [ARGS]
+
+Wrapper for compilers which do not understand '-c -o'.
+Remove '-o dest.o' from ARGS, run PROGRAM with the remaining
+arguments, and rename the output as expected.
+
+If you are trying to build a whole package this is not the
+right script to run: please start by reading the file 'INSTALL'.
+
+Report bugs to <bug-automake@gnu.org>.
+EOF
+    exit $?
+    ;;
+  -v | --v*)
+    echo "compile $scriptversion"
+    exit $?
+    ;;
+  cl | *[/\\]cl | cl.exe | *[/\\]cl.exe )
+    func_cl_wrapper "$@"      # Doesn't return...
+    ;;
+esac
+
+ofile=
+cfile=
+
+for arg
+do
+  if test -n "$eat"; then
+    eat=
+  else
+    case $1 in
+      -o)
+	# configure might choose to run compile as 'compile cc -o foo foo.c'.
+	# So we strip '-o arg' only if arg is an object.
+	eat=1
+	case $2 in
+	  *.o | *.obj)
+	    ofile=$2
+	    ;;
+	  *)
+	    set x "$@" -o "$2"
+	    shift
+	    ;;
+	esac
+	;;
+      *.c)
+	cfile=$1
+	set x "$@" "$1"
+	shift
+	;;
+      *)
+	set x "$@" "$1"
+	shift
+	;;
+    esac
+  fi
+  shift
+done
+
+if test -z "$ofile" || test -z "$cfile"; then
+  # If no '-o' option was seen then we might have been invoked from a
+  # pattern rule where we don't need one.  That is ok -- this is a
+  # normal compilation that the losing compiler can handle.  If no
+  # '.c' file was seen then we are probably linking.  That is also
+  # ok.
+  exec "$@"
+fi
+
+# Name of file we expect compiler to create.
+cofile=`echo "$cfile" | sed 's|^.*[\\/]||; s|^[a-zA-Z]:||; s/\.c$/.o/'`
+
+# Create the lock directory.
+# Note: use '[/\\:.-]' here to ensure that we don't use the same name
+# that we are using for the .o file.  Also, base the name on the expected
+# object file name, since that is what matters with a parallel build.
+lockdir=`echo "$cofile" | sed -e 's|[/\\:.-]|_|g'`.d
+while true; do
+  if mkdir "$lockdir" >/dev/null 2>&1; then
+    break
+  fi
+  sleep 1
+done
+# FIXME: race condition here if user kills between mkdir and trap.
+trap "rmdir '$lockdir'; exit 1" 1 2 15
+
+# Run the compile.
+"$@"
+ret=$?
+
+if test -f "$cofile"; then
+  test "$cofile" = "$ofile" || mv "$cofile" "$ofile"
+elif test -f "${cofile}bj"; then
+  test "${cofile}bj" = "$ofile" || mv "${cofile}bj" "$ofile"
+fi
+
+rmdir "$lockdir"
+exit $ret
+
+# Local Variables:
+# mode: shell-script
+# sh-indentation: 2
+# eval: (add-hook 'write-file-hooks 'time-stamp)
+# time-stamp-start: "scriptversion="
+# time-stamp-format: "%:y-%02m-%02d.%02H"
+# time-stamp-time-zone: "UTC"
+# time-stamp-end: "; # UTC"
+# End:
diff --git a/config.guess b/config.guess
new file mode 100644
index 0000000000..f32079abda
--- /dev/null
+++ b/config.guess
@@ -0,0 +1,1526 @@
+#! /bin/sh
+# Attempt to guess a canonical system name.
+#   Copyright (C) 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
+#   2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
+#   Free Software Foundation, Inc.
+
+timestamp='2008-01-23'
+
+# This file is free software; you can redistribute it and/or modify it
+# under the terms of the GNU General Public License as published by
+# the Free Software Foundation; either version 2 of the License, or
+# (at your option) any later version.
+#
+# This program is distributed in the hope that it will be useful, but
+# WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+# General Public License for more details.
+#
+# You should have received a copy of the GNU General Public License
+# along with this program; if not, write to the Free Software
+# Foundation, Inc., 51 Franklin Street - Fifth Floor, Boston, MA
+# 02110-1301, USA.
+#
+# As a special exception to the GNU General Public License, if you
+# distribute this file as part of a program that contains a
+# configuration script generated by Autoconf, you may include it under
+# the same distribution terms that you use for the rest of that program.
+
+
+# Originally written by Per Bothner <per@bothner.com>.
+# Please send patches to <config-patches@gnu.org>.  Submit a context
+# diff and a properly formatted ChangeLog entry.
+#
+# This script attempts to guess a canonical system name similar to
+# config.sub.  If it succeeds, it prints the system name on stdout, and
+# exits with 0.  Otherwise, it exits with 1.
+#
+# The plan is that this can be called by configure scripts if you
+# don't specify an explicit build system type.
+
+me=`echo "$0" | sed -e 's,.*/,,'`
+
+usage="\
+Usage: $0 [OPTION]
+
+Output the configuration name of the system \`$me' is run on.
+
+Operation modes:
+  -h, --help         print this help, then exit
+  -t, --time-stamp   print date of last modification, then exit
+  -v, --version      print version number, then exit
+
+Report bugs and patches to <config-patches@gnu.org>."
+
+version="\
+GNU config.guess ($timestamp)
+
+Originally written by Per Bothner.
+Copyright (C) 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000, 2001,
+2002, 2003, 2004, 2005, 2006, 2007, 2008 Free Software Foundation, Inc.
+
+This is free software; see the source for copying conditions.  There is NO
+warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE."
+
+help="
+Try \`$me --help' for more information."
+
+# Parse command line
+while test $# -gt 0 ; do
+  case $1 in
+    --time-stamp | --time* | -t )
+       echo "$timestamp" ; exit ;;
+    --version | -v )
+       echo "$version" ; exit ;;
+    --help | --h* | -h )
+       echo "$usage"; exit ;;
+    -- )     # Stop option processing
+       shift; break ;;
+    - )	# Use stdin as input.
+       break ;;
+    -* )
+       echo "$me: invalid option $1$help" >&2
+       exit 1 ;;
+    * )
+       break ;;
+  esac
+done
+
+if test $# != 0; then
+  echo "$me: too many arguments$help" >&2
+  exit 1
+fi
+
+trap 'exit 1' 1 2 15
+
+# CC_FOR_BUILD -- compiler used by this script. Note that the use of a
+# compiler to aid in system detection is discouraged as it requires
+# temporary files to be created and, as you can see below, it is a
+# headache to deal with in a portable fashion.
+
+# Historically, `CC_FOR_BUILD' used to be named `HOST_CC'. We still
+# use `HOST_CC' if defined, but it is deprecated.
+
+# Portable tmp directory creation inspired by the Autoconf team.
+
+set_cc_for_build='
+trap "exitcode=\$?; (rm -f \$tmpfiles 2>/dev/null; rmdir \$tmp 2>/dev/null) && exit \$exitcode" 0 ;
+trap "rm -f \$tmpfiles 2>/dev/null; rmdir \$tmp 2>/dev/null; exit 1" 1 2 13 15 ;
+: ${TMPDIR=/tmp} ;
+ { tmp=`(umask 077 && mktemp -d "$TMPDIR/cgXXXXXX") 2>/dev/null` && test -n "$tmp" && test -d "$tmp" ; } ||
+ { test -n "$RANDOM" && tmp=$TMPDIR/cg$$-$RANDOM && (umask 077 && mkdir $tmp) ; } ||
+ { tmp=$TMPDIR/cg-$$ && (umask 077 && mkdir $tmp) && echo "Warning: creating insecure temp directory" >&2 ; } ||
+ { echo "$me: cannot create a temporary directory in $TMPDIR" >&2 ; exit 1 ; } ;
+dummy=$tmp/dummy ;
+tmpfiles="$dummy.c $dummy.o $dummy.rel $dummy" ;
+case $CC_FOR_BUILD,$HOST_CC,$CC in
+ ,,)    echo "int x;" > $dummy.c ;
+	for c in cc gcc c89 c99 ; do
+	  if ($c -c -o $dummy.o $dummy.c) >/dev/null 2>&1 ; then
+	     CC_FOR_BUILD="$c"; break ;
+	  fi ;
+	done ;
+	if test x"$CC_FOR_BUILD" = x ; then
+	  CC_FOR_BUILD=no_compiler_found ;
+	fi
+	;;
+ ,,*)   CC_FOR_BUILD=$CC ;;
+ ,*,*)  CC_FOR_BUILD=$HOST_CC ;;
+esac ; set_cc_for_build= ;'
+
+# This is needed to find uname on a Pyramid OSx when run in the BSD universe.
+# (ghazi@noc.rutgers.edu 1994-08-24)
+if (test -f /.attbin/uname) >/dev/null 2>&1 ; then
+	PATH=$PATH:/.attbin ; export PATH
+fi
+
+UNAME_MACHINE=`(uname -m) 2>/dev/null` || UNAME_MACHINE=unknown
+UNAME_RELEASE=`(uname -r) 2>/dev/null` || UNAME_RELEASE=unknown
+UNAME_SYSTEM=`(uname -s) 2>/dev/null`  || UNAME_SYSTEM=unknown
+UNAME_VERSION=`(uname -v) 2>/dev/null` || UNAME_VERSION=unknown
+
+# Note: order is significant - the case branches are not exclusive.
+
+case "${UNAME_MACHINE}:${UNAME_SYSTEM}:${UNAME_RELEASE}:${UNAME_VERSION}" in
+    *:NetBSD:*:*)
+	# NetBSD (nbsd) targets should (where applicable) match one or
+	# more of the tupples: *-*-netbsdelf*, *-*-netbsdaout*,
+	# *-*-netbsdecoff* and *-*-netbsd*.  For targets that recently
+	# switched to ELF, *-*-netbsd* would select the old
+	# object file format.  This provides both forward
+	# compatibility and a consistent mechanism for selecting the
+	# object file format.
+	#
+	# Note: NetBSD doesn't particularly care about the vendor
+	# portion of the name.  We always set it to "unknown".
+	sysctl="sysctl -n hw.machine_arch"
+	UNAME_MACHINE_ARCH=`(/sbin/$sysctl 2>/dev/null || \
+	    /usr/sbin/$sysctl 2>/dev/null || echo unknown)`
+	case "${UNAME_MACHINE_ARCH}" in
+	    armeb) machine=armeb-unknown ;;
+	    arm*) machine=arm-unknown ;;
+	    sh3el) machine=shl-unknown ;;
+	    sh3eb) machine=sh-unknown ;;
+	    sh5el) machine=sh5le-unknown ;;
+	    *) machine=${UNAME_MACHINE_ARCH}-unknown ;;
+	esac
+	# The Operating System including object format, if it has switched
+	# to ELF recently, or will in the future.
+	case "${UNAME_MACHINE_ARCH}" in
+	    arm*|i386|m68k|ns32k|sh3*|sparc|vax)
+		eval $set_cc_for_build
+		if echo __ELF__ | $CC_FOR_BUILD -E - 2>/dev/null \
+			| grep __ELF__ >/dev/null
+		then
+		    # Once all utilities can be ECOFF (netbsdecoff) or a.out (netbsdaout).
+		    # Return netbsd for either.  FIX?
+		    os=netbsd
+		else
+		    os=netbsdelf
+		fi
+		;;
+	    *)
+	        os=netbsd
+		;;
+	esac
+	# The OS release
+	# Debian GNU/NetBSD machines have a different userland, and
+	# thus, need a distinct triplet. However, they do not need
+	# kernel version information, so it can be replaced with a
+	# suitable tag, in the style of linux-gnu.
+	case "${UNAME_VERSION}" in
+	    Debian*)
+		release='-gnu'
+		;;
+	    *)
+		release=`echo ${UNAME_RELEASE}|sed -e 's/[-_].*/\./'`
+		;;
+	esac
+	# Since CPU_TYPE-MANUFACTURER-KERNEL-OPERATING_SYSTEM:
+	# contains redundant information, the shorter form:
+	# CPU_TYPE-MANUFACTURER-OPERATING_SYSTEM is used.
+	echo "${machine}-${os}${release}"
+	exit ;;
+    *:OpenBSD:*:*)
+	UNAME_MACHINE_ARCH=`arch | sed 's/OpenBSD.//'`
+	echo ${UNAME_MACHINE_ARCH}-unknown-openbsd${UNAME_RELEASE}
+	exit ;;
+    *:ekkoBSD:*:*)
+	echo ${UNAME_MACHINE}-unknown-ekkobsd${UNAME_RELEASE}
+	exit ;;
+    *:SolidBSD:*:*)
+	echo ${UNAME_MACHINE}-unknown-solidbsd${UNAME_RELEASE}
+	exit ;;
+    macppc:MirBSD:*:*)
+	echo powerpc-unknown-mirbsd${UNAME_RELEASE}
+	exit ;;
+    *:MirBSD:*:*)
+	echo ${UNAME_MACHINE}-unknown-mirbsd${UNAME_RELEASE}
+	exit ;;
+    alpha:OSF1:*:*)
+	case $UNAME_RELEASE in
+	*4.0)
+		UNAME_RELEASE=`/usr/sbin/sizer -v | awk '{print $3}'`
+		;;
+	*5.*)
+	        UNAME_RELEASE=`/usr/sbin/sizer -v | awk '{print $4}'`
+		;;
+	esac
+	# According to Compaq, /usr/sbin/psrinfo has been available on
+	# OSF/1 and Tru64 systems produced since 1995.  I hope that
+	# covers most systems running today.  This code pipes the CPU
+	# types through head -n 1, so we only detect the type of CPU 0.
+	ALPHA_CPU_TYPE=`/usr/sbin/psrinfo -v | sed -n -e 's/^  The alpha \(.*\) processor.*$/\1/p' | head -n 1`
+	case "$ALPHA_CPU_TYPE" in
+	    "EV4 (21064)")
+		UNAME_MACHINE="alpha" ;;
+	    "EV4.5 (21064)")
+		UNAME_MACHINE="alpha" ;;
+	    "LCA4 (21066/21068)")
+		UNAME_MACHINE="alpha" ;;
+	    "EV5 (21164)")
+		UNAME_MACHINE="alphaev5" ;;
+	    "EV5.6 (21164A)")
+		UNAME_MACHINE="alphaev56" ;;
+	    "EV5.6 (21164PC)")
+		UNAME_MACHINE="alphapca56" ;;
+	    "EV5.7 (21164PC)")
+		UNAME_MACHINE="alphapca57" ;;
+	    "EV6 (21264)")
+		UNAME_MACHINE="alphaev6" ;;
+	    "EV6.7 (21264A)")
+		UNAME_MACHINE="alphaev67" ;;
+	    "EV6.8CB (21264C)")
+		UNAME_MACHINE="alphaev68" ;;
+	    "EV6.8AL (21264B)")
+		UNAME_MACHINE="alphaev68" ;;
+	    "EV6.8CX (21264D)")
+		UNAME_MACHINE="alphaev68" ;;
+	    "EV6.9A (21264/EV69A)")
+		UNAME_MACHINE="alphaev69" ;;
+	    "EV7 (21364)")
+		UNAME_MACHINE="alphaev7" ;;
+	    "EV7.9 (21364A)")
+		UNAME_MACHINE="alphaev79" ;;
+	esac
+	# A Pn.n version is a patched version.
+	# A Vn.n version is a released version.
+	# A Tn.n version is a released field test version.
+	# A Xn.n version is an unreleased experimental baselevel.
+	# 1.2 uses "1.2" for uname -r.
+	echo ${UNAME_MACHINE}-dec-osf`echo ${UNAME_RELEASE} | sed -e 's/^[PVTX]//' | tr 'ABCDEFGHIJKLMNOPQRSTUVWXYZ' 'abcdefghijklmnopqrstuvwxyz'`
+	exit ;;
+    Alpha\ *:Windows_NT*:*)
+	# How do we know it's Interix rather than the generic POSIX subsystem?
+	# Should we change UNAME_MACHINE based on the output of uname instead
+	# of the specific Alpha model?
+	echo alpha-pc-interix
+	exit ;;
+    21064:Windows_NT:50:3)
+	echo alpha-dec-winnt3.5
+	exit ;;
+    Amiga*:UNIX_System_V:4.0:*)
+	echo m68k-unknown-sysv4
+	exit ;;
+    *:[Aa]miga[Oo][Ss]:*:*)
+	echo ${UNAME_MACHINE}-unknown-amigaos
+	exit ;;
+    *:[Mm]orph[Oo][Ss]:*:*)
+	echo ${UNAME_MACHINE}-unknown-morphos
+	exit ;;
+    *:OS/390:*:*)
+	echo i370-ibm-openedition
+	exit ;;
+    *:z/VM:*:*)
+	echo s390-ibm-zvmoe
+	exit ;;
+    *:OS400:*:*)
+        echo powerpc-ibm-os400
+	exit ;;
+    arm:RISC*:1.[012]*:*|arm:riscix:1.[012]*:*)
+	echo arm-acorn-riscix${UNAME_RELEASE}
+	exit ;;
+    arm:riscos:*:*|arm:RISCOS:*:*)
+	echo arm-unknown-riscos
+	exit ;;
+    SR2?01:HI-UX/MPP:*:* | SR8000:HI-UX/MPP:*:*)
+	echo hppa1.1-hitachi-hiuxmpp
+	exit ;;
+    Pyramid*:OSx*:*:* | MIS*:OSx*:*:* | MIS*:SMP_DC-OSx*:*:*)
+	# akee@wpdis03.wpafb.af.mil (Earle F. Ake) contributed MIS and NILE.
+	if test "`(/bin/universe) 2>/dev/null`" = att ; then
+		echo pyramid-pyramid-sysv3
+	else
+		echo pyramid-pyramid-bsd
+	fi
+	exit ;;
+    NILE*:*:*:dcosx)
+	echo pyramid-pyramid-svr4
+	exit ;;
+    DRS?6000:unix:4.0:6*)
+	echo sparc-icl-nx6
+	exit ;;
+    DRS?6000:UNIX_SV:4.2*:7* | DRS?6000:isis:4.2*:7*)
+	case `/usr/bin/uname -p` in
+	    sparc) echo sparc-icl-nx7; exit ;;
+	esac ;;
+    sun4H:SunOS:5.*:*)
+	echo sparc-hal-solaris2`echo ${UNAME_RELEASE}|sed -e 's/[^.]*//'`
+	exit ;;
+    sun4*:SunOS:5.*:* | tadpole*:SunOS:5.*:*)
+	echo sparc-sun-solaris2`echo ${UNAME_RELEASE}|sed -e 's/[^.]*//'`
+	exit ;;
+    i86pc:SunOS:5.*:* | i86xen:SunOS:5.*:*)
+	echo i386-pc-solaris2`echo ${UNAME_RELEASE}|sed -e 's/[^.]*//'`
+	exit ;;
+    sun4*:SunOS:6*:*)
+	# According to config.sub, this is the proper way to canonicalize
+	# SunOS6.  Hard to guess exactly what SunOS6 will be like, but
+	# it's likely to be more like Solaris than SunOS4.
+	echo sparc-sun-solaris3`echo ${UNAME_RELEASE}|sed -e 's/[^.]*//'`
+	exit ;;
+    sun4*:SunOS:*:*)
+	case "`/usr/bin/arch -k`" in
+	    Series*|S4*)
+		UNAME_RELEASE=`uname -v`
+		;;
+	esac
+	# Japanese Language versions have a version number like `4.1.3-JL'.
+	echo sparc-sun-sunos`echo ${UNAME_RELEASE}|sed -e 's/-/_/'`
+	exit ;;
+    sun3*:SunOS:*:*)
+	echo m68k-sun-sunos${UNAME_RELEASE}
+	exit ;;
+    sun*:*:4.2BSD:*)
+	UNAME_RELEASE=`(sed 1q /etc/motd | awk '{print substr($5,1,3)}') 2>/dev/null`
+	test "x${UNAME_RELEASE}" = "x" && UNAME_RELEASE=3
+	case "`/bin/arch`" in
+	    sun3)
+		echo m68k-sun-sunos${UNAME_RELEASE}
+		;;
+	    sun4)
+		echo sparc-sun-sunos${UNAME_RELEASE}
+		;;
+	esac
+	exit ;;
+    aushp:SunOS:*:*)
+	echo sparc-auspex-sunos${UNAME_RELEASE}
+	exit ;;
+    # The situation for MiNT is a little confusing.  The machine name
+    # can be virtually everything (everything which is not
+    # "atarist" or "atariste" at least should have a processor
+    # > m68000).  The system name ranges from "MiNT" over "FreeMiNT"
+    # to the lowercase version "mint" (or "freemint").  Finally
+    # the system name "TOS" denotes a system which is actually not
+    # MiNT.  But MiNT is downward compatible to TOS, so this should
+    # be no problem.
+    atarist[e]:*MiNT:*:* | atarist[e]:*mint:*:* | atarist[e]:*TOS:*:*)
+        echo m68k-atari-mint${UNAME_RELEASE}
+	exit ;;
+    atari*:*MiNT:*:* | atari*:*mint:*:* | atarist[e]:*TOS:*:*)
+	echo m68k-atari-mint${UNAME_RELEASE}
+        exit ;;
+    *falcon*:*MiNT:*:* | *falcon*:*mint:*:* | *falcon*:*TOS:*:*)
+        echo m68k-atari-mint${UNAME_RELEASE}
+	exit ;;
+    milan*:*MiNT:*:* | milan*:*mint:*:* | *milan*:*TOS:*:*)
+        echo m68k-milan-mint${UNAME_RELEASE}
+        exit ;;
+    hades*:*MiNT:*:* | hades*:*mint:*:* | *hades*:*TOS:*:*)
+        echo m68k-hades-mint${UNAME_RELEASE}
+        exit ;;
+    *:*MiNT:*:* | *:*mint:*:* | *:*TOS:*:*)
+        echo m68k-unknown-mint${UNAME_RELEASE}
+        exit ;;
+    m68k:machten:*:*)
+	echo m68k-apple-machten${UNAME_RELEASE}
+	exit ;;
+    powerpc:machten:*:*)
+	echo powerpc-apple-machten${UNAME_RELEASE}
+	exit ;;
+    RISC*:Mach:*:*)
+	echo mips-dec-mach_bsd4.3
+	exit ;;
+    RISC*:ULTRIX:*:*)
+	echo mips-dec-ultrix${UNAME_RELEASE}
+	exit ;;
+    VAX*:ULTRIX*:*:*)
+	echo vax-dec-ultrix${UNAME_RELEASE}
+	exit ;;
+    2020:CLIX:*:* | 2430:CLIX:*:*)
+	echo clipper-intergraph-clix${UNAME_RELEASE}
+	exit ;;
+    mips:*:*:UMIPS | mips:*:*:RISCos)
+	eval $set_cc_for_build
+	sed 's/^	//' << EOF >$dummy.c
+#ifdef __cplusplus
+#include <stdio.h>  /* for printf() prototype */
+	int main (int argc, char *argv[]) {
+#else
+	int main (argc, argv) int argc; char *argv[]; {
+#endif
+	#if defined (host_mips) && defined (MIPSEB)
+	#if defined (SYSTYPE_SYSV)
+	  printf ("mips-mips-riscos%ssysv\n", argv[1]); exit (0);
+	#endif
+	#if defined (SYSTYPE_SVR4)
+	  printf ("mips-mips-riscos%ssvr4\n", argv[1]); exit (0);
+	#endif
+	#if defined (SYSTYPE_BSD43) || defined(SYSTYPE_BSD)
+	  printf ("mips-mips-riscos%sbsd\n", argv[1]); exit (0);
+	#endif
+	#endif
+	  exit (-1);
+	}
+EOF
+	$CC_FOR_BUILD -o $dummy $dummy.c &&
+	  dummyarg=`echo "${UNAME_RELEASE}" | sed -n 's/\([0-9]*\).*/\1/p'` &&
+	  SYSTEM_NAME=`$dummy $dummyarg` &&
+	    { echo "$SYSTEM_NAME"; exit; }
+	echo mips-mips-riscos${UNAME_RELEASE}
+	exit ;;
+    Motorola:PowerMAX_OS:*:*)
+	echo powerpc-motorola-powermax
+	exit ;;
+    Motorola:*:4.3:PL8-*)
+	echo powerpc-harris-powermax
+	exit ;;
+    Night_Hawk:*:*:PowerMAX_OS | Synergy:PowerMAX_OS:*:*)
+	echo powerpc-harris-powermax
+	exit ;;
+    Night_Hawk:Power_UNIX:*:*)
+	echo powerpc-harris-powerunix
+	exit ;;
+    m88k:CX/UX:7*:*)
+	echo m88k-harris-cxux7
+	exit ;;
+    m88k:*:4*:R4*)
+	echo m88k-motorola-sysv4
+	exit ;;
+    m88k:*:3*:R3*)
+	echo m88k-motorola-sysv3
+	exit ;;
+    AViiON:dgux:*:*)
+        # DG/UX returns AViiON for all architectures
+        UNAME_PROCESSOR=`/usr/bin/uname -p`
+	if [ $UNAME_PROCESSOR = mc88100 ] || [ $UNAME_PROCESSOR = mc88110 ]
+	then
+	    if [ ${TARGET_BINARY_INTERFACE}x = m88kdguxelfx ] || \
+	       [ ${TARGET_BINARY_INTERFACE}x = x ]
+	    then
+		echo m88k-dg-dgux${UNAME_RELEASE}
+	    else
+		echo m88k-dg-dguxbcs${UNAME_RELEASE}
+	    fi
+	else
+	    echo i586-dg-dgux${UNAME_RELEASE}
+	fi
+ 	exit ;;
+    M88*:DolphinOS:*:*)	# DolphinOS (SVR3)
+	echo m88k-dolphin-sysv3
+	exit ;;
+    M88*:*:R3*:*)
+	# Delta 88k system running SVR3
+	echo m88k-motorola-sysv3
+	exit ;;
+    XD88*:*:*:*) # Tektronix XD88 system running UTekV (SVR3)
+	echo m88k-tektronix-sysv3
+	exit ;;
+    Tek43[0-9][0-9]:UTek:*:*) # Tektronix 4300 system running UTek (BSD)
+	echo m68k-tektronix-bsd
+	exit ;;
+    *:IRIX*:*:*)
+	echo mips-sgi-irix`echo ${UNAME_RELEASE}|sed -e 's/-/_/g'`
+	exit ;;
+    ????????:AIX?:[12].1:2)   # AIX 2.2.1 or AIX 2.1.1 is RT/PC AIX.
+	echo romp-ibm-aix     # uname -m gives an 8 hex-code CPU id
+	exit ;;               # Note that: echo "'`uname -s`'" gives 'AIX '
+    i*86:AIX:*:*)
+	echo i386-ibm-aix
+	exit ;;
+    ia64:AIX:*:*)
+	if [ -x /usr/bin/oslevel ] ; then
+		IBM_REV=`/usr/bin/oslevel`
+	else
+		IBM_REV=${UNAME_VERSION}.${UNAME_RELEASE}
+	fi
+	echo ${UNAME_MACHINE}-ibm-aix${IBM_REV}
+	exit ;;
+    *:AIX:2:3)
+	if grep bos325 /usr/include/stdio.h >/dev/null 2>&1; then
+		eval $set_cc_for_build
+		sed 's/^		//' << EOF >$dummy.c
+		#include <sys/systemcfg.h>
+
+		main()
+			{
+			if (!__power_pc())
+				exit(1);
+			puts("powerpc-ibm-aix3.2.5");
+			exit(0);
+			}
+EOF
+		if $CC_FOR_BUILD -o $dummy $dummy.c && SYSTEM_NAME=`$dummy`
+		then
+			echo "$SYSTEM_NAME"
+		else
+			echo rs6000-ibm-aix3.2.5
+		fi
+	elif grep bos324 /usr/include/stdio.h >/dev/null 2>&1; then
+		echo rs6000-ibm-aix3.2.4
+	else
+		echo rs6000-ibm-aix3.2
+	fi
+	exit ;;
+    *:AIX:*:[456])
+	IBM_CPU_ID=`/usr/sbin/lsdev -C -c processor -S available | sed 1q | awk '{ print $1 }'`
+	if /usr/sbin/lsattr -El ${IBM_CPU_ID} | grep ' POWER' >/dev/null 2>&1; then
+		IBM_ARCH=rs6000
+	else
+		IBM_ARCH=powerpc
+	fi
+	if [ -x /usr/bin/oslevel ] ; then
+		IBM_REV=`/usr/bin/oslevel`
+	else
+		IBM_REV=${UNAME_VERSION}.${UNAME_RELEASE}
+	fi
+	echo ${IBM_ARCH}-ibm-aix${IBM_REV}
+	exit ;;
+    *:AIX:*:*)
+	echo rs6000-ibm-aix
+	exit ;;
+    ibmrt:4.4BSD:*|romp-ibm:BSD:*)
+	echo romp-ibm-bsd4.4
+	exit ;;
+    ibmrt:*BSD:*|romp-ibm:BSD:*)            # covers RT/PC BSD and
+	echo romp-ibm-bsd${UNAME_RELEASE}   # 4.3 with uname added to
+	exit ;;                             # report: romp-ibm BSD 4.3
+    *:BOSX:*:*)
+	echo rs6000-bull-bosx
+	exit ;;
+    DPX/2?00:B.O.S.:*:*)
+	echo m68k-bull-sysv3
+	exit ;;
+    9000/[34]??:4.3bsd:1.*:*)
+	echo m68k-hp-bsd
+	exit ;;
+    hp300:4.4BSD:*:* | 9000/[34]??:4.3bsd:2.*:*)
+	echo m68k-hp-bsd4.4
+	exit ;;
+    9000/[34678]??:HP-UX:*:*)
+	HPUX_REV=`echo ${UNAME_RELEASE}|sed -e 's/[^.]*.[0B]*//'`
+	case "${UNAME_MACHINE}" in
+	    9000/31? )            HP_ARCH=m68000 ;;
+	    9000/[34]?? )         HP_ARCH=m68k ;;
+	    9000/[678][0-9][0-9])
+		if [ -x /usr/bin/getconf ]; then
+		    sc_cpu_version=`/usr/bin/getconf SC_CPU_VERSION 2>/dev/null`
+                    sc_kernel_bits=`/usr/bin/getconf SC_KERNEL_BITS 2>/dev/null`
+                    case "${sc_cpu_version}" in
+                      523) HP_ARCH="hppa1.0" ;; # CPU_PA_RISC1_0
+                      528) HP_ARCH="hppa1.1" ;; # CPU_PA_RISC1_1
+                      532)                      # CPU_PA_RISC2_0
+                        case "${sc_kernel_bits}" in
+                          32) HP_ARCH="hppa2.0n" ;;
+                          64) HP_ARCH="hppa2.0w" ;;
+			  '') HP_ARCH="hppa2.0" ;;   # HP-UX 10.20
+                        esac ;;
+                    esac
+		fi
+		if [ "${HP_ARCH}" = "" ]; then
+		    eval $set_cc_for_build
+		    sed 's/^              //' << EOF >$dummy.c
+
+              #define _HPUX_SOURCE
+              #include <stdlib.h>
+              #include <unistd.h>
+
+              int main ()
+              {
+              #if defined(_SC_KERNEL_BITS)
+                  long bits = sysconf(_SC_KERNEL_BITS);
+              #endif
+                  long cpu  = sysconf (_SC_CPU_VERSION);
+
+                  switch (cpu)
+              	{
+              	case CPU_PA_RISC1_0: puts ("hppa1.0"); break;
+              	case CPU_PA_RISC1_1: puts ("hppa1.1"); break;
+              	case CPU_PA_RISC2_0:
+              #if defined(_SC_KERNEL_BITS)
+              	    switch (bits)
+              		{
+              		case 64: puts ("hppa2.0w"); break;
+              		case 32: puts ("hppa2.0n"); break;
+              		default: puts ("hppa2.0"); break;
+              		} break;
+              #else  /* !defined(_SC_KERNEL_BITS) */
+              	    puts ("hppa2.0"); break;
+              #endif
+              	default: puts ("hppa1.0"); break;
+              	}
+                  exit (0);
+              }
+EOF
+		    (CCOPTS= $CC_FOR_BUILD -o $dummy $dummy.c 2>/dev/null) && HP_ARCH=`$dummy`
+		    test -z "$HP_ARCH" && HP_ARCH=hppa
+		fi ;;
+	esac
+	if [ ${HP_ARCH} = "hppa2.0w" ]
+	then
+	    eval $set_cc_for_build
+
+	    # hppa2.0w-hp-hpux* has a 64-bit kernel and a compiler generating
+	    # 32-bit code.  hppa64-hp-hpux* has the same kernel and a compiler
+	    # generating 64-bit code.  GNU and HP use different nomenclature:
+	    #
+	    # $ CC_FOR_BUILD=cc ./config.guess
+	    # => hppa2.0w-hp-hpux11.23
+	    # $ CC_FOR_BUILD="cc +DA2.0w" ./config.guess
+	    # => hppa64-hp-hpux11.23
+
+	    if echo __LP64__ | (CCOPTS= $CC_FOR_BUILD -E - 2>/dev/null) |
+		grep __LP64__ >/dev/null
+	    then
+		HP_ARCH="hppa2.0w"
+	    else
+		HP_ARCH="hppa64"
+	    fi
+	fi
+	echo ${HP_ARCH}-hp-hpux${HPUX_REV}
+	exit ;;
+    ia64:HP-UX:*:*)
+	HPUX_REV=`echo ${UNAME_RELEASE}|sed -e 's/[^.]*.[0B]*//'`
+	echo ia64-hp-hpux${HPUX_REV}
+	exit ;;
+    3050*:HI-UX:*:*)
+	eval $set_cc_for_build
+	sed 's/^	//' << EOF >$dummy.c
+	#include <unistd.h>
+	int
+	main ()
+	{
+	  long cpu = sysconf (_SC_CPU_VERSION);
+	  /* The order matters, because CPU_IS_HP_MC68K erroneously returns
+	     true for CPU_PA_RISC1_0.  CPU_IS_PA_RISC returns correct
+	     results, however.  */
+	  if (CPU_IS_PA_RISC (cpu))
+	    {
+	      switch (cpu)
+		{
+		  case CPU_PA_RISC1_0: puts ("hppa1.0-hitachi-hiuxwe2"); break;
+		  case CPU_PA_RISC1_1: puts ("hppa1.1-hitachi-hiuxwe2"); break;
+		  case CPU_PA_RISC2_0: puts ("hppa2.0-hitachi-hiuxwe2"); break;
+		  default: puts ("hppa-hitachi-hiuxwe2"); break;
+		}
+	    }
+	  else if (CPU_IS_HP_MC68K (cpu))
+	    puts ("m68k-hitachi-hiuxwe2");
+	  else puts ("unknown-hitachi-hiuxwe2");
+	  exit (0);
+	}
+EOF
+	$CC_FOR_BUILD -o $dummy $dummy.c && SYSTEM_NAME=`$dummy` &&
+		{ echo "$SYSTEM_NAME"; exit; }
+	echo unknown-hitachi-hiuxwe2
+	exit ;;
+    9000/7??:4.3bsd:*:* | 9000/8?[79]:4.3bsd:*:* )
+	echo hppa1.1-hp-bsd
+	exit ;;
+    9000/8??:4.3bsd:*:*)
+	echo hppa1.0-hp-bsd
+	exit ;;
+    *9??*:MPE/iX:*:* | *3000*:MPE/iX:*:*)
+	echo hppa1.0-hp-mpeix
+	exit ;;
+    hp7??:OSF1:*:* | hp8?[79]:OSF1:*:* )
+	echo hppa1.1-hp-osf
+	exit ;;
+    hp8??:OSF1:*:*)
+	echo hppa1.0-hp-osf
+	exit ;;
+    i*86:OSF1:*:*)
+	if [ -x /usr/sbin/sysversion ] ; then
+	    echo ${UNAME_MACHINE}-unknown-osf1mk
+	else
+	    echo ${UNAME_MACHINE}-unknown-osf1
+	fi
+	exit ;;
+    parisc*:Lites*:*:*)
+	echo hppa1.1-hp-lites
+	exit ;;
+    C1*:ConvexOS:*:* | convex:ConvexOS:C1*:*)
+	echo c1-convex-bsd
+        exit ;;
+    C2*:ConvexOS:*:* | convex:ConvexOS:C2*:*)
+	if getsysinfo -f scalar_acc
+	then echo c32-convex-bsd
+	else echo c2-convex-bsd
+	fi
+        exit ;;
+    C34*:ConvexOS:*:* | convex:ConvexOS:C34*:*)
+	echo c34-convex-bsd
+        exit ;;
+    C38*:ConvexOS:*:* | convex:ConvexOS:C38*:*)
+	echo c38-convex-bsd
+        exit ;;
+    C4*:ConvexOS:*:* | convex:ConvexOS:C4*:*)
+	echo c4-convex-bsd
+        exit ;;
+    CRAY*Y-MP:*:*:*)
+	echo ymp-cray-unicos${UNAME_RELEASE} | sed -e 's/\.[^.]*$/.X/'
+	exit ;;
+    CRAY*[A-Z]90:*:*:*)
+	echo ${UNAME_MACHINE}-cray-unicos${UNAME_RELEASE} \
+	| sed -e 's/CRAY.*\([A-Z]90\)/\1/' \
+	      -e y/ABCDEFGHIJKLMNOPQRSTUVWXYZ/abcdefghijklmnopqrstuvwxyz/ \
+	      -e 's/\.[^.]*$/.X/'
+	exit ;;
+    CRAY*TS:*:*:*)
+	echo t90-cray-unicos${UNAME_RELEASE} | sed -e 's/\.[^.]*$/.X/'
+	exit ;;
+    CRAY*T3E:*:*:*)
+	echo alphaev5-cray-unicosmk${UNAME_RELEASE} | sed -e 's/\.[^.]*$/.X/'
+	exit ;;
+    CRAY*SV1:*:*:*)
+	echo sv1-cray-unicos${UNAME_RELEASE} | sed -e 's/\.[^.]*$/.X/'
+	exit ;;
+    *:UNICOS/mp:*:*)
+	echo craynv-cray-unicosmp${UNAME_RELEASE} | sed -e 's/\.[^.]*$/.X/'
+	exit ;;
+    F30[01]:UNIX_System_V:*:* | F700:UNIX_System_V:*:*)
+	FUJITSU_PROC=`uname -m | tr 'ABCDEFGHIJKLMNOPQRSTUVWXYZ' 'abcdefghijklmnopqrstuvwxyz'`
+        FUJITSU_SYS=`uname -p | tr 'ABCDEFGHIJKLMNOPQRSTUVWXYZ' 'abcdefghijklmnopqrstuvwxyz' | sed -e 's/\///'`
+        FUJITSU_REL=`echo ${UNAME_RELEASE} | sed -e 's/ /_/'`
+        echo "${FUJITSU_PROC}-fujitsu-${FUJITSU_SYS}${FUJITSU_REL}"
+        exit ;;
+    5000:UNIX_System_V:4.*:*)
+        FUJITSU_SYS=`uname -p | tr 'ABCDEFGHIJKLMNOPQRSTUVWXYZ' 'abcdefghijklmnopqrstuvwxyz' | sed -e 's/\///'`
+        FUJITSU_REL=`echo ${UNAME_RELEASE} | tr 'ABCDEFGHIJKLMNOPQRSTUVWXYZ' 'abcdefghijklmnopqrstuvwxyz' | sed -e 's/ /_/'`
+        echo "sparc-fujitsu-${FUJITSU_SYS}${FUJITSU_REL}"
+	exit ;;
+    i*86:BSD/386:*:* | i*86:BSD/OS:*:* | *:Ascend\ Embedded/OS:*:*)
+	echo ${UNAME_MACHINE}-pc-bsdi${UNAME_RELEASE}
+	exit ;;
+    sparc*:BSD/OS:*:*)
+	echo sparc-unknown-bsdi${UNAME_RELEASE}
+	exit ;;
+    *:BSD/OS:*:*)
+	echo ${UNAME_MACHINE}-unknown-bsdi${UNAME_RELEASE}
+	exit ;;
+    *:FreeBSD:*:*)
+	case ${UNAME_MACHINE} in
+	    pc98)
+		echo i386-unknown-freebsd`echo ${UNAME_RELEASE}|sed -e 's/[-(].*//'` ;;
+	    amd64)
+		echo x86_64-unknown-freebsd`echo ${UNAME_RELEASE}|sed -e 's/[-(].*//'` ;;
+	    *)
+		echo ${UNAME_MACHINE}-unknown-freebsd`echo ${UNAME_RELEASE}|sed -e 's/[-(].*//'` ;;
+	esac
+	exit ;;
+    i*:CYGWIN*:*)
+	echo ${UNAME_MACHINE}-pc-cygwin
+	exit ;;
+    *:MINGW*:*)
+	echo ${UNAME_MACHINE}-pc-mingw32
+	exit ;;
+    i*:windows32*:*)
+    	# uname -m includes "-pc" on this system.
+    	echo ${UNAME_MACHINE}-mingw32
+	exit ;;
+    i*:PW*:*)
+	echo ${UNAME_MACHINE}-pc-pw32
+	exit ;;
+    *:Interix*:[3456]*)
+    	case ${UNAME_MACHINE} in
+	    x86)
+		echo i586-pc-interix${UNAME_RELEASE}
+		exit ;;
+	    EM64T | authenticamd)
+		echo x86_64-unknown-interix${UNAME_RELEASE}
+		exit ;;
+	    IA64)
+		echo ia64-unknown-interix${UNAME_RELEASE}
+		exit ;;
+	esac ;;
+    [345]86:Windows_95:* | [345]86:Windows_98:* | [345]86:Windows_NT:*)
+	echo i${UNAME_MACHINE}-pc-mks
+	exit ;;
+    i*:Windows_NT*:* | Pentium*:Windows_NT*:*)
+	# How do we know it's Interix rather than the generic POSIX subsystem?
+	# It also conflicts with pre-2.0 versions of AT&T UWIN. Should we
+	# UNAME_MACHINE based on the output of uname instead of i386?
+	echo i586-pc-interix
+	exit ;;
+    i*:UWIN*:*)
+	echo ${UNAME_MACHINE}-pc-uwin
+	exit ;;
+    amd64:CYGWIN*:*:* | x86_64:CYGWIN*:*:*)
+	echo x86_64-unknown-cygwin
+	exit ;;
+    p*:CYGWIN*:*)
+	echo powerpcle-unknown-cygwin
+	exit ;;
+    prep*:SunOS:5.*:*)
+	echo powerpcle-unknown-solaris2`echo ${UNAME_RELEASE}|sed -e 's/[^.]*//'`
+	exit ;;
+    *:GNU:*:*)
+	# the GNU system
+	echo `echo ${UNAME_MACHINE}|sed -e 's,[-/].*$,,'`-unknown-gnu`echo ${UNAME_RELEASE}|sed -e 's,/.*$,,'`
+	exit ;;
+    *:GNU/*:*:*)
+	# other systems with GNU libc and userland
+	echo ${UNAME_MACHINE}-unknown-`echo ${UNAME_SYSTEM} | sed 's,^[^/]*/,,' | tr '[A-Z]' '[a-z]'``echo ${UNAME_RELEASE}|sed -e 's/[-(].*//'`-gnu
+	exit ;;
+    i*86:Minix:*:*)
+	echo ${UNAME_MACHINE}-pc-minix
+	exit ;;
+    arm*:Linux:*:*)
+	eval $set_cc_for_build
+	if echo __ARM_EABI__ | $CC_FOR_BUILD -E - 2>/dev/null \
+	    | grep -q __ARM_EABI__
+	then
+	    echo ${UNAME_MACHINE}-unknown-linux-gnu
+	else
+	    echo ${UNAME_MACHINE}-unknown-linux-gnueabi
+	fi
+	exit ;;
+    avr32*:Linux:*:*)
+	echo ${UNAME_MACHINE}-unknown-linux-gnu
+	exit ;;
+    cris:Linux:*:*)
+	echo cris-axis-linux-gnu
+	exit ;;
+    crisv32:Linux:*:*)
+	echo crisv32-axis-linux-gnu
+	exit ;;
+    frv:Linux:*:*)
+    	echo frv-unknown-linux-gnu
+	exit ;;
+    ia64:Linux:*:*)
+	echo ${UNAME_MACHINE}-unknown-linux-gnu
+	exit ;;
+    m32r*:Linux:*:*)
+	echo ${UNAME_MACHINE}-unknown-linux-gnu
+	exit ;;
+    m68*:Linux:*:*)
+	echo ${UNAME_MACHINE}-unknown-linux-gnu
+	exit ;;
+    mips:Linux:*:*)
+	eval $set_cc_for_build
+	sed 's/^	//' << EOF >$dummy.c
+	#undef CPU
+	#undef mips
+	#undef mipsel
+	#if defined(__MIPSEL__) || defined(__MIPSEL) || defined(_MIPSEL) || defined(MIPSEL)
+	CPU=mipsel
+	#else
+	#if defined(__MIPSEB__) || defined(__MIPSEB) || defined(_MIPSEB) || defined(MIPSEB)
+	CPU=mips
+	#else
+	CPU=
+	#endif
+	#endif
+EOF
+	eval "`$CC_FOR_BUILD -E $dummy.c 2>/dev/null | sed -n '
+	    /^CPU/{
+		s: ::g
+		p
+	    }'`"
+	test x"${CPU}" != x && { echo "${CPU}-unknown-linux-gnu"; exit; }
+	;;
+    mips64:Linux:*:*)
+	eval $set_cc_for_build
+	sed 's/^	//' << EOF >$dummy.c
+	#undef CPU
+	#undef mips64
+	#undef mips64el
+	#if defined(__MIPSEL__) || defined(__MIPSEL) || defined(_MIPSEL) || defined(MIPSEL)
+	CPU=mips64el
+	#else
+	#if defined(__MIPSEB__) || defined(__MIPSEB) || defined(_MIPSEB) || defined(MIPSEB)
+	CPU=mips64
+	#else
+	CPU=
+	#endif
+	#endif
+EOF
+	eval "`$CC_FOR_BUILD -E $dummy.c 2>/dev/null | sed -n '
+	    /^CPU/{
+		s: ::g
+		p
+	    }'`"
+	test x"${CPU}" != x && { echo "${CPU}-unknown-linux-gnu"; exit; }
+	;;
+    or32:Linux:*:*)
+	echo or32-unknown-linux-gnu
+	exit ;;
+    ppc:Linux:*:*)
+	echo powerpc-unknown-linux-gnu
+	exit ;;
+    ppc64:Linux:*:*)
+	echo powerpc64-unknown-linux-gnu
+	exit ;;
+    alpha:Linux:*:*)
+	case `sed -n '/^cpu model/s/^.*: \(.*\)/\1/p' < /proc/cpuinfo` in
+	  EV5)   UNAME_MACHINE=alphaev5 ;;
+	  EV56)  UNAME_MACHINE=alphaev56 ;;
+	  PCA56) UNAME_MACHINE=alphapca56 ;;
+	  PCA57) UNAME_MACHINE=alphapca56 ;;
+	  EV6)   UNAME_MACHINE=alphaev6 ;;
+	  EV67)  UNAME_MACHINE=alphaev67 ;;
+	  EV68*) UNAME_MACHINE=alphaev68 ;;
+        esac
+	objdump --private-headers /bin/sh | grep ld.so.1 >/dev/null
+	if test "$?" = 0 ; then LIBC="libc1" ; else LIBC="" ; fi
+	echo ${UNAME_MACHINE}-unknown-linux-gnu${LIBC}
+	exit ;;
+    parisc:Linux:*:* | hppa:Linux:*:*)
+	# Look for CPU level
+	case `grep '^cpu[^a-z]*:' /proc/cpuinfo 2>/dev/null | cut -d' ' -f2` in
+	  PA7*) echo hppa1.1-unknown-linux-gnu ;;
+	  PA8*) echo hppa2.0-unknown-linux-gnu ;;
+	  *)    echo hppa-unknown-linux-gnu ;;
+	esac
+	exit ;;
+    parisc64:Linux:*:* | hppa64:Linux:*:*)
+	echo hppa64-unknown-linux-gnu
+	exit ;;
+    s390:Linux:*:* | s390x:Linux:*:*)
+	echo ${UNAME_MACHINE}-ibm-linux
+	exit ;;
+    sh64*:Linux:*:*)
+    	echo ${UNAME_MACHINE}-unknown-linux-gnu
+	exit ;;
+    sh*:Linux:*:*)
+	echo ${UNAME_MACHINE}-unknown-linux-gnu
+	exit ;;
+    sparc:Linux:*:* | sparc64:Linux:*:*)
+	echo ${UNAME_MACHINE}-unknown-linux-gnu
+	exit ;;
+    vax:Linux:*:*)
+	echo ${UNAME_MACHINE}-dec-linux-gnu
+	exit ;;
+    x86_64:Linux:*:*)
+	echo x86_64-unknown-linux-gnu
+	exit ;;
+    xtensa*:Linux:*:*)
+    	echo ${UNAME_MACHINE}-unknown-linux-gnu
+	exit ;;
+    i*86:Linux:*:*)
+	# The BFD linker knows what the default object file format is, so
+	# first see if it will tell us. cd to the root directory to prevent
+	# problems with other programs or directories called `ld' in the path.
+	# Set LC_ALL=C to ensure ld outputs messages in English.
+	ld_supported_targets=`cd /; LC_ALL=C ld --help 2>&1 \
+			 | sed -ne '/supported targets:/!d
+				    s/[ 	][ 	]*/ /g
+				    s/.*supported targets: *//
+				    s/ .*//
+				    p'`
+        case "$ld_supported_targets" in
+	  elf32-i386)
+		TENTATIVE="${UNAME_MACHINE}-pc-linux-gnu"
+		;;
+	  a.out-i386-linux)
+		echo "${UNAME_MACHINE}-pc-linux-gnuaout"
+		exit ;;
+	  coff-i386)
+		echo "${UNAME_MACHINE}-pc-linux-gnucoff"
+		exit ;;
+	  "")
+		# Either a pre-BFD a.out linker (linux-gnuoldld) or
+		# one that does not give us useful --help.
+		echo "${UNAME_MACHINE}-pc-linux-gnuoldld"
+		exit ;;
+	esac
+	# Determine whether the default compiler is a.out or elf
+	eval $set_cc_for_build
+	sed 's/^	//' << EOF >$dummy.c
+	#include <features.h>
+	#ifdef __ELF__
+	# ifdef __GLIBC__
+	#  if __GLIBC__ >= 2
+	LIBC=gnu
+	#  else
+	LIBC=gnulibc1
+	#  endif
+	# else
+	LIBC=gnulibc1
+	# endif
+	#else
+	#if defined(__INTEL_COMPILER) || defined(__PGI) || defined(__SUNPRO_C) || defined(__SUNPRO_CC)
+	LIBC=gnu
+	#else
+	LIBC=gnuaout
+	#endif
+	#endif
+	#ifdef __dietlibc__
+	LIBC=dietlibc
+	#endif
+EOF
+	eval "`$CC_FOR_BUILD -E $dummy.c 2>/dev/null | sed -n '
+	    /^LIBC/{
+		s: ::g
+		p
+	    }'`"
+	test x"${LIBC}" != x && {
+		echo "${UNAME_MACHINE}-pc-linux-${LIBC}"
+		exit
+	}
+	test x"${TENTATIVE}" != x && { echo "${TENTATIVE}"; exit; }
+	;;
+    i*86:DYNIX/ptx:4*:*)
+	# ptx 4.0 does uname -s correctly, with DYNIX/ptx in there.
+	# earlier versions are messed up and put the nodename in both
+	# sysname and nodename.
+	echo i386-sequent-sysv4
+	exit ;;
+    i*86:UNIX_SV:4.2MP:2.*)
+        # Unixware is an offshoot of SVR4, but it has its own version
+        # number series starting with 2...
+        # I am not positive that other SVR4 systems won't match this,
+	# I just have to hope.  -- rms.
+        # Use sysv4.2uw... so that sysv4* matches it.
+	echo ${UNAME_MACHINE}-pc-sysv4.2uw${UNAME_VERSION}
+	exit ;;
+    i*86:OS/2:*:*)
+	# If we were able to find `uname', then EMX Unix compatibility
+	# is probably installed.
+	echo ${UNAME_MACHINE}-pc-os2-emx
+	exit ;;
+    i*86:XTS-300:*:STOP)
+	echo ${UNAME_MACHINE}-unknown-stop
+	exit ;;
+    i*86:atheos:*:*)
+	echo ${UNAME_MACHINE}-unknown-atheos
+	exit ;;
+    i*86:syllable:*:*)
+	echo ${UNAME_MACHINE}-pc-syllable
+	exit ;;
+    i*86:LynxOS:2.*:* | i*86:LynxOS:3.[01]*:* | i*86:LynxOS:4.0*:*)
+	echo i386-unknown-lynxos${UNAME_RELEASE}
+	exit ;;
+    i*86:*DOS:*:*)
+	echo ${UNAME_MACHINE}-pc-msdosdjgpp
+	exit ;;
+    i*86:*:4.*:* | i*86:SYSTEM_V:4.*:*)
+	UNAME_REL=`echo ${UNAME_RELEASE} | sed 's/\/MP$//'`
+	if grep Novell /usr/include/link.h >/dev/null 2>/dev/null; then
+		echo ${UNAME_MACHINE}-univel-sysv${UNAME_REL}
+	else
+		echo ${UNAME_MACHINE}-pc-sysv${UNAME_REL}
+	fi
+	exit ;;
+    i*86:*:5:[678]*)
+    	# UnixWare 7.x, OpenUNIX and OpenServer 6.
+	case `/bin/uname -X | grep "^Machine"` in
+	    *486*)	     UNAME_MACHINE=i486 ;;
+	    *Pentium)	     UNAME_MACHINE=i586 ;;
+	    *Pent*|*Celeron) UNAME_MACHINE=i686 ;;
+	esac
+	echo ${UNAME_MACHINE}-unknown-sysv${UNAME_RELEASE}${UNAME_SYSTEM}${UNAME_VERSION}
+	exit ;;
+    i*86:*:3.2:*)
+	if test -f /usr/options/cb.name; then
+		UNAME_REL=`sed -n 's/.*Version //p' </usr/options/cb.name`
+		echo ${UNAME_MACHINE}-pc-isc$UNAME_REL
+	elif /bin/uname -X 2>/dev/null >/dev/null ; then
+		UNAME_REL=`(/bin/uname -X|grep Release|sed -e 's/.*= //')`
+		(/bin/uname -X|grep i80486 >/dev/null) && UNAME_MACHINE=i486
+		(/bin/uname -X|grep '^Machine.*Pentium' >/dev/null) \
+			&& UNAME_MACHINE=i586
+		(/bin/uname -X|grep '^Machine.*Pent *II' >/dev/null) \
+			&& UNAME_MACHINE=i686
+		(/bin/uname -X|grep '^Machine.*Pentium Pro' >/dev/null) \
+			&& UNAME_MACHINE=i686
+		echo ${UNAME_MACHINE}-pc-sco$UNAME_REL
+	else
+		echo ${UNAME_MACHINE}-pc-sysv32
+	fi
+	exit ;;
+    pc:*:*:*)
+	# Left here for compatibility:
+        # uname -m prints for DJGPP always 'pc', but it prints nothing about
+        # the processor, so we play safe by assuming i386.
+	echo i386-pc-msdosdjgpp
+        exit ;;
+    Intel:Mach:3*:*)
+	echo i386-pc-mach3
+	exit ;;
+    paragon:*:*:*)
+	echo i860-intel-osf1
+	exit ;;
+    i860:*:4.*:*) # i860-SVR4
+	if grep Stardent /usr/include/sys/uadmin.h >/dev/null 2>&1 ; then
+	  echo i860-stardent-sysv${UNAME_RELEASE} # Stardent Vistra i860-SVR4
+	else # Add other i860-SVR4 vendors below as they are discovered.
+	  echo i860-unknown-sysv${UNAME_RELEASE}  # Unknown i860-SVR4
+	fi
+	exit ;;
+    mini*:CTIX:SYS*5:*)
+	# "miniframe"
+	echo m68010-convergent-sysv
+	exit ;;
+    mc68k:UNIX:SYSTEM5:3.51m)
+	echo m68k-convergent-sysv
+	exit ;;
+    M680?0:D-NIX:5.3:*)
+	echo m68k-diab-dnix
+	exit ;;
+    M68*:*:R3V[5678]*:*)
+	test -r /sysV68 && { echo 'm68k-motorola-sysv'; exit; } ;;
+    3[345]??:*:4.0:3.0 | 3[34]??A:*:4.0:3.0 | 3[34]??,*:*:4.0:3.0 | 3[34]??/*:*:4.0:3.0 | 4400:*:4.0:3.0 | 4850:*:4.0:3.0 | SKA40:*:4.0:3.0 | SDS2:*:4.0:3.0 | SHG2:*:4.0:3.0 | S7501*:*:4.0:3.0)
+	OS_REL=''
+	test -r /etc/.relid \
+	&& OS_REL=.`sed -n 's/[^ ]* [^ ]* \([0-9][0-9]\).*/\1/p' < /etc/.relid`
+	/bin/uname -p 2>/dev/null | grep 86 >/dev/null \
+	  && { echo i486-ncr-sysv4.3${OS_REL}; exit; }
+	/bin/uname -p 2>/dev/null | /bin/grep entium >/dev/null \
+	  && { echo i586-ncr-sysv4.3${OS_REL}; exit; } ;;
+    3[34]??:*:4.0:* | 3[34]??,*:*:4.0:*)
+        /bin/uname -p 2>/dev/null | grep 86 >/dev/null \
+          && { echo i486-ncr-sysv4; exit; } ;;
+    m68*:LynxOS:2.*:* | m68*:LynxOS:3.0*:*)
+	echo m68k-unknown-lynxos${UNAME_RELEASE}
+	exit ;;
+    mc68030:UNIX_System_V:4.*:*)
+	echo m68k-atari-sysv4
+	exit ;;
+    TSUNAMI:LynxOS:2.*:*)
+	echo sparc-unknown-lynxos${UNAME_RELEASE}
+	exit ;;
+    rs6000:LynxOS:2.*:*)
+	echo rs6000-unknown-lynxos${UNAME_RELEASE}
+	exit ;;
+    PowerPC:LynxOS:2.*:* | PowerPC:LynxOS:3.[01]*:* | PowerPC:LynxOS:4.0*:*)
+	echo powerpc-unknown-lynxos${UNAME_RELEASE}
+	exit ;;
+    SM[BE]S:UNIX_SV:*:*)
+	echo mips-dde-sysv${UNAME_RELEASE}
+	exit ;;
+    RM*:ReliantUNIX-*:*:*)
+	echo mips-sni-sysv4
+	exit ;;
+    RM*:SINIX-*:*:*)
+	echo mips-sni-sysv4
+	exit ;;
+    *:SINIX-*:*:*)
+	if uname -p 2>/dev/null >/dev/null ; then
+		UNAME_MACHINE=`(uname -p) 2>/dev/null`
+		echo ${UNAME_MACHINE}-sni-sysv4
+	else
+		echo ns32k-sni-sysv
+	fi
+	exit ;;
+    PENTIUM:*:4.0*:*) # Unisys `ClearPath HMP IX 4000' SVR4/MP effort
+                      # says <Richard.M.Bartel@ccMail.Census.GOV>
+        echo i586-unisys-sysv4
+        exit ;;
+    *:UNIX_System_V:4*:FTX*)
+	# From Gerald Hewes <hewes@openmarket.com>.
+	# How about differentiating between stratus architectures? -djm
+	echo hppa1.1-stratus-sysv4
+	exit ;;
+    *:*:*:FTX*)
+	# From seanf@swdc.stratus.com.
+	echo i860-stratus-sysv4
+	exit ;;
+    i*86:VOS:*:*)
+	# From Paul.Green@stratus.com.
+	echo ${UNAME_MACHINE}-stratus-vos
+	exit ;;
+    *:VOS:*:*)
+	# From Paul.Green@stratus.com.
+	echo hppa1.1-stratus-vos
+	exit ;;
+    mc68*:A/UX:*:*)
+	echo m68k-apple-aux${UNAME_RELEASE}
+	exit ;;
+    news*:NEWS-OS:6*:*)
+	echo mips-sony-newsos6
+	exit ;;
+    R[34]000:*System_V*:*:* | R4000:UNIX_SYSV:*:* | R*000:UNIX_SV:*:*)
+	if [ -d /usr/nec ]; then
+	        echo mips-nec-sysv${UNAME_RELEASE}
+	else
+	        echo mips-unknown-sysv${UNAME_RELEASE}
+	fi
+        exit ;;
+    BeBox:BeOS:*:*)	# BeOS running on hardware made by Be, PPC only.
+	echo powerpc-be-beos
+	exit ;;
+    BeMac:BeOS:*:*)	# BeOS running on Mac or Mac clone, PPC only.
+	echo powerpc-apple-beos
+	exit ;;
+    BePC:BeOS:*:*)	# BeOS running on Intel PC compatible.
+	echo i586-pc-beos
+	exit ;;
+    SX-4:SUPER-UX:*:*)
+	echo sx4-nec-superux${UNAME_RELEASE}
+	exit ;;
+    SX-5:SUPER-UX:*:*)
+	echo sx5-nec-superux${UNAME_RELEASE}
+	exit ;;
+    SX-6:SUPER-UX:*:*)
+	echo sx6-nec-superux${UNAME_RELEASE}
+	exit ;;
+    SX-7:SUPER-UX:*:*)
+	echo sx7-nec-superux${UNAME_RELEASE}
+	exit ;;
+    SX-8:SUPER-UX:*:*)
+	echo sx8-nec-superux${UNAME_RELEASE}
+	exit ;;
+    SX-8R:SUPER-UX:*:*)
+	echo sx8r-nec-superux${UNAME_RELEASE}
+	exit ;;
+    Power*:Rhapsody:*:*)
+	echo powerpc-apple-rhapsody${UNAME_RELEASE}
+	exit ;;
+    *:Rhapsody:*:*)
+	echo ${UNAME_MACHINE}-apple-rhapsody${UNAME_RELEASE}
+	exit ;;
+    *:Darwin:*:*)
+	UNAME_PROCESSOR=`uname -p` || UNAME_PROCESSOR=unknown
+	case $UNAME_PROCESSOR in
+	    unknown) UNAME_PROCESSOR=powerpc ;;
+	esac
+	echo ${UNAME_PROCESSOR}-apple-darwin${UNAME_RELEASE}
+	exit ;;
+    *:procnto*:*:* | *:QNX:[0123456789]*:*)
+	UNAME_PROCESSOR=`uname -p`
+	if test "$UNAME_PROCESSOR" = "x86"; then
+		UNAME_PROCESSOR=i386
+		UNAME_MACHINE=pc
+	fi
+	echo ${UNAME_PROCESSOR}-${UNAME_MACHINE}-nto-qnx${UNAME_RELEASE}
+	exit ;;
+    *:QNX:*:4*)
+	echo i386-pc-qnx
+	exit ;;
+    NSE-?:NONSTOP_KERNEL:*:*)
+	echo nse-tandem-nsk${UNAME_RELEASE}
+	exit ;;
+    NSR-?:NONSTOP_KERNEL:*:*)
+	echo nsr-tandem-nsk${UNAME_RELEASE}
+	exit ;;
+    *:NonStop-UX:*:*)
+	echo mips-compaq-nonstopux
+	exit ;;
+    BS2000:POSIX*:*:*)
+	echo bs2000-siemens-sysv
+	exit ;;
+    DS/*:UNIX_System_V:*:*)
+	echo ${UNAME_MACHINE}-${UNAME_SYSTEM}-${UNAME_RELEASE}
+	exit ;;
+    *:Plan9:*:*)
+	# "uname -m" is not consistent, so use $cputype instead. 386
+	# is converted to i386 for consistency with other x86
+	# operating systems.
+	if test "$cputype" = "386"; then
+	    UNAME_MACHINE=i386
+	else
+	    UNAME_MACHINE="$cputype"
+	fi
+	echo ${UNAME_MACHINE}-unknown-plan9
+	exit ;;
+    *:TOPS-10:*:*)
+	echo pdp10-unknown-tops10
+	exit ;;
+    *:TENEX:*:*)
+	echo pdp10-unknown-tenex
+	exit ;;
+    KS10:TOPS-20:*:* | KL10:TOPS-20:*:* | TYPE4:TOPS-20:*:*)
+	echo pdp10-dec-tops20
+	exit ;;
+    XKL-1:TOPS-20:*:* | TYPE5:TOPS-20:*:*)
+	echo pdp10-xkl-tops20
+	exit ;;
+    *:TOPS-20:*:*)
+	echo pdp10-unknown-tops20
+	exit ;;
+    *:ITS:*:*)
+	echo pdp10-unknown-its
+	exit ;;
+    SEI:*:*:SEIUX)
+        echo mips-sei-seiux${UNAME_RELEASE}
+	exit ;;
+    *:DragonFly:*:*)
+	echo ${UNAME_MACHINE}-unknown-dragonfly`echo ${UNAME_RELEASE}|sed -e 's/[-(].*//'`
+	exit ;;
+    *:*VMS:*:*)
+    	UNAME_MACHINE=`(uname -p) 2>/dev/null`
+	case "${UNAME_MACHINE}" in
+	    A*) echo alpha-dec-vms ; exit ;;
+	    I*) echo ia64-dec-vms ; exit ;;
+	    V*) echo vax-dec-vms ; exit ;;
+	esac ;;
+    *:XENIX:*:SysV)
+	echo i386-pc-xenix
+	exit ;;
+    i*86:skyos:*:*)
+	echo ${UNAME_MACHINE}-pc-skyos`echo ${UNAME_RELEASE}` | sed -e 's/ .*$//'
+	exit ;;
+    i*86:rdos:*:*)
+	echo ${UNAME_MACHINE}-pc-rdos
+	exit ;;
+esac
+
+#echo '(No uname command or uname output not recognized.)' 1>&2
+#echo "${UNAME_MACHINE}:${UNAME_SYSTEM}:${UNAME_RELEASE}:${UNAME_VERSION}" 1>&2
+
+eval $set_cc_for_build
+cat >$dummy.c <<EOF
+#ifdef _SEQUENT_
+# include <sys/types.h>
+# include <sys/utsname.h>
+#endif
+main ()
+{
+#if defined (sony)
+#if defined (MIPSEB)
+  /* BFD wants "bsd" instead of "newsos".  Perhaps BFD should be changed,
+     I don't know....  */
+  printf ("mips-sony-bsd\n"); exit (0);
+#else
+#include <sys/param.h>
+  printf ("m68k-sony-newsos%s\n",
+#ifdef NEWSOS4
+          "4"
+#else
+	  ""
+#endif
+         ); exit (0);
+#endif
+#endif
+
+#if defined (__arm) && defined (__acorn) && defined (__unix)
+  printf ("arm-acorn-riscix\n"); exit (0);
+#endif
+
+#if defined (hp300) && !defined (hpux)
+  printf ("m68k-hp-bsd\n"); exit (0);
+#endif
+
+#if defined (NeXT)
+#if !defined (__ARCHITECTURE__)
+#define __ARCHITECTURE__ "m68k"
+#endif
+  int version;
+  version=`(hostinfo | sed -n 's/.*NeXT Mach \([0-9]*\).*/\1/p') 2>/dev/null`;
+  if (version < 4)
+    printf ("%s-next-nextstep%d\n", __ARCHITECTURE__, version);
+  else
+    printf ("%s-next-openstep%d\n", __ARCHITECTURE__, version);
+  exit (0);
+#endif
+
+#if defined (MULTIMAX) || defined (n16)
+#if defined (UMAXV)
+  printf ("ns32k-encore-sysv\n"); exit (0);
+#else
+#if defined (CMU)
+  printf ("ns32k-encore-mach\n"); exit (0);
+#else
+  printf ("ns32k-encore-bsd\n"); exit (0);
+#endif
+#endif
+#endif
+
+#if defined (__386BSD__)
+  printf ("i386-pc-bsd\n"); exit (0);
+#endif
+
+#if defined (sequent)
+#if defined (i386)
+  printf ("i386-sequent-dynix\n"); exit (0);
+#endif
+#if defined (ns32000)
+  printf ("ns32k-sequent-dynix\n"); exit (0);
+#endif
+#endif
+
+#if defined (_SEQUENT_)
+    struct utsname un;
+
+    uname(&un);
+
+    if (strncmp(un.version, "V2", 2) == 0) {
+	printf ("i386-sequent-ptx2\n"); exit (0);
+    }
+    if (strncmp(un.version, "V1", 2) == 0) { /* XXX is V1 correct? */
+	printf ("i386-sequent-ptx1\n"); exit (0);
+    }
+    printf ("i386-sequent-ptx\n"); exit (0);
+
+#endif
+
+#if defined (vax)
+# if !defined (ultrix)
+#  include <sys/param.h>
+#  if defined (BSD)
+#   if BSD == 43
+      printf ("vax-dec-bsd4.3\n"); exit (0);
+#   else
+#    if BSD == 199006
+      printf ("vax-dec-bsd4.3reno\n"); exit (0);
+#    else
+      printf ("vax-dec-bsd\n"); exit (0);
+#    endif
+#   endif
+#  else
+    printf ("vax-dec-bsd\n"); exit (0);
+#  endif
+# else
+    printf ("vax-dec-ultrix\n"); exit (0);
+# endif
+#endif
+
+#if defined (alliant) && defined (i860)
+  printf ("i860-alliant-bsd\n"); exit (0);
+#endif
+
+  exit (1);
+}
+EOF
+
+$CC_FOR_BUILD -o $dummy $dummy.c 2>/dev/null && SYSTEM_NAME=`$dummy` &&
+	{ echo "$SYSTEM_NAME"; exit; }
+
+# Apollos put the system type in the environment.
+
+test -d /usr/apollo && { echo ${ISP}-apollo-${SYSTYPE}; exit; }
+
+# Convex versions that predate uname can use getsysinfo(1)
+
+if [ -x /usr/convex/getsysinfo ]
+then
+    case `getsysinfo -f cpu_type` in
+    c1*)
+	echo c1-convex-bsd
+	exit ;;
+    c2*)
+	if getsysinfo -f scalar_acc
+	then echo c32-convex-bsd
+	else echo c2-convex-bsd
+	fi
+	exit ;;
+    c34*)
+	echo c34-convex-bsd
+	exit ;;
+    c38*)
+	echo c38-convex-bsd
+	exit ;;
+    c4*)
+	echo c4-convex-bsd
+	exit ;;
+    esac
+fi
+
+cat >&2 <<EOF
+$0: unable to guess system type
+
+This script, last modified $timestamp, has failed to recognize
+the operating system you are using. It is advised that you
+download the most up to date version of the config scripts from
+
+  http://git.savannah.gnu.org/gitweb/?p=config.git;a=blob_plain;f=config.guess;hb=HEAD
+and
+  http://git.savannah.gnu.org/gitweb/?p=config.git;a=blob_plain;f=config.sub;hb=HEAD
+
+If the version you run ($0) is already up to date, please
+send the following data and any information you think might be
+pertinent to <config-patches@gnu.org> in order to provide the needed
+information to handle your system.
+
+config.guess timestamp = $timestamp
+
+uname -m = `(uname -m) 2>/dev/null || echo unknown`
+uname -r = `(uname -r) 2>/dev/null || echo unknown`
+uname -s = `(uname -s) 2>/dev/null || echo unknown`
+uname -v = `(uname -v) 2>/dev/null || echo unknown`
+
+/usr/bin/uname -p = `(/usr/bin/uname -p) 2>/dev/null`
+/bin/uname -X     = `(/bin/uname -X) 2>/dev/null`
+
+hostinfo               = `(hostinfo) 2>/dev/null`
+/bin/universe          = `(/bin/universe) 2>/dev/null`
+/usr/bin/arch -k       = `(/usr/bin/arch -k) 2>/dev/null`
+/bin/arch              = `(/bin/arch) 2>/dev/null`
+/usr/bin/oslevel       = `(/usr/bin/oslevel) 2>/dev/null`
+/usr/convex/getsysinfo = `(/usr/convex/getsysinfo) 2>/dev/null`
+
+UNAME_MACHINE = ${UNAME_MACHINE}
+UNAME_RELEASE = ${UNAME_RELEASE}
+UNAME_SYSTEM  = ${UNAME_SYSTEM}
+UNAME_VERSION = ${UNAME_VERSION}
+EOF
+
+exit 1
+
+# Local variables:
+# eval: (add-hook 'write-file-hooks 'time-stamp)
+# time-stamp-start: "timestamp='"
+# time-stamp-format: "%:y-%02m-%02d"
+# time-stamp-end: "'"
+# End:
diff --git a/config.sub b/config.sub
new file mode 100644
index 0000000000..6759825a5b
--- /dev/null
+++ b/config.sub
@@ -0,0 +1,1658 @@
+#! /bin/sh
+# Configuration validation subroutine script.
+#   Copyright (C) 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
+#   2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
+#   Free Software Foundation, Inc.
+
+timestamp='2008-01-16'
+
+# This file is (in principle) common to ALL GNU software.
+# The presence of a machine in this file suggests that SOME GNU software
+# can handle that machine.  It does not imply ALL GNU software can.
+#
+# This file is free software; you can redistribute it and/or modify
+# it under the terms of the GNU General Public License as published by
+# the Free Software Foundation; either version 2 of the License, or
+# (at your option) any later version.
+#
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU General Public License
+# along with this program; if not, write to the Free Software
+# Foundation, Inc., 51 Franklin Street - Fifth Floor, Boston, MA
+# 02110-1301, USA.
+#
+# As a special exception to the GNU General Public License, if you
+# distribute this file as part of a program that contains a
+# configuration script generated by Autoconf, you may include it under
+# the same distribution terms that you use for the rest of that program.
+
+
+# Please send patches to <config-patches@gnu.org>.  Submit a context
+# diff and a properly formatted ChangeLog entry.
+#
+# Configuration subroutine to validate and canonicalize a configuration type.
+# Supply the specified configuration type as an argument.
+# If it is invalid, we print an error message on stderr and exit with code 1.
+# Otherwise, we print the canonical config type on stdout and succeed.
+
+# This file is supposed to be the same for all GNU packages
+# and recognize all the CPU types, system types and aliases
+# that are meaningful with *any* GNU software.
+# Each package is responsible for reporting which valid configurations
+# it does not support.  The user should be able to distinguish
+# a failure to support a valid configuration from a meaningless
+# configuration.
+
+# The goal of this file is to map all the various variations of a given
+# machine specification into a single specification in the form:
+#	CPU_TYPE-MANUFACTURER-OPERATING_SYSTEM
+# or in some cases, the newer four-part form:
+#	CPU_TYPE-MANUFACTURER-KERNEL-OPERATING_SYSTEM
+# It is wrong to echo any other type of specification.
+
+me=`echo "$0" | sed -e 's,.*/,,'`
+
+usage="\
+Usage: $0 [OPTION] CPU-MFR-OPSYS
+       $0 [OPTION] ALIAS
+
+Canonicalize a configuration name.
+
+Operation modes:
+  -h, --help         print this help, then exit
+  -t, --time-stamp   print date of last modification, then exit
+  -v, --version      print version number, then exit
+
+Report bugs and patches to <config-patches@gnu.org>."
+
+version="\
+GNU config.sub ($timestamp)
+
+Copyright (C) 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000, 2001,
+2002, 2003, 2004, 2005, 2006, 2007, 2008 Free Software Foundation, Inc.
+
+This is free software; see the source for copying conditions.  There is NO
+warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE."
+
+help="
+Try \`$me --help' for more information."
+
+# Parse command line
+while test $# -gt 0 ; do
+  case $1 in
+    --time-stamp | --time* | -t )
+       echo "$timestamp" ; exit ;;
+    --version | -v )
+       echo "$version" ; exit ;;
+    --help | --h* | -h )
+       echo "$usage"; exit ;;
+    -- )     # Stop option processing
+       shift; break ;;
+    - )	# Use stdin as input.
+       break ;;
+    -* )
+       echo "$me: invalid option $1$help"
+       exit 1 ;;
+
+    *local*)
+       # First pass through any local machine types.
+       echo $1
+       exit ;;
+
+    * )
+       break ;;
+  esac
+done
+
+case $# in
+ 0) echo "$me: missing argument$help" >&2
+    exit 1;;
+ 1) ;;
+ *) echo "$me: too many arguments$help" >&2
+    exit 1;;
+esac
+
+# Separate what the user gave into CPU-COMPANY and OS or KERNEL-OS (if any).
+# Here we must recognize all the valid KERNEL-OS combinations.
+maybe_os=`echo $1 | sed 's/^\(.*\)-\([^-]*-[^-]*\)$/\2/'`
+case $maybe_os in
+  nto-qnx* | linux-gnu* | linux-dietlibc | linux-newlib* | linux-uclibc* | \
+  uclinux-uclibc* | uclinux-gnu* | kfreebsd*-gnu* | knetbsd*-gnu* | netbsd*-gnu* | \
+  storm-chaos* | os2-emx* | rtmk-nova*)
+    os=-$maybe_os
+    basic_machine=`echo $1 | sed 's/^\(.*\)-\([^-]*-[^-]*\)$/\1/'`
+    ;;
+  *)
+    basic_machine=`echo $1 | sed 's/-[^-]*$//'`
+    if [ $basic_machine != $1 ]
+    then os=`echo $1 | sed 's/.*-/-/'`
+    else os=; fi
+    ;;
+esac
+
+### Let's recognize common machines as not being operating systems so
+### that things like config.sub decstation-3100 work.  We also
+### recognize some manufacturers as not being operating systems, so we
+### can provide default operating systems below.
+case $os in
+	-sun*os*)
+		# Prevent following clause from handling this invalid input.
+		;;
+	-dec* | -mips* | -sequent* | -encore* | -pc532* | -sgi* | -sony* | \
+	-att* | -7300* | -3300* | -delta* | -motorola* | -sun[234]* | \
+	-unicom* | -ibm* | -next | -hp | -isi* | -apollo | -altos* | \
+	-convergent* | -ncr* | -news | -32* | -3600* | -3100* | -hitachi* |\
+	-c[123]* | -convex* | -sun | -crds | -omron* | -dg | -ultra | -tti* | \
+	-harris | -dolphin | -highlevel | -gould | -cbm | -ns | -masscomp | \
+	-apple | -axis | -knuth | -cray)
+		os=
+		basic_machine=$1
+		;;
+	-sim | -cisco | -oki | -wec | -winbond)
+		os=
+		basic_machine=$1
+		;;
+	-scout)
+		;;
+	-wrs)
+		os=-vxworks
+		basic_machine=$1
+		;;
+	-chorusos*)
+		os=-chorusos
+		basic_machine=$1
+		;;
+ 	-chorusrdb)
+ 		os=-chorusrdb
+		basic_machine=$1
+ 		;;
+	-hiux*)
+		os=-hiuxwe2
+		;;
+	-sco6)
+		os=-sco5v6
+		basic_machine=`echo $1 | sed -e 's/86-.*/86-pc/'`
+		;;
+	-sco5)
+		os=-sco3.2v5
+		basic_machine=`echo $1 | sed -e 's/86-.*/86-pc/'`
+		;;
+	-sco4)
+		os=-sco3.2v4
+		basic_machine=`echo $1 | sed -e 's/86-.*/86-pc/'`
+		;;
+	-sco3.2.[4-9]*)
+		os=`echo $os | sed -e 's/sco3.2./sco3.2v/'`
+		basic_machine=`echo $1 | sed -e 's/86-.*/86-pc/'`
+		;;
+	-sco3.2v[4-9]*)
+		# Don't forget version if it is 3.2v4 or newer.
+		basic_machine=`echo $1 | sed -e 's/86-.*/86-pc/'`
+		;;
+	-sco5v6*)
+		# Don't forget version if it is 3.2v4 or newer.
+		basic_machine=`echo $1 | sed -e 's/86-.*/86-pc/'`
+		;;
+	-sco*)
+		os=-sco3.2v2
+		basic_machine=`echo $1 | sed -e 's/86-.*/86-pc/'`
+		;;
+	-udk*)
+		basic_machine=`echo $1 | sed -e 's/86-.*/86-pc/'`
+		;;
+	-isc)
+		os=-isc2.2
+		basic_machine=`echo $1 | sed -e 's/86-.*/86-pc/'`
+		;;
+	-clix*)
+		basic_machine=clipper-intergraph
+		;;
+	-isc*)
+		basic_machine=`echo $1 | sed -e 's/86-.*/86-pc/'`
+		;;
+	-lynx*)
+		os=-lynxos
+		;;
+	-ptx*)
+		basic_machine=`echo $1 | sed -e 's/86-.*/86-sequent/'`
+		;;
+	-windowsnt*)
+		os=`echo $os | sed -e 's/windowsnt/winnt/'`
+		;;
+	-psos*)
+		os=-psos
+		;;
+	-mint | -mint[0-9]*)
+		basic_machine=m68k-atari
+		os=-mint
+		;;
+esac
+
+# Decode aliases for certain CPU-COMPANY combinations.
+case $basic_machine in
+	# Recognize the basic CPU types without company name.
+	# Some are omitted here because they have special meanings below.
+	1750a | 580 \
+	| a29k \
+	| alpha | alphaev[4-8] | alphaev56 | alphaev6[78] | alphapca5[67] \
+	| alpha64 | alpha64ev[4-8] | alpha64ev56 | alpha64ev6[78] | alpha64pca5[67] \
+	| am33_2.0 \
+	| arc | arm | arm[bl]e | arme[lb] | armv[2345] | armv[345][lb] | avr | avr32 \
+	| bfin \
+	| c4x | clipper \
+	| d10v | d30v | dlx | dsp16xx \
+	| fido | fr30 | frv \
+	| h8300 | h8500 | hppa | hppa1.[01] | hppa2.0 | hppa2.0[nw] | hppa64 \
+	| i370 | i860 | i960 | ia64 \
+	| ip2k | iq2000 \
+	| m32c | m32r | m32rle | m68000 | m68k | m88k \
+	| maxq | mb | microblaze | mcore | mep \
+	| mips | mipsbe | mipseb | mipsel | mipsle \
+	| mips16 \
+	| mips64 | mips64el \
+	| mips64vr | mips64vrel \
+	| mips64orion | mips64orionel \
+	| mips64vr4100 | mips64vr4100el \
+	| mips64vr4300 | mips64vr4300el \
+	| mips64vr5000 | mips64vr5000el \
+	| mips64vr5900 | mips64vr5900el \
+	| mipsisa32 | mipsisa32el \
+	| mipsisa32r2 | mipsisa32r2el \
+	| mipsisa64 | mipsisa64el \
+	| mipsisa64r2 | mipsisa64r2el \
+	| mipsisa64sb1 | mipsisa64sb1el \
+	| mipsisa64sr71k | mipsisa64sr71kel \
+	| mipstx39 | mipstx39el \
+	| mn10200 | mn10300 \
+	| mt \
+	| msp430 \
+	| nios | nios2 \
+	| ns16k | ns32k \
+	| or32 \
+	| pdp10 | pdp11 | pj | pjl \
+	| powerpc | powerpc64 | powerpc64le | powerpcle | ppcbe \
+	| pyramid \
+	| score \
+	| sh | sh[1234] | sh[24]a | sh[23]e | sh[34]eb | sheb | shbe | shle | sh[1234]le | sh3ele \
+	| sh64 | sh64le \
+	| sparc | sparc64 | sparc64b | sparc64v | sparc86x | sparclet | sparclite \
+	| sparcv8 | sparcv9 | sparcv9b | sparcv9v \
+	| spu | strongarm \
+	| tahoe | thumb | tic4x | tic80 | tron \
+	| v850 | v850e \
+	| we32k \
+	| x86 | xc16x | xscale | xscalee[bl] | xstormy16 | xtensa \
+	| z8k)
+		basic_machine=$basic_machine-unknown
+		;;
+	m6811 | m68hc11 | m6812 | m68hc12)
+		# Motorola 68HC11/12.
+		basic_machine=$basic_machine-unknown
+		os=-none
+		;;
+	m88110 | m680[12346]0 | m683?2 | m68360 | m5200 | v70 | w65 | z8k)
+		;;
+	ms1)
+		basic_machine=mt-unknown
+		;;
+
+	# We use `pc' rather than `unknown'
+	# because (1) that's what they normally are, and
+	# (2) the word "unknown" tends to confuse beginning users.
+	i*86 | x86_64)
+	  basic_machine=$basic_machine-pc
+	  ;;
+	# Object if more than one company name word.
+	*-*-*)
+		echo Invalid configuration \`$1\': machine \`$basic_machine\' not recognized 1>&2
+		exit 1
+		;;
+	# Recognize the basic CPU types with company name.
+	580-* \
+	| a29k-* \
+	| alpha-* | alphaev[4-8]-* | alphaev56-* | alphaev6[78]-* \
+	| alpha64-* | alpha64ev[4-8]-* | alpha64ev56-* | alpha64ev6[78]-* \
+	| alphapca5[67]-* | alpha64pca5[67]-* | arc-* \
+	| arm-*  | armbe-* | armle-* | armeb-* | armv*-* \
+	| avr-* | avr32-* \
+	| bfin-* | bs2000-* \
+	| c[123]* | c30-* | [cjt]90-* | c4x-* | c54x-* | c55x-* | c6x-* \
+	| clipper-* | craynv-* | cydra-* \
+	| d10v-* | d30v-* | dlx-* \
+	| elxsi-* \
+	| f30[01]-* | f700-* | fido-* | fr30-* | frv-* | fx80-* \
+	| h8300-* | h8500-* \
+	| hppa-* | hppa1.[01]-* | hppa2.0-* | hppa2.0[nw]-* | hppa64-* \
+	| i*86-* | i860-* | i960-* | ia64-* \
+	| ip2k-* | iq2000-* \
+	| m32c-* | m32r-* | m32rle-* \
+	| m68000-* | m680[012346]0-* | m68360-* | m683?2-* | m68k-* \
+	| m88110-* | m88k-* | maxq-* | mcore-* \
+	| mips-* | mipsbe-* | mipseb-* | mipsel-* | mipsle-* \
+	| mips16-* \
+	| mips64-* | mips64el-* \
+	| mips64vr-* | mips64vrel-* \
+	| mips64orion-* | mips64orionel-* \
+	| mips64vr4100-* | mips64vr4100el-* \
+	| mips64vr4300-* | mips64vr4300el-* \
+	| mips64vr5000-* | mips64vr5000el-* \
+	| mips64vr5900-* | mips64vr5900el-* \
+	| mipsisa32-* | mipsisa32el-* \
+	| mipsisa32r2-* | mipsisa32r2el-* \
+	| mipsisa64-* | mipsisa64el-* \
+	| mipsisa64r2-* | mipsisa64r2el-* \
+	| mipsisa64sb1-* | mipsisa64sb1el-* \
+	| mipsisa64sr71k-* | mipsisa64sr71kel-* \
+	| mipstx39-* | mipstx39el-* \
+	| mmix-* \
+	| mt-* \
+	| msp430-* \
+	| nios-* | nios2-* \
+	| none-* | np1-* | ns16k-* | ns32k-* \
+	| orion-* \
+	| pdp10-* | pdp11-* | pj-* | pjl-* | pn-* | power-* \
+	| powerpc-* | powerpc64-* | powerpc64le-* | powerpcle-* | ppcbe-* \
+	| pyramid-* \
+	| romp-* | rs6000-* \
+	| sh-* | sh[1234]-* | sh[24]a-* | sh[23]e-* | sh[34]eb-* | sheb-* | shbe-* \
+	| shle-* | sh[1234]le-* | sh3ele-* | sh64-* | sh64le-* \
+	| sparc-* | sparc64-* | sparc64b-* | sparc64v-* | sparc86x-* | sparclet-* \
+	| sparclite-* \
+	| sparcv8-* | sparcv9-* | sparcv9b-* | sparcv9v-* | strongarm-* | sv1-* | sx?-* \
+	| tahoe-* | thumb-* \
+	| tic30-* | tic4x-* | tic54x-* | tic55x-* | tic6x-* | tic80-* \
+	| tron-* \
+	| v850-* | v850e-* | vax-* \
+	| we32k-* \
+	| x86-* | x86_64-* | xc16x-* | xps100-* | xscale-* | xscalee[bl]-* \
+	| xstormy16-* | xtensa*-* \
+	| ymp-* \
+	| z8k-*)
+		;;
+	# Recognize the basic CPU types without company name, with glob match.
+	xtensa*)
+		basic_machine=$basic_machine-unknown
+		;;
+	# Recognize the various machine names and aliases which stand
+	# for a CPU type and a company and sometimes even an OS.
+	386bsd)
+		basic_machine=i386-unknown
+		os=-bsd
+		;;
+	3b1 | 7300 | 7300-att | att-7300 | pc7300 | safari | unixpc)
+		basic_machine=m68000-att
+		;;
+	3b*)
+		basic_machine=we32k-att
+		;;
+	a29khif)
+		basic_machine=a29k-amd
+		os=-udi
+		;;
+    	abacus)
+		basic_machine=abacus-unknown
+		;;
+	adobe68k)
+		basic_machine=m68010-adobe
+		os=-scout
+		;;
+	alliant | fx80)
+		basic_machine=fx80-alliant
+		;;
+	altos | altos3068)
+		basic_machine=m68k-altos
+		;;
+	am29k)
+		basic_machine=a29k-none
+		os=-bsd
+		;;
+	amd64)
+		basic_machine=x86_64-pc
+		;;
+	amd64-*)
+		basic_machine=x86_64-`echo $basic_machine | sed 's/^[^-]*-//'`
+		;;
+	amdahl)
+		basic_machine=580-amdahl
+		os=-sysv
+		;;
+	amiga | amiga-*)
+		basic_machine=m68k-unknown
+		;;
+	amigaos | amigados)
+		basic_machine=m68k-unknown
+		os=-amigaos
+		;;
+	amigaunix | amix)
+		basic_machine=m68k-unknown
+		os=-sysv4
+		;;
+	apollo68)
+		basic_machine=m68k-apollo
+		os=-sysv
+		;;
+	apollo68bsd)
+		basic_machine=m68k-apollo
+		os=-bsd
+		;;
+	aux)
+		basic_machine=m68k-apple
+		os=-aux
+		;;
+	balance)
+		basic_machine=ns32k-sequent
+		os=-dynix
+		;;
+	blackfin)
+		basic_machine=bfin-unknown
+		os=-linux
+		;;
+	blackfin-*)
+		basic_machine=bfin-`echo $basic_machine | sed 's/^[^-]*-//'`
+		os=-linux
+		;;
+	c90)
+		basic_machine=c90-cray
+		os=-unicos
+		;;
+	convex-c1)
+		basic_machine=c1-convex
+		os=-bsd
+		;;
+	convex-c2)
+		basic_machine=c2-convex
+		os=-bsd
+		;;
+	convex-c32)
+		basic_machine=c32-convex
+		os=-bsd
+		;;
+	convex-c34)
+		basic_machine=c34-convex
+		os=-bsd
+		;;
+	convex-c38)
+		basic_machine=c38-convex
+		os=-bsd
+		;;
+	cray | j90)
+		basic_machine=j90-cray
+		os=-unicos
+		;;
+	craynv)
+		basic_machine=craynv-cray
+		os=-unicosmp
+		;;
+	cr16)
+		basic_machine=cr16-unknown
+		os=-elf
+		;;
+	crds | unos)
+		basic_machine=m68k-crds
+		;;
+	crisv32 | crisv32-* | etraxfs*)
+		basic_machine=crisv32-axis
+		;;
+	cris | cris-* | etrax*)
+		basic_machine=cris-axis
+		;;
+	crx)
+		basic_machine=crx-unknown
+		os=-elf
+		;;
+	da30 | da30-*)
+		basic_machine=m68k-da30
+		;;
+	decstation | decstation-3100 | pmax | pmax-* | pmin | dec3100 | decstatn)
+		basic_machine=mips-dec
+		;;
+	decsystem10* | dec10*)
+		basic_machine=pdp10-dec
+		os=-tops10
+		;;
+	decsystem20* | dec20*)
+		basic_machine=pdp10-dec
+		os=-tops20
+		;;
+	delta | 3300 | motorola-3300 | motorola-delta \
+	      | 3300-motorola | delta-motorola)
+		basic_machine=m68k-motorola
+		;;
+	delta88)
+		basic_machine=m88k-motorola
+		os=-sysv3
+		;;
+	djgpp)
+		basic_machine=i586-pc
+		os=-msdosdjgpp
+		;;
+	dpx20 | dpx20-*)
+		basic_machine=rs6000-bull
+		os=-bosx
+		;;
+	dpx2* | dpx2*-bull)
+		basic_machine=m68k-bull
+		os=-sysv3
+		;;
+	ebmon29k)
+		basic_machine=a29k-amd
+		os=-ebmon
+		;;
+	elxsi)
+		basic_machine=elxsi-elxsi
+		os=-bsd
+		;;
+	encore | umax | mmax)
+		basic_machine=ns32k-encore
+		;;
+	es1800 | OSE68k | ose68k | ose | OSE)
+		basic_machine=m68k-ericsson
+		os=-ose
+		;;
+	fx2800)
+		basic_machine=i860-alliant
+		;;
+	genix)
+		basic_machine=ns32k-ns
+		;;
+	gmicro)
+		basic_machine=tron-gmicro
+		os=-sysv
+		;;
+	go32)
+		basic_machine=i386-pc
+		os=-go32
+		;;
+	h3050r* | hiux*)
+		basic_machine=hppa1.1-hitachi
+		os=-hiuxwe2
+		;;
+	h8300hms)
+		basic_machine=h8300-hitachi
+		os=-hms
+		;;
+	h8300xray)
+		basic_machine=h8300-hitachi
+		os=-xray
+		;;
+	h8500hms)
+		basic_machine=h8500-hitachi
+		os=-hms
+		;;
+	harris)
+		basic_machine=m88k-harris
+		os=-sysv3
+		;;
+	hp300-*)
+		basic_machine=m68k-hp
+		;;
+	hp300bsd)
+		basic_machine=m68k-hp
+		os=-bsd
+		;;
+	hp300hpux)
+		basic_machine=m68k-hp
+		os=-hpux
+		;;
+	hp3k9[0-9][0-9] | hp9[0-9][0-9])
+		basic_machine=hppa1.0-hp
+		;;
+	hp9k2[0-9][0-9] | hp9k31[0-9])
+		basic_machine=m68000-hp
+		;;
+	hp9k3[2-9][0-9])
+		basic_machine=m68k-hp
+		;;
+	hp9k6[0-9][0-9] | hp6[0-9][0-9])
+		basic_machine=hppa1.0-hp
+		;;
+	hp9k7[0-79][0-9] | hp7[0-79][0-9])
+		basic_machine=hppa1.1-hp
+		;;
+	hp9k78[0-9] | hp78[0-9])
+		# FIXME: really hppa2.0-hp
+		basic_machine=hppa1.1-hp
+		;;
+	hp9k8[67]1 | hp8[67]1 | hp9k80[24] | hp80[24] | hp9k8[78]9 | hp8[78]9 | hp9k893 | hp893)
+		# FIXME: really hppa2.0-hp
+		basic_machine=hppa1.1-hp
+		;;
+	hp9k8[0-9][13679] | hp8[0-9][13679])
+		basic_machine=hppa1.1-hp
+		;;
+	hp9k8[0-9][0-9] | hp8[0-9][0-9])
+		basic_machine=hppa1.0-hp
+		;;
+	hppa-next)
+		os=-nextstep3
+		;;
+	hppaosf)
+		basic_machine=hppa1.1-hp
+		os=-osf
+		;;
+	hppro)
+		basic_machine=hppa1.1-hp
+		os=-proelf
+		;;
+	i370-ibm* | ibm*)
+		basic_machine=i370-ibm
+		;;
+# I'm not sure what "Sysv32" means.  Should this be sysv3.2?
+	i*86v32)
+		basic_machine=`echo $1 | sed -e 's/86.*/86-pc/'`
+		os=-sysv32
+		;;
+	i*86v4*)
+		basic_machine=`echo $1 | sed -e 's/86.*/86-pc/'`
+		os=-sysv4
+		;;
+	i*86v)
+		basic_machine=`echo $1 | sed -e 's/86.*/86-pc/'`
+		os=-sysv
+		;;
+	i*86sol2)
+		basic_machine=`echo $1 | sed -e 's/86.*/86-pc/'`
+		os=-solaris2
+		;;
+	i386mach)
+		basic_machine=i386-mach
+		os=-mach
+		;;
+	i386-vsta | vsta)
+		basic_machine=i386-unknown
+		os=-vsta
+		;;
+	iris | iris4d)
+		basic_machine=mips-sgi
+		case $os in
+		    -irix*)
+			;;
+		    *)
+			os=-irix4
+			;;
+		esac
+		;;
+	isi68 | isi)
+		basic_machine=m68k-isi
+		os=-sysv
+		;;
+	m68knommu)
+		basic_machine=m68k-unknown
+		os=-linux
+		;;
+	m68knommu-*)
+		basic_machine=m68k-`echo $basic_machine | sed 's/^[^-]*-//'`
+		os=-linux
+		;;
+	m88k-omron*)
+		basic_machine=m88k-omron
+		;;
+	magnum | m3230)
+		basic_machine=mips-mips
+		os=-sysv
+		;;
+	merlin)
+		basic_machine=ns32k-utek
+		os=-sysv
+		;;
+	mingw32)
+		basic_machine=i386-pc
+		os=-mingw32
+		;;
+	mingw32ce)
+		basic_machine=arm-unknown
+		os=-mingw32ce
+		;;
+	miniframe)
+		basic_machine=m68000-convergent
+		;;
+	*mint | -mint[0-9]* | *MiNT | *MiNT[0-9]*)
+		basic_machine=m68k-atari
+		os=-mint
+		;;
+	mips3*-*)
+		basic_machine=`echo $basic_machine | sed -e 's/mips3/mips64/'`
+		;;
+	mips3*)
+		basic_machine=`echo $basic_machine | sed -e 's/mips3/mips64/'`-unknown
+		;;
+	monitor)
+		basic_machine=m68k-rom68k
+		os=-coff
+		;;
+	morphos)
+		basic_machine=powerpc-unknown
+		os=-morphos
+		;;
+	msdos)
+		basic_machine=i386-pc
+		os=-msdos
+		;;
+	ms1-*)
+		basic_machine=`echo $basic_machine | sed -e 's/ms1-/mt-/'`
+		;;
+	mvs)
+		basic_machine=i370-ibm
+		os=-mvs
+		;;
+	ncr3000)
+		basic_machine=i486-ncr
+		os=-sysv4
+		;;
+	netbsd386)
+		basic_machine=i386-unknown
+		os=-netbsd
+		;;
+	netwinder)
+		basic_machine=armv4l-rebel
+		os=-linux
+		;;
+	news | news700 | news800 | news900)
+		basic_machine=m68k-sony
+		os=-newsos
+		;;
+	news1000)
+		basic_machine=m68030-sony
+		os=-newsos
+		;;
+	news-3600 | risc-news)
+		basic_machine=mips-sony
+		os=-newsos
+		;;
+	necv70)
+		basic_machine=v70-nec
+		os=-sysv
+		;;
+	next | m*-next )
+		basic_machine=m68k-next
+		case $os in
+		    -nextstep* )
+			;;
+		    -ns2*)
+		      os=-nextstep2
+			;;
+		    *)
+		      os=-nextstep3
+			;;
+		esac
+		;;
+	nh3000)
+		basic_machine=m68k-harris
+		os=-cxux
+		;;
+	nh[45]000)
+		basic_machine=m88k-harris
+		os=-cxux
+		;;
+	nindy960)
+		basic_machine=i960-intel
+		os=-nindy
+		;;
+	mon960)
+		basic_machine=i960-intel
+		os=-mon960
+		;;
+	nonstopux)
+		basic_machine=mips-compaq
+		os=-nonstopux
+		;;
+	np1)
+		basic_machine=np1-gould
+		;;
+	nsr-tandem)
+		basic_machine=nsr-tandem
+		;;
+	op50n-* | op60c-*)
+		basic_machine=hppa1.1-oki
+		os=-proelf
+		;;
+	openrisc | openrisc-*)
+		basic_machine=or32-unknown
+		;;
+	os400)
+		basic_machine=powerpc-ibm
+		os=-os400
+		;;
+	OSE68000 | ose68000)
+		basic_machine=m68000-ericsson
+		os=-ose
+		;;
+	os68k)
+		basic_machine=m68k-none
+		os=-os68k
+		;;
+	pa-hitachi)
+		basic_machine=hppa1.1-hitachi
+		os=-hiuxwe2
+		;;
+	paragon)
+		basic_machine=i860-intel
+		os=-osf
+		;;
+	parisc)
+		basic_machine=hppa-unknown
+		os=-linux
+		;;
+	parisc-*)
+		basic_machine=hppa-`echo $basic_machine | sed 's/^[^-]*-//'`
+		os=-linux
+		;;
+	pbd)
+		basic_machine=sparc-tti
+		;;
+	pbb)
+		basic_machine=m68k-tti
+		;;
+	pc532 | pc532-*)
+		basic_machine=ns32k-pc532
+		;;
+	pc98)
+		basic_machine=i386-pc
+		;;
+	pc98-*)
+		basic_machine=i386-`echo $basic_machine | sed 's/^[^-]*-//'`
+		;;
+	pentium | p5 | k5 | k6 | nexgen | viac3)
+		basic_machine=i586-pc
+		;;
+	pentiumpro | p6 | 6x86 | athlon | athlon_*)
+		basic_machine=i686-pc
+		;;
+	pentiumii | pentium2 | pentiumiii | pentium3)
+		basic_machine=i686-pc
+		;;
+	pentium4)
+		basic_machine=i786-pc
+		;;
+	pentium-* | p5-* | k5-* | k6-* | nexgen-* | viac3-*)
+		basic_machine=i586-`echo $basic_machine | sed 's/^[^-]*-//'`
+		;;
+	pentiumpro-* | p6-* | 6x86-* | athlon-*)
+		basic_machine=i686-`echo $basic_machine | sed 's/^[^-]*-//'`
+		;;
+	pentiumii-* | pentium2-* | pentiumiii-* | pentium3-*)
+		basic_machine=i686-`echo $basic_machine | sed 's/^[^-]*-//'`
+		;;
+	pentium4-*)
+		basic_machine=i786-`echo $basic_machine | sed 's/^[^-]*-//'`
+		;;
+	pn)
+		basic_machine=pn-gould
+		;;
+	power)	basic_machine=power-ibm
+		;;
+	ppc)	basic_machine=powerpc-unknown
+		;;
+	ppc-*)	basic_machine=powerpc-`echo $basic_machine | sed 's/^[^-]*-//'`
+		;;
+	ppcle | powerpclittle | ppc-le | powerpc-little)
+		basic_machine=powerpcle-unknown
+		;;
+	ppcle-* | powerpclittle-*)
+		basic_machine=powerpcle-`echo $basic_machine | sed 's/^[^-]*-//'`
+		;;
+	ppc64)	basic_machine=powerpc64-unknown
+		;;
+	ppc64-*) basic_machine=powerpc64-`echo $basic_machine | sed 's/^[^-]*-//'`
+		;;
+	ppc64le | powerpc64little | ppc64-le | powerpc64-little)
+		basic_machine=powerpc64le-unknown
+		;;
+	ppc64le-* | powerpc64little-*)
+		basic_machine=powerpc64le-`echo $basic_machine | sed 's/^[^-]*-//'`
+		;;
+	ps2)
+		basic_machine=i386-ibm
+		;;
+	pw32)
+		basic_machine=i586-unknown
+		os=-pw32
+		;;
+	rdos)
+		basic_machine=i386-pc
+		os=-rdos
+		;;
+	rom68k)
+		basic_machine=m68k-rom68k
+		os=-coff
+		;;
+	rm[46]00)
+		basic_machine=mips-siemens
+		;;
+	rtpc | rtpc-*)
+		basic_machine=romp-ibm
+		;;
+	s390 | s390-*)
+		basic_machine=s390-ibm
+		;;
+	s390x | s390x-*)
+		basic_machine=s390x-ibm
+		;;
+	sa29200)
+		basic_machine=a29k-amd
+		os=-udi
+		;;
+	sb1)
+		basic_machine=mipsisa64sb1-unknown
+		;;
+	sb1el)
+		basic_machine=mipsisa64sb1el-unknown
+		;;
+	sde)
+		basic_machine=mipsisa32-sde
+		os=-elf
+		;;
+	sei)
+		basic_machine=mips-sei
+		os=-seiux
+		;;
+	sequent)
+		basic_machine=i386-sequent
+		;;
+	sh)
+		basic_machine=sh-hitachi
+		os=-hms
+		;;
+	sh5el)
+		basic_machine=sh5le-unknown
+		;;
+	sh64)
+		basic_machine=sh64-unknown
+		;;
+	sparclite-wrs | simso-wrs)
+		basic_machine=sparclite-wrs
+		os=-vxworks
+		;;
+	sps7)
+		basic_machine=m68k-bull
+		os=-sysv2
+		;;
+	spur)
+		basic_machine=spur-unknown
+		;;
+	st2000)
+		basic_machine=m68k-tandem
+		;;
+	stratus)
+		basic_machine=i860-stratus
+		os=-sysv4
+		;;
+	sun2)
+		basic_machine=m68000-sun
+		;;
+	sun2os3)
+		basic_machine=m68000-sun
+		os=-sunos3
+		;;
+	sun2os4)
+		basic_machine=m68000-sun
+		os=-sunos4
+		;;
+	sun3os3)
+		basic_machine=m68k-sun
+		os=-sunos3
+		;;
+	sun3os4)
+		basic_machine=m68k-sun
+		os=-sunos4
+		;;
+	sun4os3)
+		basic_machine=sparc-sun
+		os=-sunos3
+		;;
+	sun4os4)
+		basic_machine=sparc-sun
+		os=-sunos4
+		;;
+	sun4sol2)
+		basic_machine=sparc-sun
+		os=-solaris2
+		;;
+	sun3 | sun3-*)
+		basic_machine=m68k-sun
+		;;
+	sun4)
+		basic_machine=sparc-sun
+		;;
+	sun386 | sun386i | roadrunner)
+		basic_machine=i386-sun
+		;;
+	sv1)
+		basic_machine=sv1-cray
+		os=-unicos
+		;;
+	symmetry)
+		basic_machine=i386-sequent
+		os=-dynix
+		;;
+	t3e)
+		basic_machine=alphaev5-cray
+		os=-unicos
+		;;
+	t90)
+		basic_machine=t90-cray
+		os=-unicos
+		;;
+	tic54x | c54x*)
+		basic_machine=tic54x-unknown
+		os=-coff
+		;;
+	tic55x | c55x*)
+		basic_machine=tic55x-unknown
+		os=-coff
+		;;
+	tic6x | c6x*)
+		basic_machine=tic6x-unknown
+		os=-coff
+		;;
+	tile*)
+		basic_machine=tile-unknown
+		os=-linux-gnu
+		;;
+	tx39)
+		basic_machine=mipstx39-unknown
+		;;
+	tx39el)
+		basic_machine=mipstx39el-unknown
+		;;
+	toad1)
+		basic_machine=pdp10-xkl
+		os=-tops20
+		;;
+	tower | tower-32)
+		basic_machine=m68k-ncr
+		;;
+	tpf)
+		basic_machine=s390x-ibm
+		os=-tpf
+		;;
+	udi29k)
+		basic_machine=a29k-amd
+		os=-udi
+		;;
+	ultra3)
+		basic_machine=a29k-nyu
+		os=-sym1
+		;;
+	v810 | necv810)
+		basic_machine=v810-nec
+		os=-none
+		;;
+	vaxv)
+		basic_machine=vax-dec
+		os=-sysv
+		;;
+	vms)
+		basic_machine=vax-dec
+		os=-vms
+		;;
+	vpp*|vx|vx-*)
+		basic_machine=f301-fujitsu
+		;;
+	vxworks960)
+		basic_machine=i960-wrs
+		os=-vxworks
+		;;
+	vxworks68)
+		basic_machine=m68k-wrs
+		os=-vxworks
+		;;
+	vxworks29k)
+		basic_machine=a29k-wrs
+		os=-vxworks
+		;;
+	w65*)
+		basic_machine=w65-wdc
+		os=-none
+		;;
+	w89k-*)
+		basic_machine=hppa1.1-winbond
+		os=-proelf
+		;;
+	xbox)
+		basic_machine=i686-pc
+		os=-mingw32
+		;;
+	xps | xps100)
+		basic_machine=xps100-honeywell
+		;;
+	ymp)
+		basic_machine=ymp-cray
+		os=-unicos
+		;;
+	z8k-*-coff)
+		basic_machine=z8k-unknown
+		os=-sim
+		;;
+	none)
+		basic_machine=none-none
+		os=-none
+		;;
+
+# Here we handle the default manufacturer of certain CPU types.  It is in
+# some cases the only manufacturer, in others, it is the most popular.
+	w89k)
+		basic_machine=hppa1.1-winbond
+		;;
+	op50n)
+		basic_machine=hppa1.1-oki
+		;;
+	op60c)
+		basic_machine=hppa1.1-oki
+		;;
+	romp)
+		basic_machine=romp-ibm
+		;;
+	mmix)
+		basic_machine=mmix-knuth
+		;;
+	rs6000)
+		basic_machine=rs6000-ibm
+		;;
+	vax)
+		basic_machine=vax-dec
+		;;
+	pdp10)
+		# there are many clones, so DEC is not a safe bet
+		basic_machine=pdp10-unknown
+		;;
+	pdp11)
+		basic_machine=pdp11-dec
+		;;
+	we32k)
+		basic_machine=we32k-att
+		;;
+	sh[1234] | sh[24]a | sh[34]eb | sh[1234]le | sh[23]ele)
+		basic_machine=sh-unknown
+		;;
+	sparc | sparcv8 | sparcv9 | sparcv9b | sparcv9v)
+		basic_machine=sparc-sun
+		;;
+	cydra)
+		basic_machine=cydra-cydrome
+		;;
+	orion)
+		basic_machine=orion-highlevel
+		;;
+	orion105)
+		basic_machine=clipper-highlevel
+		;;
+	mac | mpw | mac-mpw)
+		basic_machine=m68k-apple
+		;;
+	pmac | pmac-mpw)
+		basic_machine=powerpc-apple
+		;;
+	*-unknown)
+		# Make sure to match an already-canonicalized machine name.
+		;;
+	*)
+		echo Invalid configuration \`$1\': machine \`$basic_machine\' not recognized 1>&2
+		exit 1
+		;;
+esac
+
+# Here we canonicalize certain aliases for manufacturers.
+case $basic_machine in
+	*-digital*)
+		basic_machine=`echo $basic_machine | sed 's/digital.*/dec/'`
+		;;
+	*-commodore*)
+		basic_machine=`echo $basic_machine | sed 's/commodore.*/cbm/'`
+		;;
+	*)
+		;;
+esac
+
+# Decode manufacturer-specific aliases for certain operating systems.
+
+if [ x"$os" != x"" ]
+then
+case $os in
+        # First match some system type aliases
+        # that might get confused with valid system types.
+	# -solaris* is a basic system type, with this one exception.
+	-solaris1 | -solaris1.*)
+		os=`echo $os | sed -e 's|solaris1|sunos4|'`
+		;;
+	-solaris)
+		os=-solaris2
+		;;
+	-svr4*)
+		os=-sysv4
+		;;
+	-unixware*)
+		os=-sysv4.2uw
+		;;
+	-gnu/linux*)
+		os=`echo $os | sed -e 's|gnu/linux|linux-gnu|'`
+		;;
+	# First accept the basic system types.
+	# The portable systems comes first.
+	# Each alternative MUST END IN A *, to match a version number.
+	# -sysv* is not here because it comes later, after sysvr4.
+	-gnu* | -bsd* | -mach* | -minix* | -genix* | -ultrix* | -irix* \
+	      | -*vms* | -sco* | -esix* | -isc* | -aix* | -sunos | -sunos[34]*\
+	      | -hpux* | -unos* | -osf* | -luna* | -dgux* | -solaris* | -sym* \
+	      | -amigaos* | -amigados* | -msdos* | -newsos* | -unicos* | -aof* \
+	      | -aos* \
+	      | -nindy* | -vxsim* | -vxworks* | -ebmon* | -hms* | -mvs* \
+	      | -clix* | -riscos* | -uniplus* | -iris* | -rtu* | -xenix* \
+	      | -hiux* | -386bsd* | -knetbsd* | -mirbsd* | -netbsd* \
+	      | -openbsd* | -solidbsd* \
+	      | -ekkobsd* | -kfreebsd* | -freebsd* | -riscix* | -lynxos* \
+	      | -bosx* | -nextstep* | -cxux* | -aout* | -elf* | -oabi* \
+	      | -ptx* | -coff* | -ecoff* | -winnt* | -domain* | -vsta* \
+	      | -udi* | -eabi* | -lites* | -ieee* | -go32* | -aux* \
+	      | -chorusos* | -chorusrdb* \
+	      | -cygwin* | -pe* | -psos* | -moss* | -proelf* | -rtems* \
+	      | -mingw32* | -linux-gnu* | -linux-newlib* | -linux-uclibc* \
+	      | -uxpv* | -beos* | -mpeix* | -udk* \
+	      | -interix* | -uwin* | -mks* | -rhapsody* | -darwin* | -opened* \
+	      | -openstep* | -oskit* | -conix* | -pw32* | -nonstopux* \
+	      | -storm-chaos* | -tops10* | -tenex* | -tops20* | -its* \
+	      | -os2* | -vos* | -palmos* | -uclinux* | -nucleus* \
+	      | -morphos* | -superux* | -rtmk* | -rtmk-nova* | -windiss* \
+	      | -powermax* | -dnix* | -nx6 | -nx7 | -sei* | -dragonfly* \
+	      | -skyos* | -haiku* | -rdos* | -toppers* | -drops*)
+	# Remember, each alternative MUST END IN *, to match a version number.
+		;;
+	-qnx*)
+		case $basic_machine in
+		    x86-* | i*86-*)
+			;;
+		    *)
+			os=-nto$os
+			;;
+		esac
+		;;
+	-nto-qnx*)
+		;;
+	-nto*)
+		os=`echo $os | sed -e 's|nto|nto-qnx|'`
+		;;
+	-sim | -es1800* | -hms* | -xray | -os68k* | -none* | -v88r* \
+	      | -windows* | -osx | -abug | -netware* | -os9* | -beos* | -haiku* \
+	      | -macos* | -mpw* | -magic* | -mmixware* | -mon960* | -lnews*)
+		;;
+	-mac*)
+		os=`echo $os | sed -e 's|mac|macos|'`
+		;;
+	-linux-dietlibc)
+		os=-linux-dietlibc
+		;;
+	-linux*)
+		os=`echo $os | sed -e 's|linux|linux-gnu|'`
+		;;
+	-sunos5*)
+		os=`echo $os | sed -e 's|sunos5|solaris2|'`
+		;;
+	-sunos6*)
+		os=`echo $os | sed -e 's|sunos6|solaris3|'`
+		;;
+	-opened*)
+		os=-openedition
+		;;
+        -os400*)
+		os=-os400
+		;;
+	-wince*)
+		os=-wince
+		;;
+	-osfrose*)
+		os=-osfrose
+		;;
+	-osf*)
+		os=-osf
+		;;
+	-utek*)
+		os=-bsd
+		;;
+	-dynix*)
+		os=-bsd
+		;;
+	-acis*)
+		os=-aos
+		;;
+	-atheos*)
+		os=-atheos
+		;;
+	-syllable*)
+		os=-syllable
+		;;
+	-386bsd)
+		os=-bsd
+		;;
+	-ctix* | -uts*)
+		os=-sysv
+		;;
+	-nova*)
+		os=-rtmk-nova
+		;;
+	-ns2 )
+		os=-nextstep2
+		;;
+	-nsk*)
+		os=-nsk
+		;;
+	# Preserve the version number of sinix5.
+	-sinix5.*)
+		os=`echo $os | sed -e 's|sinix|sysv|'`
+		;;
+	-sinix*)
+		os=-sysv4
+		;;
+        -tpf*)
+		os=-tpf
+		;;
+	-triton*)
+		os=-sysv3
+		;;
+	-oss*)
+		os=-sysv3
+		;;
+	-svr4)
+		os=-sysv4
+		;;
+	-svr3)
+		os=-sysv3
+		;;
+	-sysvr4)
+		os=-sysv4
+		;;
+	# This must come after -sysvr4.
+	-sysv*)
+		;;
+	-ose*)
+		os=-ose
+		;;
+	-es1800*)
+		os=-ose
+		;;
+	-xenix)
+		os=-xenix
+		;;
+	-*mint | -mint[0-9]* | -*MiNT | -MiNT[0-9]*)
+		os=-mint
+		;;
+	-aros*)
+		os=-aros
+		;;
+	-kaos*)
+		os=-kaos
+		;;
+	-zvmoe)
+		os=-zvmoe
+		;;
+	-none)
+		;;
+	*)
+		# Get rid of the `-' at the beginning of $os.
+		os=`echo $os | sed 's/[^-]*-//'`
+		echo Invalid configuration \`$1\': system \`$os\' not recognized 1>&2
+		exit 1
+		;;
+esac
+else
+
+# Here we handle the default operating systems that come with various machines.
+# The value should be what the vendor currently ships out the door with their
+# machine or put another way, the most popular os provided with the machine.
+
+# Note that if you're going to try to match "-MANUFACTURER" here (say,
+# "-sun"), then you have to tell the case statement up towards the top
+# that MANUFACTURER isn't an operating system.  Otherwise, code above
+# will signal an error saying that MANUFACTURER isn't an operating
+# system, and we'll never get to this point.
+
+case $basic_machine in
+        score-*)
+		os=-elf
+		;;
+        spu-*)
+		os=-elf
+		;;
+	*-acorn)
+		os=-riscix1.2
+		;;
+	arm*-rebel)
+		os=-linux
+		;;
+	arm*-semi)
+		os=-aout
+		;;
+        c4x-* | tic4x-*)
+        	os=-coff
+		;;
+	# This must come before the *-dec entry.
+	pdp10-*)
+		os=-tops20
+		;;
+	pdp11-*)
+		os=-none
+		;;
+	*-dec | vax-*)
+		os=-ultrix4.2
+		;;
+	m68*-apollo)
+		os=-domain
+		;;
+	i386-sun)
+		os=-sunos4.0.2
+		;;
+	m68000-sun)
+		os=-sunos3
+		# This also exists in the configure program, but was not the
+		# default.
+		# os=-sunos4
+		;;
+	m68*-cisco)
+		os=-aout
+		;;
+        mep-*)
+		os=-elf
+		;;
+	mips*-cisco)
+		os=-elf
+		;;
+	mips*-*)
+		os=-elf
+		;;
+	or32-*)
+		os=-coff
+		;;
+	*-tti)	# must be before sparc entry or we get the wrong os.
+		os=-sysv3
+		;;
+	sparc-* | *-sun)
+		os=-sunos4.1.1
+		;;
+	*-be)
+		os=-beos
+		;;
+	*-haiku)
+		os=-haiku
+		;;
+	*-ibm)
+		os=-aix
+		;;
+    	*-knuth)
+		os=-mmixware
+		;;
+	*-wec)
+		os=-proelf
+		;;
+	*-winbond)
+		os=-proelf
+		;;
+	*-oki)
+		os=-proelf
+		;;
+	*-hp)
+		os=-hpux
+		;;
+	*-hitachi)
+		os=-hiux
+		;;
+	i860-* | *-att | *-ncr | *-altos | *-motorola | *-convergent)
+		os=-sysv
+		;;
+	*-cbm)
+		os=-amigaos
+		;;
+	*-dg)
+		os=-dgux
+		;;
+	*-dolphin)
+		os=-sysv3
+		;;
+	m68k-ccur)
+		os=-rtu
+		;;
+	m88k-omron*)
+		os=-luna
+		;;
+	*-next )
+		os=-nextstep
+		;;
+	*-sequent)
+		os=-ptx
+		;;
+	*-crds)
+		os=-unos
+		;;
+	*-ns)
+		os=-genix
+		;;
+	i370-*)
+		os=-mvs
+		;;
+	*-next)
+		os=-nextstep3
+		;;
+	*-gould)
+		os=-sysv
+		;;
+	*-highlevel)
+		os=-bsd
+		;;
+	*-encore)
+		os=-bsd
+		;;
+	*-sgi)
+		os=-irix
+		;;
+	*-siemens)
+		os=-sysv4
+		;;
+	*-masscomp)
+		os=-rtu
+		;;
+	f30[01]-fujitsu | f700-fujitsu)
+		os=-uxpv
+		;;
+	*-rom68k)
+		os=-coff
+		;;
+	*-*bug)
+		os=-coff
+		;;
+	*-apple)
+		os=-macos
+		;;
+	*-atari*)
+		os=-mint
+		;;
+	*)
+		os=-none
+		;;
+esac
+fi
+
+# Here we handle the case where we know the os, and the CPU type, but not the
+# manufacturer.  We pick the logical manufacturer.
+vendor=unknown
+case $basic_machine in
+	*-unknown)
+		case $os in
+			-riscix*)
+				vendor=acorn
+				;;
+			-sunos*)
+				vendor=sun
+				;;
+			-aix*)
+				vendor=ibm
+				;;
+			-beos*)
+				vendor=be
+				;;
+			-hpux*)
+				vendor=hp
+				;;
+			-mpeix*)
+				vendor=hp
+				;;
+			-hiux*)
+				vendor=hitachi
+				;;
+			-unos*)
+				vendor=crds
+				;;
+			-dgux*)
+				vendor=dg
+				;;
+			-luna*)
+				vendor=omron
+				;;
+			-genix*)
+				vendor=ns
+				;;
+			-mvs* | -opened*)
+				vendor=ibm
+				;;
+			-os400*)
+				vendor=ibm
+				;;
+			-ptx*)
+				vendor=sequent
+				;;
+			-tpf*)
+				vendor=ibm
+				;;
+			-vxsim* | -vxworks* | -windiss*)
+				vendor=wrs
+				;;
+			-aux*)
+				vendor=apple
+				;;
+			-hms*)
+				vendor=hitachi
+				;;
+			-mpw* | -macos*)
+				vendor=apple
+				;;
+			-*mint | -mint[0-9]* | -*MiNT | -MiNT[0-9]*)
+				vendor=atari
+				;;
+			-vos*)
+				vendor=stratus
+				;;
+		esac
+		basic_machine=`echo $basic_machine | sed "s/unknown/$vendor/"`
+		;;
+esac
+
+echo $basic_machine$os
+exit
+
+# Local variables:
+# eval: (add-hook 'write-file-hooks 'time-stamp)
+# time-stamp-start: "timestamp='"
+# time-stamp-format: "%:y-%02m-%02d"
+# time-stamp-end: "'"
+# End:
diff --git a/configure b/configure
new file mode 100755
index 0000000000..30ecab469a
--- /dev/null
+++ b/configure
@@ -0,0 +1,8127 @@
+#! /bin/sh
+# Guess values for system-dependent variables and create Makefiles.
+# Generated by GNU Autoconf 2.69 for ccminer 2014.03.18.
+#
+#
+# Copyright (C) 1992-1996, 1998-2012 Free Software Foundation, Inc.
+#
+#
+# This configure script is free software; the Free Software Foundation
+# gives unlimited permission to copy, distribute and modify it.
+## -------------------- ##
+## M4sh Initialization. ##
+## -------------------- ##
+
+# Be more Bourne compatible
+DUALCASE=1; export DUALCASE # for MKS sh
+if test -n "${ZSH_VERSION+set}" && (emulate sh) >/dev/null 2>&1; then :
+  emulate sh
+  NULLCMD=:
+  # Pre-4.2 versions of Zsh do word splitting on ${1+"$@"}, which
+  # is contrary to our usage.  Disable this feature.
+  alias -g '${1+"$@"}'='"$@"'
+  setopt NO_GLOB_SUBST
+else
+  case `(set -o) 2>/dev/null` in #(
+  *posix*) :
+    set -o posix ;; #(
+  *) :
+     ;;
+esac
+fi
+
+
+as_nl='
+'
+export as_nl
+# Printing a long string crashes Solaris 7 /usr/bin/printf.
+as_echo='\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\'
+as_echo=$as_echo$as_echo$as_echo$as_echo$as_echo
+as_echo=$as_echo$as_echo$as_echo$as_echo$as_echo$as_echo
+# Prefer a ksh shell builtin over an external printf program on Solaris,
+# but without wasting forks for bash or zsh.
+if test -z "$BASH_VERSION$ZSH_VERSION" \
+    && (test "X`print -r -- $as_echo`" = "X$as_echo") 2>/dev/null; then
+  as_echo='print -r --'
+  as_echo_n='print -rn --'
+elif (test "X`printf %s $as_echo`" = "X$as_echo") 2>/dev/null; then
+  as_echo='printf %s\n'
+  as_echo_n='printf %s'
+else
+  if test "X`(/usr/ucb/echo -n -n $as_echo) 2>/dev/null`" = "X-n $as_echo"; then
+    as_echo_body='eval /usr/ucb/echo -n "$1$as_nl"'
+    as_echo_n='/usr/ucb/echo -n'
+  else
+    as_echo_body='eval expr "X$1" : "X\\(.*\\)"'
+    as_echo_n_body='eval
+      arg=$1;
+      case $arg in #(
+      *"$as_nl"*)
+	expr "X$arg" : "X\\(.*\\)$as_nl";
+	arg=`expr "X$arg" : ".*$as_nl\\(.*\\)"`;;
+      esac;
+      expr "X$arg" : "X\\(.*\\)" | tr -d "$as_nl"
+    '
+    export as_echo_n_body
+    as_echo_n='sh -c $as_echo_n_body as_echo'
+  fi
+  export as_echo_body
+  as_echo='sh -c $as_echo_body as_echo'
+fi
+
+# The user is always right.
+if test "${PATH_SEPARATOR+set}" != set; then
+  PATH_SEPARATOR=:
+  (PATH='/bin;/bin'; FPATH=$PATH; sh -c :) >/dev/null 2>&1 && {
+    (PATH='/bin:/bin'; FPATH=$PATH; sh -c :) >/dev/null 2>&1 ||
+      PATH_SEPARATOR=';'
+  }
+fi
+
+
+# IFS
+# We need space, tab and new line, in precisely that order.  Quoting is
+# there to prevent editors from complaining about space-tab.
+# (If _AS_PATH_WALK were called with IFS unset, it would disable word
+# splitting by setting IFS to empty value.)
+IFS=" ""	$as_nl"
+
+# Find who we are.  Look in the path if we contain no directory separator.
+as_myself=
+case $0 in #((
+  *[\\/]* ) as_myself=$0 ;;
+  *) as_save_IFS=$IFS; IFS=$PATH_SEPARATOR
+for as_dir in $PATH
+do
+  IFS=$as_save_IFS
+  test -z "$as_dir" && as_dir=.
+    test -r "$as_dir/$0" && as_myself=$as_dir/$0 && break
+  done
+IFS=$as_save_IFS
+
+     ;;
+esac
+# We did not find ourselves, most probably we were run as `sh COMMAND'
+# in which case we are not to be found in the path.
+if test "x$as_myself" = x; then
+  as_myself=$0
+fi
+if test ! -f "$as_myself"; then
+  $as_echo "$as_myself: error: cannot find myself; rerun with an absolute file name" >&2
+  exit 1
+fi
+
+# Unset variables that we do not need and which cause bugs (e.g. in
+# pre-3.0 UWIN ksh).  But do not cause bugs in bash 2.01; the "|| exit 1"
+# suppresses any "Segmentation fault" message there.  '((' could
+# trigger a bug in pdksh 5.2.14.
+for as_var in BASH_ENV ENV MAIL MAILPATH
+do eval test x\${$as_var+set} = xset \
+  && ( (unset $as_var) || exit 1) >/dev/null 2>&1 && unset $as_var || :
+done
+PS1='$ '
+PS2='> '
+PS4='+ '
+
+# NLS nuisances.
+LC_ALL=C
+export LC_ALL
+LANGUAGE=C
+export LANGUAGE
+
+# CDPATH.
+(unset CDPATH) >/dev/null 2>&1 && unset CDPATH
+
+# Use a proper internal environment variable to ensure we don't fall
+  # into an infinite loop, continuously re-executing ourselves.
+  if test x"${_as_can_reexec}" != xno && test "x$CONFIG_SHELL" != x; then
+    _as_can_reexec=no; export _as_can_reexec;
+    # We cannot yet assume a decent shell, so we have to provide a
+# neutralization value for shells without unset; and this also
+# works around shells that cannot unset nonexistent variables.
+# Preserve -v and -x to the replacement shell.
+BASH_ENV=/dev/null
+ENV=/dev/null
+(unset BASH_ENV) >/dev/null 2>&1 && unset BASH_ENV ENV
+case $- in # ((((
+  *v*x* | *x*v* ) as_opts=-vx ;;
+  *v* ) as_opts=-v ;;
+  *x* ) as_opts=-x ;;
+  * ) as_opts= ;;
+esac
+exec $CONFIG_SHELL $as_opts "$as_myself" ${1+"$@"}
+# Admittedly, this is quite paranoid, since all the known shells bail
+# out after a failed `exec'.
+$as_echo "$0: could not re-execute with $CONFIG_SHELL" >&2
+as_fn_exit 255
+  fi
+  # We don't want this to propagate to other subprocesses.
+          { _as_can_reexec=; unset _as_can_reexec;}
+if test "x$CONFIG_SHELL" = x; then
+  as_bourne_compatible="if test -n \"\${ZSH_VERSION+set}\" && (emulate sh) >/dev/null 2>&1; then :
+  emulate sh
+  NULLCMD=:
+  # Pre-4.2 versions of Zsh do word splitting on \${1+\"\$@\"}, which
+  # is contrary to our usage.  Disable this feature.
+  alias -g '\${1+\"\$@\"}'='\"\$@\"'
+  setopt NO_GLOB_SUBST
+else
+  case \`(set -o) 2>/dev/null\` in #(
+  *posix*) :
+    set -o posix ;; #(
+  *) :
+     ;;
+esac
+fi
+"
+  as_required="as_fn_return () { (exit \$1); }
+as_fn_success () { as_fn_return 0; }
+as_fn_failure () { as_fn_return 1; }
+as_fn_ret_success () { return 0; }
+as_fn_ret_failure () { return 1; }
+
+exitcode=0
+as_fn_success || { exitcode=1; echo as_fn_success failed.; }
+as_fn_failure && { exitcode=1; echo as_fn_failure succeeded.; }
+as_fn_ret_success || { exitcode=1; echo as_fn_ret_success failed.; }
+as_fn_ret_failure && { exitcode=1; echo as_fn_ret_failure succeeded.; }
+if ( set x; as_fn_ret_success y && test x = \"\$1\" ); then :
+
+else
+  exitcode=1; echo positional parameters were not saved.
+fi
+test x\$exitcode = x0 || exit 1
+test -x / || exit 1"
+  as_suggested="  as_lineno_1=";as_suggested=$as_suggested$LINENO;as_suggested=$as_suggested" as_lineno_1a=\$LINENO
+  as_lineno_2=";as_suggested=$as_suggested$LINENO;as_suggested=$as_suggested" as_lineno_2a=\$LINENO
+  eval 'test \"x\$as_lineno_1'\$as_run'\" != \"x\$as_lineno_2'\$as_run'\" &&
+  test \"x\`expr \$as_lineno_1'\$as_run' + 1\`\" = \"x\$as_lineno_2'\$as_run'\"' || exit 1
+test \$(( 1 + 1 )) = 2 || exit 1"
+  if (eval "$as_required") 2>/dev/null; then :
+  as_have_required=yes
+else
+  as_have_required=no
+fi
+  if test x$as_have_required = xyes && (eval "$as_suggested") 2>/dev/null; then :
+
+else
+  as_save_IFS=$IFS; IFS=$PATH_SEPARATOR
+as_found=false
+for as_dir in /bin$PATH_SEPARATOR/usr/bin$PATH_SEPARATOR$PATH
+do
+  IFS=$as_save_IFS
+  test -z "$as_dir" && as_dir=.
+  as_found=:
+  case $as_dir in #(
+	 /*)
+	   for as_base in sh bash ksh sh5; do
+	     # Try only shells that exist, to save several forks.
+	     as_shell=$as_dir/$as_base
+	     if { test -f "$as_shell" || test -f "$as_shell.exe"; } &&
+		    { $as_echo "$as_bourne_compatible""$as_required" | as_run=a "$as_shell"; } 2>/dev/null; then :
+  CONFIG_SHELL=$as_shell as_have_required=yes
+		   if { $as_echo "$as_bourne_compatible""$as_suggested" | as_run=a "$as_shell"; } 2>/dev/null; then :
+  break 2
+fi
+fi
+	   done;;
+       esac
+  as_found=false
+done
+$as_found || { if { test -f "$SHELL" || test -f "$SHELL.exe"; } &&
+	      { $as_echo "$as_bourne_compatible""$as_required" | as_run=a "$SHELL"; } 2>/dev/null; then :
+  CONFIG_SHELL=$SHELL as_have_required=yes
+fi; }
+IFS=$as_save_IFS
+
+
+      if test "x$CONFIG_SHELL" != x; then :
+  export CONFIG_SHELL
+             # We cannot yet assume a decent shell, so we have to provide a
+# neutralization value for shells without unset; and this also
+# works around shells that cannot unset nonexistent variables.
+# Preserve -v and -x to the replacement shell.
+BASH_ENV=/dev/null
+ENV=/dev/null
+(unset BASH_ENV) >/dev/null 2>&1 && unset BASH_ENV ENV
+case $- in # ((((
+  *v*x* | *x*v* ) as_opts=-vx ;;
+  *v* ) as_opts=-v ;;
+  *x* ) as_opts=-x ;;
+  * ) as_opts= ;;
+esac
+exec $CONFIG_SHELL $as_opts "$as_myself" ${1+"$@"}
+# Admittedly, this is quite paranoid, since all the known shells bail
+# out after a failed `exec'.
+$as_echo "$0: could not re-execute with $CONFIG_SHELL" >&2
+exit 255
+fi
+
+    if test x$as_have_required = xno; then :
+  $as_echo "$0: This script requires a shell more modern than all"
+  $as_echo "$0: the shells that I found on your system."
+  if test x${ZSH_VERSION+set} = xset ; then
+    $as_echo "$0: In particular, zsh $ZSH_VERSION has bugs and should"
+    $as_echo "$0: be upgraded to zsh 4.3.4 or later."
+  else
+    $as_echo "$0: Please tell bug-autoconf@gnu.org about your system,
+$0: including any error possibly output before this
+$0: message. Then install a modern shell, or manually run
+$0: the script under such a shell if you do have one."
+  fi
+  exit 1
+fi
+fi
+fi
+SHELL=${CONFIG_SHELL-/bin/sh}
+export SHELL
+# Unset more variables known to interfere with behavior of common tools.
+CLICOLOR_FORCE= GREP_OPTIONS=
+unset CLICOLOR_FORCE GREP_OPTIONS
+
+## --------------------- ##
+## M4sh Shell Functions. ##
+## --------------------- ##
+# as_fn_unset VAR
+# ---------------
+# Portably unset VAR.
+as_fn_unset ()
+{
+  { eval $1=; unset $1;}
+}
+as_unset=as_fn_unset
+
+# as_fn_set_status STATUS
+# -----------------------
+# Set $? to STATUS, without forking.
+as_fn_set_status ()
+{
+  return $1
+} # as_fn_set_status
+
+# as_fn_exit STATUS
+# -----------------
+# Exit the shell with STATUS, even in a "trap 0" or "set -e" context.
+as_fn_exit ()
+{
+  set +e
+  as_fn_set_status $1
+  exit $1
+} # as_fn_exit
+
+# as_fn_mkdir_p
+# -------------
+# Create "$as_dir" as a directory, including parents if necessary.
+as_fn_mkdir_p ()
+{
+
+  case $as_dir in #(
+  -*) as_dir=./$as_dir;;
+  esac
+  test -d "$as_dir" || eval $as_mkdir_p || {
+    as_dirs=
+    while :; do
+      case $as_dir in #(
+      *\'*) as_qdir=`$as_echo "$as_dir" | sed "s/'/'\\\\\\\\''/g"`;; #'(
+      *) as_qdir=$as_dir;;
+      esac
+      as_dirs="'$as_qdir' $as_dirs"
+      as_dir=`$as_dirname -- "$as_dir" ||
+$as_expr X"$as_dir" : 'X\(.*[^/]\)//*[^/][^/]*/*$' \| \
+	 X"$as_dir" : 'X\(//\)[^/]' \| \
+	 X"$as_dir" : 'X\(//\)$' \| \
+	 X"$as_dir" : 'X\(/\)' \| . 2>/dev/null ||
+$as_echo X"$as_dir" |
+    sed '/^X\(.*[^/]\)\/\/*[^/][^/]*\/*$/{
+	    s//\1/
+	    q
+	  }
+	  /^X\(\/\/\)[^/].*/{
+	    s//\1/
+	    q
+	  }
+	  /^X\(\/\/\)$/{
+	    s//\1/
+	    q
+	  }
+	  /^X\(\/\).*/{
+	    s//\1/
+	    q
+	  }
+	  s/.*/./; q'`
+      test -d "$as_dir" && break
+    done
+    test -z "$as_dirs" || eval "mkdir $as_dirs"
+  } || test -d "$as_dir" || as_fn_error $? "cannot create directory $as_dir"
+
+
+} # as_fn_mkdir_p
+
+# as_fn_executable_p FILE
+# -----------------------
+# Test if FILE is an executable regular file.
+as_fn_executable_p ()
+{
+  test -f "$1" && test -x "$1"
+} # as_fn_executable_p
+# as_fn_append VAR VALUE
+# ----------------------
+# Append the text in VALUE to the end of the definition contained in VAR. Take
+# advantage of any shell optimizations that allow amortized linear growth over
+# repeated appends, instead of the typical quadratic growth present in naive
+# implementations.
+if (eval "as_var=1; as_var+=2; test x\$as_var = x12") 2>/dev/null; then :
+  eval 'as_fn_append ()
+  {
+    eval $1+=\$2
+  }'
+else
+  as_fn_append ()
+  {
+    eval $1=\$$1\$2
+  }
+fi # as_fn_append
+
+# as_fn_arith ARG...
+# ------------------
+# Perform arithmetic evaluation on the ARGs, and store the result in the
+# global $as_val. Take advantage of shells that can avoid forks. The arguments
+# must be portable across $(()) and expr.
+if (eval "test \$(( 1 + 1 )) = 2") 2>/dev/null; then :
+  eval 'as_fn_arith ()
+  {
+    as_val=$(( $* ))
+  }'
+else
+  as_fn_arith ()
+  {
+    as_val=`expr "$@" || test $? -eq 1`
+  }
+fi # as_fn_arith
+
+
+# as_fn_error STATUS ERROR [LINENO LOG_FD]
+# ----------------------------------------
+# Output "`basename $0`: error: ERROR" to stderr. If LINENO and LOG_FD are
+# provided, also output the error to LOG_FD, referencing LINENO. Then exit the
+# script with STATUS, using 1 if that was 0.
+as_fn_error ()
+{
+  as_status=$1; test $as_status -eq 0 && as_status=1
+  if test "$4"; then
+    as_lineno=${as_lineno-"$3"} as_lineno_stack=as_lineno_stack=$as_lineno_stack
+    $as_echo "$as_me:${as_lineno-$LINENO}: error: $2" >&$4
+  fi
+  $as_echo "$as_me: error: $2" >&2
+  as_fn_exit $as_status
+} # as_fn_error
+
+if expr a : '\(a\)' >/dev/null 2>&1 &&
+   test "X`expr 00001 : '.*\(...\)'`" = X001; then
+  as_expr=expr
+else
+  as_expr=false
+fi
+
+if (basename -- /) >/dev/null 2>&1 && test "X`basename -- / 2>&1`" = "X/"; then
+  as_basename=basename
+else
+  as_basename=false
+fi
+
+if (as_dir=`dirname -- /` && test "X$as_dir" = X/) >/dev/null 2>&1; then
+  as_dirname=dirname
+else
+  as_dirname=false
+fi
+
+as_me=`$as_basename -- "$0" ||
+$as_expr X/"$0" : '.*/\([^/][^/]*\)/*$' \| \
+	 X"$0" : 'X\(//\)$' \| \
+	 X"$0" : 'X\(/\)' \| . 2>/dev/null ||
+$as_echo X/"$0" |
+    sed '/^.*\/\([^/][^/]*\)\/*$/{
+	    s//\1/
+	    q
+	  }
+	  /^X\/\(\/\/\)$/{
+	    s//\1/
+	    q
+	  }
+	  /^X\/\(\/\).*/{
+	    s//\1/
+	    q
+	  }
+	  s/.*/./; q'`
+
+# Avoid depending upon Character Ranges.
+as_cr_letters='abcdefghijklmnopqrstuvwxyz'
+as_cr_LETTERS='ABCDEFGHIJKLMNOPQRSTUVWXYZ'
+as_cr_Letters=$as_cr_letters$as_cr_LETTERS
+as_cr_digits='0123456789'
+as_cr_alnum=$as_cr_Letters$as_cr_digits
+
+
+  as_lineno_1=$LINENO as_lineno_1a=$LINENO
+  as_lineno_2=$LINENO as_lineno_2a=$LINENO
+  eval 'test "x$as_lineno_1'$as_run'" != "x$as_lineno_2'$as_run'" &&
+  test "x`expr $as_lineno_1'$as_run' + 1`" = "x$as_lineno_2'$as_run'"' || {
+  # Blame Lee E. McMahon (1931-1989) for sed's syntax.  :-)
+  sed -n '
+    p
+    /[$]LINENO/=
+  ' <$as_myself |
+    sed '
+      s/[$]LINENO.*/&-/
+      t lineno
+      b
+      :lineno
+      N
+      :loop
+      s/[$]LINENO\([^'$as_cr_alnum'_].*\n\)\(.*\)/\2\1\2/
+      t loop
+      s/-\n.*//
+    ' >$as_me.lineno &&
+  chmod +x "$as_me.lineno" ||
+    { $as_echo "$as_me: error: cannot create $as_me.lineno; rerun with a POSIX shell" >&2; as_fn_exit 1; }
+
+  # If we had to re-execute with $CONFIG_SHELL, we're ensured to have
+  # already done that, so ensure we don't try to do so again and fall
+  # in an infinite loop.  This has already happened in practice.
+  _as_can_reexec=no; export _as_can_reexec
+  # Don't try to exec as it changes $[0], causing all sort of problems
+  # (the dirname of $[0] is not the place where we might find the
+  # original and so on.  Autoconf is especially sensitive to this).
+  . "./$as_me.lineno"
+  # Exit status is that of the last command.
+  exit
+}
+
+ECHO_C= ECHO_N= ECHO_T=
+case `echo -n x` in #(((((
+-n*)
+  case `echo 'xy\c'` in
+  *c*) ECHO_T='	';;	# ECHO_T is single tab character.
+  xy)  ECHO_C='\c';;
+  *)   echo `echo ksh88 bug on AIX 6.1` > /dev/null
+       ECHO_T='	';;
+  esac;;
+*)
+  ECHO_N='-n';;
+esac
+
+rm -f conf$$ conf$$.exe conf$$.file
+if test -d conf$$.dir; then
+  rm -f conf$$.dir/conf$$.file
+else
+  rm -f conf$$.dir
+  mkdir conf$$.dir 2>/dev/null
+fi
+if (echo >conf$$.file) 2>/dev/null; then
+  if ln -s conf$$.file conf$$ 2>/dev/null; then
+    as_ln_s='ln -s'
+    # ... but there are two gotchas:
+    # 1) On MSYS, both `ln -s file dir' and `ln file dir' fail.
+    # 2) DJGPP < 2.04 has no symlinks; `ln -s' creates a wrapper executable.
+    # In both cases, we have to default to `cp -pR'.
+    ln -s conf$$.file conf$$.dir 2>/dev/null && test ! -f conf$$.exe ||
+      as_ln_s='cp -pR'
+  elif ln conf$$.file conf$$ 2>/dev/null; then
+    as_ln_s=ln
+  else
+    as_ln_s='cp -pR'
+  fi
+else
+  as_ln_s='cp -pR'
+fi
+rm -f conf$$ conf$$.exe conf$$.dir/conf$$.file conf$$.file
+rmdir conf$$.dir 2>/dev/null
+
+if mkdir -p . 2>/dev/null; then
+  as_mkdir_p='mkdir -p "$as_dir"'
+else
+  test -d ./-p && rmdir ./-p
+  as_mkdir_p=false
+fi
+
+as_test_x='test -x'
+as_executable_p=as_fn_executable_p
+
+# Sed expression to map a string onto a valid CPP name.
+as_tr_cpp="eval sed 'y%*$as_cr_letters%P$as_cr_LETTERS%;s%[^_$as_cr_alnum]%_%g'"
+
+# Sed expression to map a string onto a valid variable name.
+as_tr_sh="eval sed 'y%*+%pp%;s%[^_$as_cr_alnum]%_%g'"
+
+
+test -n "$DJDIR" || exec 7<&0 </dev/null
+exec 6>&1
+
+# Name of the host.
+# hostname on some systems (SVR3.2, old GNU/Linux) returns a bogus exit status,
+# so uname gets run too.
+ac_hostname=`(hostname || uname -n) 2>/dev/null | sed 1q`
+
+#
+# Initializations.
+#
+ac_default_prefix=/usr/local
+ac_clean_files=
+ac_config_libobj_dir=.
+LIBOBJS=
+cross_compiling=no
+subdirs=
+MFLAGS=
+MAKEFLAGS=
+
+# Identity of this package.
+PACKAGE_NAME='ccminer'
+PACKAGE_TARNAME='ccminer'
+PACKAGE_VERSION='2014.03.18'
+PACKAGE_STRING='ccminer 2014.03.18'
+PACKAGE_BUGREPORT=''
+PACKAGE_URL=''
+
+ac_unique_file="cpu-miner.c"
+# Factoring default headers for most tests.
+ac_includes_default="\
+#include <stdio.h>
+#ifdef HAVE_SYS_TYPES_H
+# include <sys/types.h>
+#endif
+#ifdef HAVE_SYS_STAT_H
+# include <sys/stat.h>
+#endif
+#ifdef STDC_HEADERS
+# include <stdlib.h>
+# include <stddef.h>
+#else
+# ifdef HAVE_STDLIB_H
+#  include <stdlib.h>
+# endif
+#endif
+#ifdef HAVE_STRING_H
+# if !defined STDC_HEADERS && defined HAVE_MEMORY_H
+#  include <memory.h>
+# endif
+# include <string.h>
+#endif
+#ifdef HAVE_STRINGS_H
+# include <strings.h>
+#endif
+#ifdef HAVE_INTTYPES_H
+# include <inttypes.h>
+#endif
+#ifdef HAVE_STDINT_H
+# include <stdint.h>
+#endif
+#ifdef HAVE_UNISTD_H
+# include <unistd.h>
+#endif"
+
+ac_subst_vars='am__EXEEXT_FALSE
+am__EXEEXT_TRUE
+LTLIBOBJS
+LIBOBJS
+NVCC
+CUDA_LDFLAGS
+CUDA_LIBS
+CUDA_CFLAGS
+WS2_LIBS
+PTHREAD_LIBS
+PTHREAD_FLAGS
+JANSSON_LIBS
+LIBCURL
+LIBCURL_CPPFLAGS
+_libcurl_config
+ARCH_x86_64_FALSE
+ARCH_x86_64_TRUE
+ARCH_x86_FALSE
+ARCH_x86_TRUE
+HAVE_WINDOWS_FALSE
+HAVE_WINDOWS_TRUE
+WANT_JANSSON_FALSE
+WANT_JANSSON_TRUE
+ALLOCA
+OPENMP_CFLAGS
+am__fastdepCXX_FALSE
+am__fastdepCXX_TRUE
+CXXDEPMODE
+ac_ct_CXX
+CXXFLAGS
+CXX
+RANLIB
+am__fastdepCCAS_FALSE
+am__fastdepCCAS_TRUE
+CCASDEPMODE
+CCASFLAGS
+CCAS
+EGREP
+GREP
+CPP
+am__fastdepCC_FALSE
+am__fastdepCC_TRUE
+CCDEPMODE
+am__nodep
+AMDEPBACKSLASH
+AMDEP_FALSE
+AMDEP_TRUE
+am__quote
+am__include
+DEPDIR
+OBJEXT
+EXEEXT
+ac_ct_CC
+CPPFLAGS
+LDFLAGS
+CFLAGS
+CC
+MAINT
+MAINTAINER_MODE_FALSE
+MAINTAINER_MODE_TRUE
+AM_BACKSLASH
+AM_DEFAULT_VERBOSITY
+AM_DEFAULT_V
+AM_V
+am__untar
+am__tar
+AMTAR
+am__leading_dot
+SET_MAKE
+AWK
+mkdir_p
+MKDIR_P
+INSTALL_STRIP_PROGRAM
+STRIP
+install_sh
+MAKEINFO
+AUTOHEADER
+AUTOMAKE
+AUTOCONF
+ACLOCAL
+VERSION
+PACKAGE
+CYGPATH_W
+am__isrc
+INSTALL_DATA
+INSTALL_SCRIPT
+INSTALL_PROGRAM
+target_os
+target_vendor
+target_cpu
+target
+host_os
+host_vendor
+host_cpu
+host
+build_os
+build_vendor
+build_cpu
+build
+target_alias
+host_alias
+build_alias
+LIBS
+ECHO_T
+ECHO_N
+ECHO_C
+DEFS
+mandir
+localedir
+libdir
+psdir
+pdfdir
+dvidir
+htmldir
+infodir
+docdir
+oldincludedir
+includedir
+localstatedir
+sharedstatedir
+sysconfdir
+datadir
+datarootdir
+libexecdir
+sbindir
+bindir
+program_transform_name
+prefix
+exec_prefix
+PACKAGE_URL
+PACKAGE_BUGREPORT
+PACKAGE_STRING
+PACKAGE_VERSION
+PACKAGE_TARNAME
+PACKAGE_NAME
+PATH_SEPARATOR
+SHELL'
+ac_subst_files=''
+ac_user_opts='
+enable_option_checking
+enable_silent_rules
+enable_maintainer_mode
+enable_dependency_tracking
+enable_openmp
+with_libcurl
+with_cuda
+'
+      ac_precious_vars='build_alias
+host_alias
+target_alias
+CC
+CFLAGS
+LDFLAGS
+LIBS
+CPPFLAGS
+CPP
+CCAS
+CCASFLAGS
+CXX
+CXXFLAGS
+CCC'
+
+
+# Initialize some variables set by options.
+ac_init_help=
+ac_init_version=false
+ac_unrecognized_opts=
+ac_unrecognized_sep=
+# The variables have the same names as the options, with
+# dashes changed to underlines.
+cache_file=/dev/null
+exec_prefix=NONE
+no_create=
+no_recursion=
+prefix=NONE
+program_prefix=NONE
+program_suffix=NONE
+program_transform_name=s,x,x,
+silent=
+site=
+srcdir=
+verbose=
+x_includes=NONE
+x_libraries=NONE
+
+# Installation directory options.
+# These are left unexpanded so users can "make install exec_prefix=/foo"
+# and all the variables that are supposed to be based on exec_prefix
+# by default will actually change.
+# Use braces instead of parens because sh, perl, etc. also accept them.
+# (The list follows the same order as the GNU Coding Standards.)
+bindir='${exec_prefix}/bin'
+sbindir='${exec_prefix}/sbin'
+libexecdir='${exec_prefix}/libexec'
+datarootdir='${prefix}/share'
+datadir='${datarootdir}'
+sysconfdir='${prefix}/etc'
+sharedstatedir='${prefix}/com'
+localstatedir='${prefix}/var'
+includedir='${prefix}/include'
+oldincludedir='/usr/include'
+docdir='${datarootdir}/doc/${PACKAGE_TARNAME}'
+infodir='${datarootdir}/info'
+htmldir='${docdir}'
+dvidir='${docdir}'
+pdfdir='${docdir}'
+psdir='${docdir}'
+libdir='${exec_prefix}/lib'
+localedir='${datarootdir}/locale'
+mandir='${datarootdir}/man'
+
+ac_prev=
+ac_dashdash=
+for ac_option
+do
+  # If the previous option needs an argument, assign it.
+  if test -n "$ac_prev"; then
+    eval $ac_prev=\$ac_option
+    ac_prev=
+    continue
+  fi
+
+  case $ac_option in
+  *=?*) ac_optarg=`expr "X$ac_option" : '[^=]*=\(.*\)'` ;;
+  *=)   ac_optarg= ;;
+  *)    ac_optarg=yes ;;
+  esac
+
+  # Accept the important Cygnus configure options, so we can diagnose typos.
+
+  case $ac_dashdash$ac_option in
+  --)
+    ac_dashdash=yes ;;
+
+  -bindir | --bindir | --bindi | --bind | --bin | --bi)
+    ac_prev=bindir ;;
+  -bindir=* | --bindir=* | --bindi=* | --bind=* | --bin=* | --bi=*)
+    bindir=$ac_optarg ;;
+
+  -build | --build | --buil | --bui | --bu)
+    ac_prev=build_alias ;;
+  -build=* | --build=* | --buil=* | --bui=* | --bu=*)
+    build_alias=$ac_optarg ;;
+
+  -cache-file | --cache-file | --cache-fil | --cache-fi \
+  | --cache-f | --cache- | --cache | --cach | --cac | --ca | --c)
+    ac_prev=cache_file ;;
+  -cache-file=* | --cache-file=* | --cache-fil=* | --cache-fi=* \
+  | --cache-f=* | --cache-=* | --cache=* | --cach=* | --cac=* | --ca=* | --c=*)
+    cache_file=$ac_optarg ;;
+
+  --config-cache | -C)
+    cache_file=config.cache ;;
+
+  -datadir | --datadir | --datadi | --datad)
+    ac_prev=datadir ;;
+  -datadir=* | --datadir=* | --datadi=* | --datad=*)
+    datadir=$ac_optarg ;;
+
+  -datarootdir | --datarootdir | --datarootdi | --datarootd | --dataroot \
+  | --dataroo | --dataro | --datar)
+    ac_prev=datarootdir ;;
+  -datarootdir=* | --datarootdir=* | --datarootdi=* | --datarootd=* \
+  | --dataroot=* | --dataroo=* | --dataro=* | --datar=*)
+    datarootdir=$ac_optarg ;;
+
+  -disable-* | --disable-*)
+    ac_useropt=`expr "x$ac_option" : 'x-*disable-\(.*\)'`
+    # Reject names that are not valid shell variable names.
+    expr "x$ac_useropt" : ".*[^-+._$as_cr_alnum]" >/dev/null &&
+      as_fn_error $? "invalid feature name: $ac_useropt"
+    ac_useropt_orig=$ac_useropt
+    ac_useropt=`$as_echo "$ac_useropt" | sed 's/[-+.]/_/g'`
+    case $ac_user_opts in
+      *"
+"enable_$ac_useropt"
+"*) ;;
+      *) ac_unrecognized_opts="$ac_unrecognized_opts$ac_unrecognized_sep--disable-$ac_useropt_orig"
+	 ac_unrecognized_sep=', ';;
+    esac
+    eval enable_$ac_useropt=no ;;
+
+  -docdir | --docdir | --docdi | --doc | --do)
+    ac_prev=docdir ;;
+  -docdir=* | --docdir=* | --docdi=* | --doc=* | --do=*)
+    docdir=$ac_optarg ;;
+
+  -dvidir | --dvidir | --dvidi | --dvid | --dvi | --dv)
+    ac_prev=dvidir ;;
+  -dvidir=* | --dvidir=* | --dvidi=* | --dvid=* | --dvi=* | --dv=*)
+    dvidir=$ac_optarg ;;
+
+  -enable-* | --enable-*)
+    ac_useropt=`expr "x$ac_option" : 'x-*enable-\([^=]*\)'`
+    # Reject names that are not valid shell variable names.
+    expr "x$ac_useropt" : ".*[^-+._$as_cr_alnum]" >/dev/null &&
+      as_fn_error $? "invalid feature name: $ac_useropt"
+    ac_useropt_orig=$ac_useropt
+    ac_useropt=`$as_echo "$ac_useropt" | sed 's/[-+.]/_/g'`
+    case $ac_user_opts in
+      *"
+"enable_$ac_useropt"
+"*) ;;
+      *) ac_unrecognized_opts="$ac_unrecognized_opts$ac_unrecognized_sep--enable-$ac_useropt_orig"
+	 ac_unrecognized_sep=', ';;
+    esac
+    eval enable_$ac_useropt=\$ac_optarg ;;
+
+  -exec-prefix | --exec_prefix | --exec-prefix | --exec-prefi \
+  | --exec-pref | --exec-pre | --exec-pr | --exec-p | --exec- \
+  | --exec | --exe | --ex)
+    ac_prev=exec_prefix ;;
+  -exec-prefix=* | --exec_prefix=* | --exec-prefix=* | --exec-prefi=* \
+  | --exec-pref=* | --exec-pre=* | --exec-pr=* | --exec-p=* | --exec-=* \
+  | --exec=* | --exe=* | --ex=*)
+    exec_prefix=$ac_optarg ;;
+
+  -gas | --gas | --ga | --g)
+    # Obsolete; use --with-gas.
+    with_gas=yes ;;
+
+  -help | --help | --hel | --he | -h)
+    ac_init_help=long ;;
+  -help=r* | --help=r* | --hel=r* | --he=r* | -hr*)
+    ac_init_help=recursive ;;
+  -help=s* | --help=s* | --hel=s* | --he=s* | -hs*)
+    ac_init_help=short ;;
+
+  -host | --host | --hos | --ho)
+    ac_prev=host_alias ;;
+  -host=* | --host=* | --hos=* | --ho=*)
+    host_alias=$ac_optarg ;;
+
+  -htmldir | --htmldir | --htmldi | --htmld | --html | --htm | --ht)
+    ac_prev=htmldir ;;
+  -htmldir=* | --htmldir=* | --htmldi=* | --htmld=* | --html=* | --htm=* \
+  | --ht=*)
+    htmldir=$ac_optarg ;;
+
+  -includedir | --includedir | --includedi | --included | --include \
+  | --includ | --inclu | --incl | --inc)
+    ac_prev=includedir ;;
+  -includedir=* | --includedir=* | --includedi=* | --included=* | --include=* \
+  | --includ=* | --inclu=* | --incl=* | --inc=*)
+    includedir=$ac_optarg ;;
+
+  -infodir | --infodir | --infodi | --infod | --info | --inf)
+    ac_prev=infodir ;;
+  -infodir=* | --infodir=* | --infodi=* | --infod=* | --info=* | --inf=*)
+    infodir=$ac_optarg ;;
+
+  -libdir | --libdir | --libdi | --libd)
+    ac_prev=libdir ;;
+  -libdir=* | --libdir=* | --libdi=* | --libd=*)
+    libdir=$ac_optarg ;;
+
+  -libexecdir | --libexecdir | --libexecdi | --libexecd | --libexec \
+  | --libexe | --libex | --libe)
+    ac_prev=libexecdir ;;
+  -libexecdir=* | --libexecdir=* | --libexecdi=* | --libexecd=* | --libexec=* \
+  | --libexe=* | --libex=* | --libe=*)
+    libexecdir=$ac_optarg ;;
+
+  -localedir | --localedir | --localedi | --localed | --locale)
+    ac_prev=localedir ;;
+  -localedir=* | --localedir=* | --localedi=* | --localed=* | --locale=*)
+    localedir=$ac_optarg ;;
+
+  -localstatedir | --localstatedir | --localstatedi | --localstated \
+  | --localstate | --localstat | --localsta | --localst | --locals)
+    ac_prev=localstatedir ;;
+  -localstatedir=* | --localstatedir=* | --localstatedi=* | --localstated=* \
+  | --localstate=* | --localstat=* | --localsta=* | --localst=* | --locals=*)
+    localstatedir=$ac_optarg ;;
+
+  -mandir | --mandir | --mandi | --mand | --man | --ma | --m)
+    ac_prev=mandir ;;
+  -mandir=* | --mandir=* | --mandi=* | --mand=* | --man=* | --ma=* | --m=*)
+    mandir=$ac_optarg ;;
+
+  -nfp | --nfp | --nf)
+    # Obsolete; use --without-fp.
+    with_fp=no ;;
+
+  -no-create | --no-create | --no-creat | --no-crea | --no-cre \
+  | --no-cr | --no-c | -n)
+    no_create=yes ;;
+
+  -no-recursion | --no-recursion | --no-recursio | --no-recursi \
+  | --no-recurs | --no-recur | --no-recu | --no-rec | --no-re | --no-r)
+    no_recursion=yes ;;
+
+  -oldincludedir | --oldincludedir | --oldincludedi | --oldincluded \
+  | --oldinclude | --oldinclud | --oldinclu | --oldincl | --oldinc \
+  | --oldin | --oldi | --old | --ol | --o)
+    ac_prev=oldincludedir ;;
+  -oldincludedir=* | --oldincludedir=* | --oldincludedi=* | --oldincluded=* \
+  | --oldinclude=* | --oldinclud=* | --oldinclu=* | --oldincl=* | --oldinc=* \
+  | --oldin=* | --oldi=* | --old=* | --ol=* | --o=*)
+    oldincludedir=$ac_optarg ;;
+
+  -prefix | --prefix | --prefi | --pref | --pre | --pr | --p)
+    ac_prev=prefix ;;
+  -prefix=* | --prefix=* | --prefi=* | --pref=* | --pre=* | --pr=* | --p=*)
+    prefix=$ac_optarg ;;
+
+  -program-prefix | --program-prefix | --program-prefi | --program-pref \
+  | --program-pre | --program-pr | --program-p)
+    ac_prev=program_prefix ;;
+  -program-prefix=* | --program-prefix=* | --program-prefi=* \
+  | --program-pref=* | --program-pre=* | --program-pr=* | --program-p=*)
+    program_prefix=$ac_optarg ;;
+
+  -program-suffix | --program-suffix | --program-suffi | --program-suff \
+  | --program-suf | --program-su | --program-s)
+    ac_prev=program_suffix ;;
+  -program-suffix=* | --program-suffix=* | --program-suffi=* \
+  | --program-suff=* | --program-suf=* | --program-su=* | --program-s=*)
+    program_suffix=$ac_optarg ;;
+
+  -program-transform-name | --program-transform-name \
+  | --program-transform-nam | --program-transform-na \
+  | --program-transform-n | --program-transform- \
+  | --program-transform | --program-transfor \
+  | --program-transfo | --program-transf \
+  | --program-trans | --program-tran \
+  | --progr-tra | --program-tr | --program-t)
+    ac_prev=program_transform_name ;;
+  -program-transform-name=* | --program-transform-name=* \
+  | --program-transform-nam=* | --program-transform-na=* \
+  | --program-transform-n=* | --program-transform-=* \
+  | --program-transform=* | --program-transfor=* \
+  | --program-transfo=* | --program-transf=* \
+  | --program-trans=* | --program-tran=* \
+  | --progr-tra=* | --program-tr=* | --program-t=*)
+    program_transform_name=$ac_optarg ;;
+
+  -pdfdir | --pdfdir | --pdfdi | --pdfd | --pdf | --pd)
+    ac_prev=pdfdir ;;
+  -pdfdir=* | --pdfdir=* | --pdfdi=* | --pdfd=* | --pdf=* | --pd=*)
+    pdfdir=$ac_optarg ;;
+
+  -psdir | --psdir | --psdi | --psd | --ps)
+    ac_prev=psdir ;;
+  -psdir=* | --psdir=* | --psdi=* | --psd=* | --ps=*)
+    psdir=$ac_optarg ;;
+
+  -q | -quiet | --quiet | --quie | --qui | --qu | --q \
+  | -silent | --silent | --silen | --sile | --sil)
+    silent=yes ;;
+
+  -sbindir | --sbindir | --sbindi | --sbind | --sbin | --sbi | --sb)
+    ac_prev=sbindir ;;
+  -sbindir=* | --sbindir=* | --sbindi=* | --sbind=* | --sbin=* \
+  | --sbi=* | --sb=*)
+    sbindir=$ac_optarg ;;
+
+  -sharedstatedir | --sharedstatedir | --sharedstatedi \
+  | --sharedstated | --sharedstate | --sharedstat | --sharedsta \
+  | --sharedst | --shareds | --shared | --share | --shar \
+  | --sha | --sh)
+    ac_prev=sharedstatedir ;;
+  -sharedstatedir=* | --sharedstatedir=* | --sharedstatedi=* \
+  | --sharedstated=* | --sharedstate=* | --sharedstat=* | --sharedsta=* \
+  | --sharedst=* | --shareds=* | --shared=* | --share=* | --shar=* \
+  | --sha=* | --sh=*)
+    sharedstatedir=$ac_optarg ;;
+
+  -site | --site | --sit)
+    ac_prev=site ;;
+  -site=* | --site=* | --sit=*)
+    site=$ac_optarg ;;
+
+  -srcdir | --srcdir | --srcdi | --srcd | --src | --sr)
+    ac_prev=srcdir ;;
+  -srcdir=* | --srcdir=* | --srcdi=* | --srcd=* | --src=* | --sr=*)
+    srcdir=$ac_optarg ;;
+
+  -sysconfdir | --sysconfdir | --sysconfdi | --sysconfd | --sysconf \
+  | --syscon | --sysco | --sysc | --sys | --sy)
+    ac_prev=sysconfdir ;;
+  -sysconfdir=* | --sysconfdir=* | --sysconfdi=* | --sysconfd=* | --sysconf=* \
+  | --syscon=* | --sysco=* | --sysc=* | --sys=* | --sy=*)
+    sysconfdir=$ac_optarg ;;
+
+  -target | --target | --targe | --targ | --tar | --ta | --t)
+    ac_prev=target_alias ;;
+  -target=* | --target=* | --targe=* | --targ=* | --tar=* | --ta=* | --t=*)
+    target_alias=$ac_optarg ;;
+
+  -v | -verbose | --verbose | --verbos | --verbo | --verb)
+    verbose=yes ;;
+
+  -version | --version | --versio | --versi | --vers | -V)
+    ac_init_version=: ;;
+
+  -with-* | --with-*)
+    ac_useropt=`expr "x$ac_option" : 'x-*with-\([^=]*\)'`
+    # Reject names that are not valid shell variable names.
+    expr "x$ac_useropt" : ".*[^-+._$as_cr_alnum]" >/dev/null &&
+      as_fn_error $? "invalid package name: $ac_useropt"
+    ac_useropt_orig=$ac_useropt
+    ac_useropt=`$as_echo "$ac_useropt" | sed 's/[-+.]/_/g'`
+    case $ac_user_opts in
+      *"
+"with_$ac_useropt"
+"*) ;;
+      *) ac_unrecognized_opts="$ac_unrecognized_opts$ac_unrecognized_sep--with-$ac_useropt_orig"
+	 ac_unrecognized_sep=', ';;
+    esac
+    eval with_$ac_useropt=\$ac_optarg ;;
+
+  -without-* | --without-*)
+    ac_useropt=`expr "x$ac_option" : 'x-*without-\(.*\)'`
+    # Reject names that are not valid shell variable names.
+    expr "x$ac_useropt" : ".*[^-+._$as_cr_alnum]" >/dev/null &&
+      as_fn_error $? "invalid package name: $ac_useropt"
+    ac_useropt_orig=$ac_useropt
+    ac_useropt=`$as_echo "$ac_useropt" | sed 's/[-+.]/_/g'`
+    case $ac_user_opts in
+      *"
+"with_$ac_useropt"
+"*) ;;
+      *) ac_unrecognized_opts="$ac_unrecognized_opts$ac_unrecognized_sep--without-$ac_useropt_orig"
+	 ac_unrecognized_sep=', ';;
+    esac
+    eval with_$ac_useropt=no ;;
+
+  --x)
+    # Obsolete; use --with-x.
+    with_x=yes ;;
+
+  -x-includes | --x-includes | --x-include | --x-includ | --x-inclu \
+  | --x-incl | --x-inc | --x-in | --x-i)
+    ac_prev=x_includes ;;
+  -x-includes=* | --x-includes=* | --x-include=* | --x-includ=* | --x-inclu=* \
+  | --x-incl=* | --x-inc=* | --x-in=* | --x-i=*)
+    x_includes=$ac_optarg ;;
+
+  -x-libraries | --x-libraries | --x-librarie | --x-librari \
+  | --x-librar | --x-libra | --x-libr | --x-lib | --x-li | --x-l)
+    ac_prev=x_libraries ;;
+  -x-libraries=* | --x-libraries=* | --x-librarie=* | --x-librari=* \
+  | --x-librar=* | --x-libra=* | --x-libr=* | --x-lib=* | --x-li=* | --x-l=*)
+    x_libraries=$ac_optarg ;;
+
+  -*) as_fn_error $? "unrecognized option: \`$ac_option'
+Try \`$0 --help' for more information"
+    ;;
+
+  *=*)
+    ac_envvar=`expr "x$ac_option" : 'x\([^=]*\)='`
+    # Reject names that are not valid shell variable names.
+    case $ac_envvar in #(
+      '' | [0-9]* | *[!_$as_cr_alnum]* )
+      as_fn_error $? "invalid variable name: \`$ac_envvar'" ;;
+    esac
+    eval $ac_envvar=\$ac_optarg
+    export $ac_envvar ;;
+
+  *)
+    # FIXME: should be removed in autoconf 3.0.
+    $as_echo "$as_me: WARNING: you should use --build, --host, --target" >&2
+    expr "x$ac_option" : ".*[^-._$as_cr_alnum]" >/dev/null &&
+      $as_echo "$as_me: WARNING: invalid host type: $ac_option" >&2
+    : "${build_alias=$ac_option} ${host_alias=$ac_option} ${target_alias=$ac_option}"
+    ;;
+
+  esac
+done
+
+if test -n "$ac_prev"; then
+  ac_option=--`echo $ac_prev | sed 's/_/-/g'`
+  as_fn_error $? "missing argument to $ac_option"
+fi
+
+if test -n "$ac_unrecognized_opts"; then
+  case $enable_option_checking in
+    no) ;;
+    fatal) as_fn_error $? "unrecognized options: $ac_unrecognized_opts" ;;
+    *)     $as_echo "$as_me: WARNING: unrecognized options: $ac_unrecognized_opts" >&2 ;;
+  esac
+fi
+
+# Check all directory arguments for consistency.
+for ac_var in	exec_prefix prefix bindir sbindir libexecdir datarootdir \
+		datadir sysconfdir sharedstatedir localstatedir includedir \
+		oldincludedir docdir infodir htmldir dvidir pdfdir psdir \
+		libdir localedir mandir
+do
+  eval ac_val=\$$ac_var
+  # Remove trailing slashes.
+  case $ac_val in
+    */ )
+      ac_val=`expr "X$ac_val" : 'X\(.*[^/]\)' \| "X$ac_val" : 'X\(.*\)'`
+      eval $ac_var=\$ac_val;;
+  esac
+  # Be sure to have absolute directory names.
+  case $ac_val in
+    [\\/$]* | ?:[\\/]* )  continue;;
+    NONE | '' ) case $ac_var in *prefix ) continue;; esac;;
+  esac
+  as_fn_error $? "expected an absolute directory name for --$ac_var: $ac_val"
+done
+
+# There might be people who depend on the old broken behavior: `$host'
+# used to hold the argument of --host etc.
+# FIXME: To remove some day.
+build=$build_alias
+host=$host_alias
+target=$target_alias
+
+# FIXME: To remove some day.
+if test "x$host_alias" != x; then
+  if test "x$build_alias" = x; then
+    cross_compiling=maybe
+  elif test "x$build_alias" != "x$host_alias"; then
+    cross_compiling=yes
+  fi
+fi
+
+ac_tool_prefix=
+test -n "$host_alias" && ac_tool_prefix=$host_alias-
+
+test "$silent" = yes && exec 6>/dev/null
+
+
+ac_pwd=`pwd` && test -n "$ac_pwd" &&
+ac_ls_di=`ls -di .` &&
+ac_pwd_ls_di=`cd "$ac_pwd" && ls -di .` ||
+  as_fn_error $? "working directory cannot be determined"
+test "X$ac_ls_di" = "X$ac_pwd_ls_di" ||
+  as_fn_error $? "pwd does not report name of working directory"
+
+
+# Find the source files, if location was not specified.
+if test -z "$srcdir"; then
+  ac_srcdir_defaulted=yes
+  # Try the directory containing this script, then the parent directory.
+  ac_confdir=`$as_dirname -- "$as_myself" ||
+$as_expr X"$as_myself" : 'X\(.*[^/]\)//*[^/][^/]*/*$' \| \
+	 X"$as_myself" : 'X\(//\)[^/]' \| \
+	 X"$as_myself" : 'X\(//\)$' \| \
+	 X"$as_myself" : 'X\(/\)' \| . 2>/dev/null ||
+$as_echo X"$as_myself" |
+    sed '/^X\(.*[^/]\)\/\/*[^/][^/]*\/*$/{
+	    s//\1/
+	    q
+	  }
+	  /^X\(\/\/\)[^/].*/{
+	    s//\1/
+	    q
+	  }
+	  /^X\(\/\/\)$/{
+	    s//\1/
+	    q
+	  }
+	  /^X\(\/\).*/{
+	    s//\1/
+	    q
+	  }
+	  s/.*/./; q'`
+  srcdir=$ac_confdir
+  if test ! -r "$srcdir/$ac_unique_file"; then
+    srcdir=..
+  fi
+else
+  ac_srcdir_defaulted=no
+fi
+if test ! -r "$srcdir/$ac_unique_file"; then
+  test "$ac_srcdir_defaulted" = yes && srcdir="$ac_confdir or .."
+  as_fn_error $? "cannot find sources ($ac_unique_file) in $srcdir"
+fi
+ac_msg="sources are in $srcdir, but \`cd $srcdir' does not work"
+ac_abs_confdir=`(
+	cd "$srcdir" && test -r "./$ac_unique_file" || as_fn_error $? "$ac_msg"
+	pwd)`
+# When building in place, set srcdir=.
+if test "$ac_abs_confdir" = "$ac_pwd"; then
+  srcdir=.
+fi
+# Remove unnecessary trailing slashes from srcdir.
+# Double slashes in file names in object file debugging info
+# mess up M-x gdb in Emacs.
+case $srcdir in
+*/) srcdir=`expr "X$srcdir" : 'X\(.*[^/]\)' \| "X$srcdir" : 'X\(.*\)'`;;
+esac
+for ac_var in $ac_precious_vars; do
+  eval ac_env_${ac_var}_set=\${${ac_var}+set}
+  eval ac_env_${ac_var}_value=\$${ac_var}
+  eval ac_cv_env_${ac_var}_set=\${${ac_var}+set}
+  eval ac_cv_env_${ac_var}_value=\$${ac_var}
+done
+
+#
+# Report the --help message.
+#
+if test "$ac_init_help" = "long"; then
+  # Omit some internal or obsolete options to make the list less imposing.
+  # This message is too long to be a string in the A/UX 3.1 sh.
+  cat <<_ACEOF
+\`configure' configures ccminer 2014.03.18 to adapt to many kinds of systems.
+
+Usage: $0 [OPTION]... [VAR=VALUE]...
+
+To assign environment variables (e.g., CC, CFLAGS...), specify them as
+VAR=VALUE.  See below for descriptions of some of the useful variables.
+
+Defaults for the options are specified in brackets.
+
+Configuration:
+  -h, --help              display this help and exit
+      --help=short        display options specific to this package
+      --help=recursive    display the short help of all the included packages
+  -V, --version           display version information and exit
+  -q, --quiet, --silent   do not print \`checking ...' messages
+      --cache-file=FILE   cache test results in FILE [disabled]
+  -C, --config-cache      alias for \`--cache-file=config.cache'
+  -n, --no-create         do not create output files
+      --srcdir=DIR        find the sources in DIR [configure dir or \`..']
+
+Installation directories:
+  --prefix=PREFIX         install architecture-independent files in PREFIX
+                          [$ac_default_prefix]
+  --exec-prefix=EPREFIX   install architecture-dependent files in EPREFIX
+                          [PREFIX]
+
+By default, \`make install' will install all the files in
+\`$ac_default_prefix/bin', \`$ac_default_prefix/lib' etc.  You can specify
+an installation prefix other than \`$ac_default_prefix' using \`--prefix',
+for instance \`--prefix=\$HOME'.
+
+For better control, use the options below.
+
+Fine tuning of the installation directories:
+  --bindir=DIR            user executables [EPREFIX/bin]
+  --sbindir=DIR           system admin executables [EPREFIX/sbin]
+  --libexecdir=DIR        program executables [EPREFIX/libexec]
+  --sysconfdir=DIR        read-only single-machine data [PREFIX/etc]
+  --sharedstatedir=DIR    modifiable architecture-independent data [PREFIX/com]
+  --localstatedir=DIR     modifiable single-machine data [PREFIX/var]
+  --libdir=DIR            object code libraries [EPREFIX/lib]
+  --includedir=DIR        C header files [PREFIX/include]
+  --oldincludedir=DIR     C header files for non-gcc [/usr/include]
+  --datarootdir=DIR       read-only arch.-independent data root [PREFIX/share]
+  --datadir=DIR           read-only architecture-independent data [DATAROOTDIR]
+  --infodir=DIR           info documentation [DATAROOTDIR/info]
+  --localedir=DIR         locale-dependent data [DATAROOTDIR/locale]
+  --mandir=DIR            man documentation [DATAROOTDIR/man]
+  --docdir=DIR            documentation root [DATAROOTDIR/doc/ccminer]
+  --htmldir=DIR           html documentation [DOCDIR]
+  --dvidir=DIR            dvi documentation [DOCDIR]
+  --pdfdir=DIR            pdf documentation [DOCDIR]
+  --psdir=DIR             ps documentation [DOCDIR]
+_ACEOF
+
+  cat <<\_ACEOF
+
+Program names:
+  --program-prefix=PREFIX            prepend PREFIX to installed program names
+  --program-suffix=SUFFIX            append SUFFIX to installed program names
+  --program-transform-name=PROGRAM   run sed PROGRAM on installed program names
+
+System types:
+  --build=BUILD     configure for building on BUILD [guessed]
+  --host=HOST       cross-compile to build programs to run on HOST [BUILD]
+  --target=TARGET   configure for building compilers for TARGET [HOST]
+_ACEOF
+fi
+
+if test -n "$ac_init_help"; then
+  case $ac_init_help in
+     short | recursive ) echo "Configuration of ccminer 2014.03.18:";;
+   esac
+  cat <<\_ACEOF
+
+Optional Features:
+  --disable-option-checking  ignore unrecognized --enable/--with options
+  --disable-FEATURE       do not include FEATURE (same as --enable-FEATURE=no)
+  --enable-FEATURE[=ARG]  include FEATURE [ARG=yes]
+  --enable-silent-rules   less verbose build output (undo: "make V=1")
+  --disable-silent-rules  verbose build output (undo: "make V=0")
+  --enable-maintainer-mode
+                          enable make rules and dependencies not useful (and
+                          sometimes confusing) to the casual installer
+  --enable-dependency-tracking
+                          do not reject slow dependency extractors
+  --disable-dependency-tracking
+                          speeds up one-time build
+  --disable-openmp        do not use OpenMP
+
+Optional Packages:
+  --with-PACKAGE[=ARG]    use PACKAGE [ARG=yes]
+  --without-PACKAGE       do not use PACKAGE (same as --with-PACKAGE=no)
+  --with-libcurl=PREFIX   look for the curl library in PREFIX/lib and headers
+                          in PREFIX/include
+  --with-cuda=PATH    prefix where cuda is installed default=/usr/local/cuda
+
+Some influential environment variables:
+  CC          C compiler command
+  CFLAGS      C compiler flags
+  LDFLAGS     linker flags, e.g. -L<lib dir> if you have libraries in a
+              nonstandard directory <lib dir>
+  LIBS        libraries to pass to the linker, e.g. -l<library>
+  CPPFLAGS    (Objective) C/C++ preprocessor flags, e.g. -I<include dir> if
+              you have headers in a nonstandard directory <include dir>
+  CPP         C preprocessor
+  CCAS        assembler compiler command (defaults to CC)
+  CCASFLAGS   assembler compiler flags (defaults to CFLAGS)
+  CXX         C++ compiler command
+  CXXFLAGS    C++ compiler flags
+
+Use these variables to override the choices made by `configure' or to help
+it to find libraries and programs with nonstandard names/locations.
+
+Report bugs to the package provider.
+_ACEOF
+ac_status=$?
+fi
+
+if test "$ac_init_help" = "recursive"; then
+  # If there are subdirs, report their specific --help.
+  for ac_dir in : $ac_subdirs_all; do test "x$ac_dir" = x: && continue
+    test -d "$ac_dir" ||
+      { cd "$srcdir" && ac_pwd=`pwd` && srcdir=. && test -d "$ac_dir"; } ||
+      continue
+    ac_builddir=.
+
+case "$ac_dir" in
+.) ac_dir_suffix= ac_top_builddir_sub=. ac_top_build_prefix= ;;
+*)
+  ac_dir_suffix=/`$as_echo "$ac_dir" | sed 's|^\.[\\/]||'`
+  # A ".." for each directory in $ac_dir_suffix.
+  ac_top_builddir_sub=`$as_echo "$ac_dir_suffix" | sed 's|/[^\\/]*|/..|g;s|/||'`
+  case $ac_top_builddir_sub in
+  "") ac_top_builddir_sub=. ac_top_build_prefix= ;;
+  *)  ac_top_build_prefix=$ac_top_builddir_sub/ ;;
+  esac ;;
+esac
+ac_abs_top_builddir=$ac_pwd
+ac_abs_builddir=$ac_pwd$ac_dir_suffix
+# for backward compatibility:
+ac_top_builddir=$ac_top_build_prefix
+
+case $srcdir in
+  .)  # We are building in place.
+    ac_srcdir=.
+    ac_top_srcdir=$ac_top_builddir_sub
+    ac_abs_top_srcdir=$ac_pwd ;;
+  [\\/]* | ?:[\\/]* )  # Absolute name.
+    ac_srcdir=$srcdir$ac_dir_suffix;
+    ac_top_srcdir=$srcdir
+    ac_abs_top_srcdir=$srcdir ;;
+  *) # Relative name.
+    ac_srcdir=$ac_top_build_prefix$srcdir$ac_dir_suffix
+    ac_top_srcdir=$ac_top_build_prefix$srcdir
+    ac_abs_top_srcdir=$ac_pwd/$srcdir ;;
+esac
+ac_abs_srcdir=$ac_abs_top_srcdir$ac_dir_suffix
+
+    cd "$ac_dir" || { ac_status=$?; continue; }
+    # Check for guested configure.
+    if test -f "$ac_srcdir/configure.gnu"; then
+      echo &&
+      $SHELL "$ac_srcdir/configure.gnu" --help=recursive
+    elif test -f "$ac_srcdir/configure"; then
+      echo &&
+      $SHELL "$ac_srcdir/configure" --help=recursive
+    else
+      $as_echo "$as_me: WARNING: no configuration information is in $ac_dir" >&2
+    fi || ac_status=$?
+    cd "$ac_pwd" || { ac_status=$?; break; }
+  done
+fi
+
+test -n "$ac_init_help" && exit $ac_status
+if $ac_init_version; then
+  cat <<\_ACEOF
+ccminer configure 2014.03.18
+generated by GNU Autoconf 2.69
+
+Copyright (C) 2012 Free Software Foundation, Inc.
+This configure script is free software; the Free Software Foundation
+gives unlimited permission to copy, distribute and modify it.
+_ACEOF
+  exit
+fi
+
+## ------------------------ ##
+## Autoconf initialization. ##
+## ------------------------ ##
+
+# ac_fn_c_try_compile LINENO
+# --------------------------
+# Try to compile conftest.$ac_ext, and return whether this succeeded.
+ac_fn_c_try_compile ()
+{
+  as_lineno=${as_lineno-"$1"} as_lineno_stack=as_lineno_stack=$as_lineno_stack
+  rm -f conftest.$ac_objext
+  if { { ac_try="$ac_compile"
+case "(($ac_try" in
+  *\"* | *\`* | *\\*) ac_try_echo=\$ac_try;;
+  *) ac_try_echo=$ac_try;;
+esac
+eval ac_try_echo="\"\$as_me:${as_lineno-$LINENO}: $ac_try_echo\""
+$as_echo "$ac_try_echo"; } >&5
+  (eval "$ac_compile") 2>conftest.err
+  ac_status=$?
+  if test -s conftest.err; then
+    grep -v '^ *+' conftest.err >conftest.er1
+    cat conftest.er1 >&5
+    mv -f conftest.er1 conftest.err
+  fi
+  $as_echo "$as_me:${as_lineno-$LINENO}: \$? = $ac_status" >&5
+  test $ac_status = 0; } && {
+	 test -z "$ac_c_werror_flag" ||
+	 test ! -s conftest.err
+       } && test -s conftest.$ac_objext; then :
+  ac_retval=0
+else
+  $as_echo "$as_me: failed program was:" >&5
+sed 's/^/| /' conftest.$ac_ext >&5
+
+	ac_retval=1
+fi
+  eval $as_lineno_stack; ${as_lineno_stack:+:} unset as_lineno
+  as_fn_set_status $ac_retval
+
+} # ac_fn_c_try_compile
+
+# ac_fn_c_try_cpp LINENO
+# ----------------------
+# Try to preprocess conftest.$ac_ext, and return whether this succeeded.
+ac_fn_c_try_cpp ()
+{
+  as_lineno=${as_lineno-"$1"} as_lineno_stack=as_lineno_stack=$as_lineno_stack
+  if { { ac_try="$ac_cpp conftest.$ac_ext"
+case "(($ac_try" in
+  *\"* | *\`* | *\\*) ac_try_echo=\$ac_try;;
+  *) ac_try_echo=$ac_try;;
+esac
+eval ac_try_echo="\"\$as_me:${as_lineno-$LINENO}: $ac_try_echo\""
+$as_echo "$ac_try_echo"; } >&5
+  (eval "$ac_cpp conftest.$ac_ext") 2>conftest.err
+  ac_status=$?
+  if test -s conftest.err; then
+    grep -v '^ *+' conftest.err >conftest.er1
+    cat conftest.er1 >&5
+    mv -f conftest.er1 conftest.err
+  fi
+  $as_echo "$as_me:${as_lineno-$LINENO}: \$? = $ac_status" >&5
+  test $ac_status = 0; } > conftest.i && {
+	 test -z "$ac_c_preproc_warn_flag$ac_c_werror_flag" ||
+	 test ! -s conftest.err
+       }; then :
+  ac_retval=0
+else
+  $as_echo "$as_me: failed program was:" >&5
+sed 's/^/| /' conftest.$ac_ext >&5
+
+    ac_retval=1
+fi
+  eval $as_lineno_stack; ${as_lineno_stack:+:} unset as_lineno
+  as_fn_set_status $ac_retval
+
+} # ac_fn_c_try_cpp
+
+# ac_fn_cxx_try_compile LINENO
+# ----------------------------
+# Try to compile conftest.$ac_ext, and return whether this succeeded.
+ac_fn_cxx_try_compile ()
+{
+  as_lineno=${as_lineno-"$1"} as_lineno_stack=as_lineno_stack=$as_lineno_stack
+  rm -f conftest.$ac_objext
+  if { { ac_try="$ac_compile"
+case "(($ac_try" in
+  *\"* | *\`* | *\\*) ac_try_echo=\$ac_try;;
+  *) ac_try_echo=$ac_try;;
+esac
+eval ac_try_echo="\"\$as_me:${as_lineno-$LINENO}: $ac_try_echo\""
+$as_echo "$ac_try_echo"; } >&5
+  (eval "$ac_compile") 2>conftest.err
+  ac_status=$?
+  if test -s conftest.err; then
+    grep -v '^ *+' conftest.err >conftest.er1
+    cat conftest.er1 >&5
+    mv -f conftest.er1 conftest.err
+  fi
+  $as_echo "$as_me:${as_lineno-$LINENO}: \$? = $ac_status" >&5
+  test $ac_status = 0; } && {
+	 test -z "$ac_cxx_werror_flag" ||
+	 test ! -s conftest.err
+       } && test -s conftest.$ac_objext; then :
+  ac_retval=0
+else
+  $as_echo "$as_me: failed program was:" >&5
+sed 's/^/| /' conftest.$ac_ext >&5
+
+	ac_retval=1
+fi
+  eval $as_lineno_stack; ${as_lineno_stack:+:} unset as_lineno
+  as_fn_set_status $ac_retval
+
+} # ac_fn_cxx_try_compile
+
+# ac_fn_c_try_link LINENO
+# -----------------------
+# Try to link conftest.$ac_ext, and return whether this succeeded.
+ac_fn_c_try_link ()
+{
+  as_lineno=${as_lineno-"$1"} as_lineno_stack=as_lineno_stack=$as_lineno_stack
+  rm -f conftest.$ac_objext conftest$ac_exeext
+  if { { ac_try="$ac_link"
+case "(($ac_try" in
+  *\"* | *\`* | *\\*) ac_try_echo=\$ac_try;;
+  *) ac_try_echo=$ac_try;;
+esac
+eval ac_try_echo="\"\$as_me:${as_lineno-$LINENO}: $ac_try_echo\""
+$as_echo "$ac_try_echo"; } >&5
+  (eval "$ac_link") 2>conftest.err
+  ac_status=$?
+  if test -s conftest.err; then
+    grep -v '^ *+' conftest.err >conftest.er1
+    cat conftest.er1 >&5
+    mv -f conftest.er1 conftest.err
+  fi
+  $as_echo "$as_me:${as_lineno-$LINENO}: \$? = $ac_status" >&5
+  test $ac_status = 0; } && {
+	 test -z "$ac_c_werror_flag" ||
+	 test ! -s conftest.err
+       } && test -s conftest$ac_exeext && {
+	 test "$cross_compiling" = yes ||
+	 test -x conftest$ac_exeext
+       }; then :
+  ac_retval=0
+else
+  $as_echo "$as_me: failed program was:" >&5
+sed 's/^/| /' conftest.$ac_ext >&5
+
+	ac_retval=1
+fi
+  # Delete the IPA/IPO (Inter Procedural Analysis/Optimization) information
+  # created by the PGI compiler (conftest_ipa8_conftest.oo), as it would
+  # interfere with the next link command; also delete a directory that is
+  # left behind by Apple's compiler.  We do this before executing the actions.
+  rm -rf conftest.dSYM conftest_ipa8_conftest.oo
+  eval $as_lineno_stack; ${as_lineno_stack:+:} unset as_lineno
+  as_fn_set_status $ac_retval
+
+} # ac_fn_c_try_link
+
+# ac_fn_c_try_run LINENO
+# ----------------------
+# Try to link conftest.$ac_ext, and return whether this succeeded. Assumes
+# that executables *can* be run.
+ac_fn_c_try_run ()
+{
+  as_lineno=${as_lineno-"$1"} as_lineno_stack=as_lineno_stack=$as_lineno_stack
+  if { { ac_try="$ac_link"
+case "(($ac_try" in
+  *\"* | *\`* | *\\*) ac_try_echo=\$ac_try;;
+  *) ac_try_echo=$ac_try;;
+esac
+eval ac_try_echo="\"\$as_me:${as_lineno-$LINENO}: $ac_try_echo\""
+$as_echo "$ac_try_echo"; } >&5
+  (eval "$ac_link") 2>&5
+  ac_status=$?
+  $as_echo "$as_me:${as_lineno-$LINENO}: \$? = $ac_status" >&5
+  test $ac_status = 0; } && { ac_try='./conftest$ac_exeext'
+  { { case "(($ac_try" in
+  *\"* | *\`* | *\\*) ac_try_echo=\$ac_try;;
+  *) ac_try_echo=$ac_try;;
+esac
+eval ac_try_echo="\"\$as_me:${as_lineno-$LINENO}: $ac_try_echo\""
+$as_echo "$ac_try_echo"; } >&5
+  (eval "$ac_try") 2>&5
+  ac_status=$?
+  $as_echo "$as_me:${as_lineno-$LINENO}: \$? = $ac_status" >&5
+  test $ac_status = 0; }; }; then :
+  ac_retval=0
+else
+  $as_echo "$as_me: program exited with status $ac_status" >&5
+       $as_echo "$as_me: failed program was:" >&5
+sed 's/^/| /' conftest.$ac_ext >&5
+
+       ac_retval=$ac_status
+fi
+  rm -rf conftest.dSYM conftest_ipa8_conftest.oo
+  eval $as_lineno_stack; ${as_lineno_stack:+:} unset as_lineno
+  as_fn_set_status $ac_retval
+
+} # ac_fn_c_try_run
+
+# ac_fn_c_check_header_mongrel LINENO HEADER VAR INCLUDES
+# -------------------------------------------------------
+# Tests whether HEADER exists, giving a warning if it cannot be compiled using
+# the include files in INCLUDES and setting the cache variable VAR
+# accordingly.
+ac_fn_c_check_header_mongrel ()
+{
+  as_lineno=${as_lineno-"$1"} as_lineno_stack=as_lineno_stack=$as_lineno_stack
+  if eval \${$3+:} false; then :
+  { $as_echo "$as_me:${as_lineno-$LINENO}: checking for $2" >&5
+$as_echo_n "checking for $2... " >&6; }
+if eval \${$3+:} false; then :
+  $as_echo_n "(cached) " >&6
+fi
+eval ac_res=\$$3
+	       { $as_echo "$as_me:${as_lineno-$LINENO}: result: $ac_res" >&5
+$as_echo "$ac_res" >&6; }
+else
+  # Is the header compilable?
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking $2 usability" >&5
+$as_echo_n "checking $2 usability... " >&6; }
+cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+/* end confdefs.h.  */
+$4
+#include <$2>
+_ACEOF
+if ac_fn_c_try_compile "$LINENO"; then :
+  ac_header_compiler=yes
+else
+  ac_header_compiler=no
+fi
+rm -f core conftest.err conftest.$ac_objext conftest.$ac_ext
+{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $ac_header_compiler" >&5
+$as_echo "$ac_header_compiler" >&6; }
+
+# Is the header present?
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking $2 presence" >&5
+$as_echo_n "checking $2 presence... " >&6; }
+cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+/* end confdefs.h.  */
+#include <$2>
+_ACEOF
+if ac_fn_c_try_cpp "$LINENO"; then :
+  ac_header_preproc=yes
+else
+  ac_header_preproc=no
+fi
+rm -f conftest.err conftest.i conftest.$ac_ext
+{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $ac_header_preproc" >&5
+$as_echo "$ac_header_preproc" >&6; }
+
+# So?  What about this header?
+case $ac_header_compiler:$ac_header_preproc:$ac_c_preproc_warn_flag in #((
+  yes:no: )
+    { $as_echo "$as_me:${as_lineno-$LINENO}: WARNING: $2: accepted by the compiler, rejected by the preprocessor!" >&5
+$as_echo "$as_me: WARNING: $2: accepted by the compiler, rejected by the preprocessor!" >&2;}
+    { $as_echo "$as_me:${as_lineno-$LINENO}: WARNING: $2: proceeding with the compiler's result" >&5
+$as_echo "$as_me: WARNING: $2: proceeding with the compiler's result" >&2;}
+    ;;
+  no:yes:* )
+    { $as_echo "$as_me:${as_lineno-$LINENO}: WARNING: $2: present but cannot be compiled" >&5
+$as_echo "$as_me: WARNING: $2: present but cannot be compiled" >&2;}
+    { $as_echo "$as_me:${as_lineno-$LINENO}: WARNING: $2:     check for missing prerequisite headers?" >&5
+$as_echo "$as_me: WARNING: $2:     check for missing prerequisite headers?" >&2;}
+    { $as_echo "$as_me:${as_lineno-$LINENO}: WARNING: $2: see the Autoconf documentation" >&5
+$as_echo "$as_me: WARNING: $2: see the Autoconf documentation" >&2;}
+    { $as_echo "$as_me:${as_lineno-$LINENO}: WARNING: $2:     section \"Present But Cannot Be Compiled\"" >&5
+$as_echo "$as_me: WARNING: $2:     section \"Present But Cannot Be Compiled\"" >&2;}
+    { $as_echo "$as_me:${as_lineno-$LINENO}: WARNING: $2: proceeding with the compiler's result" >&5
+$as_echo "$as_me: WARNING: $2: proceeding with the compiler's result" >&2;}
+    ;;
+esac
+  { $as_echo "$as_me:${as_lineno-$LINENO}: checking for $2" >&5
+$as_echo_n "checking for $2... " >&6; }
+if eval \${$3+:} false; then :
+  $as_echo_n "(cached) " >&6
+else
+  eval "$3=\$ac_header_compiler"
+fi
+eval ac_res=\$$3
+	       { $as_echo "$as_me:${as_lineno-$LINENO}: result: $ac_res" >&5
+$as_echo "$ac_res" >&6; }
+fi
+  eval $as_lineno_stack; ${as_lineno_stack:+:} unset as_lineno
+
+} # ac_fn_c_check_header_mongrel
+
+# ac_fn_c_check_header_compile LINENO HEADER VAR INCLUDES
+# -------------------------------------------------------
+# Tests whether HEADER exists and can be compiled using the include files in
+# INCLUDES, setting the cache variable VAR accordingly.
+ac_fn_c_check_header_compile ()
+{
+  as_lineno=${as_lineno-"$1"} as_lineno_stack=as_lineno_stack=$as_lineno_stack
+  { $as_echo "$as_me:${as_lineno-$LINENO}: checking for $2" >&5
+$as_echo_n "checking for $2... " >&6; }
+if eval \${$3+:} false; then :
+  $as_echo_n "(cached) " >&6
+else
+  cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+/* end confdefs.h.  */
+$4
+#include <$2>
+_ACEOF
+if ac_fn_c_try_compile "$LINENO"; then :
+  eval "$3=yes"
+else
+  eval "$3=no"
+fi
+rm -f core conftest.err conftest.$ac_objext conftest.$ac_ext
+fi
+eval ac_res=\$$3
+	       { $as_echo "$as_me:${as_lineno-$LINENO}: result: $ac_res" >&5
+$as_echo "$ac_res" >&6; }
+  eval $as_lineno_stack; ${as_lineno_stack:+:} unset as_lineno
+
+} # ac_fn_c_check_header_compile
+
+# ac_fn_c_check_decl LINENO SYMBOL VAR INCLUDES
+# ---------------------------------------------
+# Tests whether SYMBOL is declared in INCLUDES, setting cache variable VAR
+# accordingly.
+ac_fn_c_check_decl ()
+{
+  as_lineno=${as_lineno-"$1"} as_lineno_stack=as_lineno_stack=$as_lineno_stack
+  as_decl_name=`echo $2|sed 's/ *(.*//'`
+  as_decl_use=`echo $2|sed -e 's/(/((/' -e 's/)/) 0&/' -e 's/,/) 0& (/g'`
+  { $as_echo "$as_me:${as_lineno-$LINENO}: checking whether $as_decl_name is declared" >&5
+$as_echo_n "checking whether $as_decl_name is declared... " >&6; }
+if eval \${$3+:} false; then :
+  $as_echo_n "(cached) " >&6
+else
+  cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+/* end confdefs.h.  */
+$4
+int
+main ()
+{
+#ifndef $as_decl_name
+#ifdef __cplusplus
+  (void) $as_decl_use;
+#else
+  (void) $as_decl_name;
+#endif
+#endif
+
+  ;
+  return 0;
+}
+_ACEOF
+if ac_fn_c_try_compile "$LINENO"; then :
+  eval "$3=yes"
+else
+  eval "$3=no"
+fi
+rm -f core conftest.err conftest.$ac_objext conftest.$ac_ext
+fi
+eval ac_res=\$$3
+	       { $as_echo "$as_me:${as_lineno-$LINENO}: result: $ac_res" >&5
+$as_echo "$ac_res" >&6; }
+  eval $as_lineno_stack; ${as_lineno_stack:+:} unset as_lineno
+
+} # ac_fn_c_check_decl
+
+# ac_fn_c_check_type LINENO TYPE VAR INCLUDES
+# -------------------------------------------
+# Tests whether TYPE exists after having included INCLUDES, setting cache
+# variable VAR accordingly.
+ac_fn_c_check_type ()
+{
+  as_lineno=${as_lineno-"$1"} as_lineno_stack=as_lineno_stack=$as_lineno_stack
+  { $as_echo "$as_me:${as_lineno-$LINENO}: checking for $2" >&5
+$as_echo_n "checking for $2... " >&6; }
+if eval \${$3+:} false; then :
+  $as_echo_n "(cached) " >&6
+else
+  eval "$3=no"
+  cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+/* end confdefs.h.  */
+$4
+int
+main ()
+{
+if (sizeof ($2))
+	 return 0;
+  ;
+  return 0;
+}
+_ACEOF
+if ac_fn_c_try_compile "$LINENO"; then :
+  cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+/* end confdefs.h.  */
+$4
+int
+main ()
+{
+if (sizeof (($2)))
+	    return 0;
+  ;
+  return 0;
+}
+_ACEOF
+if ac_fn_c_try_compile "$LINENO"; then :
+
+else
+  eval "$3=yes"
+fi
+rm -f core conftest.err conftest.$ac_objext conftest.$ac_ext
+fi
+rm -f core conftest.err conftest.$ac_objext conftest.$ac_ext
+fi
+eval ac_res=\$$3
+	       { $as_echo "$as_me:${as_lineno-$LINENO}: result: $ac_res" >&5
+$as_echo "$ac_res" >&6; }
+  eval $as_lineno_stack; ${as_lineno_stack:+:} unset as_lineno
+
+} # ac_fn_c_check_type
+
+# ac_fn_c_check_func LINENO FUNC VAR
+# ----------------------------------
+# Tests whether FUNC exists, setting the cache variable VAR accordingly
+ac_fn_c_check_func ()
+{
+  as_lineno=${as_lineno-"$1"} as_lineno_stack=as_lineno_stack=$as_lineno_stack
+  { $as_echo "$as_me:${as_lineno-$LINENO}: checking for $2" >&5
+$as_echo_n "checking for $2... " >&6; }
+if eval \${$3+:} false; then :
+  $as_echo_n "(cached) " >&6
+else
+  cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+/* end confdefs.h.  */
+/* Define $2 to an innocuous variant, in case <limits.h> declares $2.
+   For example, HP-UX 11i <limits.h> declares gettimeofday.  */
+#define $2 innocuous_$2
+
+/* System header to define __stub macros and hopefully few prototypes,
+    which can conflict with char $2 (); below.
+    Prefer <limits.h> to <assert.h> if __STDC__ is defined, since
+    <limits.h> exists even on freestanding compilers.  */
+
+#ifdef __STDC__
+# include <limits.h>
+#else
+# include <assert.h>
+#endif
+
+#undef $2
+
+/* Override any GCC internal prototype to avoid an error.
+   Use char because int might match the return type of a GCC
+   builtin and then its argument prototype would still apply.  */
+#ifdef __cplusplus
+extern "C"
+#endif
+char $2 ();
+/* The GNU C library defines this for functions which it implements
+    to always fail with ENOSYS.  Some functions are actually named
+    something starting with __ and the normal name is an alias.  */
+#if defined __stub_$2 || defined __stub___$2
+choke me
+#endif
+
+int
+main ()
+{
+return $2 ();
+  ;
+  return 0;
+}
+_ACEOF
+if ac_fn_c_try_link "$LINENO"; then :
+  eval "$3=yes"
+else
+  eval "$3=no"
+fi
+rm -f core conftest.err conftest.$ac_objext \
+    conftest$ac_exeext conftest.$ac_ext
+fi
+eval ac_res=\$$3
+	       { $as_echo "$as_me:${as_lineno-$LINENO}: result: $ac_res" >&5
+$as_echo "$ac_res" >&6; }
+  eval $as_lineno_stack; ${as_lineno_stack:+:} unset as_lineno
+
+} # ac_fn_c_check_func
+cat >config.log <<_ACEOF
+This file contains any messages produced by compilers while
+running configure, to aid debugging if configure makes a mistake.
+
+It was created by ccminer $as_me 2014.03.18, which was
+generated by GNU Autoconf 2.69.  Invocation command line was
+
+  $ $0 $@
+
+_ACEOF
+exec 5>>config.log
+{
+cat <<_ASUNAME
+## --------- ##
+## Platform. ##
+## --------- ##
+
+hostname = `(hostname || uname -n) 2>/dev/null | sed 1q`
+uname -m = `(uname -m) 2>/dev/null || echo unknown`
+uname -r = `(uname -r) 2>/dev/null || echo unknown`
+uname -s = `(uname -s) 2>/dev/null || echo unknown`
+uname -v = `(uname -v) 2>/dev/null || echo unknown`
+
+/usr/bin/uname -p = `(/usr/bin/uname -p) 2>/dev/null || echo unknown`
+/bin/uname -X     = `(/bin/uname -X) 2>/dev/null     || echo unknown`
+
+/bin/arch              = `(/bin/arch) 2>/dev/null              || echo unknown`
+/usr/bin/arch -k       = `(/usr/bin/arch -k) 2>/dev/null       || echo unknown`
+/usr/convex/getsysinfo = `(/usr/convex/getsysinfo) 2>/dev/null || echo unknown`
+/usr/bin/hostinfo      = `(/usr/bin/hostinfo) 2>/dev/null      || echo unknown`
+/bin/machine           = `(/bin/machine) 2>/dev/null           || echo unknown`
+/usr/bin/oslevel       = `(/usr/bin/oslevel) 2>/dev/null       || echo unknown`
+/bin/universe          = `(/bin/universe) 2>/dev/null          || echo unknown`
+
+_ASUNAME
+
+as_save_IFS=$IFS; IFS=$PATH_SEPARATOR
+for as_dir in $PATH
+do
+  IFS=$as_save_IFS
+  test -z "$as_dir" && as_dir=.
+    $as_echo "PATH: $as_dir"
+  done
+IFS=$as_save_IFS
+
+} >&5
+
+cat >&5 <<_ACEOF
+
+
+## ----------- ##
+## Core tests. ##
+## ----------- ##
+
+_ACEOF
+
+
+# Keep a trace of the command line.
+# Strip out --no-create and --no-recursion so they do not pile up.
+# Strip out --silent because we don't want to record it for future runs.
+# Also quote any args containing shell meta-characters.
+# Make two passes to allow for proper duplicate-argument suppression.
+ac_configure_args=
+ac_configure_args0=
+ac_configure_args1=
+ac_must_keep_next=false
+for ac_pass in 1 2
+do
+  for ac_arg
+  do
+    case $ac_arg in
+    -no-create | --no-c* | -n | -no-recursion | --no-r*) continue ;;
+    -q | -quiet | --quiet | --quie | --qui | --qu | --q \
+    | -silent | --silent | --silen | --sile | --sil)
+      continue ;;
+    *\'*)
+      ac_arg=`$as_echo "$ac_arg" | sed "s/'/'\\\\\\\\''/g"` ;;
+    esac
+    case $ac_pass in
+    1) as_fn_append ac_configure_args0 " '$ac_arg'" ;;
+    2)
+      as_fn_append ac_configure_args1 " '$ac_arg'"
+      if test $ac_must_keep_next = true; then
+	ac_must_keep_next=false # Got value, back to normal.
+      else
+	case $ac_arg in
+	  *=* | --config-cache | -C | -disable-* | --disable-* \
+	  | -enable-* | --enable-* | -gas | --g* | -nfp | --nf* \
+	  | -q | -quiet | --q* | -silent | --sil* | -v | -verb* \
+	  | -with-* | --with-* | -without-* | --without-* | --x)
+	    case "$ac_configure_args0 " in
+	      "$ac_configure_args1"*" '$ac_arg' "* ) continue ;;
+	    esac
+	    ;;
+	  -* ) ac_must_keep_next=true ;;
+	esac
+      fi
+      as_fn_append ac_configure_args " '$ac_arg'"
+      ;;
+    esac
+  done
+done
+{ ac_configure_args0=; unset ac_configure_args0;}
+{ ac_configure_args1=; unset ac_configure_args1;}
+
+# When interrupted or exit'd, cleanup temporary files, and complete
+# config.log.  We remove comments because anyway the quotes in there
+# would cause problems or look ugly.
+# WARNING: Use '\'' to represent an apostrophe within the trap.
+# WARNING: Do not start the trap code with a newline, due to a FreeBSD 4.0 bug.
+trap 'exit_status=$?
+  # Save into config.log some information that might help in debugging.
+  {
+    echo
+
+    $as_echo "## ---------------- ##
+## Cache variables. ##
+## ---------------- ##"
+    echo
+    # The following way of writing the cache mishandles newlines in values,
+(
+  for ac_var in `(set) 2>&1 | sed -n '\''s/^\([a-zA-Z_][a-zA-Z0-9_]*\)=.*/\1/p'\''`; do
+    eval ac_val=\$$ac_var
+    case $ac_val in #(
+    *${as_nl}*)
+      case $ac_var in #(
+      *_cv_*) { $as_echo "$as_me:${as_lineno-$LINENO}: WARNING: cache variable $ac_var contains a newline" >&5
+$as_echo "$as_me: WARNING: cache variable $ac_var contains a newline" >&2;} ;;
+      esac
+      case $ac_var in #(
+      _ | IFS | as_nl) ;; #(
+      BASH_ARGV | BASH_SOURCE) eval $ac_var= ;; #(
+      *) { eval $ac_var=; unset $ac_var;} ;;
+      esac ;;
+    esac
+  done
+  (set) 2>&1 |
+    case $as_nl`(ac_space='\'' '\''; set) 2>&1` in #(
+    *${as_nl}ac_space=\ *)
+      sed -n \
+	"s/'\''/'\''\\\\'\'''\''/g;
+	  s/^\\([_$as_cr_alnum]*_cv_[_$as_cr_alnum]*\\)=\\(.*\\)/\\1='\''\\2'\''/p"
+      ;; #(
+    *)
+      sed -n "/^[_$as_cr_alnum]*_cv_[_$as_cr_alnum]*=/p"
+      ;;
+    esac |
+    sort
+)
+    echo
+
+    $as_echo "## ----------------- ##
+## Output variables. ##
+## ----------------- ##"
+    echo
+    for ac_var in $ac_subst_vars
+    do
+      eval ac_val=\$$ac_var
+      case $ac_val in
+      *\'\''*) ac_val=`$as_echo "$ac_val" | sed "s/'\''/'\''\\\\\\\\'\'''\''/g"`;;
+      esac
+      $as_echo "$ac_var='\''$ac_val'\''"
+    done | sort
+    echo
+
+    if test -n "$ac_subst_files"; then
+      $as_echo "## ------------------- ##
+## File substitutions. ##
+## ------------------- ##"
+      echo
+      for ac_var in $ac_subst_files
+      do
+	eval ac_val=\$$ac_var
+	case $ac_val in
+	*\'\''*) ac_val=`$as_echo "$ac_val" | sed "s/'\''/'\''\\\\\\\\'\'''\''/g"`;;
+	esac
+	$as_echo "$ac_var='\''$ac_val'\''"
+      done | sort
+      echo
+    fi
+
+    if test -s confdefs.h; then
+      $as_echo "## ----------- ##
+## confdefs.h. ##
+## ----------- ##"
+      echo
+      cat confdefs.h
+      echo
+    fi
+    test "$ac_signal" != 0 &&
+      $as_echo "$as_me: caught signal $ac_signal"
+    $as_echo "$as_me: exit $exit_status"
+  } >&5
+  rm -f core *.core core.conftest.* &&
+    rm -f -r conftest* confdefs* conf$$* $ac_clean_files &&
+    exit $exit_status
+' 0
+for ac_signal in 1 2 13 15; do
+  trap 'ac_signal='$ac_signal'; as_fn_exit 1' $ac_signal
+done
+ac_signal=0
+
+# confdefs.h avoids OS command line length limits that DEFS can exceed.
+rm -f -r conftest* confdefs.h
+
+$as_echo "/* confdefs.h */" > confdefs.h
+
+# Predefined preprocessor variables.
+
+cat >>confdefs.h <<_ACEOF
+#define PACKAGE_NAME "$PACKAGE_NAME"
+_ACEOF
+
+cat >>confdefs.h <<_ACEOF
+#define PACKAGE_TARNAME "$PACKAGE_TARNAME"
+_ACEOF
+
+cat >>confdefs.h <<_ACEOF
+#define PACKAGE_VERSION "$PACKAGE_VERSION"
+_ACEOF
+
+cat >>confdefs.h <<_ACEOF
+#define PACKAGE_STRING "$PACKAGE_STRING"
+_ACEOF
+
+cat >>confdefs.h <<_ACEOF
+#define PACKAGE_BUGREPORT "$PACKAGE_BUGREPORT"
+_ACEOF
+
+cat >>confdefs.h <<_ACEOF
+#define PACKAGE_URL "$PACKAGE_URL"
+_ACEOF
+
+
+# Let the site file select an alternate cache file if it wants to.
+# Prefer an explicitly selected file to automatically selected ones.
+ac_site_file1=NONE
+ac_site_file2=NONE
+if test -n "$CONFIG_SITE"; then
+  # We do not want a PATH search for config.site.
+  case $CONFIG_SITE in #((
+    -*)  ac_site_file1=./$CONFIG_SITE;;
+    */*) ac_site_file1=$CONFIG_SITE;;
+    *)   ac_site_file1=./$CONFIG_SITE;;
+  esac
+elif test "x$prefix" != xNONE; then
+  ac_site_file1=$prefix/share/config.site
+  ac_site_file2=$prefix/etc/config.site
+else
+  ac_site_file1=$ac_default_prefix/share/config.site
+  ac_site_file2=$ac_default_prefix/etc/config.site
+fi
+for ac_site_file in "$ac_site_file1" "$ac_site_file2"
+do
+  test "x$ac_site_file" = xNONE && continue
+  if test /dev/null != "$ac_site_file" && test -r "$ac_site_file"; then
+    { $as_echo "$as_me:${as_lineno-$LINENO}: loading site script $ac_site_file" >&5
+$as_echo "$as_me: loading site script $ac_site_file" >&6;}
+    sed 's/^/| /' "$ac_site_file" >&5
+    . "$ac_site_file" \
+      || { { $as_echo "$as_me:${as_lineno-$LINENO}: error: in \`$ac_pwd':" >&5
+$as_echo "$as_me: error: in \`$ac_pwd':" >&2;}
+as_fn_error $? "failed to load site script $ac_site_file
+See \`config.log' for more details" "$LINENO" 5; }
+  fi
+done
+
+if test -r "$cache_file"; then
+  # Some versions of bash will fail to source /dev/null (special files
+  # actually), so we avoid doing that.  DJGPP emulates it as a regular file.
+  if test /dev/null != "$cache_file" && test -f "$cache_file"; then
+    { $as_echo "$as_me:${as_lineno-$LINENO}: loading cache $cache_file" >&5
+$as_echo "$as_me: loading cache $cache_file" >&6;}
+    case $cache_file in
+      [\\/]* | ?:[\\/]* ) . "$cache_file";;
+      *)                      . "./$cache_file";;
+    esac
+  fi
+else
+  { $as_echo "$as_me:${as_lineno-$LINENO}: creating cache $cache_file" >&5
+$as_echo "$as_me: creating cache $cache_file" >&6;}
+  >$cache_file
+fi
+
+# Check that the precious variables saved in the cache have kept the same
+# value.
+ac_cache_corrupted=false
+for ac_var in $ac_precious_vars; do
+  eval ac_old_set=\$ac_cv_env_${ac_var}_set
+  eval ac_new_set=\$ac_env_${ac_var}_set
+  eval ac_old_val=\$ac_cv_env_${ac_var}_value
+  eval ac_new_val=\$ac_env_${ac_var}_value
+  case $ac_old_set,$ac_new_set in
+    set,)
+      { $as_echo "$as_me:${as_lineno-$LINENO}: error: \`$ac_var' was set to \`$ac_old_val' in the previous run" >&5
+$as_echo "$as_me: error: \`$ac_var' was set to \`$ac_old_val' in the previous run" >&2;}
+      ac_cache_corrupted=: ;;
+    ,set)
+      { $as_echo "$as_me:${as_lineno-$LINENO}: error: \`$ac_var' was not set in the previous run" >&5
+$as_echo "$as_me: error: \`$ac_var' was not set in the previous run" >&2;}
+      ac_cache_corrupted=: ;;
+    ,);;
+    *)
+      if test "x$ac_old_val" != "x$ac_new_val"; then
+	# differences in whitespace do not lead to failure.
+	ac_old_val_w=`echo x $ac_old_val`
+	ac_new_val_w=`echo x $ac_new_val`
+	if test "$ac_old_val_w" != "$ac_new_val_w"; then
+	  { $as_echo "$as_me:${as_lineno-$LINENO}: error: \`$ac_var' has changed since the previous run:" >&5
+$as_echo "$as_me: error: \`$ac_var' has changed since the previous run:" >&2;}
+	  ac_cache_corrupted=:
+	else
+	  { $as_echo "$as_me:${as_lineno-$LINENO}: warning: ignoring whitespace changes in \`$ac_var' since the previous run:" >&5
+$as_echo "$as_me: warning: ignoring whitespace changes in \`$ac_var' since the previous run:" >&2;}
+	  eval $ac_var=\$ac_old_val
+	fi
+	{ $as_echo "$as_me:${as_lineno-$LINENO}:   former value:  \`$ac_old_val'" >&5
+$as_echo "$as_me:   former value:  \`$ac_old_val'" >&2;}
+	{ $as_echo "$as_me:${as_lineno-$LINENO}:   current value: \`$ac_new_val'" >&5
+$as_echo "$as_me:   current value: \`$ac_new_val'" >&2;}
+      fi;;
+  esac
+  # Pass precious variables to config.status.
+  if test "$ac_new_set" = set; then
+    case $ac_new_val in
+    *\'*) ac_arg=$ac_var=`$as_echo "$ac_new_val" | sed "s/'/'\\\\\\\\''/g"` ;;
+    *) ac_arg=$ac_var=$ac_new_val ;;
+    esac
+    case " $ac_configure_args " in
+      *" '$ac_arg' "*) ;; # Avoid dups.  Use of quotes ensures accuracy.
+      *) as_fn_append ac_configure_args " '$ac_arg'" ;;
+    esac
+  fi
+done
+if $ac_cache_corrupted; then
+  { $as_echo "$as_me:${as_lineno-$LINENO}: error: in \`$ac_pwd':" >&5
+$as_echo "$as_me: error: in \`$ac_pwd':" >&2;}
+  { $as_echo "$as_me:${as_lineno-$LINENO}: error: changes in the environment can compromise the build" >&5
+$as_echo "$as_me: error: changes in the environment can compromise the build" >&2;}
+  as_fn_error $? "run \`make distclean' and/or \`rm $cache_file' and start over" "$LINENO" 5
+fi
+## -------------------- ##
+## Main body of script. ##
+## -------------------- ##
+
+ac_ext=c
+ac_cpp='$CPP $CPPFLAGS'
+ac_compile='$CC -c $CFLAGS $CPPFLAGS conftest.$ac_ext >&5'
+ac_link='$CC -o conftest$ac_exeext $CFLAGS $CPPFLAGS $LDFLAGS conftest.$ac_ext $LIBS >&5'
+ac_compiler_gnu=$ac_cv_c_compiler_gnu
+
+
+
+
+ac_aux_dir=
+for ac_dir in "$srcdir" "$srcdir/.." "$srcdir/../.."; do
+  if test -f "$ac_dir/install-sh"; then
+    ac_aux_dir=$ac_dir
+    ac_install_sh="$ac_aux_dir/install-sh -c"
+    break
+  elif test -f "$ac_dir/install.sh"; then
+    ac_aux_dir=$ac_dir
+    ac_install_sh="$ac_aux_dir/install.sh -c"
+    break
+  elif test -f "$ac_dir/shtool"; then
+    ac_aux_dir=$ac_dir
+    ac_install_sh="$ac_aux_dir/shtool install -c"
+    break
+  fi
+done
+if test -z "$ac_aux_dir"; then
+  as_fn_error $? "cannot find install-sh, install.sh, or shtool in \"$srcdir\" \"$srcdir/..\" \"$srcdir/../..\"" "$LINENO" 5
+fi
+
+# These three variables are undocumented and unsupported,
+# and are intended to be withdrawn in a future Autoconf release.
+# They can cause serious problems if a builder's source tree is in a directory
+# whose full name contains unusual characters.
+ac_config_guess="$SHELL $ac_aux_dir/config.guess"  # Please don't use this var.
+ac_config_sub="$SHELL $ac_aux_dir/config.sub"  # Please don't use this var.
+ac_configure="$SHELL $ac_aux_dir/configure"  # Please don't use this var.
+
+
+# Make sure we can run config.sub.
+$SHELL "$ac_aux_dir/config.sub" sun4 >/dev/null 2>&1 ||
+  as_fn_error $? "cannot run $SHELL $ac_aux_dir/config.sub" "$LINENO" 5
+
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking build system type" >&5
+$as_echo_n "checking build system type... " >&6; }
+if ${ac_cv_build+:} false; then :
+  $as_echo_n "(cached) " >&6
+else
+  ac_build_alias=$build_alias
+test "x$ac_build_alias" = x &&
+  ac_build_alias=`$SHELL "$ac_aux_dir/config.guess"`
+test "x$ac_build_alias" = x &&
+  as_fn_error $? "cannot guess build type; you must specify one" "$LINENO" 5
+ac_cv_build=`$SHELL "$ac_aux_dir/config.sub" $ac_build_alias` ||
+  as_fn_error $? "$SHELL $ac_aux_dir/config.sub $ac_build_alias failed" "$LINENO" 5
+
+fi
+{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $ac_cv_build" >&5
+$as_echo "$ac_cv_build" >&6; }
+case $ac_cv_build in
+*-*-*) ;;
+*) as_fn_error $? "invalid value of canonical build" "$LINENO" 5;;
+esac
+build=$ac_cv_build
+ac_save_IFS=$IFS; IFS='-'
+set x $ac_cv_build
+shift
+build_cpu=$1
+build_vendor=$2
+shift; shift
+# Remember, the first character of IFS is used to create $*,
+# except with old shells:
+build_os=$*
+IFS=$ac_save_IFS
+case $build_os in *\ *) build_os=`echo "$build_os" | sed 's/ /-/g'`;; esac
+
+
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking host system type" >&5
+$as_echo_n "checking host system type... " >&6; }
+if ${ac_cv_host+:} false; then :
+  $as_echo_n "(cached) " >&6
+else
+  if test "x$host_alias" = x; then
+  ac_cv_host=$ac_cv_build
+else
+  ac_cv_host=`$SHELL "$ac_aux_dir/config.sub" $host_alias` ||
+    as_fn_error $? "$SHELL $ac_aux_dir/config.sub $host_alias failed" "$LINENO" 5
+fi
+
+fi
+{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $ac_cv_host" >&5
+$as_echo "$ac_cv_host" >&6; }
+case $ac_cv_host in
+*-*-*) ;;
+*) as_fn_error $? "invalid value of canonical host" "$LINENO" 5;;
+esac
+host=$ac_cv_host
+ac_save_IFS=$IFS; IFS='-'
+set x $ac_cv_host
+shift
+host_cpu=$1
+host_vendor=$2
+shift; shift
+# Remember, the first character of IFS is used to create $*,
+# except with old shells:
+host_os=$*
+IFS=$ac_save_IFS
+case $host_os in *\ *) host_os=`echo "$host_os" | sed 's/ /-/g'`;; esac
+
+
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking target system type" >&5
+$as_echo_n "checking target system type... " >&6; }
+if ${ac_cv_target+:} false; then :
+  $as_echo_n "(cached) " >&6
+else
+  if test "x$target_alias" = x; then
+  ac_cv_target=$ac_cv_host
+else
+  ac_cv_target=`$SHELL "$ac_aux_dir/config.sub" $target_alias` ||
+    as_fn_error $? "$SHELL $ac_aux_dir/config.sub $target_alias failed" "$LINENO" 5
+fi
+
+fi
+{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $ac_cv_target" >&5
+$as_echo "$ac_cv_target" >&6; }
+case $ac_cv_target in
+*-*-*) ;;
+*) as_fn_error $? "invalid value of canonical target" "$LINENO" 5;;
+esac
+target=$ac_cv_target
+ac_save_IFS=$IFS; IFS='-'
+set x $ac_cv_target
+shift
+target_cpu=$1
+target_vendor=$2
+shift; shift
+# Remember, the first character of IFS is used to create $*,
+# except with old shells:
+target_os=$*
+IFS=$ac_save_IFS
+case $target_os in *\ *) target_os=`echo "$target_os" | sed 's/ /-/g'`;; esac
+
+
+# The aliases save the names the user supplied, while $host etc.
+# will get canonicalized.
+test -n "$target_alias" &&
+  test "$program_prefix$program_suffix$program_transform_name" = \
+    NONENONEs,x,x, &&
+  program_prefix=${target_alias}-
+
+
+am__api_version='1.13'
+
+# Find a good install program.  We prefer a C program (faster),
+# so one script is as good as another.  But avoid the broken or
+# incompatible versions:
+# SysV /etc/install, /usr/sbin/install
+# SunOS /usr/etc/install
+# IRIX /sbin/install
+# AIX /bin/install
+# AmigaOS /C/install, which installs bootblocks on floppy discs
+# AIX 4 /usr/bin/installbsd, which doesn't work without a -g flag
+# AFS /usr/afsws/bin/install, which mishandles nonexistent args
+# SVR4 /usr/ucb/install, which tries to use the nonexistent group "staff"
+# OS/2's system install, which has a completely different semantic
+# ./install, which can be erroneously created by make from ./install.sh.
+# Reject install programs that cannot install multiple files.
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for a BSD-compatible install" >&5
+$as_echo_n "checking for a BSD-compatible install... " >&6; }
+if test -z "$INSTALL"; then
+if ${ac_cv_path_install+:} false; then :
+  $as_echo_n "(cached) " >&6
+else
+  as_save_IFS=$IFS; IFS=$PATH_SEPARATOR
+for as_dir in $PATH
+do
+  IFS=$as_save_IFS
+  test -z "$as_dir" && as_dir=.
+    # Account for people who put trailing slashes in PATH elements.
+case $as_dir/ in #((
+  ./ | .// | /[cC]/* | \
+  /etc/* | /usr/sbin/* | /usr/etc/* | /sbin/* | /usr/afsws/bin/* | \
+  ?:[\\/]os2[\\/]install[\\/]* | ?:[\\/]OS2[\\/]INSTALL[\\/]* | \
+  /usr/ucb/* ) ;;
+  *)
+    # OSF1 and SCO ODT 3.0 have their own names for install.
+    # Don't use installbsd from OSF since it installs stuff as root
+    # by default.
+    for ac_prog in ginstall scoinst install; do
+      for ac_exec_ext in '' $ac_executable_extensions; do
+	if as_fn_executable_p "$as_dir/$ac_prog$ac_exec_ext"; then
+	  if test $ac_prog = install &&
+	    grep dspmsg "$as_dir/$ac_prog$ac_exec_ext" >/dev/null 2>&1; then
+	    # AIX install.  It has an incompatible calling convention.
+	    :
+	  elif test $ac_prog = install &&
+	    grep pwplus "$as_dir/$ac_prog$ac_exec_ext" >/dev/null 2>&1; then
+	    # program-specific install script used by HP pwplus--don't use.
+	    :
+	  else
+	    rm -rf conftest.one conftest.two conftest.dir
+	    echo one > conftest.one
+	    echo two > conftest.two
+	    mkdir conftest.dir
+	    if "$as_dir/$ac_prog$ac_exec_ext" -c conftest.one conftest.two "`pwd`/conftest.dir" &&
+	      test -s conftest.one && test -s conftest.two &&
+	      test -s conftest.dir/conftest.one &&
+	      test -s conftest.dir/conftest.two
+	    then
+	      ac_cv_path_install="$as_dir/$ac_prog$ac_exec_ext -c"
+	      break 3
+	    fi
+	  fi
+	fi
+      done
+    done
+    ;;
+esac
+
+  done
+IFS=$as_save_IFS
+
+rm -rf conftest.one conftest.two conftest.dir
+
+fi
+  if test "${ac_cv_path_install+set}" = set; then
+    INSTALL=$ac_cv_path_install
+  else
+    # As a last resort, use the slow shell script.  Don't cache a
+    # value for INSTALL within a source directory, because that will
+    # break other packages using the cache if that directory is
+    # removed, or if the value is a relative name.
+    INSTALL=$ac_install_sh
+  fi
+fi
+{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $INSTALL" >&5
+$as_echo "$INSTALL" >&6; }
+
+# Use test -z because SunOS4 sh mishandles braces in ${var-val}.
+# It thinks the first close brace ends the variable substitution.
+test -z "$INSTALL_PROGRAM" && INSTALL_PROGRAM='${INSTALL}'
+
+test -z "$INSTALL_SCRIPT" && INSTALL_SCRIPT='${INSTALL}'
+
+test -z "$INSTALL_DATA" && INSTALL_DATA='${INSTALL} -m 644'
+
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking whether build environment is sane" >&5
+$as_echo_n "checking whether build environment is sane... " >&6; }
+# Reject unsafe characters in $srcdir or the absolute working directory
+# name.  Accept space and tab only in the latter.
+am_lf='
+'
+case `pwd` in
+  *[\\\"\#\$\&\'\`$am_lf]*)
+    as_fn_error $? "unsafe absolute working directory name" "$LINENO" 5;;
+esac
+case $srcdir in
+  *[\\\"\#\$\&\'\`$am_lf\ \	]*)
+    as_fn_error $? "unsafe srcdir value: '$srcdir'" "$LINENO" 5;;
+esac
+
+# Do 'set' in a subshell so we don't clobber the current shell's
+# arguments.  Must try -L first in case configure is actually a
+# symlink; some systems play weird games with the mod time of symlinks
+# (eg FreeBSD returns the mod time of the symlink's containing
+# directory).
+if (
+   am_has_slept=no
+   for am_try in 1 2; do
+     echo "timestamp, slept: $am_has_slept" > conftest.file
+     set X `ls -Lt "$srcdir/configure" conftest.file 2> /dev/null`
+     if test "$*" = "X"; then
+	# -L didn't work.
+	set X `ls -t "$srcdir/configure" conftest.file`
+     fi
+     if test "$*" != "X $srcdir/configure conftest.file" \
+	&& test "$*" != "X conftest.file $srcdir/configure"; then
+
+	# If neither matched, then we have a broken ls.  This can happen
+	# if, for instance, CONFIG_SHELL is bash and it inherits a
+	# broken ls alias from the environment.  This has actually
+	# happened.  Such a system could not be considered "sane".
+	as_fn_error $? "ls -t appears to fail.  Make sure there is not a broken
+  alias in your environment" "$LINENO" 5
+     fi
+     if test "$2" = conftest.file || test $am_try -eq 2; then
+       break
+     fi
+     # Just in case.
+     sleep 1
+     am_has_slept=yes
+   done
+   test "$2" = conftest.file
+   )
+then
+   # Ok.
+   :
+else
+   as_fn_error $? "newly created file is older than distributed files!
+Check your system clock" "$LINENO" 5
+fi
+{ $as_echo "$as_me:${as_lineno-$LINENO}: result: yes" >&5
+$as_echo "yes" >&6; }
+# If we didn't sleep, we still need to ensure time stamps of config.status and
+# generated files are strictly newer.
+am_sleep_pid=
+if grep 'slept: no' conftest.file >/dev/null 2>&1; then
+  ( sleep 1 ) &
+  am_sleep_pid=$!
+fi
+
+rm -f conftest.file
+
+test "$program_prefix" != NONE &&
+  program_transform_name="s&^&$program_prefix&;$program_transform_name"
+# Use a double $ so make ignores it.
+test "$program_suffix" != NONE &&
+  program_transform_name="s&\$&$program_suffix&;$program_transform_name"
+# Double any \ or $.
+# By default was `s,x,x', remove it if useless.
+ac_script='s/[\\$]/&&/g;s/;s,x,x,$//'
+program_transform_name=`$as_echo "$program_transform_name" | sed "$ac_script"`
+
+# expand $ac_aux_dir to an absolute path
+am_aux_dir=`cd $ac_aux_dir && pwd`
+
+if test x"${MISSING+set}" != xset; then
+  case $am_aux_dir in
+  *\ * | *\	*)
+    MISSING="\${SHELL} \"$am_aux_dir/missing\"" ;;
+  *)
+    MISSING="\${SHELL} $am_aux_dir/missing" ;;
+  esac
+fi
+# Use eval to expand $SHELL
+if eval "$MISSING --is-lightweight"; then
+  am_missing_run="$MISSING "
+else
+  am_missing_run=
+  { $as_echo "$as_me:${as_lineno-$LINENO}: WARNING: 'missing' script is too old or missing" >&5
+$as_echo "$as_me: WARNING: 'missing' script is too old or missing" >&2;}
+fi
+
+if test x"${install_sh}" != xset; then
+  case $am_aux_dir in
+  *\ * | *\	*)
+    install_sh="\${SHELL} '$am_aux_dir/install-sh'" ;;
+  *)
+    install_sh="\${SHELL} $am_aux_dir/install-sh"
+  esac
+fi
+
+# Installed binaries are usually stripped using 'strip' when the user
+# run "make install-strip".  However 'strip' might not be the right
+# tool to use in cross-compilation environments, therefore Automake
+# will honor the 'STRIP' environment variable to overrule this program.
+if test "$cross_compiling" != no; then
+  if test -n "$ac_tool_prefix"; then
+  # Extract the first word of "${ac_tool_prefix}strip", so it can be a program name with args.
+set dummy ${ac_tool_prefix}strip; ac_word=$2
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for $ac_word" >&5
+$as_echo_n "checking for $ac_word... " >&6; }
+if ${ac_cv_prog_STRIP+:} false; then :
+  $as_echo_n "(cached) " >&6
+else
+  if test -n "$STRIP"; then
+  ac_cv_prog_STRIP="$STRIP" # Let the user override the test.
+else
+as_save_IFS=$IFS; IFS=$PATH_SEPARATOR
+for as_dir in $PATH
+do
+  IFS=$as_save_IFS
+  test -z "$as_dir" && as_dir=.
+    for ac_exec_ext in '' $ac_executable_extensions; do
+  if as_fn_executable_p "$as_dir/$ac_word$ac_exec_ext"; then
+    ac_cv_prog_STRIP="${ac_tool_prefix}strip"
+    $as_echo "$as_me:${as_lineno-$LINENO}: found $as_dir/$ac_word$ac_exec_ext" >&5
+    break 2
+  fi
+done
+  done
+IFS=$as_save_IFS
+
+fi
+fi
+STRIP=$ac_cv_prog_STRIP
+if test -n "$STRIP"; then
+  { $as_echo "$as_me:${as_lineno-$LINENO}: result: $STRIP" >&5
+$as_echo "$STRIP" >&6; }
+else
+  { $as_echo "$as_me:${as_lineno-$LINENO}: result: no" >&5
+$as_echo "no" >&6; }
+fi
+
+
+fi
+if test -z "$ac_cv_prog_STRIP"; then
+  ac_ct_STRIP=$STRIP
+  # Extract the first word of "strip", so it can be a program name with args.
+set dummy strip; ac_word=$2
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for $ac_word" >&5
+$as_echo_n "checking for $ac_word... " >&6; }
+if ${ac_cv_prog_ac_ct_STRIP+:} false; then :
+  $as_echo_n "(cached) " >&6
+else
+  if test -n "$ac_ct_STRIP"; then
+  ac_cv_prog_ac_ct_STRIP="$ac_ct_STRIP" # Let the user override the test.
+else
+as_save_IFS=$IFS; IFS=$PATH_SEPARATOR
+for as_dir in $PATH
+do
+  IFS=$as_save_IFS
+  test -z "$as_dir" && as_dir=.
+    for ac_exec_ext in '' $ac_executable_extensions; do
+  if as_fn_executable_p "$as_dir/$ac_word$ac_exec_ext"; then
+    ac_cv_prog_ac_ct_STRIP="strip"
+    $as_echo "$as_me:${as_lineno-$LINENO}: found $as_dir/$ac_word$ac_exec_ext" >&5
+    break 2
+  fi
+done
+  done
+IFS=$as_save_IFS
+
+fi
+fi
+ac_ct_STRIP=$ac_cv_prog_ac_ct_STRIP
+if test -n "$ac_ct_STRIP"; then
+  { $as_echo "$as_me:${as_lineno-$LINENO}: result: $ac_ct_STRIP" >&5
+$as_echo "$ac_ct_STRIP" >&6; }
+else
+  { $as_echo "$as_me:${as_lineno-$LINENO}: result: no" >&5
+$as_echo "no" >&6; }
+fi
+
+  if test "x$ac_ct_STRIP" = x; then
+    STRIP=":"
+  else
+    case $cross_compiling:$ac_tool_warned in
+yes:)
+{ $as_echo "$as_me:${as_lineno-$LINENO}: WARNING: using cross tools not prefixed with host triplet" >&5
+$as_echo "$as_me: WARNING: using cross tools not prefixed with host triplet" >&2;}
+ac_tool_warned=yes ;;
+esac
+    STRIP=$ac_ct_STRIP
+  fi
+else
+  STRIP="$ac_cv_prog_STRIP"
+fi
+
+fi
+INSTALL_STRIP_PROGRAM="\$(install_sh) -c -s"
+
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for a thread-safe mkdir -p" >&5
+$as_echo_n "checking for a thread-safe mkdir -p... " >&6; }
+if test -z "$MKDIR_P"; then
+  if ${ac_cv_path_mkdir+:} false; then :
+  $as_echo_n "(cached) " >&6
+else
+  as_save_IFS=$IFS; IFS=$PATH_SEPARATOR
+for as_dir in $PATH$PATH_SEPARATOR/opt/sfw/bin
+do
+  IFS=$as_save_IFS
+  test -z "$as_dir" && as_dir=.
+    for ac_prog in mkdir gmkdir; do
+	 for ac_exec_ext in '' $ac_executable_extensions; do
+	   as_fn_executable_p "$as_dir/$ac_prog$ac_exec_ext" || continue
+	   case `"$as_dir/$ac_prog$ac_exec_ext" --version 2>&1` in #(
+	     'mkdir (GNU coreutils) '* | \
+	     'mkdir (coreutils) '* | \
+	     'mkdir (fileutils) '4.1*)
+	       ac_cv_path_mkdir=$as_dir/$ac_prog$ac_exec_ext
+	       break 3;;
+	   esac
+	 done
+       done
+  done
+IFS=$as_save_IFS
+
+fi
+
+  test -d ./--version && rmdir ./--version
+  if test "${ac_cv_path_mkdir+set}" = set; then
+    MKDIR_P="$ac_cv_path_mkdir -p"
+  else
+    # As a last resort, use the slow shell script.  Don't cache a
+    # value for MKDIR_P within a source directory, because that will
+    # break other packages using the cache if that directory is
+    # removed, or if the value is a relative name.
+    MKDIR_P="$ac_install_sh -d"
+  fi
+fi
+{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $MKDIR_P" >&5
+$as_echo "$MKDIR_P" >&6; }
+
+for ac_prog in gawk mawk nawk awk
+do
+  # Extract the first word of "$ac_prog", so it can be a program name with args.
+set dummy $ac_prog; ac_word=$2
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for $ac_word" >&5
+$as_echo_n "checking for $ac_word... " >&6; }
+if ${ac_cv_prog_AWK+:} false; then :
+  $as_echo_n "(cached) " >&6
+else
+  if test -n "$AWK"; then
+  ac_cv_prog_AWK="$AWK" # Let the user override the test.
+else
+as_save_IFS=$IFS; IFS=$PATH_SEPARATOR
+for as_dir in $PATH
+do
+  IFS=$as_save_IFS
+  test -z "$as_dir" && as_dir=.
+    for ac_exec_ext in '' $ac_executable_extensions; do
+  if as_fn_executable_p "$as_dir/$ac_word$ac_exec_ext"; then
+    ac_cv_prog_AWK="$ac_prog"
+    $as_echo "$as_me:${as_lineno-$LINENO}: found $as_dir/$ac_word$ac_exec_ext" >&5
+    break 2
+  fi
+done
+  done
+IFS=$as_save_IFS
+
+fi
+fi
+AWK=$ac_cv_prog_AWK
+if test -n "$AWK"; then
+  { $as_echo "$as_me:${as_lineno-$LINENO}: result: $AWK" >&5
+$as_echo "$AWK" >&6; }
+else
+  { $as_echo "$as_me:${as_lineno-$LINENO}: result: no" >&5
+$as_echo "no" >&6; }
+fi
+
+
+  test -n "$AWK" && break
+done
+
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking whether ${MAKE-make} sets \$(MAKE)" >&5
+$as_echo_n "checking whether ${MAKE-make} sets \$(MAKE)... " >&6; }
+set x ${MAKE-make}
+ac_make=`$as_echo "$2" | sed 's/+/p/g; s/[^a-zA-Z0-9_]/_/g'`
+if eval \${ac_cv_prog_make_${ac_make}_set+:} false; then :
+  $as_echo_n "(cached) " >&6
+else
+  cat >conftest.make <<\_ACEOF
+SHELL = /bin/sh
+all:
+	@echo '@@@%%%=$(MAKE)=@@@%%%'
+_ACEOF
+# GNU make sometimes prints "make[1]: Entering ...", which would confuse us.
+case `${MAKE-make} -f conftest.make 2>/dev/null` in
+  *@@@%%%=?*=@@@%%%*)
+    eval ac_cv_prog_make_${ac_make}_set=yes;;
+  *)
+    eval ac_cv_prog_make_${ac_make}_set=no;;
+esac
+rm -f conftest.make
+fi
+if eval test \$ac_cv_prog_make_${ac_make}_set = yes; then
+  { $as_echo "$as_me:${as_lineno-$LINENO}: result: yes" >&5
+$as_echo "yes" >&6; }
+  SET_MAKE=
+else
+  { $as_echo "$as_me:${as_lineno-$LINENO}: result: no" >&5
+$as_echo "no" >&6; }
+  SET_MAKE="MAKE=${MAKE-make}"
+fi
+
+rm -rf .tst 2>/dev/null
+mkdir .tst 2>/dev/null
+if test -d .tst; then
+  am__leading_dot=.
+else
+  am__leading_dot=_
+fi
+rmdir .tst 2>/dev/null
+
+# Check whether --enable-silent-rules was given.
+if test "${enable_silent_rules+set}" = set; then :
+  enableval=$enable_silent_rules;
+fi
+
+case $enable_silent_rules in # (((
+  yes) AM_DEFAULT_VERBOSITY=0;;
+   no) AM_DEFAULT_VERBOSITY=1;;
+    *) AM_DEFAULT_VERBOSITY=1;;
+esac
+am_make=${MAKE-make}
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking whether $am_make supports nested variables" >&5
+$as_echo_n "checking whether $am_make supports nested variables... " >&6; }
+if ${am_cv_make_support_nested_variables+:} false; then :
+  $as_echo_n "(cached) " >&6
+else
+  if $as_echo 'TRUE=$(BAR$(V))
+BAR0=false
+BAR1=true
+V=1
+am__doit:
+	@$(TRUE)
+.PHONY: am__doit' | $am_make -f - >/dev/null 2>&1; then
+  am_cv_make_support_nested_variables=yes
+else
+  am_cv_make_support_nested_variables=no
+fi
+fi
+{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $am_cv_make_support_nested_variables" >&5
+$as_echo "$am_cv_make_support_nested_variables" >&6; }
+if test $am_cv_make_support_nested_variables = yes; then
+    AM_V='$(V)'
+  AM_DEFAULT_V='$(AM_DEFAULT_VERBOSITY)'
+else
+  AM_V=$AM_DEFAULT_VERBOSITY
+  AM_DEFAULT_V=$AM_DEFAULT_VERBOSITY
+fi
+AM_BACKSLASH='\'
+
+if test "`cd $srcdir && pwd`" != "`pwd`"; then
+  # Use -I$(srcdir) only when $(srcdir) != ., so that make's output
+  # is not polluted with repeated "-I."
+  am__isrc=' -I$(srcdir)'
+  # test to see if srcdir already configured
+  if test -f $srcdir/config.status; then
+    as_fn_error $? "source directory already configured; run \"make distclean\" there first" "$LINENO" 5
+  fi
+fi
+
+# test whether we have cygpath
+if test -z "$CYGPATH_W"; then
+  if (cygpath --version) >/dev/null 2>/dev/null; then
+    CYGPATH_W='cygpath -w'
+  else
+    CYGPATH_W=echo
+  fi
+fi
+
+
+# Define the identity of the package.
+ PACKAGE='ccminer'
+ VERSION='2014.03.18'
+
+
+cat >>confdefs.h <<_ACEOF
+#define PACKAGE "$PACKAGE"
+_ACEOF
+
+
+cat >>confdefs.h <<_ACEOF
+#define VERSION "$VERSION"
+_ACEOF
+
+# Some tools Automake needs.
+
+ACLOCAL=${ACLOCAL-"${am_missing_run}aclocal-${am__api_version}"}
+
+
+AUTOCONF=${AUTOCONF-"${am_missing_run}autoconf"}
+
+
+AUTOMAKE=${AUTOMAKE-"${am_missing_run}automake-${am__api_version}"}
+
+
+AUTOHEADER=${AUTOHEADER-"${am_missing_run}autoheader"}
+
+
+MAKEINFO=${MAKEINFO-"${am_missing_run}makeinfo"}
+
+# For better backward compatibility.  To be removed once Automake 1.9.x
+# dies out for good.  For more background, see:
+# <http://lists.gnu.org/archive/html/automake/2012-07/msg00001.html>
+# <http://lists.gnu.org/archive/html/automake/2012-07/msg00014.html>
+mkdir_p='$(MKDIR_P)'
+
+# We need awk for the "check" target.  The system "awk" is bad on
+# some platforms.
+# Always define AMTAR for backward compatibility.  Yes, it's still used
+# in the wild :-(  We should find a proper way to deprecate it ...
+AMTAR='$${TAR-tar}'
+
+
+# We'll loop over all known methods to create a tar archive until one works.
+_am_tools='gnutar  pax cpio none'
+
+am__tar='$${TAR-tar} chof - "$$tardir"' am__untar='$${TAR-tar} xf -'
+
+
+
+
+
+
+ac_config_headers="$ac_config_headers cpuminer-config.h"
+
+
+
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking whether to enable maintainer-specific portions of Makefiles" >&5
+$as_echo_n "checking whether to enable maintainer-specific portions of Makefiles... " >&6; }
+    # Check whether --enable-maintainer-mode was given.
+if test "${enable_maintainer_mode+set}" = set; then :
+  enableval=$enable_maintainer_mode; USE_MAINTAINER_MODE=$enableval
+else
+  USE_MAINTAINER_MODE=no
+fi
+
+  { $as_echo "$as_me:${as_lineno-$LINENO}: result: $USE_MAINTAINER_MODE" >&5
+$as_echo "$USE_MAINTAINER_MODE" >&6; }
+   if test $USE_MAINTAINER_MODE = yes; then
+  MAINTAINER_MODE_TRUE=
+  MAINTAINER_MODE_FALSE='#'
+else
+  MAINTAINER_MODE_TRUE='#'
+  MAINTAINER_MODE_FALSE=
+fi
+
+  MAINT=$MAINTAINER_MODE_TRUE
+
+
+
+DEPDIR="${am__leading_dot}deps"
+
+ac_config_commands="$ac_config_commands depfiles"
+
+
+am_make=${MAKE-make}
+cat > confinc << 'END'
+am__doit:
+	@echo this is the am__doit target
+.PHONY: am__doit
+END
+# If we don't find an include directive, just comment out the code.
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for style of include used by $am_make" >&5
+$as_echo_n "checking for style of include used by $am_make... " >&6; }
+am__include="#"
+am__quote=
+_am_result=none
+# First try GNU make style include.
+echo "include confinc" > confmf
+# Ignore all kinds of additional output from 'make'.
+case `$am_make -s -f confmf 2> /dev/null` in #(
+*the\ am__doit\ target*)
+  am__include=include
+  am__quote=
+  _am_result=GNU
+  ;;
+esac
+# Now try BSD make style include.
+if test "$am__include" = "#"; then
+   echo '.include "confinc"' > confmf
+   case `$am_make -s -f confmf 2> /dev/null` in #(
+   *the\ am__doit\ target*)
+     am__include=.include
+     am__quote="\""
+     _am_result=BSD
+     ;;
+   esac
+fi
+
+
+{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $_am_result" >&5
+$as_echo "$_am_result" >&6; }
+rm -f confinc confmf
+
+# Check whether --enable-dependency-tracking was given.
+if test "${enable_dependency_tracking+set}" = set; then :
+  enableval=$enable_dependency_tracking;
+fi
+
+if test "x$enable_dependency_tracking" != xno; then
+  am_depcomp="$ac_aux_dir/depcomp"
+  AMDEPBACKSLASH='\'
+  am__nodep='_no'
+fi
+ if test "x$enable_dependency_tracking" != xno; then
+  AMDEP_TRUE=
+  AMDEP_FALSE='#'
+else
+  AMDEP_TRUE='#'
+  AMDEP_FALSE=
+fi
+
+
+ac_ext=c
+ac_cpp='$CPP $CPPFLAGS'
+ac_compile='$CC -c $CFLAGS $CPPFLAGS conftest.$ac_ext >&5'
+ac_link='$CC -o conftest$ac_exeext $CFLAGS $CPPFLAGS $LDFLAGS conftest.$ac_ext $LIBS >&5'
+ac_compiler_gnu=$ac_cv_c_compiler_gnu
+if test -n "$ac_tool_prefix"; then
+  # Extract the first word of "${ac_tool_prefix}gcc", so it can be a program name with args.
+set dummy ${ac_tool_prefix}gcc; ac_word=$2
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for $ac_word" >&5
+$as_echo_n "checking for $ac_word... " >&6; }
+if ${ac_cv_prog_CC+:} false; then :
+  $as_echo_n "(cached) " >&6
+else
+  if test -n "$CC"; then
+  ac_cv_prog_CC="$CC" # Let the user override the test.
+else
+as_save_IFS=$IFS; IFS=$PATH_SEPARATOR
+for as_dir in $PATH
+do
+  IFS=$as_save_IFS
+  test -z "$as_dir" && as_dir=.
+    for ac_exec_ext in '' $ac_executable_extensions; do
+  if as_fn_executable_p "$as_dir/$ac_word$ac_exec_ext"; then
+    ac_cv_prog_CC="${ac_tool_prefix}gcc"
+    $as_echo "$as_me:${as_lineno-$LINENO}: found $as_dir/$ac_word$ac_exec_ext" >&5
+    break 2
+  fi
+done
+  done
+IFS=$as_save_IFS
+
+fi
+fi
+CC=$ac_cv_prog_CC
+if test -n "$CC"; then
+  { $as_echo "$as_me:${as_lineno-$LINENO}: result: $CC" >&5
+$as_echo "$CC" >&6; }
+else
+  { $as_echo "$as_me:${as_lineno-$LINENO}: result: no" >&5
+$as_echo "no" >&6; }
+fi
+
+
+fi
+if test -z "$ac_cv_prog_CC"; then
+  ac_ct_CC=$CC
+  # Extract the first word of "gcc", so it can be a program name with args.
+set dummy gcc; ac_word=$2
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for $ac_word" >&5
+$as_echo_n "checking for $ac_word... " >&6; }
+if ${ac_cv_prog_ac_ct_CC+:} false; then :
+  $as_echo_n "(cached) " >&6
+else
+  if test -n "$ac_ct_CC"; then
+  ac_cv_prog_ac_ct_CC="$ac_ct_CC" # Let the user override the test.
+else
+as_save_IFS=$IFS; IFS=$PATH_SEPARATOR
+for as_dir in $PATH
+do
+  IFS=$as_save_IFS
+  test -z "$as_dir" && as_dir=.
+    for ac_exec_ext in '' $ac_executable_extensions; do
+  if as_fn_executable_p "$as_dir/$ac_word$ac_exec_ext"; then
+    ac_cv_prog_ac_ct_CC="gcc"
+    $as_echo "$as_me:${as_lineno-$LINENO}: found $as_dir/$ac_word$ac_exec_ext" >&5
+    break 2
+  fi
+done
+  done
+IFS=$as_save_IFS
+
+fi
+fi
+ac_ct_CC=$ac_cv_prog_ac_ct_CC
+if test -n "$ac_ct_CC"; then
+  { $as_echo "$as_me:${as_lineno-$LINENO}: result: $ac_ct_CC" >&5
+$as_echo "$ac_ct_CC" >&6; }
+else
+  { $as_echo "$as_me:${as_lineno-$LINENO}: result: no" >&5
+$as_echo "no" >&6; }
+fi
+
+  if test "x$ac_ct_CC" = x; then
+    CC=""
+  else
+    case $cross_compiling:$ac_tool_warned in
+yes:)
+{ $as_echo "$as_me:${as_lineno-$LINENO}: WARNING: using cross tools not prefixed with host triplet" >&5
+$as_echo "$as_me: WARNING: using cross tools not prefixed with host triplet" >&2;}
+ac_tool_warned=yes ;;
+esac
+    CC=$ac_ct_CC
+  fi
+else
+  CC="$ac_cv_prog_CC"
+fi
+
+if test -z "$CC"; then
+          if test -n "$ac_tool_prefix"; then
+    # Extract the first word of "${ac_tool_prefix}cc", so it can be a program name with args.
+set dummy ${ac_tool_prefix}cc; ac_word=$2
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for $ac_word" >&5
+$as_echo_n "checking for $ac_word... " >&6; }
+if ${ac_cv_prog_CC+:} false; then :
+  $as_echo_n "(cached) " >&6
+else
+  if test -n "$CC"; then
+  ac_cv_prog_CC="$CC" # Let the user override the test.
+else
+as_save_IFS=$IFS; IFS=$PATH_SEPARATOR
+for as_dir in $PATH
+do
+  IFS=$as_save_IFS
+  test -z "$as_dir" && as_dir=.
+    for ac_exec_ext in '' $ac_executable_extensions; do
+  if as_fn_executable_p "$as_dir/$ac_word$ac_exec_ext"; then
+    ac_cv_prog_CC="${ac_tool_prefix}cc"
+    $as_echo "$as_me:${as_lineno-$LINENO}: found $as_dir/$ac_word$ac_exec_ext" >&5
+    break 2
+  fi
+done
+  done
+IFS=$as_save_IFS
+
+fi
+fi
+CC=$ac_cv_prog_CC
+if test -n "$CC"; then
+  { $as_echo "$as_me:${as_lineno-$LINENO}: result: $CC" >&5
+$as_echo "$CC" >&6; }
+else
+  { $as_echo "$as_me:${as_lineno-$LINENO}: result: no" >&5
+$as_echo "no" >&6; }
+fi
+
+
+  fi
+fi
+if test -z "$CC"; then
+  # Extract the first word of "cc", so it can be a program name with args.
+set dummy cc; ac_word=$2
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for $ac_word" >&5
+$as_echo_n "checking for $ac_word... " >&6; }
+if ${ac_cv_prog_CC+:} false; then :
+  $as_echo_n "(cached) " >&6
+else
+  if test -n "$CC"; then
+  ac_cv_prog_CC="$CC" # Let the user override the test.
+else
+  ac_prog_rejected=no
+as_save_IFS=$IFS; IFS=$PATH_SEPARATOR
+for as_dir in $PATH
+do
+  IFS=$as_save_IFS
+  test -z "$as_dir" && as_dir=.
+    for ac_exec_ext in '' $ac_executable_extensions; do
+  if as_fn_executable_p "$as_dir/$ac_word$ac_exec_ext"; then
+    if test "$as_dir/$ac_word$ac_exec_ext" = "/usr/ucb/cc"; then
+       ac_prog_rejected=yes
+       continue
+     fi
+    ac_cv_prog_CC="cc"
+    $as_echo "$as_me:${as_lineno-$LINENO}: found $as_dir/$ac_word$ac_exec_ext" >&5
+    break 2
+  fi
+done
+  done
+IFS=$as_save_IFS
+
+if test $ac_prog_rejected = yes; then
+  # We found a bogon in the path, so make sure we never use it.
+  set dummy $ac_cv_prog_CC
+  shift
+  if test $# != 0; then
+    # We chose a different compiler from the bogus one.
+    # However, it has the same basename, so the bogon will be chosen
+    # first if we set CC to just the basename; use the full file name.
+    shift
+    ac_cv_prog_CC="$as_dir/$ac_word${1+' '}$@"
+  fi
+fi
+fi
+fi
+CC=$ac_cv_prog_CC
+if test -n "$CC"; then
+  { $as_echo "$as_me:${as_lineno-$LINENO}: result: $CC" >&5
+$as_echo "$CC" >&6; }
+else
+  { $as_echo "$as_me:${as_lineno-$LINENO}: result: no" >&5
+$as_echo "no" >&6; }
+fi
+
+
+fi
+if test -z "$CC"; then
+  if test -n "$ac_tool_prefix"; then
+  for ac_prog in cl.exe
+  do
+    # Extract the first word of "$ac_tool_prefix$ac_prog", so it can be a program name with args.
+set dummy $ac_tool_prefix$ac_prog; ac_word=$2
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for $ac_word" >&5
+$as_echo_n "checking for $ac_word... " >&6; }
+if ${ac_cv_prog_CC+:} false; then :
+  $as_echo_n "(cached) " >&6
+else
+  if test -n "$CC"; then
+  ac_cv_prog_CC="$CC" # Let the user override the test.
+else
+as_save_IFS=$IFS; IFS=$PATH_SEPARATOR
+for as_dir in $PATH
+do
+  IFS=$as_save_IFS
+  test -z "$as_dir" && as_dir=.
+    for ac_exec_ext in '' $ac_executable_extensions; do
+  if as_fn_executable_p "$as_dir/$ac_word$ac_exec_ext"; then
+    ac_cv_prog_CC="$ac_tool_prefix$ac_prog"
+    $as_echo "$as_me:${as_lineno-$LINENO}: found $as_dir/$ac_word$ac_exec_ext" >&5
+    break 2
+  fi
+done
+  done
+IFS=$as_save_IFS
+
+fi
+fi
+CC=$ac_cv_prog_CC
+if test -n "$CC"; then
+  { $as_echo "$as_me:${as_lineno-$LINENO}: result: $CC" >&5
+$as_echo "$CC" >&6; }
+else
+  { $as_echo "$as_me:${as_lineno-$LINENO}: result: no" >&5
+$as_echo "no" >&6; }
+fi
+
+
+    test -n "$CC" && break
+  done
+fi
+if test -z "$CC"; then
+  ac_ct_CC=$CC
+  for ac_prog in cl.exe
+do
+  # Extract the first word of "$ac_prog", so it can be a program name with args.
+set dummy $ac_prog; ac_word=$2
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for $ac_word" >&5
+$as_echo_n "checking for $ac_word... " >&6; }
+if ${ac_cv_prog_ac_ct_CC+:} false; then :
+  $as_echo_n "(cached) " >&6
+else
+  if test -n "$ac_ct_CC"; then
+  ac_cv_prog_ac_ct_CC="$ac_ct_CC" # Let the user override the test.
+else
+as_save_IFS=$IFS; IFS=$PATH_SEPARATOR
+for as_dir in $PATH
+do
+  IFS=$as_save_IFS
+  test -z "$as_dir" && as_dir=.
+    for ac_exec_ext in '' $ac_executable_extensions; do
+  if as_fn_executable_p "$as_dir/$ac_word$ac_exec_ext"; then
+    ac_cv_prog_ac_ct_CC="$ac_prog"
+    $as_echo "$as_me:${as_lineno-$LINENO}: found $as_dir/$ac_word$ac_exec_ext" >&5
+    break 2
+  fi
+done
+  done
+IFS=$as_save_IFS
+
+fi
+fi
+ac_ct_CC=$ac_cv_prog_ac_ct_CC
+if test -n "$ac_ct_CC"; then
+  { $as_echo "$as_me:${as_lineno-$LINENO}: result: $ac_ct_CC" >&5
+$as_echo "$ac_ct_CC" >&6; }
+else
+  { $as_echo "$as_me:${as_lineno-$LINENO}: result: no" >&5
+$as_echo "no" >&6; }
+fi
+
+
+  test -n "$ac_ct_CC" && break
+done
+
+  if test "x$ac_ct_CC" = x; then
+    CC=""
+  else
+    case $cross_compiling:$ac_tool_warned in
+yes:)
+{ $as_echo "$as_me:${as_lineno-$LINENO}: WARNING: using cross tools not prefixed with host triplet" >&5
+$as_echo "$as_me: WARNING: using cross tools not prefixed with host triplet" >&2;}
+ac_tool_warned=yes ;;
+esac
+    CC=$ac_ct_CC
+  fi
+fi
+
+fi
+
+
+test -z "$CC" && { { $as_echo "$as_me:${as_lineno-$LINENO}: error: in \`$ac_pwd':" >&5
+$as_echo "$as_me: error: in \`$ac_pwd':" >&2;}
+as_fn_error $? "no acceptable C compiler found in \$PATH
+See \`config.log' for more details" "$LINENO" 5; }
+
+# Provide some information about the compiler.
+$as_echo "$as_me:${as_lineno-$LINENO}: checking for C compiler version" >&5
+set X $ac_compile
+ac_compiler=$2
+for ac_option in --version -v -V -qversion; do
+  { { ac_try="$ac_compiler $ac_option >&5"
+case "(($ac_try" in
+  *\"* | *\`* | *\\*) ac_try_echo=\$ac_try;;
+  *) ac_try_echo=$ac_try;;
+esac
+eval ac_try_echo="\"\$as_me:${as_lineno-$LINENO}: $ac_try_echo\""
+$as_echo "$ac_try_echo"; } >&5
+  (eval "$ac_compiler $ac_option >&5") 2>conftest.err
+  ac_status=$?
+  if test -s conftest.err; then
+    sed '10a\
+... rest of stderr output deleted ...
+         10q' conftest.err >conftest.er1
+    cat conftest.er1 >&5
+  fi
+  rm -f conftest.er1 conftest.err
+  $as_echo "$as_me:${as_lineno-$LINENO}: \$? = $ac_status" >&5
+  test $ac_status = 0; }
+done
+
+cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+/* end confdefs.h.  */
+
+int
+main ()
+{
+
+  ;
+  return 0;
+}
+_ACEOF
+ac_clean_files_save=$ac_clean_files
+ac_clean_files="$ac_clean_files a.out a.out.dSYM a.exe b.out"
+# Try to create an executable without -o first, disregard a.out.
+# It will help us diagnose broken compilers, and finding out an intuition
+# of exeext.
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking whether the C compiler works" >&5
+$as_echo_n "checking whether the C compiler works... " >&6; }
+ac_link_default=`$as_echo "$ac_link" | sed 's/ -o *conftest[^ ]*//'`
+
+# The possible output files:
+ac_files="a.out conftest.exe conftest a.exe a_out.exe b.out conftest.*"
+
+ac_rmfiles=
+for ac_file in $ac_files
+do
+  case $ac_file in
+    *.$ac_ext | *.xcoff | *.tds | *.d | *.pdb | *.xSYM | *.bb | *.bbg | *.map | *.inf | *.dSYM | *.o | *.obj ) ;;
+    * ) ac_rmfiles="$ac_rmfiles $ac_file";;
+  esac
+done
+rm -f $ac_rmfiles
+
+if { { ac_try="$ac_link_default"
+case "(($ac_try" in
+  *\"* | *\`* | *\\*) ac_try_echo=\$ac_try;;
+  *) ac_try_echo=$ac_try;;
+esac
+eval ac_try_echo="\"\$as_me:${as_lineno-$LINENO}: $ac_try_echo\""
+$as_echo "$ac_try_echo"; } >&5
+  (eval "$ac_link_default") 2>&5
+  ac_status=$?
+  $as_echo "$as_me:${as_lineno-$LINENO}: \$? = $ac_status" >&5
+  test $ac_status = 0; }; then :
+  # Autoconf-2.13 could set the ac_cv_exeext variable to `no'.
+# So ignore a value of `no', otherwise this would lead to `EXEEXT = no'
+# in a Makefile.  We should not override ac_cv_exeext if it was cached,
+# so that the user can short-circuit this test for compilers unknown to
+# Autoconf.
+for ac_file in $ac_files ''
+do
+  test -f "$ac_file" || continue
+  case $ac_file in
+    *.$ac_ext | *.xcoff | *.tds | *.d | *.pdb | *.xSYM | *.bb | *.bbg | *.map | *.inf | *.dSYM | *.o | *.obj )
+	;;
+    [ab].out )
+	# We found the default executable, but exeext='' is most
+	# certainly right.
+	break;;
+    *.* )
+	if test "${ac_cv_exeext+set}" = set && test "$ac_cv_exeext" != no;
+	then :; else
+	   ac_cv_exeext=`expr "$ac_file" : '[^.]*\(\..*\)'`
+	fi
+	# We set ac_cv_exeext here because the later test for it is not
+	# safe: cross compilers may not add the suffix if given an `-o'
+	# argument, so we may need to know it at that point already.
+	# Even if this section looks crufty: it has the advantage of
+	# actually working.
+	break;;
+    * )
+	break;;
+  esac
+done
+test "$ac_cv_exeext" = no && ac_cv_exeext=
+
+else
+  ac_file=''
+fi
+if test -z "$ac_file"; then :
+  { $as_echo "$as_me:${as_lineno-$LINENO}: result: no" >&5
+$as_echo "no" >&6; }
+$as_echo "$as_me: failed program was:" >&5
+sed 's/^/| /' conftest.$ac_ext >&5
+
+{ { $as_echo "$as_me:${as_lineno-$LINENO}: error: in \`$ac_pwd':" >&5
+$as_echo "$as_me: error: in \`$ac_pwd':" >&2;}
+as_fn_error 77 "C compiler cannot create executables
+See \`config.log' for more details" "$LINENO" 5; }
+else
+  { $as_echo "$as_me:${as_lineno-$LINENO}: result: yes" >&5
+$as_echo "yes" >&6; }
+fi
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for C compiler default output file name" >&5
+$as_echo_n "checking for C compiler default output file name... " >&6; }
+{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $ac_file" >&5
+$as_echo "$ac_file" >&6; }
+ac_exeext=$ac_cv_exeext
+
+rm -f -r a.out a.out.dSYM a.exe conftest$ac_cv_exeext b.out
+ac_clean_files=$ac_clean_files_save
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for suffix of executables" >&5
+$as_echo_n "checking for suffix of executables... " >&6; }
+if { { ac_try="$ac_link"
+case "(($ac_try" in
+  *\"* | *\`* | *\\*) ac_try_echo=\$ac_try;;
+  *) ac_try_echo=$ac_try;;
+esac
+eval ac_try_echo="\"\$as_me:${as_lineno-$LINENO}: $ac_try_echo\""
+$as_echo "$ac_try_echo"; } >&5
+  (eval "$ac_link") 2>&5
+  ac_status=$?
+  $as_echo "$as_me:${as_lineno-$LINENO}: \$? = $ac_status" >&5
+  test $ac_status = 0; }; then :
+  # If both `conftest.exe' and `conftest' are `present' (well, observable)
+# catch `conftest.exe'.  For instance with Cygwin, `ls conftest' will
+# work properly (i.e., refer to `conftest.exe'), while it won't with
+# `rm'.
+for ac_file in conftest.exe conftest conftest.*; do
+  test -f "$ac_file" || continue
+  case $ac_file in
+    *.$ac_ext | *.xcoff | *.tds | *.d | *.pdb | *.xSYM | *.bb | *.bbg | *.map | *.inf | *.dSYM | *.o | *.obj ) ;;
+    *.* ) ac_cv_exeext=`expr "$ac_file" : '[^.]*\(\..*\)'`
+	  break;;
+    * ) break;;
+  esac
+done
+else
+  { { $as_echo "$as_me:${as_lineno-$LINENO}: error: in \`$ac_pwd':" >&5
+$as_echo "$as_me: error: in \`$ac_pwd':" >&2;}
+as_fn_error $? "cannot compute suffix of executables: cannot compile and link
+See \`config.log' for more details" "$LINENO" 5; }
+fi
+rm -f conftest conftest$ac_cv_exeext
+{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $ac_cv_exeext" >&5
+$as_echo "$ac_cv_exeext" >&6; }
+
+rm -f conftest.$ac_ext
+EXEEXT=$ac_cv_exeext
+ac_exeext=$EXEEXT
+cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+/* end confdefs.h.  */
+#include <stdio.h>
+int
+main ()
+{
+FILE *f = fopen ("conftest.out", "w");
+ return ferror (f) || fclose (f) != 0;
+
+  ;
+  return 0;
+}
+_ACEOF
+ac_clean_files="$ac_clean_files conftest.out"
+# Check that the compiler produces executables we can run.  If not, either
+# the compiler is broken, or we cross compile.
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking whether we are cross compiling" >&5
+$as_echo_n "checking whether we are cross compiling... " >&6; }
+if test "$cross_compiling" != yes; then
+  { { ac_try="$ac_link"
+case "(($ac_try" in
+  *\"* | *\`* | *\\*) ac_try_echo=\$ac_try;;
+  *) ac_try_echo=$ac_try;;
+esac
+eval ac_try_echo="\"\$as_me:${as_lineno-$LINENO}: $ac_try_echo\""
+$as_echo "$ac_try_echo"; } >&5
+  (eval "$ac_link") 2>&5
+  ac_status=$?
+  $as_echo "$as_me:${as_lineno-$LINENO}: \$? = $ac_status" >&5
+  test $ac_status = 0; }
+  if { ac_try='./conftest$ac_cv_exeext'
+  { { case "(($ac_try" in
+  *\"* | *\`* | *\\*) ac_try_echo=\$ac_try;;
+  *) ac_try_echo=$ac_try;;
+esac
+eval ac_try_echo="\"\$as_me:${as_lineno-$LINENO}: $ac_try_echo\""
+$as_echo "$ac_try_echo"; } >&5
+  (eval "$ac_try") 2>&5
+  ac_status=$?
+  $as_echo "$as_me:${as_lineno-$LINENO}: \$? = $ac_status" >&5
+  test $ac_status = 0; }; }; then
+    cross_compiling=no
+  else
+    if test "$cross_compiling" = maybe; then
+	cross_compiling=yes
+    else
+	{ { $as_echo "$as_me:${as_lineno-$LINENO}: error: in \`$ac_pwd':" >&5
+$as_echo "$as_me: error: in \`$ac_pwd':" >&2;}
+as_fn_error $? "cannot run C compiled programs.
+If you meant to cross compile, use \`--host'.
+See \`config.log' for more details" "$LINENO" 5; }
+    fi
+  fi
+fi
+{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $cross_compiling" >&5
+$as_echo "$cross_compiling" >&6; }
+
+rm -f conftest.$ac_ext conftest$ac_cv_exeext conftest.out
+ac_clean_files=$ac_clean_files_save
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for suffix of object files" >&5
+$as_echo_n "checking for suffix of object files... " >&6; }
+if ${ac_cv_objext+:} false; then :
+  $as_echo_n "(cached) " >&6
+else
+  cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+/* end confdefs.h.  */
+
+int
+main ()
+{
+
+  ;
+  return 0;
+}
+_ACEOF
+rm -f conftest.o conftest.obj
+if { { ac_try="$ac_compile"
+case "(($ac_try" in
+  *\"* | *\`* | *\\*) ac_try_echo=\$ac_try;;
+  *) ac_try_echo=$ac_try;;
+esac
+eval ac_try_echo="\"\$as_me:${as_lineno-$LINENO}: $ac_try_echo\""
+$as_echo "$ac_try_echo"; } >&5
+  (eval "$ac_compile") 2>&5
+  ac_status=$?
+  $as_echo "$as_me:${as_lineno-$LINENO}: \$? = $ac_status" >&5
+  test $ac_status = 0; }; then :
+  for ac_file in conftest.o conftest.obj conftest.*; do
+  test -f "$ac_file" || continue;
+  case $ac_file in
+    *.$ac_ext | *.xcoff | *.tds | *.d | *.pdb | *.xSYM | *.bb | *.bbg | *.map | *.inf | *.dSYM ) ;;
+    *) ac_cv_objext=`expr "$ac_file" : '.*\.\(.*\)'`
+       break;;
+  esac
+done
+else
+  $as_echo "$as_me: failed program was:" >&5
+sed 's/^/| /' conftest.$ac_ext >&5
+
+{ { $as_echo "$as_me:${as_lineno-$LINENO}: error: in \`$ac_pwd':" >&5
+$as_echo "$as_me: error: in \`$ac_pwd':" >&2;}
+as_fn_error $? "cannot compute suffix of object files: cannot compile
+See \`config.log' for more details" "$LINENO" 5; }
+fi
+rm -f conftest.$ac_cv_objext conftest.$ac_ext
+fi
+{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $ac_cv_objext" >&5
+$as_echo "$ac_cv_objext" >&6; }
+OBJEXT=$ac_cv_objext
+ac_objext=$OBJEXT
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking whether we are using the GNU C compiler" >&5
+$as_echo_n "checking whether we are using the GNU C compiler... " >&6; }
+if ${ac_cv_c_compiler_gnu+:} false; then :
+  $as_echo_n "(cached) " >&6
+else
+  cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+/* end confdefs.h.  */
+
+int
+main ()
+{
+#ifndef __GNUC__
+       choke me
+#endif
+
+  ;
+  return 0;
+}
+_ACEOF
+if ac_fn_c_try_compile "$LINENO"; then :
+  ac_compiler_gnu=yes
+else
+  ac_compiler_gnu=no
+fi
+rm -f core conftest.err conftest.$ac_objext conftest.$ac_ext
+ac_cv_c_compiler_gnu=$ac_compiler_gnu
+
+fi
+{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $ac_cv_c_compiler_gnu" >&5
+$as_echo "$ac_cv_c_compiler_gnu" >&6; }
+if test $ac_compiler_gnu = yes; then
+  GCC=yes
+else
+  GCC=
+fi
+ac_test_CFLAGS=${CFLAGS+set}
+ac_save_CFLAGS=$CFLAGS
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking whether $CC accepts -g" >&5
+$as_echo_n "checking whether $CC accepts -g... " >&6; }
+if ${ac_cv_prog_cc_g+:} false; then :
+  $as_echo_n "(cached) " >&6
+else
+  ac_save_c_werror_flag=$ac_c_werror_flag
+   ac_c_werror_flag=yes
+   ac_cv_prog_cc_g=no
+   CFLAGS="-g"
+   cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+/* end confdefs.h.  */
+
+int
+main ()
+{
+
+  ;
+  return 0;
+}
+_ACEOF
+if ac_fn_c_try_compile "$LINENO"; then :
+  ac_cv_prog_cc_g=yes
+else
+  CFLAGS=""
+      cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+/* end confdefs.h.  */
+
+int
+main ()
+{
+
+  ;
+  return 0;
+}
+_ACEOF
+if ac_fn_c_try_compile "$LINENO"; then :
+
+else
+  ac_c_werror_flag=$ac_save_c_werror_flag
+	 CFLAGS="-g"
+	 cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+/* end confdefs.h.  */
+
+int
+main ()
+{
+
+  ;
+  return 0;
+}
+_ACEOF
+if ac_fn_c_try_compile "$LINENO"; then :
+  ac_cv_prog_cc_g=yes
+fi
+rm -f core conftest.err conftest.$ac_objext conftest.$ac_ext
+fi
+rm -f core conftest.err conftest.$ac_objext conftest.$ac_ext
+fi
+rm -f core conftest.err conftest.$ac_objext conftest.$ac_ext
+   ac_c_werror_flag=$ac_save_c_werror_flag
+fi
+{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $ac_cv_prog_cc_g" >&5
+$as_echo "$ac_cv_prog_cc_g" >&6; }
+if test "$ac_test_CFLAGS" = set; then
+  CFLAGS=$ac_save_CFLAGS
+elif test $ac_cv_prog_cc_g = yes; then
+  if test "$GCC" = yes; then
+    CFLAGS="-g -O2"
+  else
+    CFLAGS="-g"
+  fi
+else
+  if test "$GCC" = yes; then
+    CFLAGS="-O2"
+  else
+    CFLAGS=
+  fi
+fi
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for $CC option to accept ISO C89" >&5
+$as_echo_n "checking for $CC option to accept ISO C89... " >&6; }
+if ${ac_cv_prog_cc_c89+:} false; then :
+  $as_echo_n "(cached) " >&6
+else
+  ac_cv_prog_cc_c89=no
+ac_save_CC=$CC
+cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+/* end confdefs.h.  */
+#include <stdarg.h>
+#include <stdio.h>
+struct stat;
+/* Most of the following tests are stolen from RCS 5.7's src/conf.sh.  */
+struct buf { int x; };
+FILE * (*rcsopen) (struct buf *, struct stat *, int);
+static char *e (p, i)
+     char **p;
+     int i;
+{
+  return p[i];
+}
+static char *f (char * (*g) (char **, int), char **p, ...)
+{
+  char *s;
+  va_list v;
+  va_start (v,p);
+  s = g (p, va_arg (v,int));
+  va_end (v);
+  return s;
+}
+
+/* OSF 4.0 Compaq cc is some sort of almost-ANSI by default.  It has
+   function prototypes and stuff, but not '\xHH' hex character constants.
+   These don't provoke an error unfortunately, instead are silently treated
+   as 'x'.  The following induces an error, until -std is added to get
+   proper ANSI mode.  Curiously '\x00'!='x' always comes out true, for an
+   array size at least.  It's necessary to write '\x00'==0 to get something
+   that's true only with -std.  */
+int osf4_cc_array ['\x00' == 0 ? 1 : -1];
+
+/* IBM C 6 for AIX is almost-ANSI by default, but it replaces macro parameters
+   inside strings and character constants.  */
+#define FOO(x) 'x'
+int xlc6_cc_array[FOO(a) == 'x' ? 1 : -1];
+
+int test (int i, double x);
+struct s1 {int (*f) (int a);};
+struct s2 {int (*f) (double a);};
+int pairnames (int, char **, FILE *(*)(struct buf *, struct stat *, int), int, int);
+int argc;
+char **argv;
+int
+main ()
+{
+return f (e, argv, 0) != argv[0]  ||  f (e, argv, 1) != argv[1];
+  ;
+  return 0;
+}
+_ACEOF
+for ac_arg in '' -qlanglvl=extc89 -qlanglvl=ansi -std \
+	-Ae "-Aa -D_HPUX_SOURCE" "-Xc -D__EXTENSIONS__"
+do
+  CC="$ac_save_CC $ac_arg"
+  if ac_fn_c_try_compile "$LINENO"; then :
+  ac_cv_prog_cc_c89=$ac_arg
+fi
+rm -f core conftest.err conftest.$ac_objext
+  test "x$ac_cv_prog_cc_c89" != "xno" && break
+done
+rm -f conftest.$ac_ext
+CC=$ac_save_CC
+
+fi
+# AC_CACHE_VAL
+case "x$ac_cv_prog_cc_c89" in
+  x)
+    { $as_echo "$as_me:${as_lineno-$LINENO}: result: none needed" >&5
+$as_echo "none needed" >&6; } ;;
+  xno)
+    { $as_echo "$as_me:${as_lineno-$LINENO}: result: unsupported" >&5
+$as_echo "unsupported" >&6; } ;;
+  *)
+    CC="$CC $ac_cv_prog_cc_c89"
+    { $as_echo "$as_me:${as_lineno-$LINENO}: result: $ac_cv_prog_cc_c89" >&5
+$as_echo "$ac_cv_prog_cc_c89" >&6; } ;;
+esac
+if test "x$ac_cv_prog_cc_c89" != xno; then :
+
+fi
+
+ac_ext=c
+ac_cpp='$CPP $CPPFLAGS'
+ac_compile='$CC -c $CFLAGS $CPPFLAGS conftest.$ac_ext >&5'
+ac_link='$CC -o conftest$ac_exeext $CFLAGS $CPPFLAGS $LDFLAGS conftest.$ac_ext $LIBS >&5'
+ac_compiler_gnu=$ac_cv_c_compiler_gnu
+
+depcc="$CC"   am_compiler_list=
+
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking dependency style of $depcc" >&5
+$as_echo_n "checking dependency style of $depcc... " >&6; }
+if ${am_cv_CC_dependencies_compiler_type+:} false; then :
+  $as_echo_n "(cached) " >&6
+else
+  if test -z "$AMDEP_TRUE" && test -f "$am_depcomp"; then
+  # We make a subdir and do the tests there.  Otherwise we can end up
+  # making bogus files that we don't know about and never remove.  For
+  # instance it was reported that on HP-UX the gcc test will end up
+  # making a dummy file named 'D' -- because '-MD' means "put the output
+  # in D".
+  rm -rf conftest.dir
+  mkdir conftest.dir
+  # Copy depcomp to subdir because otherwise we won't find it if we're
+  # using a relative directory.
+  cp "$am_depcomp" conftest.dir
+  cd conftest.dir
+  # We will build objects and dependencies in a subdirectory because
+  # it helps to detect inapplicable dependency modes.  For instance
+  # both Tru64's cc and ICC support -MD to output dependencies as a
+  # side effect of compilation, but ICC will put the dependencies in
+  # the current directory while Tru64 will put them in the object
+  # directory.
+  mkdir sub
+
+  am_cv_CC_dependencies_compiler_type=none
+  if test "$am_compiler_list" = ""; then
+     am_compiler_list=`sed -n 's/^#*\([a-zA-Z0-9]*\))$/\1/p' < ./depcomp`
+  fi
+  am__universal=false
+  case " $depcc " in #(
+     *\ -arch\ *\ -arch\ *) am__universal=true ;;
+     esac
+
+  for depmode in $am_compiler_list; do
+    # Setup a source with many dependencies, because some compilers
+    # like to wrap large dependency lists on column 80 (with \), and
+    # we should not choose a depcomp mode which is confused by this.
+    #
+    # We need to recreate these files for each test, as the compiler may
+    # overwrite some of them when testing with obscure command lines.
+    # This happens at least with the AIX C compiler.
+    : > sub/conftest.c
+    for i in 1 2 3 4 5 6; do
+      echo '#include "conftst'$i'.h"' >> sub/conftest.c
+      # Using ": > sub/conftst$i.h" creates only sub/conftst1.h with
+      # Solaris 10 /bin/sh.
+      echo '/* dummy */' > sub/conftst$i.h
+    done
+    echo "${am__include} ${am__quote}sub/conftest.Po${am__quote}" > confmf
+
+    # We check with '-c' and '-o' for the sake of the "dashmstdout"
+    # mode.  It turns out that the SunPro C++ compiler does not properly
+    # handle '-M -o', and we need to detect this.  Also, some Intel
+    # versions had trouble with output in subdirs.
+    am__obj=sub/conftest.${OBJEXT-o}
+    am__minus_obj="-o $am__obj"
+    case $depmode in
+    gcc)
+      # This depmode causes a compiler race in universal mode.
+      test "$am__universal" = false || continue
+      ;;
+    nosideeffect)
+      # After this tag, mechanisms are not by side-effect, so they'll
+      # only be used when explicitly requested.
+      if test "x$enable_dependency_tracking" = xyes; then
+	continue
+      else
+	break
+      fi
+      ;;
+    msvc7 | msvc7msys | msvisualcpp | msvcmsys)
+      # This compiler won't grok '-c -o', but also, the minuso test has
+      # not run yet.  These depmodes are late enough in the game, and
+      # so weak that their functioning should not be impacted.
+      am__obj=conftest.${OBJEXT-o}
+      am__minus_obj=
+      ;;
+    none) break ;;
+    esac
+    if depmode=$depmode \
+       source=sub/conftest.c object=$am__obj \
+       depfile=sub/conftest.Po tmpdepfile=sub/conftest.TPo \
+       $SHELL ./depcomp $depcc -c $am__minus_obj sub/conftest.c \
+         >/dev/null 2>conftest.err &&
+       grep sub/conftst1.h sub/conftest.Po > /dev/null 2>&1 &&
+       grep sub/conftst6.h sub/conftest.Po > /dev/null 2>&1 &&
+       grep $am__obj sub/conftest.Po > /dev/null 2>&1 &&
+       ${MAKE-make} -s -f confmf > /dev/null 2>&1; then
+      # icc doesn't choke on unknown options, it will just issue warnings
+      # or remarks (even with -Werror).  So we grep stderr for any message
+      # that says an option was ignored or not supported.
+      # When given -MP, icc 7.0 and 7.1 complain thusly:
+      #   icc: Command line warning: ignoring option '-M'; no argument required
+      # The diagnosis changed in icc 8.0:
+      #   icc: Command line remark: option '-MP' not supported
+      if (grep 'ignoring option' conftest.err ||
+          grep 'not supported' conftest.err) >/dev/null 2>&1; then :; else
+        am_cv_CC_dependencies_compiler_type=$depmode
+        break
+      fi
+    fi
+  done
+
+  cd ..
+  rm -rf conftest.dir
+else
+  am_cv_CC_dependencies_compiler_type=none
+fi
+
+fi
+{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $am_cv_CC_dependencies_compiler_type" >&5
+$as_echo "$am_cv_CC_dependencies_compiler_type" >&6; }
+CCDEPMODE=depmode=$am_cv_CC_dependencies_compiler_type
+
+ if
+  test "x$enable_dependency_tracking" != xno \
+  && test "$am_cv_CC_dependencies_compiler_type" = gcc3; then
+  am__fastdepCC_TRUE=
+  am__fastdepCC_FALSE='#'
+else
+  am__fastdepCC_TRUE='#'
+  am__fastdepCC_FALSE=
+fi
+
+
+   { $as_echo "$as_me:${as_lineno-$LINENO}: checking for $CC option to accept ISO C99" >&5
+$as_echo_n "checking for $CC option to accept ISO C99... " >&6; }
+if ${ac_cv_prog_cc_c99+:} false; then :
+  $as_echo_n "(cached) " >&6
+else
+  ac_cv_prog_cc_c99=no
+ac_save_CC=$CC
+cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+/* end confdefs.h.  */
+#include <stdarg.h>
+#include <stdbool.h>
+#include <stdlib.h>
+#include <wchar.h>
+#include <stdio.h>
+
+// Check varargs macros.  These examples are taken from C99 6.10.3.5.
+#define debug(...) fprintf (stderr, __VA_ARGS__)
+#define showlist(...) puts (#__VA_ARGS__)
+#define report(test,...) ((test) ? puts (#test) : printf (__VA_ARGS__))
+static void
+test_varargs_macros (void)
+{
+  int x = 1234;
+  int y = 5678;
+  debug ("Flag");
+  debug ("X = %d\n", x);
+  showlist (The first, second, and third items.);
+  report (x>y, "x is %d but y is %d", x, y);
+}
+
+// Check long long types.
+#define BIG64 18446744073709551615ull
+#define BIG32 4294967295ul
+#define BIG_OK (BIG64 / BIG32 == 4294967297ull && BIG64 % BIG32 == 0)
+#if !BIG_OK
+  your preprocessor is broken;
+#endif
+#if BIG_OK
+#else
+  your preprocessor is broken;
+#endif
+static long long int bignum = -9223372036854775807LL;
+static unsigned long long int ubignum = BIG64;
+
+struct incomplete_array
+{
+  int datasize;
+  double data[];
+};
+
+struct named_init {
+  int number;
+  const wchar_t *name;
+  double average;
+};
+
+typedef const char *ccp;
+
+static inline int
+test_restrict (ccp restrict text)
+{
+  // See if C++-style comments work.
+  // Iterate through items via the restricted pointer.
+  // Also check for declarations in for loops.
+  for (unsigned int i = 0; *(text+i) != '\0'; ++i)
+    continue;
+  return 0;
+}
+
+// Check varargs and va_copy.
+static void
+test_varargs (const char *format, ...)
+{
+  va_list args;
+  va_start (args, format);
+  va_list args_copy;
+  va_copy (args_copy, args);
+
+  const char *str;
+  int number;
+  float fnumber;
+
+  while (*format)
+    {
+      switch (*format++)
+	{
+	case 's': // string
+	  str = va_arg (args_copy, const char *);
+	  break;
+	case 'd': // int
+	  number = va_arg (args_copy, int);
+	  break;
+	case 'f': // float
+	  fnumber = va_arg (args_copy, double);
+	  break;
+	default:
+	  break;
+	}
+    }
+  va_end (args_copy);
+  va_end (args);
+}
+
+int
+main ()
+{
+
+  // Check bool.
+  _Bool success = false;
+
+  // Check restrict.
+  if (test_restrict ("String literal") == 0)
+    success = true;
+  char *restrict newvar = "Another string";
+
+  // Check varargs.
+  test_varargs ("s, d' f .", "string", 65, 34.234);
+  test_varargs_macros ();
+
+  // Check flexible array members.
+  struct incomplete_array *ia =
+    malloc (sizeof (struct incomplete_array) + (sizeof (double) * 10));
+  ia->datasize = 10;
+  for (int i = 0; i < ia->datasize; ++i)
+    ia->data[i] = i * 1.234;
+
+  // Check named initializers.
+  struct named_init ni = {
+    .number = 34,
+    .name = L"Test wide string",
+    .average = 543.34343,
+  };
+
+  ni.number = 58;
+
+  int dynamic_array[ni.number];
+  dynamic_array[ni.number - 1] = 543;
+
+  // work around unused variable warnings
+  return (!success || bignum == 0LL || ubignum == 0uLL || newvar[0] == 'x'
+	  || dynamic_array[ni.number - 1] != 543);
+
+  ;
+  return 0;
+}
+_ACEOF
+for ac_arg in '' -std=gnu99 -std=c99 -c99 -AC99 -D_STDC_C99= -qlanglvl=extc99
+do
+  CC="$ac_save_CC $ac_arg"
+  if ac_fn_c_try_compile "$LINENO"; then :
+  ac_cv_prog_cc_c99=$ac_arg
+fi
+rm -f core conftest.err conftest.$ac_objext
+  test "x$ac_cv_prog_cc_c99" != "xno" && break
+done
+rm -f conftest.$ac_ext
+CC=$ac_save_CC
+
+fi
+# AC_CACHE_VAL
+case "x$ac_cv_prog_cc_c99" in
+  x)
+    { $as_echo "$as_me:${as_lineno-$LINENO}: result: none needed" >&5
+$as_echo "none needed" >&6; } ;;
+  xno)
+    { $as_echo "$as_me:${as_lineno-$LINENO}: result: unsupported" >&5
+$as_echo "unsupported" >&6; } ;;
+  *)
+    CC="$CC $ac_cv_prog_cc_c99"
+    { $as_echo "$as_me:${as_lineno-$LINENO}: result: $ac_cv_prog_cc_c99" >&5
+$as_echo "$ac_cv_prog_cc_c99" >&6; } ;;
+esac
+if test "x$ac_cv_prog_cc_c99" != xno; then :
+
+fi
+
+
+
+ac_ext=c
+ac_cpp='$CPP $CPPFLAGS'
+ac_compile='$CC -c $CFLAGS $CPPFLAGS conftest.$ac_ext >&5'
+ac_link='$CC -o conftest$ac_exeext $CFLAGS $CPPFLAGS $LDFLAGS conftest.$ac_ext $LIBS >&5'
+ac_compiler_gnu=$ac_cv_c_compiler_gnu
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking how to run the C preprocessor" >&5
+$as_echo_n "checking how to run the C preprocessor... " >&6; }
+# On Suns, sometimes $CPP names a directory.
+if test -n "$CPP" && test -d "$CPP"; then
+  CPP=
+fi
+if test -z "$CPP"; then
+  if ${ac_cv_prog_CPP+:} false; then :
+  $as_echo_n "(cached) " >&6
+else
+      # Double quotes because CPP needs to be expanded
+    for CPP in "$CC -E" "$CC -E -traditional-cpp" "/lib/cpp"
+    do
+      ac_preproc_ok=false
+for ac_c_preproc_warn_flag in '' yes
+do
+  # Use a header file that comes with gcc, so configuring glibc
+  # with a fresh cross-compiler works.
+  # Prefer <limits.h> to <assert.h> if __STDC__ is defined, since
+  # <limits.h> exists even on freestanding compilers.
+  # On the NeXT, cc -E runs the code through the compiler's parser,
+  # not just through cpp. "Syntax error" is here to catch this case.
+  cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+/* end confdefs.h.  */
+#ifdef __STDC__
+# include <limits.h>
+#else
+# include <assert.h>
+#endif
+		     Syntax error
+_ACEOF
+if ac_fn_c_try_cpp "$LINENO"; then :
+
+else
+  # Broken: fails on valid input.
+continue
+fi
+rm -f conftest.err conftest.i conftest.$ac_ext
+
+  # OK, works on sane cases.  Now check whether nonexistent headers
+  # can be detected and how.
+  cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+/* end confdefs.h.  */
+#include <ac_nonexistent.h>
+_ACEOF
+if ac_fn_c_try_cpp "$LINENO"; then :
+  # Broken: success on invalid input.
+continue
+else
+  # Passes both tests.
+ac_preproc_ok=:
+break
+fi
+rm -f conftest.err conftest.i conftest.$ac_ext
+
+done
+# Because of `break', _AC_PREPROC_IFELSE's cleaning code was skipped.
+rm -f conftest.i conftest.err conftest.$ac_ext
+if $ac_preproc_ok; then :
+  break
+fi
+
+    done
+    ac_cv_prog_CPP=$CPP
+
+fi
+  CPP=$ac_cv_prog_CPP
+else
+  ac_cv_prog_CPP=$CPP
+fi
+{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $CPP" >&5
+$as_echo "$CPP" >&6; }
+ac_preproc_ok=false
+for ac_c_preproc_warn_flag in '' yes
+do
+  # Use a header file that comes with gcc, so configuring glibc
+  # with a fresh cross-compiler works.
+  # Prefer <limits.h> to <assert.h> if __STDC__ is defined, since
+  # <limits.h> exists even on freestanding compilers.
+  # On the NeXT, cc -E runs the code through the compiler's parser,
+  # not just through cpp. "Syntax error" is here to catch this case.
+  cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+/* end confdefs.h.  */
+#ifdef __STDC__
+# include <limits.h>
+#else
+# include <assert.h>
+#endif
+		     Syntax error
+_ACEOF
+if ac_fn_c_try_cpp "$LINENO"; then :
+
+else
+  # Broken: fails on valid input.
+continue
+fi
+rm -f conftest.err conftest.i conftest.$ac_ext
+
+  # OK, works on sane cases.  Now check whether nonexistent headers
+  # can be detected and how.
+  cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+/* end confdefs.h.  */
+#include <ac_nonexistent.h>
+_ACEOF
+if ac_fn_c_try_cpp "$LINENO"; then :
+  # Broken: success on invalid input.
+continue
+else
+  # Passes both tests.
+ac_preproc_ok=:
+break
+fi
+rm -f conftest.err conftest.i conftest.$ac_ext
+
+done
+# Because of `break', _AC_PREPROC_IFELSE's cleaning code was skipped.
+rm -f conftest.i conftest.err conftest.$ac_ext
+if $ac_preproc_ok; then :
+
+else
+  { { $as_echo "$as_me:${as_lineno-$LINENO}: error: in \`$ac_pwd':" >&5
+$as_echo "$as_me: error: in \`$ac_pwd':" >&2;}
+as_fn_error $? "C preprocessor \"$CPP\" fails sanity check
+See \`config.log' for more details" "$LINENO" 5; }
+fi
+
+ac_ext=c
+ac_cpp='$CPP $CPPFLAGS'
+ac_compile='$CC -c $CFLAGS $CPPFLAGS conftest.$ac_ext >&5'
+ac_link='$CC -o conftest$ac_exeext $CFLAGS $CPPFLAGS $LDFLAGS conftest.$ac_ext $LIBS >&5'
+ac_compiler_gnu=$ac_cv_c_compiler_gnu
+
+
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for grep that handles long lines and -e" >&5
+$as_echo_n "checking for grep that handles long lines and -e... " >&6; }
+if ${ac_cv_path_GREP+:} false; then :
+  $as_echo_n "(cached) " >&6
+else
+  if test -z "$GREP"; then
+  ac_path_GREP_found=false
+  # Loop through the user's path and test for each of PROGNAME-LIST
+  as_save_IFS=$IFS; IFS=$PATH_SEPARATOR
+for as_dir in $PATH$PATH_SEPARATOR/usr/xpg4/bin
+do
+  IFS=$as_save_IFS
+  test -z "$as_dir" && as_dir=.
+    for ac_prog in grep ggrep; do
+    for ac_exec_ext in '' $ac_executable_extensions; do
+      ac_path_GREP="$as_dir/$ac_prog$ac_exec_ext"
+      as_fn_executable_p "$ac_path_GREP" || continue
+# Check for GNU ac_path_GREP and select it if it is found.
+  # Check for GNU $ac_path_GREP
+case `"$ac_path_GREP" --version 2>&1` in
+*GNU*)
+  ac_cv_path_GREP="$ac_path_GREP" ac_path_GREP_found=:;;
+*)
+  ac_count=0
+  $as_echo_n 0123456789 >"conftest.in"
+  while :
+  do
+    cat "conftest.in" "conftest.in" >"conftest.tmp"
+    mv "conftest.tmp" "conftest.in"
+    cp "conftest.in" "conftest.nl"
+    $as_echo 'GREP' >> "conftest.nl"
+    "$ac_path_GREP" -e 'GREP$' -e '-(cannot match)-' < "conftest.nl" >"conftest.out" 2>/dev/null || break
+    diff "conftest.out" "conftest.nl" >/dev/null 2>&1 || break
+    as_fn_arith $ac_count + 1 && ac_count=$as_val
+    if test $ac_count -gt ${ac_path_GREP_max-0}; then
+      # Best one so far, save it but keep looking for a better one
+      ac_cv_path_GREP="$ac_path_GREP"
+      ac_path_GREP_max=$ac_count
+    fi
+    # 10*(2^10) chars as input seems more than enough
+    test $ac_count -gt 10 && break
+  done
+  rm -f conftest.in conftest.tmp conftest.nl conftest.out;;
+esac
+
+      $ac_path_GREP_found && break 3
+    done
+  done
+  done
+IFS=$as_save_IFS
+  if test -z "$ac_cv_path_GREP"; then
+    as_fn_error $? "no acceptable grep could be found in $PATH$PATH_SEPARATOR/usr/xpg4/bin" "$LINENO" 5
+  fi
+else
+  ac_cv_path_GREP=$GREP
+fi
+
+fi
+{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $ac_cv_path_GREP" >&5
+$as_echo "$ac_cv_path_GREP" >&6; }
+ GREP="$ac_cv_path_GREP"
+
+
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for egrep" >&5
+$as_echo_n "checking for egrep... " >&6; }
+if ${ac_cv_path_EGREP+:} false; then :
+  $as_echo_n "(cached) " >&6
+else
+  if echo a | $GREP -E '(a|b)' >/dev/null 2>&1
+   then ac_cv_path_EGREP="$GREP -E"
+   else
+     if test -z "$EGREP"; then
+  ac_path_EGREP_found=false
+  # Loop through the user's path and test for each of PROGNAME-LIST
+  as_save_IFS=$IFS; IFS=$PATH_SEPARATOR
+for as_dir in $PATH$PATH_SEPARATOR/usr/xpg4/bin
+do
+  IFS=$as_save_IFS
+  test -z "$as_dir" && as_dir=.
+    for ac_prog in egrep; do
+    for ac_exec_ext in '' $ac_executable_extensions; do
+      ac_path_EGREP="$as_dir/$ac_prog$ac_exec_ext"
+      as_fn_executable_p "$ac_path_EGREP" || continue
+# Check for GNU ac_path_EGREP and select it if it is found.
+  # Check for GNU $ac_path_EGREP
+case `"$ac_path_EGREP" --version 2>&1` in
+*GNU*)
+  ac_cv_path_EGREP="$ac_path_EGREP" ac_path_EGREP_found=:;;
+*)
+  ac_count=0
+  $as_echo_n 0123456789 >"conftest.in"
+  while :
+  do
+    cat "conftest.in" "conftest.in" >"conftest.tmp"
+    mv "conftest.tmp" "conftest.in"
+    cp "conftest.in" "conftest.nl"
+    $as_echo 'EGREP' >> "conftest.nl"
+    "$ac_path_EGREP" 'EGREP$' < "conftest.nl" >"conftest.out" 2>/dev/null || break
+    diff "conftest.out" "conftest.nl" >/dev/null 2>&1 || break
+    as_fn_arith $ac_count + 1 && ac_count=$as_val
+    if test $ac_count -gt ${ac_path_EGREP_max-0}; then
+      # Best one so far, save it but keep looking for a better one
+      ac_cv_path_EGREP="$ac_path_EGREP"
+      ac_path_EGREP_max=$ac_count
+    fi
+    # 10*(2^10) chars as input seems more than enough
+    test $ac_count -gt 10 && break
+  done
+  rm -f conftest.in conftest.tmp conftest.nl conftest.out;;
+esac
+
+      $ac_path_EGREP_found && break 3
+    done
+  done
+  done
+IFS=$as_save_IFS
+  if test -z "$ac_cv_path_EGREP"; then
+    as_fn_error $? "no acceptable egrep could be found in $PATH$PATH_SEPARATOR/usr/xpg4/bin" "$LINENO" 5
+  fi
+else
+  ac_cv_path_EGREP=$EGREP
+fi
+
+   fi
+fi
+{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $ac_cv_path_EGREP" >&5
+$as_echo "$ac_cv_path_EGREP" >&6; }
+ EGREP="$ac_cv_path_EGREP"
+
+
+if test $ac_cv_c_compiler_gnu = yes; then
+    { $as_echo "$as_me:${as_lineno-$LINENO}: checking whether $CC needs -traditional" >&5
+$as_echo_n "checking whether $CC needs -traditional... " >&6; }
+if ${ac_cv_prog_gcc_traditional+:} false; then :
+  $as_echo_n "(cached) " >&6
+else
+    ac_pattern="Autoconf.*'x'"
+  cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+/* end confdefs.h.  */
+#include <sgtty.h>
+Autoconf TIOCGETP
+_ACEOF
+if (eval "$ac_cpp conftest.$ac_ext") 2>&5 |
+  $EGREP "$ac_pattern" >/dev/null 2>&1; then :
+  ac_cv_prog_gcc_traditional=yes
+else
+  ac_cv_prog_gcc_traditional=no
+fi
+rm -f conftest*
+
+
+  if test $ac_cv_prog_gcc_traditional = no; then
+    cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+/* end confdefs.h.  */
+#include <termio.h>
+Autoconf TCGETA
+_ACEOF
+if (eval "$ac_cpp conftest.$ac_ext") 2>&5 |
+  $EGREP "$ac_pattern" >/dev/null 2>&1; then :
+  ac_cv_prog_gcc_traditional=yes
+fi
+rm -f conftest*
+
+  fi
+fi
+{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $ac_cv_prog_gcc_traditional" >&5
+$as_echo "$ac_cv_prog_gcc_traditional" >&6; }
+  if test $ac_cv_prog_gcc_traditional = yes; then
+    CC="$CC -traditional"
+  fi
+fi
+
+if test "x$CC" != xcc; then
+  { $as_echo "$as_me:${as_lineno-$LINENO}: checking whether $CC and cc understand -c and -o together" >&5
+$as_echo_n "checking whether $CC and cc understand -c and -o together... " >&6; }
+else
+  { $as_echo "$as_me:${as_lineno-$LINENO}: checking whether cc understands -c and -o together" >&5
+$as_echo_n "checking whether cc understands -c and -o together... " >&6; }
+fi
+set dummy $CC; ac_cc=`$as_echo "$2" |
+		      sed 's/[^a-zA-Z0-9_]/_/g;s/^[0-9]/_/'`
+if eval \${ac_cv_prog_cc_${ac_cc}_c_o+:} false; then :
+  $as_echo_n "(cached) " >&6
+else
+  cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+/* end confdefs.h.  */
+
+int
+main ()
+{
+
+  ;
+  return 0;
+}
+_ACEOF
+# Make sure it works both with $CC and with simple cc.
+# We do the test twice because some compilers refuse to overwrite an
+# existing .o file with -o, though they will create one.
+ac_try='$CC -c conftest.$ac_ext -o conftest2.$ac_objext >&5'
+rm -f conftest2.*
+if { { case "(($ac_try" in
+  *\"* | *\`* | *\\*) ac_try_echo=\$ac_try;;
+  *) ac_try_echo=$ac_try;;
+esac
+eval ac_try_echo="\"\$as_me:${as_lineno-$LINENO}: $ac_try_echo\""
+$as_echo "$ac_try_echo"; } >&5
+  (eval "$ac_try") 2>&5
+  ac_status=$?
+  $as_echo "$as_me:${as_lineno-$LINENO}: \$? = $ac_status" >&5
+  test $ac_status = 0; } &&
+   test -f conftest2.$ac_objext && { { case "(($ac_try" in
+  *\"* | *\`* | *\\*) ac_try_echo=\$ac_try;;
+  *) ac_try_echo=$ac_try;;
+esac
+eval ac_try_echo="\"\$as_me:${as_lineno-$LINENO}: $ac_try_echo\""
+$as_echo "$ac_try_echo"; } >&5
+  (eval "$ac_try") 2>&5
+  ac_status=$?
+  $as_echo "$as_me:${as_lineno-$LINENO}: \$? = $ac_status" >&5
+  test $ac_status = 0; };
+then
+  eval ac_cv_prog_cc_${ac_cc}_c_o=yes
+  if test "x$CC" != xcc; then
+    # Test first that cc exists at all.
+    if { ac_try='cc -c conftest.$ac_ext >&5'
+  { { case "(($ac_try" in
+  *\"* | *\`* | *\\*) ac_try_echo=\$ac_try;;
+  *) ac_try_echo=$ac_try;;
+esac
+eval ac_try_echo="\"\$as_me:${as_lineno-$LINENO}: $ac_try_echo\""
+$as_echo "$ac_try_echo"; } >&5
+  (eval "$ac_try") 2>&5
+  ac_status=$?
+  $as_echo "$as_me:${as_lineno-$LINENO}: \$? = $ac_status" >&5
+  test $ac_status = 0; }; }; then
+      ac_try='cc -c conftest.$ac_ext -o conftest2.$ac_objext >&5'
+      rm -f conftest2.*
+      if { { case "(($ac_try" in
+  *\"* | *\`* | *\\*) ac_try_echo=\$ac_try;;
+  *) ac_try_echo=$ac_try;;
+esac
+eval ac_try_echo="\"\$as_me:${as_lineno-$LINENO}: $ac_try_echo\""
+$as_echo "$ac_try_echo"; } >&5
+  (eval "$ac_try") 2>&5
+  ac_status=$?
+  $as_echo "$as_me:${as_lineno-$LINENO}: \$? = $ac_status" >&5
+  test $ac_status = 0; } &&
+	 test -f conftest2.$ac_objext && { { case "(($ac_try" in
+  *\"* | *\`* | *\\*) ac_try_echo=\$ac_try;;
+  *) ac_try_echo=$ac_try;;
+esac
+eval ac_try_echo="\"\$as_me:${as_lineno-$LINENO}: $ac_try_echo\""
+$as_echo "$ac_try_echo"; } >&5
+  (eval "$ac_try") 2>&5
+  ac_status=$?
+  $as_echo "$as_me:${as_lineno-$LINENO}: \$? = $ac_status" >&5
+  test $ac_status = 0; };
+      then
+	# cc works too.
+	:
+      else
+	# cc exists but doesn't like -o.
+	eval ac_cv_prog_cc_${ac_cc}_c_o=no
+      fi
+    fi
+  fi
+else
+  eval ac_cv_prog_cc_${ac_cc}_c_o=no
+fi
+rm -f core conftest*
+
+fi
+if eval test \$ac_cv_prog_cc_${ac_cc}_c_o = yes; then
+  { $as_echo "$as_me:${as_lineno-$LINENO}: result: yes" >&5
+$as_echo "yes" >&6; }
+else
+  { $as_echo "$as_me:${as_lineno-$LINENO}: result: no" >&5
+$as_echo "no" >&6; }
+
+$as_echo "#define NO_MINUS_C_MINUS_O 1" >>confdefs.h
+
+fi
+
+# FIXME: we rely on the cache variable name because
+# there is no other way.
+set dummy $CC
+am_cc=`echo $2 | sed 's/[^a-zA-Z0-9_]/_/g;s/^[0-9]/_/'`
+eval am_t=\$ac_cv_prog_cc_${am_cc}_c_o
+if test "$am_t" != yes; then
+   # Losing compiler, so override with the script.
+   # FIXME: It is wrong to rewrite CC.
+   # But if we don't then we get into trouble of one sort or another.
+   # A longer-term fix would be to have automake use am__CC in this case,
+   # and then we could set am__CC="\$(top_srcdir)/compile \$(CC)"
+   CC="$am_aux_dir/compile $CC"
+fi
+
+
+# By default we simply use the C compiler to build assembly code.
+
+test "${CCAS+set}" = set || CCAS=$CC
+test "${CCASFLAGS+set}" = set || CCASFLAGS=$CFLAGS
+
+
+
+depcc="$CCAS"   am_compiler_list=
+
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking dependency style of $depcc" >&5
+$as_echo_n "checking dependency style of $depcc... " >&6; }
+if ${am_cv_CCAS_dependencies_compiler_type+:} false; then :
+  $as_echo_n "(cached) " >&6
+else
+  if test -z "$AMDEP_TRUE" && test -f "$am_depcomp"; then
+  # We make a subdir and do the tests there.  Otherwise we can end up
+  # making bogus files that we don't know about and never remove.  For
+  # instance it was reported that on HP-UX the gcc test will end up
+  # making a dummy file named 'D' -- because '-MD' means "put the output
+  # in D".
+  rm -rf conftest.dir
+  mkdir conftest.dir
+  # Copy depcomp to subdir because otherwise we won't find it if we're
+  # using a relative directory.
+  cp "$am_depcomp" conftest.dir
+  cd conftest.dir
+  # We will build objects and dependencies in a subdirectory because
+  # it helps to detect inapplicable dependency modes.  For instance
+  # both Tru64's cc and ICC support -MD to output dependencies as a
+  # side effect of compilation, but ICC will put the dependencies in
+  # the current directory while Tru64 will put them in the object
+  # directory.
+  mkdir sub
+
+  am_cv_CCAS_dependencies_compiler_type=none
+  if test "$am_compiler_list" = ""; then
+     am_compiler_list=`sed -n 's/^#*\([a-zA-Z0-9]*\))$/\1/p' < ./depcomp`
+  fi
+  am__universal=false
+
+
+  for depmode in $am_compiler_list; do
+    # Setup a source with many dependencies, because some compilers
+    # like to wrap large dependency lists on column 80 (with \), and
+    # we should not choose a depcomp mode which is confused by this.
+    #
+    # We need to recreate these files for each test, as the compiler may
+    # overwrite some of them when testing with obscure command lines.
+    # This happens at least with the AIX C compiler.
+    : > sub/conftest.c
+    for i in 1 2 3 4 5 6; do
+      echo '#include "conftst'$i'.h"' >> sub/conftest.c
+      # Using ": > sub/conftst$i.h" creates only sub/conftst1.h with
+      # Solaris 10 /bin/sh.
+      echo '/* dummy */' > sub/conftst$i.h
+    done
+    echo "${am__include} ${am__quote}sub/conftest.Po${am__quote}" > confmf
+
+    # We check with '-c' and '-o' for the sake of the "dashmstdout"
+    # mode.  It turns out that the SunPro C++ compiler does not properly
+    # handle '-M -o', and we need to detect this.  Also, some Intel
+    # versions had trouble with output in subdirs.
+    am__obj=sub/conftest.${OBJEXT-o}
+    am__minus_obj="-o $am__obj"
+    case $depmode in
+    gcc)
+      # This depmode causes a compiler race in universal mode.
+      test "$am__universal" = false || continue
+      ;;
+    nosideeffect)
+      # After this tag, mechanisms are not by side-effect, so they'll
+      # only be used when explicitly requested.
+      if test "x$enable_dependency_tracking" = xyes; then
+	continue
+      else
+	break
+      fi
+      ;;
+    msvc7 | msvc7msys | msvisualcpp | msvcmsys)
+      # This compiler won't grok '-c -o', but also, the minuso test has
+      # not run yet.  These depmodes are late enough in the game, and
+      # so weak that their functioning should not be impacted.
+      am__obj=conftest.${OBJEXT-o}
+      am__minus_obj=
+      ;;
+    none) break ;;
+    esac
+    if depmode=$depmode \
+       source=sub/conftest.c object=$am__obj \
+       depfile=sub/conftest.Po tmpdepfile=sub/conftest.TPo \
+       $SHELL ./depcomp $depcc -c $am__minus_obj sub/conftest.c \
+         >/dev/null 2>conftest.err &&
+       grep sub/conftst1.h sub/conftest.Po > /dev/null 2>&1 &&
+       grep sub/conftst6.h sub/conftest.Po > /dev/null 2>&1 &&
+       grep $am__obj sub/conftest.Po > /dev/null 2>&1 &&
+       ${MAKE-make} -s -f confmf > /dev/null 2>&1; then
+      # icc doesn't choke on unknown options, it will just issue warnings
+      # or remarks (even with -Werror).  So we grep stderr for any message
+      # that says an option was ignored or not supported.
+      # When given -MP, icc 7.0 and 7.1 complain thusly:
+      #   icc: Command line warning: ignoring option '-M'; no argument required
+      # The diagnosis changed in icc 8.0:
+      #   icc: Command line remark: option '-MP' not supported
+      if (grep 'ignoring option' conftest.err ||
+          grep 'not supported' conftest.err) >/dev/null 2>&1; then :; else
+        am_cv_CCAS_dependencies_compiler_type=$depmode
+        break
+      fi
+    fi
+  done
+
+  cd ..
+  rm -rf conftest.dir
+else
+  am_cv_CCAS_dependencies_compiler_type=none
+fi
+
+fi
+{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $am_cv_CCAS_dependencies_compiler_type" >&5
+$as_echo "$am_cv_CCAS_dependencies_compiler_type" >&6; }
+CCASDEPMODE=depmode=$am_cv_CCAS_dependencies_compiler_type
+
+ if
+  test "x$enable_dependency_tracking" != xno \
+  && test "$am_cv_CCAS_dependencies_compiler_type" = gcc3; then
+  am__fastdepCCAS_TRUE=
+  am__fastdepCCAS_FALSE='#'
+else
+  am__fastdepCCAS_TRUE='#'
+  am__fastdepCCAS_FALSE=
+fi
+
+
+if test -n "$ac_tool_prefix"; then
+  # Extract the first word of "${ac_tool_prefix}ranlib", so it can be a program name with args.
+set dummy ${ac_tool_prefix}ranlib; ac_word=$2
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for $ac_word" >&5
+$as_echo_n "checking for $ac_word... " >&6; }
+if ${ac_cv_prog_RANLIB+:} false; then :
+  $as_echo_n "(cached) " >&6
+else
+  if test -n "$RANLIB"; then
+  ac_cv_prog_RANLIB="$RANLIB" # Let the user override the test.
+else
+as_save_IFS=$IFS; IFS=$PATH_SEPARATOR
+for as_dir in $PATH
+do
+  IFS=$as_save_IFS
+  test -z "$as_dir" && as_dir=.
+    for ac_exec_ext in '' $ac_executable_extensions; do
+  if as_fn_executable_p "$as_dir/$ac_word$ac_exec_ext"; then
+    ac_cv_prog_RANLIB="${ac_tool_prefix}ranlib"
+    $as_echo "$as_me:${as_lineno-$LINENO}: found $as_dir/$ac_word$ac_exec_ext" >&5
+    break 2
+  fi
+done
+  done
+IFS=$as_save_IFS
+
+fi
+fi
+RANLIB=$ac_cv_prog_RANLIB
+if test -n "$RANLIB"; then
+  { $as_echo "$as_me:${as_lineno-$LINENO}: result: $RANLIB" >&5
+$as_echo "$RANLIB" >&6; }
+else
+  { $as_echo "$as_me:${as_lineno-$LINENO}: result: no" >&5
+$as_echo "no" >&6; }
+fi
+
+
+fi
+if test -z "$ac_cv_prog_RANLIB"; then
+  ac_ct_RANLIB=$RANLIB
+  # Extract the first word of "ranlib", so it can be a program name with args.
+set dummy ranlib; ac_word=$2
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for $ac_word" >&5
+$as_echo_n "checking for $ac_word... " >&6; }
+if ${ac_cv_prog_ac_ct_RANLIB+:} false; then :
+  $as_echo_n "(cached) " >&6
+else
+  if test -n "$ac_ct_RANLIB"; then
+  ac_cv_prog_ac_ct_RANLIB="$ac_ct_RANLIB" # Let the user override the test.
+else
+as_save_IFS=$IFS; IFS=$PATH_SEPARATOR
+for as_dir in $PATH
+do
+  IFS=$as_save_IFS
+  test -z "$as_dir" && as_dir=.
+    for ac_exec_ext in '' $ac_executable_extensions; do
+  if as_fn_executable_p "$as_dir/$ac_word$ac_exec_ext"; then
+    ac_cv_prog_ac_ct_RANLIB="ranlib"
+    $as_echo "$as_me:${as_lineno-$LINENO}: found $as_dir/$ac_word$ac_exec_ext" >&5
+    break 2
+  fi
+done
+  done
+IFS=$as_save_IFS
+
+fi
+fi
+ac_ct_RANLIB=$ac_cv_prog_ac_ct_RANLIB
+if test -n "$ac_ct_RANLIB"; then
+  { $as_echo "$as_me:${as_lineno-$LINENO}: result: $ac_ct_RANLIB" >&5
+$as_echo "$ac_ct_RANLIB" >&6; }
+else
+  { $as_echo "$as_me:${as_lineno-$LINENO}: result: no" >&5
+$as_echo "no" >&6; }
+fi
+
+  if test "x$ac_ct_RANLIB" = x; then
+    RANLIB=":"
+  else
+    case $cross_compiling:$ac_tool_warned in
+yes:)
+{ $as_echo "$as_me:${as_lineno-$LINENO}: WARNING: using cross tools not prefixed with host triplet" >&5
+$as_echo "$as_me: WARNING: using cross tools not prefixed with host triplet" >&2;}
+ac_tool_warned=yes ;;
+esac
+    RANLIB=$ac_ct_RANLIB
+  fi
+else
+  RANLIB="$ac_cv_prog_RANLIB"
+fi
+
+ac_ext=cpp
+ac_cpp='$CXXCPP $CPPFLAGS'
+ac_compile='$CXX -c $CXXFLAGS $CPPFLAGS conftest.$ac_ext >&5'
+ac_link='$CXX -o conftest$ac_exeext $CXXFLAGS $CPPFLAGS $LDFLAGS conftest.$ac_ext $LIBS >&5'
+ac_compiler_gnu=$ac_cv_cxx_compiler_gnu
+if test -z "$CXX"; then
+  if test -n "$CCC"; then
+    CXX=$CCC
+  else
+    if test -n "$ac_tool_prefix"; then
+  for ac_prog in g++ c++ gpp aCC CC cxx cc++ cl.exe FCC KCC RCC xlC_r xlC
+  do
+    # Extract the first word of "$ac_tool_prefix$ac_prog", so it can be a program name with args.
+set dummy $ac_tool_prefix$ac_prog; ac_word=$2
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for $ac_word" >&5
+$as_echo_n "checking for $ac_word... " >&6; }
+if ${ac_cv_prog_CXX+:} false; then :
+  $as_echo_n "(cached) " >&6
+else
+  if test -n "$CXX"; then
+  ac_cv_prog_CXX="$CXX" # Let the user override the test.
+else
+as_save_IFS=$IFS; IFS=$PATH_SEPARATOR
+for as_dir in $PATH
+do
+  IFS=$as_save_IFS
+  test -z "$as_dir" && as_dir=.
+    for ac_exec_ext in '' $ac_executable_extensions; do
+  if as_fn_executable_p "$as_dir/$ac_word$ac_exec_ext"; then
+    ac_cv_prog_CXX="$ac_tool_prefix$ac_prog"
+    $as_echo "$as_me:${as_lineno-$LINENO}: found $as_dir/$ac_word$ac_exec_ext" >&5
+    break 2
+  fi
+done
+  done
+IFS=$as_save_IFS
+
+fi
+fi
+CXX=$ac_cv_prog_CXX
+if test -n "$CXX"; then
+  { $as_echo "$as_me:${as_lineno-$LINENO}: result: $CXX" >&5
+$as_echo "$CXX" >&6; }
+else
+  { $as_echo "$as_me:${as_lineno-$LINENO}: result: no" >&5
+$as_echo "no" >&6; }
+fi
+
+
+    test -n "$CXX" && break
+  done
+fi
+if test -z "$CXX"; then
+  ac_ct_CXX=$CXX
+  for ac_prog in g++ c++ gpp aCC CC cxx cc++ cl.exe FCC KCC RCC xlC_r xlC
+do
+  # Extract the first word of "$ac_prog", so it can be a program name with args.
+set dummy $ac_prog; ac_word=$2
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for $ac_word" >&5
+$as_echo_n "checking for $ac_word... " >&6; }
+if ${ac_cv_prog_ac_ct_CXX+:} false; then :
+  $as_echo_n "(cached) " >&6
+else
+  if test -n "$ac_ct_CXX"; then
+  ac_cv_prog_ac_ct_CXX="$ac_ct_CXX" # Let the user override the test.
+else
+as_save_IFS=$IFS; IFS=$PATH_SEPARATOR
+for as_dir in $PATH
+do
+  IFS=$as_save_IFS
+  test -z "$as_dir" && as_dir=.
+    for ac_exec_ext in '' $ac_executable_extensions; do
+  if as_fn_executable_p "$as_dir/$ac_word$ac_exec_ext"; then
+    ac_cv_prog_ac_ct_CXX="$ac_prog"
+    $as_echo "$as_me:${as_lineno-$LINENO}: found $as_dir/$ac_word$ac_exec_ext" >&5
+    break 2
+  fi
+done
+  done
+IFS=$as_save_IFS
+
+fi
+fi
+ac_ct_CXX=$ac_cv_prog_ac_ct_CXX
+if test -n "$ac_ct_CXX"; then
+  { $as_echo "$as_me:${as_lineno-$LINENO}: result: $ac_ct_CXX" >&5
+$as_echo "$ac_ct_CXX" >&6; }
+else
+  { $as_echo "$as_me:${as_lineno-$LINENO}: result: no" >&5
+$as_echo "no" >&6; }
+fi
+
+
+  test -n "$ac_ct_CXX" && break
+done
+
+  if test "x$ac_ct_CXX" = x; then
+    CXX="g++"
+  else
+    case $cross_compiling:$ac_tool_warned in
+yes:)
+{ $as_echo "$as_me:${as_lineno-$LINENO}: WARNING: using cross tools not prefixed with host triplet" >&5
+$as_echo "$as_me: WARNING: using cross tools not prefixed with host triplet" >&2;}
+ac_tool_warned=yes ;;
+esac
+    CXX=$ac_ct_CXX
+  fi
+fi
+
+  fi
+fi
+# Provide some information about the compiler.
+$as_echo "$as_me:${as_lineno-$LINENO}: checking for C++ compiler version" >&5
+set X $ac_compile
+ac_compiler=$2
+for ac_option in --version -v -V -qversion; do
+  { { ac_try="$ac_compiler $ac_option >&5"
+case "(($ac_try" in
+  *\"* | *\`* | *\\*) ac_try_echo=\$ac_try;;
+  *) ac_try_echo=$ac_try;;
+esac
+eval ac_try_echo="\"\$as_me:${as_lineno-$LINENO}: $ac_try_echo\""
+$as_echo "$ac_try_echo"; } >&5
+  (eval "$ac_compiler $ac_option >&5") 2>conftest.err
+  ac_status=$?
+  if test -s conftest.err; then
+    sed '10a\
+... rest of stderr output deleted ...
+         10q' conftest.err >conftest.er1
+    cat conftest.er1 >&5
+  fi
+  rm -f conftest.er1 conftest.err
+  $as_echo "$as_me:${as_lineno-$LINENO}: \$? = $ac_status" >&5
+  test $ac_status = 0; }
+done
+
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking whether we are using the GNU C++ compiler" >&5
+$as_echo_n "checking whether we are using the GNU C++ compiler... " >&6; }
+if ${ac_cv_cxx_compiler_gnu+:} false; then :
+  $as_echo_n "(cached) " >&6
+else
+  cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+/* end confdefs.h.  */
+
+int
+main ()
+{
+#ifndef __GNUC__
+       choke me
+#endif
+
+  ;
+  return 0;
+}
+_ACEOF
+if ac_fn_cxx_try_compile "$LINENO"; then :
+  ac_compiler_gnu=yes
+else
+  ac_compiler_gnu=no
+fi
+rm -f core conftest.err conftest.$ac_objext conftest.$ac_ext
+ac_cv_cxx_compiler_gnu=$ac_compiler_gnu
+
+fi
+{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $ac_cv_cxx_compiler_gnu" >&5
+$as_echo "$ac_cv_cxx_compiler_gnu" >&6; }
+if test $ac_compiler_gnu = yes; then
+  GXX=yes
+else
+  GXX=
+fi
+ac_test_CXXFLAGS=${CXXFLAGS+set}
+ac_save_CXXFLAGS=$CXXFLAGS
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking whether $CXX accepts -g" >&5
+$as_echo_n "checking whether $CXX accepts -g... " >&6; }
+if ${ac_cv_prog_cxx_g+:} false; then :
+  $as_echo_n "(cached) " >&6
+else
+  ac_save_cxx_werror_flag=$ac_cxx_werror_flag
+   ac_cxx_werror_flag=yes
+   ac_cv_prog_cxx_g=no
+   CXXFLAGS="-g"
+   cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+/* end confdefs.h.  */
+
+int
+main ()
+{
+
+  ;
+  return 0;
+}
+_ACEOF
+if ac_fn_cxx_try_compile "$LINENO"; then :
+  ac_cv_prog_cxx_g=yes
+else
+  CXXFLAGS=""
+      cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+/* end confdefs.h.  */
+
+int
+main ()
+{
+
+  ;
+  return 0;
+}
+_ACEOF
+if ac_fn_cxx_try_compile "$LINENO"; then :
+
+else
+  ac_cxx_werror_flag=$ac_save_cxx_werror_flag
+	 CXXFLAGS="-g"
+	 cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+/* end confdefs.h.  */
+
+int
+main ()
+{
+
+  ;
+  return 0;
+}
+_ACEOF
+if ac_fn_cxx_try_compile "$LINENO"; then :
+  ac_cv_prog_cxx_g=yes
+fi
+rm -f core conftest.err conftest.$ac_objext conftest.$ac_ext
+fi
+rm -f core conftest.err conftest.$ac_objext conftest.$ac_ext
+fi
+rm -f core conftest.err conftest.$ac_objext conftest.$ac_ext
+   ac_cxx_werror_flag=$ac_save_cxx_werror_flag
+fi
+{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $ac_cv_prog_cxx_g" >&5
+$as_echo "$ac_cv_prog_cxx_g" >&6; }
+if test "$ac_test_CXXFLAGS" = set; then
+  CXXFLAGS=$ac_save_CXXFLAGS
+elif test $ac_cv_prog_cxx_g = yes; then
+  if test "$GXX" = yes; then
+    CXXFLAGS="-g -O2"
+  else
+    CXXFLAGS="-g"
+  fi
+else
+  if test "$GXX" = yes; then
+    CXXFLAGS="-O2"
+  else
+    CXXFLAGS=
+  fi
+fi
+ac_ext=c
+ac_cpp='$CPP $CPPFLAGS'
+ac_compile='$CC -c $CFLAGS $CPPFLAGS conftest.$ac_ext >&5'
+ac_link='$CC -o conftest$ac_exeext $CFLAGS $CPPFLAGS $LDFLAGS conftest.$ac_ext $LIBS >&5'
+ac_compiler_gnu=$ac_cv_c_compiler_gnu
+
+depcc="$CXX"  am_compiler_list=
+
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking dependency style of $depcc" >&5
+$as_echo_n "checking dependency style of $depcc... " >&6; }
+if ${am_cv_CXX_dependencies_compiler_type+:} false; then :
+  $as_echo_n "(cached) " >&6
+else
+  if test -z "$AMDEP_TRUE" && test -f "$am_depcomp"; then
+  # We make a subdir and do the tests there.  Otherwise we can end up
+  # making bogus files that we don't know about and never remove.  For
+  # instance it was reported that on HP-UX the gcc test will end up
+  # making a dummy file named 'D' -- because '-MD' means "put the output
+  # in D".
+  rm -rf conftest.dir
+  mkdir conftest.dir
+  # Copy depcomp to subdir because otherwise we won't find it if we're
+  # using a relative directory.
+  cp "$am_depcomp" conftest.dir
+  cd conftest.dir
+  # We will build objects and dependencies in a subdirectory because
+  # it helps to detect inapplicable dependency modes.  For instance
+  # both Tru64's cc and ICC support -MD to output dependencies as a
+  # side effect of compilation, but ICC will put the dependencies in
+  # the current directory while Tru64 will put them in the object
+  # directory.
+  mkdir sub
+
+  am_cv_CXX_dependencies_compiler_type=none
+  if test "$am_compiler_list" = ""; then
+     am_compiler_list=`sed -n 's/^#*\([a-zA-Z0-9]*\))$/\1/p' < ./depcomp`
+  fi
+  am__universal=false
+  case " $depcc " in #(
+     *\ -arch\ *\ -arch\ *) am__universal=true ;;
+     esac
+
+  for depmode in $am_compiler_list; do
+    # Setup a source with many dependencies, because some compilers
+    # like to wrap large dependency lists on column 80 (with \), and
+    # we should not choose a depcomp mode which is confused by this.
+    #
+    # We need to recreate these files for each test, as the compiler may
+    # overwrite some of them when testing with obscure command lines.
+    # This happens at least with the AIX C compiler.
+    : > sub/conftest.c
+    for i in 1 2 3 4 5 6; do
+      echo '#include "conftst'$i'.h"' >> sub/conftest.c
+      # Using ": > sub/conftst$i.h" creates only sub/conftst1.h with
+      # Solaris 10 /bin/sh.
+      echo '/* dummy */' > sub/conftst$i.h
+    done
+    echo "${am__include} ${am__quote}sub/conftest.Po${am__quote}" > confmf
+
+    # We check with '-c' and '-o' for the sake of the "dashmstdout"
+    # mode.  It turns out that the SunPro C++ compiler does not properly
+    # handle '-M -o', and we need to detect this.  Also, some Intel
+    # versions had trouble with output in subdirs.
+    am__obj=sub/conftest.${OBJEXT-o}
+    am__minus_obj="-o $am__obj"
+    case $depmode in
+    gcc)
+      # This depmode causes a compiler race in universal mode.
+      test "$am__universal" = false || continue
+      ;;
+    nosideeffect)
+      # After this tag, mechanisms are not by side-effect, so they'll
+      # only be used when explicitly requested.
+      if test "x$enable_dependency_tracking" = xyes; then
+	continue
+      else
+	break
+      fi
+      ;;
+    msvc7 | msvc7msys | msvisualcpp | msvcmsys)
+      # This compiler won't grok '-c -o', but also, the minuso test has
+      # not run yet.  These depmodes are late enough in the game, and
+      # so weak that their functioning should not be impacted.
+      am__obj=conftest.${OBJEXT-o}
+      am__minus_obj=
+      ;;
+    none) break ;;
+    esac
+    if depmode=$depmode \
+       source=sub/conftest.c object=$am__obj \
+       depfile=sub/conftest.Po tmpdepfile=sub/conftest.TPo \
+       $SHELL ./depcomp $depcc -c $am__minus_obj sub/conftest.c \
+         >/dev/null 2>conftest.err &&
+       grep sub/conftst1.h sub/conftest.Po > /dev/null 2>&1 &&
+       grep sub/conftst6.h sub/conftest.Po > /dev/null 2>&1 &&
+       grep $am__obj sub/conftest.Po > /dev/null 2>&1 &&
+       ${MAKE-make} -s -f confmf > /dev/null 2>&1; then
+      # icc doesn't choke on unknown options, it will just issue warnings
+      # or remarks (even with -Werror).  So we grep stderr for any message
+      # that says an option was ignored or not supported.
+      # When given -MP, icc 7.0 and 7.1 complain thusly:
+      #   icc: Command line warning: ignoring option '-M'; no argument required
+      # The diagnosis changed in icc 8.0:
+      #   icc: Command line remark: option '-MP' not supported
+      if (grep 'ignoring option' conftest.err ||
+          grep 'not supported' conftest.err) >/dev/null 2>&1; then :; else
+        am_cv_CXX_dependencies_compiler_type=$depmode
+        break
+      fi
+    fi
+  done
+
+  cd ..
+  rm -rf conftest.dir
+else
+  am_cv_CXX_dependencies_compiler_type=none
+fi
+
+fi
+{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $am_cv_CXX_dependencies_compiler_type" >&5
+$as_echo "$am_cv_CXX_dependencies_compiler_type" >&6; }
+CXXDEPMODE=depmode=$am_cv_CXX_dependencies_compiler_type
+
+ if
+  test "x$enable_dependency_tracking" != xno \
+  && test "$am_cv_CXX_dependencies_compiler_type" = gcc3; then
+  am__fastdepCXX_TRUE=
+  am__fastdepCXX_FALSE='#'
+else
+  am__fastdepCXX_TRUE='#'
+  am__fastdepCXX_FALSE=
+fi
+
+
+
+  OPENMP_CFLAGS=
+  # Check whether --enable-openmp was given.
+if test "${enable_openmp+set}" = set; then :
+  enableval=$enable_openmp;
+fi
+
+  if test "$enable_openmp" != no; then
+    { $as_echo "$as_me:${as_lineno-$LINENO}: checking for $CC option to support OpenMP" >&5
+$as_echo_n "checking for $CC option to support OpenMP... " >&6; }
+if ${ac_cv_prog_c_openmp+:} false; then :
+  $as_echo_n "(cached) " >&6
+else
+  cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+/* end confdefs.h.  */
+
+#ifndef _OPENMP
+ choke me
+#endif
+#include <omp.h>
+int main () { return omp_get_num_threads (); }
+
+_ACEOF
+if ac_fn_c_try_link "$LINENO"; then :
+  ac_cv_prog_c_openmp='none needed'
+else
+  ac_cv_prog_c_openmp='unsupported'
+	  	  	  	  	  	  	                                	  	  	  	  	  	  for ac_option in -fopenmp -xopenmp -openmp -mp -omp -qsmp=omp -homp \
+                           -Popenmp --openmp; do
+	    ac_save_CFLAGS=$CFLAGS
+	    CFLAGS="$CFLAGS $ac_option"
+	    cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+/* end confdefs.h.  */
+
+#ifndef _OPENMP
+ choke me
+#endif
+#include <omp.h>
+int main () { return omp_get_num_threads (); }
+
+_ACEOF
+if ac_fn_c_try_link "$LINENO"; then :
+  ac_cv_prog_c_openmp=$ac_option
+fi
+rm -f core conftest.err conftest.$ac_objext \
+    conftest$ac_exeext conftest.$ac_ext
+	    CFLAGS=$ac_save_CFLAGS
+	    if test "$ac_cv_prog_c_openmp" != unsupported; then
+	      break
+	    fi
+	  done
+fi
+rm -f core conftest.err conftest.$ac_objext \
+    conftest$ac_exeext conftest.$ac_ext
+fi
+{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $ac_cv_prog_c_openmp" >&5
+$as_echo "$ac_cv_prog_c_openmp" >&6; }
+    case $ac_cv_prog_c_openmp in #(
+      "none needed" | unsupported)
+	;; #(
+      *)
+	OPENMP_CFLAGS=$ac_cv_prog_c_openmp ;;
+    esac
+  fi
+
+
+
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for ANSI C header files" >&5
+$as_echo_n "checking for ANSI C header files... " >&6; }
+if ${ac_cv_header_stdc+:} false; then :
+  $as_echo_n "(cached) " >&6
+else
+  cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+/* end confdefs.h.  */
+#include <stdlib.h>
+#include <stdarg.h>
+#include <string.h>
+#include <float.h>
+
+int
+main ()
+{
+
+  ;
+  return 0;
+}
+_ACEOF
+if ac_fn_c_try_compile "$LINENO"; then :
+  ac_cv_header_stdc=yes
+else
+  ac_cv_header_stdc=no
+fi
+rm -f core conftest.err conftest.$ac_objext conftest.$ac_ext
+
+if test $ac_cv_header_stdc = yes; then
+  # SunOS 4.x string.h does not declare mem*, contrary to ANSI.
+  cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+/* end confdefs.h.  */
+#include <string.h>
+
+_ACEOF
+if (eval "$ac_cpp conftest.$ac_ext") 2>&5 |
+  $EGREP "memchr" >/dev/null 2>&1; then :
+
+else
+  ac_cv_header_stdc=no
+fi
+rm -f conftest*
+
+fi
+
+if test $ac_cv_header_stdc = yes; then
+  # ISC 2.0.2 stdlib.h does not declare free, contrary to ANSI.
+  cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+/* end confdefs.h.  */
+#include <stdlib.h>
+
+_ACEOF
+if (eval "$ac_cpp conftest.$ac_ext") 2>&5 |
+  $EGREP "free" >/dev/null 2>&1; then :
+
+else
+  ac_cv_header_stdc=no
+fi
+rm -f conftest*
+
+fi
+
+if test $ac_cv_header_stdc = yes; then
+  # /bin/cc in Irix-4.0.5 gets non-ANSI ctype macros unless using -ansi.
+  if test "$cross_compiling" = yes; then :
+  :
+else
+  cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+/* end confdefs.h.  */
+#include <ctype.h>
+#include <stdlib.h>
+#if ((' ' & 0x0FF) == 0x020)
+# define ISLOWER(c) ('a' <= (c) && (c) <= 'z')
+# define TOUPPER(c) (ISLOWER(c) ? 'A' + ((c) - 'a') : (c))
+#else
+# define ISLOWER(c) \
+		   (('a' <= (c) && (c) <= 'i') \
+		     || ('j' <= (c) && (c) <= 'r') \
+		     || ('s' <= (c) && (c) <= 'z'))
+# define TOUPPER(c) (ISLOWER(c) ? ((c) | 0x40) : (c))
+#endif
+
+#define XOR(e, f) (((e) && !(f)) || (!(e) && (f)))
+int
+main ()
+{
+  int i;
+  for (i = 0; i < 256; i++)
+    if (XOR (islower (i), ISLOWER (i))
+	|| toupper (i) != TOUPPER (i))
+      return 2;
+  return 0;
+}
+_ACEOF
+if ac_fn_c_try_run "$LINENO"; then :
+
+else
+  ac_cv_header_stdc=no
+fi
+rm -f core *.core core.conftest.* gmon.out bb.out conftest$ac_exeext \
+  conftest.$ac_objext conftest.beam conftest.$ac_ext
+fi
+
+fi
+fi
+{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $ac_cv_header_stdc" >&5
+$as_echo "$ac_cv_header_stdc" >&6; }
+if test $ac_cv_header_stdc = yes; then
+
+$as_echo "#define STDC_HEADERS 1" >>confdefs.h
+
+fi
+
+# On IRIX 5.3, sys/types and inttypes.h are conflicting.
+for ac_header in sys/types.h sys/stat.h stdlib.h string.h memory.h strings.h \
+		  inttypes.h stdint.h unistd.h
+do :
+  as_ac_Header=`$as_echo "ac_cv_header_$ac_header" | $as_tr_sh`
+ac_fn_c_check_header_compile "$LINENO" "$ac_header" "$as_ac_Header" "$ac_includes_default
+"
+if eval test \"x\$"$as_ac_Header"\" = x"yes"; then :
+  cat >>confdefs.h <<_ACEOF
+#define `$as_echo "HAVE_$ac_header" | $as_tr_cpp` 1
+_ACEOF
+
+fi
+
+done
+
+
+for ac_header in sys/endian.h sys/param.h syslog.h
+do :
+  as_ac_Header=`$as_echo "ac_cv_header_$ac_header" | $as_tr_sh`
+ac_fn_c_check_header_mongrel "$LINENO" "$ac_header" "$as_ac_Header" "$ac_includes_default"
+if eval test \"x\$"$as_ac_Header"\" = x"yes"; then :
+  cat >>confdefs.h <<_ACEOF
+#define `$as_echo "HAVE_$ac_header" | $as_tr_cpp` 1
+_ACEOF
+
+fi
+
+done
+
+# sys/sysctl.h requires sys/types.h on FreeBSD
+# sys/sysctl.h requires sys/param.h on OpenBSD
+for ac_header in sys/sysctl.h
+do :
+  ac_fn_c_check_header_compile "$LINENO" "sys/sysctl.h" "ac_cv_header_sys_sysctl_h" "#include <sys/types.h>
+#ifdef HAVE_SYS_PARAM_H
+#include <sys/param.h>
+#endif
+
+"
+if test "x$ac_cv_header_sys_sysctl_h" = xyes; then :
+  cat >>confdefs.h <<_ACEOF
+#define HAVE_SYS_SYSCTL_H 1
+_ACEOF
+
+fi
+
+done
+
+
+ac_fn_c_check_decl "$LINENO" "be32dec" "ac_cv_have_decl_be32dec" "$ac_includes_default
+#ifdef HAVE_SYS_ENDIAN_H
+#include <sys/endian.h>
+#endif
+
+"
+if test "x$ac_cv_have_decl_be32dec" = xyes; then :
+  ac_have_decl=1
+else
+  ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_BE32DEC $ac_have_decl
+_ACEOF
+ac_fn_c_check_decl "$LINENO" "le32dec" "ac_cv_have_decl_le32dec" "$ac_includes_default
+#ifdef HAVE_SYS_ENDIAN_H
+#include <sys/endian.h>
+#endif
+
+"
+if test "x$ac_cv_have_decl_le32dec" = xyes; then :
+  ac_have_decl=1
+else
+  ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_LE32DEC $ac_have_decl
+_ACEOF
+ac_fn_c_check_decl "$LINENO" "be32enc" "ac_cv_have_decl_be32enc" "$ac_includes_default
+#ifdef HAVE_SYS_ENDIAN_H
+#include <sys/endian.h>
+#endif
+
+"
+if test "x$ac_cv_have_decl_be32enc" = xyes; then :
+  ac_have_decl=1
+else
+  ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_BE32ENC $ac_have_decl
+_ACEOF
+ac_fn_c_check_decl "$LINENO" "le32enc" "ac_cv_have_decl_le32enc" "$ac_includes_default
+#ifdef HAVE_SYS_ENDIAN_H
+#include <sys/endian.h>
+#endif
+
+"
+if test "x$ac_cv_have_decl_le32enc" = xyes; then :
+  ac_have_decl=1
+else
+  ac_have_decl=0
+fi
+
+cat >>confdefs.h <<_ACEOF
+#define HAVE_DECL_LE32ENC $ac_have_decl
+_ACEOF
+
+
+ac_fn_c_check_type "$LINENO" "size_t" "ac_cv_type_size_t" "$ac_includes_default"
+if test "x$ac_cv_type_size_t" = xyes; then :
+
+else
+
+cat >>confdefs.h <<_ACEOF
+#define size_t unsigned int
+_ACEOF
+
+fi
+
+# The Ultrix 4.2 mips builtin alloca declared by alloca.h only works
+# for constant arguments.  Useless!
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for working alloca.h" >&5
+$as_echo_n "checking for working alloca.h... " >&6; }
+if ${ac_cv_working_alloca_h+:} false; then :
+  $as_echo_n "(cached) " >&6
+else
+  cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+/* end confdefs.h.  */
+#include <alloca.h>
+int
+main ()
+{
+char *p = (char *) alloca (2 * sizeof (int));
+			  if (p) return 0;
+  ;
+  return 0;
+}
+_ACEOF
+if ac_fn_c_try_link "$LINENO"; then :
+  ac_cv_working_alloca_h=yes
+else
+  ac_cv_working_alloca_h=no
+fi
+rm -f core conftest.err conftest.$ac_objext \
+    conftest$ac_exeext conftest.$ac_ext
+fi
+{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $ac_cv_working_alloca_h" >&5
+$as_echo "$ac_cv_working_alloca_h" >&6; }
+if test $ac_cv_working_alloca_h = yes; then
+
+$as_echo "#define HAVE_ALLOCA_H 1" >>confdefs.h
+
+fi
+
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for alloca" >&5
+$as_echo_n "checking for alloca... " >&6; }
+if ${ac_cv_func_alloca_works+:} false; then :
+  $as_echo_n "(cached) " >&6
+else
+  cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+/* end confdefs.h.  */
+#ifdef __GNUC__
+# define alloca __builtin_alloca
+#else
+# ifdef _MSC_VER
+#  include <malloc.h>
+#  define alloca _alloca
+# else
+#  ifdef HAVE_ALLOCA_H
+#   include <alloca.h>
+#  else
+#   ifdef _AIX
+ #pragma alloca
+#   else
+#    ifndef alloca /* predefined by HP cc +Olibcalls */
+void *alloca (size_t);
+#    endif
+#   endif
+#  endif
+# endif
+#endif
+
+int
+main ()
+{
+char *p = (char *) alloca (1);
+				    if (p) return 0;
+  ;
+  return 0;
+}
+_ACEOF
+if ac_fn_c_try_link "$LINENO"; then :
+  ac_cv_func_alloca_works=yes
+else
+  ac_cv_func_alloca_works=no
+fi
+rm -f core conftest.err conftest.$ac_objext \
+    conftest$ac_exeext conftest.$ac_ext
+fi
+{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $ac_cv_func_alloca_works" >&5
+$as_echo "$ac_cv_func_alloca_works" >&6; }
+
+if test $ac_cv_func_alloca_works = yes; then
+
+$as_echo "#define HAVE_ALLOCA 1" >>confdefs.h
+
+else
+  # The SVR3 libPW and SVR4 libucb both contain incompatible functions
+# that cause trouble.  Some versions do not even contain alloca or
+# contain a buggy version.  If you still want to use their alloca,
+# use ar to extract alloca.o from them instead of compiling alloca.c.
+
+ALLOCA=\${LIBOBJDIR}alloca.$ac_objext
+
+$as_echo "#define C_ALLOCA 1" >>confdefs.h
+
+
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking whether \`alloca.c' needs Cray hooks" >&5
+$as_echo_n "checking whether \`alloca.c' needs Cray hooks... " >&6; }
+if ${ac_cv_os_cray+:} false; then :
+  $as_echo_n "(cached) " >&6
+else
+  cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+/* end confdefs.h.  */
+#if defined CRAY && ! defined CRAY2
+webecray
+#else
+wenotbecray
+#endif
+
+_ACEOF
+if (eval "$ac_cpp conftest.$ac_ext") 2>&5 |
+  $EGREP "webecray" >/dev/null 2>&1; then :
+  ac_cv_os_cray=yes
+else
+  ac_cv_os_cray=no
+fi
+rm -f conftest*
+
+fi
+{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $ac_cv_os_cray" >&5
+$as_echo "$ac_cv_os_cray" >&6; }
+if test $ac_cv_os_cray = yes; then
+  for ac_func in _getb67 GETB67 getb67; do
+    as_ac_var=`$as_echo "ac_cv_func_$ac_func" | $as_tr_sh`
+ac_fn_c_check_func "$LINENO" "$ac_func" "$as_ac_var"
+if eval test \"x\$"$as_ac_var"\" = x"yes"; then :
+
+cat >>confdefs.h <<_ACEOF
+#define CRAY_STACKSEG_END $ac_func
+_ACEOF
+
+    break
+fi
+
+  done
+fi
+
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking stack direction for C alloca" >&5
+$as_echo_n "checking stack direction for C alloca... " >&6; }
+if ${ac_cv_c_stack_direction+:} false; then :
+  $as_echo_n "(cached) " >&6
+else
+  if test "$cross_compiling" = yes; then :
+  ac_cv_c_stack_direction=0
+else
+  cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+/* end confdefs.h.  */
+$ac_includes_default
+int
+find_stack_direction (int *addr, int depth)
+{
+  int dir, dummy = 0;
+  if (! addr)
+    addr = &dummy;
+  *addr = addr < &dummy ? 1 : addr == &dummy ? 0 : -1;
+  dir = depth ? find_stack_direction (addr, depth - 1) : 0;
+  return dir + dummy;
+}
+
+int
+main (int argc, char **argv)
+{
+  return find_stack_direction (0, argc + !argv + 20) < 0;
+}
+_ACEOF
+if ac_fn_c_try_run "$LINENO"; then :
+  ac_cv_c_stack_direction=1
+else
+  ac_cv_c_stack_direction=-1
+fi
+rm -f core *.core core.conftest.* gmon.out bb.out conftest$ac_exeext \
+  conftest.$ac_objext conftest.beam conftest.$ac_ext
+fi
+
+fi
+{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $ac_cv_c_stack_direction" >&5
+$as_echo "$ac_cv_c_stack_direction" >&6; }
+cat >>confdefs.h <<_ACEOF
+#define STACK_DIRECTION $ac_cv_c_stack_direction
+_ACEOF
+
+
+fi
+
+for ac_func in getopt_long
+do :
+  ac_fn_c_check_func "$LINENO" "getopt_long" "ac_cv_func_getopt_long"
+if test "x$ac_cv_func_getopt_long" = xyes; then :
+  cat >>confdefs.h <<_ACEOF
+#define HAVE_GETOPT_LONG 1
+_ACEOF
+
+fi
+done
+
+
+case $target in
+  i*86-*-*)
+    have_x86=true
+    ;;
+  x86_64-*-*)
+    have_x86=true
+    have_x86_64=true
+    ;;
+  amd64-*-*)
+    have_x86=true
+    have_x86_64=true
+    ;;
+esac
+
+PTHREAD_FLAGS="-pthread"
+WS2_LIBS=""
+
+case $target in
+  *-*-mingw*)
+    have_win32=true
+    PTHREAD_FLAGS=""
+    WS2_LIBS="-lws2_32"
+    ;;
+esac
+
+if test x$have_x86 = xtrue
+then
+  { $as_echo "$as_me:${as_lineno-$LINENO}: checking whether we can compile AVX code" >&5
+$as_echo_n "checking whether we can compile AVX code... " >&6; }
+  cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+/* end confdefs.h.  */
+
+int
+main ()
+{
+asm ("vmovdqa %ymm0, %ymm1");
+  ;
+  return 0;
+}
+_ACEOF
+if ac_fn_c_try_compile "$LINENO"; then :
+
+$as_echo "#define USE_AVX 1" >>confdefs.h
+
+    { $as_echo "$as_me:${as_lineno-$LINENO}: result: yes" >&5
+$as_echo "yes" >&6; }
+    { $as_echo "$as_me:${as_lineno-$LINENO}: checking whether we can compile XOP code" >&5
+$as_echo_n "checking whether we can compile XOP code... " >&6; }
+    cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+/* end confdefs.h.  */
+
+int
+main ()
+{
+asm ("vprotd \$7, %xmm0, %xmm1");
+  ;
+  return 0;
+}
+_ACEOF
+if ac_fn_c_try_compile "$LINENO"; then :
+
+$as_echo "#define USE_XOP 1" >>confdefs.h
+
+      { $as_echo "$as_me:${as_lineno-$LINENO}: result: yes" >&5
+$as_echo "yes" >&6; }
+
+else
+  { $as_echo "$as_me:${as_lineno-$LINENO}: result: no" >&5
+$as_echo "no" >&6; }
+      { $as_echo "$as_me:${as_lineno-$LINENO}: WARNING: The assembler does not support the XOP instruction set." >&5
+$as_echo "$as_me: WARNING: The assembler does not support the XOP instruction set." >&2;}
+
+fi
+rm -f core conftest.err conftest.$ac_objext conftest.$ac_ext
+    { $as_echo "$as_me:${as_lineno-$LINENO}: checking whether we can compile AVX2 code" >&5
+$as_echo_n "checking whether we can compile AVX2 code... " >&6; }
+    cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+/* end confdefs.h.  */
+
+int
+main ()
+{
+asm ("vpaddd %ymm0, %ymm1, %ymm2");
+  ;
+  return 0;
+}
+_ACEOF
+if ac_fn_c_try_compile "$LINENO"; then :
+
+$as_echo "#define USE_AVX2 1" >>confdefs.h
+
+      { $as_echo "$as_me:${as_lineno-$LINENO}: result: yes" >&5
+$as_echo "yes" >&6; }
+
+else
+  { $as_echo "$as_me:${as_lineno-$LINENO}: result: no" >&5
+$as_echo "no" >&6; }
+      { $as_echo "$as_me:${as_lineno-$LINENO}: WARNING: The assembler does not support the AVX2 instruction set." >&5
+$as_echo "$as_me: WARNING: The assembler does not support the AVX2 instruction set." >&2;}
+
+fi
+rm -f core conftest.err conftest.$ac_objext conftest.$ac_ext
+
+else
+  { $as_echo "$as_me:${as_lineno-$LINENO}: result: no" >&5
+$as_echo "no" >&6; }
+    { $as_echo "$as_me:${as_lineno-$LINENO}: WARNING: The assembler does not support the AVX instruction set." >&5
+$as_echo "$as_me: WARNING: The assembler does not support the AVX instruction set." >&2;}
+
+fi
+rm -f core conftest.err conftest.$ac_objext conftest.$ac_ext
+fi
+
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for json_loads in -ljansson" >&5
+$as_echo_n "checking for json_loads in -ljansson... " >&6; }
+if ${ac_cv_lib_jansson_json_loads+:} false; then :
+  $as_echo_n "(cached) " >&6
+else
+  ac_check_lib_save_LIBS=$LIBS
+LIBS="-ljansson  $LIBS"
+cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+/* end confdefs.h.  */
+
+/* Override any GCC internal prototype to avoid an error.
+   Use char because int might match the return type of a GCC
+   builtin and then its argument prototype would still apply.  */
+#ifdef __cplusplus
+extern "C"
+#endif
+char json_loads ();
+int
+main ()
+{
+return json_loads ();
+  ;
+  return 0;
+}
+_ACEOF
+if ac_fn_c_try_link "$LINENO"; then :
+  ac_cv_lib_jansson_json_loads=yes
+else
+  ac_cv_lib_jansson_json_loads=no
+fi
+rm -f core conftest.err conftest.$ac_objext \
+    conftest$ac_exeext conftest.$ac_ext
+LIBS=$ac_check_lib_save_LIBS
+fi
+{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $ac_cv_lib_jansson_json_loads" >&5
+$as_echo "$ac_cv_lib_jansson_json_loads" >&6; }
+if test "x$ac_cv_lib_jansson_json_loads" = xyes; then :
+  request_jansson=false
+else
+  request_jansson=true
+fi
+
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for pthread_create in -lpthread" >&5
+$as_echo_n "checking for pthread_create in -lpthread... " >&6; }
+if ${ac_cv_lib_pthread_pthread_create+:} false; then :
+  $as_echo_n "(cached) " >&6
+else
+  ac_check_lib_save_LIBS=$LIBS
+LIBS="-lpthread  $LIBS"
+cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+/* end confdefs.h.  */
+
+/* Override any GCC internal prototype to avoid an error.
+   Use char because int might match the return type of a GCC
+   builtin and then its argument prototype would still apply.  */
+#ifdef __cplusplus
+extern "C"
+#endif
+char pthread_create ();
+int
+main ()
+{
+return pthread_create ();
+  ;
+  return 0;
+}
+_ACEOF
+if ac_fn_c_try_link "$LINENO"; then :
+  ac_cv_lib_pthread_pthread_create=yes
+else
+  ac_cv_lib_pthread_pthread_create=no
+fi
+rm -f core conftest.err conftest.$ac_objext \
+    conftest$ac_exeext conftest.$ac_ext
+LIBS=$ac_check_lib_save_LIBS
+fi
+{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $ac_cv_lib_pthread_pthread_create" >&5
+$as_echo "$ac_cv_lib_pthread_pthread_create" >&6; }
+if test "x$ac_cv_lib_pthread_pthread_create" = xyes; then :
+  PTHREAD_LIBS="-lpthread"
+else
+  { $as_echo "$as_me:${as_lineno-$LINENO}: checking for pthread_create in -lpthreadGC2" >&5
+$as_echo_n "checking for pthread_create in -lpthreadGC2... " >&6; }
+if ${ac_cv_lib_pthreadGC2_pthread_create+:} false; then :
+  $as_echo_n "(cached) " >&6
+else
+  ac_check_lib_save_LIBS=$LIBS
+LIBS="-lpthreadGC2  $LIBS"
+cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+/* end confdefs.h.  */
+
+/* Override any GCC internal prototype to avoid an error.
+   Use char because int might match the return type of a GCC
+   builtin and then its argument prototype would still apply.  */
+#ifdef __cplusplus
+extern "C"
+#endif
+char pthread_create ();
+int
+main ()
+{
+return pthread_create ();
+  ;
+  return 0;
+}
+_ACEOF
+if ac_fn_c_try_link "$LINENO"; then :
+  ac_cv_lib_pthreadGC2_pthread_create=yes
+else
+  ac_cv_lib_pthreadGC2_pthread_create=no
+fi
+rm -f core conftest.err conftest.$ac_objext \
+    conftest$ac_exeext conftest.$ac_ext
+LIBS=$ac_check_lib_save_LIBS
+fi
+{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $ac_cv_lib_pthreadGC2_pthread_create" >&5
+$as_echo "$ac_cv_lib_pthreadGC2_pthread_create" >&6; }
+if test "x$ac_cv_lib_pthreadGC2_pthread_create" = xyes; then :
+  PTHREAD_LIBS="-lpthreadGC2"
+else
+  { $as_echo "$as_me:${as_lineno-$LINENO}: checking for pthread_create in -lpthreadGC1" >&5
+$as_echo_n "checking for pthread_create in -lpthreadGC1... " >&6; }
+if ${ac_cv_lib_pthreadGC1_pthread_create+:} false; then :
+  $as_echo_n "(cached) " >&6
+else
+  ac_check_lib_save_LIBS=$LIBS
+LIBS="-lpthreadGC1  $LIBS"
+cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+/* end confdefs.h.  */
+
+/* Override any GCC internal prototype to avoid an error.
+   Use char because int might match the return type of a GCC
+   builtin and then its argument prototype would still apply.  */
+#ifdef __cplusplus
+extern "C"
+#endif
+char pthread_create ();
+int
+main ()
+{
+return pthread_create ();
+  ;
+  return 0;
+}
+_ACEOF
+if ac_fn_c_try_link "$LINENO"; then :
+  ac_cv_lib_pthreadGC1_pthread_create=yes
+else
+  ac_cv_lib_pthreadGC1_pthread_create=no
+fi
+rm -f core conftest.err conftest.$ac_objext \
+    conftest$ac_exeext conftest.$ac_ext
+LIBS=$ac_check_lib_save_LIBS
+fi
+{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $ac_cv_lib_pthreadGC1_pthread_create" >&5
+$as_echo "$ac_cv_lib_pthreadGC1_pthread_create" >&6; }
+if test "x$ac_cv_lib_pthreadGC1_pthread_create" = xyes; then :
+  PTHREAD_LIBS="-lpthreadGC1"
+else
+  { $as_echo "$as_me:${as_lineno-$LINENO}: checking for pthread_create in -lpthreadGC" >&5
+$as_echo_n "checking for pthread_create in -lpthreadGC... " >&6; }
+if ${ac_cv_lib_pthreadGC_pthread_create+:} false; then :
+  $as_echo_n "(cached) " >&6
+else
+  ac_check_lib_save_LIBS=$LIBS
+LIBS="-lpthreadGC  $LIBS"
+cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+/* end confdefs.h.  */
+
+/* Override any GCC internal prototype to avoid an error.
+   Use char because int might match the return type of a GCC
+   builtin and then its argument prototype would still apply.  */
+#ifdef __cplusplus
+extern "C"
+#endif
+char pthread_create ();
+int
+main ()
+{
+return pthread_create ();
+  ;
+  return 0;
+}
+_ACEOF
+if ac_fn_c_try_link "$LINENO"; then :
+  ac_cv_lib_pthreadGC_pthread_create=yes
+else
+  ac_cv_lib_pthreadGC_pthread_create=no
+fi
+rm -f core conftest.err conftest.$ac_objext \
+    conftest$ac_exeext conftest.$ac_ext
+LIBS=$ac_check_lib_save_LIBS
+fi
+{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $ac_cv_lib_pthreadGC_pthread_create" >&5
+$as_echo "$ac_cv_lib_pthreadGC_pthread_create" >&6; }
+if test "x$ac_cv_lib_pthreadGC_pthread_create" = xyes; then :
+  PTHREAD_LIBS="-lpthreadGC"
+
+fi
+
+fi
+
+fi
+
+fi
+
+
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for SSL_library_init in -lssl" >&5
+$as_echo_n "checking for SSL_library_init in -lssl... " >&6; }
+if ${ac_cv_lib_ssl_SSL_library_init+:} false; then :
+  $as_echo_n "(cached) " >&6
+else
+  ac_check_lib_save_LIBS=$LIBS
+LIBS="-lssl  $LIBS"
+cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+/* end confdefs.h.  */
+
+/* Override any GCC internal prototype to avoid an error.
+   Use char because int might match the return type of a GCC
+   builtin and then its argument prototype would still apply.  */
+#ifdef __cplusplus
+extern "C"
+#endif
+char SSL_library_init ();
+int
+main ()
+{
+return SSL_library_init ();
+  ;
+  return 0;
+}
+_ACEOF
+if ac_fn_c_try_link "$LINENO"; then :
+  ac_cv_lib_ssl_SSL_library_init=yes
+else
+  ac_cv_lib_ssl_SSL_library_init=no
+fi
+rm -f core conftest.err conftest.$ac_objext \
+    conftest$ac_exeext conftest.$ac_ext
+LIBS=$ac_check_lib_save_LIBS
+fi
+{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $ac_cv_lib_ssl_SSL_library_init" >&5
+$as_echo "$ac_cv_lib_ssl_SSL_library_init" >&6; }
+if test "x$ac_cv_lib_ssl_SSL_library_init" = xyes; then :
+  cat >>confdefs.h <<_ACEOF
+#define HAVE_LIBSSL 1
+_ACEOF
+
+  LIBS="-lssl $LIBS"
+
+else
+  as_fn_error $? "OpenSSL library required" "$LINENO" 5
+fi
+
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for EVP_DigestFinal_ex in -lcrypto" >&5
+$as_echo_n "checking for EVP_DigestFinal_ex in -lcrypto... " >&6; }
+if ${ac_cv_lib_crypto_EVP_DigestFinal_ex+:} false; then :
+  $as_echo_n "(cached) " >&6
+else
+  ac_check_lib_save_LIBS=$LIBS
+LIBS="-lcrypto  $LIBS"
+cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+/* end confdefs.h.  */
+
+/* Override any GCC internal prototype to avoid an error.
+   Use char because int might match the return type of a GCC
+   builtin and then its argument prototype would still apply.  */
+#ifdef __cplusplus
+extern "C"
+#endif
+char EVP_DigestFinal_ex ();
+int
+main ()
+{
+return EVP_DigestFinal_ex ();
+  ;
+  return 0;
+}
+_ACEOF
+if ac_fn_c_try_link "$LINENO"; then :
+  ac_cv_lib_crypto_EVP_DigestFinal_ex=yes
+else
+  ac_cv_lib_crypto_EVP_DigestFinal_ex=no
+fi
+rm -f core conftest.err conftest.$ac_objext \
+    conftest$ac_exeext conftest.$ac_ext
+LIBS=$ac_check_lib_save_LIBS
+fi
+{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $ac_cv_lib_crypto_EVP_DigestFinal_ex" >&5
+$as_echo "$ac_cv_lib_crypto_EVP_DigestFinal_ex" >&6; }
+if test "x$ac_cv_lib_crypto_EVP_DigestFinal_ex" = xyes; then :
+  cat >>confdefs.h <<_ACEOF
+#define HAVE_LIBCRYPTO 1
+_ACEOF
+
+  LIBS="-lcrypto $LIBS"
+
+else
+  as_fn_error $? "OpenSSL library required" "$LINENO" 5
+fi
+
+
+ if test x$request_jansson = xtrue; then
+  WANT_JANSSON_TRUE=
+  WANT_JANSSON_FALSE='#'
+else
+  WANT_JANSSON_TRUE='#'
+  WANT_JANSSON_FALSE=
+fi
+
+ if test x$have_win32 = xtrue; then
+  HAVE_WINDOWS_TRUE=
+  HAVE_WINDOWS_FALSE='#'
+else
+  HAVE_WINDOWS_TRUE='#'
+  HAVE_WINDOWS_FALSE=
+fi
+
+ if test x$have_x86 = xtrue; then
+  ARCH_x86_TRUE=
+  ARCH_x86_FALSE='#'
+else
+  ARCH_x86_TRUE='#'
+  ARCH_x86_FALSE=
+fi
+
+ if test x$have_x86_64 = xtrue; then
+  ARCH_x86_64_TRUE=
+  ARCH_x86_64_FALSE='#'
+else
+  ARCH_x86_64_TRUE='#'
+  ARCH_x86_64_FALSE=
+fi
+
+
+if test x$request_jansson = xtrue
+then
+	JANSSON_LIBS="compat/jansson/libjansson.a"
+else
+	JANSSON_LIBS=-ljansson
+fi
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+# Check whether --with-libcurl was given.
+if test "${with_libcurl+set}" = set; then :
+  withval=$with_libcurl; _libcurl_with=$withval
+else
+  _libcurl_with=yes
+fi
+
+
+  if test "$_libcurl_with" != "no" ; then
+
+     for ac_prog in gawk mawk nawk awk
+do
+  # Extract the first word of "$ac_prog", so it can be a program name with args.
+set dummy $ac_prog; ac_word=$2
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for $ac_word" >&5
+$as_echo_n "checking for $ac_word... " >&6; }
+if ${ac_cv_prog_AWK+:} false; then :
+  $as_echo_n "(cached) " >&6
+else
+  if test -n "$AWK"; then
+  ac_cv_prog_AWK="$AWK" # Let the user override the test.
+else
+as_save_IFS=$IFS; IFS=$PATH_SEPARATOR
+for as_dir in $PATH
+do
+  IFS=$as_save_IFS
+  test -z "$as_dir" && as_dir=.
+    for ac_exec_ext in '' $ac_executable_extensions; do
+  if as_fn_executable_p "$as_dir/$ac_word$ac_exec_ext"; then
+    ac_cv_prog_AWK="$ac_prog"
+    $as_echo "$as_me:${as_lineno-$LINENO}: found $as_dir/$ac_word$ac_exec_ext" >&5
+    break 2
+  fi
+done
+  done
+IFS=$as_save_IFS
+
+fi
+fi
+AWK=$ac_cv_prog_AWK
+if test -n "$AWK"; then
+  { $as_echo "$as_me:${as_lineno-$LINENO}: result: $AWK" >&5
+$as_echo "$AWK" >&6; }
+else
+  { $as_echo "$as_me:${as_lineno-$LINENO}: result: no" >&5
+$as_echo "no" >&6; }
+fi
+
+
+  test -n "$AWK" && break
+done
+
+
+     _libcurl_version_parse="eval $AWK '{split(\$NF,A,\".\"); X=256*256*A[1]+256*A[2]+A[3]; print X;}'"
+
+     _libcurl_try_link=yes
+
+     if test -d "$_libcurl_with" ; then
+        LIBCURL_CPPFLAGS="-I$withval/include"
+        _libcurl_ldflags="-L$withval/lib"
+        # Extract the first word of "curl-config", so it can be a program name with args.
+set dummy curl-config; ac_word=$2
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for $ac_word" >&5
+$as_echo_n "checking for $ac_word... " >&6; }
+if ${ac_cv_path__libcurl_config+:} false; then :
+  $as_echo_n "(cached) " >&6
+else
+  case $_libcurl_config in
+  [\\/]* | ?:[\\/]*)
+  ac_cv_path__libcurl_config="$_libcurl_config" # Let the user override the test with a path.
+  ;;
+  *)
+  as_save_IFS=$IFS; IFS=$PATH_SEPARATOR
+for as_dir in "$withval/bin"
+do
+  IFS=$as_save_IFS
+  test -z "$as_dir" && as_dir=.
+    for ac_exec_ext in '' $ac_executable_extensions; do
+  if as_fn_executable_p "$as_dir/$ac_word$ac_exec_ext"; then
+    ac_cv_path__libcurl_config="$as_dir/$ac_word$ac_exec_ext"
+    $as_echo "$as_me:${as_lineno-$LINENO}: found $as_dir/$ac_word$ac_exec_ext" >&5
+    break 2
+  fi
+done
+  done
+IFS=$as_save_IFS
+
+  ;;
+esac
+fi
+_libcurl_config=$ac_cv_path__libcurl_config
+if test -n "$_libcurl_config"; then
+  { $as_echo "$as_me:${as_lineno-$LINENO}: result: $_libcurl_config" >&5
+$as_echo "$_libcurl_config" >&6; }
+else
+  { $as_echo "$as_me:${as_lineno-$LINENO}: result: no" >&5
+$as_echo "no" >&6; }
+fi
+
+
+     else
+        # Extract the first word of "curl-config", so it can be a program name with args.
+set dummy curl-config; ac_word=$2
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for $ac_word" >&5
+$as_echo_n "checking for $ac_word... " >&6; }
+if ${ac_cv_path__libcurl_config+:} false; then :
+  $as_echo_n "(cached) " >&6
+else
+  case $_libcurl_config in
+  [\\/]* | ?:[\\/]*)
+  ac_cv_path__libcurl_config="$_libcurl_config" # Let the user override the test with a path.
+  ;;
+  *)
+  as_save_IFS=$IFS; IFS=$PATH_SEPARATOR
+for as_dir in $PATH
+do
+  IFS=$as_save_IFS
+  test -z "$as_dir" && as_dir=.
+    for ac_exec_ext in '' $ac_executable_extensions; do
+  if as_fn_executable_p "$as_dir/$ac_word$ac_exec_ext"; then
+    ac_cv_path__libcurl_config="$as_dir/$ac_word$ac_exec_ext"
+    $as_echo "$as_me:${as_lineno-$LINENO}: found $as_dir/$ac_word$ac_exec_ext" >&5
+    break 2
+  fi
+done
+  done
+IFS=$as_save_IFS
+
+  ;;
+esac
+fi
+_libcurl_config=$ac_cv_path__libcurl_config
+if test -n "$_libcurl_config"; then
+  { $as_echo "$as_me:${as_lineno-$LINENO}: result: $_libcurl_config" >&5
+$as_echo "$_libcurl_config" >&6; }
+else
+  { $as_echo "$as_me:${as_lineno-$LINENO}: result: no" >&5
+$as_echo "no" >&6; }
+fi
+
+
+     fi
+
+     if test x$_libcurl_config != "x" ; then
+        { $as_echo "$as_me:${as_lineno-$LINENO}: checking for the version of libcurl" >&5
+$as_echo_n "checking for the version of libcurl... " >&6; }
+if ${libcurl_cv_lib_curl_version+:} false; then :
+  $as_echo_n "(cached) " >&6
+else
+  libcurl_cv_lib_curl_version=`$_libcurl_config --version | $AWK '{print $2}'`
+fi
+{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $libcurl_cv_lib_curl_version" >&5
+$as_echo "$libcurl_cv_lib_curl_version" >&6; }
+
+        _libcurl_version=`echo $libcurl_cv_lib_curl_version | $_libcurl_version_parse`
+        _libcurl_wanted=`echo 7.15.2 | $_libcurl_version_parse`
+
+        if test $_libcurl_wanted -gt 0 ; then
+           { $as_echo "$as_me:${as_lineno-$LINENO}: checking for libcurl >= version 7.15.2" >&5
+$as_echo_n "checking for libcurl >= version 7.15.2... " >&6; }
+if ${libcurl_cv_lib_version_ok+:} false; then :
+  $as_echo_n "(cached) " >&6
+else
+
+              if test $_libcurl_version -ge $_libcurl_wanted ; then
+                 libcurl_cv_lib_version_ok=yes
+              else
+                 libcurl_cv_lib_version_ok=no
+              fi
+
+fi
+{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $libcurl_cv_lib_version_ok" >&5
+$as_echo "$libcurl_cv_lib_version_ok" >&6; }
+        fi
+
+        if test $_libcurl_wanted -eq 0 || test x$libcurl_cv_lib_version_ok = xyes ; then
+           if test x"$LIBCURL_CPPFLAGS" = "x" ; then
+              LIBCURL_CPPFLAGS=`$_libcurl_config --cflags`
+           fi
+           if test x"$LIBCURL" = "x" ; then
+              LIBCURL=`$_libcurl_config --libs`
+
+              # This is so silly, but Apple actually has a bug in their
+              # curl-config script.  Fixed in Tiger, but there are still
+              # lots of Panther installs around.
+              case "${host}" in
+                 powerpc-apple-darwin7*)
+                    LIBCURL=`echo $LIBCURL | sed -e 's|-arch i386||g'`
+                 ;;
+              esac
+           fi
+
+           # All curl-config scripts support --feature
+           _libcurl_features=`$_libcurl_config --feature`
+
+           # Is it modern enough to have --protocols? (7.12.4)
+           if test $_libcurl_version -ge 461828 ; then
+              _libcurl_protocols=`$_libcurl_config --protocols`
+           fi
+        else
+           _libcurl_try_link=no
+        fi
+
+        unset _libcurl_wanted
+     fi
+
+     if test $_libcurl_try_link = yes ; then
+
+        # we didn't find curl-config, so let's see if the user-supplied
+        # link line (or failing that, "-lcurl") is enough.
+        LIBCURL=${LIBCURL-"$_libcurl_ldflags -lcurl"}
+
+        { $as_echo "$as_me:${as_lineno-$LINENO}: checking whether libcurl is usable" >&5
+$as_echo_n "checking whether libcurl is usable... " >&6; }
+if ${libcurl_cv_lib_curl_usable+:} false; then :
+  $as_echo_n "(cached) " >&6
+else
+
+           _libcurl_save_cppflags=$CPPFLAGS
+           CPPFLAGS="$LIBCURL_CPPFLAGS $CPPFLAGS"
+           _libcurl_save_libs=$LIBS
+           LIBS="$LIBCURL $LIBS"
+
+           cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+/* end confdefs.h.  */
+#include <curl/curl.h>
+int
+main ()
+{
+
+/* Try and use a few common options to force a failure if we are
+   missing symbols or can't link. */
+int x;
+curl_easy_setopt(NULL,CURLOPT_URL,NULL);
+x=CURL_ERROR_SIZE;
+x=CURLOPT_WRITEFUNCTION;
+x=CURLOPT_FILE;
+x=CURLOPT_ERRORBUFFER;
+x=CURLOPT_STDERR;
+x=CURLOPT_VERBOSE;
+if (x) ;
+
+  ;
+  return 0;
+}
+_ACEOF
+if ac_fn_c_try_link "$LINENO"; then :
+  libcurl_cv_lib_curl_usable=yes
+else
+  libcurl_cv_lib_curl_usable=no
+fi
+rm -f core conftest.err conftest.$ac_objext \
+    conftest$ac_exeext conftest.$ac_ext
+
+           CPPFLAGS=$_libcurl_save_cppflags
+           LIBS=$_libcurl_save_libs
+           unset _libcurl_save_cppflags
+           unset _libcurl_save_libs
+
+fi
+{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $libcurl_cv_lib_curl_usable" >&5
+$as_echo "$libcurl_cv_lib_curl_usable" >&6; }
+
+        if test $libcurl_cv_lib_curl_usable = yes ; then
+
+           # Does curl_free() exist in this version of libcurl?
+           # If not, fake it with free()
+
+           _libcurl_save_cppflags=$CPPFLAGS
+           CPPFLAGS="$CPPFLAGS $LIBCURL_CPPFLAGS"
+           _libcurl_save_libs=$LIBS
+           LIBS="$LIBS $LIBCURL"
+
+           ac_fn_c_check_func "$LINENO" "curl_free" "ac_cv_func_curl_free"
+if test "x$ac_cv_func_curl_free" = xyes; then :
+
+else
+
+$as_echo "#define curl_free free" >>confdefs.h
+
+fi
+
+
+           CPPFLAGS=$_libcurl_save_cppflags
+           LIBS=$_libcurl_save_libs
+           unset _libcurl_save_cppflags
+           unset _libcurl_save_libs
+
+
+$as_echo "#define HAVE_LIBCURL 1" >>confdefs.h
+
+
+
+
+           for _libcurl_feature in $_libcurl_features ; do
+              cat >>confdefs.h <<_ACEOF
+#define `$as_echo "libcurl_feature_$_libcurl_feature" | $as_tr_cpp` 1
+_ACEOF
+
+              eval `$as_echo "libcurl_feature_$_libcurl_feature" | $as_tr_sh`=yes
+           done
+
+           if test "x$_libcurl_protocols" = "x" ; then
+
+              # We don't have --protocols, so just assume that all
+              # protocols are available
+              _libcurl_protocols="HTTP FTP FILE TELNET LDAP DICT TFTP"
+
+              if test x$libcurl_feature_SSL = xyes ; then
+                 _libcurl_protocols="$_libcurl_protocols HTTPS"
+
+                 # FTPS wasn't standards-compliant until version
+                 # 7.11.0 (0x070b00 == 461568)
+                 if test $_libcurl_version -ge 461568; then
+                    _libcurl_protocols="$_libcurl_protocols FTPS"
+                 fi
+              fi
+
+              # RTSP, IMAP, POP3 and SMTP were added in
+              # 7.20.0 (0x071400 == 463872)
+              if test $_libcurl_version -ge 463872; then
+                 _libcurl_protocols="$_libcurl_protocols RTSP IMAP POP3 SMTP"
+              fi
+           fi
+
+           for _libcurl_protocol in $_libcurl_protocols ; do
+              cat >>confdefs.h <<_ACEOF
+#define `$as_echo "libcurl_protocol_$_libcurl_protocol" | $as_tr_cpp` 1
+_ACEOF
+
+              eval `$as_echo "libcurl_protocol_$_libcurl_protocol" | $as_tr_sh`=yes
+           done
+        else
+           unset LIBCURL
+           unset LIBCURL_CPPFLAGS
+        fi
+     fi
+
+     unset _libcurl_try_link
+     unset _libcurl_version_parse
+     unset _libcurl_config
+     unset _libcurl_feature
+     unset _libcurl_features
+     unset _libcurl_protocol
+     unset _libcurl_protocols
+     unset _libcurl_version
+     unset _libcurl_ldflags
+  fi
+
+  if test x$_libcurl_with = xno || test x$libcurl_cv_lib_curl_usable != xyes ; then
+     # This is the IF-NO path
+     as_fn_error $? "Missing required libcurl >= 7.15.2" "$LINENO" 5
+  else
+     # This is the IF-YES path
+     :
+  fi
+
+  unset _libcurl_with
+
+
+
+
+
+
+
+ac_config_files="$ac_config_files Makefile compat/Makefile compat/jansson/Makefile"
+
+
+ARCH=`uname -m`
+if [ $ARCH == "x86_64" ];
+then
+  SUFFIX="64"
+else
+  SUFFIX=""
+fi
+
+
+# Check whether --with-cuda was given.
+if test "${with_cuda+set}" = set; then :
+  withval=$with_cuda;
+fi
+
+
+if test -n "$with_cuda"
+then
+   CUDA_CFLAGS="-I$with_cuda/include"
+   CUDA_LIBS="-lcudart"
+   CUDA_LDFLAGS="-L$with_cuda/lib$SUFFIX"
+   NVCC="$with_cuda/bin/nvcc"
+else
+   CUDA_CFLAGS="-I/usr/local/cuda/include"
+   CUDA_LIBS="-lcudart -static-libstdc++"
+   CUDA_LDFLAGS="-L/usr/local/cuda/lib$SUFFIX"
+   NVCC="nvcc"
+fi
+
+
+
+
+
+
+
+cat >confcache <<\_ACEOF
+# This file is a shell script that caches the results of configure
+# tests run on this system so they can be shared between configure
+# scripts and configure runs, see configure's option --config-cache.
+# It is not useful on other systems.  If it contains results you don't
+# want to keep, you may remove or edit it.
+#
+# config.status only pays attention to the cache file if you give it
+# the --recheck option to rerun configure.
+#
+# `ac_cv_env_foo' variables (set or unset) will be overridden when
+# loading this file, other *unset* `ac_cv_foo' will be assigned the
+# following values.
+
+_ACEOF
+
+# The following way of writing the cache mishandles newlines in values,
+# but we know of no workaround that is simple, portable, and efficient.
+# So, we kill variables containing newlines.
+# Ultrix sh set writes to stderr and can't be redirected directly,
+# and sets the high bit in the cache file unless we assign to the vars.
+(
+  for ac_var in `(set) 2>&1 | sed -n 's/^\([a-zA-Z_][a-zA-Z0-9_]*\)=.*/\1/p'`; do
+    eval ac_val=\$$ac_var
+    case $ac_val in #(
+    *${as_nl}*)
+      case $ac_var in #(
+      *_cv_*) { $as_echo "$as_me:${as_lineno-$LINENO}: WARNING: cache variable $ac_var contains a newline" >&5
+$as_echo "$as_me: WARNING: cache variable $ac_var contains a newline" >&2;} ;;
+      esac
+      case $ac_var in #(
+      _ | IFS | as_nl) ;; #(
+      BASH_ARGV | BASH_SOURCE) eval $ac_var= ;; #(
+      *) { eval $ac_var=; unset $ac_var;} ;;
+      esac ;;
+    esac
+  done
+
+  (set) 2>&1 |
+    case $as_nl`(ac_space=' '; set) 2>&1` in #(
+    *${as_nl}ac_space=\ *)
+      # `set' does not quote correctly, so add quotes: double-quote
+      # substitution turns \\\\ into \\, and sed turns \\ into \.
+      sed -n \
+	"s/'/'\\\\''/g;
+	  s/^\\([_$as_cr_alnum]*_cv_[_$as_cr_alnum]*\\)=\\(.*\\)/\\1='\\2'/p"
+      ;; #(
+    *)
+      # `set' quotes correctly as required by POSIX, so do not add quotes.
+      sed -n "/^[_$as_cr_alnum]*_cv_[_$as_cr_alnum]*=/p"
+      ;;
+    esac |
+    sort
+) |
+  sed '
+     /^ac_cv_env_/b end
+     t clear
+     :clear
+     s/^\([^=]*\)=\(.*[{}].*\)$/test "${\1+set}" = set || &/
+     t end
+     s/^\([^=]*\)=\(.*\)$/\1=${\1=\2}/
+     :end' >>confcache
+if diff "$cache_file" confcache >/dev/null 2>&1; then :; else
+  if test -w "$cache_file"; then
+    if test "x$cache_file" != "x/dev/null"; then
+      { $as_echo "$as_me:${as_lineno-$LINENO}: updating cache $cache_file" >&5
+$as_echo "$as_me: updating cache $cache_file" >&6;}
+      if test ! -f "$cache_file" || test -h "$cache_file"; then
+	cat confcache >"$cache_file"
+      else
+        case $cache_file in #(
+        */* | ?:*)
+	  mv -f confcache "$cache_file"$$ &&
+	  mv -f "$cache_file"$$ "$cache_file" ;; #(
+        *)
+	  mv -f confcache "$cache_file" ;;
+	esac
+      fi
+    fi
+  else
+    { $as_echo "$as_me:${as_lineno-$LINENO}: not updating unwritable cache $cache_file" >&5
+$as_echo "$as_me: not updating unwritable cache $cache_file" >&6;}
+  fi
+fi
+rm -f confcache
+
+test "x$prefix" = xNONE && prefix=$ac_default_prefix
+# Let make expand exec_prefix.
+test "x$exec_prefix" = xNONE && exec_prefix='${prefix}'
+
+DEFS=-DHAVE_CONFIG_H
+
+ac_libobjs=
+ac_ltlibobjs=
+U=
+for ac_i in : $LIBOBJS; do test "x$ac_i" = x: && continue
+  # 1. Remove the extension, and $U if already installed.
+  ac_script='s/\$U\././;s/\.o$//;s/\.obj$//'
+  ac_i=`$as_echo "$ac_i" | sed "$ac_script"`
+  # 2. Prepend LIBOBJDIR.  When used with automake>=1.10 LIBOBJDIR
+  #    will be set to the directory where LIBOBJS objects are built.
+  as_fn_append ac_libobjs " \${LIBOBJDIR}$ac_i\$U.$ac_objext"
+  as_fn_append ac_ltlibobjs " \${LIBOBJDIR}$ac_i"'$U.lo'
+done
+LIBOBJS=$ac_libobjs
+
+LTLIBOBJS=$ac_ltlibobjs
+
+
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking that generated files are newer than configure" >&5
+$as_echo_n "checking that generated files are newer than configure... " >&6; }
+   if test -n "$am_sleep_pid"; then
+     # Hide warnings about reused PIDs.
+     wait $am_sleep_pid 2>/dev/null
+   fi
+   { $as_echo "$as_me:${as_lineno-$LINENO}: result: done" >&5
+$as_echo "done" >&6; }
+ if test -n "$EXEEXT"; then
+  am__EXEEXT_TRUE=
+  am__EXEEXT_FALSE='#'
+else
+  am__EXEEXT_TRUE='#'
+  am__EXEEXT_FALSE=
+fi
+
+if test -z "${MAINTAINER_MODE_TRUE}" && test -z "${MAINTAINER_MODE_FALSE}"; then
+  as_fn_error $? "conditional \"MAINTAINER_MODE\" was never defined.
+Usually this means the macro was only invoked conditionally." "$LINENO" 5
+fi
+if test -z "${AMDEP_TRUE}" && test -z "${AMDEP_FALSE}"; then
+  as_fn_error $? "conditional \"AMDEP\" was never defined.
+Usually this means the macro was only invoked conditionally." "$LINENO" 5
+fi
+if test -z "${am__fastdepCC_TRUE}" && test -z "${am__fastdepCC_FALSE}"; then
+  as_fn_error $? "conditional \"am__fastdepCC\" was never defined.
+Usually this means the macro was only invoked conditionally." "$LINENO" 5
+fi
+if test -z "${am__fastdepCCAS_TRUE}" && test -z "${am__fastdepCCAS_FALSE}"; then
+  as_fn_error $? "conditional \"am__fastdepCCAS\" was never defined.
+Usually this means the macro was only invoked conditionally." "$LINENO" 5
+fi
+if test -z "${am__fastdepCXX_TRUE}" && test -z "${am__fastdepCXX_FALSE}"; then
+  as_fn_error $? "conditional \"am__fastdepCXX\" was never defined.
+Usually this means the macro was only invoked conditionally." "$LINENO" 5
+fi
+if test -z "${WANT_JANSSON_TRUE}" && test -z "${WANT_JANSSON_FALSE}"; then
+  as_fn_error $? "conditional \"WANT_JANSSON\" was never defined.
+Usually this means the macro was only invoked conditionally." "$LINENO" 5
+fi
+if test -z "${HAVE_WINDOWS_TRUE}" && test -z "${HAVE_WINDOWS_FALSE}"; then
+  as_fn_error $? "conditional \"HAVE_WINDOWS\" was never defined.
+Usually this means the macro was only invoked conditionally." "$LINENO" 5
+fi
+if test -z "${ARCH_x86_TRUE}" && test -z "${ARCH_x86_FALSE}"; then
+  as_fn_error $? "conditional \"ARCH_x86\" was never defined.
+Usually this means the macro was only invoked conditionally." "$LINENO" 5
+fi
+if test -z "${ARCH_x86_64_TRUE}" && test -z "${ARCH_x86_64_FALSE}"; then
+  as_fn_error $? "conditional \"ARCH_x86_64\" was never defined.
+Usually this means the macro was only invoked conditionally." "$LINENO" 5
+fi
+
+: "${CONFIG_STATUS=./config.status}"
+ac_write_fail=0
+ac_clean_files_save=$ac_clean_files
+ac_clean_files="$ac_clean_files $CONFIG_STATUS"
+{ $as_echo "$as_me:${as_lineno-$LINENO}: creating $CONFIG_STATUS" >&5
+$as_echo "$as_me: creating $CONFIG_STATUS" >&6;}
+as_write_fail=0
+cat >$CONFIG_STATUS <<_ASEOF || as_write_fail=1
+#! $SHELL
+# Generated by $as_me.
+# Run this file to recreate the current configuration.
+# Compiler output produced by configure, useful for debugging
+# configure, is in config.log if it exists.
+
+debug=false
+ac_cs_recheck=false
+ac_cs_silent=false
+
+SHELL=\${CONFIG_SHELL-$SHELL}
+export SHELL
+_ASEOF
+cat >>$CONFIG_STATUS <<\_ASEOF || as_write_fail=1
+## -------------------- ##
+## M4sh Initialization. ##
+## -------------------- ##
+
+# Be more Bourne compatible
+DUALCASE=1; export DUALCASE # for MKS sh
+if test -n "${ZSH_VERSION+set}" && (emulate sh) >/dev/null 2>&1; then :
+  emulate sh
+  NULLCMD=:
+  # Pre-4.2 versions of Zsh do word splitting on ${1+"$@"}, which
+  # is contrary to our usage.  Disable this feature.
+  alias -g '${1+"$@"}'='"$@"'
+  setopt NO_GLOB_SUBST
+else
+  case `(set -o) 2>/dev/null` in #(
+  *posix*) :
+    set -o posix ;; #(
+  *) :
+     ;;
+esac
+fi
+
+
+as_nl='
+'
+export as_nl
+# Printing a long string crashes Solaris 7 /usr/bin/printf.
+as_echo='\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\'
+as_echo=$as_echo$as_echo$as_echo$as_echo$as_echo
+as_echo=$as_echo$as_echo$as_echo$as_echo$as_echo$as_echo
+# Prefer a ksh shell builtin over an external printf program on Solaris,
+# but without wasting forks for bash or zsh.
+if test -z "$BASH_VERSION$ZSH_VERSION" \
+    && (test "X`print -r -- $as_echo`" = "X$as_echo") 2>/dev/null; then
+  as_echo='print -r --'
+  as_echo_n='print -rn --'
+elif (test "X`printf %s $as_echo`" = "X$as_echo") 2>/dev/null; then
+  as_echo='printf %s\n'
+  as_echo_n='printf %s'
+else
+  if test "X`(/usr/ucb/echo -n -n $as_echo) 2>/dev/null`" = "X-n $as_echo"; then
+    as_echo_body='eval /usr/ucb/echo -n "$1$as_nl"'
+    as_echo_n='/usr/ucb/echo -n'
+  else
+    as_echo_body='eval expr "X$1" : "X\\(.*\\)"'
+    as_echo_n_body='eval
+      arg=$1;
+      case $arg in #(
+      *"$as_nl"*)
+	expr "X$arg" : "X\\(.*\\)$as_nl";
+	arg=`expr "X$arg" : ".*$as_nl\\(.*\\)"`;;
+      esac;
+      expr "X$arg" : "X\\(.*\\)" | tr -d "$as_nl"
+    '
+    export as_echo_n_body
+    as_echo_n='sh -c $as_echo_n_body as_echo'
+  fi
+  export as_echo_body
+  as_echo='sh -c $as_echo_body as_echo'
+fi
+
+# The user is always right.
+if test "${PATH_SEPARATOR+set}" != set; then
+  PATH_SEPARATOR=:
+  (PATH='/bin;/bin'; FPATH=$PATH; sh -c :) >/dev/null 2>&1 && {
+    (PATH='/bin:/bin'; FPATH=$PATH; sh -c :) >/dev/null 2>&1 ||
+      PATH_SEPARATOR=';'
+  }
+fi
+
+
+# IFS
+# We need space, tab and new line, in precisely that order.  Quoting is
+# there to prevent editors from complaining about space-tab.
+# (If _AS_PATH_WALK were called with IFS unset, it would disable word
+# splitting by setting IFS to empty value.)
+IFS=" ""	$as_nl"
+
+# Find who we are.  Look in the path if we contain no directory separator.
+as_myself=
+case $0 in #((
+  *[\\/]* ) as_myself=$0 ;;
+  *) as_save_IFS=$IFS; IFS=$PATH_SEPARATOR
+for as_dir in $PATH
+do
+  IFS=$as_save_IFS
+  test -z "$as_dir" && as_dir=.
+    test -r "$as_dir/$0" && as_myself=$as_dir/$0 && break
+  done
+IFS=$as_save_IFS
+
+     ;;
+esac
+# We did not find ourselves, most probably we were run as `sh COMMAND'
+# in which case we are not to be found in the path.
+if test "x$as_myself" = x; then
+  as_myself=$0
+fi
+if test ! -f "$as_myself"; then
+  $as_echo "$as_myself: error: cannot find myself; rerun with an absolute file name" >&2
+  exit 1
+fi
+
+# Unset variables that we do not need and which cause bugs (e.g. in
+# pre-3.0 UWIN ksh).  But do not cause bugs in bash 2.01; the "|| exit 1"
+# suppresses any "Segmentation fault" message there.  '((' could
+# trigger a bug in pdksh 5.2.14.
+for as_var in BASH_ENV ENV MAIL MAILPATH
+do eval test x\${$as_var+set} = xset \
+  && ( (unset $as_var) || exit 1) >/dev/null 2>&1 && unset $as_var || :
+done
+PS1='$ '
+PS2='> '
+PS4='+ '
+
+# NLS nuisances.
+LC_ALL=C
+export LC_ALL
+LANGUAGE=C
+export LANGUAGE
+
+# CDPATH.
+(unset CDPATH) >/dev/null 2>&1 && unset CDPATH
+
+
+# as_fn_error STATUS ERROR [LINENO LOG_FD]
+# ----------------------------------------
+# Output "`basename $0`: error: ERROR" to stderr. If LINENO and LOG_FD are
+# provided, also output the error to LOG_FD, referencing LINENO. Then exit the
+# script with STATUS, using 1 if that was 0.
+as_fn_error ()
+{
+  as_status=$1; test $as_status -eq 0 && as_status=1
+  if test "$4"; then
+    as_lineno=${as_lineno-"$3"} as_lineno_stack=as_lineno_stack=$as_lineno_stack
+    $as_echo "$as_me:${as_lineno-$LINENO}: error: $2" >&$4
+  fi
+  $as_echo "$as_me: error: $2" >&2
+  as_fn_exit $as_status
+} # as_fn_error
+
+
+# as_fn_set_status STATUS
+# -----------------------
+# Set $? to STATUS, without forking.
+as_fn_set_status ()
+{
+  return $1
+} # as_fn_set_status
+
+# as_fn_exit STATUS
+# -----------------
+# Exit the shell with STATUS, even in a "trap 0" or "set -e" context.
+as_fn_exit ()
+{
+  set +e
+  as_fn_set_status $1
+  exit $1
+} # as_fn_exit
+
+# as_fn_unset VAR
+# ---------------
+# Portably unset VAR.
+as_fn_unset ()
+{
+  { eval $1=; unset $1;}
+}
+as_unset=as_fn_unset
+# as_fn_append VAR VALUE
+# ----------------------
+# Append the text in VALUE to the end of the definition contained in VAR. Take
+# advantage of any shell optimizations that allow amortized linear growth over
+# repeated appends, instead of the typical quadratic growth present in naive
+# implementations.
+if (eval "as_var=1; as_var+=2; test x\$as_var = x12") 2>/dev/null; then :
+  eval 'as_fn_append ()
+  {
+    eval $1+=\$2
+  }'
+else
+  as_fn_append ()
+  {
+    eval $1=\$$1\$2
+  }
+fi # as_fn_append
+
+# as_fn_arith ARG...
+# ------------------
+# Perform arithmetic evaluation on the ARGs, and store the result in the
+# global $as_val. Take advantage of shells that can avoid forks. The arguments
+# must be portable across $(()) and expr.
+if (eval "test \$(( 1 + 1 )) = 2") 2>/dev/null; then :
+  eval 'as_fn_arith ()
+  {
+    as_val=$(( $* ))
+  }'
+else
+  as_fn_arith ()
+  {
+    as_val=`expr "$@" || test $? -eq 1`
+  }
+fi # as_fn_arith
+
+
+if expr a : '\(a\)' >/dev/null 2>&1 &&
+   test "X`expr 00001 : '.*\(...\)'`" = X001; then
+  as_expr=expr
+else
+  as_expr=false
+fi
+
+if (basename -- /) >/dev/null 2>&1 && test "X`basename -- / 2>&1`" = "X/"; then
+  as_basename=basename
+else
+  as_basename=false
+fi
+
+if (as_dir=`dirname -- /` && test "X$as_dir" = X/) >/dev/null 2>&1; then
+  as_dirname=dirname
+else
+  as_dirname=false
+fi
+
+as_me=`$as_basename -- "$0" ||
+$as_expr X/"$0" : '.*/\([^/][^/]*\)/*$' \| \
+	 X"$0" : 'X\(//\)$' \| \
+	 X"$0" : 'X\(/\)' \| . 2>/dev/null ||
+$as_echo X/"$0" |
+    sed '/^.*\/\([^/][^/]*\)\/*$/{
+	    s//\1/
+	    q
+	  }
+	  /^X\/\(\/\/\)$/{
+	    s//\1/
+	    q
+	  }
+	  /^X\/\(\/\).*/{
+	    s//\1/
+	    q
+	  }
+	  s/.*/./; q'`
+
+# Avoid depending upon Character Ranges.
+as_cr_letters='abcdefghijklmnopqrstuvwxyz'
+as_cr_LETTERS='ABCDEFGHIJKLMNOPQRSTUVWXYZ'
+as_cr_Letters=$as_cr_letters$as_cr_LETTERS
+as_cr_digits='0123456789'
+as_cr_alnum=$as_cr_Letters$as_cr_digits
+
+ECHO_C= ECHO_N= ECHO_T=
+case `echo -n x` in #(((((
+-n*)
+  case `echo 'xy\c'` in
+  *c*) ECHO_T='	';;	# ECHO_T is single tab character.
+  xy)  ECHO_C='\c';;
+  *)   echo `echo ksh88 bug on AIX 6.1` > /dev/null
+       ECHO_T='	';;
+  esac;;
+*)
+  ECHO_N='-n';;
+esac
+
+rm -f conf$$ conf$$.exe conf$$.file
+if test -d conf$$.dir; then
+  rm -f conf$$.dir/conf$$.file
+else
+  rm -f conf$$.dir
+  mkdir conf$$.dir 2>/dev/null
+fi
+if (echo >conf$$.file) 2>/dev/null; then
+  if ln -s conf$$.file conf$$ 2>/dev/null; then
+    as_ln_s='ln -s'
+    # ... but there are two gotchas:
+    # 1) On MSYS, both `ln -s file dir' and `ln file dir' fail.
+    # 2) DJGPP < 2.04 has no symlinks; `ln -s' creates a wrapper executable.
+    # In both cases, we have to default to `cp -pR'.
+    ln -s conf$$.file conf$$.dir 2>/dev/null && test ! -f conf$$.exe ||
+      as_ln_s='cp -pR'
+  elif ln conf$$.file conf$$ 2>/dev/null; then
+    as_ln_s=ln
+  else
+    as_ln_s='cp -pR'
+  fi
+else
+  as_ln_s='cp -pR'
+fi
+rm -f conf$$ conf$$.exe conf$$.dir/conf$$.file conf$$.file
+rmdir conf$$.dir 2>/dev/null
+
+
+# as_fn_mkdir_p
+# -------------
+# Create "$as_dir" as a directory, including parents if necessary.
+as_fn_mkdir_p ()
+{
+
+  case $as_dir in #(
+  -*) as_dir=./$as_dir;;
+  esac
+  test -d "$as_dir" || eval $as_mkdir_p || {
+    as_dirs=
+    while :; do
+      case $as_dir in #(
+      *\'*) as_qdir=`$as_echo "$as_dir" | sed "s/'/'\\\\\\\\''/g"`;; #'(
+      *) as_qdir=$as_dir;;
+      esac
+      as_dirs="'$as_qdir' $as_dirs"
+      as_dir=`$as_dirname -- "$as_dir" ||
+$as_expr X"$as_dir" : 'X\(.*[^/]\)//*[^/][^/]*/*$' \| \
+	 X"$as_dir" : 'X\(//\)[^/]' \| \
+	 X"$as_dir" : 'X\(//\)$' \| \
+	 X"$as_dir" : 'X\(/\)' \| . 2>/dev/null ||
+$as_echo X"$as_dir" |
+    sed '/^X\(.*[^/]\)\/\/*[^/][^/]*\/*$/{
+	    s//\1/
+	    q
+	  }
+	  /^X\(\/\/\)[^/].*/{
+	    s//\1/
+	    q
+	  }
+	  /^X\(\/\/\)$/{
+	    s//\1/
+	    q
+	  }
+	  /^X\(\/\).*/{
+	    s//\1/
+	    q
+	  }
+	  s/.*/./; q'`
+      test -d "$as_dir" && break
+    done
+    test -z "$as_dirs" || eval "mkdir $as_dirs"
+  } || test -d "$as_dir" || as_fn_error $? "cannot create directory $as_dir"
+
+
+} # as_fn_mkdir_p
+if mkdir -p . 2>/dev/null; then
+  as_mkdir_p='mkdir -p "$as_dir"'
+else
+  test -d ./-p && rmdir ./-p
+  as_mkdir_p=false
+fi
+
+
+# as_fn_executable_p FILE
+# -----------------------
+# Test if FILE is an executable regular file.
+as_fn_executable_p ()
+{
+  test -f "$1" && test -x "$1"
+} # as_fn_executable_p
+as_test_x='test -x'
+as_executable_p=as_fn_executable_p
+
+# Sed expression to map a string onto a valid CPP name.
+as_tr_cpp="eval sed 'y%*$as_cr_letters%P$as_cr_LETTERS%;s%[^_$as_cr_alnum]%_%g'"
+
+# Sed expression to map a string onto a valid variable name.
+as_tr_sh="eval sed 'y%*+%pp%;s%[^_$as_cr_alnum]%_%g'"
+
+
+exec 6>&1
+## ----------------------------------- ##
+## Main body of $CONFIG_STATUS script. ##
+## ----------------------------------- ##
+_ASEOF
+test $as_write_fail = 0 && chmod +x $CONFIG_STATUS || ac_write_fail=1
+
+cat >>$CONFIG_STATUS <<\_ACEOF || ac_write_fail=1
+# Save the log message, to keep $0 and so on meaningful, and to
+# report actual input values of CONFIG_FILES etc. instead of their
+# values after options handling.
+ac_log="
+This file was extended by ccminer $as_me 2014.03.18, which was
+generated by GNU Autoconf 2.69.  Invocation command line was
+
+  CONFIG_FILES    = $CONFIG_FILES
+  CONFIG_HEADERS  = $CONFIG_HEADERS
+  CONFIG_LINKS    = $CONFIG_LINKS
+  CONFIG_COMMANDS = $CONFIG_COMMANDS
+  $ $0 $@
+
+on `(hostname || uname -n) 2>/dev/null | sed 1q`
+"
+
+_ACEOF
+
+case $ac_config_files in *"
+"*) set x $ac_config_files; shift; ac_config_files=$*;;
+esac
+
+case $ac_config_headers in *"
+"*) set x $ac_config_headers; shift; ac_config_headers=$*;;
+esac
+
+
+cat >>$CONFIG_STATUS <<_ACEOF || ac_write_fail=1
+# Files that config.status was made for.
+config_files="$ac_config_files"
+config_headers="$ac_config_headers"
+config_commands="$ac_config_commands"
+
+_ACEOF
+
+cat >>$CONFIG_STATUS <<\_ACEOF || ac_write_fail=1
+ac_cs_usage="\
+\`$as_me' instantiates files and other configuration actions
+from templates according to the current configuration.  Unless the files
+and actions are specified as TAGs, all are instantiated by default.
+
+Usage: $0 [OPTION]... [TAG]...
+
+  -h, --help       print this help, then exit
+  -V, --version    print version number and configuration settings, then exit
+      --config     print configuration, then exit
+  -q, --quiet, --silent
+                   do not print progress messages
+  -d, --debug      don't remove temporary files
+      --recheck    update $as_me by reconfiguring in the same conditions
+      --file=FILE[:TEMPLATE]
+                   instantiate the configuration file FILE
+      --header=FILE[:TEMPLATE]
+                   instantiate the configuration header FILE
+
+Configuration files:
+$config_files
+
+Configuration headers:
+$config_headers
+
+Configuration commands:
+$config_commands
+
+Report bugs to the package provider."
+
+_ACEOF
+cat >>$CONFIG_STATUS <<_ACEOF || ac_write_fail=1
+ac_cs_config="`$as_echo "$ac_configure_args" | sed 's/^ //; s/[\\""\`\$]/\\\\&/g'`"
+ac_cs_version="\\
+ccminer config.status 2014.03.18
+configured by $0, generated by GNU Autoconf 2.69,
+  with options \\"\$ac_cs_config\\"
+
+Copyright (C) 2012 Free Software Foundation, Inc.
+This config.status script is free software; the Free Software Foundation
+gives unlimited permission to copy, distribute and modify it."
+
+ac_pwd='$ac_pwd'
+srcdir='$srcdir'
+INSTALL='$INSTALL'
+MKDIR_P='$MKDIR_P'
+AWK='$AWK'
+test -n "\$AWK" || AWK=awk
+_ACEOF
+
+cat >>$CONFIG_STATUS <<\_ACEOF || ac_write_fail=1
+# The default lists apply if the user does not specify any file.
+ac_need_defaults=:
+while test $# != 0
+do
+  case $1 in
+  --*=?*)
+    ac_option=`expr "X$1" : 'X\([^=]*\)='`
+    ac_optarg=`expr "X$1" : 'X[^=]*=\(.*\)'`
+    ac_shift=:
+    ;;
+  --*=)
+    ac_option=`expr "X$1" : 'X\([^=]*\)='`
+    ac_optarg=
+    ac_shift=:
+    ;;
+  *)
+    ac_option=$1
+    ac_optarg=$2
+    ac_shift=shift
+    ;;
+  esac
+
+  case $ac_option in
+  # Handling of the options.
+  -recheck | --recheck | --rechec | --reche | --rech | --rec | --re | --r)
+    ac_cs_recheck=: ;;
+  --version | --versio | --versi | --vers | --ver | --ve | --v | -V )
+    $as_echo "$ac_cs_version"; exit ;;
+  --config | --confi | --conf | --con | --co | --c )
+    $as_echo "$ac_cs_config"; exit ;;
+  --debug | --debu | --deb | --de | --d | -d )
+    debug=: ;;
+  --file | --fil | --fi | --f )
+    $ac_shift
+    case $ac_optarg in
+    *\'*) ac_optarg=`$as_echo "$ac_optarg" | sed "s/'/'\\\\\\\\''/g"` ;;
+    '') as_fn_error $? "missing file argument" ;;
+    esac
+    as_fn_append CONFIG_FILES " '$ac_optarg'"
+    ac_need_defaults=false;;
+  --header | --heade | --head | --hea )
+    $ac_shift
+    case $ac_optarg in
+    *\'*) ac_optarg=`$as_echo "$ac_optarg" | sed "s/'/'\\\\\\\\''/g"` ;;
+    esac
+    as_fn_append CONFIG_HEADERS " '$ac_optarg'"
+    ac_need_defaults=false;;
+  --he | --h)
+    # Conflict between --help and --header
+    as_fn_error $? "ambiguous option: \`$1'
+Try \`$0 --help' for more information.";;
+  --help | --hel | -h )
+    $as_echo "$ac_cs_usage"; exit ;;
+  -q | -quiet | --quiet | --quie | --qui | --qu | --q \
+  | -silent | --silent | --silen | --sile | --sil | --si | --s)
+    ac_cs_silent=: ;;
+
+  # This is an error.
+  -*) as_fn_error $? "unrecognized option: \`$1'
+Try \`$0 --help' for more information." ;;
+
+  *) as_fn_append ac_config_targets " $1"
+     ac_need_defaults=false ;;
+
+  esac
+  shift
+done
+
+ac_configure_extra_args=
+
+if $ac_cs_silent; then
+  exec 6>/dev/null
+  ac_configure_extra_args="$ac_configure_extra_args --silent"
+fi
+
+_ACEOF
+cat >>$CONFIG_STATUS <<_ACEOF || ac_write_fail=1
+if \$ac_cs_recheck; then
+  set X $SHELL '$0' $ac_configure_args \$ac_configure_extra_args --no-create --no-recursion
+  shift
+  \$as_echo "running CONFIG_SHELL=$SHELL \$*" >&6
+  CONFIG_SHELL='$SHELL'
+  export CONFIG_SHELL
+  exec "\$@"
+fi
+
+_ACEOF
+cat >>$CONFIG_STATUS <<\_ACEOF || ac_write_fail=1
+exec 5>>config.log
+{
+  echo
+  sed 'h;s/./-/g;s/^.../## /;s/...$/ ##/;p;x;p;x' <<_ASBOX
+## Running $as_me. ##
+_ASBOX
+  $as_echo "$ac_log"
+} >&5
+
+_ACEOF
+cat >>$CONFIG_STATUS <<_ACEOF || ac_write_fail=1
+#
+# INIT-COMMANDS
+#
+AMDEP_TRUE="$AMDEP_TRUE" ac_aux_dir="$ac_aux_dir"
+
+_ACEOF
+
+cat >>$CONFIG_STATUS <<\_ACEOF || ac_write_fail=1
+
+# Handling of arguments.
+for ac_config_target in $ac_config_targets
+do
+  case $ac_config_target in
+    "cpuminer-config.h") CONFIG_HEADERS="$CONFIG_HEADERS cpuminer-config.h" ;;
+    "depfiles") CONFIG_COMMANDS="$CONFIG_COMMANDS depfiles" ;;
+    "Makefile") CONFIG_FILES="$CONFIG_FILES Makefile" ;;
+    "compat/Makefile") CONFIG_FILES="$CONFIG_FILES compat/Makefile" ;;
+    "compat/jansson/Makefile") CONFIG_FILES="$CONFIG_FILES compat/jansson/Makefile" ;;
+
+  *) as_fn_error $? "invalid argument: \`$ac_config_target'" "$LINENO" 5;;
+  esac
+done
+
+
+# If the user did not use the arguments to specify the items to instantiate,
+# then the envvar interface is used.  Set only those that are not.
+# We use the long form for the default assignment because of an extremely
+# bizarre bug on SunOS 4.1.3.
+if $ac_need_defaults; then
+  test "${CONFIG_FILES+set}" = set || CONFIG_FILES=$config_files
+  test "${CONFIG_HEADERS+set}" = set || CONFIG_HEADERS=$config_headers
+  test "${CONFIG_COMMANDS+set}" = set || CONFIG_COMMANDS=$config_commands
+fi
+
+# Have a temporary directory for convenience.  Make it in the build tree
+# simply because there is no reason against having it here, and in addition,
+# creating and moving files from /tmp can sometimes cause problems.
+# Hook for its removal unless debugging.
+# Note that there is a small window in which the directory will not be cleaned:
+# after its creation but before its name has been assigned to `$tmp'.
+$debug ||
+{
+  tmp= ac_tmp=
+  trap 'exit_status=$?
+  : "${ac_tmp:=$tmp}"
+  { test ! -d "$ac_tmp" || rm -fr "$ac_tmp"; } && exit $exit_status
+' 0
+  trap 'as_fn_exit 1' 1 2 13 15
+}
+# Create a (secure) tmp directory for tmp files.
+
+{
+  tmp=`(umask 077 && mktemp -d "./confXXXXXX") 2>/dev/null` &&
+  test -d "$tmp"
+}  ||
+{
+  tmp=./conf$$-$RANDOM
+  (umask 077 && mkdir "$tmp")
+} || as_fn_error $? "cannot create a temporary directory in ." "$LINENO" 5
+ac_tmp=$tmp
+
+# Set up the scripts for CONFIG_FILES section.
+# No need to generate them if there are no CONFIG_FILES.
+# This happens for instance with `./config.status config.h'.
+if test -n "$CONFIG_FILES"; then
+
+
+ac_cr=`echo X | tr X '\015'`
+# On cygwin, bash can eat \r inside `` if the user requested igncr.
+# But we know of no other shell where ac_cr would be empty at this
+# point, so we can use a bashism as a fallback.
+if test "x$ac_cr" = x; then
+  eval ac_cr=\$\'\\r\'
+fi
+ac_cs_awk_cr=`$AWK 'BEGIN { print "a\rb" }' </dev/null 2>/dev/null`
+if test "$ac_cs_awk_cr" = "a${ac_cr}b"; then
+  ac_cs_awk_cr='\\r'
+else
+  ac_cs_awk_cr=$ac_cr
+fi
+
+echo 'BEGIN {' >"$ac_tmp/subs1.awk" &&
+_ACEOF
+
+
+{
+  echo "cat >conf$$subs.awk <<_ACEOF" &&
+  echo "$ac_subst_vars" | sed 's/.*/&!$&$ac_delim/' &&
+  echo "_ACEOF"
+} >conf$$subs.sh ||
+  as_fn_error $? "could not make $CONFIG_STATUS" "$LINENO" 5
+ac_delim_num=`echo "$ac_subst_vars" | grep -c '^'`
+ac_delim='%!_!# '
+for ac_last_try in false false false false false :; do
+  . ./conf$$subs.sh ||
+    as_fn_error $? "could not make $CONFIG_STATUS" "$LINENO" 5
+
+  ac_delim_n=`sed -n "s/.*$ac_delim\$/X/p" conf$$subs.awk | grep -c X`
+  if test $ac_delim_n = $ac_delim_num; then
+    break
+  elif $ac_last_try; then
+    as_fn_error $? "could not make $CONFIG_STATUS" "$LINENO" 5
+  else
+    ac_delim="$ac_delim!$ac_delim _$ac_delim!! "
+  fi
+done
+rm -f conf$$subs.sh
+
+cat >>$CONFIG_STATUS <<_ACEOF || ac_write_fail=1
+cat >>"\$ac_tmp/subs1.awk" <<\\_ACAWK &&
+_ACEOF
+sed -n '
+h
+s/^/S["/; s/!.*/"]=/
+p
+g
+s/^[^!]*!//
+:repl
+t repl
+s/'"$ac_delim"'$//
+t delim
+:nl
+h
+s/\(.\{148\}\)..*/\1/
+t more1
+s/["\\]/\\&/g; s/^/"/; s/$/\\n"\\/
+p
+n
+b repl
+:more1
+s/["\\]/\\&/g; s/^/"/; s/$/"\\/
+p
+g
+s/.\{148\}//
+t nl
+:delim
+h
+s/\(.\{148\}\)..*/\1/
+t more2
+s/["\\]/\\&/g; s/^/"/; s/$/"/
+p
+b
+:more2
+s/["\\]/\\&/g; s/^/"/; s/$/"\\/
+p
+g
+s/.\{148\}//
+t delim
+' <conf$$subs.awk | sed '
+/^[^""]/{
+  N
+  s/\n//
+}
+' >>$CONFIG_STATUS || ac_write_fail=1
+rm -f conf$$subs.awk
+cat >>$CONFIG_STATUS <<_ACEOF || ac_write_fail=1
+_ACAWK
+cat >>"\$ac_tmp/subs1.awk" <<_ACAWK &&
+  for (key in S) S_is_set[key] = 1
+  FS = ""
+
+}
+{
+  line = $ 0
+  nfields = split(line, field, "@")
+  substed = 0
+  len = length(field[1])
+  for (i = 2; i < nfields; i++) {
+    key = field[i]
+    keylen = length(key)
+    if (S_is_set[key]) {
+      value = S[key]
+      line = substr(line, 1, len) "" value "" substr(line, len + keylen + 3)
+      len += length(value) + length(field[++i])
+      substed = 1
+    } else
+      len += 1 + keylen
+  }
+
+  print line
+}
+
+_ACAWK
+_ACEOF
+cat >>$CONFIG_STATUS <<\_ACEOF || ac_write_fail=1
+if sed "s/$ac_cr//" < /dev/null > /dev/null 2>&1; then
+  sed "s/$ac_cr\$//; s/$ac_cr/$ac_cs_awk_cr/g"
+else
+  cat
+fi < "$ac_tmp/subs1.awk" > "$ac_tmp/subs.awk" \
+  || as_fn_error $? "could not setup config files machinery" "$LINENO" 5
+_ACEOF
+
+# VPATH may cause trouble with some makes, so we remove sole $(srcdir),
+# ${srcdir} and @srcdir@ entries from VPATH if srcdir is ".", strip leading and
+# trailing colons and then remove the whole line if VPATH becomes empty
+# (actually we leave an empty line to preserve line numbers).
+if test "x$srcdir" = x.; then
+  ac_vpsub='/^[	 ]*VPATH[	 ]*=[	 ]*/{
+h
+s///
+s/^/:/
+s/[	 ]*$/:/
+s/:\$(srcdir):/:/g
+s/:\${srcdir}:/:/g
+s/:@srcdir@:/:/g
+s/^:*//
+s/:*$//
+x
+s/\(=[	 ]*\).*/\1/
+G
+s/\n//
+s/^[^=]*=[	 ]*$//
+}'
+fi
+
+cat >>$CONFIG_STATUS <<\_ACEOF || ac_write_fail=1
+fi # test -n "$CONFIG_FILES"
+
+# Set up the scripts for CONFIG_HEADERS section.
+# No need to generate them if there are no CONFIG_HEADERS.
+# This happens for instance with `./config.status Makefile'.
+if test -n "$CONFIG_HEADERS"; then
+cat >"$ac_tmp/defines.awk" <<\_ACAWK ||
+BEGIN {
+_ACEOF
+
+# Transform confdefs.h into an awk script `defines.awk', embedded as
+# here-document in config.status, that substitutes the proper values into
+# config.h.in to produce config.h.
+
+# Create a delimiter string that does not exist in confdefs.h, to ease
+# handling of long lines.
+ac_delim='%!_!# '
+for ac_last_try in false false :; do
+  ac_tt=`sed -n "/$ac_delim/p" confdefs.h`
+  if test -z "$ac_tt"; then
+    break
+  elif $ac_last_try; then
+    as_fn_error $? "could not make $CONFIG_HEADERS" "$LINENO" 5
+  else
+    ac_delim="$ac_delim!$ac_delim _$ac_delim!! "
+  fi
+done
+
+# For the awk script, D is an array of macro values keyed by name,
+# likewise P contains macro parameters if any.  Preserve backslash
+# newline sequences.
+
+ac_word_re=[_$as_cr_Letters][_$as_cr_alnum]*
+sed -n '
+s/.\{148\}/&'"$ac_delim"'/g
+t rset
+:rset
+s/^[	 ]*#[	 ]*define[	 ][	 ]*/ /
+t def
+d
+:def
+s/\\$//
+t bsnl
+s/["\\]/\\&/g
+s/^ \('"$ac_word_re"'\)\(([^()]*)\)[	 ]*\(.*\)/P["\1"]="\2"\
+D["\1"]=" \3"/p
+s/^ \('"$ac_word_re"'\)[	 ]*\(.*\)/D["\1"]=" \2"/p
+d
+:bsnl
+s/["\\]/\\&/g
+s/^ \('"$ac_word_re"'\)\(([^()]*)\)[	 ]*\(.*\)/P["\1"]="\2"\
+D["\1"]=" \3\\\\\\n"\\/p
+t cont
+s/^ \('"$ac_word_re"'\)[	 ]*\(.*\)/D["\1"]=" \2\\\\\\n"\\/p
+t cont
+d
+:cont
+n
+s/.\{148\}/&'"$ac_delim"'/g
+t clear
+:clear
+s/\\$//
+t bsnlc
+s/["\\]/\\&/g; s/^/"/; s/$/"/p
+d
+:bsnlc
+s/["\\]/\\&/g; s/^/"/; s/$/\\\\\\n"\\/p
+b cont
+' <confdefs.h | sed '
+s/'"$ac_delim"'/"\\\
+"/g' >>$CONFIG_STATUS || ac_write_fail=1
+
+cat >>$CONFIG_STATUS <<_ACEOF || ac_write_fail=1
+  for (key in D) D_is_set[key] = 1
+  FS = ""
+}
+/^[\t ]*#[\t ]*(define|undef)[\t ]+$ac_word_re([\t (]|\$)/ {
+  line = \$ 0
+  split(line, arg, " ")
+  if (arg[1] == "#") {
+    defundef = arg[2]
+    mac1 = arg[3]
+  } else {
+    defundef = substr(arg[1], 2)
+    mac1 = arg[2]
+  }
+  split(mac1, mac2, "(") #)
+  macro = mac2[1]
+  prefix = substr(line, 1, index(line, defundef) - 1)
+  if (D_is_set[macro]) {
+    # Preserve the white space surrounding the "#".
+    print prefix "define", macro P[macro] D[macro]
+    next
+  } else {
+    # Replace #undef with comments.  This is necessary, for example,
+    # in the case of _POSIX_SOURCE, which is predefined and required
+    # on some systems where configure will not decide to define it.
+    if (defundef == "undef") {
+      print "/*", prefix defundef, macro, "*/"
+      next
+    }
+  }
+}
+{ print }
+_ACAWK
+_ACEOF
+cat >>$CONFIG_STATUS <<\_ACEOF || ac_write_fail=1
+  as_fn_error $? "could not setup config headers machinery" "$LINENO" 5
+fi # test -n "$CONFIG_HEADERS"
+
+
+eval set X "  :F $CONFIG_FILES  :H $CONFIG_HEADERS    :C $CONFIG_COMMANDS"
+shift
+for ac_tag
+do
+  case $ac_tag in
+  :[FHLC]) ac_mode=$ac_tag; continue;;
+  esac
+  case $ac_mode$ac_tag in
+  :[FHL]*:*);;
+  :L* | :C*:*) as_fn_error $? "invalid tag \`$ac_tag'" "$LINENO" 5;;
+  :[FH]-) ac_tag=-:-;;
+  :[FH]*) ac_tag=$ac_tag:$ac_tag.in;;
+  esac
+  ac_save_IFS=$IFS
+  IFS=:
+  set x $ac_tag
+  IFS=$ac_save_IFS
+  shift
+  ac_file=$1
+  shift
+
+  case $ac_mode in
+  :L) ac_source=$1;;
+  :[FH])
+    ac_file_inputs=
+    for ac_f
+    do
+      case $ac_f in
+      -) ac_f="$ac_tmp/stdin";;
+      *) # Look for the file first in the build tree, then in the source tree
+	 # (if the path is not absolute).  The absolute path cannot be DOS-style,
+	 # because $ac_f cannot contain `:'.
+	 test -f "$ac_f" ||
+	   case $ac_f in
+	   [\\/$]*) false;;
+	   *) test -f "$srcdir/$ac_f" && ac_f="$srcdir/$ac_f";;
+	   esac ||
+	   as_fn_error 1 "cannot find input file: \`$ac_f'" "$LINENO" 5;;
+      esac
+      case $ac_f in *\'*) ac_f=`$as_echo "$ac_f" | sed "s/'/'\\\\\\\\''/g"`;; esac
+      as_fn_append ac_file_inputs " '$ac_f'"
+    done
+
+    # Let's still pretend it is `configure' which instantiates (i.e., don't
+    # use $as_me), people would be surprised to read:
+    #    /* config.h.  Generated by config.status.  */
+    configure_input='Generated from '`
+	  $as_echo "$*" | sed 's|^[^:]*/||;s|:[^:]*/|, |g'
+	`' by configure.'
+    if test x"$ac_file" != x-; then
+      configure_input="$ac_file.  $configure_input"
+      { $as_echo "$as_me:${as_lineno-$LINENO}: creating $ac_file" >&5
+$as_echo "$as_me: creating $ac_file" >&6;}
+    fi
+    # Neutralize special characters interpreted by sed in replacement strings.
+    case $configure_input in #(
+    *\&* | *\|* | *\\* )
+       ac_sed_conf_input=`$as_echo "$configure_input" |
+       sed 's/[\\\\&|]/\\\\&/g'`;; #(
+    *) ac_sed_conf_input=$configure_input;;
+    esac
+
+    case $ac_tag in
+    *:-:* | *:-) cat >"$ac_tmp/stdin" \
+      || as_fn_error $? "could not create $ac_file" "$LINENO" 5 ;;
+    esac
+    ;;
+  esac
+
+  ac_dir=`$as_dirname -- "$ac_file" ||
+$as_expr X"$ac_file" : 'X\(.*[^/]\)//*[^/][^/]*/*$' \| \
+	 X"$ac_file" : 'X\(//\)[^/]' \| \
+	 X"$ac_file" : 'X\(//\)$' \| \
+	 X"$ac_file" : 'X\(/\)' \| . 2>/dev/null ||
+$as_echo X"$ac_file" |
+    sed '/^X\(.*[^/]\)\/\/*[^/][^/]*\/*$/{
+	    s//\1/
+	    q
+	  }
+	  /^X\(\/\/\)[^/].*/{
+	    s//\1/
+	    q
+	  }
+	  /^X\(\/\/\)$/{
+	    s//\1/
+	    q
+	  }
+	  /^X\(\/\).*/{
+	    s//\1/
+	    q
+	  }
+	  s/.*/./; q'`
+  as_dir="$ac_dir"; as_fn_mkdir_p
+  ac_builddir=.
+
+case "$ac_dir" in
+.) ac_dir_suffix= ac_top_builddir_sub=. ac_top_build_prefix= ;;
+*)
+  ac_dir_suffix=/`$as_echo "$ac_dir" | sed 's|^\.[\\/]||'`
+  # A ".." for each directory in $ac_dir_suffix.
+  ac_top_builddir_sub=`$as_echo "$ac_dir_suffix" | sed 's|/[^\\/]*|/..|g;s|/||'`
+  case $ac_top_builddir_sub in
+  "") ac_top_builddir_sub=. ac_top_build_prefix= ;;
+  *)  ac_top_build_prefix=$ac_top_builddir_sub/ ;;
+  esac ;;
+esac
+ac_abs_top_builddir=$ac_pwd
+ac_abs_builddir=$ac_pwd$ac_dir_suffix
+# for backward compatibility:
+ac_top_builddir=$ac_top_build_prefix
+
+case $srcdir in
+  .)  # We are building in place.
+    ac_srcdir=.
+    ac_top_srcdir=$ac_top_builddir_sub
+    ac_abs_top_srcdir=$ac_pwd ;;
+  [\\/]* | ?:[\\/]* )  # Absolute name.
+    ac_srcdir=$srcdir$ac_dir_suffix;
+    ac_top_srcdir=$srcdir
+    ac_abs_top_srcdir=$srcdir ;;
+  *) # Relative name.
+    ac_srcdir=$ac_top_build_prefix$srcdir$ac_dir_suffix
+    ac_top_srcdir=$ac_top_build_prefix$srcdir
+    ac_abs_top_srcdir=$ac_pwd/$srcdir ;;
+esac
+ac_abs_srcdir=$ac_abs_top_srcdir$ac_dir_suffix
+
+
+  case $ac_mode in
+  :F)
+  #
+  # CONFIG_FILE
+  #
+
+  case $INSTALL in
+  [\\/$]* | ?:[\\/]* ) ac_INSTALL=$INSTALL ;;
+  *) ac_INSTALL=$ac_top_build_prefix$INSTALL ;;
+  esac
+  ac_MKDIR_P=$MKDIR_P
+  case $MKDIR_P in
+  [\\/$]* | ?:[\\/]* ) ;;
+  */*) ac_MKDIR_P=$ac_top_build_prefix$MKDIR_P ;;
+  esac
+_ACEOF
+
+cat >>$CONFIG_STATUS <<\_ACEOF || ac_write_fail=1
+# If the template does not know about datarootdir, expand it.
+# FIXME: This hack should be removed a few years after 2.60.
+ac_datarootdir_hack=; ac_datarootdir_seen=
+ac_sed_dataroot='
+/datarootdir/ {
+  p
+  q
+}
+/@datadir@/p
+/@docdir@/p
+/@infodir@/p
+/@localedir@/p
+/@mandir@/p'
+case `eval "sed -n \"\$ac_sed_dataroot\" $ac_file_inputs"` in
+*datarootdir*) ac_datarootdir_seen=yes;;
+*@datadir@*|*@docdir@*|*@infodir@*|*@localedir@*|*@mandir@*)
+  { $as_echo "$as_me:${as_lineno-$LINENO}: WARNING: $ac_file_inputs seems to ignore the --datarootdir setting" >&5
+$as_echo "$as_me: WARNING: $ac_file_inputs seems to ignore the --datarootdir setting" >&2;}
+_ACEOF
+cat >>$CONFIG_STATUS <<_ACEOF || ac_write_fail=1
+  ac_datarootdir_hack='
+  s&@datadir@&$datadir&g
+  s&@docdir@&$docdir&g
+  s&@infodir@&$infodir&g
+  s&@localedir@&$localedir&g
+  s&@mandir@&$mandir&g
+  s&\\\${datarootdir}&$datarootdir&g' ;;
+esac
+_ACEOF
+
+# Neutralize VPATH when `$srcdir' = `.'.
+# Shell code in configure.ac might set extrasub.
+# FIXME: do we really want to maintain this feature?
+cat >>$CONFIG_STATUS <<_ACEOF || ac_write_fail=1
+ac_sed_extra="$ac_vpsub
+$extrasub
+_ACEOF
+cat >>$CONFIG_STATUS <<\_ACEOF || ac_write_fail=1
+:t
+/@[a-zA-Z_][a-zA-Z_0-9]*@/!b
+s|@configure_input@|$ac_sed_conf_input|;t t
+s&@top_builddir@&$ac_top_builddir_sub&;t t
+s&@top_build_prefix@&$ac_top_build_prefix&;t t
+s&@srcdir@&$ac_srcdir&;t t
+s&@abs_srcdir@&$ac_abs_srcdir&;t t
+s&@top_srcdir@&$ac_top_srcdir&;t t
+s&@abs_top_srcdir@&$ac_abs_top_srcdir&;t t
+s&@builddir@&$ac_builddir&;t t
+s&@abs_builddir@&$ac_abs_builddir&;t t
+s&@abs_top_builddir@&$ac_abs_top_builddir&;t t
+s&@INSTALL@&$ac_INSTALL&;t t
+s&@MKDIR_P@&$ac_MKDIR_P&;t t
+$ac_datarootdir_hack
+"
+eval sed \"\$ac_sed_extra\" "$ac_file_inputs" | $AWK -f "$ac_tmp/subs.awk" \
+  >$ac_tmp/out || as_fn_error $? "could not create $ac_file" "$LINENO" 5
+
+test -z "$ac_datarootdir_hack$ac_datarootdir_seen" &&
+  { ac_out=`sed -n '/\${datarootdir}/p' "$ac_tmp/out"`; test -n "$ac_out"; } &&
+  { ac_out=`sed -n '/^[	 ]*datarootdir[	 ]*:*=/p' \
+      "$ac_tmp/out"`; test -z "$ac_out"; } &&
+  { $as_echo "$as_me:${as_lineno-$LINENO}: WARNING: $ac_file contains a reference to the variable \`datarootdir'
+which seems to be undefined.  Please make sure it is defined" >&5
+$as_echo "$as_me: WARNING: $ac_file contains a reference to the variable \`datarootdir'
+which seems to be undefined.  Please make sure it is defined" >&2;}
+
+  rm -f "$ac_tmp/stdin"
+  case $ac_file in
+  -) cat "$ac_tmp/out" && rm -f "$ac_tmp/out";;
+  *) rm -f "$ac_file" && mv "$ac_tmp/out" "$ac_file";;
+  esac \
+  || as_fn_error $? "could not create $ac_file" "$LINENO" 5
+ ;;
+  :H)
+  #
+  # CONFIG_HEADER
+  #
+  if test x"$ac_file" != x-; then
+    {
+      $as_echo "/* $configure_input  */" \
+      && eval '$AWK -f "$ac_tmp/defines.awk"' "$ac_file_inputs"
+    } >"$ac_tmp/config.h" \
+      || as_fn_error $? "could not create $ac_file" "$LINENO" 5
+    if diff "$ac_file" "$ac_tmp/config.h" >/dev/null 2>&1; then
+      { $as_echo "$as_me:${as_lineno-$LINENO}: $ac_file is unchanged" >&5
+$as_echo "$as_me: $ac_file is unchanged" >&6;}
+    else
+      rm -f "$ac_file"
+      mv "$ac_tmp/config.h" "$ac_file" \
+	|| as_fn_error $? "could not create $ac_file" "$LINENO" 5
+    fi
+  else
+    $as_echo "/* $configure_input  */" \
+      && eval '$AWK -f "$ac_tmp/defines.awk"' "$ac_file_inputs" \
+      || as_fn_error $? "could not create -" "$LINENO" 5
+  fi
+# Compute "$ac_file"'s index in $config_headers.
+_am_arg="$ac_file"
+_am_stamp_count=1
+for _am_header in $config_headers :; do
+  case $_am_header in
+    $_am_arg | $_am_arg:* )
+      break ;;
+    * )
+      _am_stamp_count=`expr $_am_stamp_count + 1` ;;
+  esac
+done
+echo "timestamp for $_am_arg" >`$as_dirname -- "$_am_arg" ||
+$as_expr X"$_am_arg" : 'X\(.*[^/]\)//*[^/][^/]*/*$' \| \
+	 X"$_am_arg" : 'X\(//\)[^/]' \| \
+	 X"$_am_arg" : 'X\(//\)$' \| \
+	 X"$_am_arg" : 'X\(/\)' \| . 2>/dev/null ||
+$as_echo X"$_am_arg" |
+    sed '/^X\(.*[^/]\)\/\/*[^/][^/]*\/*$/{
+	    s//\1/
+	    q
+	  }
+	  /^X\(\/\/\)[^/].*/{
+	    s//\1/
+	    q
+	  }
+	  /^X\(\/\/\)$/{
+	    s//\1/
+	    q
+	  }
+	  /^X\(\/\).*/{
+	    s//\1/
+	    q
+	  }
+	  s/.*/./; q'`/stamp-h$_am_stamp_count
+ ;;
+
+  :C)  { $as_echo "$as_me:${as_lineno-$LINENO}: executing $ac_file commands" >&5
+$as_echo "$as_me: executing $ac_file commands" >&6;}
+ ;;
+  esac
+
+
+  case $ac_file$ac_mode in
+    "depfiles":C) test x"$AMDEP_TRUE" != x"" || {
+  # Older Autoconf quotes --file arguments for eval, but not when files
+  # are listed without --file.  Let's play safe and only enable the eval
+  # if we detect the quoting.
+  case $CONFIG_FILES in
+  *\'*) eval set x "$CONFIG_FILES" ;;
+  *)   set x $CONFIG_FILES ;;
+  esac
+  shift
+  for mf
+  do
+    # Strip MF so we end up with the name of the file.
+    mf=`echo "$mf" | sed -e 's/:.*$//'`
+    # Check whether this is an Automake generated Makefile or not.
+    # We used to match only the files named 'Makefile.in', but
+    # some people rename them; so instead we look at the file content.
+    # Grep'ing the first line is not enough: some people post-process
+    # each Makefile.in and add a new line on top of each file to say so.
+    # Grep'ing the whole file is not good either: AIX grep has a line
+    # limit of 2048, but all sed's we know have understand at least 4000.
+    if sed -n 's,^#.*generated by automake.*,X,p' "$mf" | grep X >/dev/null 2>&1; then
+      dirpart=`$as_dirname -- "$mf" ||
+$as_expr X"$mf" : 'X\(.*[^/]\)//*[^/][^/]*/*$' \| \
+	 X"$mf" : 'X\(//\)[^/]' \| \
+	 X"$mf" : 'X\(//\)$' \| \
+	 X"$mf" : 'X\(/\)' \| . 2>/dev/null ||
+$as_echo X"$mf" |
+    sed '/^X\(.*[^/]\)\/\/*[^/][^/]*\/*$/{
+	    s//\1/
+	    q
+	  }
+	  /^X\(\/\/\)[^/].*/{
+	    s//\1/
+	    q
+	  }
+	  /^X\(\/\/\)$/{
+	    s//\1/
+	    q
+	  }
+	  /^X\(\/\).*/{
+	    s//\1/
+	    q
+	  }
+	  s/.*/./; q'`
+    else
+      continue
+    fi
+    # Extract the definition of DEPDIR, am__include, and am__quote
+    # from the Makefile without running 'make'.
+    DEPDIR=`sed -n 's/^DEPDIR = //p' < "$mf"`
+    test -z "$DEPDIR" && continue
+    am__include=`sed -n 's/^am__include = //p' < "$mf"`
+    test -z "$am__include" && continue
+    am__quote=`sed -n 's/^am__quote = //p' < "$mf"`
+    # Find all dependency output files, they are included files with
+    # $(DEPDIR) in their names.  We invoke sed twice because it is the
+    # simplest approach to changing $(DEPDIR) to its actual value in the
+    # expansion.
+    for file in `sed -n "
+      s/^$am__include $am__quote\(.*(DEPDIR).*\)$am__quote"'$/\1/p' <"$mf" | \
+	 sed -e 's/\$(DEPDIR)/'"$DEPDIR"'/g'`; do
+      # Make sure the directory exists.
+      test -f "$dirpart/$file" && continue
+      fdir=`$as_dirname -- "$file" ||
+$as_expr X"$file" : 'X\(.*[^/]\)//*[^/][^/]*/*$' \| \
+	 X"$file" : 'X\(//\)[^/]' \| \
+	 X"$file" : 'X\(//\)$' \| \
+	 X"$file" : 'X\(/\)' \| . 2>/dev/null ||
+$as_echo X"$file" |
+    sed '/^X\(.*[^/]\)\/\/*[^/][^/]*\/*$/{
+	    s//\1/
+	    q
+	  }
+	  /^X\(\/\/\)[^/].*/{
+	    s//\1/
+	    q
+	  }
+	  /^X\(\/\/\)$/{
+	    s//\1/
+	    q
+	  }
+	  /^X\(\/\).*/{
+	    s//\1/
+	    q
+	  }
+	  s/.*/./; q'`
+      as_dir=$dirpart/$fdir; as_fn_mkdir_p
+      # echo "creating $dirpart/$file"
+      echo '# dummy' > "$dirpart/$file"
+    done
+  done
+}
+ ;;
+
+  esac
+done # for ac_tag
+
+
+as_fn_exit 0
+_ACEOF
+ac_clean_files=$ac_clean_files_save
+
+test $ac_write_fail = 0 ||
+  as_fn_error $? "write failure creating $CONFIG_STATUS" "$LINENO" 5
+
+
+# configure is writing to config.log, and then calls config.status.
+# config.status does its own redirection, appending to config.log.
+# Unfortunately, on DOS this fails, as config.log is still kept open
+# by configure, so config.status won't be able to write to it; its
+# output is simply discarded.  So we exec the FD to /dev/null,
+# effectively closing config.log, so it can be properly (re)opened and
+# appended to by config.status.  When coming back to configure, we
+# need to make the FD available again.
+if test "$no_create" != yes; then
+  ac_cs_success=:
+  ac_config_status_args=
+  test "$silent" = yes &&
+    ac_config_status_args="$ac_config_status_args --quiet"
+  exec 5>/dev/null
+  $SHELL $CONFIG_STATUS $ac_config_status_args || ac_cs_success=false
+  exec 5>>config.log
+  # Use ||, not &&, to avoid exiting from the if with $? = 1, which
+  # would make configure fail if this is the last instruction.
+  $ac_cs_success || as_fn_exit 1
+fi
+if test -n "$ac_unrecognized_opts" && test "$enable_option_checking" != no; then
+  { $as_echo "$as_me:${as_lineno-$LINENO}: WARNING: unrecognized options: $ac_unrecognized_opts" >&5
+$as_echo "$as_me: WARNING: unrecognized options: $ac_unrecognized_opts" >&2;}
+fi
+
diff --git a/configure.ac b/configure.ac
new file mode 100644
index 0000000000..52572fc92d
--- /dev/null
+++ b/configure.ac
@@ -0,0 +1,164 @@
+AC_INIT([ccminer], [2014.03.18])
+
+AC_PREREQ([2.59c])
+AC_CANONICAL_SYSTEM
+AC_CONFIG_SRCDIR([cpu-miner.c])
+AM_INIT_AUTOMAKE([gnu])
+AC_CONFIG_HEADERS([cpuminer-config.h])
+
+dnl Make sure anyone changing configure.ac/Makefile.am has a clue
+AM_MAINTAINER_MODE
+
+dnl Checks for programs
+AC_PROG_CC_C99
+AC_PROG_GCC_TRADITIONAL
+AM_PROG_CC_C_O
+AM_PROG_AS
+AC_PROG_RANLIB
+AC_PROG_CXX
+AC_OPENMP
+
+dnl Checks for header files
+AC_HEADER_STDC
+AC_CHECK_HEADERS([sys/endian.h sys/param.h syslog.h])
+# sys/sysctl.h requires sys/types.h on FreeBSD
+# sys/sysctl.h requires sys/param.h on OpenBSD
+AC_CHECK_HEADERS([sys/sysctl.h], [], [],
+[#include <sys/types.h>
+#ifdef HAVE_SYS_PARAM_H
+#include <sys/param.h>
+#endif
+])
+
+AC_CHECK_DECLS([be32dec, le32dec, be32enc, le32enc], [], [],
+[AC_INCLUDES_DEFAULT
+#ifdef HAVE_SYS_ENDIAN_H
+#include <sys/endian.h>
+#endif
+])
+
+AC_FUNC_ALLOCA
+AC_CHECK_FUNCS([getopt_long])
+
+case $target in
+  i*86-*-*)
+    have_x86=true
+    ;;
+  x86_64-*-*)
+    have_x86=true
+    have_x86_64=true
+    ;;
+  amd64-*-*)
+    have_x86=true
+    have_x86_64=true
+    ;;
+esac
+
+PTHREAD_FLAGS="-pthread"
+WS2_LIBS=""
+
+case $target in
+  *-*-mingw*)
+    have_win32=true
+    PTHREAD_FLAGS=""
+    WS2_LIBS="-lws2_32"
+    ;;
+esac
+
+if test x$have_x86 = xtrue
+then
+  AC_MSG_CHECKING(whether we can compile AVX code)
+  AC_COMPILE_IFELSE([AC_LANG_PROGRAM(,[asm ("vmovdqa %ymm0, %ymm1");])],
+    AC_DEFINE(USE_AVX, 1, [Define to 1 if AVX assembly is available.])
+    AC_MSG_RESULT(yes)
+    AC_MSG_CHECKING(whether we can compile XOP code)
+    AC_COMPILE_IFELSE([AC_LANG_PROGRAM(,[asm ("vprotd \$7, %xmm0, %xmm1");])],
+      AC_DEFINE(USE_XOP, 1, [Define to 1 if XOP assembly is available.])
+      AC_MSG_RESULT(yes)
+    ,
+      AC_MSG_RESULT(no)
+      AC_MSG_WARN([The assembler does not support the XOP instruction set.])
+    )
+    AC_MSG_CHECKING(whether we can compile AVX2 code)
+    AC_COMPILE_IFELSE([AC_LANG_PROGRAM(,[asm ("vpaddd %ymm0, %ymm1, %ymm2");])],
+      AC_DEFINE(USE_AVX2, 1, [Define to 1 if AVX2 assembly is available.])
+      AC_MSG_RESULT(yes)
+    ,
+      AC_MSG_RESULT(no)
+      AC_MSG_WARN([The assembler does not support the AVX2 instruction set.])
+    )
+  ,
+    AC_MSG_RESULT(no)
+    AC_MSG_WARN([The assembler does not support the AVX instruction set.])
+  )
+fi
+
+AC_CHECK_LIB(jansson, json_loads, request_jansson=false, request_jansson=true)
+AC_CHECK_LIB([pthread], [pthread_create], PTHREAD_LIBS="-lpthread",
+  AC_CHECK_LIB([pthreadGC2], [pthread_create], PTHREAD_LIBS="-lpthreadGC2",
+    AC_CHECK_LIB([pthreadGC1], [pthread_create], PTHREAD_LIBS="-lpthreadGC1",
+      AC_CHECK_LIB([pthreadGC], [pthread_create], PTHREAD_LIBS="-lpthreadGC"
+))))
+
+AC_CHECK_LIB([ssl],[SSL_library_init], [], [AC_MSG_ERROR([OpenSSL library required])])
+AC_CHECK_LIB([crypto],[EVP_DigestFinal_ex], [], [AC_MSG_ERROR([OpenSSL library required])])
+
+AM_CONDITIONAL([WANT_JANSSON], [test x$request_jansson = xtrue])
+AM_CONDITIONAL([HAVE_WINDOWS], [test x$have_win32 = xtrue])
+AM_CONDITIONAL([ARCH_x86], [test x$have_x86 = xtrue])
+AM_CONDITIONAL([ARCH_x86_64], [test x$have_x86_64 = xtrue])
+
+if test x$request_jansson = xtrue
+then
+	JANSSON_LIBS="compat/jansson/libjansson.a"
+else
+	JANSSON_LIBS=-ljansson
+fi
+
+LIBCURL_CHECK_CONFIG(, 7.15.2, ,
+  [AC_MSG_ERROR([Missing required libcurl >= 7.15.2])])
+
+AC_SUBST(JANSSON_LIBS)
+AC_SUBST(PTHREAD_FLAGS)
+AC_SUBST(PTHREAD_LIBS)
+AC_SUBST(WS2_LIBS)
+
+AC_CONFIG_FILES([
+	Makefile
+	compat/Makefile
+	compat/jansson/Makefile
+	])
+
+dnl find out what version we are running
+ARCH=`uname -m`
+if [[ $ARCH == "x86_64" ]];
+then
+  SUFFIX="64"
+else
+  SUFFIX=""
+fi
+
+dnl Setup CUDA paths
+AC_ARG_WITH([cuda],
+   [  --with-cuda=PATH    prefix where cuda is installed [default=/usr/local/cuda]])
+
+if test -n "$with_cuda"
+then
+   CUDA_CFLAGS="-I$with_cuda/include"
+   CUDA_LIBS="-lcudart"
+   CUDA_LDFLAGS="-L$with_cuda/lib$SUFFIX"
+   NVCC="$with_cuda/bin/nvcc"
+else
+   CUDA_CFLAGS="-I/usr/local/cuda/include"
+   CUDA_LIBS="-lcudart -static-libstdc++"
+   CUDA_LDFLAGS="-L/usr/local/cuda/lib$SUFFIX"
+   NVCC="nvcc"
+fi
+AC_SUBST(CUDA_CFLAGS)
+AC_SUBST(CUDA_LIBS)
+AC_SUBST(CUDA_LDFLAGS)
+AC_SUBST(NVCC)
+
+AC_SUBST(OPENMP_CFLAGS) 
+
+AC_OUTPUT
diff --git a/configure.sh b/configure.sh
new file mode 100644
index 0000000000..134abd163b
--- /dev/null
+++ b/configure.sh
@@ -0,0 +1 @@
+./configure "CFLAGS=-O3" "CXXFLAGS=-O3" --with-cuda=/usr/local/cuda
diff --git a/cpu-miner.c b/cpu-miner.c
new file mode 100644
index 0000000000..d3cb4efe7f
--- /dev/null
+++ b/cpu-miner.c
@@ -0,0 +1,1523 @@
+/*
+ * Copyright 2010 Jeff Garzik
+ * Copyright 2012-2014 pooler
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License as published by the Free
+ * Software Foundation; either version 2 of the License, or (at your option)
+ * any later version.  See COPYING for more details.
+ */
+
+#include "cpuminer-config.h"
+#define _GNU_SOURCE
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <stdbool.h>
+#include <inttypes.h>
+#include <unistd.h>
+#include <sys/time.h>
+#include <time.h>
+#ifdef WIN32
+#include <windows.h>
+#else
+#include <errno.h>
+#include <signal.h>
+#include <sys/resource.h>
+#if HAVE_SYS_SYSCTL_H
+#include <sys/types.h>
+#if HAVE_SYS_PARAM_H
+#include <sys/param.h>
+#endif
+#include <sys/sysctl.h>
+#endif
+#endif
+#include <jansson.h>
+#include <curl/curl.h>
+#include "compat.h"
+#include "miner.h"
+
+#define PROGRAM_NAME		"minerd"
+#define LP_SCANTIME		60
+#define HEAVYCOIN_BLKHDR_SZ		84
+
+// from heavy.cu
+#ifdef __cplusplus
+extern "C"
+#endif
+int cuda_num_devices();
+
+
+#ifdef __linux /* Linux specific policy and affinity management */
+#include <sched.h>
+static inline void drop_policy(void)
+{
+	struct sched_param param;
+	param.sched_priority = 0;
+
+#ifdef SCHED_IDLE
+	if (unlikely(sched_setscheduler(0, SCHED_IDLE, &param) == -1))
+#endif
+#ifdef SCHED_BATCH
+		sched_setscheduler(0, SCHED_BATCH, &param);
+#endif
+}
+
+static inline void affine_to_cpu(int id, int cpu)
+{
+	cpu_set_t set;
+
+	CPU_ZERO(&set);
+	CPU_SET(cpu, &set);
+	sched_setaffinity(0, sizeof(&set), &set);
+}
+#elif defined(__FreeBSD__) /* FreeBSD specific policy and affinity management */
+#include <sys/cpuset.h>
+static inline void drop_policy(void)
+{
+}
+
+static inline void affine_to_cpu(int id, int cpu)
+{
+	cpuset_t set;
+	CPU_ZERO(&set);
+	CPU_SET(cpu, &set);
+	cpuset_setaffinity(CPU_LEVEL_WHICH, CPU_WHICH_TID, -1, sizeof(cpuset_t), &set);
+}
+#else
+static inline void drop_policy(void)
+{
+}
+
+static inline void affine_to_cpu(int id, int cpu)
+{
+}
+#endif
+		
+enum workio_commands {
+	WC_GET_WORK,
+	WC_SUBMIT_WORK,
+};
+
+struct workio_cmd {
+	enum workio_commands	cmd;
+	struct thr_info		*thr;
+	union {
+		struct work	*work;
+	} u;
+};
+
+typedef enum {
+	ALGO_SCRYPT,		/* scrypt(1024,1,1) */
+	ALGO_SHA256D,		/* SHA-256d */
+	ALGO_HEAVY,		/* Heavycoin hash */
+	ALGO_FUGUE256,		/* Fugue256 */
+} sha256_algos;
+
+static const char *algo_names[] = {
+	"scrypt",
+	"sha256d",
+	"heavy",
+	"fugue256"
+};
+
+bool opt_debug = false;
+bool opt_protocol = false;
+static bool opt_benchmark = false;
+bool want_longpoll = true;
+bool have_longpoll = false;
+bool want_stratum = true;
+bool have_stratum = false;
+static bool submit_old = false;
+bool use_syslog = false;
+static bool opt_background = false;
+static bool opt_quiet = false;
+static int opt_retries = -1;
+static int opt_fail_pause = 30;
+int opt_timeout = 270;
+static int opt_scantime = 5;
+static json_t *opt_config;
+static const bool opt_time = true;
+static sha256_algos opt_algo = ALGO_HEAVY;
+static int opt_n_threads;
+bool opt_trust_pool = false;
+uint16_t opt_vote = 9999;
+static int num_processors;
+static char *rpc_url;
+static char *rpc_userpass;
+static char *rpc_user, *rpc_pass;
+char *opt_cert;
+char *opt_proxy;
+long opt_proxy_type;
+struct thr_info *thr_info;
+static int work_thr_id;
+int longpoll_thr_id = -1;
+int stratum_thr_id = -1;
+struct work_restart *work_restart = NULL;
+static struct stratum_ctx stratum;
+
+pthread_mutex_t applog_lock;
+static pthread_mutex_t stats_lock;
+
+static unsigned long accepted_count = 0L;
+static unsigned long rejected_count = 0L;
+static double *thr_hashrates;
+
+#ifdef HAVE_GETOPT_LONG
+#include <getopt.h>
+#else
+struct option {
+	const char *name;
+	int has_arg;
+	int *flag;
+	int val;
+};
+#endif
+
+static char const usage[] = "\
+Usage: " PROGRAM_NAME " [OPTIONS]\n\
+Options:\n\
+  -a, --algo=ALGO       specify the algorithm to use\n\
+                          scrypt    scrypt(1024, 1, 1) (default)\n\
+                          sha256d   SHA-256d\n\
+                          heavy     Heavycoin hash\n\
+  -v, --vote=VOTE       block reward vote\n\
+  -m, --trust-pool      trust the max block reward vote (maxvote) sent by the pool\n\
+  -o, --url=URL         URL of mining server\n\
+  -O, --userpass=U:P    username:password pair for mining server\n\
+  -u, --user=USERNAME   username for mining server\n\
+  -p, --pass=PASSWORD   password for mining server\n\
+      --cert=FILE       certificate for mining server using SSL\n\
+  -x, --proxy=[PROTOCOL://]HOST[:PORT]  connect through a proxy\n\
+  -t, --threads=N       number of miner threads (default: number of processors)\n\
+  -r, --retries=N       number of times to retry if a network call fails\n\
+                          (default: retry indefinitely)\n\
+  -R, --retry-pause=N   time to pause between retries, in seconds (default: 30)\n\
+  -T, --timeout=N       network timeout, in seconds (default: 270)\n\
+  -s, --scantime=N      upper bound on time spent scanning current work when\n\
+                          long polling is unavailable, in seconds (default: 5)\n\
+      --no-longpoll     disable X-Long-Polling support\n\
+      --no-stratum      disable X-Stratum support\n\
+  -q, --quiet           disable per-thread hashmeter output\n\
+  -D, --debug           enable debug output\n\
+  -P, --protocol-dump   verbose dump of protocol-level activities\n"
+#ifdef HAVE_SYSLOG_H
+"\
+  -S, --syslog          use system log for output messages\n"
+#endif
+#ifndef WIN32
+"\
+  -B, --background      run the miner in the background\n"
+#endif
+"\
+      --benchmark       run in offline benchmark mode\n\
+  -c, --config=FILE     load a JSON-format configuration file\n\
+  -V, --version         display version information and exit\n\
+  -h, --help            display this help text and exit\n\
+";
+
+static char const short_options[] =
+#ifndef WIN32
+	"B"
+#endif
+#ifdef HAVE_SYSLOG_H
+	"S"
+#endif
+	"a:c:Dhp:Px:qr:R:s:t:T:o:u:O:Vmv:";
+
+static struct option const options[] = {
+	{ "algo", 1, NULL, 'a' },
+#ifndef WIN32
+	{ "background", 0, NULL, 'B' },
+#endif
+	{ "benchmark", 0, NULL, 1005 },
+	{ "cert", 1, NULL, 1001 },
+	{ "config", 1, NULL, 'c' },
+	{ "debug", 0, NULL, 'D' },
+	{ "help", 0, NULL, 'h' },
+	{ "no-longpoll", 0, NULL, 1003 },
+	{ "no-stratum", 0, NULL, 1007 },
+	{ "pass", 1, NULL, 'p' },
+	{ "protocol-dump", 0, NULL, 'P' },
+	{ "proxy", 1, NULL, 'x' },
+	{ "quiet", 0, NULL, 'q' },
+	{ "retries", 1, NULL, 'r' },
+	{ "retry-pause", 1, NULL, 'R' },
+	{ "scantime", 1, NULL, 's' },
+#ifdef HAVE_SYSLOG_H
+	{ "syslog", 0, NULL, 'S' },
+#endif
+	{ "threads", 1, NULL, 't' },
+	{ "vote", 1, NULL, 'v' },
+	{ "trust-pool", 0, NULL, 'm' },
+	{ "timeout", 1, NULL, 'T' },
+	{ "url", 1, NULL, 'o' },
+	{ "user", 1, NULL, 'u' },
+	{ "userpass", 1, NULL, 'O' },
+	{ "version", 0, NULL, 'V' },
+	{ 0, 0, 0, 0 }
+};
+
+struct work {
+	uint32_t data[32];
+	uint32_t target[8];
+	uint32_t maxvote;
+
+	char job_id[128];
+	size_t xnonce2_len;
+	unsigned char xnonce2[32];
+};
+
+static struct work g_work;
+static time_t g_work_time;
+static pthread_mutex_t g_work_lock;
+
+static bool jobj_binary(const json_t *obj, const char *key,
+			void *buf, size_t buflen)
+{
+	const char *hexstr;
+	json_t *tmp;
+
+	tmp = json_object_get(obj, key);
+	if (unlikely(!tmp)) {
+		applog(LOG_ERR, "JSON key '%s' not found", key);
+		return false;
+	}
+	hexstr = json_string_value(tmp);
+	if (unlikely(!hexstr)) {
+		applog(LOG_ERR, "JSON key '%s' is not a string", key);
+		return false;
+	}
+	if (!hex2bin((unsigned char*)buf, hexstr, buflen))
+		return false;
+
+	return true;
+}
+
+static bool work_decode(const json_t *val, struct work *work)
+{
+	int i;
+	
+	if (unlikely(!jobj_binary(val, "data", work->data, sizeof(work->data)))) {
+		applog(LOG_ERR, "JSON inval data");
+		goto err_out;
+	}
+	if (unlikely(!jobj_binary(val, "target", work->target, sizeof(work->target)))) {
+		applog(LOG_ERR, "JSON inval target");
+		goto err_out;
+	}
+	if (opt_algo == ALGO_HEAVY) {
+		if (unlikely(!jobj_binary(val, "maxvote", &work->maxvote, sizeof(work->maxvote)))) {
+			work->maxvote = 1024;
+		}
+	} else work->maxvote = 0;
+
+	for (i = 0; i < ARRAY_SIZE(work->data); i++)
+		work->data[i] = le32dec(work->data + i);
+	for (i = 0; i < ARRAY_SIZE(work->target); i++)
+		work->target[i] = le32dec(work->target + i);
+
+	return true;
+
+err_out:
+	return false;
+}
+
+static void share_result(int result, const char *reason)
+{
+	char s[345];
+	double hashrate;
+	int i;
+
+	hashrate = 0.;
+	pthread_mutex_lock(&stats_lock);
+	for (i = 0; i < opt_n_threads; i++)
+		hashrate += thr_hashrates[i];
+	result ? accepted_count++ : rejected_count++;
+	pthread_mutex_unlock(&stats_lock);
+	
+	sprintf(s, hashrate >= 1e6 ? "%.0f" : "%.2f", 1e-3 * hashrate);
+	applog(LOG_INFO, "accepted: %lu/%lu (%.2f%%), %s khash/s %s",
+		   accepted_count,
+		   accepted_count + rejected_count,
+		   100. * accepted_count / (accepted_count + rejected_count),
+		   s,
+		   result ? "(yay!!!)" : "(booooo)");
+
+	if (opt_debug && reason)
+		applog(LOG_DEBUG, "DEBUG: reject reason: %s", reason);
+}
+
+static bool submit_upstream_work(CURL *curl, struct work *work)
+{
+	char *str = NULL;
+	json_t *val, *res, *reason;
+	char s[345];
+	int i;
+	bool rc = false;
+
+	/* pass if the previous hash is not the current previous hash */
+	if (memcmp(work->data + 1, g_work.data + 1, 32)) {
+		if (opt_debug)
+			applog(LOG_DEBUG, "DEBUG: stale work detected, discarding");
+		return true;
+	}
+
+	if (have_stratum) {
+		uint32_t ntime, nonce;
+		uint16_t nvote;
+		char *ntimestr, *noncestr, *xnonce2str, *nvotestr;
+
+		le32enc(&ntime, work->data[17]);
+		le32enc(&nonce, work->data[19]);
+		be16enc(&nvote, *((uint16_t*)&work->data[20]));
+
+		ntimestr = bin2hex((const unsigned char *)(&ntime), 4);
+		noncestr = bin2hex((const unsigned char *)(&nonce), 4);
+		xnonce2str = bin2hex(work->xnonce2, work->xnonce2_len);
+		nvotestr = bin2hex((const unsigned char *)(&nvote), 2);
+		if (opt_algo == ALGO_HEAVY) {
+			sprintf(s,
+				"{\"method\": \"mining.submit\", \"params\": [\"%s\", \"%s\", \"%s\", \"%s\", \"%s\", \"%s\"], \"id\":4}",
+				rpc_user, work->job_id, xnonce2str, ntimestr, noncestr, nvotestr);
+		} else {
+			sprintf(s,
+				"{\"method\": \"mining.submit\", \"params\": [\"%s\", \"%s\", \"%s\", \"%s\", \"%s\"], \"id\":4}",
+				rpc_user, work->job_id, xnonce2str, ntimestr, noncestr);
+		}
+		free(ntimestr);
+		free(noncestr);
+		free(xnonce2str);
+		free(nvotestr);
+
+		if (unlikely(!stratum_send_line(&stratum, s))) {
+			applog(LOG_ERR, "submit_upstream_work stratum_send_line failed");
+			goto out;
+		}
+	} else {
+
+		/* build hex string */
+
+                if (opt_algo != ALGO_HEAVY) {
+                    for (i = 0; i < ARRAY_SIZE(work->data); i++)
+			le32enc(work->data + i, work->data[i]);
+                }
+		str = bin2hex((unsigned char *)work->data, sizeof(work->data));
+		if (unlikely(!str)) {
+			applog(LOG_ERR, "submit_upstream_work OOM");
+			goto out;
+		}
+
+		/* build JSON-RPC request */
+		sprintf(s,
+			"{\"method\": \"getwork\", \"params\": [ \"%s\" ], \"id\":1}\r\n",
+			str);
+
+		/* issue JSON-RPC request */
+		val = json_rpc_call(curl, rpc_url, rpc_userpass, s, false, false, NULL);
+		if (unlikely(!val)) {
+			applog(LOG_ERR, "submit_upstream_work json_rpc_call failed");
+			goto out;
+		}
+
+		res = json_object_get(val, "result");
+		reason = json_object_get(val, "reject-reason");
+		share_result(json_is_true(res), reason ? json_string_value(reason) : NULL);
+
+		json_decref(val);
+	}
+
+	rc = true;
+
+out:
+	free(str);
+	return rc;
+}
+
+static const char *rpc_req =
+	"{\"method\": \"getwork\", \"params\": [], \"id\":0}\r\n";
+
+static bool get_upstream_work(CURL *curl, struct work *work)
+{
+	json_t *val;
+	bool rc;
+	struct timeval tv_start, tv_end, diff;
+
+	gettimeofday(&tv_start, NULL);
+	val = json_rpc_call(curl, rpc_url, rpc_userpass, rpc_req,
+			    want_longpoll, false, NULL);
+	gettimeofday(&tv_end, NULL);
+
+	if (have_stratum) {
+		if (val)
+			json_decref(val);
+		return true;
+	}
+
+	if (!val)
+		return false;
+
+	rc = work_decode(json_object_get(val, "result"), work);
+
+	if (opt_debug && rc) {
+		timeval_subtract(&diff, &tv_end, &tv_start);
+		applog(LOG_DEBUG, "DEBUG: got new work in %d ms",
+		       diff.tv_sec * 1000 + diff.tv_usec / 1000);
+	}
+
+	json_decref(val);
+
+	return rc;
+}
+
+static void workio_cmd_free(struct workio_cmd *wc)
+{
+	if (!wc)
+		return;
+
+	switch (wc->cmd) {
+	case WC_SUBMIT_WORK:
+		free(wc->u.work);
+		break;
+	default: /* do nothing */
+		break;
+	}
+
+	memset(wc, 0, sizeof(*wc));	/* poison */
+	free(wc);
+}
+
+static bool workio_get_work(struct workio_cmd *wc, CURL *curl)
+{
+	struct work *ret_work;
+	int failures = 0;
+
+	ret_work = (struct work*)calloc(1, sizeof(*ret_work));
+	if (!ret_work)
+		return false;
+
+	/* obtain new work from bitcoin via JSON-RPC */
+	while (!get_upstream_work(curl, ret_work)) {
+		if (unlikely((opt_retries >= 0) && (++failures > opt_retries))) {
+			applog(LOG_ERR, "json_rpc_call failed, terminating workio thread");
+			free(ret_work);
+			return false;
+		}
+
+		/* pause, then restart work-request loop */
+		applog(LOG_ERR, "json_rpc_call failed, retry after %d seconds",
+			opt_fail_pause);
+		sleep(opt_fail_pause);
+	}
+
+	/* send work to requesting thread */
+	if (!tq_push(wc->thr->q, ret_work))
+		free(ret_work);
+
+	return true;
+}
+
+static bool workio_submit_work(struct workio_cmd *wc, CURL *curl)
+{
+	int failures = 0;
+
+	/* submit solution to bitcoin via JSON-RPC */
+	while (!submit_upstream_work(curl, wc->u.work)) {
+		if (unlikely((opt_retries >= 0) && (++failures > opt_retries))) {
+			applog(LOG_ERR, "...terminating workio thread");
+			return false;
+		}
+
+		/* pause, then restart work-request loop */
+		applog(LOG_ERR, "...retry after %d seconds",
+			opt_fail_pause);
+		sleep(opt_fail_pause);
+	}
+
+	return true;
+}
+
+static void *workio_thread(void *userdata)
+{
+	struct thr_info *mythr = (struct thr_info*)userdata;
+	CURL *curl;
+	bool ok = true;
+
+	curl = curl_easy_init();
+	if (unlikely(!curl)) {
+		applog(LOG_ERR, "CURL initialization failed");
+		return NULL;
+	}
+
+	while (ok) {
+		struct workio_cmd *wc;
+
+		/* wait for workio_cmd sent to us, on our queue */
+		wc = (struct workio_cmd *)tq_pop(mythr->q, NULL);
+		if (!wc) {
+			ok = false;
+			break;
+		}
+
+		/* process workio_cmd */
+		switch (wc->cmd) {
+		case WC_GET_WORK:
+			ok = workio_get_work(wc, curl);
+			break;
+		case WC_SUBMIT_WORK:
+			ok = workio_submit_work(wc, curl);
+			break;
+
+		default:		/* should never happen */
+			ok = false;
+			break;
+		}
+
+		workio_cmd_free(wc);
+	}
+
+	tq_freeze(mythr->q);
+	curl_easy_cleanup(curl);
+
+	return NULL;
+}
+
+static bool get_work(struct thr_info *thr, struct work *work)
+{
+	struct workio_cmd *wc;
+	struct work *work_heap;
+
+	if (opt_benchmark) {
+		memset(work->data, 0x55, 76);
+		work->data[17] = swab32((uint32_t)time(NULL));
+		memset(work->data + 19, 0x00, 52);
+		work->data[20] = 0x80000000;
+		work->data[31] = 0x00000280;
+		memset(work->target, 0x00, sizeof(work->target));
+		return true;
+	}
+
+	/* fill out work request message */
+	wc = (struct workio_cmd *)calloc(1, sizeof(*wc));
+	if (!wc)
+		return false;
+
+	wc->cmd = WC_GET_WORK;
+	wc->thr = thr;
+
+	/* send work request to workio thread */
+	if (!tq_push(thr_info[work_thr_id].q, wc)) {
+		workio_cmd_free(wc);
+		return false;
+	}
+
+	/* wait for response, a unit of work */
+	work_heap = (struct work *)tq_pop(thr->q, NULL);
+	if (!work_heap)
+		return false;
+
+	/* copy returned work into storage provided by caller */
+	memcpy(work, work_heap, sizeof(*work));
+	free(work_heap);
+
+	return true;
+}
+
+static bool submit_work(struct thr_info *thr, const struct work *work_in)
+{
+	struct workio_cmd *wc;
+	/* fill out work request message */
+	wc = (struct workio_cmd *)calloc(1, sizeof(*wc));
+	if (!wc)
+		return false;
+
+	wc->u.work = (struct work *)malloc(sizeof(*work_in));
+	if (!wc->u.work)
+		goto err_out;
+
+	wc->cmd = WC_SUBMIT_WORK;
+	wc->thr = thr;
+	memcpy(wc->u.work, work_in, sizeof(*work_in));
+
+	/* send solution to workio thread */
+	if (!tq_push(thr_info[work_thr_id].q, wc))
+		goto err_out;
+
+	return true;
+
+err_out:
+	workio_cmd_free(wc);
+	return false;
+}
+
+static void stratum_gen_work(struct stratum_ctx *sctx, struct work *work)
+{
+	unsigned char merkle_root[64];
+	int i;
+
+	pthread_mutex_lock(&sctx->work_lock);
+
+	strcpy(work->job_id, sctx->job.job_id);
+	work->xnonce2_len = sctx->xnonce2_size;
+	memcpy(work->xnonce2, sctx->job.xnonce2, sctx->xnonce2_size);
+
+	/* Generate merkle root */
+	if (opt_algo == ALGO_HEAVY)
+		heavycoin_hash(merkle_root, sctx->job.coinbase, (int)sctx->job.coinbase_size);
+	else
+		sha256d(merkle_root, sctx->job.coinbase, (int)sctx->job.coinbase_size);
+	for (i = 0; i < sctx->job.merkle_count; i++) {
+		memcpy(merkle_root + 32, sctx->job.merkle[i], 32);
+		if (opt_algo == ALGO_HEAVY)
+			heavycoin_hash(merkle_root, merkle_root, 64);
+		else
+			sha256d(merkle_root, merkle_root, 64);
+	}
+	
+	/* Increment extranonce2 */
+	for (i = 0; i < (int)sctx->xnonce2_size && !++sctx->job.xnonce2[i]; i++);
+
+	/* Assemble block header */
+	memset(work->data, 0, 128);
+	work->data[0] = le32dec(sctx->job.version);
+	for (i = 0; i < 8; i++)
+		work->data[1 + i] = le32dec((uint32_t *)sctx->job.prevhash + i);
+	for (i = 0; i < 8; i++)
+		work->data[9 + i] = be32dec((uint32_t *)merkle_root + i);
+	work->data[17] = le32dec(sctx->job.ntime);
+	work->data[18] = le32dec(sctx->job.nbits);
+	work->data[20] = 0x80000000;
+	work->data[31] = 0x00000280;
+
+	// HeavyCoin
+	if (opt_algo == ALGO_HEAVY) {
+		uint16_t *ext;
+		work->maxvote = 1024;
+		ext = (uint16_t*)(&work->data[20]);
+		ext[0] = opt_vote;
+		ext[1] = be16dec(sctx->job.nreward);
+
+		for (i = 0; i < 20; i++)
+			work->data[i] = be32dec((uint32_t *)&work->data[i]);
+	}
+	//
+
+	pthread_mutex_unlock(&sctx->work_lock);
+
+	if (opt_debug) {
+		char *xnonce2str = bin2hex(work->xnonce2, sctx->xnonce2_size);
+		applog(LOG_DEBUG, "DEBUG: job_id='%s' extranonce2=%s ntime=%08x",
+		       work->job_id, xnonce2str, swab32(work->data[17]));
+		free(xnonce2str);
+	}
+
+	if (opt_algo == ALGO_SCRYPT)
+		diff_to_target(work->target, sctx->job.diff / 65536.0);
+	else
+		diff_to_target(work->target, sctx->job.diff);
+}
+
+static void *miner_thread(void *userdata)
+{
+	struct thr_info *mythr = (struct thr_info *)userdata;
+	int thr_id = mythr->id;
+	struct work work;
+	uint32_t max_nonce;
+	uint32_t end_nonce = 0xffffffffU / opt_n_threads * (thr_id + 1) - 0x20;
+	unsigned char *scratchbuf = NULL;
+	char s[16];
+	int i;
+
+	memset(&work, 0, sizeof(work)); // prevent work from being used uninitialized
+
+	/* Set worker threads to nice 19 and then preferentially to SCHED_IDLE
+	 * and if that fails, then SCHED_BATCH. No need for this to be an
+	 * error if it fails */
+	if (!opt_benchmark) {
+		setpriority(PRIO_PROCESS, 0, 19);
+		drop_policy();
+	}
+
+	/* Cpu affinity only makes sense if the number of threads is a multiple
+	 * of the number of CPUs */
+	if (num_processors > 1 && opt_n_threads % num_processors == 0) {
+		if (!opt_quiet)
+			applog(LOG_INFO, "Binding thread %d to cpu %d",
+			       thr_id, thr_id % num_processors);
+		affine_to_cpu(thr_id, thr_id % num_processors);
+	}
+
+	if (opt_algo == ALGO_SCRYPT)
+	{
+		scratchbuf = scrypt_buffer_alloc();
+	}
+
+	while (1) {
+		unsigned long hashes_done;
+		struct timeval tv_start, tv_end, diff;
+		int64_t max64;
+		int rc;
+
+		if (have_stratum) {
+			while (time(NULL) >= g_work_time + 120)
+				sleep(1);
+			pthread_mutex_lock(&g_work_lock);
+			if (work.data[19] >= end_nonce)
+				stratum_gen_work(&stratum, &g_work);
+		} else {
+			/* obtain new work from internal workio thread */
+			pthread_mutex_lock(&g_work_lock);
+			if (!have_stratum && (!have_longpoll ||
+					time(NULL) >= g_work_time + LP_SCANTIME*3/4 ||
+					work.data[19] >= end_nonce)) {
+				if (unlikely(!get_work(mythr, &g_work))) {
+					applog(LOG_ERR, "work retrieval failed, exiting "
+						"mining thread %d", mythr->id);
+					pthread_mutex_unlock(&g_work_lock);
+					goto out;
+				}
+				g_work_time = have_stratum ? 0 : time(NULL);
+			}
+			if (have_stratum) {
+				pthread_mutex_unlock(&g_work_lock);
+				continue;
+			}
+		}
+		if (memcmp(work.data, g_work.data, 76)) {
+			memcpy(&work, &g_work, sizeof(struct work));
+			work.data[19] = 0xffffffffU / opt_n_threads * thr_id;
+		} else
+			work.data[19]++;
+		pthread_mutex_unlock(&g_work_lock);
+		work_restart[thr_id].restart = 0;
+
+		/* adjust max_nonce to meet target scan time */
+		if (have_stratum)
+			max64 = LP_SCANTIME;
+		else
+			max64 = g_work_time + (have_longpoll ? LP_SCANTIME : opt_scantime)
+			      - time(NULL);
+		max64 *= (int64_t)thr_hashrates[thr_id];
+		if (max64 <= 0)
+			max64 = opt_algo == ALGO_SCRYPT ? 0xfffLL : 0x1fffffLL;
+		if (work.data[19] + max64 > end_nonce)
+			max_nonce = end_nonce;
+		else
+			max_nonce = (uint32_t)(work.data[19] + max64);
+
+		hashes_done = 0;
+		gettimeofday(&tv_start, NULL);
+
+		/* scan nonces for a proof-of-work hash */
+		switch (opt_algo) {
+		case ALGO_SCRYPT:
+			rc = scanhash_scrypt(thr_id, work.data, scratchbuf, work.target,
+			                     max_nonce, &hashes_done);
+			break;
+
+		case ALGO_SHA256D:
+			rc = scanhash_sha256d(thr_id, work.data, work.target,
+			                      max_nonce, &hashes_done);
+			break;
+
+		case ALGO_HEAVY:
+			rc = scanhash_heavy(thr_id, work.data, work.target,
+			                      max_nonce, &hashes_done, work.maxvote);
+			break;
+
+		case ALGO_FUGUE256:
+			rc = scanhash_fugue256(thr_id, work.data, work.target,
+			                      max_nonce, &hashes_done);
+			break;
+
+		default:
+			/* should never happen */
+			goto out;
+		}
+
+		/* record scanhash elapsed time */
+		gettimeofday(&tv_end, NULL);
+		timeval_subtract(&diff, &tv_end, &tv_start);
+		if (diff.tv_usec || diff.tv_sec) {
+			pthread_mutex_lock(&stats_lock);
+			thr_hashrates[thr_id] =
+				hashes_done / (diff.tv_sec + 1e-6 * diff.tv_usec);
+			pthread_mutex_unlock(&stats_lock);
+		}
+		if (!opt_quiet) {
+			sprintf(s, thr_hashrates[thr_id] >= 1e6 ? "%.0f" : "%.2f",
+				1e-3 * thr_hashrates[thr_id]);
+			applog(LOG_INFO, "thread %d: %lu hashes, %s khash/s",
+				thr_id, hashes_done, s);
+		}
+		if (opt_benchmark && thr_id == opt_n_threads - 1) {
+			double hashrate = 0.;
+			for (i = 0; i < opt_n_threads && thr_hashrates[i]; i++)
+				hashrate += thr_hashrates[i];
+			if (i == opt_n_threads) {
+				sprintf(s, hashrate >= 1e6 ? "%.0f" : "%.2f", 1e-3 * hashrate);
+				applog(LOG_INFO, "Total: %s khash/s", s);
+			}
+		}
+
+		/* if nonce found, submit work */
+		if (rc && !opt_benchmark && !submit_work(mythr, &work))
+			break;
+	}
+
+out:
+	tq_freeze(mythr->q);
+
+	return NULL;
+}
+
+static void restart_threads(void)
+{
+	int i;
+
+	for (i = 0; i < opt_n_threads; i++)
+		work_restart[i].restart = 1;
+}
+
+static void *longpoll_thread(void *userdata)
+{
+	struct thr_info *mythr = (struct thr_info *)userdata;
+	CURL *curl = NULL;
+	char *copy_start, *hdr_path = NULL, *lp_url = NULL;
+	bool need_slash = false;
+
+	curl = curl_easy_init();
+	if (unlikely(!curl)) {
+		applog(LOG_ERR, "CURL initialization failed");
+		goto out;
+	}
+
+start:
+	hdr_path = (char*)tq_pop(mythr->q, NULL);
+	if (!hdr_path)
+		goto out;
+
+	/* full URL */
+	if (strstr(hdr_path, "://")) {
+		lp_url = hdr_path;
+		hdr_path = NULL;
+	}
+	
+	/* absolute path, on current server */
+	else {
+		copy_start = (*hdr_path == '/') ? (hdr_path + 1) : hdr_path;
+		if (rpc_url[strlen(rpc_url) - 1] != '/')
+			need_slash = true;
+
+		lp_url = (char*)malloc(strlen(rpc_url) + strlen(copy_start) + 2);
+		if (!lp_url)
+			goto out;
+
+		sprintf(lp_url, "%s%s%s", rpc_url, need_slash ? "/" : "", copy_start);
+	}
+
+	applog(LOG_INFO, "Long-polling activated for %s", lp_url);
+
+	while (1) {
+		json_t *val, *soval;
+		int err;
+
+		val = json_rpc_call(curl, lp_url, rpc_userpass, rpc_req,
+				    false, true, &err);
+		if (have_stratum) {
+			if (val)
+				json_decref(val);
+			goto out;
+		}
+		if (likely(val)) {
+			applog(LOG_INFO, "LONGPOLL detected new block");
+			soval = json_object_get(json_object_get(val, "result"), "submitold");
+			submit_old = soval ? json_is_true(soval) : false;
+			pthread_mutex_lock(&g_work_lock);
+			if (work_decode(json_object_get(val, "result"), &g_work)) {
+				if (opt_debug)
+					applog(LOG_DEBUG, "DEBUG: got new work");
+				time(&g_work_time);
+				restart_threads();
+			}
+			pthread_mutex_unlock(&g_work_lock);
+			json_decref(val);
+		} else {
+			pthread_mutex_lock(&g_work_lock);
+			g_work_time -= LP_SCANTIME;
+			pthread_mutex_unlock(&g_work_lock);
+			if (err == CURLE_OPERATION_TIMEDOUT) {
+				restart_threads();
+			} else {
+				have_longpoll = false;
+				restart_threads();
+				free(hdr_path);
+				free(lp_url);
+				lp_url = NULL;
+				sleep(opt_fail_pause);
+				goto start;
+			}
+		}
+	}
+
+out:
+	free(hdr_path);
+	free(lp_url);
+	tq_freeze(mythr->q);
+	if (curl)
+		curl_easy_cleanup(curl);
+
+	return NULL;
+}
+
+static bool stratum_handle_response(char *buf)
+{
+	json_t *val, *err_val, *res_val, *id_val;
+	json_error_t err;
+	bool ret = false;
+
+	val = JSON_LOADS(buf, &err);
+	if (!val) {
+		applog(LOG_INFO, "JSON decode failed(%d): %s", err.line, err.text);
+		goto out;
+	}
+
+	res_val = json_object_get(val, "result");
+	err_val = json_object_get(val, "error");
+	id_val = json_object_get(val, "id");
+
+	if (!id_val || json_is_null(id_val) || !res_val)
+		goto out;
+
+	share_result(json_is_true(res_val),
+		err_val ? json_string_value(json_array_get(err_val, 1)) : NULL);
+
+	ret = true;
+out:
+	if (val)
+		json_decref(val);
+
+	return ret;
+}
+
+static void *stratum_thread(void *userdata)
+{
+	struct thr_info *mythr = (struct thr_info *)userdata;
+	char *s;
+
+	stratum.url = (char*)tq_pop(mythr->q, NULL);
+	if (!stratum.url)
+		goto out;
+	applog(LOG_INFO, "Starting Stratum on %s", stratum.url);
+
+	while (1) {
+		int failures = 0;
+
+		while (!stratum.curl) {
+			pthread_mutex_lock(&g_work_lock);
+			g_work_time = 0;
+			pthread_mutex_unlock(&g_work_lock);
+			restart_threads();
+
+			if (!stratum_connect(&stratum, stratum.url) ||
+			    !stratum_subscribe(&stratum) ||
+			    !stratum_authorize(&stratum, rpc_user, rpc_pass)) {
+				stratum_disconnect(&stratum);
+				if (opt_retries >= 0 && ++failures > opt_retries) {
+					applog(LOG_ERR, "...terminating workio thread");
+					tq_push(thr_info[work_thr_id].q, NULL);
+					goto out;
+				}
+				applog(LOG_ERR, "...retry after %d seconds", opt_fail_pause);
+				sleep(opt_fail_pause);
+			}
+		}
+
+		if (stratum.job.job_id &&
+		    (strcmp(stratum.job.job_id, g_work.job_id) || !g_work_time)) {
+			pthread_mutex_lock(&g_work_lock);
+			stratum_gen_work(&stratum, &g_work);
+			time(&g_work_time);
+			pthread_mutex_unlock(&g_work_lock);
+			if (stratum.job.clean) {
+				applog(LOG_INFO, "Stratum detected new block");
+				restart_threads();
+			}
+		}
+		
+		if (!stratum_socket_full(&stratum, 120)) {
+			applog(LOG_ERR, "Stratum connection timed out");
+			s = NULL;
+		} else
+			s = stratum_recv_line(&stratum);
+		if (!s) {
+			stratum_disconnect(&stratum);
+			applog(LOG_ERR, "Stratum connection interrupted");
+			continue;
+		}
+		if (!stratum_handle_method(&stratum, s))
+			stratum_handle_response(s);
+		free(s);
+	}
+
+out:
+	return NULL;
+}
+
+static void show_version_and_exit(void)
+{
+	printf("%s\n%s\n", PACKAGE_STRING, curl_version());
+	exit(0);
+}
+
+static void show_usage_and_exit(int status)
+{
+	if (status)
+		fprintf(stderr, "Try `" PROGRAM_NAME " --help' for more information.\n");
+	else
+		printf(usage);
+	exit(status);
+}
+
+static void parse_arg (int key, char *arg)
+{
+	char *p;
+	int v, i;
+
+	switch(key) {
+	case 'a':
+		for (i = 0; i < ARRAY_SIZE(algo_names); i++) {
+			if (algo_names[i] &&
+			    !strcmp(arg, algo_names[i])) {
+				opt_algo = (sha256_algos)i;
+				break;
+			}
+		}
+		if (i == ARRAY_SIZE(algo_names))
+			show_usage_and_exit(1);
+		break;
+	case 'B':
+		opt_background = true;
+		break;
+	case 'c': {
+		json_error_t err;
+		if (opt_config)
+			json_decref(opt_config);
+#if JANSSON_VERSION_HEX >= 0x020000
+		opt_config = json_load_file(arg, 0, &err);
+#else
+		opt_config = json_load_file(arg, &err);
+#endif
+		if (!json_is_object(opt_config)) {
+			applog(LOG_ERR, "JSON decode of %s failed", arg);
+			exit(1);
+		}
+		break;
+	}
+	case 'q':
+		opt_quiet = true;
+		break;
+	case 'D':
+		opt_debug = true;
+		break;
+	case 'p':
+		free(rpc_pass);
+		rpc_pass = strdup(arg);
+		break;
+	case 'P':
+		opt_protocol = true;
+		break;
+	case 'r':
+		v = atoi(arg);
+		if (v < -1 || v > 9999)	/* sanity check */
+			show_usage_and_exit(1);
+		opt_retries = v;
+		break;
+	case 'R':
+		v = atoi(arg);
+		if (v < 1 || v > 9999)	/* sanity check */
+			show_usage_and_exit(1);
+		opt_fail_pause = v;
+		break;
+	case 's':
+		v = atoi(arg);
+		if (v < 1 || v > 9999)	/* sanity check */
+			show_usage_and_exit(1);
+		opt_scantime = v;
+		break;
+	case 'T':
+		v = atoi(arg);
+		if (v < 1 || v > 99999)	/* sanity check */
+			show_usage_and_exit(1);
+		opt_timeout = v;
+		break;
+	case 't':
+		v = atoi(arg);
+		if (v < 1 || v > 9999)	/* sanity check */
+			show_usage_and_exit(1);
+		opt_n_threads = v;
+		break;
+	case 'v':
+		v = atoi(arg);
+		if (v < 0 || v > 1024)	/* sanity check */
+			show_usage_and_exit(1);
+		opt_vote = (uint16_t)v;
+		break;
+	case 'm':
+		opt_trust_pool = true;
+		break;
+	case 'u':
+		free(rpc_user);
+		rpc_user = strdup(arg);
+		break;
+	case 'o':			/* --url */
+		p = strstr(arg, "://");
+		if (p) {
+			if (strncasecmp(arg, "http://", 7) && strncasecmp(arg, "https://", 8) &&
+					strncasecmp(arg, "stratum+tcp://", 14))
+				show_usage_and_exit(1);
+			free(rpc_url);
+			rpc_url = strdup(arg);
+		} else {
+			if (!strlen(arg) || *arg == '/')
+				show_usage_and_exit(1);
+			free(rpc_url);
+			rpc_url = (char*)malloc(strlen(arg) + 8);
+			sprintf(rpc_url, "http://%s", arg);
+		}
+		p = strrchr(rpc_url, '@');
+		if (p) {
+			char *sp, *ap;
+			*p = '\0';
+			ap = strstr(rpc_url, "://") + 3;
+			sp = strchr(ap, ':');
+			if (sp) {
+				free(rpc_userpass);
+				rpc_userpass = strdup(ap);
+				free(rpc_user);
+				rpc_user = (char*)calloc(sp - ap + 1, 1);
+				strncpy(rpc_user, ap, sp - ap);
+				free(rpc_pass);
+				rpc_pass = strdup(sp + 1);
+			} else {
+				free(rpc_user);
+				rpc_user = strdup(ap);
+			}
+			memmove(ap, p + 1, strlen(p + 1) + 1);
+		}
+		have_stratum = !opt_benchmark && !strncasecmp(rpc_url, "stratum", 7);
+		break;
+	case 'O':			/* --userpass */
+		p = strchr(arg, ':');
+		if (!p)
+			show_usage_and_exit(1);
+		free(rpc_userpass);
+		rpc_userpass = strdup(arg);
+		free(rpc_user);
+		rpc_user = (char*)calloc(p - arg + 1, 1);
+		strncpy(rpc_user, arg, p - arg);
+		free(rpc_pass);
+		rpc_pass = strdup(p + 1);
+		break;
+	case 'x':			/* --proxy */
+		if (!strncasecmp(arg, "socks4://", 9))
+			opt_proxy_type = CURLPROXY_SOCKS4;
+		else if (!strncasecmp(arg, "socks5://", 9))
+			opt_proxy_type = CURLPROXY_SOCKS5;
+#if LIBCURL_VERSION_NUM >= 0x071200
+		else if (!strncasecmp(arg, "socks4a://", 10))
+			opt_proxy_type = CURLPROXY_SOCKS4A;
+		else if (!strncasecmp(arg, "socks5h://", 10))
+			opt_proxy_type = CURLPROXY_SOCKS5_HOSTNAME;
+#endif
+		else
+			opt_proxy_type = CURLPROXY_HTTP;
+		free(opt_proxy);
+		opt_proxy = strdup(arg);
+		break;
+	case 1001:
+		free(opt_cert);
+		opt_cert = strdup(arg);
+		break;
+	case 1005:
+		opt_benchmark = true;
+		want_longpoll = false;
+		want_stratum = false;
+		have_stratum = false;
+		break;
+	case 1003:
+		want_longpoll = false;
+		break;
+	case 1007:
+		want_stratum = false;
+		break;
+	case 'S':
+		use_syslog = true;
+		break;
+	case 'V':
+		show_version_and_exit();
+	case 'h':
+		show_usage_and_exit(0);
+	default:
+		show_usage_and_exit(1);
+	}
+}
+
+static void parse_config(void)
+{
+	int i;
+	json_t *val;
+
+	if (!json_is_object(opt_config))
+		return;
+
+	for (i = 0; i < ARRAY_SIZE(options); i++) {
+		if (!options[i].name)
+			break;
+		if (!strcmp(options[i].name, "config"))
+			continue;
+
+		val = json_object_get(opt_config, options[i].name);
+		if (!val)
+			continue;
+
+		if (options[i].has_arg && json_is_string(val)) {
+			char *s = strdup(json_string_value(val));
+			if (!s)
+				break;
+			parse_arg(options[i].val, s);
+			free(s);
+		} else if (!options[i].has_arg && json_is_true(val))
+			parse_arg(options[i].val, "");
+		else
+			applog(LOG_ERR, "JSON option %s invalid",
+				options[i].name);
+	}
+
+        if (opt_algo == ALGO_HEAVY && opt_vote == 9999) {
+            fprintf(stderr, "Heavycoin hash requires block reward vote parameter (see --vote)\n");
+            show_usage_and_exit(1);
+        }
+}
+
+static void parse_cmdline(int argc, char *argv[])
+{
+	int key;
+
+	while (1) {
+#if HAVE_GETOPT_LONG
+		key = getopt_long(argc, argv, short_options, options, NULL);
+#else
+		key = getopt(argc, argv, short_options);
+#endif
+		if (key < 0)
+			break;
+
+		parse_arg(key, optarg);
+	}
+	if (optind < argc) {
+		fprintf(stderr, "%s: unsupported non-option argument '%s'\n",
+			argv[0], argv[optind]);
+		show_usage_and_exit(1);
+	}
+
+        if (opt_algo == ALGO_HEAVY && opt_vote == 9999) {
+		fprintf(stderr, "%s: Heavycoin hash requires block reward vote parameter (see --vote)\n",
+			argv[0]);
+		show_usage_and_exit(1);
+        }
+
+	parse_config();
+}
+
+#ifndef WIN32
+static void signal_handler(int sig)
+{
+	switch (sig) {
+	case SIGHUP:
+		applog(LOG_INFO, "SIGHUP received");
+		break;
+	case SIGINT:
+		applog(LOG_INFO, "SIGINT received, exiting");
+		exit(0);
+		break;
+	case SIGTERM:
+		applog(LOG_INFO, "SIGTERM received, exiting");
+		exit(0);
+		break;
+	}
+}
+#endif
+
+#define PROGRAM_VERSION "0.1"
+int main(int argc, char *argv[])
+{
+	struct thr_info *thr;
+	long flags;
+	int i;
+
+#ifdef WIN32
+	SYSTEM_INFO sysinfo;
+#endif
+
+	printf("     *** ccMiner for nVidia GPUs by Christian Buchner and Christian H. ***\n");
+	printf("\t             This is version "PROGRAM_VERSION" (beta)\n");
+	printf("\t  based on pooler-cpuminer 2.3.2 (c) 2010 Jeff Garzik, 2012 pooler\n");
+	printf("\t  based on pooler-cpuminer extension for HVC from\n\t       https://github.com/heavycoin/cpuminer-heavycoin\n");
+	printf("\t\t\tand\n\t       http://hvc.1gh.com/\n");
+	printf("\tCuda additions Copyright 2014 Christian Buchner, Christian H.\n");
+	printf("\t  LTC donation address: LKS1WDKGED647msBQfLBHV3Ls8sveGncnm\n");
+	printf("\t  BTC donation address: 16hJF5mceSojnTD3ZTUDqdRhDyPJzoRakM\n");
+	printf("\t  YAC donation address: Y87sptDEcpLkLeAuex6qZioDbvy1qXZEj4\n");
+
+	rpc_user = strdup("");
+	rpc_pass = strdup("");
+
+	/* parse command line */
+	parse_cmdline(argc, argv);
+
+	if (!opt_benchmark && !rpc_url) {
+		fprintf(stderr, "%s: no URL supplied\n", argv[0]);
+		show_usage_and_exit(1);
+	}
+
+	if (!rpc_userpass) {
+		rpc_userpass = (char*)malloc(strlen(rpc_user) + strlen(rpc_pass) + 2);
+		if (!rpc_userpass)
+			return 1;
+		sprintf(rpc_userpass, "%s:%s", rpc_user, rpc_pass);
+	}
+
+	pthread_mutex_init(&applog_lock, NULL);
+	pthread_mutex_init(&stats_lock, NULL);
+	pthread_mutex_init(&g_work_lock, NULL);
+	pthread_mutex_init(&stratum.sock_lock, NULL);
+	pthread_mutex_init(&stratum.work_lock, NULL);
+
+	flags = !opt_benchmark && strncmp(rpc_url, "https:", 6)
+	      ? (CURL_GLOBAL_ALL & ~CURL_GLOBAL_SSL)
+	      : CURL_GLOBAL_ALL;
+	if (curl_global_init(flags)) {
+		applog(LOG_ERR, "CURL initialization failed");
+		return 1;
+	}
+
+#ifndef WIN32
+	if (opt_background) {
+		i = fork();
+		if (i < 0) exit(1);
+		if (i > 0) exit(0);
+		i = setsid();
+		if (i < 0)
+			applog(LOG_ERR, "setsid() failed (errno = %d)", errno);
+		i = chdir("/");
+		if (i < 0)
+			applog(LOG_ERR, "chdir() failed (errno = %d)", errno);
+		signal(SIGHUP, signal_handler);
+		signal(SIGINT, signal_handler);
+		signal(SIGTERM, signal_handler);
+	}
+#endif
+
+	num_processors = cuda_num_devices();
+	if (num_processors == 0)
+	{
+		applog(LOG_ERR, "No CUDA devices found! terminating.");
+		exit(1);
+	}
+	if (!opt_n_threads)
+		opt_n_threads = num_processors;
+
+#ifdef HAVE_SYSLOG_H
+	if (use_syslog)
+		openlog("cpuminer", LOG_PID, LOG_USER);
+#endif
+
+	work_restart = (struct work_restart *)calloc(opt_n_threads, sizeof(*work_restart));
+	if (!work_restart)
+		return 1;
+
+	thr_info = (struct thr_info *)calloc(opt_n_threads + 3, sizeof(*thr));
+	if (!thr_info)
+		return 1;
+	
+	thr_hashrates = (double *) calloc(opt_n_threads, sizeof(double));
+	if (!thr_hashrates)
+		return 1;
+
+	/* init workio thread info */
+	work_thr_id = opt_n_threads;
+	thr = &thr_info[work_thr_id];
+	thr->id = work_thr_id;
+	thr->q = tq_new();
+	if (!thr->q)
+		return 1;
+
+	/* start work I/O thread */
+	if (pthread_create(&thr->pth, NULL, workio_thread, thr)) {
+		applog(LOG_ERR, "workio thread create failed");
+		return 1;
+	}
+
+	if (want_longpoll && !have_stratum) {
+		/* init longpoll thread info */
+		longpoll_thr_id = opt_n_threads + 1;
+		thr = &thr_info[longpoll_thr_id];
+		thr->id = longpoll_thr_id;
+		thr->q = tq_new();
+		if (!thr->q)
+			return 1;
+
+		/* start longpoll thread */
+		if (unlikely(pthread_create(&thr->pth, NULL, longpoll_thread, thr))) {
+			applog(LOG_ERR, "longpoll thread create failed");
+			return 1;
+		}
+	}
+	if (want_stratum) {
+		/* init stratum thread info */
+		stratum_thr_id = opt_n_threads + 2;
+		thr = &thr_info[stratum_thr_id];
+		thr->id = stratum_thr_id;
+		thr->q = tq_new();
+		if (!thr->q)
+			return 1;
+
+		/* start stratum thread */
+		if (unlikely(pthread_create(&thr->pth, NULL, stratum_thread, thr))) {
+			applog(LOG_ERR, "stratum thread create failed");
+			return 1;
+		}
+
+		if (have_stratum)
+			tq_push(thr_info[stratum_thr_id].q, strdup(rpc_url));
+	}
+
+	/* start mining threads */
+	for (i = 0; i < opt_n_threads; i++) {
+		thr = &thr_info[i];
+
+		thr->id = i;
+		thr->q = tq_new();
+		if (!thr->q)
+			return 1;
+
+		if (unlikely(pthread_create(&thr->pth, NULL, miner_thread, thr))) {
+			applog(LOG_ERR, "thread %d create failed", i);
+			return 1;
+		}
+	}
+
+	applog(LOG_INFO, "%d miner threads started, "
+		"using '%s' algorithm.",
+		opt_n_threads,
+		algo_names[opt_algo]);
+
+	/* main loop - simply wait for workio thread to exit */
+	pthread_join(thr_info[work_thr_id].pth, NULL);
+
+	applog(LOG_INFO, "workio thread dead, exiting.");
+
+	return 0;
+}
diff --git a/cpuminer-config.h b/cpuminer-config.h
new file mode 100644
index 0000000000..bdd09fe4fb
--- /dev/null
+++ b/cpuminer-config.h
@@ -0,0 +1,190 @@
+/* cpuminer-config.h.in.  Generated from configure.ac by autoheader.  */
+
+/* Define to one of `_getb67', `GETB67', `getb67' for Cray-2 and Cray-YMP
+   systems. This function is required for `alloca.c' support on those systems.
+   */
+#undef CRAY_STACKSEG_END
+
+/* Define to 1 if using `alloca.c'. */
+#undef C_ALLOCA
+
+/* Define to 1 if you have `alloca', as a function or macro. */
+#undef HAVE_ALLOCA
+
+/* Define to 1 if you have <alloca.h> and it should be used (not on Ultrix).
+   */
+#undef HAVE_ALLOCA_H
+
+/* Define to 1 if you have the declaration of `be32dec', and to 0 if you
+   don't. */
+#undef HAVE_DECL_BE32DEC
+
+/* Define to 1 if you have the declaration of `be32enc', and to 0 if you
+   don't. */
+#undef HAVE_DECL_BE32ENC
+
+/* Define to 1 if you have the declaration of `le32dec', and to 0 if you
+   don't. */
+#undef HAVE_DECL_LE32DEC
+
+/* Define to 1 if you have the declaration of `le32enc', and to 0 if you
+   don't. */
+#undef HAVE_DECL_LE32ENC
+
+/* Define to 1 if you have the `getopt_long' function. */
+#define HAVE_GETOPT_LONG 1
+
+/* Define to 1 if you have the <inttypes.h> header file. */
+#undef HAVE_INTTYPES_H
+
+/* Define to 1 if you have a functional curl library. */
+#undef HAVE_LIBCURL
+
+/* Define to 1 if you have the <memory.h> header file. */
+#undef HAVE_MEMORY_H
+
+/* Define to 1 if you have the <stdint.h> header file. */
+#undef HAVE_STDINT_H
+
+/* Define to 1 if you have the <stdlib.h> header file. */
+#undef HAVE_STDLIB_H
+
+/* Define to 1 if you have the <strings.h> header file. */
+#undef HAVE_STRINGS_H
+
+/* Define to 1 if you have the <string.h> header file. */
+#undef HAVE_STRING_H
+
+/* Define to 1 if you have the <syslog.h> header file. */
+#undef HAVE_SYSLOG_H
+
+/* Define to 1 if you have the <sys/endian.h> header file. */
+#undef HAVE_SYS_ENDIAN_H
+
+/* Define to 1 if you have the <sys/param.h> header file. */
+#undef HAVE_SYS_PARAM_H
+
+/* Define to 1 if you have the <sys/stat.h> header file. */
+#undef HAVE_SYS_STAT_H
+
+/* Define to 1 if you have the <sys/sysctl.h> header file. */
+#undef HAVE_SYS_SYSCTL_H
+
+/* Define to 1 if you have the <sys/types.h> header file. */
+#undef HAVE_SYS_TYPES_H
+
+/* Define to 1 if you have the <unistd.h> header file. */
+#undef HAVE_UNISTD_H
+
+/* Defined if libcurl supports AsynchDNS */
+#undef LIBCURL_FEATURE_ASYNCHDNS
+
+/* Defined if libcurl supports IDN */
+#undef LIBCURL_FEATURE_IDN
+
+/* Defined if libcurl supports IPv6 */
+#undef LIBCURL_FEATURE_IPV6
+
+/* Defined if libcurl supports KRB4 */
+#undef LIBCURL_FEATURE_KRB4
+
+/* Defined if libcurl supports libz */
+#undef LIBCURL_FEATURE_LIBZ
+
+/* Defined if libcurl supports NTLM */
+#undef LIBCURL_FEATURE_NTLM
+
+/* Defined if libcurl supports SSL */
+#undef LIBCURL_FEATURE_SSL
+
+/* Defined if libcurl supports SSPI */
+#undef LIBCURL_FEATURE_SSPI
+
+/* Defined if libcurl supports DICT */
+#undef LIBCURL_PROTOCOL_DICT
+
+/* Defined if libcurl supports FILE */
+#undef LIBCURL_PROTOCOL_FILE
+
+/* Defined if libcurl supports FTP */
+#undef LIBCURL_PROTOCOL_FTP
+
+/* Defined if libcurl supports FTPS */
+#undef LIBCURL_PROTOCOL_FTPS
+
+/* Defined if libcurl supports HTTP */
+#undef LIBCURL_PROTOCOL_HTTP
+
+/* Defined if libcurl supports HTTPS */
+#undef LIBCURL_PROTOCOL_HTTPS
+
+/* Defined if libcurl supports IMAP */
+#undef LIBCURL_PROTOCOL_IMAP
+
+/* Defined if libcurl supports LDAP */
+#undef LIBCURL_PROTOCOL_LDAP
+
+/* Defined if libcurl supports POP3 */
+#undef LIBCURL_PROTOCOL_POP3
+
+/* Defined if libcurl supports RTSP */
+#undef LIBCURL_PROTOCOL_RTSP
+
+/* Defined if libcurl supports SMTP */
+#undef LIBCURL_PROTOCOL_SMTP
+
+/* Defined if libcurl supports TELNET */
+#undef LIBCURL_PROTOCOL_TELNET
+
+/* Defined if libcurl supports TFTP */
+#undef LIBCURL_PROTOCOL_TFTP
+
+/* Define to 1 if your C compiler doesn't accept -c and -o together. */
+#undef NO_MINUS_C_MINUS_O
+
+/* Name of package */
+#undef PACKAGE
+
+/* Define to the address where bug reports for this package should be sent. */
+#undef PACKAGE_BUGREPORT
+
+/* Define to the full name of this package. */
+#define PACKAGE_NAME "ccminer"
+
+/* Define to the full name and version of this package. */
+#define PACKAGE_STRING "ccminer 2014.03.18"
+
+/* Define to the one symbol short name of this package. */
+#undef PACKAGE_TARNAME
+
+/* Define to the home page for this package. */
+#undef PACKAGE_URL
+
+/* Define to the version of this package. */
+#define PACKAGE_VERSION "2014.03.18"
+
+/* If using the C implementation of alloca, define if you know the
+   direction of stack growth for your system; otherwise it will be
+   automatically deduced at runtime.
+	STACK_DIRECTION > 0 => grows toward higher addresses
+	STACK_DIRECTION < 0 => grows toward lower addresses
+	STACK_DIRECTION = 0 => direction of growth unknown */
+#undef STACK_DIRECTION
+
+/* Define to 1 if you have the ANSI C header files. */
+#undef STDC_HEADERS
+
+/* Define to 1 if AVX assembly is available. */
+#undef USE_AVX
+
+/* Define to 1 if XOP assembly is available. */
+#undef USE_XOP
+
+/* Version number of package */
+#undef VERSION
+
+/* Define curl_free() as free() if our version of curl lacks curl_free. */
+#undef curl_free
+
+/* Define to `unsigned int' if <sys/types.h> does not define. */
+#undef size_t
diff --git a/cpuminer-config.h.in b/cpuminer-config.h.in
new file mode 100644
index 0000000000..b8668be60d
--- /dev/null
+++ b/cpuminer-config.h.in
@@ -0,0 +1,199 @@
+/* cpuminer-config.h.in.  Generated from configure.ac by autoheader.  */
+
+/* Define to one of `_getb67', `GETB67', `getb67' for Cray-2 and Cray-YMP
+   systems. This function is required for `alloca.c' support on those systems.
+   */
+#undef CRAY_STACKSEG_END
+
+/* Define to 1 if using `alloca.c'. */
+#undef C_ALLOCA
+
+/* Define to 1 if you have `alloca', as a function or macro. */
+#undef HAVE_ALLOCA
+
+/* Define to 1 if you have <alloca.h> and it should be used (not on Ultrix).
+   */
+#undef HAVE_ALLOCA_H
+
+/* Define to 1 if you have the declaration of `be32dec', and to 0 if you
+   don't. */
+#undef HAVE_DECL_BE32DEC
+
+/* Define to 1 if you have the declaration of `be32enc', and to 0 if you
+   don't. */
+#undef HAVE_DECL_BE32ENC
+
+/* Define to 1 if you have the declaration of `le32dec', and to 0 if you
+   don't. */
+#undef HAVE_DECL_LE32DEC
+
+/* Define to 1 if you have the declaration of `le32enc', and to 0 if you
+   don't. */
+#undef HAVE_DECL_LE32ENC
+
+/* Define to 1 if you have the `getopt_long' function. */
+#undef HAVE_GETOPT_LONG
+
+/* Define to 1 if you have the <inttypes.h> header file. */
+#undef HAVE_INTTYPES_H
+
+/* Define to 1 if you have the `crypto' library (-lcrypto). */
+#undef HAVE_LIBCRYPTO
+
+/* Define to 1 if you have a functional curl library. */
+#undef HAVE_LIBCURL
+
+/* Define to 1 if you have the `ssl' library (-lssl). */
+#undef HAVE_LIBSSL
+
+/* Define to 1 if you have the <memory.h> header file. */
+#undef HAVE_MEMORY_H
+
+/* Define to 1 if you have the <stdint.h> header file. */
+#undef HAVE_STDINT_H
+
+/* Define to 1 if you have the <stdlib.h> header file. */
+#undef HAVE_STDLIB_H
+
+/* Define to 1 if you have the <strings.h> header file. */
+#undef HAVE_STRINGS_H
+
+/* Define to 1 if you have the <string.h> header file. */
+#undef HAVE_STRING_H
+
+/* Define to 1 if you have the <syslog.h> header file. */
+#undef HAVE_SYSLOG_H
+
+/* Define to 1 if you have the <sys/endian.h> header file. */
+#undef HAVE_SYS_ENDIAN_H
+
+/* Define to 1 if you have the <sys/param.h> header file. */
+#undef HAVE_SYS_PARAM_H
+
+/* Define to 1 if you have the <sys/stat.h> header file. */
+#undef HAVE_SYS_STAT_H
+
+/* Define to 1 if you have the <sys/sysctl.h> header file. */
+#undef HAVE_SYS_SYSCTL_H
+
+/* Define to 1 if you have the <sys/types.h> header file. */
+#undef HAVE_SYS_TYPES_H
+
+/* Define to 1 if you have the <unistd.h> header file. */
+#undef HAVE_UNISTD_H
+
+/* Defined if libcurl supports AsynchDNS */
+#undef LIBCURL_FEATURE_ASYNCHDNS
+
+/* Defined if libcurl supports IDN */
+#undef LIBCURL_FEATURE_IDN
+
+/* Defined if libcurl supports IPv6 */
+#undef LIBCURL_FEATURE_IPV6
+
+/* Defined if libcurl supports KRB4 */
+#undef LIBCURL_FEATURE_KRB4
+
+/* Defined if libcurl supports libz */
+#undef LIBCURL_FEATURE_LIBZ
+
+/* Defined if libcurl supports NTLM */
+#undef LIBCURL_FEATURE_NTLM
+
+/* Defined if libcurl supports SSL */
+#undef LIBCURL_FEATURE_SSL
+
+/* Defined if libcurl supports SSPI */
+#undef LIBCURL_FEATURE_SSPI
+
+/* Defined if libcurl supports DICT */
+#undef LIBCURL_PROTOCOL_DICT
+
+/* Defined if libcurl supports FILE */
+#undef LIBCURL_PROTOCOL_FILE
+
+/* Defined if libcurl supports FTP */
+#undef LIBCURL_PROTOCOL_FTP
+
+/* Defined if libcurl supports FTPS */
+#undef LIBCURL_PROTOCOL_FTPS
+
+/* Defined if libcurl supports HTTP */
+#undef LIBCURL_PROTOCOL_HTTP
+
+/* Defined if libcurl supports HTTPS */
+#undef LIBCURL_PROTOCOL_HTTPS
+
+/* Defined if libcurl supports IMAP */
+#undef LIBCURL_PROTOCOL_IMAP
+
+/* Defined if libcurl supports LDAP */
+#undef LIBCURL_PROTOCOL_LDAP
+
+/* Defined if libcurl supports POP3 */
+#undef LIBCURL_PROTOCOL_POP3
+
+/* Defined if libcurl supports RTSP */
+#undef LIBCURL_PROTOCOL_RTSP
+
+/* Defined if libcurl supports SMTP */
+#undef LIBCURL_PROTOCOL_SMTP
+
+/* Defined if libcurl supports TELNET */
+#undef LIBCURL_PROTOCOL_TELNET
+
+/* Defined if libcurl supports TFTP */
+#undef LIBCURL_PROTOCOL_TFTP
+
+/* Define to 1 if your C compiler doesn't accept -c and -o together. */
+#undef NO_MINUS_C_MINUS_O
+
+/* Name of package */
+#undef PACKAGE
+
+/* Define to the address where bug reports for this package should be sent. */
+#undef PACKAGE_BUGREPORT
+
+/* Define to the full name of this package. */
+#undef PACKAGE_NAME
+
+/* Define to the full name and version of this package. */
+#undef PACKAGE_STRING
+
+/* Define to the one symbol short name of this package. */
+#undef PACKAGE_TARNAME
+
+/* Define to the home page for this package. */
+#undef PACKAGE_URL
+
+/* Define to the version of this package. */
+#undef PACKAGE_VERSION
+
+/* If using the C implementation of alloca, define if you know the
+   direction of stack growth for your system; otherwise it will be
+   automatically deduced at runtime.
+	STACK_DIRECTION > 0 => grows toward higher addresses
+	STACK_DIRECTION < 0 => grows toward lower addresses
+	STACK_DIRECTION = 0 => direction of growth unknown */
+#undef STACK_DIRECTION
+
+/* Define to 1 if you have the ANSI C header files. */
+#undef STDC_HEADERS
+
+/* Define to 1 if AVX assembly is available. */
+#undef USE_AVX
+
+/* Define to 1 if AVX2 assembly is available. */
+#undef USE_AVX2
+
+/* Define to 1 if XOP assembly is available. */
+#undef USE_XOP
+
+/* Version number of package */
+#undef VERSION
+
+/* Define curl_free() as free() if our version of curl lacks curl_free. */
+#undef curl_free
+
+/* Define to `unsigned int' if <sys/types.h> does not define. */
+#undef size_t
diff --git a/cuda_blake512.cu b/cuda_blake512.cu
new file mode 100644
index 0000000000..325901d11c
--- /dev/null
+++ b/cuda_blake512.cu
@@ -0,0 +1,308 @@
+/* Diese Funktion ist auf 84+32-Byte große Eingabedaten ausgerichtet (Heavycoin) */
+#include <cuda.h>
+#include "cuda_runtime.h"
+#include "device_launch_parameters.h"
+
+#include <stdio.h>
+#include <memory.h>
+
+// Folgende Definitionen später durch header ersetzen
+typedef unsigned char uint8_t;
+typedef unsigned int uint32_t;
+typedef unsigned long long uint64_t;
+
+// globaler Speicher für alle HeftyHashes aller Threads
+extern uint32_t *d_heftyHashes[8];
+extern uint32_t *d_nonceVector[8];
+
+// globaler Speicher für unsere Ergebnisse
+uint32_t *d_hash5output[8];
+
+// die Message (116 Bytes) mit Padding zur Berechnung auf der GPU
+__constant__ uint64_t c_PaddedMessage[16]; // padded message (84+32 bytes + padding)
+
+// ---------------------------- BEGIN CUDA blake512 functions ------------------------------------
+
+__constant__ uint8_t c_sigma[16][16];
+
+const uint8_t host_sigma[16][16] =
+{
+  { 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15 },
+  {14, 10, 4, 8, 9, 15, 13, 6, 1, 12, 0, 2, 11, 7, 5, 3 },
+  {11, 8, 12, 0, 5, 2, 15, 13, 10, 14, 3, 6, 7, 1, 9, 4 },
+  { 7, 9, 3, 1, 13, 12, 11, 14, 2, 6, 5, 10, 4, 0, 15, 8 },
+  { 9, 0, 5, 7, 2, 4, 10, 15, 14, 1, 11, 12, 6, 8, 3, 13 },
+  { 2, 12, 6, 10, 0, 11, 8, 3, 4, 13, 7, 5, 15, 14, 1, 9 },
+  {12, 5, 1, 15, 14, 13, 4, 10, 0, 7, 6, 3, 9, 2, 8, 11 },
+  {13, 11, 7, 14, 12, 1, 3, 9, 5, 0, 15, 4, 8, 6, 2, 10 },
+  { 6, 15, 14, 9, 11, 3, 0, 8, 12, 2, 13, 7, 1, 4, 10, 5 },
+  {10, 2, 8, 4, 7, 6, 1, 5, 15, 11, 9, 14, 3, 12, 13 , 0 },
+  { 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15 },
+  {14, 10, 4, 8, 9, 15, 13, 6, 1, 12, 0, 2, 11, 7, 5, 3 },
+  {11, 8, 12, 0, 5, 2, 15, 13, 10, 14, 3, 6, 7, 1, 9, 4 },
+  { 7, 9, 3, 1, 13, 12, 11, 14, 2, 6, 5, 10, 4, 0, 15, 8 },
+  { 9, 0, 5, 7, 2, 4, 10, 15, 14, 1, 11, 12, 6, 8, 3, 13 },
+  { 2, 12, 6, 10, 0, 11, 8, 3, 4, 13, 7, 5, 15, 14, 1, 9 }
+};
+
+#define SWAP32(x) \
+    ((((x) << 24) & 0xff000000u) | (((x) << 8) & 0x00ff0000u)   | \
+      (((x) >> 8) & 0x0000ff00u) | (((x) >> 24) & 0x000000ffu))
+
+#define SWAP64(x) \
+    ((uint64_t)((((uint64_t)(x) & 0xff00000000000000ULL) >> 56) | \
+                (((uint64_t)(x) & 0x00ff000000000000ULL) >> 40) | \
+                (((uint64_t)(x) & 0x0000ff0000000000ULL) >> 24) | \
+                (((uint64_t)(x) & 0x000000ff00000000ULL) >>  8) | \
+                (((uint64_t)(x) & 0x00000000ff000000ULL) <<  8) | \
+                (((uint64_t)(x) & 0x0000000000ff0000ULL) << 24) | \
+                (((uint64_t)(x) & 0x000000000000ff00ULL) << 40) | \
+                (((uint64_t)(x) & 0x00000000000000ffULL) << 56)))
+
+__constant__ uint64_t c_SecondRound[16];
+
+const uint64_t host_SecondRound[16] =
+{
+  0,0,0,0,0,0,0,0,0,0,0,0,0,SWAP64(1),0,SWAP64(0x3A0)
+};
+
+__constant__ uint64_t c_u512[16];
+
+const uint64_t host_u512[16] =
+{
+  0x243f6a8885a308d3ULL, 0x13198a2e03707344ULL, 
+  0xa4093822299f31d0ULL, 0x082efa98ec4e6c89ULL,
+  0x452821e638d01377ULL, 0xbe5466cf34e90c6cULL, 
+  0xc0ac29b7c97c50ddULL, 0x3f84d5b5b5470917ULL,
+  0x9216d5d98979fb1bULL, 0xd1310ba698dfb5acULL, 
+  0x2ffd72dbd01adfb7ULL, 0xb8e1afed6a267e96ULL,
+  0xba7c9045f12c7f99ULL, 0x24a19947b3916cf7ULL, 
+  0x0801f2e2858efc16ULL, 0x636920d871574e69ULL
+};
+
+
+#define ROTR(x,n) (((x)<<(64-n))|( (x)>>(n)))
+
+#define G(a,b,c,d,e)          \
+    v[a] += (m[sigma[i][e]] ^ u512[sigma[i][e+1]]) + v[b];\
+    v[d] = ROTR( v[d] ^ v[a],32);        \
+    v[c] += v[d];           \
+    v[b] = ROTR( v[b] ^ v[c],25);        \
+    v[a] += (m[sigma[i][e+1]] ^ u512[sigma[i][e]])+v[b];  \
+    v[d] = ROTR( v[d] ^ v[a],16);        \
+    v[c] += v[d];           \
+    v[b] = ROTR( v[b] ^ v[c],11);
+
+__device__ void blake512_compress( uint64_t *h, const uint64_t *block, int nullt, const uint8_t ((*sigma)[16]), const uint64_t *u512 )
+{
+    uint64_t v[16], m[16], i;
+
+#pragma unroll 16
+    for( i = 0; i < 16; ++i )  m[i] = SWAP64(block[i]);
+
+#pragma unroll 8
+    for( i = 0; i < 8; ++i )  v[i] = h[i];
+
+    v[ 8] = u512[0];
+    v[ 9] = u512[1];
+    v[10] = u512[2];
+    v[11] = u512[3];
+    v[12] = u512[4];
+    v[13] = u512[5];
+    v[14] = u512[6];
+    v[15] = u512[7];
+
+    /* don't xor t when the block is only padding */
+    if ( !nullt ) {
+        v[12] ^= 928;
+        v[13] ^= 928;
+    }
+
+#pragma unroll 16
+    for( i = 0; i < 16; ++i )
+    {
+        /* column step */
+        G( 0, 4, 8, 12, 0 );
+        G( 1, 5, 9, 13, 2 );
+        G( 2, 6, 10, 14, 4 );
+        G( 3, 7, 11, 15, 6 );
+        /* diagonal step */
+        G( 0, 5, 10, 15, 8 );
+        G( 1, 6, 11, 12, 10 );
+        G( 2, 7, 8, 13, 12 );
+        G( 3, 4, 9, 14, 14 );
+    }
+
+#pragma unroll 16
+    for( i = 0; i < 16; ++i )  h[i % 8] ^= v[i];
+}
+
+// Endian Drehung für 32 Bit Typen
+static __device__ uint32_t cuda_swab32(uint32_t x)
+{
+    return (((x << 24) & 0xff000000u) | ((x << 8) & 0x00ff0000u)
+          | ((x >> 8) & 0x0000ff00u) | ((x >> 24) & 0x000000ffu));
+}
+
+// Endian Drehung für 64 Bit Typen
+static __device__ uint64_t cuda_swab64(uint64_t x) {
+    uint32_t h = (x >> 32);
+    uint32_t l = (x & 0xFFFFFFFFULL);
+    return (((uint64_t)cuda_swab32(l)) << 32) | ((uint64_t)cuda_swab32(h));
+}
+
+// das Hi Word aus einem 64 Bit Typen extrahieren
+static __device__ uint32_t HIWORD(const uint64_t &x) {
+#if __CUDA_ARCH__ >= 130
+	return (uint32_t)__double2hiint(__longlong_as_double(x));
+#else
+	return (uint32_t)(x >> 32);
+#endif
+}
+
+// das Hi Word in einem 64 Bit Typen ersetzen
+static __device__ uint64_t REPLACE_HIWORD(const uint64_t &x, const uint32_t &y) {
+	return (x & 0xFFFFFFFFULL) | (((uint64_t)y) << 32ULL);
+}
+
+// das Lo Word aus einem 64 Bit Typen extrahieren
+static __device__ uint32_t LOWORD(const uint64_t &x) {
+#if __CUDA_ARCH__ >= 130
+	return (uint32_t)__double2loint(__longlong_as_double(x));
+#else
+	return (uint32_t)(x & 0xFFFFFFFFULL);
+#endif
+}
+
+// das Lo Word in einem 64 Bit Typen ersetzen
+static __device__ uint64_t REPLACE_LOWORD(const uint64_t &x, const uint32_t &y) {
+	return (x & 0xFFFFFFFF00000000ULL) | ((uint64_t)y);
+}
+
+__global__ void blake512_gpu_hash(int threads, uint32_t startNounce, void *outputHash, uint32_t *heftyHashes, uint32_t *nonceVector)
+{
+	int thread = (blockDim.x * blockIdx.x + threadIdx.x);
+	if (thread < threads)
+	{
+		// bestimme den aktuellen Zähler
+		//uint32_t nounce = startNounce + thread;
+		uint32_t nounce = nonceVector[thread];
+
+		// Index-Position des Hashes in den Hash Puffern bestimmen (Hefty1 und outputHash)
+		uint32_t hashPosition = nounce - startNounce;
+
+		// State vorbereiten
+		uint64_t h[8];
+		h[0] = 0x6a09e667f3bcc908ULL;
+		h[1] = 0xbb67ae8584caa73bULL;
+		h[2] = 0x3c6ef372fe94f82bULL;
+		h[3] = 0xa54ff53a5f1d36f1ULL;
+		h[4] = 0x510e527fade682d1ULL;
+		h[5] = 0x9b05688c2b3e6c1fULL;
+		h[6] = 0x1f83d9abfb41bd6bULL;
+		h[7] = 0x5be0cd19137e2179ULL;
+
+		// 128 Byte für die Message
+		uint64_t buf[16];
+
+		// Message für die erste Runde in Register holen
+#pragma unroll 16
+		for (int i=0; i < 16; ++i) buf[i] = c_PaddedMessage[i];
+
+		// die Nounce durch die thread-spezifische ersetzen
+		buf[9] = REPLACE_HIWORD(buf[9], nounce);
+
+		// den thread-spezifischen Hefty1 hash einsetzen
+		uint32_t *hefty = heftyHashes + 8 * hashPosition;
+		buf[10] = REPLACE_HIWORD(buf[10], hefty[0]);
+		buf[11] = REPLACE_LOWORD(buf[11], hefty[1]);
+		buf[11] = REPLACE_HIWORD(buf[11], hefty[2]);
+		buf[12] = REPLACE_LOWORD(buf[12], hefty[3]);
+		buf[12] = REPLACE_HIWORD(buf[12], hefty[4]);
+		buf[13] = REPLACE_LOWORD(buf[13], hefty[5]);
+		buf[13] = REPLACE_HIWORD(buf[13], hefty[6]);
+		buf[14] = REPLACE_LOWORD(buf[14], hefty[7]);
+
+		// erste Runde
+		blake512_compress( h, buf, 0, c_sigma, c_u512 );
+
+		// zweite Runde
+#pragma unroll 16
+		for (int i=0; i < 16; ++i) buf[i] = c_SecondRound[i];
+		blake512_compress( h, buf, 1, c_sigma, c_u512 );
+
+		// Hash rauslassen
+#if 0
+		// ausschliesslich 32 bit Operationen sofern die SM1.3 double intrinsics verfügbar sind
+		uint32_t *outHash = (uint32_t *)outputHash + 16 * hashPosition;
+#pragma unroll 8
+		for (int i=0; i < 8; ++i) {
+			outHash[2*i+0] = cuda_swab32( HIWORD(h[i]) );
+			outHash[2*i+1] = cuda_swab32( LOWORD(h[i]) );
+		}
+#else
+		// in dieser Version passieren auch ein paar 64 Bit Shifts
+		uint64_t *outHash = (uint64_t *)outputHash + 8 * hashPosition;
+#pragma unroll 8
+		for (int i=0; i < 8; ++i) outHash[i] = cuda_swab64( h[i] );
+#endif
+	}
+}
+
+
+// ---------------------------- END CUDA blake512 functions ------------------------------------
+
+// Setup-Funktionen
+__host__ void blake512_cpu_init(int thr_id, int threads)
+{
+	// Kopiere die Hash-Tabellen in den GPU-Speicher
+	cudaMemcpyToSymbol( c_sigma,
+						host_sigma,
+						sizeof(host_sigma),
+						0, cudaMemcpyHostToDevice);
+
+	cudaMemcpyToSymbol( c_u512,
+						host_u512,
+						sizeof(host_u512),
+						0, cudaMemcpyHostToDevice);
+
+	cudaMemcpyToSymbol( c_SecondRound,
+						host_SecondRound,
+						sizeof(host_SecondRound),
+						0, cudaMemcpyHostToDevice);
+
+	// Speicher für alle Ergebnisse belegen
+	cudaMalloc(&d_hash5output[thr_id], 16 * sizeof(uint32_t) * threads);
+}
+
+__host__ void blake512_cpu_setBlock(void *pdata)
+	// data muss 84-Byte haben!
+	// heftyHash hat 32-Byte
+{
+	// Message mit Padding für erste Runde bereitstellen
+	unsigned char PaddedMessage[128];
+	memcpy(PaddedMessage, pdata, 84);
+	memset(PaddedMessage+84, 0, 32); // leeres Hefty Hash einfüllen
+	memset(PaddedMessage+116, 0, 12);
+	PaddedMessage[116] = 0x80;
+
+	// die Message (116 Bytes) ohne Padding zur Berechnung auf der GPU
+	cudaMemcpyToSymbol( c_PaddedMessage, PaddedMessage, 16*sizeof(uint64_t), 0, cudaMemcpyHostToDevice);
+}
+
+
+__host__ void blake512_cpu_hash(int thr_id, int threads, uint32_t startNounce)
+{
+	const int threadsperblock = 128;
+
+	// berechne wie viele Thread Blocks wir brauchen
+	dim3 grid((threads + threadsperblock-1)/threadsperblock);
+	dim3 block(threadsperblock);
+
+	// Größe des dynamischen Shared Memory Bereichs (abhängig von der Threadanzahl)
+	size_t shared_size = 0;
+
+//	fprintf(stderr, "threads=%d, %d blocks, %d threads per block, %d bytes shared\n", threads, grid.x, block.x, shared_size);
+
+	blake512_gpu_hash<<<grid, block, shared_size>>>(threads, startNounce, d_hash5output[thr_id], d_heftyHashes[thr_id], d_nonceVector[thr_id]);
+}
diff --git a/cuda_blake512.h b/cuda_blake512.h
new file mode 100644
index 0000000000..b0cf2019cc
--- /dev/null
+++ b/cuda_blake512.h
@@ -0,0 +1,8 @@
+#ifndef _CUDA_BLAKE512_H
+#define _CUDA_BLAKE512_H
+
+void blake512_cpu_init(int thr_id, int threads);
+void blake512_cpu_setBlock(void *pdata);
+void blake512_cpu_hash(int thr_id, int threads, uint32_t startNounce);
+
+#endif
diff --git a/cuda_combine.cu b/cuda_combine.cu
new file mode 100644
index 0000000000..2949765329
--- /dev/null
+++ b/cuda_combine.cu
@@ -0,0 +1,151 @@
+/* Diese Funktion ist auf 84+32 Byte große Eingabedaten ausgerichtet (Heavycoin) */
+#include <cuda.h>
+#include "cuda_runtime.h"
+#include "device_launch_parameters.h"
+
+// Folgende Definitionen später durch header ersetzen
+typedef unsigned int uint32_t;
+
+// globaler Speicher für unsere Ergebnisse
+uint32_t *d_hashoutput[8];
+
+extern uint32_t *d_hash2output[8];
+extern uint32_t *d_hash3output[8];
+extern uint32_t *d_hash4output[8];
+extern uint32_t *d_hash5output[8];
+extern uint32_t *d_nonceVector[8];
+
+/* Combines top 64-bits from each hash into a single hash */
+static void __device__ combine_hashes(uint32_t *out, uint32_t *hash1, uint32_t *hash2, uint32_t *hash3, uint32_t *hash4)
+{
+	uint32_t lout[8]; // Combining in Registern machen
+
+#pragma unroll 8
+	for (int i=0; i < 8; ++i)
+		lout[i] = 0;
+
+	// das Makro setzt jeweils 4 Bits aus vier verschiedenen Hashes zu einem Nibble zusammen
+#define MIX(bits, mask, i) \
+	lout[(255 - (bits+3))/32] <<= 4; \
+	if ((hash1[i] & mask) != 0) lout[(255 - (bits+0))/32] |= 8; \
+	if ((hash2[i] & mask) != 0) lout[(255 - (bits+1))/32] |= 4; \
+	if ((hash3[i] & mask) != 0) lout[(255 - (bits+2))/32] |= 2; \
+	if ((hash4[i] & mask) != 0) lout[(255 - (bits+3))/32] |= 1; \
+
+	/* Transpose first 64 bits of each hash into out */
+	MIX(  0, 0x80000000, 7);
+	MIX(  4, 0x40000000, 7);
+	MIX(  8, 0x20000000, 7);
+	MIX( 12, 0x10000000, 7);
+	MIX( 16, 0x08000000, 7);
+	MIX( 20, 0x04000000, 7);
+	MIX( 24, 0x02000000, 7);
+	MIX( 28, 0x01000000, 7);
+	MIX( 32, 0x00800000, 7);
+	MIX( 36, 0x00400000, 7);
+	MIX( 40, 0x00200000, 7);
+	MIX( 44, 0x00100000, 7);
+	MIX( 48, 0x00080000, 7);
+	MIX( 52, 0x00040000, 7);
+	MIX( 56, 0x00020000, 7);
+	MIX( 60, 0x00010000, 7);
+	MIX( 64, 0x00008000, 7);
+	MIX( 68, 0x00004000, 7);
+	MIX( 72, 0x00002000, 7);
+	MIX( 76, 0x00001000, 7);
+	MIX( 80, 0x00000800, 7);
+	MIX( 84, 0x00000400, 7);
+	MIX( 88, 0x00000200, 7);
+	MIX( 92, 0x00000100, 7);
+	MIX( 96, 0x00000080, 7);
+	MIX(100, 0x00000040, 7);
+	MIX(104, 0x00000020, 7);
+	MIX(108, 0x00000010, 7);
+	MIX(112, 0x00000008, 7);
+	MIX(116, 0x00000004, 7);
+	MIX(120, 0x00000002, 7);
+	MIX(124, 0x00000001, 7);
+
+	MIX(128, 0x80000000, 6);
+	MIX(132, 0x40000000, 6);
+	MIX(136, 0x20000000, 6);
+	MIX(140, 0x10000000, 6);
+	MIX(144, 0x08000000, 6);
+	MIX(148, 0x04000000, 6);
+	MIX(152, 0x02000000, 6);
+	MIX(156, 0x01000000, 6);
+	MIX(160, 0x00800000, 6);
+	MIX(164, 0x00400000, 6);
+	MIX(168, 0x00200000, 6);
+	MIX(172, 0x00100000, 6);
+	MIX(176, 0x00080000, 6);
+	MIX(180, 0x00040000, 6);
+	MIX(184, 0x00020000, 6);
+	MIX(188, 0x00010000, 6);
+	MIX(192, 0x00008000, 6);
+	MIX(196, 0x00004000, 6);
+	MIX(200, 0x00002000, 6);
+	MIX(204, 0x00001000, 6);
+	MIX(208, 0x00000800, 6);
+	MIX(212, 0x00000400, 6);
+	MIX(216, 0x00000200, 6);
+	MIX(220, 0x00000100, 6);
+	MIX(224, 0x00000080, 6);
+	MIX(228, 0x00000040, 6);
+	MIX(232, 0x00000020, 6);
+	MIX(236, 0x00000010, 6);
+	MIX(240, 0x00000008, 6);
+	MIX(244, 0x00000004, 6);
+	MIX(248, 0x00000002, 6);
+	MIX(252, 0x00000001, 6);
+
+#pragma unroll 8
+	for (int i=0; i < 8; ++i)
+		out[i] = lout[i];
+}
+
+__global__ void combine_gpu_hash(int threads, uint32_t startNounce, uint32_t *out, uint32_t *hash2, uint32_t *hash3, uint32_t *hash4, uint32_t *hash5, uint32_t *nonceVector)
+{
+	int thread = (blockDim.x * blockIdx.x + threadIdx.x);
+	if (thread < threads)
+	{
+		uint32_t nounce = nonceVector[thread];
+		uint32_t hashPosition = nounce - startNounce;
+		// Die Aufgabe der combine-funktion besteht aus zwei Teilen.
+		// 1) Komprimiere die hashes zu einem kleinen Array
+		// 2) Errechne dort den combines-value
+
+		// Die Kompression wird dadurch verwirklicht, dass im out-array weiterhin mit "thread" indiziert
+		// wird. Die anderen Werte werden mit der nonce indiziert
+
+		combine_hashes(&out[8 * thread], &hash2[8 * hashPosition], &hash3[16 * hashPosition], &hash4[16 * hashPosition], &hash5[16 * hashPosition]);
+	}
+}
+
+// Setup-Funktionen
+__host__ void combine_cpu_init(int thr_id, int threads)
+{
+	// Speicher für alle Ergebnisse belegen
+	cudaMalloc(&d_hashoutput[thr_id], 8 * sizeof(uint32_t) * threads);
+}
+
+void combine_cpu_hash(int thr_id, int threads, uint32_t startNounce, uint32_t *hash)
+{
+	// diese Kopien sind optional, da die Hashes jetzt bereits auf der GPU liegen sollten
+
+	const int threadsperblock = 128;
+
+	// berechne wie viele Thread Blocks wir brauchen
+	dim3 grid((threads + threadsperblock-1)/threadsperblock);
+	dim3 block(threadsperblock);
+
+	// Größe des dynamischen Shared Memory Bereichs (abhängig von der Threadanzahl)
+	size_t shared_size = 0;
+
+//	fprintf(stderr, "threads=%d, %d blocks, %d threads per block, %d bytes shared\n", threads, grid.x, block.x, shared_size);
+
+	combine_gpu_hash<<<grid, block, shared_size>>>(threads, startNounce, d_hashoutput[thr_id], d_hash2output[thr_id], d_hash3output[thr_id], d_hash4output[thr_id], d_hash5output[thr_id], d_nonceVector[thr_id]);
+
+	// da die Hash Auswertung noch auf der CPU erfolgt, müssen die Ergebnisse auf jeden Fall zum Host kopiert werden
+	cudaMemcpy(hash, d_hashoutput[thr_id], 8 * sizeof(uint32_t) * threads, cudaMemcpyDeviceToHost);
+}
diff --git a/cuda_combine.h b/cuda_combine.h
new file mode 100644
index 0000000000..ada3a2180a
--- /dev/null
+++ b/cuda_combine.h
@@ -0,0 +1,7 @@
+#ifndef _CUDA_COMBINE_H
+#define _CUDA_COMBINE_H
+
+void combine_cpu_init(int thr_id, int threads);
+void combine_cpu_hash(int thr_id, int threads, uint32_t startNounce, uint32_t *hash);
+
+#endif
diff --git a/cuda_fugue256.cu b/cuda_fugue256.cu
new file mode 100644
index 0000000000..0457130ecf
--- /dev/null
+++ b/cuda_fugue256.cu
@@ -0,0 +1,782 @@
+#if 1
+/* Diese Funktion ist auf 84+32 Byte große Eingabedaten ausgerichtet (Heavycoin) */
+#include <cuda.h>
+#include "cuda_runtime.h"
+#include "device_launch_parameters.h"
+
+#include <stdio.h>
+#include <memory.h>
+
+#include "sph_fugue.h"
+
+// heavy.cu
+extern cudaError_t MyStreamSynchronize(cudaStream_t stream, int situation, int thr_id);
+
+// Folgende Definitionen später durch header ersetzen
+typedef unsigned char uint8_t;
+typedef unsigned int uint32_t;
+typedef unsigned long long uint64_t;
+
+// schon in sph_fugue.h definiert
+//#define SPH_C32(x)	((uint32_t)(x ## U))
+
+uint32_t *d_fugue256_hashoutput[8];
+uint32_t *d_resultNonce[8];
+
+__constant__ uint32_t GPUstate[30]; // Single GPU
+__constant__ uint32_t pTarget[8]; // Single GPU
+
+//__constant__ uint32_t mixtab0[256]; // 1K
+//__constant__ uint32_t mixtab1[256]; // 1K
+//__constant__ uint32_t mixtab2[256]; // 1K
+//__constant__ uint32_t mixtab3[256]; // 1K
+
+texture<unsigned int, 1, cudaReadModeElementType> mixTab0Tex;
+texture<unsigned int, 1, cudaReadModeElementType> mixTab1Tex;
+texture<unsigned int, 1, cudaReadModeElementType> mixTab2Tex;
+texture<unsigned int, 1, cudaReadModeElementType> mixTab3Tex;
+
+#define mixtab0(x) tex1Dfetch(mixTab0Tex, x)
+#define mixtab1(x) tex1Dfetch(mixTab1Tex, x)
+#define mixtab2(x) tex1Dfetch(mixTab2Tex, x)
+#define mixtab3(x) tex1Dfetch(mixTab3Tex, x)
+
+/* TABELLEN */
+static const uint32_t mixtab0_cpu[] = {
+	SPH_C32(0x63633297), SPH_C32(0x7c7c6feb), SPH_C32(0x77775ec7),
+	SPH_C32(0x7b7b7af7), SPH_C32(0xf2f2e8e5), SPH_C32(0x6b6b0ab7),
+	SPH_C32(0x6f6f16a7), SPH_C32(0xc5c56d39), SPH_C32(0x303090c0),
+	SPH_C32(0x01010704), SPH_C32(0x67672e87), SPH_C32(0x2b2bd1ac),
+	SPH_C32(0xfefeccd5), SPH_C32(0xd7d71371), SPH_C32(0xabab7c9a),
+	SPH_C32(0x767659c3), SPH_C32(0xcaca4005), SPH_C32(0x8282a33e),
+	SPH_C32(0xc9c94909), SPH_C32(0x7d7d68ef), SPH_C32(0xfafad0c5),
+	SPH_C32(0x5959947f), SPH_C32(0x4747ce07), SPH_C32(0xf0f0e6ed),
+	SPH_C32(0xadad6e82), SPH_C32(0xd4d41a7d), SPH_C32(0xa2a243be),
+	SPH_C32(0xafaf608a), SPH_C32(0x9c9cf946), SPH_C32(0xa4a451a6),
+	SPH_C32(0x727245d3), SPH_C32(0xc0c0762d), SPH_C32(0xb7b728ea),
+	SPH_C32(0xfdfdc5d9), SPH_C32(0x9393d47a), SPH_C32(0x2626f298),
+	SPH_C32(0x363682d8), SPH_C32(0x3f3fbdfc), SPH_C32(0xf7f7f3f1),
+	SPH_C32(0xcccc521d), SPH_C32(0x34348cd0), SPH_C32(0xa5a556a2),
+	SPH_C32(0xe5e58db9), SPH_C32(0xf1f1e1e9), SPH_C32(0x71714cdf),
+	SPH_C32(0xd8d83e4d), SPH_C32(0x313197c4), SPH_C32(0x15156b54),
+	SPH_C32(0x04041c10), SPH_C32(0xc7c76331), SPH_C32(0x2323e98c),
+	SPH_C32(0xc3c37f21), SPH_C32(0x18184860), SPH_C32(0x9696cf6e),
+	SPH_C32(0x05051b14), SPH_C32(0x9a9aeb5e), SPH_C32(0x0707151c),
+	SPH_C32(0x12127e48), SPH_C32(0x8080ad36), SPH_C32(0xe2e298a5),
+	SPH_C32(0xebeba781), SPH_C32(0x2727f59c), SPH_C32(0xb2b233fe),
+	SPH_C32(0x757550cf), SPH_C32(0x09093f24), SPH_C32(0x8383a43a),
+	SPH_C32(0x2c2cc4b0), SPH_C32(0x1a1a4668), SPH_C32(0x1b1b416c),
+	SPH_C32(0x6e6e11a3), SPH_C32(0x5a5a9d73), SPH_C32(0xa0a04db6),
+	SPH_C32(0x5252a553), SPH_C32(0x3b3ba1ec), SPH_C32(0xd6d61475),
+	SPH_C32(0xb3b334fa), SPH_C32(0x2929dfa4), SPH_C32(0xe3e39fa1),
+	SPH_C32(0x2f2fcdbc), SPH_C32(0x8484b126), SPH_C32(0x5353a257),
+	SPH_C32(0xd1d10169), SPH_C32(0x00000000), SPH_C32(0xededb599),
+	SPH_C32(0x2020e080), SPH_C32(0xfcfcc2dd), SPH_C32(0xb1b13af2),
+	SPH_C32(0x5b5b9a77), SPH_C32(0x6a6a0db3), SPH_C32(0xcbcb4701),
+	SPH_C32(0xbebe17ce), SPH_C32(0x3939afe4), SPH_C32(0x4a4aed33),
+	SPH_C32(0x4c4cff2b), SPH_C32(0x5858937b), SPH_C32(0xcfcf5b11),
+	SPH_C32(0xd0d0066d), SPH_C32(0xefefbb91), SPH_C32(0xaaaa7b9e),
+	SPH_C32(0xfbfbd7c1), SPH_C32(0x4343d217), SPH_C32(0x4d4df82f),
+	SPH_C32(0x333399cc), SPH_C32(0x8585b622), SPH_C32(0x4545c00f),
+	SPH_C32(0xf9f9d9c9), SPH_C32(0x02020e08), SPH_C32(0x7f7f66e7),
+	SPH_C32(0x5050ab5b), SPH_C32(0x3c3cb4f0), SPH_C32(0x9f9ff04a),
+	SPH_C32(0xa8a87596), SPH_C32(0x5151ac5f), SPH_C32(0xa3a344ba),
+	SPH_C32(0x4040db1b), SPH_C32(0x8f8f800a), SPH_C32(0x9292d37e),
+	SPH_C32(0x9d9dfe42), SPH_C32(0x3838a8e0), SPH_C32(0xf5f5fdf9),
+	SPH_C32(0xbcbc19c6), SPH_C32(0xb6b62fee), SPH_C32(0xdada3045),
+	SPH_C32(0x2121e784), SPH_C32(0x10107040), SPH_C32(0xffffcbd1),
+	SPH_C32(0xf3f3efe1), SPH_C32(0xd2d20865), SPH_C32(0xcdcd5519),
+	SPH_C32(0x0c0c2430), SPH_C32(0x1313794c), SPH_C32(0xececb29d),
+	SPH_C32(0x5f5f8667), SPH_C32(0x9797c86a), SPH_C32(0x4444c70b),
+	SPH_C32(0x1717655c), SPH_C32(0xc4c46a3d), SPH_C32(0xa7a758aa),
+	SPH_C32(0x7e7e61e3), SPH_C32(0x3d3db3f4), SPH_C32(0x6464278b),
+	SPH_C32(0x5d5d886f), SPH_C32(0x19194f64), SPH_C32(0x737342d7),
+	SPH_C32(0x60603b9b), SPH_C32(0x8181aa32), SPH_C32(0x4f4ff627),
+	SPH_C32(0xdcdc225d), SPH_C32(0x2222ee88), SPH_C32(0x2a2ad6a8),
+	SPH_C32(0x9090dd76), SPH_C32(0x88889516), SPH_C32(0x4646c903),
+	SPH_C32(0xeeeebc95), SPH_C32(0xb8b805d6), SPH_C32(0x14146c50),
+	SPH_C32(0xdede2c55), SPH_C32(0x5e5e8163), SPH_C32(0x0b0b312c),
+	SPH_C32(0xdbdb3741), SPH_C32(0xe0e096ad), SPH_C32(0x32329ec8),
+	SPH_C32(0x3a3aa6e8), SPH_C32(0x0a0a3628), SPH_C32(0x4949e43f),
+	SPH_C32(0x06061218), SPH_C32(0x2424fc90), SPH_C32(0x5c5c8f6b),
+	SPH_C32(0xc2c27825), SPH_C32(0xd3d30f61), SPH_C32(0xacac6986),
+	SPH_C32(0x62623593), SPH_C32(0x9191da72), SPH_C32(0x9595c662),
+	SPH_C32(0xe4e48abd), SPH_C32(0x797974ff), SPH_C32(0xe7e783b1),
+	SPH_C32(0xc8c84e0d), SPH_C32(0x373785dc), SPH_C32(0x6d6d18af),
+	SPH_C32(0x8d8d8e02), SPH_C32(0xd5d51d79), SPH_C32(0x4e4ef123),
+	SPH_C32(0xa9a97292), SPH_C32(0x6c6c1fab), SPH_C32(0x5656b943),
+	SPH_C32(0xf4f4fafd), SPH_C32(0xeaeaa085), SPH_C32(0x6565208f),
+	SPH_C32(0x7a7a7df3), SPH_C32(0xaeae678e), SPH_C32(0x08083820),
+	SPH_C32(0xbaba0bde), SPH_C32(0x787873fb), SPH_C32(0x2525fb94),
+	SPH_C32(0x2e2ecab8), SPH_C32(0x1c1c5470), SPH_C32(0xa6a65fae),
+	SPH_C32(0xb4b421e6), SPH_C32(0xc6c66435), SPH_C32(0xe8e8ae8d),
+	SPH_C32(0xdddd2559), SPH_C32(0x747457cb), SPH_C32(0x1f1f5d7c),
+	SPH_C32(0x4b4bea37), SPH_C32(0xbdbd1ec2), SPH_C32(0x8b8b9c1a),
+	SPH_C32(0x8a8a9b1e), SPH_C32(0x70704bdb), SPH_C32(0x3e3ebaf8),
+	SPH_C32(0xb5b526e2), SPH_C32(0x66662983), SPH_C32(0x4848e33b),
+	SPH_C32(0x0303090c), SPH_C32(0xf6f6f4f5), SPH_C32(0x0e0e2a38),
+	SPH_C32(0x61613c9f), SPH_C32(0x35358bd4), SPH_C32(0x5757be47),
+	SPH_C32(0xb9b902d2), SPH_C32(0x8686bf2e), SPH_C32(0xc1c17129),
+	SPH_C32(0x1d1d5374), SPH_C32(0x9e9ef74e), SPH_C32(0xe1e191a9),
+	SPH_C32(0xf8f8decd), SPH_C32(0x9898e556), SPH_C32(0x11117744),
+	SPH_C32(0x696904bf), SPH_C32(0xd9d93949), SPH_C32(0x8e8e870e),
+	SPH_C32(0x9494c166), SPH_C32(0x9b9bec5a), SPH_C32(0x1e1e5a78),
+	SPH_C32(0x8787b82a), SPH_C32(0xe9e9a989), SPH_C32(0xcece5c15),
+	SPH_C32(0x5555b04f), SPH_C32(0x2828d8a0), SPH_C32(0xdfdf2b51),
+	SPH_C32(0x8c8c8906), SPH_C32(0xa1a14ab2), SPH_C32(0x89899212),
+	SPH_C32(0x0d0d2334), SPH_C32(0xbfbf10ca), SPH_C32(0xe6e684b5),
+	SPH_C32(0x4242d513), SPH_C32(0x686803bb), SPH_C32(0x4141dc1f),
+	SPH_C32(0x9999e252), SPH_C32(0x2d2dc3b4), SPH_C32(0x0f0f2d3c),
+	SPH_C32(0xb0b03df6), SPH_C32(0x5454b74b), SPH_C32(0xbbbb0cda),
+	SPH_C32(0x16166258)
+};
+
+static const uint32_t mixtab1_cpu[] = {
+	SPH_C32(0x97636332), SPH_C32(0xeb7c7c6f), SPH_C32(0xc777775e),
+	SPH_C32(0xf77b7b7a), SPH_C32(0xe5f2f2e8), SPH_C32(0xb76b6b0a),
+	SPH_C32(0xa76f6f16), SPH_C32(0x39c5c56d), SPH_C32(0xc0303090),
+	SPH_C32(0x04010107), SPH_C32(0x8767672e), SPH_C32(0xac2b2bd1),
+	SPH_C32(0xd5fefecc), SPH_C32(0x71d7d713), SPH_C32(0x9aabab7c),
+	SPH_C32(0xc3767659), SPH_C32(0x05caca40), SPH_C32(0x3e8282a3),
+	SPH_C32(0x09c9c949), SPH_C32(0xef7d7d68), SPH_C32(0xc5fafad0),
+	SPH_C32(0x7f595994), SPH_C32(0x074747ce), SPH_C32(0xedf0f0e6),
+	SPH_C32(0x82adad6e), SPH_C32(0x7dd4d41a), SPH_C32(0xbea2a243),
+	SPH_C32(0x8aafaf60), SPH_C32(0x469c9cf9), SPH_C32(0xa6a4a451),
+	SPH_C32(0xd3727245), SPH_C32(0x2dc0c076), SPH_C32(0xeab7b728),
+	SPH_C32(0xd9fdfdc5), SPH_C32(0x7a9393d4), SPH_C32(0x982626f2),
+	SPH_C32(0xd8363682), SPH_C32(0xfc3f3fbd), SPH_C32(0xf1f7f7f3),
+	SPH_C32(0x1dcccc52), SPH_C32(0xd034348c), SPH_C32(0xa2a5a556),
+	SPH_C32(0xb9e5e58d), SPH_C32(0xe9f1f1e1), SPH_C32(0xdf71714c),
+	SPH_C32(0x4dd8d83e), SPH_C32(0xc4313197), SPH_C32(0x5415156b),
+	SPH_C32(0x1004041c), SPH_C32(0x31c7c763), SPH_C32(0x8c2323e9),
+	SPH_C32(0x21c3c37f), SPH_C32(0x60181848), SPH_C32(0x6e9696cf),
+	SPH_C32(0x1405051b), SPH_C32(0x5e9a9aeb), SPH_C32(0x1c070715),
+	SPH_C32(0x4812127e), SPH_C32(0x368080ad), SPH_C32(0xa5e2e298),
+	SPH_C32(0x81ebeba7), SPH_C32(0x9c2727f5), SPH_C32(0xfeb2b233),
+	SPH_C32(0xcf757550), SPH_C32(0x2409093f), SPH_C32(0x3a8383a4),
+	SPH_C32(0xb02c2cc4), SPH_C32(0x681a1a46), SPH_C32(0x6c1b1b41),
+	SPH_C32(0xa36e6e11), SPH_C32(0x735a5a9d), SPH_C32(0xb6a0a04d),
+	SPH_C32(0x535252a5), SPH_C32(0xec3b3ba1), SPH_C32(0x75d6d614),
+	SPH_C32(0xfab3b334), SPH_C32(0xa42929df), SPH_C32(0xa1e3e39f),
+	SPH_C32(0xbc2f2fcd), SPH_C32(0x268484b1), SPH_C32(0x575353a2),
+	SPH_C32(0x69d1d101), SPH_C32(0x00000000), SPH_C32(0x99ededb5),
+	SPH_C32(0x802020e0), SPH_C32(0xddfcfcc2), SPH_C32(0xf2b1b13a),
+	SPH_C32(0x775b5b9a), SPH_C32(0xb36a6a0d), SPH_C32(0x01cbcb47),
+	SPH_C32(0xcebebe17), SPH_C32(0xe43939af), SPH_C32(0x334a4aed),
+	SPH_C32(0x2b4c4cff), SPH_C32(0x7b585893), SPH_C32(0x11cfcf5b),
+	SPH_C32(0x6dd0d006), SPH_C32(0x91efefbb), SPH_C32(0x9eaaaa7b),
+	SPH_C32(0xc1fbfbd7), SPH_C32(0x174343d2), SPH_C32(0x2f4d4df8),
+	SPH_C32(0xcc333399), SPH_C32(0x228585b6), SPH_C32(0x0f4545c0),
+	SPH_C32(0xc9f9f9d9), SPH_C32(0x0802020e), SPH_C32(0xe77f7f66),
+	SPH_C32(0x5b5050ab), SPH_C32(0xf03c3cb4), SPH_C32(0x4a9f9ff0),
+	SPH_C32(0x96a8a875), SPH_C32(0x5f5151ac), SPH_C32(0xbaa3a344),
+	SPH_C32(0x1b4040db), SPH_C32(0x0a8f8f80), SPH_C32(0x7e9292d3),
+	SPH_C32(0x429d9dfe), SPH_C32(0xe03838a8), SPH_C32(0xf9f5f5fd),
+	SPH_C32(0xc6bcbc19), SPH_C32(0xeeb6b62f), SPH_C32(0x45dada30),
+	SPH_C32(0x842121e7), SPH_C32(0x40101070), SPH_C32(0xd1ffffcb),
+	SPH_C32(0xe1f3f3ef), SPH_C32(0x65d2d208), SPH_C32(0x19cdcd55),
+	SPH_C32(0x300c0c24), SPH_C32(0x4c131379), SPH_C32(0x9dececb2),
+	SPH_C32(0x675f5f86), SPH_C32(0x6a9797c8), SPH_C32(0x0b4444c7),
+	SPH_C32(0x5c171765), SPH_C32(0x3dc4c46a), SPH_C32(0xaaa7a758),
+	SPH_C32(0xe37e7e61), SPH_C32(0xf43d3db3), SPH_C32(0x8b646427),
+	SPH_C32(0x6f5d5d88), SPH_C32(0x6419194f), SPH_C32(0xd7737342),
+	SPH_C32(0x9b60603b), SPH_C32(0x328181aa), SPH_C32(0x274f4ff6),
+	SPH_C32(0x5ddcdc22), SPH_C32(0x882222ee), SPH_C32(0xa82a2ad6),
+	SPH_C32(0x769090dd), SPH_C32(0x16888895), SPH_C32(0x034646c9),
+	SPH_C32(0x95eeeebc), SPH_C32(0xd6b8b805), SPH_C32(0x5014146c),
+	SPH_C32(0x55dede2c), SPH_C32(0x635e5e81), SPH_C32(0x2c0b0b31),
+	SPH_C32(0x41dbdb37), SPH_C32(0xade0e096), SPH_C32(0xc832329e),
+	SPH_C32(0xe83a3aa6), SPH_C32(0x280a0a36), SPH_C32(0x3f4949e4),
+	SPH_C32(0x18060612), SPH_C32(0x902424fc), SPH_C32(0x6b5c5c8f),
+	SPH_C32(0x25c2c278), SPH_C32(0x61d3d30f), SPH_C32(0x86acac69),
+	SPH_C32(0x93626235), SPH_C32(0x729191da), SPH_C32(0x629595c6),
+	SPH_C32(0xbde4e48a), SPH_C32(0xff797974), SPH_C32(0xb1e7e783),
+	SPH_C32(0x0dc8c84e), SPH_C32(0xdc373785), SPH_C32(0xaf6d6d18),
+	SPH_C32(0x028d8d8e), SPH_C32(0x79d5d51d), SPH_C32(0x234e4ef1),
+	SPH_C32(0x92a9a972), SPH_C32(0xab6c6c1f), SPH_C32(0x435656b9),
+	SPH_C32(0xfdf4f4fa), SPH_C32(0x85eaeaa0), SPH_C32(0x8f656520),
+	SPH_C32(0xf37a7a7d), SPH_C32(0x8eaeae67), SPH_C32(0x20080838),
+	SPH_C32(0xdebaba0b), SPH_C32(0xfb787873), SPH_C32(0x942525fb),
+	SPH_C32(0xb82e2eca), SPH_C32(0x701c1c54), SPH_C32(0xaea6a65f),
+	SPH_C32(0xe6b4b421), SPH_C32(0x35c6c664), SPH_C32(0x8de8e8ae),
+	SPH_C32(0x59dddd25), SPH_C32(0xcb747457), SPH_C32(0x7c1f1f5d),
+	SPH_C32(0x374b4bea), SPH_C32(0xc2bdbd1e), SPH_C32(0x1a8b8b9c),
+	SPH_C32(0x1e8a8a9b), SPH_C32(0xdb70704b), SPH_C32(0xf83e3eba),
+	SPH_C32(0xe2b5b526), SPH_C32(0x83666629), SPH_C32(0x3b4848e3),
+	SPH_C32(0x0c030309), SPH_C32(0xf5f6f6f4), SPH_C32(0x380e0e2a),
+	SPH_C32(0x9f61613c), SPH_C32(0xd435358b), SPH_C32(0x475757be),
+	SPH_C32(0xd2b9b902), SPH_C32(0x2e8686bf), SPH_C32(0x29c1c171),
+	SPH_C32(0x741d1d53), SPH_C32(0x4e9e9ef7), SPH_C32(0xa9e1e191),
+	SPH_C32(0xcdf8f8de), SPH_C32(0x569898e5), SPH_C32(0x44111177),
+	SPH_C32(0xbf696904), SPH_C32(0x49d9d939), SPH_C32(0x0e8e8e87),
+	SPH_C32(0x669494c1), SPH_C32(0x5a9b9bec), SPH_C32(0x781e1e5a),
+	SPH_C32(0x2a8787b8), SPH_C32(0x89e9e9a9), SPH_C32(0x15cece5c),
+	SPH_C32(0x4f5555b0), SPH_C32(0xa02828d8), SPH_C32(0x51dfdf2b),
+	SPH_C32(0x068c8c89), SPH_C32(0xb2a1a14a), SPH_C32(0x12898992),
+	SPH_C32(0x340d0d23), SPH_C32(0xcabfbf10), SPH_C32(0xb5e6e684),
+	SPH_C32(0x134242d5), SPH_C32(0xbb686803), SPH_C32(0x1f4141dc),
+	SPH_C32(0x529999e2), SPH_C32(0xb42d2dc3), SPH_C32(0x3c0f0f2d),
+	SPH_C32(0xf6b0b03d), SPH_C32(0x4b5454b7), SPH_C32(0xdabbbb0c),
+	SPH_C32(0x58161662)
+};
+
+static const uint32_t mixtab2_cpu[] = {
+	SPH_C32(0x32976363), SPH_C32(0x6feb7c7c), SPH_C32(0x5ec77777),
+	SPH_C32(0x7af77b7b), SPH_C32(0xe8e5f2f2), SPH_C32(0x0ab76b6b),
+	SPH_C32(0x16a76f6f), SPH_C32(0x6d39c5c5), SPH_C32(0x90c03030),
+	SPH_C32(0x07040101), SPH_C32(0x2e876767), SPH_C32(0xd1ac2b2b),
+	SPH_C32(0xccd5fefe), SPH_C32(0x1371d7d7), SPH_C32(0x7c9aabab),
+	SPH_C32(0x59c37676), SPH_C32(0x4005caca), SPH_C32(0xa33e8282),
+	SPH_C32(0x4909c9c9), SPH_C32(0x68ef7d7d), SPH_C32(0xd0c5fafa),
+	SPH_C32(0x947f5959), SPH_C32(0xce074747), SPH_C32(0xe6edf0f0),
+	SPH_C32(0x6e82adad), SPH_C32(0x1a7dd4d4), SPH_C32(0x43bea2a2),
+	SPH_C32(0x608aafaf), SPH_C32(0xf9469c9c), SPH_C32(0x51a6a4a4),
+	SPH_C32(0x45d37272), SPH_C32(0x762dc0c0), SPH_C32(0x28eab7b7),
+	SPH_C32(0xc5d9fdfd), SPH_C32(0xd47a9393), SPH_C32(0xf2982626),
+	SPH_C32(0x82d83636), SPH_C32(0xbdfc3f3f), SPH_C32(0xf3f1f7f7),
+	SPH_C32(0x521dcccc), SPH_C32(0x8cd03434), SPH_C32(0x56a2a5a5),
+	SPH_C32(0x8db9e5e5), SPH_C32(0xe1e9f1f1), SPH_C32(0x4cdf7171),
+	SPH_C32(0x3e4dd8d8), SPH_C32(0x97c43131), SPH_C32(0x6b541515),
+	SPH_C32(0x1c100404), SPH_C32(0x6331c7c7), SPH_C32(0xe98c2323),
+	SPH_C32(0x7f21c3c3), SPH_C32(0x48601818), SPH_C32(0xcf6e9696),
+	SPH_C32(0x1b140505), SPH_C32(0xeb5e9a9a), SPH_C32(0x151c0707),
+	SPH_C32(0x7e481212), SPH_C32(0xad368080), SPH_C32(0x98a5e2e2),
+	SPH_C32(0xa781ebeb), SPH_C32(0xf59c2727), SPH_C32(0x33feb2b2),
+	SPH_C32(0x50cf7575), SPH_C32(0x3f240909), SPH_C32(0xa43a8383),
+	SPH_C32(0xc4b02c2c), SPH_C32(0x46681a1a), SPH_C32(0x416c1b1b),
+	SPH_C32(0x11a36e6e), SPH_C32(0x9d735a5a), SPH_C32(0x4db6a0a0),
+	SPH_C32(0xa5535252), SPH_C32(0xa1ec3b3b), SPH_C32(0x1475d6d6),
+	SPH_C32(0x34fab3b3), SPH_C32(0xdfa42929), SPH_C32(0x9fa1e3e3),
+	SPH_C32(0xcdbc2f2f), SPH_C32(0xb1268484), SPH_C32(0xa2575353),
+	SPH_C32(0x0169d1d1), SPH_C32(0x00000000), SPH_C32(0xb599eded),
+	SPH_C32(0xe0802020), SPH_C32(0xc2ddfcfc), SPH_C32(0x3af2b1b1),
+	SPH_C32(0x9a775b5b), SPH_C32(0x0db36a6a), SPH_C32(0x4701cbcb),
+	SPH_C32(0x17cebebe), SPH_C32(0xafe43939), SPH_C32(0xed334a4a),
+	SPH_C32(0xff2b4c4c), SPH_C32(0x937b5858), SPH_C32(0x5b11cfcf),
+	SPH_C32(0x066dd0d0), SPH_C32(0xbb91efef), SPH_C32(0x7b9eaaaa),
+	SPH_C32(0xd7c1fbfb), SPH_C32(0xd2174343), SPH_C32(0xf82f4d4d),
+	SPH_C32(0x99cc3333), SPH_C32(0xb6228585), SPH_C32(0xc00f4545),
+	SPH_C32(0xd9c9f9f9), SPH_C32(0x0e080202), SPH_C32(0x66e77f7f),
+	SPH_C32(0xab5b5050), SPH_C32(0xb4f03c3c), SPH_C32(0xf04a9f9f),
+	SPH_C32(0x7596a8a8), SPH_C32(0xac5f5151), SPH_C32(0x44baa3a3),
+	SPH_C32(0xdb1b4040), SPH_C32(0x800a8f8f), SPH_C32(0xd37e9292),
+	SPH_C32(0xfe429d9d), SPH_C32(0xa8e03838), SPH_C32(0xfdf9f5f5),
+	SPH_C32(0x19c6bcbc), SPH_C32(0x2feeb6b6), SPH_C32(0x3045dada),
+	SPH_C32(0xe7842121), SPH_C32(0x70401010), SPH_C32(0xcbd1ffff),
+	SPH_C32(0xefe1f3f3), SPH_C32(0x0865d2d2), SPH_C32(0x5519cdcd),
+	SPH_C32(0x24300c0c), SPH_C32(0x794c1313), SPH_C32(0xb29decec),
+	SPH_C32(0x86675f5f), SPH_C32(0xc86a9797), SPH_C32(0xc70b4444),
+	SPH_C32(0x655c1717), SPH_C32(0x6a3dc4c4), SPH_C32(0x58aaa7a7),
+	SPH_C32(0x61e37e7e), SPH_C32(0xb3f43d3d), SPH_C32(0x278b6464),
+	SPH_C32(0x886f5d5d), SPH_C32(0x4f641919), SPH_C32(0x42d77373),
+	SPH_C32(0x3b9b6060), SPH_C32(0xaa328181), SPH_C32(0xf6274f4f),
+	SPH_C32(0x225ddcdc), SPH_C32(0xee882222), SPH_C32(0xd6a82a2a),
+	SPH_C32(0xdd769090), SPH_C32(0x95168888), SPH_C32(0xc9034646),
+	SPH_C32(0xbc95eeee), SPH_C32(0x05d6b8b8), SPH_C32(0x6c501414),
+	SPH_C32(0x2c55dede), SPH_C32(0x81635e5e), SPH_C32(0x312c0b0b),
+	SPH_C32(0x3741dbdb), SPH_C32(0x96ade0e0), SPH_C32(0x9ec83232),
+	SPH_C32(0xa6e83a3a), SPH_C32(0x36280a0a), SPH_C32(0xe43f4949),
+	SPH_C32(0x12180606), SPH_C32(0xfc902424), SPH_C32(0x8f6b5c5c),
+	SPH_C32(0x7825c2c2), SPH_C32(0x0f61d3d3), SPH_C32(0x6986acac),
+	SPH_C32(0x35936262), SPH_C32(0xda729191), SPH_C32(0xc6629595),
+	SPH_C32(0x8abde4e4), SPH_C32(0x74ff7979), SPH_C32(0x83b1e7e7),
+	SPH_C32(0x4e0dc8c8), SPH_C32(0x85dc3737), SPH_C32(0x18af6d6d),
+	SPH_C32(0x8e028d8d), SPH_C32(0x1d79d5d5), SPH_C32(0xf1234e4e),
+	SPH_C32(0x7292a9a9), SPH_C32(0x1fab6c6c), SPH_C32(0xb9435656),
+	SPH_C32(0xfafdf4f4), SPH_C32(0xa085eaea), SPH_C32(0x208f6565),
+	SPH_C32(0x7df37a7a), SPH_C32(0x678eaeae), SPH_C32(0x38200808),
+	SPH_C32(0x0bdebaba), SPH_C32(0x73fb7878), SPH_C32(0xfb942525),
+	SPH_C32(0xcab82e2e), SPH_C32(0x54701c1c), SPH_C32(0x5faea6a6),
+	SPH_C32(0x21e6b4b4), SPH_C32(0x6435c6c6), SPH_C32(0xae8de8e8),
+	SPH_C32(0x2559dddd), SPH_C32(0x57cb7474), SPH_C32(0x5d7c1f1f),
+	SPH_C32(0xea374b4b), SPH_C32(0x1ec2bdbd), SPH_C32(0x9c1a8b8b),
+	SPH_C32(0x9b1e8a8a), SPH_C32(0x4bdb7070), SPH_C32(0xbaf83e3e),
+	SPH_C32(0x26e2b5b5), SPH_C32(0x29836666), SPH_C32(0xe33b4848),
+	SPH_C32(0x090c0303), SPH_C32(0xf4f5f6f6), SPH_C32(0x2a380e0e),
+	SPH_C32(0x3c9f6161), SPH_C32(0x8bd43535), SPH_C32(0xbe475757),
+	SPH_C32(0x02d2b9b9), SPH_C32(0xbf2e8686), SPH_C32(0x7129c1c1),
+	SPH_C32(0x53741d1d), SPH_C32(0xf74e9e9e), SPH_C32(0x91a9e1e1),
+	SPH_C32(0xdecdf8f8), SPH_C32(0xe5569898), SPH_C32(0x77441111),
+	SPH_C32(0x04bf6969), SPH_C32(0x3949d9d9), SPH_C32(0x870e8e8e),
+	SPH_C32(0xc1669494), SPH_C32(0xec5a9b9b), SPH_C32(0x5a781e1e),
+	SPH_C32(0xb82a8787), SPH_C32(0xa989e9e9), SPH_C32(0x5c15cece),
+	SPH_C32(0xb04f5555), SPH_C32(0xd8a02828), SPH_C32(0x2b51dfdf),
+	SPH_C32(0x89068c8c), SPH_C32(0x4ab2a1a1), SPH_C32(0x92128989),
+	SPH_C32(0x23340d0d), SPH_C32(0x10cabfbf), SPH_C32(0x84b5e6e6),
+	SPH_C32(0xd5134242), SPH_C32(0x03bb6868), SPH_C32(0xdc1f4141),
+	SPH_C32(0xe2529999), SPH_C32(0xc3b42d2d), SPH_C32(0x2d3c0f0f),
+	SPH_C32(0x3df6b0b0), SPH_C32(0xb74b5454), SPH_C32(0x0cdabbbb),
+	SPH_C32(0x62581616)
+};
+
+static const uint32_t mixtab3_cpu[] = {
+	SPH_C32(0x63329763), SPH_C32(0x7c6feb7c), SPH_C32(0x775ec777),
+	SPH_C32(0x7b7af77b), SPH_C32(0xf2e8e5f2), SPH_C32(0x6b0ab76b),
+	SPH_C32(0x6f16a76f), SPH_C32(0xc56d39c5), SPH_C32(0x3090c030),
+	SPH_C32(0x01070401), SPH_C32(0x672e8767), SPH_C32(0x2bd1ac2b),
+	SPH_C32(0xfeccd5fe), SPH_C32(0xd71371d7), SPH_C32(0xab7c9aab),
+	SPH_C32(0x7659c376), SPH_C32(0xca4005ca), SPH_C32(0x82a33e82),
+	SPH_C32(0xc94909c9), SPH_C32(0x7d68ef7d), SPH_C32(0xfad0c5fa),
+	SPH_C32(0x59947f59), SPH_C32(0x47ce0747), SPH_C32(0xf0e6edf0),
+	SPH_C32(0xad6e82ad), SPH_C32(0xd41a7dd4), SPH_C32(0xa243bea2),
+	SPH_C32(0xaf608aaf), SPH_C32(0x9cf9469c), SPH_C32(0xa451a6a4),
+	SPH_C32(0x7245d372), SPH_C32(0xc0762dc0), SPH_C32(0xb728eab7),
+	SPH_C32(0xfdc5d9fd), SPH_C32(0x93d47a93), SPH_C32(0x26f29826),
+	SPH_C32(0x3682d836), SPH_C32(0x3fbdfc3f), SPH_C32(0xf7f3f1f7),
+	SPH_C32(0xcc521dcc), SPH_C32(0x348cd034), SPH_C32(0xa556a2a5),
+	SPH_C32(0xe58db9e5), SPH_C32(0xf1e1e9f1), SPH_C32(0x714cdf71),
+	SPH_C32(0xd83e4dd8), SPH_C32(0x3197c431), SPH_C32(0x156b5415),
+	SPH_C32(0x041c1004), SPH_C32(0xc76331c7), SPH_C32(0x23e98c23),
+	SPH_C32(0xc37f21c3), SPH_C32(0x18486018), SPH_C32(0x96cf6e96),
+	SPH_C32(0x051b1405), SPH_C32(0x9aeb5e9a), SPH_C32(0x07151c07),
+	SPH_C32(0x127e4812), SPH_C32(0x80ad3680), SPH_C32(0xe298a5e2),
+	SPH_C32(0xeba781eb), SPH_C32(0x27f59c27), SPH_C32(0xb233feb2),
+	SPH_C32(0x7550cf75), SPH_C32(0x093f2409), SPH_C32(0x83a43a83),
+	SPH_C32(0x2cc4b02c), SPH_C32(0x1a46681a), SPH_C32(0x1b416c1b),
+	SPH_C32(0x6e11a36e), SPH_C32(0x5a9d735a), SPH_C32(0xa04db6a0),
+	SPH_C32(0x52a55352), SPH_C32(0x3ba1ec3b), SPH_C32(0xd61475d6),
+	SPH_C32(0xb334fab3), SPH_C32(0x29dfa429), SPH_C32(0xe39fa1e3),
+	SPH_C32(0x2fcdbc2f), SPH_C32(0x84b12684), SPH_C32(0x53a25753),
+	SPH_C32(0xd10169d1), SPH_C32(0x00000000), SPH_C32(0xedb599ed),
+	SPH_C32(0x20e08020), SPH_C32(0xfcc2ddfc), SPH_C32(0xb13af2b1),
+	SPH_C32(0x5b9a775b), SPH_C32(0x6a0db36a), SPH_C32(0xcb4701cb),
+	SPH_C32(0xbe17cebe), SPH_C32(0x39afe439), SPH_C32(0x4aed334a),
+	SPH_C32(0x4cff2b4c), SPH_C32(0x58937b58), SPH_C32(0xcf5b11cf),
+	SPH_C32(0xd0066dd0), SPH_C32(0xefbb91ef), SPH_C32(0xaa7b9eaa),
+	SPH_C32(0xfbd7c1fb), SPH_C32(0x43d21743), SPH_C32(0x4df82f4d),
+	SPH_C32(0x3399cc33), SPH_C32(0x85b62285), SPH_C32(0x45c00f45),
+	SPH_C32(0xf9d9c9f9), SPH_C32(0x020e0802), SPH_C32(0x7f66e77f),
+	SPH_C32(0x50ab5b50), SPH_C32(0x3cb4f03c), SPH_C32(0x9ff04a9f),
+	SPH_C32(0xa87596a8), SPH_C32(0x51ac5f51), SPH_C32(0xa344baa3),
+	SPH_C32(0x40db1b40), SPH_C32(0x8f800a8f), SPH_C32(0x92d37e92),
+	SPH_C32(0x9dfe429d), SPH_C32(0x38a8e038), SPH_C32(0xf5fdf9f5),
+	SPH_C32(0xbc19c6bc), SPH_C32(0xb62feeb6), SPH_C32(0xda3045da),
+	SPH_C32(0x21e78421), SPH_C32(0x10704010), SPH_C32(0xffcbd1ff),
+	SPH_C32(0xf3efe1f3), SPH_C32(0xd20865d2), SPH_C32(0xcd5519cd),
+	SPH_C32(0x0c24300c), SPH_C32(0x13794c13), SPH_C32(0xecb29dec),
+	SPH_C32(0x5f86675f), SPH_C32(0x97c86a97), SPH_C32(0x44c70b44),
+	SPH_C32(0x17655c17), SPH_C32(0xc46a3dc4), SPH_C32(0xa758aaa7),
+	SPH_C32(0x7e61e37e), SPH_C32(0x3db3f43d), SPH_C32(0x64278b64),
+	SPH_C32(0x5d886f5d), SPH_C32(0x194f6419), SPH_C32(0x7342d773),
+	SPH_C32(0x603b9b60), SPH_C32(0x81aa3281), SPH_C32(0x4ff6274f),
+	SPH_C32(0xdc225ddc), SPH_C32(0x22ee8822), SPH_C32(0x2ad6a82a),
+	SPH_C32(0x90dd7690), SPH_C32(0x88951688), SPH_C32(0x46c90346),
+	SPH_C32(0xeebc95ee), SPH_C32(0xb805d6b8), SPH_C32(0x146c5014),
+	SPH_C32(0xde2c55de), SPH_C32(0x5e81635e), SPH_C32(0x0b312c0b),
+	SPH_C32(0xdb3741db), SPH_C32(0xe096ade0), SPH_C32(0x329ec832),
+	SPH_C32(0x3aa6e83a), SPH_C32(0x0a36280a), SPH_C32(0x49e43f49),
+	SPH_C32(0x06121806), SPH_C32(0x24fc9024), SPH_C32(0x5c8f6b5c),
+	SPH_C32(0xc27825c2), SPH_C32(0xd30f61d3), SPH_C32(0xac6986ac),
+	SPH_C32(0x62359362), SPH_C32(0x91da7291), SPH_C32(0x95c66295),
+	SPH_C32(0xe48abde4), SPH_C32(0x7974ff79), SPH_C32(0xe783b1e7),
+	SPH_C32(0xc84e0dc8), SPH_C32(0x3785dc37), SPH_C32(0x6d18af6d),
+	SPH_C32(0x8d8e028d), SPH_C32(0xd51d79d5), SPH_C32(0x4ef1234e),
+	SPH_C32(0xa97292a9), SPH_C32(0x6c1fab6c), SPH_C32(0x56b94356),
+	SPH_C32(0xf4fafdf4), SPH_C32(0xeaa085ea), SPH_C32(0x65208f65),
+	SPH_C32(0x7a7df37a), SPH_C32(0xae678eae), SPH_C32(0x08382008),
+	SPH_C32(0xba0bdeba), SPH_C32(0x7873fb78), SPH_C32(0x25fb9425),
+	SPH_C32(0x2ecab82e), SPH_C32(0x1c54701c), SPH_C32(0xa65faea6),
+	SPH_C32(0xb421e6b4), SPH_C32(0xc66435c6), SPH_C32(0xe8ae8de8),
+	SPH_C32(0xdd2559dd), SPH_C32(0x7457cb74), SPH_C32(0x1f5d7c1f),
+	SPH_C32(0x4bea374b), SPH_C32(0xbd1ec2bd), SPH_C32(0x8b9c1a8b),
+	SPH_C32(0x8a9b1e8a), SPH_C32(0x704bdb70), SPH_C32(0x3ebaf83e),
+	SPH_C32(0xb526e2b5), SPH_C32(0x66298366), SPH_C32(0x48e33b48),
+	SPH_C32(0x03090c03), SPH_C32(0xf6f4f5f6), SPH_C32(0x0e2a380e),
+	SPH_C32(0x613c9f61), SPH_C32(0x358bd435), SPH_C32(0x57be4757),
+	SPH_C32(0xb902d2b9), SPH_C32(0x86bf2e86), SPH_C32(0xc17129c1),
+	SPH_C32(0x1d53741d), SPH_C32(0x9ef74e9e), SPH_C32(0xe191a9e1),
+	SPH_C32(0xf8decdf8), SPH_C32(0x98e55698), SPH_C32(0x11774411),
+	SPH_C32(0x6904bf69), SPH_C32(0xd93949d9), SPH_C32(0x8e870e8e),
+	SPH_C32(0x94c16694), SPH_C32(0x9bec5a9b), SPH_C32(0x1e5a781e),
+	SPH_C32(0x87b82a87), SPH_C32(0xe9a989e9), SPH_C32(0xce5c15ce),
+	SPH_C32(0x55b04f55), SPH_C32(0x28d8a028), SPH_C32(0xdf2b51df),
+	SPH_C32(0x8c89068c), SPH_C32(0xa14ab2a1), SPH_C32(0x89921289),
+	SPH_C32(0x0d23340d), SPH_C32(0xbf10cabf), SPH_C32(0xe684b5e6),
+	SPH_C32(0x42d51342), SPH_C32(0x6803bb68), SPH_C32(0x41dc1f41),
+	SPH_C32(0x99e25299), SPH_C32(0x2dc3b42d), SPH_C32(0x0f2d3c0f),
+	SPH_C32(0xb03df6b0), SPH_C32(0x54b74b54), SPH_C32(0xbb0cdabb),
+	SPH_C32(0x16625816)
+};
+
+#define TIX2(q, x00, x01, x08, x10, x24) { \
+		x10 ^= x00; \
+		x00 = (q); \
+		x08 ^= x00; \
+		x01 ^= x24; \
+	}
+
+#define TIX3(q, x00, x01, x04, x08, x16, x27, x30) { \
+		x16 ^= x00; \
+		x00 = (q); \
+		x08 ^= x00; \
+		x01 ^= x27; \
+		x04 ^= x30; \
+	}
+
+#define TIX4(q, x00, x01, x04, x07, x08, x22, x24, x27, x30) { \
+		x22 ^= x00; \
+		x00 = (q); \
+		x08 ^= x00; \
+		x01 ^= x24; \
+		x04 ^= x27; \
+		x07 ^= x30; \
+	}
+
+#define CMIX30(x00, x01, x02, x04, x05, x06, x15, x16, x17) { \
+		x00 ^= x04; \
+		x01 ^= x05; \
+		x02 ^= x06; \
+		x15 ^= x04; \
+		x16 ^= x05; \
+		x17 ^= x06; \
+	}
+
+#define CMIX36(x00, x01, x02, x04, x05, x06, x18, x19, x20) { \
+		x00 ^= x04; \
+		x01 ^= x05; \
+		x02 ^= x06; \
+		x18 ^= x04; \
+		x19 ^= x05; \
+		x20 ^= x06; \
+	}
+
+#define SMIX(x0, x1, x2, x3) { \
+		uint32_t c0 = 0; \
+		uint32_t c1 = 0; \
+		uint32_t c2 = 0; \
+		uint32_t c3 = 0; \
+		uint32_t r0 = 0; \
+		uint32_t r1 = 0; \
+		uint32_t r2 = 0; \
+		uint32_t r3 = 0; \
+		uint32_t tmp; \
+		tmp = mixtab0(x0 >> 24); \
+		c0 ^= tmp; \
+		tmp = mixtab1((x0 >> 16) & 0xFF); \
+		c0 ^= tmp; \
+		r1 ^= tmp; \
+		tmp = mixtab2((x0 >>  8) & 0xFF); \
+		c0 ^= tmp; \
+		r2 ^= tmp; \
+		tmp = mixtab3(x0 & 0xFF); \
+		c0 ^= tmp; \
+		r3 ^= tmp; \
+		tmp = mixtab0(x1 >> 24); \
+		c1 ^= tmp; \
+		r0 ^= tmp; \
+		tmp = mixtab1((x1 >> 16) & 0xFF); \
+		c1 ^= tmp; \
+		tmp = mixtab2((x1 >>  8) & 0xFF); \
+		c1 ^= tmp; \
+		r2 ^= tmp; \
+		tmp = mixtab3(x1 & 0xFF); \
+		c1 ^= tmp; \
+		r3 ^= tmp; \
+		tmp = mixtab0(x2 >> 24); \
+		c2 ^= tmp; \
+		r0 ^= tmp; \
+		tmp = mixtab1((x2 >> 16) & 0xFF); \
+		c2 ^= tmp; \
+		r1 ^= tmp; \
+		tmp = mixtab2((x2 >>  8) & 0xFF); \
+		c2 ^= tmp; \
+		tmp = mixtab3(x2 & 0xFF); \
+		c2 ^= tmp; \
+		r3 ^= tmp; \
+		tmp = mixtab0(x3 >> 24); \
+		c3 ^= tmp; \
+		r0 ^= tmp; \
+		tmp = mixtab1((x3 >> 16) & 0xFF); \
+		c3 ^= tmp; \
+		r1 ^= tmp; \
+		tmp = mixtab2((x3 >>  8) & 0xFF); \
+		c3 ^= tmp; \
+		r2 ^= tmp; \
+		tmp = mixtab3(x3 & 0xFF); \
+		c3 ^= tmp; \
+		x0 = ((c0 ^ r0) & SPH_C32(0xFF000000)) \
+			| ((c1 ^ r1) & SPH_C32(0x00FF0000)) \
+			| ((c2 ^ r2) & SPH_C32(0x0000FF00)) \
+			| ((c3 ^ r3) & SPH_C32(0x000000FF)); \
+		x1 = ((c1 ^ (r0 << 8)) & SPH_C32(0xFF000000)) \
+			| ((c2 ^ (r1 << 8)) & SPH_C32(0x00FF0000)) \
+			| ((c3 ^ (r2 << 8)) & SPH_C32(0x0000FF00)) \
+			| ((c0 ^ (r3 >> 24)) & SPH_C32(0x000000FF)); \
+		x2 = ((c2 ^ (r0 << 16)) & SPH_C32(0xFF000000)) \
+			| ((c3 ^ (r1 << 16)) & SPH_C32(0x00FF0000)) \
+			| ((c0 ^ (r2 >> 16)) & SPH_C32(0x0000FF00)) \
+			| ((c1 ^ (r3 >> 16)) & SPH_C32(0x000000FF)); \
+		x3 = ((c3 ^ (r0 << 24)) & SPH_C32(0xFF000000)) \
+			| ((c0 ^ (r1 >> 8)) & SPH_C32(0x00FF0000)) \
+			| ((c1 ^ (r2 >> 8)) & SPH_C32(0x0000FF00)) \
+			| ((c2 ^ (r3 >> 8)) & SPH_C32(0x000000FF)); \
+		/* */ \
+	}
+
+#define S00   (sc[ 0])
+#define S01   (sc[ 1])
+#define S02   (sc[ 2])
+#define S03   (sc[ 3])
+#define S04   (sc[ 4])
+#define S05   (sc[ 5])
+#define S06   (sc[ 6])
+#define S07   (sc[ 7])
+#define S08   (sc[ 8])
+#define S09   (sc[ 9])
+#define S10   (sc[10])
+#define S11   (sc[11])
+#define S12   (sc[12])
+#define S13   (sc[13])
+#define S14   (sc[14])
+#define S15   (sc[15])
+#define S16   (sc[16])
+#define S17   (sc[17])
+#define S18   (sc[18])
+#define S19   (sc[19])
+#define S20   (sc[20])
+#define S21   (sc[21])
+#define S22   (sc[22])
+#define S23   (sc[23])
+#define S24   (sc[24])
+#define S25   (sc[25])
+#define S26   (sc[26])
+#define S27   (sc[27])
+#define S28   (sc[28])
+#define S29   (sc[29])
+#define S30   (sc[30])
+#define S31   (sc[31])
+#define S32   (sc[32])
+#define S33   (sc[33])
+#define S34   (sc[34])
+#define S35   (sc[35])
+
+#define SWAB32(x)		( ((x & 0x000000FF) << 24) | ((x & 0x0000FF00) << 8) | ((x & 0x00FF0000) >> 8) | ((x & 0xFF000000) >> 24) )
+/* GPU - FUNKTIONEN */
+
+__global__ void fugue256_gpu_hash(int thr_id, int threads, uint32_t startNounce, void *outputHash, uint32_t *resNounce)
+{
+	int thread = (blockDim.x * blockIdx.x + threadIdx.x);
+	if (thread < threads)
+	{
+		/* Nimm den State und verarbeite das letztenByte (die Nounce) */
+		uint32_t sc[30];
+
+	#pragma unroll 30
+		for(int i=0;i<30;i++)
+			sc[i] = GPUstate[i];
+
+		uint32_t nounce = startNounce + thread; // muss noch ermittelt werden	
+		uint32_t q;
+
+
+		// Bei Byte 80 laufen die Teilrunden: 4-0-1 (hier fest)
+
+		// Teilrunde 4
+		q = SWAB32(nounce);
+		TIX2(q, S06, S07, S14, S16, S00);
+		CMIX30(S03, S04, S05, S07, S08, S09, S18, S19, S20);
+		SMIX(S03, S04, S05, S06);
+		CMIX30(S00, S01, S02, S04, S05, S06, S15, S16, S17);
+		SMIX(S00, S01, S02, S03);
+
+		// Teilrunde 0
+		q = 0;
+		TIX2(q, S00, S01, S08, S10, S24);
+		CMIX30(S27, S28, S29, S01, S02, S03, S12, S13, S14);
+		SMIX(S27, S28, S29, S00);
+		CMIX30(S24, S25, S26, S28, S29, S00, S09, S10, S11);
+		SMIX(S24, S25, S26, S27);
+
+		// Teilrunde 1
+		q = 0x280; // hoffentlich richtig rum...
+		TIX2(q, S24, S25, S02, S04, S18);
+		CMIX30(S21, S22, S23, S25, S26, S27, S06, S07, S08);
+		SMIX(S21, S22, S23, S24);
+		CMIX30(S18, S19, S20, S22, S23, S24, S03, S04, S05);
+		SMIX(S18, S19, S20, S21);
+
+		// Rundenende
+		// rms = 12, d.h. 30 - 12 = 18
+
+	#pragma unroll 10
+		for(int i=0;i<10;i++)
+		{
+			//ROR(3, 30);
+			uint32_t tmp[3];
+			#pragma unroll 3
+			for(int k=0;k<3;k++)
+				tmp[k] = sc[27+k];
+			#pragma unroll 27
+			for(int k=26;k>=0;k--)
+				sc[k+3] = sc[k];
+			#pragma unroll 3
+			for(int k=0;k<3;k++)
+				sc[k] = tmp[k];
+
+
+			CMIX30(sc[18], sc[19], sc[20], sc[22], sc[23], sc[24], sc[3], sc[4], sc[5]);
+			SMIX(sc[18], sc[19], sc[20], sc[21]);
+		}
+
+	#pragma unroll 13
+		for(int i=0;i<13;i++)
+		{
+			sc[22] ^= sc[18];
+			sc[3] ^= sc[18];
+
+			// ROR(15, 30); BEGIN
+			uint32_t tmp1[15];
+			#pragma unroll 15
+			for(int k=0;k<15;k++)
+				tmp1[k] = sc[15+k];
+			#pragma unroll 15
+			for(int k=14;k>=0;k--)
+				sc[k+15] = sc[k];
+			#pragma unroll 15
+			for(int k=0;k<15;k++)
+				sc[k] = tmp1[k];
+			// ROR(15, 30); END
+
+			SMIX(sc[18], sc[19], sc[20], sc[21]);
+			sc[22] ^= sc[18];
+			sc[4] ^= sc[18];
+
+			// ROR(14, 30); BEGIN
+			uint32_t tmp2[14];
+			#pragma unroll 14
+			for(int k=0;k<14;k++)
+				tmp2[k] = sc[16+k];
+			#pragma unroll 16
+			for(int k=15;k>=0;k--)
+				sc[k+14] = sc[k];
+			#pragma unroll 14
+			for(int k=0;k<14;k++)
+				sc[k] = tmp2[k];
+			// ROR(14, 30); END
+
+			SMIX(sc[18], sc[19], sc[20], sc[21]);
+		}
+
+		sc[22] ^= sc[18];
+		sc[3] ^= sc[18];
+
+		/*
+		// SWAP32 und Daten ausgeben
+	#pragma unroll 4
+		for(int i=0;i<4;i++)
+			((uint32_t*)outputHash)[8*thread+i] = SWAB32(sc[19+i]);
+
+	#pragma unroll 4
+		for(int i=0;i<4;i++)
+			((uint32_t*)outputHash)[8*thread+i+4] = SWAB32(sc[3+i]);
+		*/
+		uint32_t hash[8];
+	#pragma unroll 4
+		for(int i=0;i<4;i++)
+			((uint32_t*)hash)[i] = SWAB32(sc[19+i]);
+
+	#pragma unroll 4
+		for(int i=0;i<4;i++)
+			((uint32_t*)hash)[i+4] = SWAB32(sc[3+i]);
+
+		int i;
+		bool rc = true;
+	
+		for (i = 7; i >= 0; i--) {
+			if (hash[i] > pTarget[i]) {
+				rc = false;
+				break;
+			}
+			if (hash[i] < pTarget[i]) {
+				rc = true;
+				break;
+			}
+		}
+
+		if(rc == true)
+		{
+			if(resNounce[0] > SWAB32(nounce))
+				resNounce[0] = SWAB32(nounce);
+		}
+	}
+}
+
+#define texDef(texname, texmem, texsource, texsize) \
+	unsigned int *texmem; \
+	cudaMalloc(&texmem, texsize); \
+	cudaMemcpy(texmem, texsource, texsize, cudaMemcpyHostToDevice); \
+	texname.normalized = 0; \
+	texname.filterMode = cudaFilterModePoint; \
+	texname.addressMode[0] = cudaAddressModeClamp; \
+	{ cudaChannelFormatDesc channelDesc = cudaCreateChannelDesc<unsigned int>(); \
+	  cudaBindTexture(NULL, &texname, texmem, &channelDesc, texsize ); } \
+
+
+void fugue256_cpu_init(int thr_id, int threads)
+{
+	cudaSetDevice(thr_id);
+
+	// Kopiere die Hash-Tabellen in den GPU-Speicher
+	/*
+	cudaMemcpyToSymbol(	mixtab0,
+						mixtab0_cpu,
+						sizeof(uint32_t) * 256 );
+	cudaMemcpyToSymbol(	mixtab1,
+						mixtab1_cpu,
+						sizeof(uint32_t) * 256 );
+	cudaMemcpyToSymbol(	mixtab2,
+						mixtab2_cpu,
+						sizeof(uint32_t) * 256 );
+	cudaMemcpyToSymbol(	mixtab3,
+						mixtab3_cpu,
+						sizeof(uint32_t) * 256 );
+	*/
+	texDef(mixTab0Tex, mixTab0m, mixtab0_cpu, sizeof(uint32_t)*256);
+	texDef(mixTab1Tex, mixTab1m, mixtab1_cpu, sizeof(uint32_t)*256);
+	texDef(mixTab2Tex, mixTab2m, mixtab2_cpu, sizeof(uint32_t)*256);
+	texDef(mixTab3Tex, mixTab3m, mixtab3_cpu, sizeof(uint32_t)*256);
+	// Speicher für alle Ergebnisse belegen
+	cudaMalloc(&d_fugue256_hashoutput[thr_id], 8 * sizeof(uint32_t) * threads);
+	cudaMalloc(&d_resultNonce[thr_id], sizeof(uint32_t)); 
+}
+
+__host__ void fugue256_cpu_setBlock(int thr_id, void *data, void *pTargetIn)
+{
+	// CPU-Vorbereitungen treffen
+	sph_fugue256_context ctx_fugue_const;
+	sph_fugue256_init(&ctx_fugue_const);
+	sph_fugue256 (&ctx_fugue_const, data, 80);	// State speichern
+
+	cudaMemcpyToSymbol(	GPUstate,
+						ctx_fugue_const.S,
+						sizeof(uint32_t) * 30 );
+
+	cudaMemcpyToSymbol(	pTarget,
+						pTargetIn,
+						sizeof(uint32_t) * 8 );
+
+	cudaMemset(d_resultNonce[thr_id], 0xFF, sizeof(uint32_t));
+}
+
+__host__ void fugue256_cpu_hash(int thr_id, int threads, int startNounce, void *outputHashes, uint32_t *nounce)
+{
+	const int threadsperblock = 512;
+
+	// berechne wie viele Thread Blocks wir brauchen
+	dim3 grid((threads + threadsperblock-1)/threadsperblock);
+	dim3 block(threadsperblock);
+
+	// Größe des dynamischen Shared Memory Bereichs (abhängig von der Threadanzahl)
+	//size_t shared_size = W_ALIGNMENT*sizeof(uint32_t)*threadsperblock;  // ein uint32_t eingefügt gegen Bank Konflikte
+	size_t shared_size = 0;
+
+	fugue256_gpu_hash<<<grid, block, shared_size>>>(thr_id, threads, startNounce, d_fugue256_hashoutput[thr_id], d_resultNonce[thr_id]);
+
+	// Strategisches Sleep Kommando zur Senkung der CPU Last
+	MyStreamSynchronize(NULL, 0, thr_id);
+
+	//cudaMemcpy(outputHashes, d_fugue256_hashoutput[thr_id], 8 * sizeof(uint32_t), cudaMemcpyDeviceToHost);
+	cudaMemcpy(nounce, d_resultNonce[thr_id], sizeof(uint32_t), cudaMemcpyDeviceToHost);
+}
+
+#endif
diff --git a/cuda_fugue256.h b/cuda_fugue256.h
new file mode 100644
index 0000000000..bb864fb70b
--- /dev/null
+++ b/cuda_fugue256.h
@@ -0,0 +1,8 @@
+#ifndef _CUDA_FUGUE512_H
+#define _CUDA_FUGUE512_H
+
+void fugue256_cpu_hash(int thr_id, int threads, int startNounce, void *outputHashes, uint32_t *nounce);
+void fugue256_cpu_setBlock(int thr_id, void *data, void *pTargetIn);
+void fugue256_cpu_init(int thr_id, int threads);
+
+#endif
diff --git a/cuda_groestl512.cu b/cuda_groestl512.cu
new file mode 100644
index 0000000000..1c1dce9c41
--- /dev/null
+++ b/cuda_groestl512.cu
@@ -0,0 +1,837 @@
+/* Diese Funktion ist auf 84+32-Byte große Eingabedaten ausgerichtet (Heavycoin) */
+#include <cuda.h>
+#include "cuda_runtime.h"
+#include "device_launch_parameters.h"
+
+#include <stdio.h>
+#include <memory.h>
+
+#define USE_SHARED 0
+#define W_ALIGNMENT 65
+
+// Folgende Definitionen später durch header ersetzen
+typedef unsigned char uint8_t;
+typedef unsigned int uint32_t;
+typedef unsigned long long uint64_t;
+
+// globaler Speicher für alle HeftyHashes aller Threads
+extern uint32_t *d_heftyHashes[8];
+extern uint32_t *d_nonceVector[8];
+
+// globaler Speicher für unsere Ergebnisse
+uint32_t *d_hash4output[8];
+
+__constant__ uint32_t groestl_gpu_state[32];
+__constant__ uint32_t groestl_gpu_msg[32];
+
+#define SPH_T32(x)    ((x) & SPH_C32(0xFFFFFFFF))
+
+#define PC32up(j, r)   ((uint32_t)((j) + (r)))
+#define PC32dn(j, r)   0
+#define QC32up(j, r)   0xFFFFFFFF
+#define QC32dn(j, r)   (((uint32_t)(r) << 24) ^ SPH_T32(~((uint32_t)(j) << 24)))
+
+#define B32_0(x)    ((x) & 0xFF)
+#define B32_1(x)    (((x) >> 8) & 0xFF)
+#define B32_2(x)    (((x) >> 16) & 0xFF)
+#define B32_3(x)    ((x) >> 24)
+
+#define SPH_C32(x)	((uint32_t)(x ## U))
+#define C32e(x)     ((SPH_C32(x) >> 24) \
+                    | ((SPH_C32(x) >>  8) & SPH_C32(0x0000FF00)) \
+                    | ((SPH_C32(x) <<  8) & SPH_C32(0x00FF0000)) \
+                    | ((SPH_C32(x) << 24) & SPH_C32(0xFF000000)))
+
+#define T0up(x) tex1Dfetch(t0up, x)
+#define T0dn(x) tex1Dfetch(t0dn, x)
+#define T1up(x) tex1Dfetch(t1up, x)
+#define T1dn(x) tex1Dfetch(t1dn, x)
+#define T2up(x) tex1Dfetch(t2up, x)
+#define T2dn(x) tex1Dfetch(t2dn, x)
+#define T3up(x) tex1Dfetch(t3up, x)
+#define T3dn(x) tex1Dfetch(t3dn, x)
+
+texture<unsigned int, 1, cudaReadModeElementType> t0up;
+texture<unsigned int, 1, cudaReadModeElementType> t0dn;
+texture<unsigned int, 1, cudaReadModeElementType> t1up;
+texture<unsigned int, 1, cudaReadModeElementType> t1dn;
+texture<unsigned int, 1, cudaReadModeElementType> t2up;
+texture<unsigned int, 1, cudaReadModeElementType> t2dn;
+texture<unsigned int, 1, cudaReadModeElementType> t3up;
+texture<unsigned int, 1, cudaReadModeElementType> t3dn;
+
+static const uint32_t T0up_cpu[] = {
+	C32e(0xc632f4a5), C32e(0xf86f9784), C32e(0xee5eb099), C32e(0xf67a8c8d),
+	C32e(0xffe8170d), C32e(0xd60adcbd), C32e(0xde16c8b1), C32e(0x916dfc54),
+	C32e(0x6090f050), C32e(0x02070503), C32e(0xce2ee0a9), C32e(0x56d1877d),
+	C32e(0xe7cc2b19), C32e(0xb513a662), C32e(0x4d7c31e6), C32e(0xec59b59a),
+	C32e(0x8f40cf45), C32e(0x1fa3bc9d), C32e(0x8949c040), C32e(0xfa689287),
+	C32e(0xefd03f15), C32e(0xb29426eb), C32e(0x8ece40c9), C32e(0xfbe61d0b),
+	C32e(0x416e2fec), C32e(0xb31aa967), C32e(0x5f431cfd), C32e(0x456025ea),
+	C32e(0x23f9dabf), C32e(0x535102f7), C32e(0xe445a196), C32e(0x9b76ed5b),
+	C32e(0x75285dc2), C32e(0xe1c5241c), C32e(0x3dd4e9ae), C32e(0x4cf2be6a),
+	C32e(0x6c82ee5a), C32e(0x7ebdc341), C32e(0xf5f30602), C32e(0x8352d14f),
+	C32e(0x688ce45c), C32e(0x515607f4), C32e(0xd18d5c34), C32e(0xf9e11808),
+	C32e(0xe24cae93), C32e(0xab3e9573), C32e(0x6297f553), C32e(0x2a6b413f),
+	C32e(0x081c140c), C32e(0x9563f652), C32e(0x46e9af65), C32e(0x9d7fe25e),
+	C32e(0x30487828), C32e(0x37cff8a1), C32e(0x0a1b110f), C32e(0x2febc4b5),
+	C32e(0x0e151b09), C32e(0x247e5a36), C32e(0x1badb69b), C32e(0xdf98473d),
+	C32e(0xcda76a26), C32e(0x4ef5bb69), C32e(0x7f334ccd), C32e(0xea50ba9f),
+	C32e(0x123f2d1b), C32e(0x1da4b99e), C32e(0x58c49c74), C32e(0x3446722e),
+	C32e(0x3641772d), C32e(0xdc11cdb2), C32e(0xb49d29ee), C32e(0x5b4d16fb),
+	C32e(0xa4a501f6), C32e(0x76a1d74d), C32e(0xb714a361), C32e(0x7d3449ce),
+	C32e(0x52df8d7b), C32e(0xdd9f423e), C32e(0x5ecd9371), C32e(0x13b1a297),
+	C32e(0xa6a204f5), C32e(0xb901b868), C32e(0x00000000), C32e(0xc1b5742c),
+	C32e(0x40e0a060), C32e(0xe3c2211f), C32e(0x793a43c8), C32e(0xb69a2ced),
+	C32e(0xd40dd9be), C32e(0x8d47ca46), C32e(0x671770d9), C32e(0x72afdd4b),
+	C32e(0x94ed79de), C32e(0x98ff67d4), C32e(0xb09323e8), C32e(0x855bde4a),
+	C32e(0xbb06bd6b), C32e(0xc5bb7e2a), C32e(0x4f7b34e5), C32e(0xedd73a16),
+	C32e(0x86d254c5), C32e(0x9af862d7), C32e(0x6699ff55), C32e(0x11b6a794),
+	C32e(0x8ac04acf), C32e(0xe9d93010), C32e(0x040e0a06), C32e(0xfe669881),
+	C32e(0xa0ab0bf0), C32e(0x78b4cc44), C32e(0x25f0d5ba), C32e(0x4b753ee3),
+	C32e(0xa2ac0ef3), C32e(0x5d4419fe), C32e(0x80db5bc0), C32e(0x0580858a),
+	C32e(0x3fd3ecad), C32e(0x21fedfbc), C32e(0x70a8d848), C32e(0xf1fd0c04),
+	C32e(0x63197adf), C32e(0x772f58c1), C32e(0xaf309f75), C32e(0x42e7a563),
+	C32e(0x20705030), C32e(0xe5cb2e1a), C32e(0xfdef120e), C32e(0xbf08b76d),
+	C32e(0x8155d44c), C32e(0x18243c14), C32e(0x26795f35), C32e(0xc3b2712f),
+	C32e(0xbe8638e1), C32e(0x35c8fda2), C32e(0x88c74fcc), C32e(0x2e654b39),
+	C32e(0x936af957), C32e(0x55580df2), C32e(0xfc619d82), C32e(0x7ab3c947),
+	C32e(0xc827efac), C32e(0xba8832e7), C32e(0x324f7d2b), C32e(0xe642a495),
+	C32e(0xc03bfba0), C32e(0x19aab398), C32e(0x9ef668d1), C32e(0xa322817f),
+	C32e(0x44eeaa66), C32e(0x54d6827e), C32e(0x3bdde6ab), C32e(0x0b959e83),
+	C32e(0x8cc945ca), C32e(0xc7bc7b29), C32e(0x6b056ed3), C32e(0x286c443c),
+	C32e(0xa72c8b79), C32e(0xbc813de2), C32e(0x1631271d), C32e(0xad379a76),
+	C32e(0xdb964d3b), C32e(0x649efa56), C32e(0x74a6d24e), C32e(0x1436221e),
+	C32e(0x92e476db), C32e(0x0c121e0a), C32e(0x48fcb46c), C32e(0xb88f37e4),
+	C32e(0x9f78e75d), C32e(0xbd0fb26e), C32e(0x43692aef), C32e(0xc435f1a6),
+	C32e(0x39dae3a8), C32e(0x31c6f7a4), C32e(0xd38a5937), C32e(0xf274868b),
+	C32e(0xd5835632), C32e(0x8b4ec543), C32e(0x6e85eb59), C32e(0xda18c2b7),
+	C32e(0x018e8f8c), C32e(0xb11dac64), C32e(0x9cf16dd2), C32e(0x49723be0),
+	C32e(0xd81fc7b4), C32e(0xacb915fa), C32e(0xf3fa0907), C32e(0xcfa06f25),
+	C32e(0xca20eaaf), C32e(0xf47d898e), C32e(0x476720e9), C32e(0x10382818),
+	C32e(0x6f0b64d5), C32e(0xf0738388), C32e(0x4afbb16f), C32e(0x5cca9672),
+	C32e(0x38546c24), C32e(0x575f08f1), C32e(0x732152c7), C32e(0x9764f351),
+	C32e(0xcbae6523), C32e(0xa125847c), C32e(0xe857bf9c), C32e(0x3e5d6321),
+	C32e(0x96ea7cdd), C32e(0x611e7fdc), C32e(0x0d9c9186), C32e(0x0f9b9485),
+	C32e(0xe04bab90), C32e(0x7cbac642), C32e(0x712657c4), C32e(0xcc29e5aa),
+	C32e(0x90e373d8), C32e(0x06090f05), C32e(0xf7f40301), C32e(0x1c2a3612),
+	C32e(0xc23cfea3), C32e(0x6a8be15f), C32e(0xaebe10f9), C32e(0x69026bd0),
+	C32e(0x17bfa891), C32e(0x9971e858), C32e(0x3a536927), C32e(0x27f7d0b9),
+	C32e(0xd9914838), C32e(0xebde3513), C32e(0x2be5ceb3), C32e(0x22775533),
+	C32e(0xd204d6bb), C32e(0xa9399070), C32e(0x07878089), C32e(0x33c1f2a7),
+	C32e(0x2decc1b6), C32e(0x3c5a6622), C32e(0x15b8ad92), C32e(0xc9a96020),
+	C32e(0x875cdb49), C32e(0xaab01aff), C32e(0x50d88878), C32e(0xa52b8e7a),
+	C32e(0x03898a8f), C32e(0x594a13f8), C32e(0x09929b80), C32e(0x1a233917),
+	C32e(0x651075da), C32e(0xd7845331), C32e(0x84d551c6), C32e(0xd003d3b8),
+	C32e(0x82dc5ec3), C32e(0x29e2cbb0), C32e(0x5ac39977), C32e(0x1e2d3311),
+	C32e(0x7b3d46cb), C32e(0xa8b71ffc), C32e(0x6d0c61d6), C32e(0x2c624e3a)
+};
+
+static const uint32_t T0dn_cpu[] = {
+	C32e(0xf497a5c6), C32e(0x97eb84f8), C32e(0xb0c799ee), C32e(0x8cf78df6),
+	C32e(0x17e50dff), C32e(0xdcb7bdd6), C32e(0xc8a7b1de), C32e(0xfc395491),
+	C32e(0xf0c05060), C32e(0x05040302), C32e(0xe087a9ce), C32e(0x87ac7d56),
+	C32e(0x2bd519e7), C32e(0xa67162b5), C32e(0x319ae64d), C32e(0xb5c39aec),
+	C32e(0xcf05458f), C32e(0xbc3e9d1f), C32e(0xc0094089), C32e(0x92ef87fa),
+	C32e(0x3fc515ef), C32e(0x267febb2), C32e(0x4007c98e), C32e(0x1ded0bfb),
+	C32e(0x2f82ec41), C32e(0xa97d67b3), C32e(0x1cbefd5f), C32e(0x258aea45),
+	C32e(0xda46bf23), C32e(0x02a6f753), C32e(0xa1d396e4), C32e(0xed2d5b9b),
+	C32e(0x5deac275), C32e(0x24d91ce1), C32e(0xe97aae3d), C32e(0xbe986a4c),
+	C32e(0xeed85a6c), C32e(0xc3fc417e), C32e(0x06f102f5), C32e(0xd11d4f83),
+	C32e(0xe4d05c68), C32e(0x07a2f451), C32e(0x5cb934d1), C32e(0x18e908f9),
+	C32e(0xaedf93e2), C32e(0x954d73ab), C32e(0xf5c45362), C32e(0x41543f2a),
+	C32e(0x14100c08), C32e(0xf6315295), C32e(0xaf8c6546), C32e(0xe2215e9d),
+	C32e(0x78602830), C32e(0xf86ea137), C32e(0x11140f0a), C32e(0xc45eb52f),
+	C32e(0x1b1c090e), C32e(0x5a483624), C32e(0xb6369b1b), C32e(0x47a53ddf),
+	C32e(0x6a8126cd), C32e(0xbb9c694e), C32e(0x4cfecd7f), C32e(0xbacf9fea),
+	C32e(0x2d241b12), C32e(0xb93a9e1d), C32e(0x9cb07458), C32e(0x72682e34),
+	C32e(0x776c2d36), C32e(0xcda3b2dc), C32e(0x2973eeb4), C32e(0x16b6fb5b),
+	C32e(0x0153f6a4), C32e(0xd7ec4d76), C32e(0xa37561b7), C32e(0x49face7d),
+	C32e(0x8da47b52), C32e(0x42a13edd), C32e(0x93bc715e), C32e(0xa2269713),
+	C32e(0x0457f5a6), C32e(0xb86968b9), C32e(0x00000000), C32e(0x74992cc1),
+	C32e(0xa0806040), C32e(0x21dd1fe3), C32e(0x43f2c879), C32e(0x2c77edb6),
+	C32e(0xd9b3bed4), C32e(0xca01468d), C32e(0x70ced967), C32e(0xdde44b72),
+	C32e(0x7933de94), C32e(0x672bd498), C32e(0x237be8b0), C32e(0xde114a85),
+	C32e(0xbd6d6bbb), C32e(0x7e912ac5), C32e(0x349ee54f), C32e(0x3ac116ed),
+	C32e(0x5417c586), C32e(0x622fd79a), C32e(0xffcc5566), C32e(0xa7229411),
+	C32e(0x4a0fcf8a), C32e(0x30c910e9), C32e(0x0a080604), C32e(0x98e781fe),
+	C32e(0x0b5bf0a0), C32e(0xccf04478), C32e(0xd54aba25), C32e(0x3e96e34b),
+	C32e(0x0e5ff3a2), C32e(0x19bafe5d), C32e(0x5b1bc080), C32e(0x850a8a05),
+	C32e(0xec7ead3f), C32e(0xdf42bc21), C32e(0xd8e04870), C32e(0x0cf904f1),
+	C32e(0x7ac6df63), C32e(0x58eec177), C32e(0x9f4575af), C32e(0xa5846342),
+	C32e(0x50403020), C32e(0x2ed11ae5), C32e(0x12e10efd), C32e(0xb7656dbf),
+	C32e(0xd4194c81), C32e(0x3c301418), C32e(0x5f4c3526), C32e(0x719d2fc3),
+	C32e(0x3867e1be), C32e(0xfd6aa235), C32e(0x4f0bcc88), C32e(0x4b5c392e),
+	C32e(0xf93d5793), C32e(0x0daaf255), C32e(0x9de382fc), C32e(0xc9f4477a),
+	C32e(0xef8bacc8), C32e(0x326fe7ba), C32e(0x7d642b32), C32e(0xa4d795e6),
+	C32e(0xfb9ba0c0), C32e(0xb3329819), C32e(0x6827d19e), C32e(0x815d7fa3),
+	C32e(0xaa886644), C32e(0x82a87e54), C32e(0xe676ab3b), C32e(0x9e16830b),
+	C32e(0x4503ca8c), C32e(0x7b9529c7), C32e(0x6ed6d36b), C32e(0x44503c28),
+	C32e(0x8b5579a7), C32e(0x3d63e2bc), C32e(0x272c1d16), C32e(0x9a4176ad),
+	C32e(0x4dad3bdb), C32e(0xfac85664), C32e(0xd2e84e74), C32e(0x22281e14),
+	C32e(0x763fdb92), C32e(0x1e180a0c), C32e(0xb4906c48), C32e(0x376be4b8),
+	C32e(0xe7255d9f), C32e(0xb2616ebd), C32e(0x2a86ef43), C32e(0xf193a6c4),
+	C32e(0xe372a839), C32e(0xf762a431), C32e(0x59bd37d3), C32e(0x86ff8bf2),
+	C32e(0x56b132d5), C32e(0xc50d438b), C32e(0xebdc596e), C32e(0xc2afb7da),
+	C32e(0x8f028c01), C32e(0xac7964b1), C32e(0x6d23d29c), C32e(0x3b92e049),
+	C32e(0xc7abb4d8), C32e(0x1543faac), C32e(0x09fd07f3), C32e(0x6f8525cf),
+	C32e(0xea8fafca), C32e(0x89f38ef4), C32e(0x208ee947), C32e(0x28201810),
+	C32e(0x64ded56f), C32e(0x83fb88f0), C32e(0xb1946f4a), C32e(0x96b8725c),
+	C32e(0x6c702438), C32e(0x08aef157), C32e(0x52e6c773), C32e(0xf3355197),
+	C32e(0x658d23cb), C32e(0x84597ca1), C32e(0xbfcb9ce8), C32e(0x637c213e),
+	C32e(0x7c37dd96), C32e(0x7fc2dc61), C32e(0x911a860d), C32e(0x941e850f),
+	C32e(0xabdb90e0), C32e(0xc6f8427c), C32e(0x57e2c471), C32e(0xe583aacc),
+	C32e(0x733bd890), C32e(0x0f0c0506), C32e(0x03f501f7), C32e(0x3638121c),
+	C32e(0xfe9fa3c2), C32e(0xe1d45f6a), C32e(0x1047f9ae), C32e(0x6bd2d069),
+	C32e(0xa82e9117), C32e(0xe8295899), C32e(0x6974273a), C32e(0xd04eb927),
+	C32e(0x48a938d9), C32e(0x35cd13eb), C32e(0xce56b32b), C32e(0x55443322),
+	C32e(0xd6bfbbd2), C32e(0x904970a9), C32e(0x800e8907), C32e(0xf266a733),
+	C32e(0xc15ab62d), C32e(0x6678223c), C32e(0xad2a9215), C32e(0x608920c9),
+	C32e(0xdb154987), C32e(0x1a4fffaa), C32e(0x88a07850), C32e(0x8e517aa5),
+	C32e(0x8a068f03), C32e(0x13b2f859), C32e(0x9b128009), C32e(0x3934171a),
+	C32e(0x75cada65), C32e(0x53b531d7), C32e(0x5113c684), C32e(0xd3bbb8d0),
+	C32e(0x5e1fc382), C32e(0xcb52b029), C32e(0x99b4775a), C32e(0x333c111e),
+	C32e(0x46f6cb7b), C32e(0x1f4bfca8), C32e(0x61dad66d), C32e(0x4e583a2c)
+};
+
+static const uint32_t T1up_cpu[] = {
+	C32e(0xc6c632f4), C32e(0xf8f86f97), C32e(0xeeee5eb0), C32e(0xf6f67a8c),
+	C32e(0xffffe817), C32e(0xd6d60adc), C32e(0xdede16c8), C32e(0x91916dfc),
+	C32e(0x606090f0), C32e(0x02020705), C32e(0xcece2ee0), C32e(0x5656d187),
+	C32e(0xe7e7cc2b), C32e(0xb5b513a6), C32e(0x4d4d7c31), C32e(0xecec59b5),
+	C32e(0x8f8f40cf), C32e(0x1f1fa3bc), C32e(0x898949c0), C32e(0xfafa6892),
+	C32e(0xefefd03f), C32e(0xb2b29426), C32e(0x8e8ece40), C32e(0xfbfbe61d),
+	C32e(0x41416e2f), C32e(0xb3b31aa9), C32e(0x5f5f431c), C32e(0x45456025),
+	C32e(0x2323f9da), C32e(0x53535102), C32e(0xe4e445a1), C32e(0x9b9b76ed),
+	C32e(0x7575285d), C32e(0xe1e1c524), C32e(0x3d3dd4e9), C32e(0x4c4cf2be),
+	C32e(0x6c6c82ee), C32e(0x7e7ebdc3), C32e(0xf5f5f306), C32e(0x838352d1),
+	C32e(0x68688ce4), C32e(0x51515607), C32e(0xd1d18d5c), C32e(0xf9f9e118),
+	C32e(0xe2e24cae), C32e(0xabab3e95), C32e(0x626297f5), C32e(0x2a2a6b41),
+	C32e(0x08081c14), C32e(0x959563f6), C32e(0x4646e9af), C32e(0x9d9d7fe2),
+	C32e(0x30304878), C32e(0x3737cff8), C32e(0x0a0a1b11), C32e(0x2f2febc4),
+	C32e(0x0e0e151b), C32e(0x24247e5a), C32e(0x1b1badb6), C32e(0xdfdf9847),
+	C32e(0xcdcda76a), C32e(0x4e4ef5bb), C32e(0x7f7f334c), C32e(0xeaea50ba),
+	C32e(0x12123f2d), C32e(0x1d1da4b9), C32e(0x5858c49c), C32e(0x34344672),
+	C32e(0x36364177), C32e(0xdcdc11cd), C32e(0xb4b49d29), C32e(0x5b5b4d16),
+	C32e(0xa4a4a501), C32e(0x7676a1d7), C32e(0xb7b714a3), C32e(0x7d7d3449),
+	C32e(0x5252df8d), C32e(0xdddd9f42), C32e(0x5e5ecd93), C32e(0x1313b1a2),
+	C32e(0xa6a6a204), C32e(0xb9b901b8), C32e(0x00000000), C32e(0xc1c1b574),
+	C32e(0x4040e0a0), C32e(0xe3e3c221), C32e(0x79793a43), C32e(0xb6b69a2c),
+	C32e(0xd4d40dd9), C32e(0x8d8d47ca), C32e(0x67671770), C32e(0x7272afdd),
+	C32e(0x9494ed79), C32e(0x9898ff67), C32e(0xb0b09323), C32e(0x85855bde),
+	C32e(0xbbbb06bd), C32e(0xc5c5bb7e), C32e(0x4f4f7b34), C32e(0xededd73a),
+	C32e(0x8686d254), C32e(0x9a9af862), C32e(0x666699ff), C32e(0x1111b6a7),
+	C32e(0x8a8ac04a), C32e(0xe9e9d930), C32e(0x04040e0a), C32e(0xfefe6698),
+	C32e(0xa0a0ab0b), C32e(0x7878b4cc), C32e(0x2525f0d5), C32e(0x4b4b753e),
+	C32e(0xa2a2ac0e), C32e(0x5d5d4419), C32e(0x8080db5b), C32e(0x05058085),
+	C32e(0x3f3fd3ec), C32e(0x2121fedf), C32e(0x7070a8d8), C32e(0xf1f1fd0c),
+	C32e(0x6363197a), C32e(0x77772f58), C32e(0xafaf309f), C32e(0x4242e7a5),
+	C32e(0x20207050), C32e(0xe5e5cb2e), C32e(0xfdfdef12), C32e(0xbfbf08b7),
+	C32e(0x818155d4), C32e(0x1818243c), C32e(0x2626795f), C32e(0xc3c3b271),
+	C32e(0xbebe8638), C32e(0x3535c8fd), C32e(0x8888c74f), C32e(0x2e2e654b),
+	C32e(0x93936af9), C32e(0x5555580d), C32e(0xfcfc619d), C32e(0x7a7ab3c9),
+	C32e(0xc8c827ef), C32e(0xbaba8832), C32e(0x32324f7d), C32e(0xe6e642a4),
+	C32e(0xc0c03bfb), C32e(0x1919aab3), C32e(0x9e9ef668), C32e(0xa3a32281),
+	C32e(0x4444eeaa), C32e(0x5454d682), C32e(0x3b3bdde6), C32e(0x0b0b959e),
+	C32e(0x8c8cc945), C32e(0xc7c7bc7b), C32e(0x6b6b056e), C32e(0x28286c44),
+	C32e(0xa7a72c8b), C32e(0xbcbc813d), C32e(0x16163127), C32e(0xadad379a),
+	C32e(0xdbdb964d), C32e(0x64649efa), C32e(0x7474a6d2), C32e(0x14143622),
+	C32e(0x9292e476), C32e(0x0c0c121e), C32e(0x4848fcb4), C32e(0xb8b88f37),
+	C32e(0x9f9f78e7), C32e(0xbdbd0fb2), C32e(0x4343692a), C32e(0xc4c435f1),
+	C32e(0x3939dae3), C32e(0x3131c6f7), C32e(0xd3d38a59), C32e(0xf2f27486),
+	C32e(0xd5d58356), C32e(0x8b8b4ec5), C32e(0x6e6e85eb), C32e(0xdada18c2),
+	C32e(0x01018e8f), C32e(0xb1b11dac), C32e(0x9c9cf16d), C32e(0x4949723b),
+	C32e(0xd8d81fc7), C32e(0xacacb915), C32e(0xf3f3fa09), C32e(0xcfcfa06f),
+	C32e(0xcaca20ea), C32e(0xf4f47d89), C32e(0x47476720), C32e(0x10103828),
+	C32e(0x6f6f0b64), C32e(0xf0f07383), C32e(0x4a4afbb1), C32e(0x5c5cca96),
+	C32e(0x3838546c), C32e(0x57575f08), C32e(0x73732152), C32e(0x979764f3),
+	C32e(0xcbcbae65), C32e(0xa1a12584), C32e(0xe8e857bf), C32e(0x3e3e5d63),
+	C32e(0x9696ea7c), C32e(0x61611e7f), C32e(0x0d0d9c91), C32e(0x0f0f9b94),
+	C32e(0xe0e04bab), C32e(0x7c7cbac6), C32e(0x71712657), C32e(0xcccc29e5),
+	C32e(0x9090e373), C32e(0x0606090f), C32e(0xf7f7f403), C32e(0x1c1c2a36),
+	C32e(0xc2c23cfe), C32e(0x6a6a8be1), C32e(0xaeaebe10), C32e(0x6969026b),
+	C32e(0x1717bfa8), C32e(0x999971e8), C32e(0x3a3a5369), C32e(0x2727f7d0),
+	C32e(0xd9d99148), C32e(0xebebde35), C32e(0x2b2be5ce), C32e(0x22227755),
+	C32e(0xd2d204d6), C32e(0xa9a93990), C32e(0x07078780), C32e(0x3333c1f2),
+	C32e(0x2d2decc1), C32e(0x3c3c5a66), C32e(0x1515b8ad), C32e(0xc9c9a960),
+	C32e(0x87875cdb), C32e(0xaaaab01a), C32e(0x5050d888), C32e(0xa5a52b8e),
+	C32e(0x0303898a), C32e(0x59594a13), C32e(0x0909929b), C32e(0x1a1a2339),
+	C32e(0x65651075), C32e(0xd7d78453), C32e(0x8484d551), C32e(0xd0d003d3),
+	C32e(0x8282dc5e), C32e(0x2929e2cb), C32e(0x5a5ac399), C32e(0x1e1e2d33),
+	C32e(0x7b7b3d46), C32e(0xa8a8b71f), C32e(0x6d6d0c61), C32e(0x2c2c624e)
+};
+
+static const uint32_t T1dn_cpu[] = {
+	C32e(0xa5f497a5), C32e(0x8497eb84), C32e(0x99b0c799), C32e(0x8d8cf78d),
+	C32e(0x0d17e50d), C32e(0xbddcb7bd), C32e(0xb1c8a7b1), C32e(0x54fc3954),
+	C32e(0x50f0c050), C32e(0x03050403), C32e(0xa9e087a9), C32e(0x7d87ac7d),
+	C32e(0x192bd519), C32e(0x62a67162), C32e(0xe6319ae6), C32e(0x9ab5c39a),
+	C32e(0x45cf0545), C32e(0x9dbc3e9d), C32e(0x40c00940), C32e(0x8792ef87),
+	C32e(0x153fc515), C32e(0xeb267feb), C32e(0xc94007c9), C32e(0x0b1ded0b),
+	C32e(0xec2f82ec), C32e(0x67a97d67), C32e(0xfd1cbefd), C32e(0xea258aea),
+	C32e(0xbfda46bf), C32e(0xf702a6f7), C32e(0x96a1d396), C32e(0x5bed2d5b),
+	C32e(0xc25deac2), C32e(0x1c24d91c), C32e(0xaee97aae), C32e(0x6abe986a),
+	C32e(0x5aeed85a), C32e(0x41c3fc41), C32e(0x0206f102), C32e(0x4fd11d4f),
+	C32e(0x5ce4d05c), C32e(0xf407a2f4), C32e(0x345cb934), C32e(0x0818e908),
+	C32e(0x93aedf93), C32e(0x73954d73), C32e(0x53f5c453), C32e(0x3f41543f),
+	C32e(0x0c14100c), C32e(0x52f63152), C32e(0x65af8c65), C32e(0x5ee2215e),
+	C32e(0x28786028), C32e(0xa1f86ea1), C32e(0x0f11140f), C32e(0xb5c45eb5),
+	C32e(0x091b1c09), C32e(0x365a4836), C32e(0x9bb6369b), C32e(0x3d47a53d),
+	C32e(0x266a8126), C32e(0x69bb9c69), C32e(0xcd4cfecd), C32e(0x9fbacf9f),
+	C32e(0x1b2d241b), C32e(0x9eb93a9e), C32e(0x749cb074), C32e(0x2e72682e),
+	C32e(0x2d776c2d), C32e(0xb2cda3b2), C32e(0xee2973ee), C32e(0xfb16b6fb),
+	C32e(0xf60153f6), C32e(0x4dd7ec4d), C32e(0x61a37561), C32e(0xce49face),
+	C32e(0x7b8da47b), C32e(0x3e42a13e), C32e(0x7193bc71), C32e(0x97a22697),
+	C32e(0xf50457f5), C32e(0x68b86968), C32e(0x00000000), C32e(0x2c74992c),
+	C32e(0x60a08060), C32e(0x1f21dd1f), C32e(0xc843f2c8), C32e(0xed2c77ed),
+	C32e(0xbed9b3be), C32e(0x46ca0146), C32e(0xd970ced9), C32e(0x4bdde44b),
+	C32e(0xde7933de), C32e(0xd4672bd4), C32e(0xe8237be8), C32e(0x4ade114a),
+	C32e(0x6bbd6d6b), C32e(0x2a7e912a), C32e(0xe5349ee5), C32e(0x163ac116),
+	C32e(0xc55417c5), C32e(0xd7622fd7), C32e(0x55ffcc55), C32e(0x94a72294),
+	C32e(0xcf4a0fcf), C32e(0x1030c910), C32e(0x060a0806), C32e(0x8198e781),
+	C32e(0xf00b5bf0), C32e(0x44ccf044), C32e(0xbad54aba), C32e(0xe33e96e3),
+	C32e(0xf30e5ff3), C32e(0xfe19bafe), C32e(0xc05b1bc0), C32e(0x8a850a8a),
+	C32e(0xadec7ead), C32e(0xbcdf42bc), C32e(0x48d8e048), C32e(0x040cf904),
+	C32e(0xdf7ac6df), C32e(0xc158eec1), C32e(0x759f4575), C32e(0x63a58463),
+	C32e(0x30504030), C32e(0x1a2ed11a), C32e(0x0e12e10e), C32e(0x6db7656d),
+	C32e(0x4cd4194c), C32e(0x143c3014), C32e(0x355f4c35), C32e(0x2f719d2f),
+	C32e(0xe13867e1), C32e(0xa2fd6aa2), C32e(0xcc4f0bcc), C32e(0x394b5c39),
+	C32e(0x57f93d57), C32e(0xf20daaf2), C32e(0x829de382), C32e(0x47c9f447),
+	C32e(0xacef8bac), C32e(0xe7326fe7), C32e(0x2b7d642b), C32e(0x95a4d795),
+	C32e(0xa0fb9ba0), C32e(0x98b33298), C32e(0xd16827d1), C32e(0x7f815d7f),
+	C32e(0x66aa8866), C32e(0x7e82a87e), C32e(0xabe676ab), C32e(0x839e1683),
+	C32e(0xca4503ca), C32e(0x297b9529), C32e(0xd36ed6d3), C32e(0x3c44503c),
+	C32e(0x798b5579), C32e(0xe23d63e2), C32e(0x1d272c1d), C32e(0x769a4176),
+	C32e(0x3b4dad3b), C32e(0x56fac856), C32e(0x4ed2e84e), C32e(0x1e22281e),
+	C32e(0xdb763fdb), C32e(0x0a1e180a), C32e(0x6cb4906c), C32e(0xe4376be4),
+	C32e(0x5de7255d), C32e(0x6eb2616e), C32e(0xef2a86ef), C32e(0xa6f193a6),
+	C32e(0xa8e372a8), C32e(0xa4f762a4), C32e(0x3759bd37), C32e(0x8b86ff8b),
+	C32e(0x3256b132), C32e(0x43c50d43), C32e(0x59ebdc59), C32e(0xb7c2afb7),
+	C32e(0x8c8f028c), C32e(0x64ac7964), C32e(0xd26d23d2), C32e(0xe03b92e0),
+	C32e(0xb4c7abb4), C32e(0xfa1543fa), C32e(0x0709fd07), C32e(0x256f8525),
+	C32e(0xafea8faf), C32e(0x8e89f38e), C32e(0xe9208ee9), C32e(0x18282018),
+	C32e(0xd564ded5), C32e(0x8883fb88), C32e(0x6fb1946f), C32e(0x7296b872),
+	C32e(0x246c7024), C32e(0xf108aef1), C32e(0xc752e6c7), C32e(0x51f33551),
+	C32e(0x23658d23), C32e(0x7c84597c), C32e(0x9cbfcb9c), C32e(0x21637c21),
+	C32e(0xdd7c37dd), C32e(0xdc7fc2dc), C32e(0x86911a86), C32e(0x85941e85),
+	C32e(0x90abdb90), C32e(0x42c6f842), C32e(0xc457e2c4), C32e(0xaae583aa),
+	C32e(0xd8733bd8), C32e(0x050f0c05), C32e(0x0103f501), C32e(0x12363812),
+	C32e(0xa3fe9fa3), C32e(0x5fe1d45f), C32e(0xf91047f9), C32e(0xd06bd2d0),
+	C32e(0x91a82e91), C32e(0x58e82958), C32e(0x27697427), C32e(0xb9d04eb9),
+	C32e(0x3848a938), C32e(0x1335cd13), C32e(0xb3ce56b3), C32e(0x33554433),
+	C32e(0xbbd6bfbb), C32e(0x70904970), C32e(0x89800e89), C32e(0xa7f266a7),
+	C32e(0xb6c15ab6), C32e(0x22667822), C32e(0x92ad2a92), C32e(0x20608920),
+	C32e(0x49db1549), C32e(0xff1a4fff), C32e(0x7888a078), C32e(0x7a8e517a),
+	C32e(0x8f8a068f), C32e(0xf813b2f8), C32e(0x809b1280), C32e(0x17393417),
+	C32e(0xda75cada), C32e(0x3153b531), C32e(0xc65113c6), C32e(0xb8d3bbb8),
+	C32e(0xc35e1fc3), C32e(0xb0cb52b0), C32e(0x7799b477), C32e(0x11333c11),
+	C32e(0xcb46f6cb), C32e(0xfc1f4bfc), C32e(0xd661dad6), C32e(0x3a4e583a)
+};
+
+static const uint32_t T2up_cpu[] = {
+	C32e(0xa5c6c632), C32e(0x84f8f86f), C32e(0x99eeee5e), C32e(0x8df6f67a),
+	C32e(0x0dffffe8), C32e(0xbdd6d60a), C32e(0xb1dede16), C32e(0x5491916d),
+	C32e(0x50606090), C32e(0x03020207), C32e(0xa9cece2e), C32e(0x7d5656d1),
+	C32e(0x19e7e7cc), C32e(0x62b5b513), C32e(0xe64d4d7c), C32e(0x9aecec59),
+	C32e(0x458f8f40), C32e(0x9d1f1fa3), C32e(0x40898949), C32e(0x87fafa68),
+	C32e(0x15efefd0), C32e(0xebb2b294), C32e(0xc98e8ece), C32e(0x0bfbfbe6),
+	C32e(0xec41416e), C32e(0x67b3b31a), C32e(0xfd5f5f43), C32e(0xea454560),
+	C32e(0xbf2323f9), C32e(0xf7535351), C32e(0x96e4e445), C32e(0x5b9b9b76),
+	C32e(0xc2757528), C32e(0x1ce1e1c5), C32e(0xae3d3dd4), C32e(0x6a4c4cf2),
+	C32e(0x5a6c6c82), C32e(0x417e7ebd), C32e(0x02f5f5f3), C32e(0x4f838352),
+	C32e(0x5c68688c), C32e(0xf4515156), C32e(0x34d1d18d), C32e(0x08f9f9e1),
+	C32e(0x93e2e24c), C32e(0x73abab3e), C32e(0x53626297), C32e(0x3f2a2a6b),
+	C32e(0x0c08081c), C32e(0x52959563), C32e(0x654646e9), C32e(0x5e9d9d7f),
+	C32e(0x28303048), C32e(0xa13737cf), C32e(0x0f0a0a1b), C32e(0xb52f2feb),
+	C32e(0x090e0e15), C32e(0x3624247e), C32e(0x9b1b1bad), C32e(0x3ddfdf98),
+	C32e(0x26cdcda7), C32e(0x694e4ef5), C32e(0xcd7f7f33), C32e(0x9feaea50),
+	C32e(0x1b12123f), C32e(0x9e1d1da4), C32e(0x745858c4), C32e(0x2e343446),
+	C32e(0x2d363641), C32e(0xb2dcdc11), C32e(0xeeb4b49d), C32e(0xfb5b5b4d),
+	C32e(0xf6a4a4a5), C32e(0x4d7676a1), C32e(0x61b7b714), C32e(0xce7d7d34),
+	C32e(0x7b5252df), C32e(0x3edddd9f), C32e(0x715e5ecd), C32e(0x971313b1),
+	C32e(0xf5a6a6a2), C32e(0x68b9b901), C32e(0x00000000), C32e(0x2cc1c1b5),
+	C32e(0x604040e0), C32e(0x1fe3e3c2), C32e(0xc879793a), C32e(0xedb6b69a),
+	C32e(0xbed4d40d), C32e(0x468d8d47), C32e(0xd9676717), C32e(0x4b7272af),
+	C32e(0xde9494ed), C32e(0xd49898ff), C32e(0xe8b0b093), C32e(0x4a85855b),
+	C32e(0x6bbbbb06), C32e(0x2ac5c5bb), C32e(0xe54f4f7b), C32e(0x16ededd7),
+	C32e(0xc58686d2), C32e(0xd79a9af8), C32e(0x55666699), C32e(0x941111b6),
+	C32e(0xcf8a8ac0), C32e(0x10e9e9d9), C32e(0x0604040e), C32e(0x81fefe66),
+	C32e(0xf0a0a0ab), C32e(0x447878b4), C32e(0xba2525f0), C32e(0xe34b4b75),
+	C32e(0xf3a2a2ac), C32e(0xfe5d5d44), C32e(0xc08080db), C32e(0x8a050580),
+	C32e(0xad3f3fd3), C32e(0xbc2121fe), C32e(0x487070a8), C32e(0x04f1f1fd),
+	C32e(0xdf636319), C32e(0xc177772f), C32e(0x75afaf30), C32e(0x634242e7),
+	C32e(0x30202070), C32e(0x1ae5e5cb), C32e(0x0efdfdef), C32e(0x6dbfbf08),
+	C32e(0x4c818155), C32e(0x14181824), C32e(0x35262679), C32e(0x2fc3c3b2),
+	C32e(0xe1bebe86), C32e(0xa23535c8), C32e(0xcc8888c7), C32e(0x392e2e65),
+	C32e(0x5793936a), C32e(0xf2555558), C32e(0x82fcfc61), C32e(0x477a7ab3),
+	C32e(0xacc8c827), C32e(0xe7baba88), C32e(0x2b32324f), C32e(0x95e6e642),
+	C32e(0xa0c0c03b), C32e(0x981919aa), C32e(0xd19e9ef6), C32e(0x7fa3a322),
+	C32e(0x664444ee), C32e(0x7e5454d6), C32e(0xab3b3bdd), C32e(0x830b0b95),
+	C32e(0xca8c8cc9), C32e(0x29c7c7bc), C32e(0xd36b6b05), C32e(0x3c28286c),
+	C32e(0x79a7a72c), C32e(0xe2bcbc81), C32e(0x1d161631), C32e(0x76adad37),
+	C32e(0x3bdbdb96), C32e(0x5664649e), C32e(0x4e7474a6), C32e(0x1e141436),
+	C32e(0xdb9292e4), C32e(0x0a0c0c12), C32e(0x6c4848fc), C32e(0xe4b8b88f),
+	C32e(0x5d9f9f78), C32e(0x6ebdbd0f), C32e(0xef434369), C32e(0xa6c4c435),
+	C32e(0xa83939da), C32e(0xa43131c6), C32e(0x37d3d38a), C32e(0x8bf2f274),
+	C32e(0x32d5d583), C32e(0x438b8b4e), C32e(0x596e6e85), C32e(0xb7dada18),
+	C32e(0x8c01018e), C32e(0x64b1b11d), C32e(0xd29c9cf1), C32e(0xe0494972),
+	C32e(0xb4d8d81f), C32e(0xfaacacb9), C32e(0x07f3f3fa), C32e(0x25cfcfa0),
+	C32e(0xafcaca20), C32e(0x8ef4f47d), C32e(0xe9474767), C32e(0x18101038),
+	C32e(0xd56f6f0b), C32e(0x88f0f073), C32e(0x6f4a4afb), C32e(0x725c5cca),
+	C32e(0x24383854), C32e(0xf157575f), C32e(0xc7737321), C32e(0x51979764),
+	C32e(0x23cbcbae), C32e(0x7ca1a125), C32e(0x9ce8e857), C32e(0x213e3e5d),
+	C32e(0xdd9696ea), C32e(0xdc61611e), C32e(0x860d0d9c), C32e(0x850f0f9b),
+	C32e(0x90e0e04b), C32e(0x427c7cba), C32e(0xc4717126), C32e(0xaacccc29),
+	C32e(0xd89090e3), C32e(0x05060609), C32e(0x01f7f7f4), C32e(0x121c1c2a),
+	C32e(0xa3c2c23c), C32e(0x5f6a6a8b), C32e(0xf9aeaebe), C32e(0xd0696902),
+	C32e(0x911717bf), C32e(0x58999971), C32e(0x273a3a53), C32e(0xb92727f7),
+	C32e(0x38d9d991), C32e(0x13ebebde), C32e(0xb32b2be5), C32e(0x33222277),
+	C32e(0xbbd2d204), C32e(0x70a9a939), C32e(0x89070787), C32e(0xa73333c1),
+	C32e(0xb62d2dec), C32e(0x223c3c5a), C32e(0x921515b8), C32e(0x20c9c9a9),
+	C32e(0x4987875c), C32e(0xffaaaab0), C32e(0x785050d8), C32e(0x7aa5a52b),
+	C32e(0x8f030389), C32e(0xf859594a), C32e(0x80090992), C32e(0x171a1a23),
+	C32e(0xda656510), C32e(0x31d7d784), C32e(0xc68484d5), C32e(0xb8d0d003),
+	C32e(0xc38282dc), C32e(0xb02929e2), C32e(0x775a5ac3), C32e(0x111e1e2d),
+	C32e(0xcb7b7b3d), C32e(0xfca8a8b7), C32e(0xd66d6d0c), C32e(0x3a2c2c62)
+};
+
+static const uint32_t T2dn_cpu[] = {
+	C32e(0xf4a5f497), C32e(0x978497eb), C32e(0xb099b0c7), C32e(0x8c8d8cf7),
+	C32e(0x170d17e5), C32e(0xdcbddcb7), C32e(0xc8b1c8a7), C32e(0xfc54fc39),
+	C32e(0xf050f0c0), C32e(0x05030504), C32e(0xe0a9e087), C32e(0x877d87ac),
+	C32e(0x2b192bd5), C32e(0xa662a671), C32e(0x31e6319a), C32e(0xb59ab5c3),
+	C32e(0xcf45cf05), C32e(0xbc9dbc3e), C32e(0xc040c009), C32e(0x928792ef),
+	C32e(0x3f153fc5), C32e(0x26eb267f), C32e(0x40c94007), C32e(0x1d0b1ded),
+	C32e(0x2fec2f82), C32e(0xa967a97d), C32e(0x1cfd1cbe), C32e(0x25ea258a),
+	C32e(0xdabfda46), C32e(0x02f702a6), C32e(0xa196a1d3), C32e(0xed5bed2d),
+	C32e(0x5dc25dea), C32e(0x241c24d9), C32e(0xe9aee97a), C32e(0xbe6abe98),
+	C32e(0xee5aeed8), C32e(0xc341c3fc), C32e(0x060206f1), C32e(0xd14fd11d),
+	C32e(0xe45ce4d0), C32e(0x07f407a2), C32e(0x5c345cb9), C32e(0x180818e9),
+	C32e(0xae93aedf), C32e(0x9573954d), C32e(0xf553f5c4), C32e(0x413f4154),
+	C32e(0x140c1410), C32e(0xf652f631), C32e(0xaf65af8c), C32e(0xe25ee221),
+	C32e(0x78287860), C32e(0xf8a1f86e), C32e(0x110f1114), C32e(0xc4b5c45e),
+	C32e(0x1b091b1c), C32e(0x5a365a48), C32e(0xb69bb636), C32e(0x473d47a5),
+	C32e(0x6a266a81), C32e(0xbb69bb9c), C32e(0x4ccd4cfe), C32e(0xba9fbacf),
+	C32e(0x2d1b2d24), C32e(0xb99eb93a), C32e(0x9c749cb0), C32e(0x722e7268),
+	C32e(0x772d776c), C32e(0xcdb2cda3), C32e(0x29ee2973), C32e(0x16fb16b6),
+	C32e(0x01f60153), C32e(0xd74dd7ec), C32e(0xa361a375), C32e(0x49ce49fa),
+	C32e(0x8d7b8da4), C32e(0x423e42a1), C32e(0x937193bc), C32e(0xa297a226),
+	C32e(0x04f50457), C32e(0xb868b869), C32e(0x00000000), C32e(0x742c7499),
+	C32e(0xa060a080), C32e(0x211f21dd), C32e(0x43c843f2), C32e(0x2ced2c77),
+	C32e(0xd9bed9b3), C32e(0xca46ca01), C32e(0x70d970ce), C32e(0xdd4bdde4),
+	C32e(0x79de7933), C32e(0x67d4672b), C32e(0x23e8237b), C32e(0xde4ade11),
+	C32e(0xbd6bbd6d), C32e(0x7e2a7e91), C32e(0x34e5349e), C32e(0x3a163ac1),
+	C32e(0x54c55417), C32e(0x62d7622f), C32e(0xff55ffcc), C32e(0xa794a722),
+	C32e(0x4acf4a0f), C32e(0x301030c9), C32e(0x0a060a08), C32e(0x988198e7),
+	C32e(0x0bf00b5b), C32e(0xcc44ccf0), C32e(0xd5bad54a), C32e(0x3ee33e96),
+	C32e(0x0ef30e5f), C32e(0x19fe19ba), C32e(0x5bc05b1b), C32e(0x858a850a),
+	C32e(0xecadec7e), C32e(0xdfbcdf42), C32e(0xd848d8e0), C32e(0x0c040cf9),
+	C32e(0x7adf7ac6), C32e(0x58c158ee), C32e(0x9f759f45), C32e(0xa563a584),
+	C32e(0x50305040), C32e(0x2e1a2ed1), C32e(0x120e12e1), C32e(0xb76db765),
+	C32e(0xd44cd419), C32e(0x3c143c30), C32e(0x5f355f4c), C32e(0x712f719d),
+	C32e(0x38e13867), C32e(0xfda2fd6a), C32e(0x4fcc4f0b), C32e(0x4b394b5c),
+	C32e(0xf957f93d), C32e(0x0df20daa), C32e(0x9d829de3), C32e(0xc947c9f4),
+	C32e(0xefacef8b), C32e(0x32e7326f), C32e(0x7d2b7d64), C32e(0xa495a4d7),
+	C32e(0xfba0fb9b), C32e(0xb398b332), C32e(0x68d16827), C32e(0x817f815d),
+	C32e(0xaa66aa88), C32e(0x827e82a8), C32e(0xe6abe676), C32e(0x9e839e16),
+	C32e(0x45ca4503), C32e(0x7b297b95), C32e(0x6ed36ed6), C32e(0x443c4450),
+	C32e(0x8b798b55), C32e(0x3de23d63), C32e(0x271d272c), C32e(0x9a769a41),
+	C32e(0x4d3b4dad), C32e(0xfa56fac8), C32e(0xd24ed2e8), C32e(0x221e2228),
+	C32e(0x76db763f), C32e(0x1e0a1e18), C32e(0xb46cb490), C32e(0x37e4376b),
+	C32e(0xe75de725), C32e(0xb26eb261), C32e(0x2aef2a86), C32e(0xf1a6f193),
+	C32e(0xe3a8e372), C32e(0xf7a4f762), C32e(0x593759bd), C32e(0x868b86ff),
+	C32e(0x563256b1), C32e(0xc543c50d), C32e(0xeb59ebdc), C32e(0xc2b7c2af),
+	C32e(0x8f8c8f02), C32e(0xac64ac79), C32e(0x6dd26d23), C32e(0x3be03b92),
+	C32e(0xc7b4c7ab), C32e(0x15fa1543), C32e(0x090709fd), C32e(0x6f256f85),
+	C32e(0xeaafea8f), C32e(0x898e89f3), C32e(0x20e9208e), C32e(0x28182820),
+	C32e(0x64d564de), C32e(0x838883fb), C32e(0xb16fb194), C32e(0x967296b8),
+	C32e(0x6c246c70), C32e(0x08f108ae), C32e(0x52c752e6), C32e(0xf351f335),
+	C32e(0x6523658d), C32e(0x847c8459), C32e(0xbf9cbfcb), C32e(0x6321637c),
+	C32e(0x7cdd7c37), C32e(0x7fdc7fc2), C32e(0x9186911a), C32e(0x9485941e),
+	C32e(0xab90abdb), C32e(0xc642c6f8), C32e(0x57c457e2), C32e(0xe5aae583),
+	C32e(0x73d8733b), C32e(0x0f050f0c), C32e(0x030103f5), C32e(0x36123638),
+	C32e(0xfea3fe9f), C32e(0xe15fe1d4), C32e(0x10f91047), C32e(0x6bd06bd2),
+	C32e(0xa891a82e), C32e(0xe858e829), C32e(0x69276974), C32e(0xd0b9d04e),
+	C32e(0x483848a9), C32e(0x351335cd), C32e(0xceb3ce56), C32e(0x55335544),
+	C32e(0xd6bbd6bf), C32e(0x90709049), C32e(0x8089800e), C32e(0xf2a7f266),
+	C32e(0xc1b6c15a), C32e(0x66226678), C32e(0xad92ad2a), C32e(0x60206089),
+	C32e(0xdb49db15), C32e(0x1aff1a4f), C32e(0x887888a0), C32e(0x8e7a8e51),
+	C32e(0x8a8f8a06), C32e(0x13f813b2), C32e(0x9b809b12), C32e(0x39173934),
+	C32e(0x75da75ca), C32e(0x533153b5), C32e(0x51c65113), C32e(0xd3b8d3bb),
+	C32e(0x5ec35e1f), C32e(0xcbb0cb52), C32e(0x997799b4), C32e(0x3311333c),
+	C32e(0x46cb46f6), C32e(0x1ffc1f4b), C32e(0x61d661da), C32e(0x4e3a4e58)
+};
+
+static const uint32_t T3up_cpu[] = {
+	C32e(0x97a5c6c6), C32e(0xeb84f8f8), C32e(0xc799eeee), C32e(0xf78df6f6),
+	C32e(0xe50dffff), C32e(0xb7bdd6d6), C32e(0xa7b1dede), C32e(0x39549191),
+	C32e(0xc0506060), C32e(0x04030202), C32e(0x87a9cece), C32e(0xac7d5656),
+	C32e(0xd519e7e7), C32e(0x7162b5b5), C32e(0x9ae64d4d), C32e(0xc39aecec),
+	C32e(0x05458f8f), C32e(0x3e9d1f1f), C32e(0x09408989), C32e(0xef87fafa),
+	C32e(0xc515efef), C32e(0x7febb2b2), C32e(0x07c98e8e), C32e(0xed0bfbfb),
+	C32e(0x82ec4141), C32e(0x7d67b3b3), C32e(0xbefd5f5f), C32e(0x8aea4545),
+	C32e(0x46bf2323), C32e(0xa6f75353), C32e(0xd396e4e4), C32e(0x2d5b9b9b),
+	C32e(0xeac27575), C32e(0xd91ce1e1), C32e(0x7aae3d3d), C32e(0x986a4c4c),
+	C32e(0xd85a6c6c), C32e(0xfc417e7e), C32e(0xf102f5f5), C32e(0x1d4f8383),
+	C32e(0xd05c6868), C32e(0xa2f45151), C32e(0xb934d1d1), C32e(0xe908f9f9),
+	C32e(0xdf93e2e2), C32e(0x4d73abab), C32e(0xc4536262), C32e(0x543f2a2a),
+	C32e(0x100c0808), C32e(0x31529595), C32e(0x8c654646), C32e(0x215e9d9d),
+	C32e(0x60283030), C32e(0x6ea13737), C32e(0x140f0a0a), C32e(0x5eb52f2f),
+	C32e(0x1c090e0e), C32e(0x48362424), C32e(0x369b1b1b), C32e(0xa53ddfdf),
+	C32e(0x8126cdcd), C32e(0x9c694e4e), C32e(0xfecd7f7f), C32e(0xcf9feaea),
+	C32e(0x241b1212), C32e(0x3a9e1d1d), C32e(0xb0745858), C32e(0x682e3434),
+	C32e(0x6c2d3636), C32e(0xa3b2dcdc), C32e(0x73eeb4b4), C32e(0xb6fb5b5b),
+	C32e(0x53f6a4a4), C32e(0xec4d7676), C32e(0x7561b7b7), C32e(0xface7d7d),
+	C32e(0xa47b5252), C32e(0xa13edddd), C32e(0xbc715e5e), C32e(0x26971313),
+	C32e(0x57f5a6a6), C32e(0x6968b9b9), C32e(0x00000000), C32e(0x992cc1c1),
+	C32e(0x80604040), C32e(0xdd1fe3e3), C32e(0xf2c87979), C32e(0x77edb6b6),
+	C32e(0xb3bed4d4), C32e(0x01468d8d), C32e(0xced96767), C32e(0xe44b7272),
+	C32e(0x33de9494), C32e(0x2bd49898), C32e(0x7be8b0b0), C32e(0x114a8585),
+	C32e(0x6d6bbbbb), C32e(0x912ac5c5), C32e(0x9ee54f4f), C32e(0xc116eded),
+	C32e(0x17c58686), C32e(0x2fd79a9a), C32e(0xcc556666), C32e(0x22941111),
+	C32e(0x0fcf8a8a), C32e(0xc910e9e9), C32e(0x08060404), C32e(0xe781fefe),
+	C32e(0x5bf0a0a0), C32e(0xf0447878), C32e(0x4aba2525), C32e(0x96e34b4b),
+	C32e(0x5ff3a2a2), C32e(0xbafe5d5d), C32e(0x1bc08080), C32e(0x0a8a0505),
+	C32e(0x7ead3f3f), C32e(0x42bc2121), C32e(0xe0487070), C32e(0xf904f1f1),
+	C32e(0xc6df6363), C32e(0xeec17777), C32e(0x4575afaf), C32e(0x84634242),
+	C32e(0x40302020), C32e(0xd11ae5e5), C32e(0xe10efdfd), C32e(0x656dbfbf),
+	C32e(0x194c8181), C32e(0x30141818), C32e(0x4c352626), C32e(0x9d2fc3c3),
+	C32e(0x67e1bebe), C32e(0x6aa23535), C32e(0x0bcc8888), C32e(0x5c392e2e),
+	C32e(0x3d579393), C32e(0xaaf25555), C32e(0xe382fcfc), C32e(0xf4477a7a),
+	C32e(0x8bacc8c8), C32e(0x6fe7baba), C32e(0x642b3232), C32e(0xd795e6e6),
+	C32e(0x9ba0c0c0), C32e(0x32981919), C32e(0x27d19e9e), C32e(0x5d7fa3a3),
+	C32e(0x88664444), C32e(0xa87e5454), C32e(0x76ab3b3b), C32e(0x16830b0b),
+	C32e(0x03ca8c8c), C32e(0x9529c7c7), C32e(0xd6d36b6b), C32e(0x503c2828),
+	C32e(0x5579a7a7), C32e(0x63e2bcbc), C32e(0x2c1d1616), C32e(0x4176adad),
+	C32e(0xad3bdbdb), C32e(0xc8566464), C32e(0xe84e7474), C32e(0x281e1414),
+	C32e(0x3fdb9292), C32e(0x180a0c0c), C32e(0x906c4848), C32e(0x6be4b8b8),
+	C32e(0x255d9f9f), C32e(0x616ebdbd), C32e(0x86ef4343), C32e(0x93a6c4c4),
+	C32e(0x72a83939), C32e(0x62a43131), C32e(0xbd37d3d3), C32e(0xff8bf2f2),
+	C32e(0xb132d5d5), C32e(0x0d438b8b), C32e(0xdc596e6e), C32e(0xafb7dada),
+	C32e(0x028c0101), C32e(0x7964b1b1), C32e(0x23d29c9c), C32e(0x92e04949),
+	C32e(0xabb4d8d8), C32e(0x43faacac), C32e(0xfd07f3f3), C32e(0x8525cfcf),
+	C32e(0x8fafcaca), C32e(0xf38ef4f4), C32e(0x8ee94747), C32e(0x20181010),
+	C32e(0xded56f6f), C32e(0xfb88f0f0), C32e(0x946f4a4a), C32e(0xb8725c5c),
+	C32e(0x70243838), C32e(0xaef15757), C32e(0xe6c77373), C32e(0x35519797),
+	C32e(0x8d23cbcb), C32e(0x597ca1a1), C32e(0xcb9ce8e8), C32e(0x7c213e3e),
+	C32e(0x37dd9696), C32e(0xc2dc6161), C32e(0x1a860d0d), C32e(0x1e850f0f),
+	C32e(0xdb90e0e0), C32e(0xf8427c7c), C32e(0xe2c47171), C32e(0x83aacccc),
+	C32e(0x3bd89090), C32e(0x0c050606), C32e(0xf501f7f7), C32e(0x38121c1c),
+	C32e(0x9fa3c2c2), C32e(0xd45f6a6a), C32e(0x47f9aeae), C32e(0xd2d06969),
+	C32e(0x2e911717), C32e(0x29589999), C32e(0x74273a3a), C32e(0x4eb92727),
+	C32e(0xa938d9d9), C32e(0xcd13ebeb), C32e(0x56b32b2b), C32e(0x44332222),
+	C32e(0xbfbbd2d2), C32e(0x4970a9a9), C32e(0x0e890707), C32e(0x66a73333),
+	C32e(0x5ab62d2d), C32e(0x78223c3c), C32e(0x2a921515), C32e(0x8920c9c9),
+	C32e(0x15498787), C32e(0x4fffaaaa), C32e(0xa0785050), C32e(0x517aa5a5),
+	C32e(0x068f0303), C32e(0xb2f85959), C32e(0x12800909), C32e(0x34171a1a),
+	C32e(0xcada6565), C32e(0xb531d7d7), C32e(0x13c68484), C32e(0xbbb8d0d0),
+	C32e(0x1fc38282), C32e(0x52b02929), C32e(0xb4775a5a), C32e(0x3c111e1e),
+	C32e(0xf6cb7b7b), C32e(0x4bfca8a8), C32e(0xdad66d6d), C32e(0x583a2c2c)
+};
+
+static const uint32_t T3dn_cpu[] = {
+	C32e(0x32f4a5f4), C32e(0x6f978497), C32e(0x5eb099b0), C32e(0x7a8c8d8c),
+	C32e(0xe8170d17), C32e(0x0adcbddc), C32e(0x16c8b1c8), C32e(0x6dfc54fc),
+	C32e(0x90f050f0), C32e(0x07050305), C32e(0x2ee0a9e0), C32e(0xd1877d87),
+	C32e(0xcc2b192b), C32e(0x13a662a6), C32e(0x7c31e631), C32e(0x59b59ab5),
+	C32e(0x40cf45cf), C32e(0xa3bc9dbc), C32e(0x49c040c0), C32e(0x68928792),
+	C32e(0xd03f153f), C32e(0x9426eb26), C32e(0xce40c940), C32e(0xe61d0b1d),
+	C32e(0x6e2fec2f), C32e(0x1aa967a9), C32e(0x431cfd1c), C32e(0x6025ea25),
+	C32e(0xf9dabfda), C32e(0x5102f702), C32e(0x45a196a1), C32e(0x76ed5bed),
+	C32e(0x285dc25d), C32e(0xc5241c24), C32e(0xd4e9aee9), C32e(0xf2be6abe),
+	C32e(0x82ee5aee), C32e(0xbdc341c3), C32e(0xf3060206), C32e(0x52d14fd1),
+	C32e(0x8ce45ce4), C32e(0x5607f407), C32e(0x8d5c345c), C32e(0xe1180818),
+	C32e(0x4cae93ae), C32e(0x3e957395), C32e(0x97f553f5), C32e(0x6b413f41),
+	C32e(0x1c140c14), C32e(0x63f652f6), C32e(0xe9af65af), C32e(0x7fe25ee2),
+	C32e(0x48782878), C32e(0xcff8a1f8), C32e(0x1b110f11), C32e(0xebc4b5c4),
+	C32e(0x151b091b), C32e(0x7e5a365a), C32e(0xadb69bb6), C32e(0x98473d47),
+	C32e(0xa76a266a), C32e(0xf5bb69bb), C32e(0x334ccd4c), C32e(0x50ba9fba),
+	C32e(0x3f2d1b2d), C32e(0xa4b99eb9), C32e(0xc49c749c), C32e(0x46722e72),
+	C32e(0x41772d77), C32e(0x11cdb2cd), C32e(0x9d29ee29), C32e(0x4d16fb16),
+	C32e(0xa501f601), C32e(0xa1d74dd7), C32e(0x14a361a3), C32e(0x3449ce49),
+	C32e(0xdf8d7b8d), C32e(0x9f423e42), C32e(0xcd937193), C32e(0xb1a297a2),
+	C32e(0xa204f504), C32e(0x01b868b8), C32e(0x00000000), C32e(0xb5742c74),
+	C32e(0xe0a060a0), C32e(0xc2211f21), C32e(0x3a43c843), C32e(0x9a2ced2c),
+	C32e(0x0dd9bed9), C32e(0x47ca46ca), C32e(0x1770d970), C32e(0xafdd4bdd),
+	C32e(0xed79de79), C32e(0xff67d467), C32e(0x9323e823), C32e(0x5bde4ade),
+	C32e(0x06bd6bbd), C32e(0xbb7e2a7e), C32e(0x7b34e534), C32e(0xd73a163a),
+	C32e(0xd254c554), C32e(0xf862d762), C32e(0x99ff55ff), C32e(0xb6a794a7),
+	C32e(0xc04acf4a), C32e(0xd9301030), C32e(0x0e0a060a), C32e(0x66988198),
+	C32e(0xab0bf00b), C32e(0xb4cc44cc), C32e(0xf0d5bad5), C32e(0x753ee33e),
+	C32e(0xac0ef30e), C32e(0x4419fe19), C32e(0xdb5bc05b), C32e(0x80858a85),
+	C32e(0xd3ecadec), C32e(0xfedfbcdf), C32e(0xa8d848d8), C32e(0xfd0c040c),
+	C32e(0x197adf7a), C32e(0x2f58c158), C32e(0x309f759f), C32e(0xe7a563a5),
+	C32e(0x70503050), C32e(0xcb2e1a2e), C32e(0xef120e12), C32e(0x08b76db7),
+	C32e(0x55d44cd4), C32e(0x243c143c), C32e(0x795f355f), C32e(0xb2712f71),
+	C32e(0x8638e138), C32e(0xc8fda2fd), C32e(0xc74fcc4f), C32e(0x654b394b),
+	C32e(0x6af957f9), C32e(0x580df20d), C32e(0x619d829d), C32e(0xb3c947c9),
+	C32e(0x27efacef), C32e(0x8832e732), C32e(0x4f7d2b7d), C32e(0x42a495a4),
+	C32e(0x3bfba0fb), C32e(0xaab398b3), C32e(0xf668d168), C32e(0x22817f81),
+	C32e(0xeeaa66aa), C32e(0xd6827e82), C32e(0xdde6abe6), C32e(0x959e839e),
+	C32e(0xc945ca45), C32e(0xbc7b297b), C32e(0x056ed36e), C32e(0x6c443c44),
+	C32e(0x2c8b798b), C32e(0x813de23d), C32e(0x31271d27), C32e(0x379a769a),
+	C32e(0x964d3b4d), C32e(0x9efa56fa), C32e(0xa6d24ed2), C32e(0x36221e22),
+	C32e(0xe476db76), C32e(0x121e0a1e), C32e(0xfcb46cb4), C32e(0x8f37e437),
+	C32e(0x78e75de7), C32e(0x0fb26eb2), C32e(0x692aef2a), C32e(0x35f1a6f1),
+	C32e(0xdae3a8e3), C32e(0xc6f7a4f7), C32e(0x8a593759), C32e(0x74868b86),
+	C32e(0x83563256), C32e(0x4ec543c5), C32e(0x85eb59eb), C32e(0x18c2b7c2),
+	C32e(0x8e8f8c8f), C32e(0x1dac64ac), C32e(0xf16dd26d), C32e(0x723be03b),
+	C32e(0x1fc7b4c7), C32e(0xb915fa15), C32e(0xfa090709), C32e(0xa06f256f),
+	C32e(0x20eaafea), C32e(0x7d898e89), C32e(0x6720e920), C32e(0x38281828),
+	C32e(0x0b64d564), C32e(0x73838883), C32e(0xfbb16fb1), C32e(0xca967296),
+	C32e(0x546c246c), C32e(0x5f08f108), C32e(0x2152c752), C32e(0x64f351f3),
+	C32e(0xae652365), C32e(0x25847c84), C32e(0x57bf9cbf), C32e(0x5d632163),
+	C32e(0xea7cdd7c), C32e(0x1e7fdc7f), C32e(0x9c918691), C32e(0x9b948594),
+	C32e(0x4bab90ab), C32e(0xbac642c6), C32e(0x2657c457), C32e(0x29e5aae5),
+	C32e(0xe373d873), C32e(0x090f050f), C32e(0xf4030103), C32e(0x2a361236),
+	C32e(0x3cfea3fe), C32e(0x8be15fe1), C32e(0xbe10f910), C32e(0x026bd06b),
+	C32e(0xbfa891a8), C32e(0x71e858e8), C32e(0x53692769), C32e(0xf7d0b9d0),
+	C32e(0x91483848), C32e(0xde351335), C32e(0xe5ceb3ce), C32e(0x77553355),
+	C32e(0x04d6bbd6), C32e(0x39907090), C32e(0x87808980), C32e(0xc1f2a7f2),
+	C32e(0xecc1b6c1), C32e(0x5a662266), C32e(0xb8ad92ad), C32e(0xa9602060),
+	C32e(0x5cdb49db), C32e(0xb01aff1a), C32e(0xd8887888), C32e(0x2b8e7a8e),
+	C32e(0x898a8f8a), C32e(0x4a13f813), C32e(0x929b809b), C32e(0x23391739),
+	C32e(0x1075da75), C32e(0x84533153), C32e(0xd551c651), C32e(0x03d3b8d3),
+	C32e(0xdc5ec35e), C32e(0xe2cbb0cb), C32e(0xc3997799), C32e(0x2d331133),
+	C32e(0x3d46cb46), C32e(0xb71ffc1f), C32e(0x0c61d661), C32e(0x624e3a4e)
+};
+
+__device__ void groestl512_perm_P(uint32_t *a)
+{
+	uint32_t t[32];
+
+//#pragma unroll 14
+	for(int r=0;r<14;r++)
+	{
+#pragma unroll 16
+		for(int k=0;k<16;k++)
+		{
+			a[(k*2)+0] ^= PC32up(k * 0x10, r);
+			//a[(k<<1)+1] ^= PC32dn(k * 0x10, r);
+		}
+
+		// RBTT
+#pragma unroll 16
+		for(int k=0;k<32;k+=2)
+		{
+			t[k + 0] =	T0up( B32_0(a[k & 0x1f]) ) ^ 
+						T1up( B32_1(a[(k + 2) & 0x1f]) ) ^ 
+						T2up( B32_2(a[(k + 4) & 0x1f]) ) ^ 
+						T3up( B32_3(a[(k + 6) & 0x1f]) ) ^ 
+						T0dn( B32_0(a[(k + 9) & 0x1f]) ) ^ 
+						T1dn( B32_1(a[(k + 11) & 0x1f]) ) ^ 
+						T2dn( B32_2(a[(k + 13) & 0x1f]) ) ^ 
+						T3dn( B32_3(a[(k + 23) & 0x1f]) );
+
+			t[k + 1] =	T0dn( B32_0(a[k & 0x1f]) ) ^ 
+						T1dn( B32_1(a[(k + 2) & 0x1f]) ) ^ 
+						T2dn( B32_2(a[(k + 4) & 0x1f]) ) ^ 
+						T3dn( B32_3(a[(k + 6) & 0x1f]) ) ^ 
+						T0up( B32_0(a[(k + 9) & 0x1f]) ) ^ 
+						T1up( B32_1(a[(k + 11) & 0x1f]) ) ^ 
+						T2up( B32_2(a[(k + 13) & 0x1f]) ) ^ 
+						T3up( B32_3(a[(k + 23) & 0x1f]) );
+		}
+#pragma unroll 32
+		for(int k=0;k<32;k++)
+			a[k] = t[k];
+	}
+}
+
+__device__ void groestl512_perm_Q(uint32_t *a)
+{
+//#pragma unroll 14
+	for(int r=0;r<14;r++)
+	{
+		uint32_t t[32];
+
+#pragma unroll 16
+		for(int k=0;k<16;k++)
+		{
+			a[(k*2)+0] ^= QC32up(k * 0x10, r);
+			a[(k*2)+1] ^= QC32dn(k * 0x10, r);
+		}
+
+		// RBTT
+#pragma unroll 16
+		for(int k=0;k<32;k+=2)
+		{
+			t[k + 0] =	T0up( B32_0(a[(k + 2) & 0x1f]) ) ^ 
+						T1up( B32_1(a[(k + 6) & 0x1f]) ) ^ 
+						T2up( B32_2(a[(k + 10) & 0x1f]) ) ^ 
+						T3up( B32_3(a[(k + 22) & 0x1f]) ) ^ 
+						T0dn( B32_0(a[(k + 1) & 0x1f]) ) ^ 
+						T1dn( B32_1(a[(k + 5) & 0x1f]) ) ^ 
+						T2dn( B32_2(a[(k + 9) & 0x1f]) ) ^ 
+						T3dn( B32_3(a[(k + 13) & 0x1f]) );
+
+			t[k + 1] =	T0dn( B32_0(a[(k + 2) & 0x1f]) ) ^ 
+						T1dn( B32_1(a[(k + 6) & 0x1f]) ) ^ 
+						T2dn( B32_2(a[(k + 10) & 0x1f]) ) ^ 
+						T3dn( B32_3(a[(k + 22) & 0x1f]) ) ^ 
+						T0up( B32_0(a[(k + 1) & 0x1f]) ) ^ 
+						T1up( B32_1(a[(k + 5) & 0x1f]) ) ^ 
+						T2up( B32_2(a[(k + 9) & 0x1f]) ) ^ 
+						T3up( B32_3(a[(k + 13) & 0x1f]) );
+		}
+#pragma unroll 32
+		for(int k=0;k<32;k++)
+			a[k] = t[k];
+	}
+}
+
+__global__ void groestl512_gpu_hash(int threads, uint32_t startNounce, void *outputHash, uint32_t *heftyHashes, uint32_t *nonceVector)
+{
+	int thread = (blockDim.x * blockIdx.x + threadIdx.x);
+	if (thread < threads)
+	{
+#if USE_SHARED
+		extern __shared__ unsigned char s[];
+		uint32_t offset = W_ALIGNMENT * sizeof(uint32_t) * threadIdx.x;
+		uint32_t *message = (uint32_t*)(&s[offset + 0]); // 128 Byte
+		uint32_t *state = (uint32_t*)(&s[offset + 128]); // 128 Byte
+#else
+		uint32_t message[32];
+		uint32_t state[32];
+#endif
+
+		// lese message ein & verknüpfe diese mit dem hash1 von hefty1
+		// lese den state ein
+
+#pragma unroll 32
+		for(int k=0;k<32;k++)
+		{
+			state[k] = groestl_gpu_state[k];
+			message[k] = groestl_gpu_msg[k];
+		}
+
+		uint32_t nounce = nonceVector[thread];
+		// nounce setzen
+		//message[19] = startNounce + thread;
+		message[19] = nounce;
+
+		uint32_t hashPosition = nounce - startNounce;
+
+		// den richtigen Hefty1 Hash holen
+//			memcpy(&message[21], &heftyHashes[8 * hashPosition], sizeof(uint32_t) * 8);
+		uint32_t *heftyHash = &heftyHashes[8 * hashPosition];
+#pragma unroll 8
+		for (int k=0; k<8; ++k)
+			message[21+k] = heftyHash[k];
+		
+		uint32_t g[32];
+#pragma unroll 32
+		for(int u=0;u<32;u++)
+			g[u] = message[u] ^ state[u];
+
+		// Perm
+		groestl512_perm_P(g);
+		groestl512_perm_Q(message);
+		
+#pragma unroll 32
+		for(int u=0;u<32;u++)
+		{
+			state[u] ^= g[u] ^ message[u];
+			g[u] = state[u];
+		}
+
+		groestl512_perm_P(g);
+
+#pragma unroll 32
+		for(int u=0;u<32;u++)
+			state[u] ^= g[u];
+
+		// kopiere Ergebnis
+#pragma unroll 16
+		for(int k=0;k<16;k++)
+			((uint32_t*)outputHash)[16*hashPosition+k] = state[k + 16];
+	}
+}
+
+#define texDef(texname, texmem, texsource, texsize) \
+	unsigned int *texmem; \
+	cudaMalloc(&texmem, texsize); \
+	cudaMemcpy(texmem, texsource, texsize, cudaMemcpyHostToDevice); \
+	texname.normalized = 0; \
+	texname.filterMode = cudaFilterModePoint; \
+	texname.addressMode[0] = cudaAddressModeClamp; \
+	{ cudaChannelFormatDesc channelDesc = cudaCreateChannelDesc<unsigned int>(); \
+	  cudaBindTexture(NULL, &texname, texmem, &channelDesc, texsize ); } \
+
+// Setup-Funktionen
+__host__ void groestl512_cpu_init(int thr_id, int threads)
+{
+	// Texturen mit obigem Makro initialisieren
+	texDef(t0up, d_T0up, T0up_cpu, sizeof(uint32_t)*256);
+	texDef(t0dn, d_T0dn, T0dn_cpu, sizeof(uint32_t)*256);
+	texDef(t1up, d_T1up, T1up_cpu, sizeof(uint32_t)*256);
+	texDef(t1dn, d_T1dn, T1dn_cpu, sizeof(uint32_t)*256);
+	texDef(t2up, d_T2up, T2up_cpu, sizeof(uint32_t)*256);
+	texDef(t2dn, d_T2dn, T2dn_cpu, sizeof(uint32_t)*256);
+	texDef(t3up, d_T3up, T3up_cpu, sizeof(uint32_t)*256);
+	texDef(t3dn, d_T3dn, T3dn_cpu, sizeof(uint32_t)*256);
+
+	// Speicher für alle Ergebnisse belegen
+	cudaMalloc(&d_hash4output[thr_id], 16 * sizeof(uint32_t) * threads);
+}
+
+__host__ void groestl512_cpu_setBlock(void *data)
+	// data muss 84-Byte haben!
+	// heftyHash hat 32-Byte
+{
+	// Nachricht expandieren und setzen
+	uint32_t msgBlock[32];
+
+	memset(msgBlock, 0, sizeof(uint32_t) * 32);
+	memcpy(&msgBlock[0], data, 84);
+
+	// Erweitere die Nachricht auf den Nachrichtenblock (padding)
+	// Unsere Nachricht hat 116 Byte
+	msgBlock[29] = 0x80;
+	msgBlock[31] = 0x01000000;
+
+	// groestl512 braucht hierfür keinen CPU-Code (die einzige Runde wird
+	// auf der GPU ausgeführt)
+
+	// setze register 
+	uint32_t groestl_state_init[32];
+	memset(groestl_state_init, 0, sizeof(uint32_t) * 32);
+	groestl_state_init[31] = 0x20000;
+
+	// state speichern
+	cudaMemcpyToSymbol(	groestl_gpu_state,
+						groestl_state_init,
+						128);
+
+	// Blockheader setzen (korrekte Nonce und Hefty Hash fehlen da drin noch)
+	cudaMemcpyToSymbol(	groestl_gpu_msg,
+						msgBlock,
+						128);
+}
+
+__host__ void groestl512_cpu_copyHeftyHash(int thr_id, int threads, void *heftyHashes, int copy)
+{
+	// Hefty1 Hashes kopieren (eigentlich nur zum debuggen)
+	if (copy)	
+		cudaMemcpy( d_heftyHashes[thr_id], heftyHashes, 8 * sizeof(uint32_t) * threads, cudaMemcpyHostToDevice );		
+}
+
+__host__ void groestl512_cpu_hash(int thr_id, int threads, uint32_t startNounce)
+{
+	const int threadsperblock = 128;
+
+	// berechne wie viele Thread Blocks wir brauchen
+	dim3 grid((threads + threadsperblock-1)/threadsperblock);
+	dim3 block(threadsperblock);
+
+	// Größe des dynamischen Shared Memory Bereichs (abhängig von der Threadanzahl)
+#if USE_SHARED
+	size_t shared_size = W_ALIGNMENT*sizeof(uint32_t)*threadsperblock;  // ein uint32_t eingefügt gegen Bank Konflikte
+#else
+	size_t shared_size = 0;
+#endif
+
+//	fprintf(stderr, "threads=%d, %d blocks, %d threads per block, %d bytes shared\n", threads, grid.x, block.x, shared_size);
+
+	groestl512_gpu_hash<<<grid, block, shared_size>>>(threads, startNounce, d_hash4output[thr_id], d_heftyHashes[thr_id], d_nonceVector[thr_id]);	
+}
diff --git a/cuda_groestl512.h b/cuda_groestl512.h
new file mode 100644
index 0000000000..bbeee40f53
--- /dev/null
+++ b/cuda_groestl512.h
@@ -0,0 +1,9 @@
+#ifndef _CUDA_GROESTL512_H
+#define _CUDA_GROESTL512_H
+
+void groestl512_cpu_init(int thr_id, int threads);
+void groestl512_cpu_copyHeftyHash(int thr_id, int threads, void *heftyHashes, int copy);
+void groestl512_cpu_setBlock(void *data);
+void groestl512_cpu_hash(int thr_id, int threads, uint32_t startNounce);
+
+#endif
\ No newline at end of file
diff --git a/cuda_hefty1.cu b/cuda_hefty1.cu
new file mode 100644
index 0000000000..c90d15dfd2
--- /dev/null
+++ b/cuda_hefty1.cu
@@ -0,0 +1,401 @@
+/* Diese Funktion ist auf 84-Byte große Eingabedaten ausgerichtet (Heavycoin) */
+#include <cuda.h>
+#include "cuda_runtime.h"
+#include "device_launch_parameters.h"
+
+#include <stdio.h>
+#include <memory.h>
+
+#define USE_SHARED 0
+#define W_ALIGNMENT 65
+
+// Folgende Definitionen später durch header ersetzen
+typedef unsigned int uint32_t;
+typedef unsigned char uint8_t;
+typedef unsigned short uint16_t;
+
+// globaler Speicher für alle HeftyHashes aller Threads
+uint32_t *d_heftyHashes[8];
+
+/* Hash-Tabellen */
+__constant__ uint32_t hefty_gpu_constantTable[64];
+
+// muss expandiert werden
+__constant__ uint32_t hefty_gpu_blockHeader[16]; // 2x512 Bit Message
+__constant__ uint32_t hefty_gpu_register[8];
+__constant__ uint32_t hefty_gpu_sponge[4];
+
+uint32_t hefty_cpu_hashTable[] = { 0x6a09e667UL,
+    0xbb67ae85UL,
+    0x3c6ef372UL,
+    0xa54ff53aUL,
+    0x510e527fUL,
+    0x9b05688cUL,
+    0x1f83d9abUL,
+    0x5be0cd19UL };
+uint32_t hefty_cpu_constantTable[] = {
+	0x428a2f98UL, 0x71374491UL, 0xb5c0fbcfUL, 0xe9b5dba5UL,
+    0x3956c25bUL, 0x59f111f1UL, 0x923f82a4UL, 0xab1c5ed5UL,
+    0xd807aa98UL, 0x12835b01UL, 0x243185beUL, 0x550c7dc3UL,
+    0x72be5d74UL, 0x80deb1feUL, 0x9bdc06a7UL, 0xc19bf174UL,
+    0xe49b69c1UL, 0xefbe4786UL, 0x0fc19dc6UL, 0x240ca1ccUL,
+    0x2de92c6fUL, 0x4a7484aaUL, 0x5cb0a9dcUL, 0x76f988daUL,
+    0x983e5152UL, 0xa831c66dUL, 0xb00327c8UL, 0xbf597fc7UL,
+    0xc6e00bf3UL, 0xd5a79147UL, 0x06ca6351UL, 0x14292967UL,
+    0x27b70a85UL, 0x2e1b2138UL, 0x4d2c6dfcUL, 0x53380d13UL,
+    0x650a7354UL, 0x766a0abbUL, 0x81c2c92eUL, 0x92722c85UL,
+    0xa2bfe8a1UL, 0xa81a664bUL, 0xc24b8b70UL, 0xc76c51a3UL,
+    0xd192e819UL, 0xd6990624UL, 0xf40e3585UL, 0x106aa070UL,
+    0x19a4c116UL, 0x1e376c08UL, 0x2748774cUL, 0x34b0bcb5UL,
+    0x391c0cb3UL, 0x4ed8aa4aUL, 0x5b9cca4fUL, 0x682e6ff3UL,
+    0x748f82eeUL, 0x78a5636fUL, 0x84c87814UL, 0x8cc70208UL,
+    0x90befffaUL, 0xa4506cebUL, 0xbef9a3f7UL, 0xc67178f2UL
+};
+
+#define S(x, n)			(((x) >> (n)) | ((x) << (32 - (n))))
+#define R(x, n)			((x) >> (n))
+#define Ch(x, y, z)		((x & (y ^ z)) ^ z)
+#define Maj(x, y, z)	((x & (y | z)) | (y & z))
+#define S0(x)			(S(x, 2) ^ S(x, 13) ^ S(x, 22))
+#define S1(x)			(S(x, 6) ^ S(x, 11) ^ S(x, 25))
+#define s0(x)			(S(x, 7) ^ S(x, 18) ^ R(x, 3))
+#define s1(x)			(S(x, 17) ^ S(x, 19) ^ R(x, 10))
+
+#define SWAB32(x)		( ((x & 0x000000FF) << 24) | ((x & 0x0000FF00) << 8) | ((x & 0x00FF0000) >> 8) | ((x & 0xFF000000) >> 24) )
+
+// uint8_t
+#define smoosh4(x)		( ((x)>>4) ^ ((x) & 0x0F) )
+__host__ __forceinline__ __device__ uint8_t smoosh2(uint32_t x)
+{
+	uint16_t w = (x >> 16) ^ (x & 0xffff);
+	uint8_t n = smoosh4( (uint8_t)( (w >> 8) ^ (w & 0xFF) ) );
+	return (n >> 2) ^ (n & 0x03);
+}
+// 4 auf einmal
+#define smoosh4Quad(x)	( (((x)>>4) ^ (x)) & 0x0F0F0F0F )
+#define getByte(x,y)	( ((x) >> (y)) & 0xFF )
+
+__host__ __device__ void Mangle(uint32_t *inp)
+{
+	uint32_t r = smoosh4Quad(inp[0]);
+	//uint8_t r0 = smoosh4( (uint8_t)(inp[0] >> 24) );
+	//uint8_t r1 = smoosh4( (uint8_t)(inp[0] >> 16) );
+	//uint8_t r2 = smoosh4( (uint8_t)(inp[0] >> 8) );
+	//uint8_t r3 = smoosh4( (uint8_t)(inp[0] & 0xFF) );
+
+	inp[1] = inp[1] ^ S(inp[0], getByte(r, 24));
+	
+	switch (smoosh2(inp[1])) {
+      case 0: inp[2] ^= S(inp[0], 1 + getByte(r,24)); break;
+      case 1: inp[2] += S(~inp[0], 1 + getByte(r,16)); break;
+      case 2: inp[2] &= S(~inp[0], 1 + getByte(r,8)); break;
+      case 3: inp[2] ^= S(inp[0], 1 + getByte(r,0)); break;
+    }
+    
+	uint32_t tmp = smoosh2(inp[1] ^ inp[2]);
+    switch (tmp) {
+      case 0: inp[3] ^= S(inp[0], 2 + getByte(r,24)); break;
+      case 1: inp[3] += S(~inp[0], 2 + getByte(r,16)); break;
+      case 2: inp[3] &= S(~inp[0], 2 + getByte(r,8)); break;
+      case 3: inp[3] ^= S(inp[0], 2 + getByte(r,0)); break;
+    }
+
+	inp[0] ^= (inp[1] ^ inp[2]) + inp[3];
+}
+
+__host__ __forceinline__ __device__ void Absorb(uint32_t *inp, uint32_t x)
+{
+	inp[0] ^= x;
+	Mangle(inp);
+}
+
+__host__ __forceinline__ __device__ uint32_t Squeeze(uint32_t *inp)
+{
+	uint32_t y = inp[0];
+	Mangle(inp);
+	return y;
+}
+
+__host__ __forceinline__ __device__ uint32_t Br(uint32_t *sponge, uint32_t x)
+{
+	uint32_t r = Squeeze(sponge);
+
+	//uint8_t r0 = r >> 8;
+	uint8_t r1 = r & 0xFF;
+	uint32_t y = 1 << ((r >> 8) & 0x1F);
+
+	//uint32_t retVal;
+	//retVal = x;
+	
+	uint32_t resArr[4];
+	resArr[0] = x;
+	resArr[1] = x & ~y;
+	resArr[2] = x | y;
+	resArr[3] = x ^ y;
+	return resArr[r1 & 0x03];
+
+	/*
+	switch(r1 & 0x03)
+	{
+	case 0:
+		break;
+    case 1:
+        retVal = x & ~y;
+		break;
+    case 2:
+        retVal = x | y;
+		break;
+    case 3:
+        retVal = x ^ y;
+		break;
+    }
+	return retVal;
+	*/
+}
+
+__forceinline__ __device__ void hefty_gpu_round(uint32_t *regs, uint32_t W, uint32_t K, uint32_t *sponge)
+{
+	uint32_t tmpBr;
+
+	uint32_t brG = Br(sponge, regs[6]);	
+	uint32_t brF = Br(sponge, regs[5]);
+	uint32_t tmp1 = Ch(regs[4], brF, brG) + regs[7] + W + K;
+	uint32_t brE = Br(sponge, regs[4]);
+	uint32_t tmp2 = tmp1 + S1(brE);
+	uint32_t brC = Br(sponge, regs[2]);
+	uint32_t brB = Br(sponge, regs[1]);
+	uint32_t brA = Br(sponge, regs[0]);
+	uint32_t tmp3 = Maj(brA, brB, brC);
+	tmpBr = Br(sponge, regs[0]);
+	uint32_t tmp4 = tmp3 + S0(tmpBr);
+	tmpBr = Br(sponge, tmp2);
+
+	#pragma unroll 7
+	for (int k=6; k >= 0; k--) regs[k+1] = regs[k];
+	regs[0] = tmp2 + tmp4;
+	regs[4] += tmpBr;
+}
+
+__host__ void hefty_cpu_round(uint32_t *regs, uint32_t W, uint32_t K, uint32_t *sponge)
+{
+	uint32_t tmpBr;
+
+	uint32_t brG = Br(sponge, regs[6]);	
+	uint32_t brF = Br(sponge, regs[5]);
+	uint32_t tmp1 = Ch(regs[4], brF, brG) + regs[7] + W + K;
+	uint32_t brE = Br(sponge, regs[4]);
+	uint32_t tmp2 = tmp1 + S1(brE);
+	uint32_t brC = Br(sponge, regs[2]);
+	uint32_t brB = Br(sponge, regs[1]);
+	uint32_t brA = Br(sponge, regs[0]);
+	uint32_t tmp3 = Maj(brA, brB, brC);
+	tmpBr = Br(sponge, regs[0]);
+	uint32_t tmp4 = tmp3 + S0(tmpBr);
+	tmpBr = Br(sponge, tmp2);
+
+	for (int k=6; k >= 0; k--) regs[k+1] = regs[k];
+	regs[0] = tmp2 + tmp4;
+	regs[4] += tmpBr;
+}
+
+// Die Hash-Funktion
+__global__ void hefty_gpu_hash(int threads, uint32_t startNounce, void *outputHash)
+{
+	int thread = (blockDim.x * blockIdx.x + threadIdx.x);
+	if (thread < threads)
+	{
+		// bestimme den aktuellen Zähler
+		uint32_t nounce = startNounce + thread;
+	
+		// jeder thread in diesem  Block bekommt sein eigenes W Array im Shared memory
+#if USE_SHARED
+		extern __shared__ unsigned char s[];
+		uint32_t *W = (uint32_t *)(&s[W_ALIGNMENT * sizeof(uint32_t) * threadIdx.x]);
+#else
+		// reduktion von 256 byte auf 128 byte
+		uint32_t W1[16];
+		uint32_t W2[16];
+#endif
+
+		// Initialisiere die register a bis h mit der Hash-Tabelle
+		uint32_t regs[8];
+		uint32_t hash[8];
+		uint32_t sponge[4];
+	
+#pragma unroll 4
+		for(int k=0; k < 4; k++)
+			sponge[k] = hefty_gpu_sponge[k];
+
+		// pre
+#pragma unroll 8
+		for (int k=0; k < 8; k++)
+		{
+			regs[k] = hefty_gpu_register[k];
+			hash[k] = regs[k];
+		}
+	
+		//memcpy(W, &hefty_gpu_blockHeader[0], sizeof(uint32_t) * 16); // verbleibende 20 bytes aus Block 2 plus padding
+#pragma unroll 16
+		for(int k=0;k<16;k++)
+			W1[k] = hefty_gpu_blockHeader[k];
+		W1[3] = SWAB32(nounce);
+
+
+		// 2. Runde
+#pragma unroll 16
+		for(int j=0;j<16;j++)
+			Absorb(sponge, W1[j] ^ hefty_gpu_constantTable[j]);
+
+// Progress W1 (Bytes 0...63)
+#pragma unroll 16
+		for(int j=0;j<16;j++)
+		{
+			Absorb(sponge, regs[3] ^ regs[7]);
+			hefty_gpu_round(regs, W1[j], hefty_gpu_constantTable[j], sponge);
+		}
+
+// Progress W2 (Bytes 64...127) then W3 (Bytes 128...191) ...
+		
+#pragma unroll 3
+		for(int k=0;k<3;k++)
+		{
+	#pragma unroll 2
+			for(int j=0;j<2;j++)
+				W2[j] = s1(W1[14+j]) + W1[9+j] + s0(W1[1+j]) + W1[j];
+	#pragma unroll 5
+			for(int j=2;j<7;j++)
+				W2[j] = s1(W2[j-2]) + W1[9+j] + s0(W1[1+j]) + W1[j];
+
+	#pragma unroll 8
+			for(int j=7;j<15;j++)
+				W2[j] = s1(W2[j-2]) + W2[j-7] + s0(W1[1+j]) + W1[j];
+
+			W2[15] = s1(W2[13]) + W2[8] + s0(W2[0]) + W1[15];
+
+	#pragma unroll 16
+			for(int j=0;j<16;j++)
+			{
+				Absorb(sponge, regs[3] + regs[7]);
+				hefty_gpu_round(regs, W2[j], hefty_gpu_constantTable[j + 16 * (k+1)], sponge);
+			}
+	#pragma unroll 16
+			for(int j=0;j<16;j++)
+				W1[j] = W2[j];
+		}
+		
+
+#pragma unroll 8
+		for(int k=0;k<8;k++)
+			hash[k] += regs[k];
+
+#pragma unroll 8
+		for(int k=0;k<8;k++)
+			((uint32_t*)outputHash)[8*thread+k] = SWAB32(hash[k]);
+	}
+}
+
+// Setup-Funktionen
+__host__ void hefty_cpu_init(int thr_id, int threads)
+{
+	cudaSetDevice(thr_id);
+
+	// Kopiere die Hash-Tabellen in den GPU-Speicher
+	cudaMemcpyToSymbol(	hefty_gpu_constantTable,
+						hefty_cpu_constantTable,
+						sizeof(uint32_t) * 64 );
+
+	// Speicher für alle Hefty1 hashes belegen
+	cudaMalloc(&d_heftyHashes[thr_id], 8 * sizeof(uint32_t) * threads);
+}
+
+__host__ void hefty_cpu_setBlock(int thr_id, int threads, void *data)
+	// data muss 84-Byte haben!
+{
+	// Nachricht expandieren und setzen
+	uint32_t msgBlock[32];
+
+	memset(msgBlock, 0, sizeof(uint32_t) * 32);
+	memcpy(&msgBlock[0], data, 84);
+	msgBlock[21] |= 0x80;
+	msgBlock[31] = 672; // bitlen
+	
+	for(int i=0;i<31;i++) // Byteorder drehen
+		msgBlock[i] = SWAB32(msgBlock[i]);
+
+	// die erste Runde wird auf der CPU durchgeführt, da diese für
+	// alle Threads gleich ist. Der Hash wird dann an die Threads
+	// übergeben
+
+	// Erstelle expandierten Block W
+	uint32_t W[64];	
+	memcpy(W, &msgBlock[0], sizeof(uint32_t) * 16);	
+	for(int j=16;j<64;j++)
+		W[j] = s1(W[j-2]) + W[j-7] + s0(W[j-15]) + W[j-16];
+
+	// Initialisiere die register a bis h mit der Hash-Tabelle
+	uint32_t regs[8];
+	uint32_t hash[8];
+	uint32_t sponge[4];
+
+	// pre
+	memset(sponge, 0, sizeof(uint32_t) * 4);
+    for (int k=0; k < 8; k++)
+	{
+		regs[k] = hefty_cpu_hashTable[k];
+		hash[k] = regs[k];
+	}	
+
+	// 1. Runde
+	for(int j=0;j<16;j++)
+		Absorb(sponge, W[j] ^ hefty_cpu_constantTable[j]);
+
+	for(int j=0;j<16;j++)
+	{
+		Absorb(sponge, regs[3] ^ regs[7]);
+		hefty_cpu_round(regs, W[j], hefty_cpu_constantTable[j], sponge);
+	}
+
+	for(int j=16;j<64;j++)
+	{
+		Absorb(sponge, regs[3] + regs[7]);
+		hefty_cpu_round(regs, W[j], hefty_cpu_constantTable[j], sponge);
+	}
+
+	for(int k=0;k<8;k++)
+		hash[k] += regs[k];
+
+	// sponge speichern
+
+	cudaMemcpyToSymbol(	hefty_gpu_sponge,
+						sponge,
+						sizeof(uint32_t) * 4 );
+	// hash speichern
+	cudaMemcpyToSymbol(	hefty_gpu_register,
+						hash,
+						sizeof(uint32_t) * 8 );
+
+	// Blockheader setzen (korrekte Nonce fehlt da drin noch)
+	cudaMemcpyToSymbol(	hefty_gpu_blockHeader,
+						&msgBlock[16],
+						64);
+}
+
+__host__ void hefty_cpu_hash(int thr_id, int threads, int startNounce)
+{
+	const int threadsperblock = 128;
+
+	// berechne wie viele Thread Blocks wir brauchen
+	dim3 grid((threads + threadsperblock-1)/threadsperblock);
+	dim3 block(threadsperblock);
+
+	// Größe des dynamischen Shared Memory Bereichs (abhängig von der Threadanzahl)
+#if USE_SHARED
+	size_t shared_size = W_ALIGNMENT*sizeof(uint32_t)*threadsperblock;  // ein uint32_t eingefügt gegen Bank Konflikte
+#else
+	size_t shared_size = 0;
+#endif
+
+//	fprintf(stderr, "threads=%d, %d blocks, %d threads per block, %d bytes shared\n", threads, grid.x, block.x, shared_size);
+
+	hefty_gpu_hash<<<grid, block, shared_size>>>(threads, startNounce, (void*)d_heftyHashes[thr_id]);
+}
diff --git a/cuda_hefty1.h b/cuda_hefty1.h
new file mode 100644
index 0000000000..08b1844480
--- /dev/null
+++ b/cuda_hefty1.h
@@ -0,0 +1,8 @@
+#ifndef _CUDA_HEFTY1_H
+#define _CUDA_HEFTY1_H
+
+void hefty_cpu_hash(int thr_id, int threads, int startNounce);
+void hefty_cpu_setBlock(int thr_id, int threads, void *data);
+void hefty_cpu_init(int thr_id, int threads);
+
+#endif
\ No newline at end of file
diff --git a/cuda_keccak512.cu b/cuda_keccak512.cu
new file mode 100644
index 0000000000..66dddafe26
--- /dev/null
+++ b/cuda_keccak512.cu
@@ -0,0 +1,274 @@
+/* Diese Funktion ist auf 84+32-Byte große Eingabedaten ausgerichtet (Heavycoin) */
+#include <cuda.h>
+#include "cuda_runtime.h"
+#include "device_launch_parameters.h"
+
+#include <stdio.h>
+#include <memory.h>
+
+// Folgende Definitionen später durch header ersetzen
+typedef unsigned char uint8_t;
+typedef unsigned int uint32_t;
+typedef unsigned long long uint64_t;
+
+// globaler Speicher für alle HeftyHashes aller Threads
+extern uint32_t *d_heftyHashes[8];
+extern uint32_t *d_nonceVector[8];
+
+// globaler Speicher für unsere Ergebnisse
+uint32_t *d_hash3output[8];
+
+// der Keccak512 State nach der ersten Runde (72 Bytes)
+__constant__ uint64_t c_State[25];
+
+// die Message (72 Bytes) für die zweite Runde auf der GPU
+__constant__ uint32_t c_PaddedMessage2[18]; // 44 bytes of remaining message (Nonce at offset 4) plus padding
+
+// ---------------------------- BEGIN CUDA keccak512 functions ------------------------------------
+
+#define ROTL64(a,b) (((a) << (b)) | ((a) >> (64 - b)))
+
+#define U32TO64_LE(p) \
+    (((uint64_t)(*p)) | (((uint64_t)(*(p + 1))) << 32))
+
+#define U64TO32_LE(p, v) \
+    *p = (uint32_t)((v)); *(p+1) = (uint32_t)((v) >> 32);
+
+static __device__ void mycpy72(uint32_t *d, const uint32_t *s) {
+#pragma unroll 18
+    for (int k=0; k < 18; ++k) d[k] = s[k];
+}
+
+static __device__ void mycpy32(uint32_t *d, const uint32_t *s) {
+#pragma unroll 8
+    for (int k=0; k < 8; ++k) d[k] = s[k];
+}
+
+typedef struct keccak_hash_state_t {
+    uint64_t state[25];                        // 25*2
+    uint32_t buffer[72/4];                     // 72
+} keccak_hash_state;
+
+__device__ void statecopy(uint64_t *d, uint64_t *s)
+{
+#pragma unroll 25
+    for (int i=0; i < 25; ++i)
+        d[i] = s[i];
+}
+
+
+static const uint64_t host_keccak_round_constants[24] = {
+    0x0000000000000001ull, 0x0000000000008082ull,
+    0x800000000000808aull, 0x8000000080008000ull,
+    0x000000000000808bull, 0x0000000080000001ull,
+    0x8000000080008081ull, 0x8000000000008009ull,
+    0x000000000000008aull, 0x0000000000000088ull,
+    0x0000000080008009ull, 0x000000008000000aull,
+    0x000000008000808bull, 0x800000000000008bull,
+    0x8000000000008089ull, 0x8000000000008003ull,
+    0x8000000000008002ull, 0x8000000000000080ull,
+    0x000000000000800aull, 0x800000008000000aull,
+    0x8000000080008081ull, 0x8000000000008080ull,
+    0x0000000080000001ull, 0x8000000080008008ull
+};
+
+__constant__ uint64_t c_keccak_round_constants[24];
+
+__host__ __device__ void
+keccak_block(uint64_t *s, const uint32_t *in, const uint64_t *keccak_round_constants) {
+    size_t i;
+    uint64_t t[5], u[5], v, w;
+
+    /* absorb input */
+#pragma unroll 9
+    for (i = 0; i < 72 / 8; i++, in += 2)
+        s[i] ^= U32TO64_LE(in);
+    
+    for (i = 0; i < 24; i++) {
+        /* theta: c = a[0,i] ^ a[1,i] ^ .. a[4,i] */
+        t[0] = s[0] ^ s[5] ^ s[10] ^ s[15] ^ s[20];
+        t[1] = s[1] ^ s[6] ^ s[11] ^ s[16] ^ s[21];
+        t[2] = s[2] ^ s[7] ^ s[12] ^ s[17] ^ s[22];
+        t[3] = s[3] ^ s[8] ^ s[13] ^ s[18] ^ s[23];
+        t[4] = s[4] ^ s[9] ^ s[14] ^ s[19] ^ s[24];
+
+        /* theta: d[i] = c[i+4] ^ rotl(c[i+1],1) */
+        u[0] = t[4] ^ ROTL64(t[1], 1);
+        u[1] = t[0] ^ ROTL64(t[2], 1);
+        u[2] = t[1] ^ ROTL64(t[3], 1);
+        u[3] = t[2] ^ ROTL64(t[4], 1);
+        u[4] = t[3] ^ ROTL64(t[0], 1);
+
+        /* theta: a[0,i], a[1,i], .. a[4,i] ^= d[i] */
+        s[0] ^= u[0]; s[5] ^= u[0]; s[10] ^= u[0]; s[15] ^= u[0]; s[20] ^= u[0];
+        s[1] ^= u[1]; s[6] ^= u[1]; s[11] ^= u[1]; s[16] ^= u[1]; s[21] ^= u[1];
+        s[2] ^= u[2]; s[7] ^= u[2]; s[12] ^= u[2]; s[17] ^= u[2]; s[22] ^= u[2];
+        s[3] ^= u[3]; s[8] ^= u[3]; s[13] ^= u[3]; s[18] ^= u[3]; s[23] ^= u[3];
+        s[4] ^= u[4]; s[9] ^= u[4]; s[14] ^= u[4]; s[19] ^= u[4]; s[24] ^= u[4];
+
+        /* rho pi: b[..] = rotl(a[..], ..) */
+        v = s[ 1];
+        s[ 1] = ROTL64(s[ 6], 44);
+        s[ 6] = ROTL64(s[ 9], 20);
+        s[ 9] = ROTL64(s[22], 61);
+        s[22] = ROTL64(s[14], 39);
+        s[14] = ROTL64(s[20], 18);
+        s[20] = ROTL64(s[ 2], 62);
+        s[ 2] = ROTL64(s[12], 43);
+        s[12] = ROTL64(s[13], 25);
+        s[13] = ROTL64(s[19],  8);
+        s[19] = ROTL64(s[23], 56);
+        s[23] = ROTL64(s[15], 41);
+        s[15] = ROTL64(s[ 4], 27);
+        s[ 4] = ROTL64(s[24], 14);
+        s[24] = ROTL64(s[21],  2);
+        s[21] = ROTL64(s[ 8], 55);
+        s[ 8] = ROTL64(s[16], 45);
+        s[16] = ROTL64(s[ 5], 36);
+        s[ 5] = ROTL64(s[ 3], 28);
+        s[ 3] = ROTL64(s[18], 21);
+        s[18] = ROTL64(s[17], 15);
+        s[17] = ROTL64(s[11], 10);
+        s[11] = ROTL64(s[ 7],  6);
+        s[ 7] = ROTL64(s[10],  3);
+        s[10] = ROTL64(    v,  1);
+
+        /* chi: a[i,j] ^= ~b[i,j+1] & b[i,j+2] */
+        v = s[ 0]; w = s[ 1]; s[ 0] ^= (~w) & s[ 2]; s[ 1] ^= (~s[ 2]) & s[ 3]; s[ 2] ^= (~s[ 3]) & s[ 4]; s[ 3] ^= (~s[ 4]) & v; s[ 4] ^= (~v) & w;
+        v = s[ 5]; w = s[ 6]; s[ 5] ^= (~w) & s[ 7]; s[ 6] ^= (~s[ 7]) & s[ 8]; s[ 7] ^= (~s[ 8]) & s[ 9]; s[ 8] ^= (~s[ 9]) & v; s[ 9] ^= (~v) & w;
+        v = s[10]; w = s[11]; s[10] ^= (~w) & s[12]; s[11] ^= (~s[12]) & s[13]; s[12] ^= (~s[13]) & s[14]; s[13] ^= (~s[14]) & v; s[14] ^= (~v) & w;
+        v = s[15]; w = s[16]; s[15] ^= (~w) & s[17]; s[16] ^= (~s[17]) & s[18]; s[17] ^= (~s[18]) & s[19]; s[18] ^= (~s[19]) & v; s[19] ^= (~v) & w;
+        v = s[20]; w = s[21]; s[20] ^= (~w) & s[22]; s[21] ^= (~s[22]) & s[23]; s[22] ^= (~s[23]) & s[24]; s[23] ^= (~s[24]) & v; s[24] ^= (~v) & w;
+
+        /* iota: a[0,0] ^= round constant */
+        s[0] ^= keccak_round_constants[i];
+    }
+}
+
+// Die Hash-Funktion
+__global__ void keccak512_gpu_hash(int threads, uint32_t startNounce, void *outputHash, uint32_t *heftyHashes, uint32_t *nonceVector)
+{
+	int thread = (blockDim.x * blockIdx.x + threadIdx.x);
+	if (thread < threads)
+	{
+		// bestimme den aktuellen Zähler
+		//uint32_t nounce = startNounce + thread;
+		uint32_t nounce = nonceVector[thread];
+
+		// Index-Position des Hashes in den Hash Puffern bestimmen (Hefty1 und outputHash)
+		uint32_t hashPosition = nounce - startNounce;
+
+		// erstmal den State der ersten Runde holen
+		uint64_t keccak_gpu_state[25];
+#pragma unroll 25
+		for (int i=0; i < 25; ++i)
+			keccak_gpu_state[i] = c_State[i];
+	
+		// Message2 in den Puffer holen
+		uint32_t msgBlock[18];
+		mycpy72(msgBlock, c_PaddedMessage2);
+
+		// die individuelle Nonce einsetzen
+		msgBlock[1] = nounce;
+
+		// den individuellen Hefty1 Hash einsetzen
+		mycpy32(&msgBlock[3], &heftyHashes[8 * hashPosition]);
+
+		// den Block einmal gut durchschütteln
+		keccak_block(keccak_gpu_state, msgBlock, c_keccak_round_constants);
+
+		// das Hash erzeugen
+		uint32_t hash[16];
+
+#pragma unroll 8
+		for (size_t i = 0; i < 64; i += 8) {
+			U64TO32_LE((&hash[i/4]), keccak_gpu_state[i / 8]);
+		}
+
+
+		// und ins Global Memory rausschreiben
+#pragma unroll 16
+		for(int k=0;k<16;k++)
+			((uint32_t*)outputHash)[16*hashPosition+k] = hash[k];
+	}
+}
+
+// ---------------------------- END CUDA keccak512 functions ------------------------------------
+
+// Setup-Funktionen
+__host__ void keccak512_cpu_init(int thr_id, int threads)
+{
+	// Kopiere die Hash-Tabellen in den GPU-Speicher
+	cudaMemcpyToSymbol( c_keccak_round_constants,
+						host_keccak_round_constants,
+						sizeof(host_keccak_round_constants),
+						0, cudaMemcpyHostToDevice);
+
+	// Speicher für alle Ergebnisse belegen
+	cudaMalloc(&d_hash3output[thr_id], 16 * sizeof(uint32_t) * threads);
+}
+
+// ----------------BEGIN keccak512 CPU version from scrypt-jane code --------------------
+
+#define SCRYPT_HASH_DIGEST_SIZE 64
+#define SCRYPT_KECCAK_F 1600
+#define SCRYPT_KECCAK_C (SCRYPT_HASH_DIGEST_SIZE * 8 * 2) /* 1024 */
+#define SCRYPT_KECCAK_R (SCRYPT_KECCAK_F - SCRYPT_KECCAK_C) /* 576 */
+#define SCRYPT_HASH_BLOCK_SIZE (SCRYPT_KECCAK_R / 8) /* 72 */
+
+// --------------- END keccak512 CPU version from scrypt-jane code --------------------
+
+__host__ void keccak512_cpu_setBlock(void *data)
+	// data muss 84-Byte haben!
+	// heftyHash hat 32-Byte
+{
+	// CH
+	// state init	
+	uint64_t keccak_cpu_state[25];
+	memset(keccak_cpu_state, 0, 200);
+
+	// keccak hat 72-Byte blöcke, d.h. in unserem Fall zwei Blöcke
+	// zu jeweils 
+	uint32_t msgBlock[18];
+	memset(msgBlock, 0, 18 * sizeof(uint32_t));
+
+	// kopiere die Daten rein (aber nur alles nach Bit 72)
+	memcpy(&msgBlock[0], &((uint8_t*)data)[72], 12);
+
+	// Nachricht abschließen
+	msgBlock[11] = 0x01;
+	msgBlock[17] = 0x80000000;
+	
+	// erste Runde	
+	keccak_block((uint64_t*)&keccak_cpu_state, (const uint32_t*)data, host_keccak_round_constants);
+
+	// Message 2 ins Constant Memory kopieren (die variable Nonce und 
+	// der Hefty1 Anteil muss aber auf der GPU erst noch ersetzt werden)
+	cudaMemcpyToSymbol( c_PaddedMessage2, msgBlock, 18*sizeof(uint32_t), 0, cudaMemcpyHostToDevice );
+
+	// state kopieren
+	cudaMemcpyToSymbol( c_State, keccak_cpu_state, 25*sizeof(uint64_t), 0, cudaMemcpyHostToDevice);
+}
+
+__host__ void keccak512_cpu_copyHeftyHash(int thr_id, int threads, void *heftyHashes, int copy)
+{
+	// Hefty1 Hashes kopieren
+	if (copy) cudaMemcpy( d_heftyHashes[thr_id], heftyHashes, 8 * sizeof(uint32_t) * threads, cudaMemcpyHostToDevice );
+	//else cudaThreadSynchronize();
+}
+
+__host__ void keccak512_cpu_hash(int thr_id, int threads, uint32_t startNounce)
+{
+	const int threadsperblock = 128;
+
+	// berechne wie viele Thread Blocks wir brauchen
+	dim3 grid((threads + threadsperblock-1)/threadsperblock);
+	dim3 block(threadsperblock);
+
+	// Größe des dynamischen Shared Memory Bereichs (abhängig von der Threadanzahl)
+	size_t shared_size = 0;
+
+//	fprintf(stderr, "threads=%d, %d blocks, %d threads per block, %d bytes shared\n", threads, grid.x, block.x, shared_size);
+
+	keccak512_gpu_hash<<<grid, block, shared_size>>>(threads, startNounce, d_hash3output[thr_id], d_heftyHashes[thr_id], d_nonceVector[thr_id]);
+}
diff --git a/cuda_keccak512.h b/cuda_keccak512.h
new file mode 100644
index 0000000000..abd4741fd9
--- /dev/null
+++ b/cuda_keccak512.h
@@ -0,0 +1,9 @@
+#ifndef _CUDA_KECCAK512_H
+#define _CUDA_KECCAK512_H
+
+void keccak512_cpu_init(int thr_id, int threads);
+void keccak512_cpu_setBlock(void *data);
+void keccak512_cpu_copyHeftyHash(int thr_id, int threads, void *heftyHashes, int copy);
+void keccak512_cpu_hash(int thr_id, int threads, uint32_t startNounce);
+
+#endif
diff --git a/cuda_sha256.cu b/cuda_sha256.cu
new file mode 100644
index 0000000000..a68f849664
--- /dev/null
+++ b/cuda_sha256.cu
@@ -0,0 +1,274 @@
+/* Diese Funktion ist auf 84+32 Byte große Eingabedaten ausgerichtet (Heavycoin) */
+#include <cuda.h>
+#include "cuda_runtime.h"
+#include "device_launch_parameters.h"
+
+#include <stdio.h>
+#include <memory.h>
+
+#define W_ALIGNMENT 65
+
+// Folgende Definitionen später durch header ersetzen
+typedef unsigned int uint32_t;
+
+// globaler Speicher für alle HeftyHashes aller Threads
+extern uint32_t *d_heftyHashes[8];
+extern uint32_t *d_nonceVector[8];
+
+// globaler Speicher für unsere Ergebnisse
+uint32_t *d_hash2output[8];
+
+
+/* Hash-Tabellen */
+__constant__ uint32_t sha256_gpu_constantTable[64];
+
+// muss expandiert werden
+__constant__ uint32_t sha256_gpu_blockHeader[16]; // 2x512 Bit Message
+__constant__ uint32_t sha256_gpu_register[8];
+
+uint32_t sha256_cpu_hashTable[] = { 0x6a09e667, 0xbb67ae85, 0x3c6ef372, 0xa54ff53a, 0x510e527f, 0x9b05688c, 0x1f83d9ab, 0x5be0cd19 };
+uint32_t sha256_cpu_constantTable[] = {
+	0x428a2f98, 0x71374491, 0xb5c0fbcf, 0xe9b5dba5, 0x3956c25b, 0x59f111f1, 0x923f82a4, 0xab1c5ed5,
+	0xd807aa98, 0x12835b01, 0x243185be, 0x550c7dc3, 0x72be5d74, 0x80deb1fe, 0x9bdc06a7, 0xc19bf174,
+	0xe49b69c1, 0xefbe4786, 0x0fc19dc6, 0x240ca1cc, 0x2de92c6f, 0x4a7484aa, 0x5cb0a9dc, 0x76f988da,
+	0x983e5152, 0xa831c66d, 0xb00327c8, 0xbf597fc7, 0xc6e00bf3, 0xd5a79147, 0x06ca6351, 0x14292967,
+	0x27b70a85, 0x2e1b2138, 0x4d2c6dfc, 0x53380d13, 0x650a7354, 0x766a0abb, 0x81c2c92e, 0x92722c85,
+	0xa2bfe8a1, 0xa81a664b, 0xc24b8b70, 0xc76c51a3, 0xd192e819, 0xd6990624, 0xf40e3585, 0x106aa070,
+	0x19a4c116, 0x1e376c08, 0x2748774c, 0x34b0bcb5, 0x391c0cb3, 0x4ed8aa4a, 0x5b9cca4f, 0x682e6ff3,
+	0x748f82ee, 0x78a5636f, 0x84c87814, 0x8cc70208, 0x90befffa, 0xa4506ceb, 0xbef9a3f7, 0xc67178f2,
+};
+
+#define S(x, n)			(((x) >> (n)) | ((x) << (32 - (n))))
+#define R(x, n)			((x) >> (n))
+#define Ch(x, y, z)		((x & (y ^ z)) ^ z)
+#define Maj(x, y, z)	((x & (y | z)) | (y & z))
+#define S0(x)			(S(x, 2) ^ S(x, 13) ^ S(x, 22))
+#define S1(x)			(S(x, 6) ^ S(x, 11) ^ S(x, 25))
+#define s0(x)			(S(x, 7) ^ S(x, 18) ^ R(x, 3))
+#define s1(x)			(S(x, 17) ^ S(x, 19) ^ R(x, 10))
+
+#define SWAB32(x)		( ((x & 0x000000FF) << 24) | ((x & 0x0000FF00) << 8) | ((x & 0x00FF0000) >> 8) | ((x & 0xFF000000) >> 24) )
+
+// Die Hash-Funktion
+__global__ void sha256_gpu_hash(int threads, uint32_t startNounce, void *outputHash, uint32_t *heftyHashes, uint32_t *nonceVector)
+{
+	int thread = (blockDim.x * blockIdx.x + threadIdx.x);
+	if (thread < threads)
+	{
+		// bestimme den aktuellen Zähler
+		uint32_t nounce = startNounce + thread;
+		nonceVector[thread] = nounce;
+	
+		// jeder thread in diesem  Block bekommt sein eigenes W Array im Shared memory
+		//extern __shared__ unsigned char s[];
+		//uint32_t *W = (uint32_t *)(&s[W_ALIGNMENT * sizeof(uint32_t) * threadIdx.x]);
+		uint32_t W1[16];
+		uint32_t W2[16];
+
+		// Initialisiere die register a bis h mit der Hash-Tabelle
+		uint32_t regs[8];
+		uint32_t hash[8];
+
+		// pre
+#pragma unroll 8
+		for (int k=0; k < 8; k++)
+		{
+			regs[k] = sha256_gpu_register[k];
+			hash[k] = regs[k];
+		}
+	
+		// 2. Runde
+		//memcpy(W, &sha256_gpu_blockHeader[0], sizeof(uint32_t) * 16); // TODO: aufsplitten in zwei Teilblöcke
+		//memcpy(&W[5], &heftyHashes[8 * (blockDim.x * blockIdx.x + threadIdx.x)], sizeof(uint32_t) * 8); // den richtigen Hefty1 Hash holen		
+#pragma unroll 16
+		for(int k=0;k<16;k++)
+			W1[k] = sha256_gpu_blockHeader[k];
+
+		uint32_t offset = 8 * (blockDim.x * blockIdx.x + threadIdx.x);
+#pragma unroll 8
+		for(int k=0;k<8;k++)
+			W1[5+k] = heftyHashes[offset + k];
+
+
+#pragma unroll 8
+		for (int i=5; i <5+8; ++i) W1[i] = SWAB32(W1[i]); // die Hefty1 Hashes brauchen eine Drehung ;)
+		W1[3] = SWAB32(nounce);
+
+// Progress W1
+#pragma unroll 16
+		for(int j=0;j<16;j++)
+		{
+			uint32_t T1, T2;
+			T1 = regs[7] + S1(regs[4]) + Ch(regs[4], regs[5], regs[6]) + sha256_gpu_constantTable[j] + W1[j];
+			T2 = S0(regs[0]) + Maj(regs[0], regs[1], regs[2]);
+		
+			#pragma unroll 7
+			for (int k=6; k >= 0; k--) regs[k+1] = regs[k];
+			regs[0] = T1 + T2;
+			regs[4] += T1;
+		}
+
+// Progress W2...W3
+#pragma unroll 3
+		for(int k=0;k<3;k++)
+		{
+	#pragma unroll 2
+			for(int j=0;j<2;j++)
+				W2[j] = s1(W1[14+j]) + W1[9+j] + s0(W1[1+j]) + W1[j];
+	#pragma unroll 5
+			for(int j=2;j<7;j++)
+				W2[j] = s1(W2[j-2]) + W1[9+j] + s0(W1[1+j]) + W1[j];
+
+	#pragma unroll 8
+			for(int j=7;j<15;j++)
+				W2[j] = s1(W2[j-2]) + W2[j-7] + s0(W1[1+j]) + W1[j];
+
+			W2[15] = s1(W2[13]) + W2[8] + s0(W2[0]) + W1[15];
+
+			// Rundenfunktion
+	#pragma unroll 16
+			for(int j=0;j<16;j++)
+			{
+				uint32_t T1, T2;
+				T1 = regs[7] + S1(regs[4]) + Ch(regs[4], regs[5], regs[6]) + sha256_gpu_constantTable[j + 16 * (k+1)] + W2[j];
+				T2 = S0(regs[0]) + Maj(regs[0], regs[1], regs[2]);
+		
+				#pragma unroll 7
+				for (int l=6; l >= 0; l--) regs[l+1] = regs[l];
+				regs[0] = T1 + T2;
+				regs[4] += T1;
+			}
+
+	#pragma unroll 16
+			for(int j=0;j<16;j++)
+				W1[j] = W2[j];
+		}
+
+/*
+		for(int j=16;j<64;j++)
+			W[j] = s1(W[j-2]) + W[j-7] + s0(W[j-15]) + W[j-16];
+	
+#pragma unroll 64
+		for(int j=0;j<64;j++)
+		{
+			uint32_t T1, T2;
+			T1 = regs[7] + S1(regs[4]) + Ch(regs[4], regs[5], regs[6]) + sha256_gpu_constantTable[j] + W[j];
+			T2 = S0(regs[0]) + Maj(regs[0], regs[1], regs[2]);
+		
+			#pragma unroll 7
+			for (int k=6; k >= 0; k--) regs[k+1] = regs[k];
+			regs[0] = T1 + T2;
+			regs[4] += T1;
+		}
+*/
+#pragma unroll 8
+		for(int k=0;k<8;k++)
+			hash[k] += regs[k];
+
+#pragma unroll 8
+		for(int k=0;k<8;k++)
+			((uint32_t*)outputHash)[8*thread+k] = SWAB32(hash[k]);
+	}
+}
+
+// Setup-Funktionen
+__host__ void sha256_cpu_init(int thr_id, int threads)
+{
+	// Kopiere die Hash-Tabellen in den GPU-Speicher
+	cudaMemcpyToSymbol(	sha256_gpu_constantTable,
+						sha256_cpu_constantTable,
+						sizeof(uint32_t) * 64 );
+
+	// Speicher für alle Ergebnisse belegen
+	cudaMalloc(&d_hash2output[thr_id], 8 * sizeof(uint32_t) * threads);
+}
+
+__host__ void sha256_cpu_setBlock(void *data)
+	// data muss 84-Byte haben!
+	// heftyHash hat 32-Byte
+{
+	// Nachricht expandieren und setzen
+	uint32_t msgBlock[32];
+
+	memset(msgBlock, 0, sizeof(uint32_t) * 32);
+	memcpy(&msgBlock[0], data, 84);
+	memset(&msgBlock[21], 0, 32); // vorläufig  Nullen anstatt der Hefty1 Hashes einfüllen
+	msgBlock[29] |= 0x80;
+	msgBlock[31] = 928; // bitlen
+	
+	for(int i=0;i<31;i++) // Byteorder drehen
+		msgBlock[i] = SWAB32(msgBlock[i]);
+
+	// die erste Runde wird auf der CPU durchgeführt, da diese für
+	// alle Threads gleich ist. Der Hash wird dann an die Threads
+	// übergeben
+	uint32_t W[64];
+
+	// Erstelle expandierten Block W
+	memcpy(W, &msgBlock[0], sizeof(uint32_t) * 16);	
+	for(int j=16;j<64;j++)
+		W[j] = s1(W[j-2]) + W[j-7] + s0(W[j-15]) + W[j-16];
+
+	// Initialisiere die register a bis h mit der Hash-Tabelle
+	uint32_t regs[8];
+	uint32_t hash[8];
+
+	// pre
+    for (int k=0; k < 8; k++)
+	{
+		regs[k] = sha256_cpu_hashTable[k];
+		hash[k] = regs[k];
+	}
+
+	// 1. Runde
+	for(int j=0;j<64;j++)
+	{
+		uint32_t T1, T2;
+		T1 = regs[7] + S1(regs[4]) + Ch(regs[4], regs[5], regs[6]) + sha256_cpu_constantTable[j] + W[j];
+		T2 = S0(regs[0]) + Maj(regs[0], regs[1], regs[2]);
+		
+		//#pragma unroll 7
+		for (int k=6; k >= 0; k--) regs[k+1] = regs[k];
+		// sollte mal noch durch memmov ersetzt werden!
+//		memcpy(&regs[1], &regs[0], sizeof(uint32_t) * 7);
+		regs[0] = T1 + T2;
+		regs[4] += T1;
+	}
+
+	for(int k=0;k<8;k++)
+		hash[k] += regs[k];
+
+	// hash speichern
+	cudaMemcpyToSymbol(	sha256_gpu_register,
+						hash,
+						sizeof(uint32_t) * 8 );
+
+	// Blockheader setzen (korrekte Nonce und Hefty Hash fehlen da drin noch)
+	cudaMemcpyToSymbol(	sha256_gpu_blockHeader,
+						&msgBlock[16],
+						64);
+}
+
+__host__ void sha256_cpu_copyHeftyHash(int thr_id, int threads, void *heftyHashes, int copy)
+{
+	// Hefty1 Hashes kopieren
+	if (copy) cudaMemcpy( d_heftyHashes[thr_id], heftyHashes, 8 * sizeof(uint32_t) * threads, cudaMemcpyHostToDevice );
+	//else cudaThreadSynchronize();
+}
+
+__host__ void sha256_cpu_hash(int thr_id, int threads, int startNounce)
+{
+	const int threadsperblock = 128;
+
+	// berechne wie viele Thread Blocks wir brauchen
+	dim3 grid((threads + threadsperblock-1)/threadsperblock);
+	dim3 block(threadsperblock);
+
+	// Größe des dynamischen Shared Memory Bereichs (abhängig von der Threadanzahl)
+	//size_t shared_size = W_ALIGNMENT*sizeof(uint32_t)*threadsperblock;  // ein uint32_t eingefügt gegen Bank Konflikte
+	size_t shared_size = 0;
+
+//	fprintf(stderr, "threads=%d, %d blocks, %d threads per block, %d bytes shared\n", threads, grid.x, block.x, shared_size);
+
+	sha256_gpu_hash<<<grid, block, shared_size>>>(threads, startNounce, d_hash2output[thr_id], d_heftyHashes[thr_id], d_nonceVector[thr_id]);
+}
diff --git a/cuda_sha256.h b/cuda_sha256.h
new file mode 100644
index 0000000000..ff03bf544b
--- /dev/null
+++ b/cuda_sha256.h
@@ -0,0 +1,8 @@
+#ifndef _CUDA_SHA256_H
+#define _CUDA_SHA256_H
+
+void sha256_cpu_init(int thr_id, int threads);
+void sha256_cpu_setBlock(void *data);
+void sha256_cpu_hash(int thr_id, int threads, int startNounce);
+void sha256_cpu_copyHeftyHash(int thr_id, int threads, void *heftyHashes, int copy);
+#endif
diff --git a/depcomp b/depcomp
new file mode 100644
index 0000000000..bd0ac08958
--- /dev/null
+++ b/depcomp
@@ -0,0 +1,688 @@
+#! /bin/sh
+# depcomp - compile a program generating dependencies as side-effects
+
+scriptversion=2011-12-04.11; # UTC
+
+# Copyright (C) 1999, 2000, 2003, 2004, 2005, 2006, 2007, 2009, 2010,
+# 2011 Free Software Foundation, Inc.
+
+# This program is free software; you can redistribute it and/or modify
+# it under the terms of the GNU General Public License as published by
+# the Free Software Foundation; either version 2, or (at your option)
+# any later version.
+
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+# GNU General Public License for more details.
+
+# You should have received a copy of the GNU General Public License
+# along with this program.  If not, see <http://www.gnu.org/licenses/>.
+
+# As a special exception to the GNU General Public License, if you
+# distribute this file as part of a program that contains a
+# configuration script generated by Autoconf, you may include it under
+# the same distribution terms that you use for the rest of that program.
+
+# Originally written by Alexandre Oliva <oliva@dcc.unicamp.br>.
+
+case $1 in
+  '')
+     echo "$0: No command.  Try \`$0 --help' for more information." 1>&2
+     exit 1;
+     ;;
+  -h | --h*)
+    cat <<\EOF
+Usage: depcomp [--help] [--version] PROGRAM [ARGS]
+
+Run PROGRAMS ARGS to compile a file, generating dependencies
+as side-effects.
+
+Environment variables:
+  depmode     Dependency tracking mode.
+  source      Source file read by `PROGRAMS ARGS'.
+  object      Object file output by `PROGRAMS ARGS'.
+  DEPDIR      directory where to store dependencies.
+  depfile     Dependency file to output.
+  tmpdepfile  Temporary file to use when outputting dependencies.
+  libtool     Whether libtool is used (yes/no).
+
+Report bugs to <bug-automake@gnu.org>.
+EOF
+    exit $?
+    ;;
+  -v | --v*)
+    echo "depcomp $scriptversion"
+    exit $?
+    ;;
+esac
+
+if test -z "$depmode" || test -z "$source" || test -z "$object"; then
+  echo "depcomp: Variables source, object and depmode must be set" 1>&2
+  exit 1
+fi
+
+# Dependencies for sub/bar.o or sub/bar.obj go into sub/.deps/bar.Po.
+depfile=${depfile-`echo "$object" |
+  sed 's|[^\\/]*$|'${DEPDIR-.deps}'/&|;s|\.\([^.]*\)$|.P\1|;s|Pobj$|Po|'`}
+tmpdepfile=${tmpdepfile-`echo "$depfile" | sed 's/\.\([^.]*\)$/.T\1/'`}
+
+rm -f "$tmpdepfile"
+
+# Some modes work just like other modes, but use different flags.  We
+# parameterize here, but still list the modes in the big case below,
+# to make depend.m4 easier to write.  Note that we *cannot* use a case
+# here, because this file can only contain one case statement.
+if test "$depmode" = hp; then
+  # HP compiler uses -M and no extra arg.
+  gccflag=-M
+  depmode=gcc
+fi
+
+if test "$depmode" = dashXmstdout; then
+   # This is just like dashmstdout with a different argument.
+   dashmflag=-xM
+   depmode=dashmstdout
+fi
+
+cygpath_u="cygpath -u -f -"
+if test "$depmode" = msvcmsys; then
+   # This is just like msvisualcpp but w/o cygpath translation.
+   # Just convert the backslash-escaped backslashes to single forward
+   # slashes to satisfy depend.m4
+   cygpath_u='sed s,\\\\,/,g'
+   depmode=msvisualcpp
+fi
+
+if test "$depmode" = msvc7msys; then
+   # This is just like msvc7 but w/o cygpath translation.
+   # Just convert the backslash-escaped backslashes to single forward
+   # slashes to satisfy depend.m4
+   cygpath_u='sed s,\\\\,/,g'
+   depmode=msvc7
+fi
+
+case "$depmode" in
+gcc3)
+## gcc 3 implements dependency tracking that does exactly what
+## we want.  Yay!  Note: for some reason libtool 1.4 doesn't like
+## it if -MD -MP comes after the -MF stuff.  Hmm.
+## Unfortunately, FreeBSD c89 acceptance of flags depends upon
+## the command line argument order; so add the flags where they
+## appear in depend2.am.  Note that the slowdown incurred here
+## affects only configure: in makefiles, %FASTDEP% shortcuts this.
+  for arg
+  do
+    case $arg in
+    -c) set fnord "$@" -MT "$object" -MD -MP -MF "$tmpdepfile" "$arg" ;;
+    *)  set fnord "$@" "$arg" ;;
+    esac
+    shift # fnord
+    shift # $arg
+  done
+  "$@"
+  stat=$?
+  if test $stat -eq 0; then :
+  else
+    rm -f "$tmpdepfile"
+    exit $stat
+  fi
+  mv "$tmpdepfile" "$depfile"
+  ;;
+
+gcc)
+## There are various ways to get dependency output from gcc.  Here's
+## why we pick this rather obscure method:
+## - Don't want to use -MD because we'd like the dependencies to end
+##   up in a subdir.  Having to rename by hand is ugly.
+##   (We might end up doing this anyway to support other compilers.)
+## - The DEPENDENCIES_OUTPUT environment variable makes gcc act like
+##   -MM, not -M (despite what the docs say).
+## - Using -M directly means running the compiler twice (even worse
+##   than renaming).
+  if test -z "$gccflag"; then
+    gccflag=-MD,
+  fi
+  "$@" -Wp,"$gccflag$tmpdepfile"
+  stat=$?
+  if test $stat -eq 0; then :
+  else
+    rm -f "$tmpdepfile"
+    exit $stat
+  fi
+  rm -f "$depfile"
+  echo "$object : \\" > "$depfile"
+  alpha=ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz
+## The second -e expression handles DOS-style file names with drive letters.
+  sed -e 's/^[^:]*: / /' \
+      -e 's/^['$alpha']:\/[^:]*: / /' < "$tmpdepfile" >> "$depfile"
+## This next piece of magic avoids the `deleted header file' problem.
+## The problem is that when a header file which appears in a .P file
+## is deleted, the dependency causes make to die (because there is
+## typically no way to rebuild the header).  We avoid this by adding
+## dummy dependencies for each header file.  Too bad gcc doesn't do
+## this for us directly.
+  tr ' ' '
+' < "$tmpdepfile" |
+## Some versions of gcc put a space before the `:'.  On the theory
+## that the space means something, we add a space to the output as
+## well.  hp depmode also adds that space, but also prefixes the VPATH
+## to the object.  Take care to not repeat it in the output.
+## Some versions of the HPUX 10.20 sed can't process this invocation
+## correctly.  Breaking it into two sed invocations is a workaround.
+    sed -e 's/^\\$//' -e '/^$/d' -e "s|.*$object$||" -e '/:$/d' \
+      | sed -e 's/$/ :/' >> "$depfile"
+  rm -f "$tmpdepfile"
+  ;;
+
+hp)
+  # This case exists only to let depend.m4 do its work.  It works by
+  # looking at the text of this script.  This case will never be run,
+  # since it is checked for above.
+  exit 1
+  ;;
+
+sgi)
+  if test "$libtool" = yes; then
+    "$@" "-Wp,-MDupdate,$tmpdepfile"
+  else
+    "$@" -MDupdate "$tmpdepfile"
+  fi
+  stat=$?
+  if test $stat -eq 0; then :
+  else
+    rm -f "$tmpdepfile"
+    exit $stat
+  fi
+  rm -f "$depfile"
+
+  if test -f "$tmpdepfile"; then  # yes, the sourcefile depend on other files
+    echo "$object : \\" > "$depfile"
+
+    # Clip off the initial element (the dependent).  Don't try to be
+    # clever and replace this with sed code, as IRIX sed won't handle
+    # lines with more than a fixed number of characters (4096 in
+    # IRIX 6.2 sed, 8192 in IRIX 6.5).  We also remove comment lines;
+    # the IRIX cc adds comments like `#:fec' to the end of the
+    # dependency line.
+    tr ' ' '
+' < "$tmpdepfile" \
+    | sed -e 's/^.*\.o://' -e 's/#.*$//' -e '/^$/ d' | \
+    tr '
+' ' ' >> "$depfile"
+    echo >> "$depfile"
+
+    # The second pass generates a dummy entry for each header file.
+    tr ' ' '
+' < "$tmpdepfile" \
+   | sed -e 's/^.*\.o://' -e 's/#.*$//' -e '/^$/ d' -e 's/$/:/' \
+   >> "$depfile"
+  else
+    # The sourcefile does not contain any dependencies, so just
+    # store a dummy comment line, to avoid errors with the Makefile
+    # "include basename.Plo" scheme.
+    echo "#dummy" > "$depfile"
+  fi
+  rm -f "$tmpdepfile"
+  ;;
+
+aix)
+  # The C for AIX Compiler uses -M and outputs the dependencies
+  # in a .u file.  In older versions, this file always lives in the
+  # current directory.  Also, the AIX compiler puts `$object:' at the
+  # start of each line; $object doesn't have directory information.
+  # Version 6 uses the directory in both cases.
+  dir=`echo "$object" | sed -e 's|/[^/]*$|/|'`
+  test "x$dir" = "x$object" && dir=
+  base=`echo "$object" | sed -e 's|^.*/||' -e 's/\.o$//' -e 's/\.lo$//'`
+  if test "$libtool" = yes; then
+    tmpdepfile1=$dir$base.u
+    tmpdepfile2=$base.u
+    tmpdepfile3=$dir.libs/$base.u
+    "$@" -Wc,-M
+  else
+    tmpdepfile1=$dir$base.u
+    tmpdepfile2=$dir$base.u
+    tmpdepfile3=$dir$base.u
+    "$@" -M
+  fi
+  stat=$?
+
+  if test $stat -eq 0; then :
+  else
+    rm -f "$tmpdepfile1" "$tmpdepfile2" "$tmpdepfile3"
+    exit $stat
+  fi
+
+  for tmpdepfile in "$tmpdepfile1" "$tmpdepfile2" "$tmpdepfile3"
+  do
+    test -f "$tmpdepfile" && break
+  done
+  if test -f "$tmpdepfile"; then
+    # Each line is of the form `foo.o: dependent.h'.
+    # Do two passes, one to just change these to
+    # `$object: dependent.h' and one to simply `dependent.h:'.
+    sed -e "s,^.*\.[a-z]*:,$object:," < "$tmpdepfile" > "$depfile"
+    # That's a tab and a space in the [].
+    sed -e 's,^.*\.[a-z]*:[	 ]*,,' -e 's,$,:,' < "$tmpdepfile" >> "$depfile"
+  else
+    # The sourcefile does not contain any dependencies, so just
+    # store a dummy comment line, to avoid errors with the Makefile
+    # "include basename.Plo" scheme.
+    echo "#dummy" > "$depfile"
+  fi
+  rm -f "$tmpdepfile"
+  ;;
+
+icc)
+  # Intel's C compiler understands `-MD -MF file'.  However on
+  #    icc -MD -MF foo.d -c -o sub/foo.o sub/foo.c
+  # ICC 7.0 will fill foo.d with something like
+  #    foo.o: sub/foo.c
+  #    foo.o: sub/foo.h
+  # which is wrong.  We want:
+  #    sub/foo.o: sub/foo.c
+  #    sub/foo.o: sub/foo.h
+  #    sub/foo.c:
+  #    sub/foo.h:
+  # ICC 7.1 will output
+  #    foo.o: sub/foo.c sub/foo.h
+  # and will wrap long lines using \ :
+  #    foo.o: sub/foo.c ... \
+  #     sub/foo.h ... \
+  #     ...
+
+  "$@" -MD -MF "$tmpdepfile"
+  stat=$?
+  if test $stat -eq 0; then :
+  else
+    rm -f "$tmpdepfile"
+    exit $stat
+  fi
+  rm -f "$depfile"
+  # Each line is of the form `foo.o: dependent.h',
+  # or `foo.o: dep1.h dep2.h \', or ` dep3.h dep4.h \'.
+  # Do two passes, one to just change these to
+  # `$object: dependent.h' and one to simply `dependent.h:'.
+  sed "s,^[^:]*:,$object :," < "$tmpdepfile" > "$depfile"
+  # Some versions of the HPUX 10.20 sed can't process this invocation
+  # correctly.  Breaking it into two sed invocations is a workaround.
+  sed 's,^[^:]*: \(.*\)$,\1,;s/^\\$//;/^$/d;/:$/d' < "$tmpdepfile" |
+    sed -e 's/$/ :/' >> "$depfile"
+  rm -f "$tmpdepfile"
+  ;;
+
+hp2)
+  # The "hp" stanza above does not work with aCC (C++) and HP's ia64
+  # compilers, which have integrated preprocessors.  The correct option
+  # to use with these is +Maked; it writes dependencies to a file named
+  # 'foo.d', which lands next to the object file, wherever that
+  # happens to be.
+  # Much of this is similar to the tru64 case; see comments there.
+  dir=`echo "$object" | sed -e 's|/[^/]*$|/|'`
+  test "x$dir" = "x$object" && dir=
+  base=`echo "$object" | sed -e 's|^.*/||' -e 's/\.o$//' -e 's/\.lo$//'`
+  if test "$libtool" = yes; then
+    tmpdepfile1=$dir$base.d
+    tmpdepfile2=$dir.libs/$base.d
+    "$@" -Wc,+Maked
+  else
+    tmpdepfile1=$dir$base.d
+    tmpdepfile2=$dir$base.d
+    "$@" +Maked
+  fi
+  stat=$?
+  if test $stat -eq 0; then :
+  else
+     rm -f "$tmpdepfile1" "$tmpdepfile2"
+     exit $stat
+  fi
+
+  for tmpdepfile in "$tmpdepfile1" "$tmpdepfile2"
+  do
+    test -f "$tmpdepfile" && break
+  done
+  if test -f "$tmpdepfile"; then
+    sed -e "s,^.*\.[a-z]*:,$object:," "$tmpdepfile" > "$depfile"
+    # Add `dependent.h:' lines.
+    sed -ne '2,${
+	       s/^ *//
+	       s/ \\*$//
+	       s/$/:/
+	       p
+	     }' "$tmpdepfile" >> "$depfile"
+  else
+    echo "#dummy" > "$depfile"
+  fi
+  rm -f "$tmpdepfile" "$tmpdepfile2"
+  ;;
+
+tru64)
+   # The Tru64 compiler uses -MD to generate dependencies as a side
+   # effect.  `cc -MD -o foo.o ...' puts the dependencies into `foo.o.d'.
+   # At least on Alpha/Redhat 6.1, Compaq CCC V6.2-504 seems to put
+   # dependencies in `foo.d' instead, so we check for that too.
+   # Subdirectories are respected.
+   dir=`echo "$object" | sed -e 's|/[^/]*$|/|'`
+   test "x$dir" = "x$object" && dir=
+   base=`echo "$object" | sed -e 's|^.*/||' -e 's/\.o$//' -e 's/\.lo$//'`
+
+   if test "$libtool" = yes; then
+      # With Tru64 cc, shared objects can also be used to make a
+      # static library.  This mechanism is used in libtool 1.4 series to
+      # handle both shared and static libraries in a single compilation.
+      # With libtool 1.4, dependencies were output in $dir.libs/$base.lo.d.
+      #
+      # With libtool 1.5 this exception was removed, and libtool now
+      # generates 2 separate objects for the 2 libraries.  These two
+      # compilations output dependencies in $dir.libs/$base.o.d and
+      # in $dir$base.o.d.  We have to check for both files, because
+      # one of the two compilations can be disabled.  We should prefer
+      # $dir$base.o.d over $dir.libs/$base.o.d because the latter is
+      # automatically cleaned when .libs/ is deleted, while ignoring
+      # the former would cause a distcleancheck panic.
+      tmpdepfile1=$dir.libs/$base.lo.d   # libtool 1.4
+      tmpdepfile2=$dir$base.o.d          # libtool 1.5
+      tmpdepfile3=$dir.libs/$base.o.d    # libtool 1.5
+      tmpdepfile4=$dir.libs/$base.d      # Compaq CCC V6.2-504
+      "$@" -Wc,-MD
+   else
+      tmpdepfile1=$dir$base.o.d
+      tmpdepfile2=$dir$base.d
+      tmpdepfile3=$dir$base.d
+      tmpdepfile4=$dir$base.d
+      "$@" -MD
+   fi
+
+   stat=$?
+   if test $stat -eq 0; then :
+   else
+      rm -f "$tmpdepfile1" "$tmpdepfile2" "$tmpdepfile3" "$tmpdepfile4"
+      exit $stat
+   fi
+
+   for tmpdepfile in "$tmpdepfile1" "$tmpdepfile2" "$tmpdepfile3" "$tmpdepfile4"
+   do
+     test -f "$tmpdepfile" && break
+   done
+   if test -f "$tmpdepfile"; then
+      sed -e "s,^.*\.[a-z]*:,$object:," < "$tmpdepfile" > "$depfile"
+      # That's a tab and a space in the [].
+      sed -e 's,^.*\.[a-z]*:[	 ]*,,' -e 's,$,:,' < "$tmpdepfile" >> "$depfile"
+   else
+      echo "#dummy" > "$depfile"
+   fi
+   rm -f "$tmpdepfile"
+   ;;
+
+msvc7)
+  if test "$libtool" = yes; then
+    showIncludes=-Wc,-showIncludes
+  else
+    showIncludes=-showIncludes
+  fi
+  "$@" $showIncludes > "$tmpdepfile"
+  stat=$?
+  grep -v '^Note: including file: ' "$tmpdepfile"
+  if test "$stat" = 0; then :
+  else
+    rm -f "$tmpdepfile"
+    exit $stat
+  fi
+  rm -f "$depfile"
+  echo "$object : \\" > "$depfile"
+  # The first sed program below extracts the file names and escapes
+  # backslashes for cygpath.  The second sed program outputs the file
+  # name when reading, but also accumulates all include files in the
+  # hold buffer in order to output them again at the end.  This only
+  # works with sed implementations that can handle large buffers.
+  sed < "$tmpdepfile" -n '
+/^Note: including file:  *\(.*\)/ {
+  s//\1/
+  s/\\/\\\\/g
+  p
+}' | $cygpath_u | sort -u | sed -n '
+s/ /\\ /g
+s/\(.*\)/	\1 \\/p
+s/.\(.*\) \\/\1:/
+H
+$ {
+  s/.*/	/
+  G
+  p
+}' >> "$depfile"
+  rm -f "$tmpdepfile"
+  ;;
+
+msvc7msys)
+  # This case exists only to let depend.m4 do its work.  It works by
+  # looking at the text of this script.  This case will never be run,
+  # since it is checked for above.
+  exit 1
+  ;;
+
+#nosideeffect)
+  # This comment above is used by automake to tell side-effect
+  # dependency tracking mechanisms from slower ones.
+
+dashmstdout)
+  # Important note: in order to support this mode, a compiler *must*
+  # always write the preprocessed file to stdout, regardless of -o.
+  "$@" || exit $?
+
+  # Remove the call to Libtool.
+  if test "$libtool" = yes; then
+    while test "X$1" != 'X--mode=compile'; do
+      shift
+    done
+    shift
+  fi
+
+  # Remove `-o $object'.
+  IFS=" "
+  for arg
+  do
+    case $arg in
+    -o)
+      shift
+      ;;
+    $object)
+      shift
+      ;;
+    *)
+      set fnord "$@" "$arg"
+      shift # fnord
+      shift # $arg
+      ;;
+    esac
+  done
+
+  test -z "$dashmflag" && dashmflag=-M
+  # Require at least two characters before searching for `:'
+  # in the target name.  This is to cope with DOS-style filenames:
+  # a dependency such as `c:/foo/bar' could be seen as target `c' otherwise.
+  "$@" $dashmflag |
+    sed 's:^[  ]*[^: ][^:][^:]*\:[    ]*:'"$object"'\: :' > "$tmpdepfile"
+  rm -f "$depfile"
+  cat < "$tmpdepfile" > "$depfile"
+  tr ' ' '
+' < "$tmpdepfile" | \
+## Some versions of the HPUX 10.20 sed can't process this invocation
+## correctly.  Breaking it into two sed invocations is a workaround.
+    sed -e 's/^\\$//' -e '/^$/d' -e '/:$/d' | sed -e 's/$/ :/' >> "$depfile"
+  rm -f "$tmpdepfile"
+  ;;
+
+dashXmstdout)
+  # This case only exists to satisfy depend.m4.  It is never actually
+  # run, as this mode is specially recognized in the preamble.
+  exit 1
+  ;;
+
+makedepend)
+  "$@" || exit $?
+  # Remove any Libtool call
+  if test "$libtool" = yes; then
+    while test "X$1" != 'X--mode=compile'; do
+      shift
+    done
+    shift
+  fi
+  # X makedepend
+  shift
+  cleared=no eat=no
+  for arg
+  do
+    case $cleared in
+    no)
+      set ""; shift
+      cleared=yes ;;
+    esac
+    if test $eat = yes; then
+      eat=no
+      continue
+    fi
+    case "$arg" in
+    -D*|-I*)
+      set fnord "$@" "$arg"; shift ;;
+    # Strip any option that makedepend may not understand.  Remove
+    # the object too, otherwise makedepend will parse it as a source file.
+    -arch)
+      eat=yes ;;
+    -*|$object)
+      ;;
+    *)
+      set fnord "$@" "$arg"; shift ;;
+    esac
+  done
+  obj_suffix=`echo "$object" | sed 's/^.*\././'`
+  touch "$tmpdepfile"
+  ${MAKEDEPEND-makedepend} -o"$obj_suffix" -f"$tmpdepfile" "$@"
+  rm -f "$depfile"
+  # makedepend may prepend the VPATH from the source file name to the object.
+  # No need to regex-escape $object, excess matching of '.' is harmless.
+  sed "s|^.*\($object *:\)|\1|" "$tmpdepfile" > "$depfile"
+  sed '1,2d' "$tmpdepfile" | tr ' ' '
+' | \
+## Some versions of the HPUX 10.20 sed can't process this invocation
+## correctly.  Breaking it into two sed invocations is a workaround.
+    sed -e 's/^\\$//' -e '/^$/d' -e '/:$/d' | sed -e 's/$/ :/' >> "$depfile"
+  rm -f "$tmpdepfile" "$tmpdepfile".bak
+  ;;
+
+cpp)
+  # Important note: in order to support this mode, a compiler *must*
+  # always write the preprocessed file to stdout.
+  "$@" || exit $?
+
+  # Remove the call to Libtool.
+  if test "$libtool" = yes; then
+    while test "X$1" != 'X--mode=compile'; do
+      shift
+    done
+    shift
+  fi
+
+  # Remove `-o $object'.
+  IFS=" "
+  for arg
+  do
+    case $arg in
+    -o)
+      shift
+      ;;
+    $object)
+      shift
+      ;;
+    *)
+      set fnord "$@" "$arg"
+      shift # fnord
+      shift # $arg
+      ;;
+    esac
+  done
+
+  "$@" -E |
+    sed -n -e '/^# [0-9][0-9]* "\([^"]*\)".*/ s:: \1 \\:p' \
+       -e '/^#line [0-9][0-9]* "\([^"]*\)".*/ s:: \1 \\:p' |
+    sed '$ s: \\$::' > "$tmpdepfile"
+  rm -f "$depfile"
+  echo "$object : \\" > "$depfile"
+  cat < "$tmpdepfile" >> "$depfile"
+  sed < "$tmpdepfile" '/^$/d;s/^ //;s/ \\$//;s/$/ :/' >> "$depfile"
+  rm -f "$tmpdepfile"
+  ;;
+
+msvisualcpp)
+  # Important note: in order to support this mode, a compiler *must*
+  # always write the preprocessed file to stdout.
+  "$@" || exit $?
+
+  # Remove the call to Libtool.
+  if test "$libtool" = yes; then
+    while test "X$1" != 'X--mode=compile'; do
+      shift
+    done
+    shift
+  fi
+
+  IFS=" "
+  for arg
+  do
+    case "$arg" in
+    -o)
+      shift
+      ;;
+    $object)
+      shift
+      ;;
+    "-Gm"|"/Gm"|"-Gi"|"/Gi"|"-ZI"|"/ZI")
+	set fnord "$@"
+	shift
+	shift
+	;;
+    *)
+	set fnord "$@" "$arg"
+	shift
+	shift
+	;;
+    esac
+  done
+  "$@" -E 2>/dev/null |
+  sed -n '/^#line [0-9][0-9]* "\([^"]*\)"/ s::\1:p' | $cygpath_u | sort -u > "$tmpdepfile"
+  rm -f "$depfile"
+  echo "$object : \\" > "$depfile"
+  sed < "$tmpdepfile" -n -e 's% %\\ %g' -e '/^\(.*\)$/ s::	\1 \\:p' >> "$depfile"
+  echo "	" >> "$depfile"
+  sed < "$tmpdepfile" -n -e 's% %\\ %g' -e '/^\(.*\)$/ s::\1\::p' >> "$depfile"
+  rm -f "$tmpdepfile"
+  ;;
+
+msvcmsys)
+  # This case exists only to let depend.m4 do its work.  It works by
+  # looking at the text of this script.  This case will never be run,
+  # since it is checked for above.
+  exit 1
+  ;;
+
+none)
+  exec "$@"
+  ;;
+
+*)
+  echo "Unknown depmode $depmode" 1>&2
+  exit 1
+  ;;
+esac
+
+exit 0
+
+# Local Variables:
+# mode: shell-script
+# sh-indentation: 2
+# eval: (add-hook 'write-file-hooks 'time-stamp)
+# time-stamp-start: "scriptversion="
+# time-stamp-format: "%:y-%02m-%02d.%02H"
+# time-stamp-time-zone: "UTC"
+# time-stamp-end: "; # UTC"
+# End:
diff --git a/elist.h b/elist.h
new file mode 100644
index 0000000000..431472f7b7
--- /dev/null
+++ b/elist.h
@@ -0,0 +1,251 @@
+#ifndef _LINUX_LIST_H
+#define _LINUX_LIST_H
+
+/*
+ * Simple doubly linked list implementation.
+ *
+ * Some of the internal functions ("__xxx") are useful when
+ * manipulating whole lists rather than single entries, as
+ * sometimes we already know the next/prev entries and we can
+ * generate better code by using them directly rather than
+ * using the generic single-entry routines.
+ */
+
+struct list_head {
+	struct list_head *next, *prev;
+};
+
+#define LIST_HEAD_INIT(name) { &(name), &(name) }
+
+#define LIST_HEAD(name) \
+	struct list_head name = LIST_HEAD_INIT(name)
+
+#define INIT_LIST_HEAD(ptr) do { \
+	(ptr)->next = (ptr); (ptr)->prev = (ptr); \
+} while (0)
+
+/*
+ * Insert a new entry between two known consecutive entries.
+ *
+ * This is only for internal list manipulation where we know
+ * the prev/next entries already!
+ */
+static __inline void __list_add(struct list_head *lnew,
+			      struct list_head *prev,
+			      struct list_head *next)
+{
+	next->prev = lnew;
+	lnew->next = next;
+	lnew->prev = prev;
+	prev->next = lnew;
+}
+
+/**
+ * list_add - add a new entry
+ * @new: new entry to be added
+ * @head: list head to add it after
+ *
+ * Insert a new entry after the specified head.
+ * This is good for implementing stacks.
+ */
+static __inline void list_add(struct list_head *lnew, struct list_head *head)
+{
+	__list_add(lnew, head, head->next);
+}
+
+/**
+ * list_add_tail - add a new entry
+ * @new: new entry to be added
+ * @head: list head to add it before
+ *
+ * Insert a new entry before the specified head.
+ * This is useful for implementing queues.
+ */
+static __inline void list_add_tail(struct list_head *lnew, struct list_head *head)
+{
+	__list_add(lnew, head->prev, head);
+}
+
+/*
+ * Delete a list entry by making the prev/next entries
+ * point to each other.
+ *
+ * This is only for internal list manipulation where we know
+ * the prev/next entries already!
+ */
+static __inline void __list_del(struct list_head *prev, struct list_head *next)
+{
+	next->prev = prev;
+	prev->next = next;
+}
+
+/**
+ * list_del - deletes entry from list.
+ * @entry: the element to delete from the list.
+ * Note: list_empty on entry does not return true after this, the entry is in an undefined state.
+ */
+static __inline void list_del(struct list_head *entry)
+{
+	__list_del(entry->prev, entry->next);
+	entry->next = (struct list_head *) 0;
+	entry->prev = (struct list_head *) 0;
+}
+
+/**
+ * list_del_init - deletes entry from list and reinitialize it.
+ * @entry: the element to delete from the list.
+ */
+static __inline void list_del_init(struct list_head *entry)
+{
+	__list_del(entry->prev, entry->next);
+	INIT_LIST_HEAD(entry);
+}
+
+/**
+ * list_move - delete from one list and add as another's head
+ * @list: the entry to move
+ * @head: the head that will precede our entry
+ */
+static __inline void list_move(struct list_head *list, struct list_head *head)
+{
+        __list_del(list->prev, list->next);
+        list_add(list, head);
+}
+
+/**
+ * list_move_tail - delete from one list and add as another's tail
+ * @list: the entry to move
+ * @head: the head that will follow our entry
+ */
+static __inline void list_move_tail(struct list_head *list,
+				  struct list_head *head)
+{
+        __list_del(list->prev, list->next);
+        list_add_tail(list, head);
+}
+
+/**
+ * list_empty - tests whether a list is empty
+ * @head: the list to test.
+ */
+static __inline int list_empty(struct list_head *head)
+{
+	return head->next == head;
+}
+
+static __inline void __list_splice(struct list_head *list,
+				 struct list_head *head)
+{
+	struct list_head *first = list->next;
+	struct list_head *last = list->prev;
+	struct list_head *at = head->next;
+
+	first->prev = head;
+	head->next = first;
+
+	last->next = at;
+	at->prev = last;
+}
+
+/**
+ * list_splice - join two lists
+ * @list: the new list to add.
+ * @head: the place to add it in the first list.
+ */
+static __inline void list_splice(struct list_head *list, struct list_head *head)
+{
+	if (!list_empty(list))
+		__list_splice(list, head);
+}
+
+/**
+ * list_splice_init - join two lists and reinitialise the emptied list.
+ * @list: the new list to add.
+ * @head: the place to add it in the first list.
+ *
+ * The list at @list is reinitialised
+ */
+static __inline void list_splice_init(struct list_head *list,
+				    struct list_head *head)
+{
+	if (!list_empty(list)) {
+		__list_splice(list, head);
+		INIT_LIST_HEAD(list);
+	}
+}
+
+/**
+ * list_entry - get the struct for this entry
+ * @ptr:	the &struct list_head pointer.
+ * @type:	the type of the struct this is embedded in.
+ * @member:	the name of the list_struct within the struct.
+ */
+#define list_entry(ptr, type, member) \
+	((type *)((char *)(ptr)-(unsigned long)(&((type *)0)->member)))
+
+/**
+ * list_for_each	-	iterate over a list
+ * @pos:	the &struct list_head to use as a loop counter.
+ * @head:	the head for your list.
+ */
+#define list_for_each(pos, head) \
+	for (pos = (head)->next; pos != (head); \
+        	pos = pos->next)
+/**
+ * list_for_each_prev	-	iterate over a list backwards
+ * @pos:	the &struct list_head to use as a loop counter.
+ * @head:	the head for your list.
+ */
+#define list_for_each_prev(pos, head) \
+	for (pos = (head)->prev; pos != (head); \
+        	pos = pos->prev)
+
+/**
+ * list_for_each_safe	-	iterate over a list safe against removal of list entry
+ * @pos:	the &struct list_head to use as a loop counter.
+ * @n:		another &struct list_head to use as temporary storage
+ * @head:	the head for your list.
+ */
+#define list_for_each_safe(pos, n, head) \
+	for (pos = (head)->next, n = pos->next; pos != (head); \
+		pos = n, n = pos->next)
+
+/**
+ * list_for_each_entry	-	iterate over list of given type
+ * @pos:	the type * to use as a loop counter.
+ * @head:	the head for your list.
+ * @member:	the name of the list_struct within the struct.
+ */
+#define list_for_each_entry(pos, head, member, tpos)				\
+	for (pos = list_entry((head)->next, tpos, member);	\
+	     &pos->member != (head); 					\
+	     pos = list_entry(pos->member.next, tpos, member))
+
+/**
+ * list_for_each_entry_safe - iterate over list of given type safe against removal of list entry
+ * @pos:	the type * to use as a loop counter.
+ * @n:		another type * to use as temporary storage
+ * @head:	the head for your list.
+ * @member:	the name of the list_struct within the struct.
+ */
+#define list_for_each_entry_safe(pos, n, head, member, tpos, tn)			\
+	for (pos = list_entry((head)->next, tpos, member),	\
+		n = list_entry(pos->member.next, tpos, member);	\
+	     &pos->member != (head); 					\
+	     pos = n, n = list_entry(n->member.next, tn, member))
+
+/**
+ * list_for_each_entry_continue -       iterate over list of given type
+ *                      continuing after existing point
+ * @pos:        the type * to use as a loop counter.
+ * @head:       the head for your list.
+ * @member:     the name of the list_struct within the struct.
+ */
+#define list_for_each_entry_continue(pos, head, member, tpos)			\
+	for (pos = list_entry(pos->member.next, tpos, member),	\
+		     prefetch(pos->member.next);			\
+	     &pos->member != (head);					\
+	     pos = list_entry(pos->member.next, tpos, member),	\
+		     prefetch(pos->member.next))
+
+#endif
diff --git a/files.txt b/files.txt
new file mode 100644
index 0000000000..2b7db9d84f
--- /dev/null
+++ b/files.txt
@@ -0,0 +1,30 @@
+blake512.cu
+blake.c
+combine.cu
+compat.h
+cpu-miner.c
+cpuminer-config.h
+cuda_blake512.h
+cuda_combine.h
+cuda_groestl512.h
+cuda_hefty1.h
+cuda_keccak512.h
+cuda_sha256.h
+elist.h
+groestl512.cu
+groestl.c
+heavy.c
+hefty1.c
+hefty1.cu
+hefty1.h
+keccak512.cu
+keccak.c
+miner.h
+scrypt.c
+sha256.cu
+sha2.c
+sph_blake.h
+sph_groestl.h
+sph_keccak.h
+sph_types.h
+util.c
diff --git a/fugue.c b/fugue.c
new file mode 100644
index 0000000000..85767c9830
--- /dev/null
+++ b/fugue.c
@@ -0,0 +1,1208 @@
+#include <stddef.h>
+#include <string.h>
+
+#include "sph_fugue.h"
+
+#ifdef __cplusplus
+extern "C"{
+#endif
+
+#ifdef _MSC_VER
+#pragma warning (disable: 4146)
+#endif
+
+static const sph_u32 IV224[] = {
+	SPH_C32(0xf4c9120d), SPH_C32(0x6286f757), SPH_C32(0xee39e01c),
+	SPH_C32(0xe074e3cb), SPH_C32(0xa1127c62), SPH_C32(0x9a43d215),
+	SPH_C32(0xbd8d679a)
+};
+
+static const sph_u32 IV256[] = {
+	SPH_C32(0xe952bdde), SPH_C32(0x6671135f), SPH_C32(0xe0d4f668),
+	SPH_C32(0xd2b0b594), SPH_C32(0xf96c621d), SPH_C32(0xfbf929de),
+	SPH_C32(0x9149e899), SPH_C32(0x34f8c248)
+};
+
+static const sph_u32 IV384[] = {
+	SPH_C32(0xaa61ec0d), SPH_C32(0x31252e1f), SPH_C32(0xa01db4c7),
+	SPH_C32(0x00600985), SPH_C32(0x215ef44a), SPH_C32(0x741b5e9c),
+	SPH_C32(0xfa693e9a), SPH_C32(0x473eb040), SPH_C32(0xe502ae8a),
+	SPH_C32(0xa99c25e0), SPH_C32(0xbc95517c), SPH_C32(0x5c1095a1)
+};
+
+static const sph_u32 IV512[] = {
+	SPH_C32(0x8807a57e), SPH_C32(0xe616af75), SPH_C32(0xc5d3e4db),
+	SPH_C32(0xac9ab027), SPH_C32(0xd915f117), SPH_C32(0xb6eecc54),
+	SPH_C32(0x06e8020b), SPH_C32(0x4a92efd1), SPH_C32(0xaac6e2c9),
+	SPH_C32(0xddb21398), SPH_C32(0xcae65838), SPH_C32(0x437f203f),
+	SPH_C32(0x25ea78e7), SPH_C32(0x951fddd6), SPH_C32(0xda6ed11d),
+	SPH_C32(0xe13e3567)
+};
+
+static const sph_u32 mixtab0[] = {
+	SPH_C32(0x63633297), SPH_C32(0x7c7c6feb), SPH_C32(0x77775ec7),
+	SPH_C32(0x7b7b7af7), SPH_C32(0xf2f2e8e5), SPH_C32(0x6b6b0ab7),
+	SPH_C32(0x6f6f16a7), SPH_C32(0xc5c56d39), SPH_C32(0x303090c0),
+	SPH_C32(0x01010704), SPH_C32(0x67672e87), SPH_C32(0x2b2bd1ac),
+	SPH_C32(0xfefeccd5), SPH_C32(0xd7d71371), SPH_C32(0xabab7c9a),
+	SPH_C32(0x767659c3), SPH_C32(0xcaca4005), SPH_C32(0x8282a33e),
+	SPH_C32(0xc9c94909), SPH_C32(0x7d7d68ef), SPH_C32(0xfafad0c5),
+	SPH_C32(0x5959947f), SPH_C32(0x4747ce07), SPH_C32(0xf0f0e6ed),
+	SPH_C32(0xadad6e82), SPH_C32(0xd4d41a7d), SPH_C32(0xa2a243be),
+	SPH_C32(0xafaf608a), SPH_C32(0x9c9cf946), SPH_C32(0xa4a451a6),
+	SPH_C32(0x727245d3), SPH_C32(0xc0c0762d), SPH_C32(0xb7b728ea),
+	SPH_C32(0xfdfdc5d9), SPH_C32(0x9393d47a), SPH_C32(0x2626f298),
+	SPH_C32(0x363682d8), SPH_C32(0x3f3fbdfc), SPH_C32(0xf7f7f3f1),
+	SPH_C32(0xcccc521d), SPH_C32(0x34348cd0), SPH_C32(0xa5a556a2),
+	SPH_C32(0xe5e58db9), SPH_C32(0xf1f1e1e9), SPH_C32(0x71714cdf),
+	SPH_C32(0xd8d83e4d), SPH_C32(0x313197c4), SPH_C32(0x15156b54),
+	SPH_C32(0x04041c10), SPH_C32(0xc7c76331), SPH_C32(0x2323e98c),
+	SPH_C32(0xc3c37f21), SPH_C32(0x18184860), SPH_C32(0x9696cf6e),
+	SPH_C32(0x05051b14), SPH_C32(0x9a9aeb5e), SPH_C32(0x0707151c),
+	SPH_C32(0x12127e48), SPH_C32(0x8080ad36), SPH_C32(0xe2e298a5),
+	SPH_C32(0xebeba781), SPH_C32(0x2727f59c), SPH_C32(0xb2b233fe),
+	SPH_C32(0x757550cf), SPH_C32(0x09093f24), SPH_C32(0x8383a43a),
+	SPH_C32(0x2c2cc4b0), SPH_C32(0x1a1a4668), SPH_C32(0x1b1b416c),
+	SPH_C32(0x6e6e11a3), SPH_C32(0x5a5a9d73), SPH_C32(0xa0a04db6),
+	SPH_C32(0x5252a553), SPH_C32(0x3b3ba1ec), SPH_C32(0xd6d61475),
+	SPH_C32(0xb3b334fa), SPH_C32(0x2929dfa4), SPH_C32(0xe3e39fa1),
+	SPH_C32(0x2f2fcdbc), SPH_C32(0x8484b126), SPH_C32(0x5353a257),
+	SPH_C32(0xd1d10169), SPH_C32(0x00000000), SPH_C32(0xededb599),
+	SPH_C32(0x2020e080), SPH_C32(0xfcfcc2dd), SPH_C32(0xb1b13af2),
+	SPH_C32(0x5b5b9a77), SPH_C32(0x6a6a0db3), SPH_C32(0xcbcb4701),
+	SPH_C32(0xbebe17ce), SPH_C32(0x3939afe4), SPH_C32(0x4a4aed33),
+	SPH_C32(0x4c4cff2b), SPH_C32(0x5858937b), SPH_C32(0xcfcf5b11),
+	SPH_C32(0xd0d0066d), SPH_C32(0xefefbb91), SPH_C32(0xaaaa7b9e),
+	SPH_C32(0xfbfbd7c1), SPH_C32(0x4343d217), SPH_C32(0x4d4df82f),
+	SPH_C32(0x333399cc), SPH_C32(0x8585b622), SPH_C32(0x4545c00f),
+	SPH_C32(0xf9f9d9c9), SPH_C32(0x02020e08), SPH_C32(0x7f7f66e7),
+	SPH_C32(0x5050ab5b), SPH_C32(0x3c3cb4f0), SPH_C32(0x9f9ff04a),
+	SPH_C32(0xa8a87596), SPH_C32(0x5151ac5f), SPH_C32(0xa3a344ba),
+	SPH_C32(0x4040db1b), SPH_C32(0x8f8f800a), SPH_C32(0x9292d37e),
+	SPH_C32(0x9d9dfe42), SPH_C32(0x3838a8e0), SPH_C32(0xf5f5fdf9),
+	SPH_C32(0xbcbc19c6), SPH_C32(0xb6b62fee), SPH_C32(0xdada3045),
+	SPH_C32(0x2121e784), SPH_C32(0x10107040), SPH_C32(0xffffcbd1),
+	SPH_C32(0xf3f3efe1), SPH_C32(0xd2d20865), SPH_C32(0xcdcd5519),
+	SPH_C32(0x0c0c2430), SPH_C32(0x1313794c), SPH_C32(0xececb29d),
+	SPH_C32(0x5f5f8667), SPH_C32(0x9797c86a), SPH_C32(0x4444c70b),
+	SPH_C32(0x1717655c), SPH_C32(0xc4c46a3d), SPH_C32(0xa7a758aa),
+	SPH_C32(0x7e7e61e3), SPH_C32(0x3d3db3f4), SPH_C32(0x6464278b),
+	SPH_C32(0x5d5d886f), SPH_C32(0x19194f64), SPH_C32(0x737342d7),
+	SPH_C32(0x60603b9b), SPH_C32(0x8181aa32), SPH_C32(0x4f4ff627),
+	SPH_C32(0xdcdc225d), SPH_C32(0x2222ee88), SPH_C32(0x2a2ad6a8),
+	SPH_C32(0x9090dd76), SPH_C32(0x88889516), SPH_C32(0x4646c903),
+	SPH_C32(0xeeeebc95), SPH_C32(0xb8b805d6), SPH_C32(0x14146c50),
+	SPH_C32(0xdede2c55), SPH_C32(0x5e5e8163), SPH_C32(0x0b0b312c),
+	SPH_C32(0xdbdb3741), SPH_C32(0xe0e096ad), SPH_C32(0x32329ec8),
+	SPH_C32(0x3a3aa6e8), SPH_C32(0x0a0a3628), SPH_C32(0x4949e43f),
+	SPH_C32(0x06061218), SPH_C32(0x2424fc90), SPH_C32(0x5c5c8f6b),
+	SPH_C32(0xc2c27825), SPH_C32(0xd3d30f61), SPH_C32(0xacac6986),
+	SPH_C32(0x62623593), SPH_C32(0x9191da72), SPH_C32(0x9595c662),
+	SPH_C32(0xe4e48abd), SPH_C32(0x797974ff), SPH_C32(0xe7e783b1),
+	SPH_C32(0xc8c84e0d), SPH_C32(0x373785dc), SPH_C32(0x6d6d18af),
+	SPH_C32(0x8d8d8e02), SPH_C32(0xd5d51d79), SPH_C32(0x4e4ef123),
+	SPH_C32(0xa9a97292), SPH_C32(0x6c6c1fab), SPH_C32(0x5656b943),
+	SPH_C32(0xf4f4fafd), SPH_C32(0xeaeaa085), SPH_C32(0x6565208f),
+	SPH_C32(0x7a7a7df3), SPH_C32(0xaeae678e), SPH_C32(0x08083820),
+	SPH_C32(0xbaba0bde), SPH_C32(0x787873fb), SPH_C32(0x2525fb94),
+	SPH_C32(0x2e2ecab8), SPH_C32(0x1c1c5470), SPH_C32(0xa6a65fae),
+	SPH_C32(0xb4b421e6), SPH_C32(0xc6c66435), SPH_C32(0xe8e8ae8d),
+	SPH_C32(0xdddd2559), SPH_C32(0x747457cb), SPH_C32(0x1f1f5d7c),
+	SPH_C32(0x4b4bea37), SPH_C32(0xbdbd1ec2), SPH_C32(0x8b8b9c1a),
+	SPH_C32(0x8a8a9b1e), SPH_C32(0x70704bdb), SPH_C32(0x3e3ebaf8),
+	SPH_C32(0xb5b526e2), SPH_C32(0x66662983), SPH_C32(0x4848e33b),
+	SPH_C32(0x0303090c), SPH_C32(0xf6f6f4f5), SPH_C32(0x0e0e2a38),
+	SPH_C32(0x61613c9f), SPH_C32(0x35358bd4), SPH_C32(0x5757be47),
+	SPH_C32(0xb9b902d2), SPH_C32(0x8686bf2e), SPH_C32(0xc1c17129),
+	SPH_C32(0x1d1d5374), SPH_C32(0x9e9ef74e), SPH_C32(0xe1e191a9),
+	SPH_C32(0xf8f8decd), SPH_C32(0x9898e556), SPH_C32(0x11117744),
+	SPH_C32(0x696904bf), SPH_C32(0xd9d93949), SPH_C32(0x8e8e870e),
+	SPH_C32(0x9494c166), SPH_C32(0x9b9bec5a), SPH_C32(0x1e1e5a78),
+	SPH_C32(0x8787b82a), SPH_C32(0xe9e9a989), SPH_C32(0xcece5c15),
+	SPH_C32(0x5555b04f), SPH_C32(0x2828d8a0), SPH_C32(0xdfdf2b51),
+	SPH_C32(0x8c8c8906), SPH_C32(0xa1a14ab2), SPH_C32(0x89899212),
+	SPH_C32(0x0d0d2334), SPH_C32(0xbfbf10ca), SPH_C32(0xe6e684b5),
+	SPH_C32(0x4242d513), SPH_C32(0x686803bb), SPH_C32(0x4141dc1f),
+	SPH_C32(0x9999e252), SPH_C32(0x2d2dc3b4), SPH_C32(0x0f0f2d3c),
+	SPH_C32(0xb0b03df6), SPH_C32(0x5454b74b), SPH_C32(0xbbbb0cda),
+	SPH_C32(0x16166258)
+};
+
+static const sph_u32 mixtab1[] = {
+	SPH_C32(0x97636332), SPH_C32(0xeb7c7c6f), SPH_C32(0xc777775e),
+	SPH_C32(0xf77b7b7a), SPH_C32(0xe5f2f2e8), SPH_C32(0xb76b6b0a),
+	SPH_C32(0xa76f6f16), SPH_C32(0x39c5c56d), SPH_C32(0xc0303090),
+	SPH_C32(0x04010107), SPH_C32(0x8767672e), SPH_C32(0xac2b2bd1),
+	SPH_C32(0xd5fefecc), SPH_C32(0x71d7d713), SPH_C32(0x9aabab7c),
+	SPH_C32(0xc3767659), SPH_C32(0x05caca40), SPH_C32(0x3e8282a3),
+	SPH_C32(0x09c9c949), SPH_C32(0xef7d7d68), SPH_C32(0xc5fafad0),
+	SPH_C32(0x7f595994), SPH_C32(0x074747ce), SPH_C32(0xedf0f0e6),
+	SPH_C32(0x82adad6e), SPH_C32(0x7dd4d41a), SPH_C32(0xbea2a243),
+	SPH_C32(0x8aafaf60), SPH_C32(0x469c9cf9), SPH_C32(0xa6a4a451),
+	SPH_C32(0xd3727245), SPH_C32(0x2dc0c076), SPH_C32(0xeab7b728),
+	SPH_C32(0xd9fdfdc5), SPH_C32(0x7a9393d4), SPH_C32(0x982626f2),
+	SPH_C32(0xd8363682), SPH_C32(0xfc3f3fbd), SPH_C32(0xf1f7f7f3),
+	SPH_C32(0x1dcccc52), SPH_C32(0xd034348c), SPH_C32(0xa2a5a556),
+	SPH_C32(0xb9e5e58d), SPH_C32(0xe9f1f1e1), SPH_C32(0xdf71714c),
+	SPH_C32(0x4dd8d83e), SPH_C32(0xc4313197), SPH_C32(0x5415156b),
+	SPH_C32(0x1004041c), SPH_C32(0x31c7c763), SPH_C32(0x8c2323e9),
+	SPH_C32(0x21c3c37f), SPH_C32(0x60181848), SPH_C32(0x6e9696cf),
+	SPH_C32(0x1405051b), SPH_C32(0x5e9a9aeb), SPH_C32(0x1c070715),
+	SPH_C32(0x4812127e), SPH_C32(0x368080ad), SPH_C32(0xa5e2e298),
+	SPH_C32(0x81ebeba7), SPH_C32(0x9c2727f5), SPH_C32(0xfeb2b233),
+	SPH_C32(0xcf757550), SPH_C32(0x2409093f), SPH_C32(0x3a8383a4),
+	SPH_C32(0xb02c2cc4), SPH_C32(0x681a1a46), SPH_C32(0x6c1b1b41),
+	SPH_C32(0xa36e6e11), SPH_C32(0x735a5a9d), SPH_C32(0xb6a0a04d),
+	SPH_C32(0x535252a5), SPH_C32(0xec3b3ba1), SPH_C32(0x75d6d614),
+	SPH_C32(0xfab3b334), SPH_C32(0xa42929df), SPH_C32(0xa1e3e39f),
+	SPH_C32(0xbc2f2fcd), SPH_C32(0x268484b1), SPH_C32(0x575353a2),
+	SPH_C32(0x69d1d101), SPH_C32(0x00000000), SPH_C32(0x99ededb5),
+	SPH_C32(0x802020e0), SPH_C32(0xddfcfcc2), SPH_C32(0xf2b1b13a),
+	SPH_C32(0x775b5b9a), SPH_C32(0xb36a6a0d), SPH_C32(0x01cbcb47),
+	SPH_C32(0xcebebe17), SPH_C32(0xe43939af), SPH_C32(0x334a4aed),
+	SPH_C32(0x2b4c4cff), SPH_C32(0x7b585893), SPH_C32(0x11cfcf5b),
+	SPH_C32(0x6dd0d006), SPH_C32(0x91efefbb), SPH_C32(0x9eaaaa7b),
+	SPH_C32(0xc1fbfbd7), SPH_C32(0x174343d2), SPH_C32(0x2f4d4df8),
+	SPH_C32(0xcc333399), SPH_C32(0x228585b6), SPH_C32(0x0f4545c0),
+	SPH_C32(0xc9f9f9d9), SPH_C32(0x0802020e), SPH_C32(0xe77f7f66),
+	SPH_C32(0x5b5050ab), SPH_C32(0xf03c3cb4), SPH_C32(0x4a9f9ff0),
+	SPH_C32(0x96a8a875), SPH_C32(0x5f5151ac), SPH_C32(0xbaa3a344),
+	SPH_C32(0x1b4040db), SPH_C32(0x0a8f8f80), SPH_C32(0x7e9292d3),
+	SPH_C32(0x429d9dfe), SPH_C32(0xe03838a8), SPH_C32(0xf9f5f5fd),
+	SPH_C32(0xc6bcbc19), SPH_C32(0xeeb6b62f), SPH_C32(0x45dada30),
+	SPH_C32(0x842121e7), SPH_C32(0x40101070), SPH_C32(0xd1ffffcb),
+	SPH_C32(0xe1f3f3ef), SPH_C32(0x65d2d208), SPH_C32(0x19cdcd55),
+	SPH_C32(0x300c0c24), SPH_C32(0x4c131379), SPH_C32(0x9dececb2),
+	SPH_C32(0x675f5f86), SPH_C32(0x6a9797c8), SPH_C32(0x0b4444c7),
+	SPH_C32(0x5c171765), SPH_C32(0x3dc4c46a), SPH_C32(0xaaa7a758),
+	SPH_C32(0xe37e7e61), SPH_C32(0xf43d3db3), SPH_C32(0x8b646427),
+	SPH_C32(0x6f5d5d88), SPH_C32(0x6419194f), SPH_C32(0xd7737342),
+	SPH_C32(0x9b60603b), SPH_C32(0x328181aa), SPH_C32(0x274f4ff6),
+	SPH_C32(0x5ddcdc22), SPH_C32(0x882222ee), SPH_C32(0xa82a2ad6),
+	SPH_C32(0x769090dd), SPH_C32(0x16888895), SPH_C32(0x034646c9),
+	SPH_C32(0x95eeeebc), SPH_C32(0xd6b8b805), SPH_C32(0x5014146c),
+	SPH_C32(0x55dede2c), SPH_C32(0x635e5e81), SPH_C32(0x2c0b0b31),
+	SPH_C32(0x41dbdb37), SPH_C32(0xade0e096), SPH_C32(0xc832329e),
+	SPH_C32(0xe83a3aa6), SPH_C32(0x280a0a36), SPH_C32(0x3f4949e4),
+	SPH_C32(0x18060612), SPH_C32(0x902424fc), SPH_C32(0x6b5c5c8f),
+	SPH_C32(0x25c2c278), SPH_C32(0x61d3d30f), SPH_C32(0x86acac69),
+	SPH_C32(0x93626235), SPH_C32(0x729191da), SPH_C32(0x629595c6),
+	SPH_C32(0xbde4e48a), SPH_C32(0xff797974), SPH_C32(0xb1e7e783),
+	SPH_C32(0x0dc8c84e), SPH_C32(0xdc373785), SPH_C32(0xaf6d6d18),
+	SPH_C32(0x028d8d8e), SPH_C32(0x79d5d51d), SPH_C32(0x234e4ef1),
+	SPH_C32(0x92a9a972), SPH_C32(0xab6c6c1f), SPH_C32(0x435656b9),
+	SPH_C32(0xfdf4f4fa), SPH_C32(0x85eaeaa0), SPH_C32(0x8f656520),
+	SPH_C32(0xf37a7a7d), SPH_C32(0x8eaeae67), SPH_C32(0x20080838),
+	SPH_C32(0xdebaba0b), SPH_C32(0xfb787873), SPH_C32(0x942525fb),
+	SPH_C32(0xb82e2eca), SPH_C32(0x701c1c54), SPH_C32(0xaea6a65f),
+	SPH_C32(0xe6b4b421), SPH_C32(0x35c6c664), SPH_C32(0x8de8e8ae),
+	SPH_C32(0x59dddd25), SPH_C32(0xcb747457), SPH_C32(0x7c1f1f5d),
+	SPH_C32(0x374b4bea), SPH_C32(0xc2bdbd1e), SPH_C32(0x1a8b8b9c),
+	SPH_C32(0x1e8a8a9b), SPH_C32(0xdb70704b), SPH_C32(0xf83e3eba),
+	SPH_C32(0xe2b5b526), SPH_C32(0x83666629), SPH_C32(0x3b4848e3),
+	SPH_C32(0x0c030309), SPH_C32(0xf5f6f6f4), SPH_C32(0x380e0e2a),
+	SPH_C32(0x9f61613c), SPH_C32(0xd435358b), SPH_C32(0x475757be),
+	SPH_C32(0xd2b9b902), SPH_C32(0x2e8686bf), SPH_C32(0x29c1c171),
+	SPH_C32(0x741d1d53), SPH_C32(0x4e9e9ef7), SPH_C32(0xa9e1e191),
+	SPH_C32(0xcdf8f8de), SPH_C32(0x569898e5), SPH_C32(0x44111177),
+	SPH_C32(0xbf696904), SPH_C32(0x49d9d939), SPH_C32(0x0e8e8e87),
+	SPH_C32(0x669494c1), SPH_C32(0x5a9b9bec), SPH_C32(0x781e1e5a),
+	SPH_C32(0x2a8787b8), SPH_C32(0x89e9e9a9), SPH_C32(0x15cece5c),
+	SPH_C32(0x4f5555b0), SPH_C32(0xa02828d8), SPH_C32(0x51dfdf2b),
+	SPH_C32(0x068c8c89), SPH_C32(0xb2a1a14a), SPH_C32(0x12898992),
+	SPH_C32(0x340d0d23), SPH_C32(0xcabfbf10), SPH_C32(0xb5e6e684),
+	SPH_C32(0x134242d5), SPH_C32(0xbb686803), SPH_C32(0x1f4141dc),
+	SPH_C32(0x529999e2), SPH_C32(0xb42d2dc3), SPH_C32(0x3c0f0f2d),
+	SPH_C32(0xf6b0b03d), SPH_C32(0x4b5454b7), SPH_C32(0xdabbbb0c),
+	SPH_C32(0x58161662)
+};
+
+static const sph_u32 mixtab2[] = {
+	SPH_C32(0x32976363), SPH_C32(0x6feb7c7c), SPH_C32(0x5ec77777),
+	SPH_C32(0x7af77b7b), SPH_C32(0xe8e5f2f2), SPH_C32(0x0ab76b6b),
+	SPH_C32(0x16a76f6f), SPH_C32(0x6d39c5c5), SPH_C32(0x90c03030),
+	SPH_C32(0x07040101), SPH_C32(0x2e876767), SPH_C32(0xd1ac2b2b),
+	SPH_C32(0xccd5fefe), SPH_C32(0x1371d7d7), SPH_C32(0x7c9aabab),
+	SPH_C32(0x59c37676), SPH_C32(0x4005caca), SPH_C32(0xa33e8282),
+	SPH_C32(0x4909c9c9), SPH_C32(0x68ef7d7d), SPH_C32(0xd0c5fafa),
+	SPH_C32(0x947f5959), SPH_C32(0xce074747), SPH_C32(0xe6edf0f0),
+	SPH_C32(0x6e82adad), SPH_C32(0x1a7dd4d4), SPH_C32(0x43bea2a2),
+	SPH_C32(0x608aafaf), SPH_C32(0xf9469c9c), SPH_C32(0x51a6a4a4),
+	SPH_C32(0x45d37272), SPH_C32(0x762dc0c0), SPH_C32(0x28eab7b7),
+	SPH_C32(0xc5d9fdfd), SPH_C32(0xd47a9393), SPH_C32(0xf2982626),
+	SPH_C32(0x82d83636), SPH_C32(0xbdfc3f3f), SPH_C32(0xf3f1f7f7),
+	SPH_C32(0x521dcccc), SPH_C32(0x8cd03434), SPH_C32(0x56a2a5a5),
+	SPH_C32(0x8db9e5e5), SPH_C32(0xe1e9f1f1), SPH_C32(0x4cdf7171),
+	SPH_C32(0x3e4dd8d8), SPH_C32(0x97c43131), SPH_C32(0x6b541515),
+	SPH_C32(0x1c100404), SPH_C32(0x6331c7c7), SPH_C32(0xe98c2323),
+	SPH_C32(0x7f21c3c3), SPH_C32(0x48601818), SPH_C32(0xcf6e9696),
+	SPH_C32(0x1b140505), SPH_C32(0xeb5e9a9a), SPH_C32(0x151c0707),
+	SPH_C32(0x7e481212), SPH_C32(0xad368080), SPH_C32(0x98a5e2e2),
+	SPH_C32(0xa781ebeb), SPH_C32(0xf59c2727), SPH_C32(0x33feb2b2),
+	SPH_C32(0x50cf7575), SPH_C32(0x3f240909), SPH_C32(0xa43a8383),
+	SPH_C32(0xc4b02c2c), SPH_C32(0x46681a1a), SPH_C32(0x416c1b1b),
+	SPH_C32(0x11a36e6e), SPH_C32(0x9d735a5a), SPH_C32(0x4db6a0a0),
+	SPH_C32(0xa5535252), SPH_C32(0xa1ec3b3b), SPH_C32(0x1475d6d6),
+	SPH_C32(0x34fab3b3), SPH_C32(0xdfa42929), SPH_C32(0x9fa1e3e3),
+	SPH_C32(0xcdbc2f2f), SPH_C32(0xb1268484), SPH_C32(0xa2575353),
+	SPH_C32(0x0169d1d1), SPH_C32(0x00000000), SPH_C32(0xb599eded),
+	SPH_C32(0xe0802020), SPH_C32(0xc2ddfcfc), SPH_C32(0x3af2b1b1),
+	SPH_C32(0x9a775b5b), SPH_C32(0x0db36a6a), SPH_C32(0x4701cbcb),
+	SPH_C32(0x17cebebe), SPH_C32(0xafe43939), SPH_C32(0xed334a4a),
+	SPH_C32(0xff2b4c4c), SPH_C32(0x937b5858), SPH_C32(0x5b11cfcf),
+	SPH_C32(0x066dd0d0), SPH_C32(0xbb91efef), SPH_C32(0x7b9eaaaa),
+	SPH_C32(0xd7c1fbfb), SPH_C32(0xd2174343), SPH_C32(0xf82f4d4d),
+	SPH_C32(0x99cc3333), SPH_C32(0xb6228585), SPH_C32(0xc00f4545),
+	SPH_C32(0xd9c9f9f9), SPH_C32(0x0e080202), SPH_C32(0x66e77f7f),
+	SPH_C32(0xab5b5050), SPH_C32(0xb4f03c3c), SPH_C32(0xf04a9f9f),
+	SPH_C32(0x7596a8a8), SPH_C32(0xac5f5151), SPH_C32(0x44baa3a3),
+	SPH_C32(0xdb1b4040), SPH_C32(0x800a8f8f), SPH_C32(0xd37e9292),
+	SPH_C32(0xfe429d9d), SPH_C32(0xa8e03838), SPH_C32(0xfdf9f5f5),
+	SPH_C32(0x19c6bcbc), SPH_C32(0x2feeb6b6), SPH_C32(0x3045dada),
+	SPH_C32(0xe7842121), SPH_C32(0x70401010), SPH_C32(0xcbd1ffff),
+	SPH_C32(0xefe1f3f3), SPH_C32(0x0865d2d2), SPH_C32(0x5519cdcd),
+	SPH_C32(0x24300c0c), SPH_C32(0x794c1313), SPH_C32(0xb29decec),
+	SPH_C32(0x86675f5f), SPH_C32(0xc86a9797), SPH_C32(0xc70b4444),
+	SPH_C32(0x655c1717), SPH_C32(0x6a3dc4c4), SPH_C32(0x58aaa7a7),
+	SPH_C32(0x61e37e7e), SPH_C32(0xb3f43d3d), SPH_C32(0x278b6464),
+	SPH_C32(0x886f5d5d), SPH_C32(0x4f641919), SPH_C32(0x42d77373),
+	SPH_C32(0x3b9b6060), SPH_C32(0xaa328181), SPH_C32(0xf6274f4f),
+	SPH_C32(0x225ddcdc), SPH_C32(0xee882222), SPH_C32(0xd6a82a2a),
+	SPH_C32(0xdd769090), SPH_C32(0x95168888), SPH_C32(0xc9034646),
+	SPH_C32(0xbc95eeee), SPH_C32(0x05d6b8b8), SPH_C32(0x6c501414),
+	SPH_C32(0x2c55dede), SPH_C32(0x81635e5e), SPH_C32(0x312c0b0b),
+	SPH_C32(0x3741dbdb), SPH_C32(0x96ade0e0), SPH_C32(0x9ec83232),
+	SPH_C32(0xa6e83a3a), SPH_C32(0x36280a0a), SPH_C32(0xe43f4949),
+	SPH_C32(0x12180606), SPH_C32(0xfc902424), SPH_C32(0x8f6b5c5c),
+	SPH_C32(0x7825c2c2), SPH_C32(0x0f61d3d3), SPH_C32(0x6986acac),
+	SPH_C32(0x35936262), SPH_C32(0xda729191), SPH_C32(0xc6629595),
+	SPH_C32(0x8abde4e4), SPH_C32(0x74ff7979), SPH_C32(0x83b1e7e7),
+	SPH_C32(0x4e0dc8c8), SPH_C32(0x85dc3737), SPH_C32(0x18af6d6d),
+	SPH_C32(0x8e028d8d), SPH_C32(0x1d79d5d5), SPH_C32(0xf1234e4e),
+	SPH_C32(0x7292a9a9), SPH_C32(0x1fab6c6c), SPH_C32(0xb9435656),
+	SPH_C32(0xfafdf4f4), SPH_C32(0xa085eaea), SPH_C32(0x208f6565),
+	SPH_C32(0x7df37a7a), SPH_C32(0x678eaeae), SPH_C32(0x38200808),
+	SPH_C32(0x0bdebaba), SPH_C32(0x73fb7878), SPH_C32(0xfb942525),
+	SPH_C32(0xcab82e2e), SPH_C32(0x54701c1c), SPH_C32(0x5faea6a6),
+	SPH_C32(0x21e6b4b4), SPH_C32(0x6435c6c6), SPH_C32(0xae8de8e8),
+	SPH_C32(0x2559dddd), SPH_C32(0x57cb7474), SPH_C32(0x5d7c1f1f),
+	SPH_C32(0xea374b4b), SPH_C32(0x1ec2bdbd), SPH_C32(0x9c1a8b8b),
+	SPH_C32(0x9b1e8a8a), SPH_C32(0x4bdb7070), SPH_C32(0xbaf83e3e),
+	SPH_C32(0x26e2b5b5), SPH_C32(0x29836666), SPH_C32(0xe33b4848),
+	SPH_C32(0x090c0303), SPH_C32(0xf4f5f6f6), SPH_C32(0x2a380e0e),
+	SPH_C32(0x3c9f6161), SPH_C32(0x8bd43535), SPH_C32(0xbe475757),
+	SPH_C32(0x02d2b9b9), SPH_C32(0xbf2e8686), SPH_C32(0x7129c1c1),
+	SPH_C32(0x53741d1d), SPH_C32(0xf74e9e9e), SPH_C32(0x91a9e1e1),
+	SPH_C32(0xdecdf8f8), SPH_C32(0xe5569898), SPH_C32(0x77441111),
+	SPH_C32(0x04bf6969), SPH_C32(0x3949d9d9), SPH_C32(0x870e8e8e),
+	SPH_C32(0xc1669494), SPH_C32(0xec5a9b9b), SPH_C32(0x5a781e1e),
+	SPH_C32(0xb82a8787), SPH_C32(0xa989e9e9), SPH_C32(0x5c15cece),
+	SPH_C32(0xb04f5555), SPH_C32(0xd8a02828), SPH_C32(0x2b51dfdf),
+	SPH_C32(0x89068c8c), SPH_C32(0x4ab2a1a1), SPH_C32(0x92128989),
+	SPH_C32(0x23340d0d), SPH_C32(0x10cabfbf), SPH_C32(0x84b5e6e6),
+	SPH_C32(0xd5134242), SPH_C32(0x03bb6868), SPH_C32(0xdc1f4141),
+	SPH_C32(0xe2529999), SPH_C32(0xc3b42d2d), SPH_C32(0x2d3c0f0f),
+	SPH_C32(0x3df6b0b0), SPH_C32(0xb74b5454), SPH_C32(0x0cdabbbb),
+	SPH_C32(0x62581616)
+};
+
+static const sph_u32 mixtab3[] = {
+	SPH_C32(0x63329763), SPH_C32(0x7c6feb7c), SPH_C32(0x775ec777),
+	SPH_C32(0x7b7af77b), SPH_C32(0xf2e8e5f2), SPH_C32(0x6b0ab76b),
+	SPH_C32(0x6f16a76f), SPH_C32(0xc56d39c5), SPH_C32(0x3090c030),
+	SPH_C32(0x01070401), SPH_C32(0x672e8767), SPH_C32(0x2bd1ac2b),
+	SPH_C32(0xfeccd5fe), SPH_C32(0xd71371d7), SPH_C32(0xab7c9aab),
+	SPH_C32(0x7659c376), SPH_C32(0xca4005ca), SPH_C32(0x82a33e82),
+	SPH_C32(0xc94909c9), SPH_C32(0x7d68ef7d), SPH_C32(0xfad0c5fa),
+	SPH_C32(0x59947f59), SPH_C32(0x47ce0747), SPH_C32(0xf0e6edf0),
+	SPH_C32(0xad6e82ad), SPH_C32(0xd41a7dd4), SPH_C32(0xa243bea2),
+	SPH_C32(0xaf608aaf), SPH_C32(0x9cf9469c), SPH_C32(0xa451a6a4),
+	SPH_C32(0x7245d372), SPH_C32(0xc0762dc0), SPH_C32(0xb728eab7),
+	SPH_C32(0xfdc5d9fd), SPH_C32(0x93d47a93), SPH_C32(0x26f29826),
+	SPH_C32(0x3682d836), SPH_C32(0x3fbdfc3f), SPH_C32(0xf7f3f1f7),
+	SPH_C32(0xcc521dcc), SPH_C32(0x348cd034), SPH_C32(0xa556a2a5),
+	SPH_C32(0xe58db9e5), SPH_C32(0xf1e1e9f1), SPH_C32(0x714cdf71),
+	SPH_C32(0xd83e4dd8), SPH_C32(0x3197c431), SPH_C32(0x156b5415),
+	SPH_C32(0x041c1004), SPH_C32(0xc76331c7), SPH_C32(0x23e98c23),
+	SPH_C32(0xc37f21c3), SPH_C32(0x18486018), SPH_C32(0x96cf6e96),
+	SPH_C32(0x051b1405), SPH_C32(0x9aeb5e9a), SPH_C32(0x07151c07),
+	SPH_C32(0x127e4812), SPH_C32(0x80ad3680), SPH_C32(0xe298a5e2),
+	SPH_C32(0xeba781eb), SPH_C32(0x27f59c27), SPH_C32(0xb233feb2),
+	SPH_C32(0x7550cf75), SPH_C32(0x093f2409), SPH_C32(0x83a43a83),
+	SPH_C32(0x2cc4b02c), SPH_C32(0x1a46681a), SPH_C32(0x1b416c1b),
+	SPH_C32(0x6e11a36e), SPH_C32(0x5a9d735a), SPH_C32(0xa04db6a0),
+	SPH_C32(0x52a55352), SPH_C32(0x3ba1ec3b), SPH_C32(0xd61475d6),
+	SPH_C32(0xb334fab3), SPH_C32(0x29dfa429), SPH_C32(0xe39fa1e3),
+	SPH_C32(0x2fcdbc2f), SPH_C32(0x84b12684), SPH_C32(0x53a25753),
+	SPH_C32(0xd10169d1), SPH_C32(0x00000000), SPH_C32(0xedb599ed),
+	SPH_C32(0x20e08020), SPH_C32(0xfcc2ddfc), SPH_C32(0xb13af2b1),
+	SPH_C32(0x5b9a775b), SPH_C32(0x6a0db36a), SPH_C32(0xcb4701cb),
+	SPH_C32(0xbe17cebe), SPH_C32(0x39afe439), SPH_C32(0x4aed334a),
+	SPH_C32(0x4cff2b4c), SPH_C32(0x58937b58), SPH_C32(0xcf5b11cf),
+	SPH_C32(0xd0066dd0), SPH_C32(0xefbb91ef), SPH_C32(0xaa7b9eaa),
+	SPH_C32(0xfbd7c1fb), SPH_C32(0x43d21743), SPH_C32(0x4df82f4d),
+	SPH_C32(0x3399cc33), SPH_C32(0x85b62285), SPH_C32(0x45c00f45),
+	SPH_C32(0xf9d9c9f9), SPH_C32(0x020e0802), SPH_C32(0x7f66e77f),
+	SPH_C32(0x50ab5b50), SPH_C32(0x3cb4f03c), SPH_C32(0x9ff04a9f),
+	SPH_C32(0xa87596a8), SPH_C32(0x51ac5f51), SPH_C32(0xa344baa3),
+	SPH_C32(0x40db1b40), SPH_C32(0x8f800a8f), SPH_C32(0x92d37e92),
+	SPH_C32(0x9dfe429d), SPH_C32(0x38a8e038), SPH_C32(0xf5fdf9f5),
+	SPH_C32(0xbc19c6bc), SPH_C32(0xb62feeb6), SPH_C32(0xda3045da),
+	SPH_C32(0x21e78421), SPH_C32(0x10704010), SPH_C32(0xffcbd1ff),
+	SPH_C32(0xf3efe1f3), SPH_C32(0xd20865d2), SPH_C32(0xcd5519cd),
+	SPH_C32(0x0c24300c), SPH_C32(0x13794c13), SPH_C32(0xecb29dec),
+	SPH_C32(0x5f86675f), SPH_C32(0x97c86a97), SPH_C32(0x44c70b44),
+	SPH_C32(0x17655c17), SPH_C32(0xc46a3dc4), SPH_C32(0xa758aaa7),
+	SPH_C32(0x7e61e37e), SPH_C32(0x3db3f43d), SPH_C32(0x64278b64),
+	SPH_C32(0x5d886f5d), SPH_C32(0x194f6419), SPH_C32(0x7342d773),
+	SPH_C32(0x603b9b60), SPH_C32(0x81aa3281), SPH_C32(0x4ff6274f),
+	SPH_C32(0xdc225ddc), SPH_C32(0x22ee8822), SPH_C32(0x2ad6a82a),
+	SPH_C32(0x90dd7690), SPH_C32(0x88951688), SPH_C32(0x46c90346),
+	SPH_C32(0xeebc95ee), SPH_C32(0xb805d6b8), SPH_C32(0x146c5014),
+	SPH_C32(0xde2c55de), SPH_C32(0x5e81635e), SPH_C32(0x0b312c0b),
+	SPH_C32(0xdb3741db), SPH_C32(0xe096ade0), SPH_C32(0x329ec832),
+	SPH_C32(0x3aa6e83a), SPH_C32(0x0a36280a), SPH_C32(0x49e43f49),
+	SPH_C32(0x06121806), SPH_C32(0x24fc9024), SPH_C32(0x5c8f6b5c),
+	SPH_C32(0xc27825c2), SPH_C32(0xd30f61d3), SPH_C32(0xac6986ac),
+	SPH_C32(0x62359362), SPH_C32(0x91da7291), SPH_C32(0x95c66295),
+	SPH_C32(0xe48abde4), SPH_C32(0x7974ff79), SPH_C32(0xe783b1e7),
+	SPH_C32(0xc84e0dc8), SPH_C32(0x3785dc37), SPH_C32(0x6d18af6d),
+	SPH_C32(0x8d8e028d), SPH_C32(0xd51d79d5), SPH_C32(0x4ef1234e),
+	SPH_C32(0xa97292a9), SPH_C32(0x6c1fab6c), SPH_C32(0x56b94356),
+	SPH_C32(0xf4fafdf4), SPH_C32(0xeaa085ea), SPH_C32(0x65208f65),
+	SPH_C32(0x7a7df37a), SPH_C32(0xae678eae), SPH_C32(0x08382008),
+	SPH_C32(0xba0bdeba), SPH_C32(0x7873fb78), SPH_C32(0x25fb9425),
+	SPH_C32(0x2ecab82e), SPH_C32(0x1c54701c), SPH_C32(0xa65faea6),
+	SPH_C32(0xb421e6b4), SPH_C32(0xc66435c6), SPH_C32(0xe8ae8de8),
+	SPH_C32(0xdd2559dd), SPH_C32(0x7457cb74), SPH_C32(0x1f5d7c1f),
+	SPH_C32(0x4bea374b), SPH_C32(0xbd1ec2bd), SPH_C32(0x8b9c1a8b),
+	SPH_C32(0x8a9b1e8a), SPH_C32(0x704bdb70), SPH_C32(0x3ebaf83e),
+	SPH_C32(0xb526e2b5), SPH_C32(0x66298366), SPH_C32(0x48e33b48),
+	SPH_C32(0x03090c03), SPH_C32(0xf6f4f5f6), SPH_C32(0x0e2a380e),
+	SPH_C32(0x613c9f61), SPH_C32(0x358bd435), SPH_C32(0x57be4757),
+	SPH_C32(0xb902d2b9), SPH_C32(0x86bf2e86), SPH_C32(0xc17129c1),
+	SPH_C32(0x1d53741d), SPH_C32(0x9ef74e9e), SPH_C32(0xe191a9e1),
+	SPH_C32(0xf8decdf8), SPH_C32(0x98e55698), SPH_C32(0x11774411),
+	SPH_C32(0x6904bf69), SPH_C32(0xd93949d9), SPH_C32(0x8e870e8e),
+	SPH_C32(0x94c16694), SPH_C32(0x9bec5a9b), SPH_C32(0x1e5a781e),
+	SPH_C32(0x87b82a87), SPH_C32(0xe9a989e9), SPH_C32(0xce5c15ce),
+	SPH_C32(0x55b04f55), SPH_C32(0x28d8a028), SPH_C32(0xdf2b51df),
+	SPH_C32(0x8c89068c), SPH_C32(0xa14ab2a1), SPH_C32(0x89921289),
+	SPH_C32(0x0d23340d), SPH_C32(0xbf10cabf), SPH_C32(0xe684b5e6),
+	SPH_C32(0x42d51342), SPH_C32(0x6803bb68), SPH_C32(0x41dc1f41),
+	SPH_C32(0x99e25299), SPH_C32(0x2dc3b42d), SPH_C32(0x0f2d3c0f),
+	SPH_C32(0xb03df6b0), SPH_C32(0x54b74b54), SPH_C32(0xbb0cdabb),
+	SPH_C32(0x16625816)
+};
+
+#define TIX2(q, x00, x01, x08, x10, x24)   do { \
+		x10 ^= x00; \
+		x00 = (q); \
+		x08 ^= x00; \
+		x01 ^= x24; \
+	} while (0)
+
+#define TIX3(q, x00, x01, x04, x08, x16, x27, x30)   do { \
+		x16 ^= x00; \
+		x00 = (q); \
+		x08 ^= x00; \
+		x01 ^= x27; \
+		x04 ^= x30; \
+	} while (0)
+
+#define TIX4(q, x00, x01, x04, x07, x08, x22, x24, x27, x30)   do { \
+		x22 ^= x00; \
+		x00 = (q); \
+		x08 ^= x00; \
+		x01 ^= x24; \
+		x04 ^= x27; \
+		x07 ^= x30; \
+	} while (0)
+
+#define CMIX30(x00, x01, x02, x04, x05, x06, x15, x16, x17)   do { \
+		x00 ^= x04; \
+		x01 ^= x05; \
+		x02 ^= x06; \
+		x15 ^= x04; \
+		x16 ^= x05; \
+		x17 ^= x06; \
+	} while (0)
+
+#define CMIX36(x00, x01, x02, x04, x05, x06, x18, x19, x20)   do { \
+		x00 ^= x04; \
+		x01 ^= x05; \
+		x02 ^= x06; \
+		x18 ^= x04; \
+		x19 ^= x05; \
+		x20 ^= x06; \
+	} while (0)
+
+#define SMIX(x0, x1, x2, x3)   do { \
+		sph_u32 c0 = 0; \
+		sph_u32 c1 = 0; \
+		sph_u32 c2 = 0; \
+		sph_u32 c3 = 0; \
+		sph_u32 r0 = 0; \
+		sph_u32 r1 = 0; \
+		sph_u32 r2 = 0; \
+		sph_u32 r3 = 0; \
+		sph_u32 tmp; \
+		tmp = mixtab0[x0 >> 24]; \
+		c0 ^= tmp; \
+		tmp = mixtab1[(x0 >> 16) & 0xFF]; \
+		c0 ^= tmp; \
+		r1 ^= tmp; \
+		tmp = mixtab2[(x0 >>  8) & 0xFF]; \
+		c0 ^= tmp; \
+		r2 ^= tmp; \
+		tmp = mixtab3[x0 & 0xFF]; \
+		c0 ^= tmp; \
+		r3 ^= tmp; \
+		tmp = mixtab0[x1 >> 24]; \
+		c1 ^= tmp; \
+		r0 ^= tmp; \
+		tmp = mixtab1[(x1 >> 16) & 0xFF]; \
+		c1 ^= tmp; \
+		tmp = mixtab2[(x1 >>  8) & 0xFF]; \
+		c1 ^= tmp; \
+		r2 ^= tmp; \
+		tmp = mixtab3[x1 & 0xFF]; \
+		c1 ^= tmp; \
+		r3 ^= tmp; \
+		tmp = mixtab0[x2 >> 24]; \
+		c2 ^= tmp; \
+		r0 ^= tmp; \
+		tmp = mixtab1[(x2 >> 16) & 0xFF]; \
+		c2 ^= tmp; \
+		r1 ^= tmp; \
+		tmp = mixtab2[(x2 >>  8) & 0xFF]; \
+		c2 ^= tmp; \
+		tmp = mixtab3[x2 & 0xFF]; \
+		c2 ^= tmp; \
+		r3 ^= tmp; \
+		tmp = mixtab0[x3 >> 24]; \
+		c3 ^= tmp; \
+		r0 ^= tmp; \
+		tmp = mixtab1[(x3 >> 16) & 0xFF]; \
+		c3 ^= tmp; \
+		r1 ^= tmp; \
+		tmp = mixtab2[(x3 >>  8) & 0xFF]; \
+		c3 ^= tmp; \
+		r2 ^= tmp; \
+		tmp = mixtab3[x3 & 0xFF]; \
+		c3 ^= tmp; \
+		x0 = ((c0 ^ r0) & SPH_C32(0xFF000000)) \
+			| ((c1 ^ r1) & SPH_C32(0x00FF0000)) \
+			| ((c2 ^ r2) & SPH_C32(0x0000FF00)) \
+			| ((c3 ^ r3) & SPH_C32(0x000000FF)); \
+		x1 = ((c1 ^ (r0 << 8)) & SPH_C32(0xFF000000)) \
+			| ((c2 ^ (r1 << 8)) & SPH_C32(0x00FF0000)) \
+			| ((c3 ^ (r2 << 8)) & SPH_C32(0x0000FF00)) \
+			| ((c0 ^ (r3 >> 24)) & SPH_C32(0x000000FF)); \
+		x2 = ((c2 ^ (r0 << 16)) & SPH_C32(0xFF000000)) \
+			| ((c3 ^ (r1 << 16)) & SPH_C32(0x00FF0000)) \
+			| ((c0 ^ (r2 >> 16)) & SPH_C32(0x0000FF00)) \
+			| ((c1 ^ (r3 >> 16)) & SPH_C32(0x000000FF)); \
+		x3 = ((c3 ^ (r0 << 24)) & SPH_C32(0xFF000000)) \
+			| ((c0 ^ (r1 >> 8)) & SPH_C32(0x00FF0000)) \
+			| ((c1 ^ (r2 >> 8)) & SPH_C32(0x0000FF00)) \
+			| ((c2 ^ (r3 >> 8)) & SPH_C32(0x000000FF)); \
+		/* */ \
+	} while (0)
+
+#if SPH_FUGUE_NOCOPY
+
+#define DECL_STATE_SMALL
+#define READ_STATE_SMALL(state)
+#define WRITE_STATE_SMALL(state)
+#define DECL_STATE_BIG
+#define READ_STATE_BIG(state)
+#define WRITE_STATE_BIG(state)
+
+#define S00   ((sc)->S[ 0])
+#define S01   ((sc)->S[ 1])
+#define S02   ((sc)->S[ 2])
+#define S03   ((sc)->S[ 3])
+#define S04   ((sc)->S[ 4])
+#define S05   ((sc)->S[ 5])
+#define S06   ((sc)->S[ 6])
+#define S07   ((sc)->S[ 7])
+#define S08   ((sc)->S[ 8])
+#define S09   ((sc)->S[ 9])
+#define S10   ((sc)->S[10])
+#define S11   ((sc)->S[11])
+#define S12   ((sc)->S[12])
+#define S13   ((sc)->S[13])
+#define S14   ((sc)->S[14])
+#define S15   ((sc)->S[15])
+#define S16   ((sc)->S[16])
+#define S17   ((sc)->S[17])
+#define S18   ((sc)->S[18])
+#define S19   ((sc)->S[19])
+#define S20   ((sc)->S[20])
+#define S21   ((sc)->S[21])
+#define S22   ((sc)->S[22])
+#define S23   ((sc)->S[23])
+#define S24   ((sc)->S[24])
+#define S25   ((sc)->S[25])
+#define S26   ((sc)->S[26])
+#define S27   ((sc)->S[27])
+#define S28   ((sc)->S[28])
+#define S29   ((sc)->S[29])
+#define S30   ((sc)->S[30])
+#define S31   ((sc)->S[31])
+#define S32   ((sc)->S[32])
+#define S33   ((sc)->S[33])
+#define S34   ((sc)->S[34])
+#define S35   ((sc)->S[35])
+
+#else
+
+#define DECL_STATE_SMALL \
+	sph_u32 S00, S01, S02, S03, S04, S05, S06, S07, S08, S09; \
+	sph_u32 S10, S11, S12, S13, S14, S15, S16, S17, S18, S19; \
+	sph_u32 S20, S21, S22, S23, S24, S25, S26, S27, S28, S29;
+
+#define DECL_STATE_BIG \
+	DECL_STATE_SMALL \
+	sph_u32 S30, S31, S32, S33, S34, S35;
+
+#define READ_STATE_SMALL(state)   do { \
+		S00 = (state)->S[ 0]; \
+		S01 = (state)->S[ 1]; \
+		S02 = (state)->S[ 2]; \
+		S03 = (state)->S[ 3]; \
+		S04 = (state)->S[ 4]; \
+		S05 = (state)->S[ 5]; \
+		S06 = (state)->S[ 6]; \
+		S07 = (state)->S[ 7]; \
+		S08 = (state)->S[ 8]; \
+		S09 = (state)->S[ 9]; \
+		S10 = (state)->S[10]; \
+		S11 = (state)->S[11]; \
+		S12 = (state)->S[12]; \
+		S13 = (state)->S[13]; \
+		S14 = (state)->S[14]; \
+		S15 = (state)->S[15]; \
+		S16 = (state)->S[16]; \
+		S17 = (state)->S[17]; \
+		S18 = (state)->S[18]; \
+		S19 = (state)->S[19]; \
+		S20 = (state)->S[20]; \
+		S21 = (state)->S[21]; \
+		S22 = (state)->S[22]; \
+		S23 = (state)->S[23]; \
+		S24 = (state)->S[24]; \
+		S25 = (state)->S[25]; \
+		S26 = (state)->S[26]; \
+		S27 = (state)->S[27]; \
+		S28 = (state)->S[28]; \
+		S29 = (state)->S[29]; \
+	} while (0)
+
+#define READ_STATE_BIG(state)   do { \
+		READ_STATE_SMALL(state); \
+		S30 = (state)->S[30]; \
+		S31 = (state)->S[31]; \
+		S32 = (state)->S[32]; \
+		S33 = (state)->S[33]; \
+		S34 = (state)->S[34]; \
+		S35 = (state)->S[35]; \
+	} while (0)
+
+#define WRITE_STATE_SMALL(state)   do { \
+		(state)->S[ 0] = S00; \
+		(state)->S[ 1] = S01; \
+		(state)->S[ 2] = S02; \
+		(state)->S[ 3] = S03; \
+		(state)->S[ 4] = S04; \
+		(state)->S[ 5] = S05; \
+		(state)->S[ 6] = S06; \
+		(state)->S[ 7] = S07; \
+		(state)->S[ 8] = S08; \
+		(state)->S[ 9] = S09; \
+		(state)->S[10] = S10; \
+		(state)->S[11] = S11; \
+		(state)->S[12] = S12; \
+		(state)->S[13] = S13; \
+		(state)->S[14] = S14; \
+		(state)->S[15] = S15; \
+		(state)->S[16] = S16; \
+		(state)->S[17] = S17; \
+		(state)->S[18] = S18; \
+		(state)->S[19] = S19; \
+		(state)->S[20] = S20; \
+		(state)->S[21] = S21; \
+		(state)->S[22] = S22; \
+		(state)->S[23] = S23; \
+		(state)->S[24] = S24; \
+		(state)->S[25] = S25; \
+		(state)->S[26] = S26; \
+		(state)->S[27] = S27; \
+		(state)->S[28] = S28; \
+		(state)->S[29] = S29; \
+	} while (0)
+
+#define WRITE_STATE_BIG(state)   do { \
+		WRITE_STATE_SMALL(state); \
+		(state)->S[30] = S30; \
+		(state)->S[31] = S31; \
+		(state)->S[32] = S32; \
+		(state)->S[33] = S33; \
+		(state)->S[34] = S34; \
+		(state)->S[35] = S35; \
+	} while (0)
+
+#endif
+
+static void
+fugue_init(sph_fugue_context *sc, size_t z_len,
+	const sph_u32 *iv, size_t iv_len)
+{
+	size_t u;
+
+	for (u = 0; u < z_len; u ++)
+		sc->S[u] = 0;
+	memcpy(&sc->S[z_len], iv, iv_len * sizeof *iv);
+	sc->partial = 0;
+	sc->partial_len = 0;
+	sc->round_shift = 0;
+#if SPH_64
+	sc->bit_count = 0;
+#else
+	sc->bit_count_high = 0;
+	sc->bit_count_low = 0;
+#endif
+}
+
+#if SPH_64
+
+#define INCR_COUNTER   do { \
+		sc->bit_count += (sph_u64)len << 3; \
+	} while (0)
+
+#else
+
+#define INCR_COUNTER   do { \
+		sph_u32 tmp = SPH_T32((sph_u32)len << 3); \
+		sc->bit_count_low = SPH_T32(sc->bit_count_low + tmp); \
+		if (sc->bit_count_low < tmp) \
+			sc->bit_count_high ++; \
+		sc->bit_count_high = SPH_T32(sc->bit_count_high \
+			+ ((sph_u32)len >> 29)); \
+	} while (0)
+
+#endif
+
+#define CORE_ENTRY \
+	sph_u32 p; \
+	unsigned plen, rshift; \
+	INCR_COUNTER; \
+	p = sc->partial; \
+	plen = sc->partial_len; \
+	if (plen < 4) { \
+		unsigned count = 4 - plen; \
+		if (len < count) \
+			count = len; \
+		plen += count; \
+		while (count -- > 0) { \
+			p = (p << 8) | *(const unsigned char *)data; \
+			data = (const unsigned char *)data + 1; \
+			len --; \
+		} \
+		if (len == 0) { \
+			sc->partial = p; \
+			sc->partial_len = plen; \
+			return; \
+		} \
+	}
+
+#define CORE_EXIT \
+	p = 0; \
+	sc->partial_len = (unsigned)len; \
+	while (len -- > 0) { \
+		p = (p << 8) | *(const unsigned char *)data; \
+		data = (const unsigned char *)data + 1; \
+	} \
+	sc->partial = p; \
+	sc->round_shift = rshift;
+
+/*
+ * Not in a do..while: the 'break' must exit the outer loop.
+ */
+#define NEXT(rc) \
+	if (len <= 4) { \
+		rshift = (rc); \
+		break; \
+	} \
+	p = sph_dec32be(data); \
+	data = (const unsigned char *)data + 4; \
+	len -= 4
+
+static void
+fugue2_core(sph_fugue_context *sc, const void *data, size_t len)
+{
+	DECL_STATE_SMALL
+	CORE_ENTRY
+	READ_STATE_SMALL(sc);
+	rshift = sc->round_shift;
+	switch (rshift) {
+		for (;;) {
+			sph_u32 q;
+
+		case 0:
+			q = p;
+			TIX2(q, S00, S01, S08, S10, S24);
+			CMIX30(S27, S28, S29, S01, S02, S03, S12, S13, S14);
+			SMIX(S27, S28, S29, S00);
+			CMIX30(S24, S25, S26, S28, S29, S00, S09, S10, S11);
+			SMIX(S24, S25, S26, S27);
+			NEXT(1);
+			/* fall through */
+		case 1:
+			q = p;
+			TIX2(q, S24, S25, S02, S04, S18);
+			CMIX30(S21, S22, S23, S25, S26, S27, S06, S07, S08);
+			SMIX(S21, S22, S23, S24);
+			CMIX30(S18, S19, S20, S22, S23, S24, S03, S04, S05);
+			SMIX(S18, S19, S20, S21);
+			NEXT(2);
+			/* fall through */
+		case 2:
+			q = p;
+			TIX2(q, S18, S19, S26, S28, S12);
+			CMIX30(S15, S16, S17, S19, S20, S21, S00, S01, S02);
+			SMIX(S15, S16, S17, S18);
+			CMIX30(S12, S13, S14, S16, S17, S18, S27, S28, S29);
+			SMIX(S12, S13, S14, S15);
+			NEXT(3);
+			/* fall through */
+		case 3:
+			q = p;
+			TIX2(q, S12, S13, S20, S22, S06);
+			CMIX30(S09, S10, S11, S13, S14, S15, S24, S25, S26);
+			SMIX(S09, S10, S11, S12);
+			CMIX30(S06, S07, S08, S10, S11, S12, S21, S22, S23);
+			SMIX(S06, S07, S08, S09);
+			NEXT(4);
+			/* fall through */
+		case 4:
+			q = p;
+			TIX2(q, S06, S07, S14, S16, S00);
+			CMIX30(S03, S04, S05, S07, S08, S09, S18, S19, S20);
+			SMIX(S03, S04, S05, S06);
+			CMIX30(S00, S01, S02, S04, S05, S06, S15, S16, S17);
+			SMIX(S00, S01, S02, S03);
+			NEXT(0);
+		}
+	}
+	CORE_EXIT
+	WRITE_STATE_SMALL(sc);
+}
+
+static void
+fugue3_core(sph_fugue_context *sc, const void *data, size_t len)
+{
+	DECL_STATE_BIG
+	CORE_ENTRY
+	READ_STATE_BIG(sc);
+	rshift = sc->round_shift;
+	switch (rshift) {
+		for (;;) {
+			sph_u32 q;
+
+		case 0:
+			q = p;
+			TIX3(q, S00, S01, S04, S08, S16, S27, S30);
+			CMIX36(S33, S34, S35, S01, S02, S03, S15, S16, S17);
+			SMIX(S33, S34, S35, S00);
+			CMIX36(S30, S31, S32, S34, S35, S00, S12, S13, S14);
+			SMIX(S30, S31, S32, S33);
+			CMIX36(S27, S28, S29, S31, S32, S33, S09, S10, S11);
+			SMIX(S27, S28, S29, S30);
+			NEXT(1);
+			/* fall through */
+		case 1:
+			q = p;
+			TIX3(q, S27, S28, S31, S35, S07, S18, S21);
+			CMIX36(S24, S25, S26, S28, S29, S30, S06, S07, S08);
+			SMIX(S24, S25, S26, S27);
+			CMIX36(S21, S22, S23, S25, S26, S27, S03, S04, S05);
+			SMIX(S21, S22, S23, S24);
+			CMIX36(S18, S19, S20, S22, S23, S24, S00, S01, S02);
+			SMIX(S18, S19, S20, S21);
+			NEXT(2);
+			/* fall through */
+		case 2:
+			q = p;
+			TIX3(q, S18, S19, S22, S26, S34, S09, S12);
+			CMIX36(S15, S16, S17, S19, S20, S21, S33, S34, S35);
+			SMIX(S15, S16, S17, S18);
+			CMIX36(S12, S13, S14, S16, S17, S18, S30, S31, S32);
+			SMIX(S12, S13, S14, S15);
+			CMIX36(S09, S10, S11, S13, S14, S15, S27, S28, S29);
+			SMIX(S09, S10, S11, S12);
+			NEXT(3);
+			/* fall through */
+		case 3:
+			q = p;
+			TIX3(q, S09, S10, S13, S17, S25, S00, S03);
+			CMIX36(S06, S07, S08, S10, S11, S12, S24, S25, S26);
+			SMIX(S06, S07, S08, S09);
+			CMIX36(S03, S04, S05, S07, S08, S09, S21, S22, S23);
+			SMIX(S03, S04, S05, S06);
+			CMIX36(S00, S01, S02, S04, S05, S06, S18, S19, S20);
+			SMIX(S00, S01, S02, S03);
+			NEXT(0);
+		}
+	}
+	CORE_EXIT
+	WRITE_STATE_BIG(sc);
+}
+
+static void
+fugue4_core(sph_fugue_context *sc, const void *data, size_t len)
+{
+	DECL_STATE_BIG
+	CORE_ENTRY
+	READ_STATE_BIG(sc);
+	rshift = sc->round_shift;
+	switch (rshift) {
+		for (;;) {
+			sph_u32 q;
+
+		case 0:
+			q = p;
+			TIX4(q, S00, S01, S04, S07, S08, S22, S24, S27, S30);
+			CMIX36(S33, S34, S35, S01, S02, S03, S15, S16, S17);
+			SMIX(S33, S34, S35, S00);
+			CMIX36(S30, S31, S32, S34, S35, S00, S12, S13, S14);
+			SMIX(S30, S31, S32, S33);
+			CMIX36(S27, S28, S29, S31, S32, S33, S09, S10, S11);
+			SMIX(S27, S28, S29, S30);
+			CMIX36(S24, S25, S26, S28, S29, S30, S06, S07, S08);
+			SMIX(S24, S25, S26, S27);
+			NEXT(1);
+			/* fall through */
+		case 1:
+			q = p;
+			TIX4(q, S24, S25, S28, S31, S32, S10, S12, S15, S18);
+			CMIX36(S21, S22, S23, S25, S26, S27, S03, S04, S05);
+			SMIX(S21, S22, S23, S24);
+			CMIX36(S18, S19, S20, S22, S23, S24, S00, S01, S02);
+			SMIX(S18, S19, S20, S21);
+			CMIX36(S15, S16, S17, S19, S20, S21, S33, S34, S35);
+			SMIX(S15, S16, S17, S18);
+			CMIX36(S12, S13, S14, S16, S17, S18, S30, S31, S32);
+			SMIX(S12, S13, S14, S15);
+			NEXT(2);
+			/* fall through */
+		case 2:
+			q = p;
+			TIX4(q, S12, S13, S16, S19, S20, S34, S00, S03, S06);
+			CMIX36(S09, S10, S11, S13, S14, S15, S27, S28, S29);
+			SMIX(S09, S10, S11, S12);
+			CMIX36(S06, S07, S08, S10, S11, S12, S24, S25, S26);
+			SMIX(S06, S07, S08, S09);
+			CMIX36(S03, S04, S05, S07, S08, S09, S21, S22, S23);
+			SMIX(S03, S04, S05, S06);
+			CMIX36(S00, S01, S02, S04, S05, S06, S18, S19, S20);
+			SMIX(S00, S01, S02, S03);
+			NEXT(0);
+		}
+	}
+	CORE_EXIT
+	WRITE_STATE_BIG(sc);
+}
+
+#if SPH_64
+
+#define WRITE_COUNTER   do { \
+		sph_enc64be(buf + 4, sc->bit_count + n); \
+	} while (0)
+
+#else
+
+#define WRITE_COUNTER   do { \
+		sph_enc32be(buf + 4, sc->bit_count_high); \
+		sph_enc32be(buf + 8, sc->bit_count_low + n); \
+	} while (0)
+
+#endif
+
+#define CLOSE_ENTRY(s, rcm, core) \
+	unsigned char buf[16]; \
+	unsigned plen, rms; \
+	unsigned char *out; \
+	sph_u32 S[s]; \
+	plen = sc->partial_len; \
+	WRITE_COUNTER; \
+	if (plen == 0 && n == 0) { \
+		plen = 4; \
+	} else if (plen < 4 || n != 0) { \
+		unsigned u; \
+ \
+		if (plen == 4) \
+			plen = 0; \
+		buf[plen] = ub & ~(0xFFU >> n); \
+		for (u = plen + 1; u < 4; u ++) \
+			buf[u] = 0; \
+	} \
+	core(sc, buf + plen, (sizeof buf) - plen); \
+	rms = sc->round_shift * (rcm); \
+	memcpy(S, sc->S + (s) - rms, rms * sizeof(sph_u32)); \
+	memcpy(S + rms, sc->S, ((s) - rms) * sizeof(sph_u32));
+
+#define ROR(n, s)   do { \
+		sph_u32 tmp[n]; \
+		memcpy(tmp, S + ((s) - (n)), (n) * sizeof(sph_u32)); \
+		memmove(S + (n), S, ((s) - (n)) * sizeof(sph_u32)); \
+		memcpy(S, tmp, (n) * sizeof(sph_u32)); \
+	} while (0)
+
+static void
+fugue2_close(sph_fugue_context *sc, unsigned ub, unsigned n,
+	void *dst, size_t out_size_w32)
+{
+	int i;
+
+	CLOSE_ENTRY(30, 6, fugue2_core)
+	for (i = 0; i < 10; i ++) {
+		ROR(3, 30);
+		CMIX30(S[0], S[1], S[2], S[4], S[5], S[6], S[15], S[16], S[17]);
+		SMIX(S[0], S[1], S[2], S[3]);
+	}
+	for (i = 0; i < 13; i ++) {
+		S[4] ^= S[0];
+		S[15] ^= S[0];
+		ROR(15, 30);
+		SMIX(S[0], S[1], S[2], S[3]);
+		S[4] ^= S[0];
+		S[16] ^= S[0];
+		ROR(14, 30);
+		SMIX(S[0], S[1], S[2], S[3]);
+	}
+	S[4] ^= S[0];
+	S[15] ^= S[0];
+	out = dst;
+	sph_enc32be(out +  0, S[ 1]);
+	sph_enc32be(out +  4, S[ 2]);
+	sph_enc32be(out +  8, S[ 3]);
+	sph_enc32be(out + 12, S[ 4]);
+	sph_enc32be(out + 16, S[15]);
+	sph_enc32be(out + 20, S[16]);
+	sph_enc32be(out + 24, S[17]);
+	if (out_size_w32 == 8) {
+		sph_enc32be(out + 28, S[18]);
+		sph_fugue256_init(sc);
+	} else {
+		sph_fugue224_init(sc);
+	}
+}
+
+static void
+fugue3_close(sph_fugue_context *sc, unsigned ub, unsigned n, void *dst)
+{
+	int i;
+
+	CLOSE_ENTRY(36, 9, fugue3_core)
+	for (i = 0; i < 18; i ++) {
+		ROR(3, 36);
+		CMIX36(S[0], S[1], S[2], S[4], S[5], S[6], S[18], S[19], S[20]);
+		SMIX(S[0], S[1], S[2], S[3]);
+	}
+	for (i = 0; i < 13; i ++) {
+		S[4] ^= S[0];
+		S[12] ^= S[0];
+		S[24] ^= S[0];
+		ROR(12, 36);
+		SMIX(S[0], S[1], S[2], S[3]);
+		S[4] ^= S[0];
+		S[13] ^= S[0];
+		S[24] ^= S[0];
+		ROR(12, 36);
+		SMIX(S[0], S[1], S[2], S[3]);
+		S[4] ^= S[0];
+		S[13] ^= S[0];
+		S[25] ^= S[0];
+		ROR(11, 36);
+		SMIX(S[0], S[1], S[2], S[3]);
+	}
+	S[4] ^= S[0];
+	S[12] ^= S[0];
+	S[24] ^= S[0];
+	out = dst;
+	sph_enc32be(out +  0, S[ 1]);
+	sph_enc32be(out +  4, S[ 2]);
+	sph_enc32be(out +  8, S[ 3]);
+	sph_enc32be(out + 12, S[ 4]);
+	sph_enc32be(out + 16, S[12]);
+	sph_enc32be(out + 20, S[13]);
+	sph_enc32be(out + 24, S[14]);
+	sph_enc32be(out + 28, S[15]);
+	sph_enc32be(out + 32, S[24]);
+	sph_enc32be(out + 36, S[25]);
+	sph_enc32be(out + 40, S[26]);
+	sph_enc32be(out + 44, S[27]);
+	sph_fugue384_init(sc);
+}
+
+static void
+fugue4_close(sph_fugue_context *sc, unsigned ub, unsigned n, void *dst)
+{
+	int i;
+
+	CLOSE_ENTRY(36, 12, fugue4_core)
+	for (i = 0; i < 32; i ++) {
+		ROR(3, 36);
+		CMIX36(S[0], S[1], S[2], S[4], S[5], S[6], S[18], S[19], S[20]);
+		SMIX(S[0], S[1], S[2], S[3]);
+	}
+	for (i = 0; i < 13; i ++) {
+		S[4] ^= S[0];
+		S[9] ^= S[0];
+		S[18] ^= S[0];
+		S[27] ^= S[0];
+		ROR(9, 36);
+		SMIX(S[0], S[1], S[2], S[3]);
+		S[4] ^= S[0];
+		S[10] ^= S[0];
+		S[18] ^= S[0];
+		S[27] ^= S[0];
+		ROR(9, 36);
+		SMIX(S[0], S[1], S[2], S[3]);
+		S[4] ^= S[0];
+		S[10] ^= S[0];
+		S[19] ^= S[0];
+		S[27] ^= S[0];
+		ROR(9, 36);
+		SMIX(S[0], S[1], S[2], S[3]);
+		S[4] ^= S[0];
+		S[10] ^= S[0];
+		S[19] ^= S[0];
+		S[28] ^= S[0];
+		ROR(8, 36);
+		SMIX(S[0], S[1], S[2], S[3]);
+	}
+	S[4] ^= S[0];
+	S[9] ^= S[0];
+	S[18] ^= S[0];
+	S[27] ^= S[0];
+	out = dst;
+	sph_enc32be(out +  0, S[ 1]);
+	sph_enc32be(out +  4, S[ 2]);
+	sph_enc32be(out +  8, S[ 3]);
+	sph_enc32be(out + 12, S[ 4]);
+	sph_enc32be(out + 16, S[ 9]);
+	sph_enc32be(out + 20, S[10]);
+	sph_enc32be(out + 24, S[11]);
+	sph_enc32be(out + 28, S[12]);
+	sph_enc32be(out + 32, S[18]);
+	sph_enc32be(out + 36, S[19]);
+	sph_enc32be(out + 40, S[20]);
+	sph_enc32be(out + 44, S[21]);
+	sph_enc32be(out + 48, S[27]);
+	sph_enc32be(out + 52, S[28]);
+	sph_enc32be(out + 56, S[29]);
+	sph_enc32be(out + 60, S[30]);
+	sph_fugue512_init(sc);
+}
+
+void
+sph_fugue224_init(void *cc)
+{
+	fugue_init(cc, 23, IV224, 7);
+}
+
+void
+sph_fugue224(void *cc, const void *data, size_t len)
+{
+	fugue2_core(cc, data, len);
+}
+
+void
+sph_fugue224_close(void *cc, void *dst)
+{
+	fugue2_close(cc, 0, 0, dst, 7);
+}
+
+void
+sph_fugue224_addbits_and_close(void *cc, unsigned ub, unsigned n, void *dst)
+{
+	fugue2_close(cc, ub, n, dst, 7);
+}
+
+void
+sph_fugue256_init(void *cc)
+{
+	fugue_init(cc, 22, IV256, 8);
+}
+
+void
+sph_fugue256(void *cc, const void *data, size_t len)
+{
+	fugue2_core(cc, data, len);
+}
+
+void
+sph_fugue256_close(void *cc, void *dst)
+{
+	fugue2_close(cc, 0, 0, dst, 8);
+}
+
+void
+sph_fugue256_addbits_and_close(void *cc, unsigned ub, unsigned n, void *dst)
+{
+	fugue2_close(cc, ub, n, dst, 8);
+}
+
+void
+sph_fugue384_init(void *cc)
+{
+	fugue_init(cc, 24, IV384, 12);
+}
+
+void
+sph_fugue384(void *cc, const void *data, size_t len)
+{
+	fugue3_core(cc, data, len);
+}
+
+void
+sph_fugue384_close(void *cc, void *dst)
+{
+	fugue3_close(cc, 0, 0, dst);
+}
+
+void
+sph_fugue384_addbits_and_close(void *cc, unsigned ub, unsigned n, void *dst)
+{
+	fugue3_close(cc, ub, n, dst);
+}
+
+void
+sph_fugue512_init(void *cc)
+{
+	fugue_init(cc, 20, IV512, 16);
+}
+
+void
+sph_fugue512(void *cc, const void *data, size_t len)
+{
+	fugue4_core(cc, data, len);
+}
+
+void
+sph_fugue512_close(void *cc, void *dst)
+{
+	fugue4_close(cc, 0, 0, dst);
+}
+
+void
+sph_fugue512_addbits_and_close(void *cc, unsigned ub, unsigned n, void *dst)
+{
+	fugue4_close(cc, ub, n, dst);
+}
+#ifdef __cplusplus
+}
+#endif
diff --git a/fuguecoin.cpp b/fuguecoin.cpp
new file mode 100644
index 0000000000..64c05d1828
--- /dev/null
+++ b/fuguecoin.cpp
@@ -0,0 +1,74 @@
+#include "uint256.h"
+#include "sph_fugue.h"
+
+#include "cpuminer-config.h"
+#include "miner.h"
+
+#include <string.h>
+#include <stdint.h>
+#include <cuda_fugue256.h>
+
+extern "C" void my_fugue256_init(void *cc);
+extern "C" void my_fugue256(void *cc, const void *data, size_t len);
+extern "C" void my_fugue256_close(void *cc, void *dst);
+extern "C" void my_fugue256_addbits_and_close(void *cc, unsigned ub, unsigned n, void *dst);
+
+// vorbereitete Kontexte nach den ersten 80 Bytes
+sph_fugue256_context  ctx_fugue_const[8];
+
+#define SWAP32(x) \
+    ((((x) << 24) & 0xff000000u) | (((x) << 8) & 0x00ff0000u)   | \
+      (((x) >> 8) & 0x0000ff00u) | (((x) >> 24) & 0x000000ffu))
+
+extern "C" int scanhash_fugue256(int thr_id, uint32_t *pdata, const uint32_t *ptarget,
+	uint32_t max_nonce, unsigned long *hashes_done)
+{	
+	uint32_t start_nonce = pdata[19];
+	const uint32_t Htarg = ptarget[7];
+	const uint32_t throughPut = 4096 * 128;
+
+	// init
+	static bool init[8] = { false, false, false, false, false, false, false, false };
+	if(!init[thr_id])
+	{
+		fugue256_cpu_init(thr_id, throughPut);
+		init[thr_id] = true;
+	}
+	
+	// Endian Drehung ist notwendig
+	uint32_t endiandata[20];
+	for (int kk=0; kk < 20; kk++)
+		be32enc(&endiandata[kk], pdata[kk]);
+
+	// Context mit dem Endian gedrehten Blockheader vorbereiten (Nonce wird später ersetzt)
+	fugue256_cpu_setBlock(thr_id, endiandata, (void*)ptarget);
+
+	do {
+		// GPU
+		uint32_t foundNounce = 0xFFFFFFFF;
+		fugue256_cpu_hash(thr_id, throughPut, pdata[19], NULL, &foundNounce);
+
+		if(foundNounce < 0xffffffff)
+		{
+			uint32_t hash[8];
+			endiandata[19] = SWAP32(foundNounce);
+			sph_fugue256_context ctx_fugue;
+			sph_fugue256_init(&ctx_fugue);
+			sph_fugue256 (&ctx_fugue, endiandata, 80);
+			sph_fugue256_close(&ctx_fugue, &hash);
+
+			pdata[19] = foundNounce;
+			*hashes_done = SWAP32(foundNounce) - start_nonce + 1;
+			return 1;
+		}
+
+		if (pdata[19] + throughPut < pdata[19])
+			pdata[19] = max_nonce;
+		else pdata[19] += throughPut;
+
+
+	} while (pdata[19] < max_nonce && !work_restart[thr_id].restart);
+	
+	*hashes_done = pdata[19] - start_nonce + 1;
+	return 0;
+}
diff --git a/groestl.c b/groestl.c
new file mode 100644
index 0000000000..cc685f4457
--- /dev/null
+++ b/groestl.c
@@ -0,0 +1,3123 @@
+/* $Id: groestl.c 260 2011-07-21 01:02:38Z tp $ */
+/*
+ * Groestl implementation.
+ *
+ * ==========================(LICENSE BEGIN)============================
+ *
+ * Copyright (c) 2007-2010  Projet RNRT SAPHIR
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining
+ * a copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sublicense, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice shall be
+ * included in all copies or substantial portions of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ * IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY
+ * CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * ===========================(LICENSE END)=============================
+ *
+ * @author   Thomas Pornin <thomas.pornin@cryptolog.com>
+ */
+
+#include <stddef.h>
+#include <string.h>
+
+#include "sph_groestl.h"
+
+#ifdef __cplusplus
+extern "C"{
+#endif
+
+#if SPH_SMALL_FOOTPRINT && !defined SPH_SMALL_FOOTPRINT_GROESTL
+#define SPH_SMALL_FOOTPRINT_GROESTL   1
+#endif
+
+/*
+ * Apparently, the 32-bit-only version is not faster than the 64-bit
+ * version unless using the "small footprint" code on a 32-bit machine.
+ */
+#if !defined SPH_GROESTL_64
+#if SPH_SMALL_FOOTPRINT_GROESTL && !SPH_64_TRUE
+#define SPH_GROESTL_64   0
+#else
+#define SPH_GROESTL_64   1
+#endif
+#endif
+
+#if !SPH_64
+#undef SPH_GROESTL_64
+#endif
+
+#ifdef _MSC_VER
+#pragma warning (disable: 4146)
+#endif
+
+/*
+ * The internal representation may use either big-endian or
+ * little-endian. Using the platform default representation speeds up
+ * encoding and decoding between bytes and the matrix columns.
+ */
+
+#undef USE_LE
+#if SPH_GROESTL_LITTLE_ENDIAN
+#define USE_LE   1
+#elif SPH_GROESTL_BIG_ENDIAN
+#define USE_LE   0
+#elif SPH_LITTLE_ENDIAN
+#define USE_LE   1
+#endif
+
+#if USE_LE
+
+#define C32e(x)     ((SPH_C32(x) >> 24) \
+                    | ((SPH_C32(x) >>  8) & SPH_C32(0x0000FF00)) \
+                    | ((SPH_C32(x) <<  8) & SPH_C32(0x00FF0000)) \
+                    | ((SPH_C32(x) << 24) & SPH_C32(0xFF000000)))
+#define dec32e_aligned   sph_dec32le_aligned
+#define enc32e           sph_enc32le
+#define B32_0(x)    ((x) & 0xFF)
+#define B32_1(x)    (((x) >> 8) & 0xFF)
+#define B32_2(x)    (((x) >> 16) & 0xFF)
+#define B32_3(x)    ((x) >> 24)
+
+#define R32u(u, d)   SPH_T32(((u) << 16) | ((d) >> 16))
+#define R32d(u, d)   SPH_T32(((u) >> 16) | ((d) << 16))
+
+#define PC32up(j, r)   ((sph_u32)((j) + (r)))
+#define PC32dn(j, r)   0
+#define QC32up(j, r)   SPH_C32(0xFFFFFFFF)
+#define QC32dn(j, r)   (((sph_u32)(r) << 24) ^ SPH_T32(~((sph_u32)(j) << 24)))
+
+#if SPH_64
+#define C64e(x)     ((SPH_C64(x) >> 56) \
+                    | ((SPH_C64(x) >> 40) & SPH_C64(0x000000000000FF00)) \
+                    | ((SPH_C64(x) >> 24) & SPH_C64(0x0000000000FF0000)) \
+                    | ((SPH_C64(x) >>  8) & SPH_C64(0x00000000FF000000)) \
+                    | ((SPH_C64(x) <<  8) & SPH_C64(0x000000FF00000000)) \
+                    | ((SPH_C64(x) << 24) & SPH_C64(0x0000FF0000000000)) \
+                    | ((SPH_C64(x) << 40) & SPH_C64(0x00FF000000000000)) \
+                    | ((SPH_C64(x) << 56) & SPH_C64(0xFF00000000000000)))
+#define dec64e_aligned   sph_dec64le_aligned
+#define enc64e           sph_enc64le
+#define B64_0(x)    ((x) & 0xFF)
+#define B64_1(x)    (((x) >> 8) & 0xFF)
+#define B64_2(x)    (((x) >> 16) & 0xFF)
+#define B64_3(x)    (((x) >> 24) & 0xFF)
+#define B64_4(x)    (((x) >> 32) & 0xFF)
+#define B64_5(x)    (((x) >> 40) & 0xFF)
+#define B64_6(x)    (((x) >> 48) & 0xFF)
+#define B64_7(x)    ((x) >> 56)
+#define R64         SPH_ROTL64
+#define PC64(j, r)  ((sph_u64)((j) + (r)))
+#define QC64(j, r)  (((sph_u64)(r) << 56) ^ SPH_T64(~((sph_u64)(j) << 56)))
+#endif
+
+#else
+
+#define C32e(x)     SPH_C32(x)
+#define dec32e_aligned   sph_dec32be_aligned
+#define enc32e           sph_enc32be
+#define B32_0(x)    ((x) >> 24)
+#define B32_1(x)    (((x) >> 16) & 0xFF)
+#define B32_2(x)    (((x) >> 8) & 0xFF)
+#define B32_3(x)    ((x) & 0xFF)
+
+#define R32u(u, d)   SPH_T32(((u) >> 16) | ((d) << 16))
+#define R32d(u, d)   SPH_T32(((u) << 16) | ((d) >> 16))
+
+#define PC32up(j, r)   ((sph_u32)((j) + (r)) << 24)
+#define PC32dn(j, r)   0
+#define QC32up(j, r)   SPH_C32(0xFFFFFFFF)
+#define QC32dn(j, r)   ((sph_u32)(r) ^ SPH_T32(~(sph_u32)(j)))
+
+#if SPH_64
+#define C64e(x)     SPH_C64(x)
+#define dec64e_aligned   sph_dec64be_aligned
+#define enc64e           sph_enc64be
+#define B64_0(x)    ((x) >> 56)
+#define B64_1(x)    (((x) >> 48) & 0xFF)
+#define B64_2(x)    (((x) >> 40) & 0xFF)
+#define B64_3(x)    (((x) >> 32) & 0xFF)
+#define B64_4(x)    (((x) >> 24) & 0xFF)
+#define B64_5(x)    (((x) >> 16) & 0xFF)
+#define B64_6(x)    (((x) >> 8) & 0xFF)
+#define B64_7(x)    ((x) & 0xFF)
+#define R64         SPH_ROTR64
+#define PC64(j, r)  ((sph_u64)((j) + (r)) << 56)
+#define QC64(j, r)  ((sph_u64)(r) ^ SPH_T64(~(sph_u64)(j)))
+#endif
+
+#endif
+
+#if SPH_GROESTL_64
+
+static const sph_u64 T0[] = {
+	C64e(0xc632f4a5f497a5c6), C64e(0xf86f978497eb84f8),
+	C64e(0xee5eb099b0c799ee), C64e(0xf67a8c8d8cf78df6),
+	C64e(0xffe8170d17e50dff), C64e(0xd60adcbddcb7bdd6),
+	C64e(0xde16c8b1c8a7b1de), C64e(0x916dfc54fc395491),
+	C64e(0x6090f050f0c05060), C64e(0x0207050305040302),
+	C64e(0xce2ee0a9e087a9ce), C64e(0x56d1877d87ac7d56),
+	C64e(0xe7cc2b192bd519e7), C64e(0xb513a662a67162b5),
+	C64e(0x4d7c31e6319ae64d), C64e(0xec59b59ab5c39aec),
+	C64e(0x8f40cf45cf05458f), C64e(0x1fa3bc9dbc3e9d1f),
+	C64e(0x8949c040c0094089), C64e(0xfa68928792ef87fa),
+	C64e(0xefd03f153fc515ef), C64e(0xb29426eb267febb2),
+	C64e(0x8ece40c94007c98e), C64e(0xfbe61d0b1ded0bfb),
+	C64e(0x416e2fec2f82ec41), C64e(0xb31aa967a97d67b3),
+	C64e(0x5f431cfd1cbefd5f), C64e(0x456025ea258aea45),
+	C64e(0x23f9dabfda46bf23), C64e(0x535102f702a6f753),
+	C64e(0xe445a196a1d396e4), C64e(0x9b76ed5bed2d5b9b),
+	C64e(0x75285dc25deac275), C64e(0xe1c5241c24d91ce1),
+	C64e(0x3dd4e9aee97aae3d), C64e(0x4cf2be6abe986a4c),
+	C64e(0x6c82ee5aeed85a6c), C64e(0x7ebdc341c3fc417e),
+	C64e(0xf5f3060206f102f5), C64e(0x8352d14fd11d4f83),
+	C64e(0x688ce45ce4d05c68), C64e(0x515607f407a2f451),
+	C64e(0xd18d5c345cb934d1), C64e(0xf9e1180818e908f9),
+	C64e(0xe24cae93aedf93e2), C64e(0xab3e9573954d73ab),
+	C64e(0x6297f553f5c45362), C64e(0x2a6b413f41543f2a),
+	C64e(0x081c140c14100c08), C64e(0x9563f652f6315295),
+	C64e(0x46e9af65af8c6546), C64e(0x9d7fe25ee2215e9d),
+	C64e(0x3048782878602830), C64e(0x37cff8a1f86ea137),
+	C64e(0x0a1b110f11140f0a), C64e(0x2febc4b5c45eb52f),
+	C64e(0x0e151b091b1c090e), C64e(0x247e5a365a483624),
+	C64e(0x1badb69bb6369b1b), C64e(0xdf98473d47a53ddf),
+	C64e(0xcda76a266a8126cd), C64e(0x4ef5bb69bb9c694e),
+	C64e(0x7f334ccd4cfecd7f), C64e(0xea50ba9fbacf9fea),
+	C64e(0x123f2d1b2d241b12), C64e(0x1da4b99eb93a9e1d),
+	C64e(0x58c49c749cb07458), C64e(0x3446722e72682e34),
+	C64e(0x3641772d776c2d36), C64e(0xdc11cdb2cda3b2dc),
+	C64e(0xb49d29ee2973eeb4), C64e(0x5b4d16fb16b6fb5b),
+	C64e(0xa4a501f60153f6a4), C64e(0x76a1d74dd7ec4d76),
+	C64e(0xb714a361a37561b7), C64e(0x7d3449ce49face7d),
+	C64e(0x52df8d7b8da47b52), C64e(0xdd9f423e42a13edd),
+	C64e(0x5ecd937193bc715e), C64e(0x13b1a297a2269713),
+	C64e(0xa6a204f50457f5a6), C64e(0xb901b868b86968b9),
+	C64e(0x0000000000000000), C64e(0xc1b5742c74992cc1),
+	C64e(0x40e0a060a0806040), C64e(0xe3c2211f21dd1fe3),
+	C64e(0x793a43c843f2c879), C64e(0xb69a2ced2c77edb6),
+	C64e(0xd40dd9bed9b3bed4), C64e(0x8d47ca46ca01468d),
+	C64e(0x671770d970ced967), C64e(0x72afdd4bdde44b72),
+	C64e(0x94ed79de7933de94), C64e(0x98ff67d4672bd498),
+	C64e(0xb09323e8237be8b0), C64e(0x855bde4ade114a85),
+	C64e(0xbb06bd6bbd6d6bbb), C64e(0xc5bb7e2a7e912ac5),
+	C64e(0x4f7b34e5349ee54f), C64e(0xedd73a163ac116ed),
+	C64e(0x86d254c55417c586), C64e(0x9af862d7622fd79a),
+	C64e(0x6699ff55ffcc5566), C64e(0x11b6a794a7229411),
+	C64e(0x8ac04acf4a0fcf8a), C64e(0xe9d9301030c910e9),
+	C64e(0x040e0a060a080604), C64e(0xfe66988198e781fe),
+	C64e(0xa0ab0bf00b5bf0a0), C64e(0x78b4cc44ccf04478),
+	C64e(0x25f0d5bad54aba25), C64e(0x4b753ee33e96e34b),
+	C64e(0xa2ac0ef30e5ff3a2), C64e(0x5d4419fe19bafe5d),
+	C64e(0x80db5bc05b1bc080), C64e(0x0580858a850a8a05),
+	C64e(0x3fd3ecadec7ead3f), C64e(0x21fedfbcdf42bc21),
+	C64e(0x70a8d848d8e04870), C64e(0xf1fd0c040cf904f1),
+	C64e(0x63197adf7ac6df63), C64e(0x772f58c158eec177),
+	C64e(0xaf309f759f4575af), C64e(0x42e7a563a5846342),
+	C64e(0x2070503050403020), C64e(0xe5cb2e1a2ed11ae5),
+	C64e(0xfdef120e12e10efd), C64e(0xbf08b76db7656dbf),
+	C64e(0x8155d44cd4194c81), C64e(0x18243c143c301418),
+	C64e(0x26795f355f4c3526), C64e(0xc3b2712f719d2fc3),
+	C64e(0xbe8638e13867e1be), C64e(0x35c8fda2fd6aa235),
+	C64e(0x88c74fcc4f0bcc88), C64e(0x2e654b394b5c392e),
+	C64e(0x936af957f93d5793), C64e(0x55580df20daaf255),
+	C64e(0xfc619d829de382fc), C64e(0x7ab3c947c9f4477a),
+	C64e(0xc827efacef8bacc8), C64e(0xba8832e7326fe7ba),
+	C64e(0x324f7d2b7d642b32), C64e(0xe642a495a4d795e6),
+	C64e(0xc03bfba0fb9ba0c0), C64e(0x19aab398b3329819),
+	C64e(0x9ef668d16827d19e), C64e(0xa322817f815d7fa3),
+	C64e(0x44eeaa66aa886644), C64e(0x54d6827e82a87e54),
+	C64e(0x3bdde6abe676ab3b), C64e(0x0b959e839e16830b),
+	C64e(0x8cc945ca4503ca8c), C64e(0xc7bc7b297b9529c7),
+	C64e(0x6b056ed36ed6d36b), C64e(0x286c443c44503c28),
+	C64e(0xa72c8b798b5579a7), C64e(0xbc813de23d63e2bc),
+	C64e(0x1631271d272c1d16), C64e(0xad379a769a4176ad),
+	C64e(0xdb964d3b4dad3bdb), C64e(0x649efa56fac85664),
+	C64e(0x74a6d24ed2e84e74), C64e(0x1436221e22281e14),
+	C64e(0x92e476db763fdb92), C64e(0x0c121e0a1e180a0c),
+	C64e(0x48fcb46cb4906c48), C64e(0xb88f37e4376be4b8),
+	C64e(0x9f78e75de7255d9f), C64e(0xbd0fb26eb2616ebd),
+	C64e(0x43692aef2a86ef43), C64e(0xc435f1a6f193a6c4),
+	C64e(0x39dae3a8e372a839), C64e(0x31c6f7a4f762a431),
+	C64e(0xd38a593759bd37d3), C64e(0xf274868b86ff8bf2),
+	C64e(0xd583563256b132d5), C64e(0x8b4ec543c50d438b),
+	C64e(0x6e85eb59ebdc596e), C64e(0xda18c2b7c2afb7da),
+	C64e(0x018e8f8c8f028c01), C64e(0xb11dac64ac7964b1),
+	C64e(0x9cf16dd26d23d29c), C64e(0x49723be03b92e049),
+	C64e(0xd81fc7b4c7abb4d8), C64e(0xacb915fa1543faac),
+	C64e(0xf3fa090709fd07f3), C64e(0xcfa06f256f8525cf),
+	C64e(0xca20eaafea8fafca), C64e(0xf47d898e89f38ef4),
+	C64e(0x476720e9208ee947), C64e(0x1038281828201810),
+	C64e(0x6f0b64d564ded56f), C64e(0xf073838883fb88f0),
+	C64e(0x4afbb16fb1946f4a), C64e(0x5cca967296b8725c),
+	C64e(0x38546c246c702438), C64e(0x575f08f108aef157),
+	C64e(0x732152c752e6c773), C64e(0x9764f351f3355197),
+	C64e(0xcbae6523658d23cb), C64e(0xa125847c84597ca1),
+	C64e(0xe857bf9cbfcb9ce8), C64e(0x3e5d6321637c213e),
+	C64e(0x96ea7cdd7c37dd96), C64e(0x611e7fdc7fc2dc61),
+	C64e(0x0d9c9186911a860d), C64e(0x0f9b9485941e850f),
+	C64e(0xe04bab90abdb90e0), C64e(0x7cbac642c6f8427c),
+	C64e(0x712657c457e2c471), C64e(0xcc29e5aae583aacc),
+	C64e(0x90e373d8733bd890), C64e(0x06090f050f0c0506),
+	C64e(0xf7f4030103f501f7), C64e(0x1c2a36123638121c),
+	C64e(0xc23cfea3fe9fa3c2), C64e(0x6a8be15fe1d45f6a),
+	C64e(0xaebe10f91047f9ae), C64e(0x69026bd06bd2d069),
+	C64e(0x17bfa891a82e9117), C64e(0x9971e858e8295899),
+	C64e(0x3a5369276974273a), C64e(0x27f7d0b9d04eb927),
+	C64e(0xd991483848a938d9), C64e(0xebde351335cd13eb),
+	C64e(0x2be5ceb3ce56b32b), C64e(0x2277553355443322),
+	C64e(0xd204d6bbd6bfbbd2), C64e(0xa9399070904970a9),
+	C64e(0x07878089800e8907), C64e(0x33c1f2a7f266a733),
+	C64e(0x2decc1b6c15ab62d), C64e(0x3c5a66226678223c),
+	C64e(0x15b8ad92ad2a9215), C64e(0xc9a96020608920c9),
+	C64e(0x875cdb49db154987), C64e(0xaab01aff1a4fffaa),
+	C64e(0x50d8887888a07850), C64e(0xa52b8e7a8e517aa5),
+	C64e(0x03898a8f8a068f03), C64e(0x594a13f813b2f859),
+	C64e(0x09929b809b128009), C64e(0x1a2339173934171a),
+	C64e(0x651075da75cada65), C64e(0xd784533153b531d7),
+	C64e(0x84d551c65113c684), C64e(0xd003d3b8d3bbb8d0),
+	C64e(0x82dc5ec35e1fc382), C64e(0x29e2cbb0cb52b029),
+	C64e(0x5ac3997799b4775a), C64e(0x1e2d3311333c111e),
+	C64e(0x7b3d46cb46f6cb7b), C64e(0xa8b71ffc1f4bfca8),
+	C64e(0x6d0c61d661dad66d), C64e(0x2c624e3a4e583a2c)
+};
+
+#if !SPH_SMALL_FOOTPRINT_GROESTL
+
+static const sph_u64 T1[] = {
+	C64e(0xc6c632f4a5f497a5), C64e(0xf8f86f978497eb84),
+	C64e(0xeeee5eb099b0c799), C64e(0xf6f67a8c8d8cf78d),
+	C64e(0xffffe8170d17e50d), C64e(0xd6d60adcbddcb7bd),
+	C64e(0xdede16c8b1c8a7b1), C64e(0x91916dfc54fc3954),
+	C64e(0x606090f050f0c050), C64e(0x0202070503050403),
+	C64e(0xcece2ee0a9e087a9), C64e(0x5656d1877d87ac7d),
+	C64e(0xe7e7cc2b192bd519), C64e(0xb5b513a662a67162),
+	C64e(0x4d4d7c31e6319ae6), C64e(0xecec59b59ab5c39a),
+	C64e(0x8f8f40cf45cf0545), C64e(0x1f1fa3bc9dbc3e9d),
+	C64e(0x898949c040c00940), C64e(0xfafa68928792ef87),
+	C64e(0xefefd03f153fc515), C64e(0xb2b29426eb267feb),
+	C64e(0x8e8ece40c94007c9), C64e(0xfbfbe61d0b1ded0b),
+	C64e(0x41416e2fec2f82ec), C64e(0xb3b31aa967a97d67),
+	C64e(0x5f5f431cfd1cbefd), C64e(0x45456025ea258aea),
+	C64e(0x2323f9dabfda46bf), C64e(0x53535102f702a6f7),
+	C64e(0xe4e445a196a1d396), C64e(0x9b9b76ed5bed2d5b),
+	C64e(0x7575285dc25deac2), C64e(0xe1e1c5241c24d91c),
+	C64e(0x3d3dd4e9aee97aae), C64e(0x4c4cf2be6abe986a),
+	C64e(0x6c6c82ee5aeed85a), C64e(0x7e7ebdc341c3fc41),
+	C64e(0xf5f5f3060206f102), C64e(0x838352d14fd11d4f),
+	C64e(0x68688ce45ce4d05c), C64e(0x51515607f407a2f4),
+	C64e(0xd1d18d5c345cb934), C64e(0xf9f9e1180818e908),
+	C64e(0xe2e24cae93aedf93), C64e(0xabab3e9573954d73),
+	C64e(0x626297f553f5c453), C64e(0x2a2a6b413f41543f),
+	C64e(0x08081c140c14100c), C64e(0x959563f652f63152),
+	C64e(0x4646e9af65af8c65), C64e(0x9d9d7fe25ee2215e),
+	C64e(0x3030487828786028), C64e(0x3737cff8a1f86ea1),
+	C64e(0x0a0a1b110f11140f), C64e(0x2f2febc4b5c45eb5),
+	C64e(0x0e0e151b091b1c09), C64e(0x24247e5a365a4836),
+	C64e(0x1b1badb69bb6369b), C64e(0xdfdf98473d47a53d),
+	C64e(0xcdcda76a266a8126), C64e(0x4e4ef5bb69bb9c69),
+	C64e(0x7f7f334ccd4cfecd), C64e(0xeaea50ba9fbacf9f),
+	C64e(0x12123f2d1b2d241b), C64e(0x1d1da4b99eb93a9e),
+	C64e(0x5858c49c749cb074), C64e(0x343446722e72682e),
+	C64e(0x363641772d776c2d), C64e(0xdcdc11cdb2cda3b2),
+	C64e(0xb4b49d29ee2973ee), C64e(0x5b5b4d16fb16b6fb),
+	C64e(0xa4a4a501f60153f6), C64e(0x7676a1d74dd7ec4d),
+	C64e(0xb7b714a361a37561), C64e(0x7d7d3449ce49face),
+	C64e(0x5252df8d7b8da47b), C64e(0xdddd9f423e42a13e),
+	C64e(0x5e5ecd937193bc71), C64e(0x1313b1a297a22697),
+	C64e(0xa6a6a204f50457f5), C64e(0xb9b901b868b86968),
+	C64e(0x0000000000000000), C64e(0xc1c1b5742c74992c),
+	C64e(0x4040e0a060a08060), C64e(0xe3e3c2211f21dd1f),
+	C64e(0x79793a43c843f2c8), C64e(0xb6b69a2ced2c77ed),
+	C64e(0xd4d40dd9bed9b3be), C64e(0x8d8d47ca46ca0146),
+	C64e(0x67671770d970ced9), C64e(0x7272afdd4bdde44b),
+	C64e(0x9494ed79de7933de), C64e(0x9898ff67d4672bd4),
+	C64e(0xb0b09323e8237be8), C64e(0x85855bde4ade114a),
+	C64e(0xbbbb06bd6bbd6d6b), C64e(0xc5c5bb7e2a7e912a),
+	C64e(0x4f4f7b34e5349ee5), C64e(0xededd73a163ac116),
+	C64e(0x8686d254c55417c5), C64e(0x9a9af862d7622fd7),
+	C64e(0x666699ff55ffcc55), C64e(0x1111b6a794a72294),
+	C64e(0x8a8ac04acf4a0fcf), C64e(0xe9e9d9301030c910),
+	C64e(0x04040e0a060a0806), C64e(0xfefe66988198e781),
+	C64e(0xa0a0ab0bf00b5bf0), C64e(0x7878b4cc44ccf044),
+	C64e(0x2525f0d5bad54aba), C64e(0x4b4b753ee33e96e3),
+	C64e(0xa2a2ac0ef30e5ff3), C64e(0x5d5d4419fe19bafe),
+	C64e(0x8080db5bc05b1bc0), C64e(0x050580858a850a8a),
+	C64e(0x3f3fd3ecadec7ead), C64e(0x2121fedfbcdf42bc),
+	C64e(0x7070a8d848d8e048), C64e(0xf1f1fd0c040cf904),
+	C64e(0x6363197adf7ac6df), C64e(0x77772f58c158eec1),
+	C64e(0xafaf309f759f4575), C64e(0x4242e7a563a58463),
+	C64e(0x2020705030504030), C64e(0xe5e5cb2e1a2ed11a),
+	C64e(0xfdfdef120e12e10e), C64e(0xbfbf08b76db7656d),
+	C64e(0x818155d44cd4194c), C64e(0x1818243c143c3014),
+	C64e(0x2626795f355f4c35), C64e(0xc3c3b2712f719d2f),
+	C64e(0xbebe8638e13867e1), C64e(0x3535c8fda2fd6aa2),
+	C64e(0x8888c74fcc4f0bcc), C64e(0x2e2e654b394b5c39),
+	C64e(0x93936af957f93d57), C64e(0x5555580df20daaf2),
+	C64e(0xfcfc619d829de382), C64e(0x7a7ab3c947c9f447),
+	C64e(0xc8c827efacef8bac), C64e(0xbaba8832e7326fe7),
+	C64e(0x32324f7d2b7d642b), C64e(0xe6e642a495a4d795),
+	C64e(0xc0c03bfba0fb9ba0), C64e(0x1919aab398b33298),
+	C64e(0x9e9ef668d16827d1), C64e(0xa3a322817f815d7f),
+	C64e(0x4444eeaa66aa8866), C64e(0x5454d6827e82a87e),
+	C64e(0x3b3bdde6abe676ab), C64e(0x0b0b959e839e1683),
+	C64e(0x8c8cc945ca4503ca), C64e(0xc7c7bc7b297b9529),
+	C64e(0x6b6b056ed36ed6d3), C64e(0x28286c443c44503c),
+	C64e(0xa7a72c8b798b5579), C64e(0xbcbc813de23d63e2),
+	C64e(0x161631271d272c1d), C64e(0xadad379a769a4176),
+	C64e(0xdbdb964d3b4dad3b), C64e(0x64649efa56fac856),
+	C64e(0x7474a6d24ed2e84e), C64e(0x141436221e22281e),
+	C64e(0x9292e476db763fdb), C64e(0x0c0c121e0a1e180a),
+	C64e(0x4848fcb46cb4906c), C64e(0xb8b88f37e4376be4),
+	C64e(0x9f9f78e75de7255d), C64e(0xbdbd0fb26eb2616e),
+	C64e(0x4343692aef2a86ef), C64e(0xc4c435f1a6f193a6),
+	C64e(0x3939dae3a8e372a8), C64e(0x3131c6f7a4f762a4),
+	C64e(0xd3d38a593759bd37), C64e(0xf2f274868b86ff8b),
+	C64e(0xd5d583563256b132), C64e(0x8b8b4ec543c50d43),
+	C64e(0x6e6e85eb59ebdc59), C64e(0xdada18c2b7c2afb7),
+	C64e(0x01018e8f8c8f028c), C64e(0xb1b11dac64ac7964),
+	C64e(0x9c9cf16dd26d23d2), C64e(0x4949723be03b92e0),
+	C64e(0xd8d81fc7b4c7abb4), C64e(0xacacb915fa1543fa),
+	C64e(0xf3f3fa090709fd07), C64e(0xcfcfa06f256f8525),
+	C64e(0xcaca20eaafea8faf), C64e(0xf4f47d898e89f38e),
+	C64e(0x47476720e9208ee9), C64e(0x1010382818282018),
+	C64e(0x6f6f0b64d564ded5), C64e(0xf0f073838883fb88),
+	C64e(0x4a4afbb16fb1946f), C64e(0x5c5cca967296b872),
+	C64e(0x3838546c246c7024), C64e(0x57575f08f108aef1),
+	C64e(0x73732152c752e6c7), C64e(0x979764f351f33551),
+	C64e(0xcbcbae6523658d23), C64e(0xa1a125847c84597c),
+	C64e(0xe8e857bf9cbfcb9c), C64e(0x3e3e5d6321637c21),
+	C64e(0x9696ea7cdd7c37dd), C64e(0x61611e7fdc7fc2dc),
+	C64e(0x0d0d9c9186911a86), C64e(0x0f0f9b9485941e85),
+	C64e(0xe0e04bab90abdb90), C64e(0x7c7cbac642c6f842),
+	C64e(0x71712657c457e2c4), C64e(0xcccc29e5aae583aa),
+	C64e(0x9090e373d8733bd8), C64e(0x0606090f050f0c05),
+	C64e(0xf7f7f4030103f501), C64e(0x1c1c2a3612363812),
+	C64e(0xc2c23cfea3fe9fa3), C64e(0x6a6a8be15fe1d45f),
+	C64e(0xaeaebe10f91047f9), C64e(0x6969026bd06bd2d0),
+	C64e(0x1717bfa891a82e91), C64e(0x999971e858e82958),
+	C64e(0x3a3a536927697427), C64e(0x2727f7d0b9d04eb9),
+	C64e(0xd9d991483848a938), C64e(0xebebde351335cd13),
+	C64e(0x2b2be5ceb3ce56b3), C64e(0x2222775533554433),
+	C64e(0xd2d204d6bbd6bfbb), C64e(0xa9a9399070904970),
+	C64e(0x0707878089800e89), C64e(0x3333c1f2a7f266a7),
+	C64e(0x2d2decc1b6c15ab6), C64e(0x3c3c5a6622667822),
+	C64e(0x1515b8ad92ad2a92), C64e(0xc9c9a96020608920),
+	C64e(0x87875cdb49db1549), C64e(0xaaaab01aff1a4fff),
+	C64e(0x5050d8887888a078), C64e(0xa5a52b8e7a8e517a),
+	C64e(0x0303898a8f8a068f), C64e(0x59594a13f813b2f8),
+	C64e(0x0909929b809b1280), C64e(0x1a1a233917393417),
+	C64e(0x65651075da75cada), C64e(0xd7d784533153b531),
+	C64e(0x8484d551c65113c6), C64e(0xd0d003d3b8d3bbb8),
+	C64e(0x8282dc5ec35e1fc3), C64e(0x2929e2cbb0cb52b0),
+	C64e(0x5a5ac3997799b477), C64e(0x1e1e2d3311333c11),
+	C64e(0x7b7b3d46cb46f6cb), C64e(0xa8a8b71ffc1f4bfc),
+	C64e(0x6d6d0c61d661dad6), C64e(0x2c2c624e3a4e583a)
+};
+
+static const sph_u64 T2[] = {
+	C64e(0xa5c6c632f4a5f497), C64e(0x84f8f86f978497eb),
+	C64e(0x99eeee5eb099b0c7), C64e(0x8df6f67a8c8d8cf7),
+	C64e(0x0dffffe8170d17e5), C64e(0xbdd6d60adcbddcb7),
+	C64e(0xb1dede16c8b1c8a7), C64e(0x5491916dfc54fc39),
+	C64e(0x50606090f050f0c0), C64e(0x0302020705030504),
+	C64e(0xa9cece2ee0a9e087), C64e(0x7d5656d1877d87ac),
+	C64e(0x19e7e7cc2b192bd5), C64e(0x62b5b513a662a671),
+	C64e(0xe64d4d7c31e6319a), C64e(0x9aecec59b59ab5c3),
+	C64e(0x458f8f40cf45cf05), C64e(0x9d1f1fa3bc9dbc3e),
+	C64e(0x40898949c040c009), C64e(0x87fafa68928792ef),
+	C64e(0x15efefd03f153fc5), C64e(0xebb2b29426eb267f),
+	C64e(0xc98e8ece40c94007), C64e(0x0bfbfbe61d0b1ded),
+	C64e(0xec41416e2fec2f82), C64e(0x67b3b31aa967a97d),
+	C64e(0xfd5f5f431cfd1cbe), C64e(0xea45456025ea258a),
+	C64e(0xbf2323f9dabfda46), C64e(0xf753535102f702a6),
+	C64e(0x96e4e445a196a1d3), C64e(0x5b9b9b76ed5bed2d),
+	C64e(0xc27575285dc25dea), C64e(0x1ce1e1c5241c24d9),
+	C64e(0xae3d3dd4e9aee97a), C64e(0x6a4c4cf2be6abe98),
+	C64e(0x5a6c6c82ee5aeed8), C64e(0x417e7ebdc341c3fc),
+	C64e(0x02f5f5f3060206f1), C64e(0x4f838352d14fd11d),
+	C64e(0x5c68688ce45ce4d0), C64e(0xf451515607f407a2),
+	C64e(0x34d1d18d5c345cb9), C64e(0x08f9f9e1180818e9),
+	C64e(0x93e2e24cae93aedf), C64e(0x73abab3e9573954d),
+	C64e(0x53626297f553f5c4), C64e(0x3f2a2a6b413f4154),
+	C64e(0x0c08081c140c1410), C64e(0x52959563f652f631),
+	C64e(0x654646e9af65af8c), C64e(0x5e9d9d7fe25ee221),
+	C64e(0x2830304878287860), C64e(0xa13737cff8a1f86e),
+	C64e(0x0f0a0a1b110f1114), C64e(0xb52f2febc4b5c45e),
+	C64e(0x090e0e151b091b1c), C64e(0x3624247e5a365a48),
+	C64e(0x9b1b1badb69bb636), C64e(0x3ddfdf98473d47a5),
+	C64e(0x26cdcda76a266a81), C64e(0x694e4ef5bb69bb9c),
+	C64e(0xcd7f7f334ccd4cfe), C64e(0x9feaea50ba9fbacf),
+	C64e(0x1b12123f2d1b2d24), C64e(0x9e1d1da4b99eb93a),
+	C64e(0x745858c49c749cb0), C64e(0x2e343446722e7268),
+	C64e(0x2d363641772d776c), C64e(0xb2dcdc11cdb2cda3),
+	C64e(0xeeb4b49d29ee2973), C64e(0xfb5b5b4d16fb16b6),
+	C64e(0xf6a4a4a501f60153), C64e(0x4d7676a1d74dd7ec),
+	C64e(0x61b7b714a361a375), C64e(0xce7d7d3449ce49fa),
+	C64e(0x7b5252df8d7b8da4), C64e(0x3edddd9f423e42a1),
+	C64e(0x715e5ecd937193bc), C64e(0x971313b1a297a226),
+	C64e(0xf5a6a6a204f50457), C64e(0x68b9b901b868b869),
+	C64e(0x0000000000000000), C64e(0x2cc1c1b5742c7499),
+	C64e(0x604040e0a060a080), C64e(0x1fe3e3c2211f21dd),
+	C64e(0xc879793a43c843f2), C64e(0xedb6b69a2ced2c77),
+	C64e(0xbed4d40dd9bed9b3), C64e(0x468d8d47ca46ca01),
+	C64e(0xd967671770d970ce), C64e(0x4b7272afdd4bdde4),
+	C64e(0xde9494ed79de7933), C64e(0xd49898ff67d4672b),
+	C64e(0xe8b0b09323e8237b), C64e(0x4a85855bde4ade11),
+	C64e(0x6bbbbb06bd6bbd6d), C64e(0x2ac5c5bb7e2a7e91),
+	C64e(0xe54f4f7b34e5349e), C64e(0x16ededd73a163ac1),
+	C64e(0xc58686d254c55417), C64e(0xd79a9af862d7622f),
+	C64e(0x55666699ff55ffcc), C64e(0x941111b6a794a722),
+	C64e(0xcf8a8ac04acf4a0f), C64e(0x10e9e9d9301030c9),
+	C64e(0x0604040e0a060a08), C64e(0x81fefe66988198e7),
+	C64e(0xf0a0a0ab0bf00b5b), C64e(0x447878b4cc44ccf0),
+	C64e(0xba2525f0d5bad54a), C64e(0xe34b4b753ee33e96),
+	C64e(0xf3a2a2ac0ef30e5f), C64e(0xfe5d5d4419fe19ba),
+	C64e(0xc08080db5bc05b1b), C64e(0x8a050580858a850a),
+	C64e(0xad3f3fd3ecadec7e), C64e(0xbc2121fedfbcdf42),
+	C64e(0x487070a8d848d8e0), C64e(0x04f1f1fd0c040cf9),
+	C64e(0xdf6363197adf7ac6), C64e(0xc177772f58c158ee),
+	C64e(0x75afaf309f759f45), C64e(0x634242e7a563a584),
+	C64e(0x3020207050305040), C64e(0x1ae5e5cb2e1a2ed1),
+	C64e(0x0efdfdef120e12e1), C64e(0x6dbfbf08b76db765),
+	C64e(0x4c818155d44cd419), C64e(0x141818243c143c30),
+	C64e(0x352626795f355f4c), C64e(0x2fc3c3b2712f719d),
+	C64e(0xe1bebe8638e13867), C64e(0xa23535c8fda2fd6a),
+	C64e(0xcc8888c74fcc4f0b), C64e(0x392e2e654b394b5c),
+	C64e(0x5793936af957f93d), C64e(0xf25555580df20daa),
+	C64e(0x82fcfc619d829de3), C64e(0x477a7ab3c947c9f4),
+	C64e(0xacc8c827efacef8b), C64e(0xe7baba8832e7326f),
+	C64e(0x2b32324f7d2b7d64), C64e(0x95e6e642a495a4d7),
+	C64e(0xa0c0c03bfba0fb9b), C64e(0x981919aab398b332),
+	C64e(0xd19e9ef668d16827), C64e(0x7fa3a322817f815d),
+	C64e(0x664444eeaa66aa88), C64e(0x7e5454d6827e82a8),
+	C64e(0xab3b3bdde6abe676), C64e(0x830b0b959e839e16),
+	C64e(0xca8c8cc945ca4503), C64e(0x29c7c7bc7b297b95),
+	C64e(0xd36b6b056ed36ed6), C64e(0x3c28286c443c4450),
+	C64e(0x79a7a72c8b798b55), C64e(0xe2bcbc813de23d63),
+	C64e(0x1d161631271d272c), C64e(0x76adad379a769a41),
+	C64e(0x3bdbdb964d3b4dad), C64e(0x5664649efa56fac8),
+	C64e(0x4e7474a6d24ed2e8), C64e(0x1e141436221e2228),
+	C64e(0xdb9292e476db763f), C64e(0x0a0c0c121e0a1e18),
+	C64e(0x6c4848fcb46cb490), C64e(0xe4b8b88f37e4376b),
+	C64e(0x5d9f9f78e75de725), C64e(0x6ebdbd0fb26eb261),
+	C64e(0xef4343692aef2a86), C64e(0xa6c4c435f1a6f193),
+	C64e(0xa83939dae3a8e372), C64e(0xa43131c6f7a4f762),
+	C64e(0x37d3d38a593759bd), C64e(0x8bf2f274868b86ff),
+	C64e(0x32d5d583563256b1), C64e(0x438b8b4ec543c50d),
+	C64e(0x596e6e85eb59ebdc), C64e(0xb7dada18c2b7c2af),
+	C64e(0x8c01018e8f8c8f02), C64e(0x64b1b11dac64ac79),
+	C64e(0xd29c9cf16dd26d23), C64e(0xe04949723be03b92),
+	C64e(0xb4d8d81fc7b4c7ab), C64e(0xfaacacb915fa1543),
+	C64e(0x07f3f3fa090709fd), C64e(0x25cfcfa06f256f85),
+	C64e(0xafcaca20eaafea8f), C64e(0x8ef4f47d898e89f3),
+	C64e(0xe947476720e9208e), C64e(0x1810103828182820),
+	C64e(0xd56f6f0b64d564de), C64e(0x88f0f073838883fb),
+	C64e(0x6f4a4afbb16fb194), C64e(0x725c5cca967296b8),
+	C64e(0x243838546c246c70), C64e(0xf157575f08f108ae),
+	C64e(0xc773732152c752e6), C64e(0x51979764f351f335),
+	C64e(0x23cbcbae6523658d), C64e(0x7ca1a125847c8459),
+	C64e(0x9ce8e857bf9cbfcb), C64e(0x213e3e5d6321637c),
+	C64e(0xdd9696ea7cdd7c37), C64e(0xdc61611e7fdc7fc2),
+	C64e(0x860d0d9c9186911a), C64e(0x850f0f9b9485941e),
+	C64e(0x90e0e04bab90abdb), C64e(0x427c7cbac642c6f8),
+	C64e(0xc471712657c457e2), C64e(0xaacccc29e5aae583),
+	C64e(0xd89090e373d8733b), C64e(0x050606090f050f0c),
+	C64e(0x01f7f7f4030103f5), C64e(0x121c1c2a36123638),
+	C64e(0xa3c2c23cfea3fe9f), C64e(0x5f6a6a8be15fe1d4),
+	C64e(0xf9aeaebe10f91047), C64e(0xd06969026bd06bd2),
+	C64e(0x911717bfa891a82e), C64e(0x58999971e858e829),
+	C64e(0x273a3a5369276974), C64e(0xb92727f7d0b9d04e),
+	C64e(0x38d9d991483848a9), C64e(0x13ebebde351335cd),
+	C64e(0xb32b2be5ceb3ce56), C64e(0x3322227755335544),
+	C64e(0xbbd2d204d6bbd6bf), C64e(0x70a9a93990709049),
+	C64e(0x890707878089800e), C64e(0xa73333c1f2a7f266),
+	C64e(0xb62d2decc1b6c15a), C64e(0x223c3c5a66226678),
+	C64e(0x921515b8ad92ad2a), C64e(0x20c9c9a960206089),
+	C64e(0x4987875cdb49db15), C64e(0xffaaaab01aff1a4f),
+	C64e(0x785050d8887888a0), C64e(0x7aa5a52b8e7a8e51),
+	C64e(0x8f0303898a8f8a06), C64e(0xf859594a13f813b2),
+	C64e(0x800909929b809b12), C64e(0x171a1a2339173934),
+	C64e(0xda65651075da75ca), C64e(0x31d7d784533153b5),
+	C64e(0xc68484d551c65113), C64e(0xb8d0d003d3b8d3bb),
+	C64e(0xc38282dc5ec35e1f), C64e(0xb02929e2cbb0cb52),
+	C64e(0x775a5ac3997799b4), C64e(0x111e1e2d3311333c),
+	C64e(0xcb7b7b3d46cb46f6), C64e(0xfca8a8b71ffc1f4b),
+	C64e(0xd66d6d0c61d661da), C64e(0x3a2c2c624e3a4e58)
+};
+
+static const sph_u64 T3[] = {
+	C64e(0x97a5c6c632f4a5f4), C64e(0xeb84f8f86f978497),
+	C64e(0xc799eeee5eb099b0), C64e(0xf78df6f67a8c8d8c),
+	C64e(0xe50dffffe8170d17), C64e(0xb7bdd6d60adcbddc),
+	C64e(0xa7b1dede16c8b1c8), C64e(0x395491916dfc54fc),
+	C64e(0xc050606090f050f0), C64e(0x0403020207050305),
+	C64e(0x87a9cece2ee0a9e0), C64e(0xac7d5656d1877d87),
+	C64e(0xd519e7e7cc2b192b), C64e(0x7162b5b513a662a6),
+	C64e(0x9ae64d4d7c31e631), C64e(0xc39aecec59b59ab5),
+	C64e(0x05458f8f40cf45cf), C64e(0x3e9d1f1fa3bc9dbc),
+	C64e(0x0940898949c040c0), C64e(0xef87fafa68928792),
+	C64e(0xc515efefd03f153f), C64e(0x7febb2b29426eb26),
+	C64e(0x07c98e8ece40c940), C64e(0xed0bfbfbe61d0b1d),
+	C64e(0x82ec41416e2fec2f), C64e(0x7d67b3b31aa967a9),
+	C64e(0xbefd5f5f431cfd1c), C64e(0x8aea45456025ea25),
+	C64e(0x46bf2323f9dabfda), C64e(0xa6f753535102f702),
+	C64e(0xd396e4e445a196a1), C64e(0x2d5b9b9b76ed5bed),
+	C64e(0xeac27575285dc25d), C64e(0xd91ce1e1c5241c24),
+	C64e(0x7aae3d3dd4e9aee9), C64e(0x986a4c4cf2be6abe),
+	C64e(0xd85a6c6c82ee5aee), C64e(0xfc417e7ebdc341c3),
+	C64e(0xf102f5f5f3060206), C64e(0x1d4f838352d14fd1),
+	C64e(0xd05c68688ce45ce4), C64e(0xa2f451515607f407),
+	C64e(0xb934d1d18d5c345c), C64e(0xe908f9f9e1180818),
+	C64e(0xdf93e2e24cae93ae), C64e(0x4d73abab3e957395),
+	C64e(0xc453626297f553f5), C64e(0x543f2a2a6b413f41),
+	C64e(0x100c08081c140c14), C64e(0x3152959563f652f6),
+	C64e(0x8c654646e9af65af), C64e(0x215e9d9d7fe25ee2),
+	C64e(0x6028303048782878), C64e(0x6ea13737cff8a1f8),
+	C64e(0x140f0a0a1b110f11), C64e(0x5eb52f2febc4b5c4),
+	C64e(0x1c090e0e151b091b), C64e(0x483624247e5a365a),
+	C64e(0x369b1b1badb69bb6), C64e(0xa53ddfdf98473d47),
+	C64e(0x8126cdcda76a266a), C64e(0x9c694e4ef5bb69bb),
+	C64e(0xfecd7f7f334ccd4c), C64e(0xcf9feaea50ba9fba),
+	C64e(0x241b12123f2d1b2d), C64e(0x3a9e1d1da4b99eb9),
+	C64e(0xb0745858c49c749c), C64e(0x682e343446722e72),
+	C64e(0x6c2d363641772d77), C64e(0xa3b2dcdc11cdb2cd),
+	C64e(0x73eeb4b49d29ee29), C64e(0xb6fb5b5b4d16fb16),
+	C64e(0x53f6a4a4a501f601), C64e(0xec4d7676a1d74dd7),
+	C64e(0x7561b7b714a361a3), C64e(0xface7d7d3449ce49),
+	C64e(0xa47b5252df8d7b8d), C64e(0xa13edddd9f423e42),
+	C64e(0xbc715e5ecd937193), C64e(0x26971313b1a297a2),
+	C64e(0x57f5a6a6a204f504), C64e(0x6968b9b901b868b8),
+	C64e(0x0000000000000000), C64e(0x992cc1c1b5742c74),
+	C64e(0x80604040e0a060a0), C64e(0xdd1fe3e3c2211f21),
+	C64e(0xf2c879793a43c843), C64e(0x77edb6b69a2ced2c),
+	C64e(0xb3bed4d40dd9bed9), C64e(0x01468d8d47ca46ca),
+	C64e(0xced967671770d970), C64e(0xe44b7272afdd4bdd),
+	C64e(0x33de9494ed79de79), C64e(0x2bd49898ff67d467),
+	C64e(0x7be8b0b09323e823), C64e(0x114a85855bde4ade),
+	C64e(0x6d6bbbbb06bd6bbd), C64e(0x912ac5c5bb7e2a7e),
+	C64e(0x9ee54f4f7b34e534), C64e(0xc116ededd73a163a),
+	C64e(0x17c58686d254c554), C64e(0x2fd79a9af862d762),
+	C64e(0xcc55666699ff55ff), C64e(0x22941111b6a794a7),
+	C64e(0x0fcf8a8ac04acf4a), C64e(0xc910e9e9d9301030),
+	C64e(0x080604040e0a060a), C64e(0xe781fefe66988198),
+	C64e(0x5bf0a0a0ab0bf00b), C64e(0xf0447878b4cc44cc),
+	C64e(0x4aba2525f0d5bad5), C64e(0x96e34b4b753ee33e),
+	C64e(0x5ff3a2a2ac0ef30e), C64e(0xbafe5d5d4419fe19),
+	C64e(0x1bc08080db5bc05b), C64e(0x0a8a050580858a85),
+	C64e(0x7ead3f3fd3ecadec), C64e(0x42bc2121fedfbcdf),
+	C64e(0xe0487070a8d848d8), C64e(0xf904f1f1fd0c040c),
+	C64e(0xc6df6363197adf7a), C64e(0xeec177772f58c158),
+	C64e(0x4575afaf309f759f), C64e(0x84634242e7a563a5),
+	C64e(0x4030202070503050), C64e(0xd11ae5e5cb2e1a2e),
+	C64e(0xe10efdfdef120e12), C64e(0x656dbfbf08b76db7),
+	C64e(0x194c818155d44cd4), C64e(0x30141818243c143c),
+	C64e(0x4c352626795f355f), C64e(0x9d2fc3c3b2712f71),
+	C64e(0x67e1bebe8638e138), C64e(0x6aa23535c8fda2fd),
+	C64e(0x0bcc8888c74fcc4f), C64e(0x5c392e2e654b394b),
+	C64e(0x3d5793936af957f9), C64e(0xaaf25555580df20d),
+	C64e(0xe382fcfc619d829d), C64e(0xf4477a7ab3c947c9),
+	C64e(0x8bacc8c827efacef), C64e(0x6fe7baba8832e732),
+	C64e(0x642b32324f7d2b7d), C64e(0xd795e6e642a495a4),
+	C64e(0x9ba0c0c03bfba0fb), C64e(0x32981919aab398b3),
+	C64e(0x27d19e9ef668d168), C64e(0x5d7fa3a322817f81),
+	C64e(0x88664444eeaa66aa), C64e(0xa87e5454d6827e82),
+	C64e(0x76ab3b3bdde6abe6), C64e(0x16830b0b959e839e),
+	C64e(0x03ca8c8cc945ca45), C64e(0x9529c7c7bc7b297b),
+	C64e(0xd6d36b6b056ed36e), C64e(0x503c28286c443c44),
+	C64e(0x5579a7a72c8b798b), C64e(0x63e2bcbc813de23d),
+	C64e(0x2c1d161631271d27), C64e(0x4176adad379a769a),
+	C64e(0xad3bdbdb964d3b4d), C64e(0xc85664649efa56fa),
+	C64e(0xe84e7474a6d24ed2), C64e(0x281e141436221e22),
+	C64e(0x3fdb9292e476db76), C64e(0x180a0c0c121e0a1e),
+	C64e(0x906c4848fcb46cb4), C64e(0x6be4b8b88f37e437),
+	C64e(0x255d9f9f78e75de7), C64e(0x616ebdbd0fb26eb2),
+	C64e(0x86ef4343692aef2a), C64e(0x93a6c4c435f1a6f1),
+	C64e(0x72a83939dae3a8e3), C64e(0x62a43131c6f7a4f7),
+	C64e(0xbd37d3d38a593759), C64e(0xff8bf2f274868b86),
+	C64e(0xb132d5d583563256), C64e(0x0d438b8b4ec543c5),
+	C64e(0xdc596e6e85eb59eb), C64e(0xafb7dada18c2b7c2),
+	C64e(0x028c01018e8f8c8f), C64e(0x7964b1b11dac64ac),
+	C64e(0x23d29c9cf16dd26d), C64e(0x92e04949723be03b),
+	C64e(0xabb4d8d81fc7b4c7), C64e(0x43faacacb915fa15),
+	C64e(0xfd07f3f3fa090709), C64e(0x8525cfcfa06f256f),
+	C64e(0x8fafcaca20eaafea), C64e(0xf38ef4f47d898e89),
+	C64e(0x8ee947476720e920), C64e(0x2018101038281828),
+	C64e(0xded56f6f0b64d564), C64e(0xfb88f0f073838883),
+	C64e(0x946f4a4afbb16fb1), C64e(0xb8725c5cca967296),
+	C64e(0x70243838546c246c), C64e(0xaef157575f08f108),
+	C64e(0xe6c773732152c752), C64e(0x3551979764f351f3),
+	C64e(0x8d23cbcbae652365), C64e(0x597ca1a125847c84),
+	C64e(0xcb9ce8e857bf9cbf), C64e(0x7c213e3e5d632163),
+	C64e(0x37dd9696ea7cdd7c), C64e(0xc2dc61611e7fdc7f),
+	C64e(0x1a860d0d9c918691), C64e(0x1e850f0f9b948594),
+	C64e(0xdb90e0e04bab90ab), C64e(0xf8427c7cbac642c6),
+	C64e(0xe2c471712657c457), C64e(0x83aacccc29e5aae5),
+	C64e(0x3bd89090e373d873), C64e(0x0c050606090f050f),
+	C64e(0xf501f7f7f4030103), C64e(0x38121c1c2a361236),
+	C64e(0x9fa3c2c23cfea3fe), C64e(0xd45f6a6a8be15fe1),
+	C64e(0x47f9aeaebe10f910), C64e(0xd2d06969026bd06b),
+	C64e(0x2e911717bfa891a8), C64e(0x2958999971e858e8),
+	C64e(0x74273a3a53692769), C64e(0x4eb92727f7d0b9d0),
+	C64e(0xa938d9d991483848), C64e(0xcd13ebebde351335),
+	C64e(0x56b32b2be5ceb3ce), C64e(0x4433222277553355),
+	C64e(0xbfbbd2d204d6bbd6), C64e(0x4970a9a939907090),
+	C64e(0x0e89070787808980), C64e(0x66a73333c1f2a7f2),
+	C64e(0x5ab62d2decc1b6c1), C64e(0x78223c3c5a662266),
+	C64e(0x2a921515b8ad92ad), C64e(0x8920c9c9a9602060),
+	C64e(0x154987875cdb49db), C64e(0x4fffaaaab01aff1a),
+	C64e(0xa0785050d8887888), C64e(0x517aa5a52b8e7a8e),
+	C64e(0x068f0303898a8f8a), C64e(0xb2f859594a13f813),
+	C64e(0x12800909929b809b), C64e(0x34171a1a23391739),
+	C64e(0xcada65651075da75), C64e(0xb531d7d784533153),
+	C64e(0x13c68484d551c651), C64e(0xbbb8d0d003d3b8d3),
+	C64e(0x1fc38282dc5ec35e), C64e(0x52b02929e2cbb0cb),
+	C64e(0xb4775a5ac3997799), C64e(0x3c111e1e2d331133),
+	C64e(0xf6cb7b7b3d46cb46), C64e(0x4bfca8a8b71ffc1f),
+	C64e(0xdad66d6d0c61d661), C64e(0x583a2c2c624e3a4e)
+};
+
+#endif
+
+static const sph_u64 T4[] = {
+	C64e(0xf497a5c6c632f4a5), C64e(0x97eb84f8f86f9784),
+	C64e(0xb0c799eeee5eb099), C64e(0x8cf78df6f67a8c8d),
+	C64e(0x17e50dffffe8170d), C64e(0xdcb7bdd6d60adcbd),
+	C64e(0xc8a7b1dede16c8b1), C64e(0xfc395491916dfc54),
+	C64e(0xf0c050606090f050), C64e(0x0504030202070503),
+	C64e(0xe087a9cece2ee0a9), C64e(0x87ac7d5656d1877d),
+	C64e(0x2bd519e7e7cc2b19), C64e(0xa67162b5b513a662),
+	C64e(0x319ae64d4d7c31e6), C64e(0xb5c39aecec59b59a),
+	C64e(0xcf05458f8f40cf45), C64e(0xbc3e9d1f1fa3bc9d),
+	C64e(0xc00940898949c040), C64e(0x92ef87fafa689287),
+	C64e(0x3fc515efefd03f15), C64e(0x267febb2b29426eb),
+	C64e(0x4007c98e8ece40c9), C64e(0x1ded0bfbfbe61d0b),
+	C64e(0x2f82ec41416e2fec), C64e(0xa97d67b3b31aa967),
+	C64e(0x1cbefd5f5f431cfd), C64e(0x258aea45456025ea),
+	C64e(0xda46bf2323f9dabf), C64e(0x02a6f753535102f7),
+	C64e(0xa1d396e4e445a196), C64e(0xed2d5b9b9b76ed5b),
+	C64e(0x5deac27575285dc2), C64e(0x24d91ce1e1c5241c),
+	C64e(0xe97aae3d3dd4e9ae), C64e(0xbe986a4c4cf2be6a),
+	C64e(0xeed85a6c6c82ee5a), C64e(0xc3fc417e7ebdc341),
+	C64e(0x06f102f5f5f30602), C64e(0xd11d4f838352d14f),
+	C64e(0xe4d05c68688ce45c), C64e(0x07a2f451515607f4),
+	C64e(0x5cb934d1d18d5c34), C64e(0x18e908f9f9e11808),
+	C64e(0xaedf93e2e24cae93), C64e(0x954d73abab3e9573),
+	C64e(0xf5c453626297f553), C64e(0x41543f2a2a6b413f),
+	C64e(0x14100c08081c140c), C64e(0xf63152959563f652),
+	C64e(0xaf8c654646e9af65), C64e(0xe2215e9d9d7fe25e),
+	C64e(0x7860283030487828), C64e(0xf86ea13737cff8a1),
+	C64e(0x11140f0a0a1b110f), C64e(0xc45eb52f2febc4b5),
+	C64e(0x1b1c090e0e151b09), C64e(0x5a483624247e5a36),
+	C64e(0xb6369b1b1badb69b), C64e(0x47a53ddfdf98473d),
+	C64e(0x6a8126cdcda76a26), C64e(0xbb9c694e4ef5bb69),
+	C64e(0x4cfecd7f7f334ccd), C64e(0xbacf9feaea50ba9f),
+	C64e(0x2d241b12123f2d1b), C64e(0xb93a9e1d1da4b99e),
+	C64e(0x9cb0745858c49c74), C64e(0x72682e343446722e),
+	C64e(0x776c2d363641772d), C64e(0xcda3b2dcdc11cdb2),
+	C64e(0x2973eeb4b49d29ee), C64e(0x16b6fb5b5b4d16fb),
+	C64e(0x0153f6a4a4a501f6), C64e(0xd7ec4d7676a1d74d),
+	C64e(0xa37561b7b714a361), C64e(0x49face7d7d3449ce),
+	C64e(0x8da47b5252df8d7b), C64e(0x42a13edddd9f423e),
+	C64e(0x93bc715e5ecd9371), C64e(0xa226971313b1a297),
+	C64e(0x0457f5a6a6a204f5), C64e(0xb86968b9b901b868),
+	C64e(0x0000000000000000), C64e(0x74992cc1c1b5742c),
+	C64e(0xa080604040e0a060), C64e(0x21dd1fe3e3c2211f),
+	C64e(0x43f2c879793a43c8), C64e(0x2c77edb6b69a2ced),
+	C64e(0xd9b3bed4d40dd9be), C64e(0xca01468d8d47ca46),
+	C64e(0x70ced967671770d9), C64e(0xdde44b7272afdd4b),
+	C64e(0x7933de9494ed79de), C64e(0x672bd49898ff67d4),
+	C64e(0x237be8b0b09323e8), C64e(0xde114a85855bde4a),
+	C64e(0xbd6d6bbbbb06bd6b), C64e(0x7e912ac5c5bb7e2a),
+	C64e(0x349ee54f4f7b34e5), C64e(0x3ac116ededd73a16),
+	C64e(0x5417c58686d254c5), C64e(0x622fd79a9af862d7),
+	C64e(0xffcc55666699ff55), C64e(0xa722941111b6a794),
+	C64e(0x4a0fcf8a8ac04acf), C64e(0x30c910e9e9d93010),
+	C64e(0x0a080604040e0a06), C64e(0x98e781fefe669881),
+	C64e(0x0b5bf0a0a0ab0bf0), C64e(0xccf0447878b4cc44),
+	C64e(0xd54aba2525f0d5ba), C64e(0x3e96e34b4b753ee3),
+	C64e(0x0e5ff3a2a2ac0ef3), C64e(0x19bafe5d5d4419fe),
+	C64e(0x5b1bc08080db5bc0), C64e(0x850a8a050580858a),
+	C64e(0xec7ead3f3fd3ecad), C64e(0xdf42bc2121fedfbc),
+	C64e(0xd8e0487070a8d848), C64e(0x0cf904f1f1fd0c04),
+	C64e(0x7ac6df6363197adf), C64e(0x58eec177772f58c1),
+	C64e(0x9f4575afaf309f75), C64e(0xa584634242e7a563),
+	C64e(0x5040302020705030), C64e(0x2ed11ae5e5cb2e1a),
+	C64e(0x12e10efdfdef120e), C64e(0xb7656dbfbf08b76d),
+	C64e(0xd4194c818155d44c), C64e(0x3c30141818243c14),
+	C64e(0x5f4c352626795f35), C64e(0x719d2fc3c3b2712f),
+	C64e(0x3867e1bebe8638e1), C64e(0xfd6aa23535c8fda2),
+	C64e(0x4f0bcc8888c74fcc), C64e(0x4b5c392e2e654b39),
+	C64e(0xf93d5793936af957), C64e(0x0daaf25555580df2),
+	C64e(0x9de382fcfc619d82), C64e(0xc9f4477a7ab3c947),
+	C64e(0xef8bacc8c827efac), C64e(0x326fe7baba8832e7),
+	C64e(0x7d642b32324f7d2b), C64e(0xa4d795e6e642a495),
+	C64e(0xfb9ba0c0c03bfba0), C64e(0xb332981919aab398),
+	C64e(0x6827d19e9ef668d1), C64e(0x815d7fa3a322817f),
+	C64e(0xaa88664444eeaa66), C64e(0x82a87e5454d6827e),
+	C64e(0xe676ab3b3bdde6ab), C64e(0x9e16830b0b959e83),
+	C64e(0x4503ca8c8cc945ca), C64e(0x7b9529c7c7bc7b29),
+	C64e(0x6ed6d36b6b056ed3), C64e(0x44503c28286c443c),
+	C64e(0x8b5579a7a72c8b79), C64e(0x3d63e2bcbc813de2),
+	C64e(0x272c1d161631271d), C64e(0x9a4176adad379a76),
+	C64e(0x4dad3bdbdb964d3b), C64e(0xfac85664649efa56),
+	C64e(0xd2e84e7474a6d24e), C64e(0x22281e141436221e),
+	C64e(0x763fdb9292e476db), C64e(0x1e180a0c0c121e0a),
+	C64e(0xb4906c4848fcb46c), C64e(0x376be4b8b88f37e4),
+	C64e(0xe7255d9f9f78e75d), C64e(0xb2616ebdbd0fb26e),
+	C64e(0x2a86ef4343692aef), C64e(0xf193a6c4c435f1a6),
+	C64e(0xe372a83939dae3a8), C64e(0xf762a43131c6f7a4),
+	C64e(0x59bd37d3d38a5937), C64e(0x86ff8bf2f274868b),
+	C64e(0x56b132d5d5835632), C64e(0xc50d438b8b4ec543),
+	C64e(0xebdc596e6e85eb59), C64e(0xc2afb7dada18c2b7),
+	C64e(0x8f028c01018e8f8c), C64e(0xac7964b1b11dac64),
+	C64e(0x6d23d29c9cf16dd2), C64e(0x3b92e04949723be0),
+	C64e(0xc7abb4d8d81fc7b4), C64e(0x1543faacacb915fa),
+	C64e(0x09fd07f3f3fa0907), C64e(0x6f8525cfcfa06f25),
+	C64e(0xea8fafcaca20eaaf), C64e(0x89f38ef4f47d898e),
+	C64e(0x208ee947476720e9), C64e(0x2820181010382818),
+	C64e(0x64ded56f6f0b64d5), C64e(0x83fb88f0f0738388),
+	C64e(0xb1946f4a4afbb16f), C64e(0x96b8725c5cca9672),
+	C64e(0x6c70243838546c24), C64e(0x08aef157575f08f1),
+	C64e(0x52e6c773732152c7), C64e(0xf33551979764f351),
+	C64e(0x658d23cbcbae6523), C64e(0x84597ca1a125847c),
+	C64e(0xbfcb9ce8e857bf9c), C64e(0x637c213e3e5d6321),
+	C64e(0x7c37dd9696ea7cdd), C64e(0x7fc2dc61611e7fdc),
+	C64e(0x911a860d0d9c9186), C64e(0x941e850f0f9b9485),
+	C64e(0xabdb90e0e04bab90), C64e(0xc6f8427c7cbac642),
+	C64e(0x57e2c471712657c4), C64e(0xe583aacccc29e5aa),
+	C64e(0x733bd89090e373d8), C64e(0x0f0c050606090f05),
+	C64e(0x03f501f7f7f40301), C64e(0x3638121c1c2a3612),
+	C64e(0xfe9fa3c2c23cfea3), C64e(0xe1d45f6a6a8be15f),
+	C64e(0x1047f9aeaebe10f9), C64e(0x6bd2d06969026bd0),
+	C64e(0xa82e911717bfa891), C64e(0xe82958999971e858),
+	C64e(0x6974273a3a536927), C64e(0xd04eb92727f7d0b9),
+	C64e(0x48a938d9d9914838), C64e(0x35cd13ebebde3513),
+	C64e(0xce56b32b2be5ceb3), C64e(0x5544332222775533),
+	C64e(0xd6bfbbd2d204d6bb), C64e(0x904970a9a9399070),
+	C64e(0x800e890707878089), C64e(0xf266a73333c1f2a7),
+	C64e(0xc15ab62d2decc1b6), C64e(0x6678223c3c5a6622),
+	C64e(0xad2a921515b8ad92), C64e(0x608920c9c9a96020),
+	C64e(0xdb154987875cdb49), C64e(0x1a4fffaaaab01aff),
+	C64e(0x88a0785050d88878), C64e(0x8e517aa5a52b8e7a),
+	C64e(0x8a068f0303898a8f), C64e(0x13b2f859594a13f8),
+	C64e(0x9b12800909929b80), C64e(0x3934171a1a233917),
+	C64e(0x75cada65651075da), C64e(0x53b531d7d7845331),
+	C64e(0x5113c68484d551c6), C64e(0xd3bbb8d0d003d3b8),
+	C64e(0x5e1fc38282dc5ec3), C64e(0xcb52b02929e2cbb0),
+	C64e(0x99b4775a5ac39977), C64e(0x333c111e1e2d3311),
+	C64e(0x46f6cb7b7b3d46cb), C64e(0x1f4bfca8a8b71ffc),
+	C64e(0x61dad66d6d0c61d6), C64e(0x4e583a2c2c624e3a)
+};
+
+#if !SPH_SMALL_FOOTPRINT_GROESTL
+
+static const sph_u64 T5[] = {
+	C64e(0xa5f497a5c6c632f4), C64e(0x8497eb84f8f86f97),
+	C64e(0x99b0c799eeee5eb0), C64e(0x8d8cf78df6f67a8c),
+	C64e(0x0d17e50dffffe817), C64e(0xbddcb7bdd6d60adc),
+	C64e(0xb1c8a7b1dede16c8), C64e(0x54fc395491916dfc),
+	C64e(0x50f0c050606090f0), C64e(0x0305040302020705),
+	C64e(0xa9e087a9cece2ee0), C64e(0x7d87ac7d5656d187),
+	C64e(0x192bd519e7e7cc2b), C64e(0x62a67162b5b513a6),
+	C64e(0xe6319ae64d4d7c31), C64e(0x9ab5c39aecec59b5),
+	C64e(0x45cf05458f8f40cf), C64e(0x9dbc3e9d1f1fa3bc),
+	C64e(0x40c00940898949c0), C64e(0x8792ef87fafa6892),
+	C64e(0x153fc515efefd03f), C64e(0xeb267febb2b29426),
+	C64e(0xc94007c98e8ece40), C64e(0x0b1ded0bfbfbe61d),
+	C64e(0xec2f82ec41416e2f), C64e(0x67a97d67b3b31aa9),
+	C64e(0xfd1cbefd5f5f431c), C64e(0xea258aea45456025),
+	C64e(0xbfda46bf2323f9da), C64e(0xf702a6f753535102),
+	C64e(0x96a1d396e4e445a1), C64e(0x5bed2d5b9b9b76ed),
+	C64e(0xc25deac27575285d), C64e(0x1c24d91ce1e1c524),
+	C64e(0xaee97aae3d3dd4e9), C64e(0x6abe986a4c4cf2be),
+	C64e(0x5aeed85a6c6c82ee), C64e(0x41c3fc417e7ebdc3),
+	C64e(0x0206f102f5f5f306), C64e(0x4fd11d4f838352d1),
+	C64e(0x5ce4d05c68688ce4), C64e(0xf407a2f451515607),
+	C64e(0x345cb934d1d18d5c), C64e(0x0818e908f9f9e118),
+	C64e(0x93aedf93e2e24cae), C64e(0x73954d73abab3e95),
+	C64e(0x53f5c453626297f5), C64e(0x3f41543f2a2a6b41),
+	C64e(0x0c14100c08081c14), C64e(0x52f63152959563f6),
+	C64e(0x65af8c654646e9af), C64e(0x5ee2215e9d9d7fe2),
+	C64e(0x2878602830304878), C64e(0xa1f86ea13737cff8),
+	C64e(0x0f11140f0a0a1b11), C64e(0xb5c45eb52f2febc4),
+	C64e(0x091b1c090e0e151b), C64e(0x365a483624247e5a),
+	C64e(0x9bb6369b1b1badb6), C64e(0x3d47a53ddfdf9847),
+	C64e(0x266a8126cdcda76a), C64e(0x69bb9c694e4ef5bb),
+	C64e(0xcd4cfecd7f7f334c), C64e(0x9fbacf9feaea50ba),
+	C64e(0x1b2d241b12123f2d), C64e(0x9eb93a9e1d1da4b9),
+	C64e(0x749cb0745858c49c), C64e(0x2e72682e34344672),
+	C64e(0x2d776c2d36364177), C64e(0xb2cda3b2dcdc11cd),
+	C64e(0xee2973eeb4b49d29), C64e(0xfb16b6fb5b5b4d16),
+	C64e(0xf60153f6a4a4a501), C64e(0x4dd7ec4d7676a1d7),
+	C64e(0x61a37561b7b714a3), C64e(0xce49face7d7d3449),
+	C64e(0x7b8da47b5252df8d), C64e(0x3e42a13edddd9f42),
+	C64e(0x7193bc715e5ecd93), C64e(0x97a226971313b1a2),
+	C64e(0xf50457f5a6a6a204), C64e(0x68b86968b9b901b8),
+	C64e(0x0000000000000000), C64e(0x2c74992cc1c1b574),
+	C64e(0x60a080604040e0a0), C64e(0x1f21dd1fe3e3c221),
+	C64e(0xc843f2c879793a43), C64e(0xed2c77edb6b69a2c),
+	C64e(0xbed9b3bed4d40dd9), C64e(0x46ca01468d8d47ca),
+	C64e(0xd970ced967671770), C64e(0x4bdde44b7272afdd),
+	C64e(0xde7933de9494ed79), C64e(0xd4672bd49898ff67),
+	C64e(0xe8237be8b0b09323), C64e(0x4ade114a85855bde),
+	C64e(0x6bbd6d6bbbbb06bd), C64e(0x2a7e912ac5c5bb7e),
+	C64e(0xe5349ee54f4f7b34), C64e(0x163ac116ededd73a),
+	C64e(0xc55417c58686d254), C64e(0xd7622fd79a9af862),
+	C64e(0x55ffcc55666699ff), C64e(0x94a722941111b6a7),
+	C64e(0xcf4a0fcf8a8ac04a), C64e(0x1030c910e9e9d930),
+	C64e(0x060a080604040e0a), C64e(0x8198e781fefe6698),
+	C64e(0xf00b5bf0a0a0ab0b), C64e(0x44ccf0447878b4cc),
+	C64e(0xbad54aba2525f0d5), C64e(0xe33e96e34b4b753e),
+	C64e(0xf30e5ff3a2a2ac0e), C64e(0xfe19bafe5d5d4419),
+	C64e(0xc05b1bc08080db5b), C64e(0x8a850a8a05058085),
+	C64e(0xadec7ead3f3fd3ec), C64e(0xbcdf42bc2121fedf),
+	C64e(0x48d8e0487070a8d8), C64e(0x040cf904f1f1fd0c),
+	C64e(0xdf7ac6df6363197a), C64e(0xc158eec177772f58),
+	C64e(0x759f4575afaf309f), C64e(0x63a584634242e7a5),
+	C64e(0x3050403020207050), C64e(0x1a2ed11ae5e5cb2e),
+	C64e(0x0e12e10efdfdef12), C64e(0x6db7656dbfbf08b7),
+	C64e(0x4cd4194c818155d4), C64e(0x143c30141818243c),
+	C64e(0x355f4c352626795f), C64e(0x2f719d2fc3c3b271),
+	C64e(0xe13867e1bebe8638), C64e(0xa2fd6aa23535c8fd),
+	C64e(0xcc4f0bcc8888c74f), C64e(0x394b5c392e2e654b),
+	C64e(0x57f93d5793936af9), C64e(0xf20daaf25555580d),
+	C64e(0x829de382fcfc619d), C64e(0x47c9f4477a7ab3c9),
+	C64e(0xacef8bacc8c827ef), C64e(0xe7326fe7baba8832),
+	C64e(0x2b7d642b32324f7d), C64e(0x95a4d795e6e642a4),
+	C64e(0xa0fb9ba0c0c03bfb), C64e(0x98b332981919aab3),
+	C64e(0xd16827d19e9ef668), C64e(0x7f815d7fa3a32281),
+	C64e(0x66aa88664444eeaa), C64e(0x7e82a87e5454d682),
+	C64e(0xabe676ab3b3bdde6), C64e(0x839e16830b0b959e),
+	C64e(0xca4503ca8c8cc945), C64e(0x297b9529c7c7bc7b),
+	C64e(0xd36ed6d36b6b056e), C64e(0x3c44503c28286c44),
+	C64e(0x798b5579a7a72c8b), C64e(0xe23d63e2bcbc813d),
+	C64e(0x1d272c1d16163127), C64e(0x769a4176adad379a),
+	C64e(0x3b4dad3bdbdb964d), C64e(0x56fac85664649efa),
+	C64e(0x4ed2e84e7474a6d2), C64e(0x1e22281e14143622),
+	C64e(0xdb763fdb9292e476), C64e(0x0a1e180a0c0c121e),
+	C64e(0x6cb4906c4848fcb4), C64e(0xe4376be4b8b88f37),
+	C64e(0x5de7255d9f9f78e7), C64e(0x6eb2616ebdbd0fb2),
+	C64e(0xef2a86ef4343692a), C64e(0xa6f193a6c4c435f1),
+	C64e(0xa8e372a83939dae3), C64e(0xa4f762a43131c6f7),
+	C64e(0x3759bd37d3d38a59), C64e(0x8b86ff8bf2f27486),
+	C64e(0x3256b132d5d58356), C64e(0x43c50d438b8b4ec5),
+	C64e(0x59ebdc596e6e85eb), C64e(0xb7c2afb7dada18c2),
+	C64e(0x8c8f028c01018e8f), C64e(0x64ac7964b1b11dac),
+	C64e(0xd26d23d29c9cf16d), C64e(0xe03b92e04949723b),
+	C64e(0xb4c7abb4d8d81fc7), C64e(0xfa1543faacacb915),
+	C64e(0x0709fd07f3f3fa09), C64e(0x256f8525cfcfa06f),
+	C64e(0xafea8fafcaca20ea), C64e(0x8e89f38ef4f47d89),
+	C64e(0xe9208ee947476720), C64e(0x1828201810103828),
+	C64e(0xd564ded56f6f0b64), C64e(0x8883fb88f0f07383),
+	C64e(0x6fb1946f4a4afbb1), C64e(0x7296b8725c5cca96),
+	C64e(0x246c70243838546c), C64e(0xf108aef157575f08),
+	C64e(0xc752e6c773732152), C64e(0x51f33551979764f3),
+	C64e(0x23658d23cbcbae65), C64e(0x7c84597ca1a12584),
+	C64e(0x9cbfcb9ce8e857bf), C64e(0x21637c213e3e5d63),
+	C64e(0xdd7c37dd9696ea7c), C64e(0xdc7fc2dc61611e7f),
+	C64e(0x86911a860d0d9c91), C64e(0x85941e850f0f9b94),
+	C64e(0x90abdb90e0e04bab), C64e(0x42c6f8427c7cbac6),
+	C64e(0xc457e2c471712657), C64e(0xaae583aacccc29e5),
+	C64e(0xd8733bd89090e373), C64e(0x050f0c050606090f),
+	C64e(0x0103f501f7f7f403), C64e(0x123638121c1c2a36),
+	C64e(0xa3fe9fa3c2c23cfe), C64e(0x5fe1d45f6a6a8be1),
+	C64e(0xf91047f9aeaebe10), C64e(0xd06bd2d06969026b),
+	C64e(0x91a82e911717bfa8), C64e(0x58e82958999971e8),
+	C64e(0x276974273a3a5369), C64e(0xb9d04eb92727f7d0),
+	C64e(0x3848a938d9d99148), C64e(0x1335cd13ebebde35),
+	C64e(0xb3ce56b32b2be5ce), C64e(0x3355443322227755),
+	C64e(0xbbd6bfbbd2d204d6), C64e(0x70904970a9a93990),
+	C64e(0x89800e8907078780), C64e(0xa7f266a73333c1f2),
+	C64e(0xb6c15ab62d2decc1), C64e(0x226678223c3c5a66),
+	C64e(0x92ad2a921515b8ad), C64e(0x20608920c9c9a960),
+	C64e(0x49db154987875cdb), C64e(0xff1a4fffaaaab01a),
+	C64e(0x7888a0785050d888), C64e(0x7a8e517aa5a52b8e),
+	C64e(0x8f8a068f0303898a), C64e(0xf813b2f859594a13),
+	C64e(0x809b12800909929b), C64e(0x173934171a1a2339),
+	C64e(0xda75cada65651075), C64e(0x3153b531d7d78453),
+	C64e(0xc65113c68484d551), C64e(0xb8d3bbb8d0d003d3),
+	C64e(0xc35e1fc38282dc5e), C64e(0xb0cb52b02929e2cb),
+	C64e(0x7799b4775a5ac399), C64e(0x11333c111e1e2d33),
+	C64e(0xcb46f6cb7b7b3d46), C64e(0xfc1f4bfca8a8b71f),
+	C64e(0xd661dad66d6d0c61), C64e(0x3a4e583a2c2c624e)
+};
+
+static const sph_u64 T6[] = {
+	C64e(0xf4a5f497a5c6c632), C64e(0x978497eb84f8f86f),
+	C64e(0xb099b0c799eeee5e), C64e(0x8c8d8cf78df6f67a),
+	C64e(0x170d17e50dffffe8), C64e(0xdcbddcb7bdd6d60a),
+	C64e(0xc8b1c8a7b1dede16), C64e(0xfc54fc395491916d),
+	C64e(0xf050f0c050606090), C64e(0x0503050403020207),
+	C64e(0xe0a9e087a9cece2e), C64e(0x877d87ac7d5656d1),
+	C64e(0x2b192bd519e7e7cc), C64e(0xa662a67162b5b513),
+	C64e(0x31e6319ae64d4d7c), C64e(0xb59ab5c39aecec59),
+	C64e(0xcf45cf05458f8f40), C64e(0xbc9dbc3e9d1f1fa3),
+	C64e(0xc040c00940898949), C64e(0x928792ef87fafa68),
+	C64e(0x3f153fc515efefd0), C64e(0x26eb267febb2b294),
+	C64e(0x40c94007c98e8ece), C64e(0x1d0b1ded0bfbfbe6),
+	C64e(0x2fec2f82ec41416e), C64e(0xa967a97d67b3b31a),
+	C64e(0x1cfd1cbefd5f5f43), C64e(0x25ea258aea454560),
+	C64e(0xdabfda46bf2323f9), C64e(0x02f702a6f7535351),
+	C64e(0xa196a1d396e4e445), C64e(0xed5bed2d5b9b9b76),
+	C64e(0x5dc25deac2757528), C64e(0x241c24d91ce1e1c5),
+	C64e(0xe9aee97aae3d3dd4), C64e(0xbe6abe986a4c4cf2),
+	C64e(0xee5aeed85a6c6c82), C64e(0xc341c3fc417e7ebd),
+	C64e(0x060206f102f5f5f3), C64e(0xd14fd11d4f838352),
+	C64e(0xe45ce4d05c68688c), C64e(0x07f407a2f4515156),
+	C64e(0x5c345cb934d1d18d), C64e(0x180818e908f9f9e1),
+	C64e(0xae93aedf93e2e24c), C64e(0x9573954d73abab3e),
+	C64e(0xf553f5c453626297), C64e(0x413f41543f2a2a6b),
+	C64e(0x140c14100c08081c), C64e(0xf652f63152959563),
+	C64e(0xaf65af8c654646e9), C64e(0xe25ee2215e9d9d7f),
+	C64e(0x7828786028303048), C64e(0xf8a1f86ea13737cf),
+	C64e(0x110f11140f0a0a1b), C64e(0xc4b5c45eb52f2feb),
+	C64e(0x1b091b1c090e0e15), C64e(0x5a365a483624247e),
+	C64e(0xb69bb6369b1b1bad), C64e(0x473d47a53ddfdf98),
+	C64e(0x6a266a8126cdcda7), C64e(0xbb69bb9c694e4ef5),
+	C64e(0x4ccd4cfecd7f7f33), C64e(0xba9fbacf9feaea50),
+	C64e(0x2d1b2d241b12123f), C64e(0xb99eb93a9e1d1da4),
+	C64e(0x9c749cb0745858c4), C64e(0x722e72682e343446),
+	C64e(0x772d776c2d363641), C64e(0xcdb2cda3b2dcdc11),
+	C64e(0x29ee2973eeb4b49d), C64e(0x16fb16b6fb5b5b4d),
+	C64e(0x01f60153f6a4a4a5), C64e(0xd74dd7ec4d7676a1),
+	C64e(0xa361a37561b7b714), C64e(0x49ce49face7d7d34),
+	C64e(0x8d7b8da47b5252df), C64e(0x423e42a13edddd9f),
+	C64e(0x937193bc715e5ecd), C64e(0xa297a226971313b1),
+	C64e(0x04f50457f5a6a6a2), C64e(0xb868b86968b9b901),
+	C64e(0x0000000000000000), C64e(0x742c74992cc1c1b5),
+	C64e(0xa060a080604040e0), C64e(0x211f21dd1fe3e3c2),
+	C64e(0x43c843f2c879793a), C64e(0x2ced2c77edb6b69a),
+	C64e(0xd9bed9b3bed4d40d), C64e(0xca46ca01468d8d47),
+	C64e(0x70d970ced9676717), C64e(0xdd4bdde44b7272af),
+	C64e(0x79de7933de9494ed), C64e(0x67d4672bd49898ff),
+	C64e(0x23e8237be8b0b093), C64e(0xde4ade114a85855b),
+	C64e(0xbd6bbd6d6bbbbb06), C64e(0x7e2a7e912ac5c5bb),
+	C64e(0x34e5349ee54f4f7b), C64e(0x3a163ac116ededd7),
+	C64e(0x54c55417c58686d2), C64e(0x62d7622fd79a9af8),
+	C64e(0xff55ffcc55666699), C64e(0xa794a722941111b6),
+	C64e(0x4acf4a0fcf8a8ac0), C64e(0x301030c910e9e9d9),
+	C64e(0x0a060a080604040e), C64e(0x988198e781fefe66),
+	C64e(0x0bf00b5bf0a0a0ab), C64e(0xcc44ccf0447878b4),
+	C64e(0xd5bad54aba2525f0), C64e(0x3ee33e96e34b4b75),
+	C64e(0x0ef30e5ff3a2a2ac), C64e(0x19fe19bafe5d5d44),
+	C64e(0x5bc05b1bc08080db), C64e(0x858a850a8a050580),
+	C64e(0xecadec7ead3f3fd3), C64e(0xdfbcdf42bc2121fe),
+	C64e(0xd848d8e0487070a8), C64e(0x0c040cf904f1f1fd),
+	C64e(0x7adf7ac6df636319), C64e(0x58c158eec177772f),
+	C64e(0x9f759f4575afaf30), C64e(0xa563a584634242e7),
+	C64e(0x5030504030202070), C64e(0x2e1a2ed11ae5e5cb),
+	C64e(0x120e12e10efdfdef), C64e(0xb76db7656dbfbf08),
+	C64e(0xd44cd4194c818155), C64e(0x3c143c3014181824),
+	C64e(0x5f355f4c35262679), C64e(0x712f719d2fc3c3b2),
+	C64e(0x38e13867e1bebe86), C64e(0xfda2fd6aa23535c8),
+	C64e(0x4fcc4f0bcc8888c7), C64e(0x4b394b5c392e2e65),
+	C64e(0xf957f93d5793936a), C64e(0x0df20daaf2555558),
+	C64e(0x9d829de382fcfc61), C64e(0xc947c9f4477a7ab3),
+	C64e(0xefacef8bacc8c827), C64e(0x32e7326fe7baba88),
+	C64e(0x7d2b7d642b32324f), C64e(0xa495a4d795e6e642),
+	C64e(0xfba0fb9ba0c0c03b), C64e(0xb398b332981919aa),
+	C64e(0x68d16827d19e9ef6), C64e(0x817f815d7fa3a322),
+	C64e(0xaa66aa88664444ee), C64e(0x827e82a87e5454d6),
+	C64e(0xe6abe676ab3b3bdd), C64e(0x9e839e16830b0b95),
+	C64e(0x45ca4503ca8c8cc9), C64e(0x7b297b9529c7c7bc),
+	C64e(0x6ed36ed6d36b6b05), C64e(0x443c44503c28286c),
+	C64e(0x8b798b5579a7a72c), C64e(0x3de23d63e2bcbc81),
+	C64e(0x271d272c1d161631), C64e(0x9a769a4176adad37),
+	C64e(0x4d3b4dad3bdbdb96), C64e(0xfa56fac85664649e),
+	C64e(0xd24ed2e84e7474a6), C64e(0x221e22281e141436),
+	C64e(0x76db763fdb9292e4), C64e(0x1e0a1e180a0c0c12),
+	C64e(0xb46cb4906c4848fc), C64e(0x37e4376be4b8b88f),
+	C64e(0xe75de7255d9f9f78), C64e(0xb26eb2616ebdbd0f),
+	C64e(0x2aef2a86ef434369), C64e(0xf1a6f193a6c4c435),
+	C64e(0xe3a8e372a83939da), C64e(0xf7a4f762a43131c6),
+	C64e(0x593759bd37d3d38a), C64e(0x868b86ff8bf2f274),
+	C64e(0x563256b132d5d583), C64e(0xc543c50d438b8b4e),
+	C64e(0xeb59ebdc596e6e85), C64e(0xc2b7c2afb7dada18),
+	C64e(0x8f8c8f028c01018e), C64e(0xac64ac7964b1b11d),
+	C64e(0x6dd26d23d29c9cf1), C64e(0x3be03b92e0494972),
+	C64e(0xc7b4c7abb4d8d81f), C64e(0x15fa1543faacacb9),
+	C64e(0x090709fd07f3f3fa), C64e(0x6f256f8525cfcfa0),
+	C64e(0xeaafea8fafcaca20), C64e(0x898e89f38ef4f47d),
+	C64e(0x20e9208ee9474767), C64e(0x2818282018101038),
+	C64e(0x64d564ded56f6f0b), C64e(0x838883fb88f0f073),
+	C64e(0xb16fb1946f4a4afb), C64e(0x967296b8725c5cca),
+	C64e(0x6c246c7024383854), C64e(0x08f108aef157575f),
+	C64e(0x52c752e6c7737321), C64e(0xf351f33551979764),
+	C64e(0x6523658d23cbcbae), C64e(0x847c84597ca1a125),
+	C64e(0xbf9cbfcb9ce8e857), C64e(0x6321637c213e3e5d),
+	C64e(0x7cdd7c37dd9696ea), C64e(0x7fdc7fc2dc61611e),
+	C64e(0x9186911a860d0d9c), C64e(0x9485941e850f0f9b),
+	C64e(0xab90abdb90e0e04b), C64e(0xc642c6f8427c7cba),
+	C64e(0x57c457e2c4717126), C64e(0xe5aae583aacccc29),
+	C64e(0x73d8733bd89090e3), C64e(0x0f050f0c05060609),
+	C64e(0x030103f501f7f7f4), C64e(0x36123638121c1c2a),
+	C64e(0xfea3fe9fa3c2c23c), C64e(0xe15fe1d45f6a6a8b),
+	C64e(0x10f91047f9aeaebe), C64e(0x6bd06bd2d0696902),
+	C64e(0xa891a82e911717bf), C64e(0xe858e82958999971),
+	C64e(0x69276974273a3a53), C64e(0xd0b9d04eb92727f7),
+	C64e(0x483848a938d9d991), C64e(0x351335cd13ebebde),
+	C64e(0xceb3ce56b32b2be5), C64e(0x5533554433222277),
+	C64e(0xd6bbd6bfbbd2d204), C64e(0x9070904970a9a939),
+	C64e(0x8089800e89070787), C64e(0xf2a7f266a73333c1),
+	C64e(0xc1b6c15ab62d2dec), C64e(0x66226678223c3c5a),
+	C64e(0xad92ad2a921515b8), C64e(0x6020608920c9c9a9),
+	C64e(0xdb49db154987875c), C64e(0x1aff1a4fffaaaab0),
+	C64e(0x887888a0785050d8), C64e(0x8e7a8e517aa5a52b),
+	C64e(0x8a8f8a068f030389), C64e(0x13f813b2f859594a),
+	C64e(0x9b809b1280090992), C64e(0x39173934171a1a23),
+	C64e(0x75da75cada656510), C64e(0x533153b531d7d784),
+	C64e(0x51c65113c68484d5), C64e(0xd3b8d3bbb8d0d003),
+	C64e(0x5ec35e1fc38282dc), C64e(0xcbb0cb52b02929e2),
+	C64e(0x997799b4775a5ac3), C64e(0x3311333c111e1e2d),
+	C64e(0x46cb46f6cb7b7b3d), C64e(0x1ffc1f4bfca8a8b7),
+	C64e(0x61d661dad66d6d0c), C64e(0x4e3a4e583a2c2c62)
+};
+
+static const sph_u64 T7[] = {
+	C64e(0x32f4a5f497a5c6c6), C64e(0x6f978497eb84f8f8),
+	C64e(0x5eb099b0c799eeee), C64e(0x7a8c8d8cf78df6f6),
+	C64e(0xe8170d17e50dffff), C64e(0x0adcbddcb7bdd6d6),
+	C64e(0x16c8b1c8a7b1dede), C64e(0x6dfc54fc39549191),
+	C64e(0x90f050f0c0506060), C64e(0x0705030504030202),
+	C64e(0x2ee0a9e087a9cece), C64e(0xd1877d87ac7d5656),
+	C64e(0xcc2b192bd519e7e7), C64e(0x13a662a67162b5b5),
+	C64e(0x7c31e6319ae64d4d), C64e(0x59b59ab5c39aecec),
+	C64e(0x40cf45cf05458f8f), C64e(0xa3bc9dbc3e9d1f1f),
+	C64e(0x49c040c009408989), C64e(0x68928792ef87fafa),
+	C64e(0xd03f153fc515efef), C64e(0x9426eb267febb2b2),
+	C64e(0xce40c94007c98e8e), C64e(0xe61d0b1ded0bfbfb),
+	C64e(0x6e2fec2f82ec4141), C64e(0x1aa967a97d67b3b3),
+	C64e(0x431cfd1cbefd5f5f), C64e(0x6025ea258aea4545),
+	C64e(0xf9dabfda46bf2323), C64e(0x5102f702a6f75353),
+	C64e(0x45a196a1d396e4e4), C64e(0x76ed5bed2d5b9b9b),
+	C64e(0x285dc25deac27575), C64e(0xc5241c24d91ce1e1),
+	C64e(0xd4e9aee97aae3d3d), C64e(0xf2be6abe986a4c4c),
+	C64e(0x82ee5aeed85a6c6c), C64e(0xbdc341c3fc417e7e),
+	C64e(0xf3060206f102f5f5), C64e(0x52d14fd11d4f8383),
+	C64e(0x8ce45ce4d05c6868), C64e(0x5607f407a2f45151),
+	C64e(0x8d5c345cb934d1d1), C64e(0xe1180818e908f9f9),
+	C64e(0x4cae93aedf93e2e2), C64e(0x3e9573954d73abab),
+	C64e(0x97f553f5c4536262), C64e(0x6b413f41543f2a2a),
+	C64e(0x1c140c14100c0808), C64e(0x63f652f631529595),
+	C64e(0xe9af65af8c654646), C64e(0x7fe25ee2215e9d9d),
+	C64e(0x4878287860283030), C64e(0xcff8a1f86ea13737),
+	C64e(0x1b110f11140f0a0a), C64e(0xebc4b5c45eb52f2f),
+	C64e(0x151b091b1c090e0e), C64e(0x7e5a365a48362424),
+	C64e(0xadb69bb6369b1b1b), C64e(0x98473d47a53ddfdf),
+	C64e(0xa76a266a8126cdcd), C64e(0xf5bb69bb9c694e4e),
+	C64e(0x334ccd4cfecd7f7f), C64e(0x50ba9fbacf9feaea),
+	C64e(0x3f2d1b2d241b1212), C64e(0xa4b99eb93a9e1d1d),
+	C64e(0xc49c749cb0745858), C64e(0x46722e72682e3434),
+	C64e(0x41772d776c2d3636), C64e(0x11cdb2cda3b2dcdc),
+	C64e(0x9d29ee2973eeb4b4), C64e(0x4d16fb16b6fb5b5b),
+	C64e(0xa501f60153f6a4a4), C64e(0xa1d74dd7ec4d7676),
+	C64e(0x14a361a37561b7b7), C64e(0x3449ce49face7d7d),
+	C64e(0xdf8d7b8da47b5252), C64e(0x9f423e42a13edddd),
+	C64e(0xcd937193bc715e5e), C64e(0xb1a297a226971313),
+	C64e(0xa204f50457f5a6a6), C64e(0x01b868b86968b9b9),
+	C64e(0x0000000000000000), C64e(0xb5742c74992cc1c1),
+	C64e(0xe0a060a080604040), C64e(0xc2211f21dd1fe3e3),
+	C64e(0x3a43c843f2c87979), C64e(0x9a2ced2c77edb6b6),
+	C64e(0x0dd9bed9b3bed4d4), C64e(0x47ca46ca01468d8d),
+	C64e(0x1770d970ced96767), C64e(0xafdd4bdde44b7272),
+	C64e(0xed79de7933de9494), C64e(0xff67d4672bd49898),
+	C64e(0x9323e8237be8b0b0), C64e(0x5bde4ade114a8585),
+	C64e(0x06bd6bbd6d6bbbbb), C64e(0xbb7e2a7e912ac5c5),
+	C64e(0x7b34e5349ee54f4f), C64e(0xd73a163ac116eded),
+	C64e(0xd254c55417c58686), C64e(0xf862d7622fd79a9a),
+	C64e(0x99ff55ffcc556666), C64e(0xb6a794a722941111),
+	C64e(0xc04acf4a0fcf8a8a), C64e(0xd9301030c910e9e9),
+	C64e(0x0e0a060a08060404), C64e(0x66988198e781fefe),
+	C64e(0xab0bf00b5bf0a0a0), C64e(0xb4cc44ccf0447878),
+	C64e(0xf0d5bad54aba2525), C64e(0x753ee33e96e34b4b),
+	C64e(0xac0ef30e5ff3a2a2), C64e(0x4419fe19bafe5d5d),
+	C64e(0xdb5bc05b1bc08080), C64e(0x80858a850a8a0505),
+	C64e(0xd3ecadec7ead3f3f), C64e(0xfedfbcdf42bc2121),
+	C64e(0xa8d848d8e0487070), C64e(0xfd0c040cf904f1f1),
+	C64e(0x197adf7ac6df6363), C64e(0x2f58c158eec17777),
+	C64e(0x309f759f4575afaf), C64e(0xe7a563a584634242),
+	C64e(0x7050305040302020), C64e(0xcb2e1a2ed11ae5e5),
+	C64e(0xef120e12e10efdfd), C64e(0x08b76db7656dbfbf),
+	C64e(0x55d44cd4194c8181), C64e(0x243c143c30141818),
+	C64e(0x795f355f4c352626), C64e(0xb2712f719d2fc3c3),
+	C64e(0x8638e13867e1bebe), C64e(0xc8fda2fd6aa23535),
+	C64e(0xc74fcc4f0bcc8888), C64e(0x654b394b5c392e2e),
+	C64e(0x6af957f93d579393), C64e(0x580df20daaf25555),
+	C64e(0x619d829de382fcfc), C64e(0xb3c947c9f4477a7a),
+	C64e(0x27efacef8bacc8c8), C64e(0x8832e7326fe7baba),
+	C64e(0x4f7d2b7d642b3232), C64e(0x42a495a4d795e6e6),
+	C64e(0x3bfba0fb9ba0c0c0), C64e(0xaab398b332981919),
+	C64e(0xf668d16827d19e9e), C64e(0x22817f815d7fa3a3),
+	C64e(0xeeaa66aa88664444), C64e(0xd6827e82a87e5454),
+	C64e(0xdde6abe676ab3b3b), C64e(0x959e839e16830b0b),
+	C64e(0xc945ca4503ca8c8c), C64e(0xbc7b297b9529c7c7),
+	C64e(0x056ed36ed6d36b6b), C64e(0x6c443c44503c2828),
+	C64e(0x2c8b798b5579a7a7), C64e(0x813de23d63e2bcbc),
+	C64e(0x31271d272c1d1616), C64e(0x379a769a4176adad),
+	C64e(0x964d3b4dad3bdbdb), C64e(0x9efa56fac8566464),
+	C64e(0xa6d24ed2e84e7474), C64e(0x36221e22281e1414),
+	C64e(0xe476db763fdb9292), C64e(0x121e0a1e180a0c0c),
+	C64e(0xfcb46cb4906c4848), C64e(0x8f37e4376be4b8b8),
+	C64e(0x78e75de7255d9f9f), C64e(0x0fb26eb2616ebdbd),
+	C64e(0x692aef2a86ef4343), C64e(0x35f1a6f193a6c4c4),
+	C64e(0xdae3a8e372a83939), C64e(0xc6f7a4f762a43131),
+	C64e(0x8a593759bd37d3d3), C64e(0x74868b86ff8bf2f2),
+	C64e(0x83563256b132d5d5), C64e(0x4ec543c50d438b8b),
+	C64e(0x85eb59ebdc596e6e), C64e(0x18c2b7c2afb7dada),
+	C64e(0x8e8f8c8f028c0101), C64e(0x1dac64ac7964b1b1),
+	C64e(0xf16dd26d23d29c9c), C64e(0x723be03b92e04949),
+	C64e(0x1fc7b4c7abb4d8d8), C64e(0xb915fa1543faacac),
+	C64e(0xfa090709fd07f3f3), C64e(0xa06f256f8525cfcf),
+	C64e(0x20eaafea8fafcaca), C64e(0x7d898e89f38ef4f4),
+	C64e(0x6720e9208ee94747), C64e(0x3828182820181010),
+	C64e(0x0b64d564ded56f6f), C64e(0x73838883fb88f0f0),
+	C64e(0xfbb16fb1946f4a4a), C64e(0xca967296b8725c5c),
+	C64e(0x546c246c70243838), C64e(0x5f08f108aef15757),
+	C64e(0x2152c752e6c77373), C64e(0x64f351f335519797),
+	C64e(0xae6523658d23cbcb), C64e(0x25847c84597ca1a1),
+	C64e(0x57bf9cbfcb9ce8e8), C64e(0x5d6321637c213e3e),
+	C64e(0xea7cdd7c37dd9696), C64e(0x1e7fdc7fc2dc6161),
+	C64e(0x9c9186911a860d0d), C64e(0x9b9485941e850f0f),
+	C64e(0x4bab90abdb90e0e0), C64e(0xbac642c6f8427c7c),
+	C64e(0x2657c457e2c47171), C64e(0x29e5aae583aacccc),
+	C64e(0xe373d8733bd89090), C64e(0x090f050f0c050606),
+	C64e(0xf4030103f501f7f7), C64e(0x2a36123638121c1c),
+	C64e(0x3cfea3fe9fa3c2c2), C64e(0x8be15fe1d45f6a6a),
+	C64e(0xbe10f91047f9aeae), C64e(0x026bd06bd2d06969),
+	C64e(0xbfa891a82e911717), C64e(0x71e858e829589999),
+	C64e(0x5369276974273a3a), C64e(0xf7d0b9d04eb92727),
+	C64e(0x91483848a938d9d9), C64e(0xde351335cd13ebeb),
+	C64e(0xe5ceb3ce56b32b2b), C64e(0x7755335544332222),
+	C64e(0x04d6bbd6bfbbd2d2), C64e(0x399070904970a9a9),
+	C64e(0x878089800e890707), C64e(0xc1f2a7f266a73333),
+	C64e(0xecc1b6c15ab62d2d), C64e(0x5a66226678223c3c),
+	C64e(0xb8ad92ad2a921515), C64e(0xa96020608920c9c9),
+	C64e(0x5cdb49db15498787), C64e(0xb01aff1a4fffaaaa),
+	C64e(0xd8887888a0785050), C64e(0x2b8e7a8e517aa5a5),
+	C64e(0x898a8f8a068f0303), C64e(0x4a13f813b2f85959),
+	C64e(0x929b809b12800909), C64e(0x2339173934171a1a),
+	C64e(0x1075da75cada6565), C64e(0x84533153b531d7d7),
+	C64e(0xd551c65113c68484), C64e(0x03d3b8d3bbb8d0d0),
+	C64e(0xdc5ec35e1fc38282), C64e(0xe2cbb0cb52b02929),
+	C64e(0xc3997799b4775a5a), C64e(0x2d3311333c111e1e),
+	C64e(0x3d46cb46f6cb7b7b), C64e(0xb71ffc1f4bfca8a8),
+	C64e(0x0c61d661dad66d6d), C64e(0x624e3a4e583a2c2c)
+};
+
+#endif
+
+#define DECL_STATE_SMALL \
+	sph_u64 H[8];
+
+#define READ_STATE_SMALL(sc)   do { \
+		memcpy(H, (sc)->state.wide, sizeof H); \
+	} while (0)
+
+#define WRITE_STATE_SMALL(sc)   do { \
+		memcpy((sc)->state.wide, H, sizeof H); \
+	} while (0)
+
+#if SPH_SMALL_FOOTPRINT_GROESTL
+
+#define RSTT(d, a, b0, b1, b2, b3, b4, b5, b6, b7)   do { \
+		t[d] = T0[B64_0(a[b0])] \
+			^ R64(T0[B64_1(a[b1])],  8) \
+			^ R64(T0[B64_2(a[b2])], 16) \
+			^ R64(T0[B64_3(a[b3])], 24) \
+			^ T4[B64_4(a[b4])] \
+			^ R64(T4[B64_5(a[b5])],  8) \
+			^ R64(T4[B64_6(a[b6])], 16) \
+			^ R64(T4[B64_7(a[b7])], 24); \
+	} while (0)
+
+#else
+
+#define RSTT(d, a, b0, b1, b2, b3, b4, b5, b6, b7)   do { \
+		t[d] = T0[B64_0(a[b0])] \
+			^ T1[B64_1(a[b1])] \
+			^ T2[B64_2(a[b2])] \
+			^ T3[B64_3(a[b3])] \
+			^ T4[B64_4(a[b4])] \
+			^ T5[B64_5(a[b5])] \
+			^ T6[B64_6(a[b6])] \
+			^ T7[B64_7(a[b7])]; \
+	} while (0)
+
+#endif
+
+#define ROUND_SMALL_P(a, r)   do { \
+		sph_u64 t[8]; \
+		a[0] ^= PC64(0x00, r); \
+		a[1] ^= PC64(0x10, r); \
+		a[2] ^= PC64(0x20, r); \
+		a[3] ^= PC64(0x30, r); \
+		a[4] ^= PC64(0x40, r); \
+		a[5] ^= PC64(0x50, r); \
+		a[6] ^= PC64(0x60, r); \
+		a[7] ^= PC64(0x70, r); \
+		RSTT(0, a, 0, 1, 2, 3, 4, 5, 6, 7); \
+		RSTT(1, a, 1, 2, 3, 4, 5, 6, 7, 0); \
+		RSTT(2, a, 2, 3, 4, 5, 6, 7, 0, 1); \
+		RSTT(3, a, 3, 4, 5, 6, 7, 0, 1, 2); \
+		RSTT(4, a, 4, 5, 6, 7, 0, 1, 2, 3); \
+		RSTT(5, a, 5, 6, 7, 0, 1, 2, 3, 4); \
+		RSTT(6, a, 6, 7, 0, 1, 2, 3, 4, 5); \
+		RSTT(7, a, 7, 0, 1, 2, 3, 4, 5, 6); \
+		a[0] = t[0]; \
+		a[1] = t[1]; \
+		a[2] = t[2]; \
+		a[3] = t[3]; \
+		a[4] = t[4]; \
+		a[5] = t[5]; \
+		a[6] = t[6]; \
+		a[7] = t[7]; \
+	} while (0)
+
+#define ROUND_SMALL_Q(a, r)   do { \
+		sph_u64 t[8]; \
+		a[0] ^= QC64(0x00, r); \
+		a[1] ^= QC64(0x10, r); \
+		a[2] ^= QC64(0x20, r); \
+		a[3] ^= QC64(0x30, r); \
+		a[4] ^= QC64(0x40, r); \
+		a[5] ^= QC64(0x50, r); \
+		a[6] ^= QC64(0x60, r); \
+		a[7] ^= QC64(0x70, r); \
+		RSTT(0, a, 1, 3, 5, 7, 0, 2, 4, 6); \
+		RSTT(1, a, 2, 4, 6, 0, 1, 3, 5, 7); \
+		RSTT(2, a, 3, 5, 7, 1, 2, 4, 6, 0); \
+		RSTT(3, a, 4, 6, 0, 2, 3, 5, 7, 1); \
+		RSTT(4, a, 5, 7, 1, 3, 4, 6, 0, 2); \
+		RSTT(5, a, 6, 0, 2, 4, 5, 7, 1, 3); \
+		RSTT(6, a, 7, 1, 3, 5, 6, 0, 2, 4); \
+		RSTT(7, a, 0, 2, 4, 6, 7, 1, 3, 5); \
+		a[0] = t[0]; \
+		a[1] = t[1]; \
+		a[2] = t[2]; \
+		a[3] = t[3]; \
+		a[4] = t[4]; \
+		a[5] = t[5]; \
+		a[6] = t[6]; \
+		a[7] = t[7]; \
+	} while (0)
+
+#if SPH_SMALL_FOOTPRINT_GROESTL
+
+#define PERM_SMALL_P(a)   do { \
+		int r; \
+		for (r = 0; r < 10; r ++) \
+			ROUND_SMALL_P(a, r); \
+	} while (0)
+
+#define PERM_SMALL_Q(a)   do { \
+		int r; \
+		for (r = 0; r < 10; r ++) \
+			ROUND_SMALL_Q(a, r); \
+	} while (0)
+
+#else
+
+/*
+ * Apparently, unrolling more than that confuses GCC, resulting in
+ * lower performance, even though L1 cache would be no problem.
+ */
+#define PERM_SMALL_P(a)   do { \
+		int r; \
+		for (r = 0; r < 10; r += 2) { \
+			ROUND_SMALL_P(a, r + 0); \
+			ROUND_SMALL_P(a, r + 1); \
+		} \
+	} while (0)
+
+#define PERM_SMALL_Q(a)   do { \
+		int r; \
+		for (r = 0; r < 10; r += 2) { \
+			ROUND_SMALL_Q(a, r + 0); \
+			ROUND_SMALL_Q(a, r + 1); \
+		} \
+	} while (0)
+
+#endif
+
+#define COMPRESS_SMALL   do { \
+		sph_u64 g[8], m[8]; \
+		size_t u; \
+		for (u = 0; u < 8; u ++) { \
+			m[u] = dec64e_aligned(buf + (u << 3)); \
+			g[u] = m[u] ^ H[u]; \
+		} \
+		PERM_SMALL_P(g); \
+		PERM_SMALL_Q(m); \
+		for (u = 0; u < 8; u ++) \
+			H[u] ^= g[u] ^ m[u]; \
+	} while (0)
+
+#define FINAL_SMALL   do { \
+		sph_u64 x[8]; \
+		size_t u; \
+		memcpy(x, H, sizeof x); \
+		PERM_SMALL_P(x); \
+		for (u = 0; u < 8; u ++) \
+			H[u] ^= x[u]; \
+	} while (0)
+
+#define DECL_STATE_BIG \
+	sph_u64 H[16];
+
+#define READ_STATE_BIG(sc)   do { \
+		memcpy(H, (sc)->state.wide, sizeof H); \
+	} while (0)
+
+#define WRITE_STATE_BIG(sc)   do { \
+		memcpy((sc)->state.wide, H, sizeof H); \
+	} while (0)
+
+#if SPH_SMALL_FOOTPRINT_GROESTL
+
+#define RBTT(d, a, b0, b1, b2, b3, b4, b5, b6, b7)   do { \
+		t[d] = T0[B64_0(a[b0])] \
+			^ R64(T0[B64_1(a[b1])],  8) \
+			^ R64(T0[B64_2(a[b2])], 16) \
+			^ R64(T0[B64_3(a[b3])], 24) \
+			^ T4[B64_4(a[b4])] \
+			^ R64(T4[B64_5(a[b5])],  8) \
+			^ R64(T4[B64_6(a[b6])], 16) \
+			^ R64(T4[B64_7(a[b7])], 24); \
+	} while (0)
+
+#else
+
+#define RBTT(d, a, b0, b1, b2, b3, b4, b5, b6, b7)   do { \
+		t[d] = T0[B64_0(a[b0])] \
+			^ T1[B64_1(a[b1])] \
+			^ T2[B64_2(a[b2])] \
+			^ T3[B64_3(a[b3])] \
+			^ T4[B64_4(a[b4])] \
+			^ T5[B64_5(a[b5])] \
+			^ T6[B64_6(a[b6])] \
+			^ T7[B64_7(a[b7])]; \
+	} while (0)
+
+#endif
+
+#if SPH_SMALL_FOOTPRINT_GROESTL
+
+#define ROUND_BIG_P(a, r)   do { \
+		sph_u64 t[16]; \
+		size_t u; \
+		a[0x0] ^= PC64(0x00, r); \
+		a[0x1] ^= PC64(0x10, r); \
+		a[0x2] ^= PC64(0x20, r); \
+		a[0x3] ^= PC64(0x30, r); \
+		a[0x4] ^= PC64(0x40, r); \
+		a[0x5] ^= PC64(0x50, r); \
+		a[0x6] ^= PC64(0x60, r); \
+		a[0x7] ^= PC64(0x70, r); \
+		a[0x8] ^= PC64(0x80, r); \
+		a[0x9] ^= PC64(0x90, r); \
+		a[0xA] ^= PC64(0xA0, r); \
+		a[0xB] ^= PC64(0xB0, r); \
+		a[0xC] ^= PC64(0xC0, r); \
+		a[0xD] ^= PC64(0xD0, r); \
+		a[0xE] ^= PC64(0xE0, r); \
+		a[0xF] ^= PC64(0xF0, r); \
+		for (u = 0; u < 16; u += 4) { \
+			RBTT(u + 0, a, u + 0, (u + 1) & 0xF, \
+				(u + 2) & 0xF, (u + 3) & 0xF, (u + 4) & 0xF, \
+				(u + 5) & 0xF, (u + 6) & 0xF, (u + 11) & 0xF); \
+			RBTT(u + 1, a, u + 1, (u + 2) & 0xF, \
+				(u + 3) & 0xF, (u + 4) & 0xF, (u + 5) & 0xF, \
+				(u + 6) & 0xF, (u + 7) & 0xF, (u + 12) & 0xF); \
+			RBTT(u + 2, a, u + 2, (u + 3) & 0xF, \
+				(u + 4) & 0xF, (u + 5) & 0xF, (u + 6) & 0xF, \
+				(u + 7) & 0xF, (u + 8) & 0xF, (u + 13) & 0xF); \
+			RBTT(u + 3, a, u + 3, (u + 4) & 0xF, \
+				(u + 5) & 0xF, (u + 6) & 0xF, (u + 7) & 0xF, \
+				(u + 8) & 0xF, (u + 9) & 0xF, (u + 14) & 0xF); \
+		} \
+		memcpy(a, t, sizeof t); \
+	} while (0)
+
+#define ROUND_BIG_Q(a, r)   do { \
+		sph_u64 t[16]; \
+		size_t u; \
+		a[0x0] ^= QC64(0x00, r); \
+		a[0x1] ^= QC64(0x10, r); \
+		a[0x2] ^= QC64(0x20, r); \
+		a[0x3] ^= QC64(0x30, r); \
+		a[0x4] ^= QC64(0x40, r); \
+		a[0x5] ^= QC64(0x50, r); \
+		a[0x6] ^= QC64(0x60, r); \
+		a[0x7] ^= QC64(0x70, r); \
+		a[0x8] ^= QC64(0x80, r); \
+		a[0x9] ^= QC64(0x90, r); \
+		a[0xA] ^= QC64(0xA0, r); \
+		a[0xB] ^= QC64(0xB0, r); \
+		a[0xC] ^= QC64(0xC0, r); \
+		a[0xD] ^= QC64(0xD0, r); \
+		a[0xE] ^= QC64(0xE0, r); \
+		a[0xF] ^= QC64(0xF0, r); \
+		for (u = 0; u < 16; u += 4) { \
+			RBTT(u + 0, a, (u + 1) & 0xF, (u + 3) & 0xF, \
+				(u + 5) & 0xF, (u + 11) & 0xF, (u + 0) & 0xF, \
+				(u + 2) & 0xF, (u + 4) & 0xF, (u + 6) & 0xF); \
+			RBTT(u + 1, a, (u + 2) & 0xF, (u + 4) & 0xF, \
+				(u + 6) & 0xF, (u + 12) & 0xF, (u + 1) & 0xF, \
+				(u + 3) & 0xF, (u + 5) & 0xF, (u + 7) & 0xF); \
+			RBTT(u + 2, a, (u + 3) & 0xF, (u + 5) & 0xF, \
+				(u + 7) & 0xF, (u + 13) & 0xF, (u + 2) & 0xF, \
+				(u + 4) & 0xF, (u + 6) & 0xF, (u + 8) & 0xF); \
+			RBTT(u + 3, a, (u + 4) & 0xF, (u + 6) & 0xF, \
+				(u + 8) & 0xF, (u + 14) & 0xF, (u + 3) & 0xF, \
+				(u + 5) & 0xF, (u + 7) & 0xF, (u + 9) & 0xF); \
+		} \
+		memcpy(a, t, sizeof t); \
+	} while (0)
+
+#else
+
+#define ROUND_BIG_P(a, r)   do { \
+		sph_u64 t[16]; \
+		a[0x0] ^= PC64(0x00, r); \
+		a[0x1] ^= PC64(0x10, r); \
+		a[0x2] ^= PC64(0x20, r); \
+		a[0x3] ^= PC64(0x30, r); \
+		a[0x4] ^= PC64(0x40, r); \
+		a[0x5] ^= PC64(0x50, r); \
+		a[0x6] ^= PC64(0x60, r); \
+		a[0x7] ^= PC64(0x70, r); \
+		a[0x8] ^= PC64(0x80, r); \
+		a[0x9] ^= PC64(0x90, r); \
+		a[0xA] ^= PC64(0xA0, r); \
+		a[0xB] ^= PC64(0xB0, r); \
+		a[0xC] ^= PC64(0xC0, r); \
+		a[0xD] ^= PC64(0xD0, r); \
+		a[0xE] ^= PC64(0xE0, r); \
+		a[0xF] ^= PC64(0xF0, r); \
+		RBTT(0x0, a, 0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0xB); \
+		RBTT(0x1, a, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7, 0xC); \
+		RBTT(0x2, a, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7, 0x8, 0xD); \
+		RBTT(0x3, a, 0x3, 0x4, 0x5, 0x6, 0x7, 0x8, 0x9, 0xE); \
+		RBTT(0x4, a, 0x4, 0x5, 0x6, 0x7, 0x8, 0x9, 0xA, 0xF); \
+		RBTT(0x5, a, 0x5, 0x6, 0x7, 0x8, 0x9, 0xA, 0xB, 0x0); \
+		RBTT(0x6, a, 0x6, 0x7, 0x8, 0x9, 0xA, 0xB, 0xC, 0x1); \
+		RBTT(0x7, a, 0x7, 0x8, 0x9, 0xA, 0xB, 0xC, 0xD, 0x2); \
+		RBTT(0x8, a, 0x8, 0x9, 0xA, 0xB, 0xC, 0xD, 0xE, 0x3); \
+		RBTT(0x9, a, 0x9, 0xA, 0xB, 0xC, 0xD, 0xE, 0xF, 0x4); \
+		RBTT(0xA, a, 0xA, 0xB, 0xC, 0xD, 0xE, 0xF, 0x0, 0x5); \
+		RBTT(0xB, a, 0xB, 0xC, 0xD, 0xE, 0xF, 0x0, 0x1, 0x6); \
+		RBTT(0xC, a, 0xC, 0xD, 0xE, 0xF, 0x0, 0x1, 0x2, 0x7); \
+		RBTT(0xD, a, 0xD, 0xE, 0xF, 0x0, 0x1, 0x2, 0x3, 0x8); \
+		RBTT(0xE, a, 0xE, 0xF, 0x0, 0x1, 0x2, 0x3, 0x4, 0x9); \
+		RBTT(0xF, a, 0xF, 0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0xA); \
+		a[0x0] = t[0x0]; \
+		a[0x1] = t[0x1]; \
+		a[0x2] = t[0x2]; \
+		a[0x3] = t[0x3]; \
+		a[0x4] = t[0x4]; \
+		a[0x5] = t[0x5]; \
+		a[0x6] = t[0x6]; \
+		a[0x7] = t[0x7]; \
+		a[0x8] = t[0x8]; \
+		a[0x9] = t[0x9]; \
+		a[0xA] = t[0xA]; \
+		a[0xB] = t[0xB]; \
+		a[0xC] = t[0xC]; \
+		a[0xD] = t[0xD]; \
+		a[0xE] = t[0xE]; \
+		a[0xF] = t[0xF]; \
+	} while (0)
+
+#define ROUND_BIG_Q(a, r)   do { \
+		sph_u64 t[16]; \
+		a[0x0] ^= QC64(0x00, r); \
+		a[0x1] ^= QC64(0x10, r); \
+		a[0x2] ^= QC64(0x20, r); \
+		a[0x3] ^= QC64(0x30, r); \
+		a[0x4] ^= QC64(0x40, r); \
+		a[0x5] ^= QC64(0x50, r); \
+		a[0x6] ^= QC64(0x60, r); \
+		a[0x7] ^= QC64(0x70, r); \
+		a[0x8] ^= QC64(0x80, r); \
+		a[0x9] ^= QC64(0x90, r); \
+		a[0xA] ^= QC64(0xA0, r); \
+		a[0xB] ^= QC64(0xB0, r); \
+		a[0xC] ^= QC64(0xC0, r); \
+		a[0xD] ^= QC64(0xD0, r); \
+		a[0xE] ^= QC64(0xE0, r); \
+		a[0xF] ^= QC64(0xF0, r); \
+		RBTT(0x0, a, 0x1, 0x3, 0x5, 0xB, 0x0, 0x2, 0x4, 0x6); \
+		RBTT(0x1, a, 0x2, 0x4, 0x6, 0xC, 0x1, 0x3, 0x5, 0x7); \
+		RBTT(0x2, a, 0x3, 0x5, 0x7, 0xD, 0x2, 0x4, 0x6, 0x8); \
+		RBTT(0x3, a, 0x4, 0x6, 0x8, 0xE, 0x3, 0x5, 0x7, 0x9); \
+		RBTT(0x4, a, 0x5, 0x7, 0x9, 0xF, 0x4, 0x6, 0x8, 0xA); \
+		RBTT(0x5, a, 0x6, 0x8, 0xA, 0x0, 0x5, 0x7, 0x9, 0xB); \
+		RBTT(0x6, a, 0x7, 0x9, 0xB, 0x1, 0x6, 0x8, 0xA, 0xC); \
+		RBTT(0x7, a, 0x8, 0xA, 0xC, 0x2, 0x7, 0x9, 0xB, 0xD); \
+		RBTT(0x8, a, 0x9, 0xB, 0xD, 0x3, 0x8, 0xA, 0xC, 0xE); \
+		RBTT(0x9, a, 0xA, 0xC, 0xE, 0x4, 0x9, 0xB, 0xD, 0xF); \
+		RBTT(0xA, a, 0xB, 0xD, 0xF, 0x5, 0xA, 0xC, 0xE, 0x0); \
+		RBTT(0xB, a, 0xC, 0xE, 0x0, 0x6, 0xB, 0xD, 0xF, 0x1); \
+		RBTT(0xC, a, 0xD, 0xF, 0x1, 0x7, 0xC, 0xE, 0x0, 0x2); \
+		RBTT(0xD, a, 0xE, 0x0, 0x2, 0x8, 0xD, 0xF, 0x1, 0x3); \
+		RBTT(0xE, a, 0xF, 0x1, 0x3, 0x9, 0xE, 0x0, 0x2, 0x4); \
+		RBTT(0xF, a, 0x0, 0x2, 0x4, 0xA, 0xF, 0x1, 0x3, 0x5); \
+		a[0x0] = t[0x0]; \
+		a[0x1] = t[0x1]; \
+		a[0x2] = t[0x2]; \
+		a[0x3] = t[0x3]; \
+		a[0x4] = t[0x4]; \
+		a[0x5] = t[0x5]; \
+		a[0x6] = t[0x6]; \
+		a[0x7] = t[0x7]; \
+		a[0x8] = t[0x8]; \
+		a[0x9] = t[0x9]; \
+		a[0xA] = t[0xA]; \
+		a[0xB] = t[0xB]; \
+		a[0xC] = t[0xC]; \
+		a[0xD] = t[0xD]; \
+		a[0xE] = t[0xE]; \
+		a[0xF] = t[0xF]; \
+	} while (0)
+
+#endif
+
+#define PERM_BIG_P(a)   do { \
+		int r; \
+		for (r = 0; r < 14; r += 2) { \
+			ROUND_BIG_P(a, r + 0); \
+			ROUND_BIG_P(a, r + 1); \
+		} \
+	} while (0)
+
+#define PERM_BIG_Q(a)   do { \
+		int r; \
+		for (r = 0; r < 14; r += 2) { \
+			ROUND_BIG_Q(a, r + 0); \
+			ROUND_BIG_Q(a, r + 1); \
+		} \
+	} while (0)
+
+/* obsolete
+#if SPH_SMALL_FOOTPRINT_GROESTL
+
+#define COMPRESS_BIG   do { \
+		sph_u64 g[16], m[16], *ya; \
+		const sph_u64 *yc; \
+		size_t u; \
+		int i; \
+		for (u = 0; u < 16; u ++) { \
+			m[u] = dec64e_aligned(buf + (u << 3)); \
+			g[u] = m[u] ^ H[u]; \
+		} \
+		ya = g; \
+		yc = CP; \
+		for (i = 0; i < 2; i ++) { \
+			PERM_BIG(ya, yc); \
+			ya = m; \
+			yc = CQ; \
+		} \
+		for (u = 0; u < 16; u ++) { \
+			H[u] ^= g[u] ^ m[u]; \
+		} \
+	} while (0)
+
+#else
+*/
+
+#define COMPRESS_BIG   do { \
+		sph_u64 g[16], m[16]; \
+		size_t u; \
+		for (u = 0; u < 16; u ++) { \
+			m[u] = dec64e_aligned(buf + (u << 3)); \
+			g[u] = m[u] ^ H[u]; \
+		} \
+		PERM_BIG_P(g); \
+		PERM_BIG_Q(m); \
+		for (u = 0; u < 16; u ++) { \
+			H[u] ^= g[u] ^ m[u]; \
+		} \
+	} while (0)
+
+/* obsolete
+#endif
+*/
+
+#define FINAL_BIG   do { \
+		sph_u64 x[16]; \
+		size_t u; \
+		memcpy(x, H, sizeof x); \
+		PERM_BIG_P(x); \
+		for (u = 0; u < 16; u ++) \
+			H[u] ^= x[u]; \
+	} while (0)
+
+#else
+
+static const sph_u32 T0up[] = {
+	C32e(0xc632f4a5), C32e(0xf86f9784), C32e(0xee5eb099), C32e(0xf67a8c8d),
+	C32e(0xffe8170d), C32e(0xd60adcbd), C32e(0xde16c8b1), C32e(0x916dfc54),
+	C32e(0x6090f050), C32e(0x02070503), C32e(0xce2ee0a9), C32e(0x56d1877d),
+	C32e(0xe7cc2b19), C32e(0xb513a662), C32e(0x4d7c31e6), C32e(0xec59b59a),
+	C32e(0x8f40cf45), C32e(0x1fa3bc9d), C32e(0x8949c040), C32e(0xfa689287),
+	C32e(0xefd03f15), C32e(0xb29426eb), C32e(0x8ece40c9), C32e(0xfbe61d0b),
+	C32e(0x416e2fec), C32e(0xb31aa967), C32e(0x5f431cfd), C32e(0x456025ea),
+	C32e(0x23f9dabf), C32e(0x535102f7), C32e(0xe445a196), C32e(0x9b76ed5b),
+	C32e(0x75285dc2), C32e(0xe1c5241c), C32e(0x3dd4e9ae), C32e(0x4cf2be6a),
+	C32e(0x6c82ee5a), C32e(0x7ebdc341), C32e(0xf5f30602), C32e(0x8352d14f),
+	C32e(0x688ce45c), C32e(0x515607f4), C32e(0xd18d5c34), C32e(0xf9e11808),
+	C32e(0xe24cae93), C32e(0xab3e9573), C32e(0x6297f553), C32e(0x2a6b413f),
+	C32e(0x081c140c), C32e(0x9563f652), C32e(0x46e9af65), C32e(0x9d7fe25e),
+	C32e(0x30487828), C32e(0x37cff8a1), C32e(0x0a1b110f), C32e(0x2febc4b5),
+	C32e(0x0e151b09), C32e(0x247e5a36), C32e(0x1badb69b), C32e(0xdf98473d),
+	C32e(0xcda76a26), C32e(0x4ef5bb69), C32e(0x7f334ccd), C32e(0xea50ba9f),
+	C32e(0x123f2d1b), C32e(0x1da4b99e), C32e(0x58c49c74), C32e(0x3446722e),
+	C32e(0x3641772d), C32e(0xdc11cdb2), C32e(0xb49d29ee), C32e(0x5b4d16fb),
+	C32e(0xa4a501f6), C32e(0x76a1d74d), C32e(0xb714a361), C32e(0x7d3449ce),
+	C32e(0x52df8d7b), C32e(0xdd9f423e), C32e(0x5ecd9371), C32e(0x13b1a297),
+	C32e(0xa6a204f5), C32e(0xb901b868), C32e(0x00000000), C32e(0xc1b5742c),
+	C32e(0x40e0a060), C32e(0xe3c2211f), C32e(0x793a43c8), C32e(0xb69a2ced),
+	C32e(0xd40dd9be), C32e(0x8d47ca46), C32e(0x671770d9), C32e(0x72afdd4b),
+	C32e(0x94ed79de), C32e(0x98ff67d4), C32e(0xb09323e8), C32e(0x855bde4a),
+	C32e(0xbb06bd6b), C32e(0xc5bb7e2a), C32e(0x4f7b34e5), C32e(0xedd73a16),
+	C32e(0x86d254c5), C32e(0x9af862d7), C32e(0x6699ff55), C32e(0x11b6a794),
+	C32e(0x8ac04acf), C32e(0xe9d93010), C32e(0x040e0a06), C32e(0xfe669881),
+	C32e(0xa0ab0bf0), C32e(0x78b4cc44), C32e(0x25f0d5ba), C32e(0x4b753ee3),
+	C32e(0xa2ac0ef3), C32e(0x5d4419fe), C32e(0x80db5bc0), C32e(0x0580858a),
+	C32e(0x3fd3ecad), C32e(0x21fedfbc), C32e(0x70a8d848), C32e(0xf1fd0c04),
+	C32e(0x63197adf), C32e(0x772f58c1), C32e(0xaf309f75), C32e(0x42e7a563),
+	C32e(0x20705030), C32e(0xe5cb2e1a), C32e(0xfdef120e), C32e(0xbf08b76d),
+	C32e(0x8155d44c), C32e(0x18243c14), C32e(0x26795f35), C32e(0xc3b2712f),
+	C32e(0xbe8638e1), C32e(0x35c8fda2), C32e(0x88c74fcc), C32e(0x2e654b39),
+	C32e(0x936af957), C32e(0x55580df2), C32e(0xfc619d82), C32e(0x7ab3c947),
+	C32e(0xc827efac), C32e(0xba8832e7), C32e(0x324f7d2b), C32e(0xe642a495),
+	C32e(0xc03bfba0), C32e(0x19aab398), C32e(0x9ef668d1), C32e(0xa322817f),
+	C32e(0x44eeaa66), C32e(0x54d6827e), C32e(0x3bdde6ab), C32e(0x0b959e83),
+	C32e(0x8cc945ca), C32e(0xc7bc7b29), C32e(0x6b056ed3), C32e(0x286c443c),
+	C32e(0xa72c8b79), C32e(0xbc813de2), C32e(0x1631271d), C32e(0xad379a76),
+	C32e(0xdb964d3b), C32e(0x649efa56), C32e(0x74a6d24e), C32e(0x1436221e),
+	C32e(0x92e476db), C32e(0x0c121e0a), C32e(0x48fcb46c), C32e(0xb88f37e4),
+	C32e(0x9f78e75d), C32e(0xbd0fb26e), C32e(0x43692aef), C32e(0xc435f1a6),
+	C32e(0x39dae3a8), C32e(0x31c6f7a4), C32e(0xd38a5937), C32e(0xf274868b),
+	C32e(0xd5835632), C32e(0x8b4ec543), C32e(0x6e85eb59), C32e(0xda18c2b7),
+	C32e(0x018e8f8c), C32e(0xb11dac64), C32e(0x9cf16dd2), C32e(0x49723be0),
+	C32e(0xd81fc7b4), C32e(0xacb915fa), C32e(0xf3fa0907), C32e(0xcfa06f25),
+	C32e(0xca20eaaf), C32e(0xf47d898e), C32e(0x476720e9), C32e(0x10382818),
+	C32e(0x6f0b64d5), C32e(0xf0738388), C32e(0x4afbb16f), C32e(0x5cca9672),
+	C32e(0x38546c24), C32e(0x575f08f1), C32e(0x732152c7), C32e(0x9764f351),
+	C32e(0xcbae6523), C32e(0xa125847c), C32e(0xe857bf9c), C32e(0x3e5d6321),
+	C32e(0x96ea7cdd), C32e(0x611e7fdc), C32e(0x0d9c9186), C32e(0x0f9b9485),
+	C32e(0xe04bab90), C32e(0x7cbac642), C32e(0x712657c4), C32e(0xcc29e5aa),
+	C32e(0x90e373d8), C32e(0x06090f05), C32e(0xf7f40301), C32e(0x1c2a3612),
+	C32e(0xc23cfea3), C32e(0x6a8be15f), C32e(0xaebe10f9), C32e(0x69026bd0),
+	C32e(0x17bfa891), C32e(0x9971e858), C32e(0x3a536927), C32e(0x27f7d0b9),
+	C32e(0xd9914838), C32e(0xebde3513), C32e(0x2be5ceb3), C32e(0x22775533),
+	C32e(0xd204d6bb), C32e(0xa9399070), C32e(0x07878089), C32e(0x33c1f2a7),
+	C32e(0x2decc1b6), C32e(0x3c5a6622), C32e(0x15b8ad92), C32e(0xc9a96020),
+	C32e(0x875cdb49), C32e(0xaab01aff), C32e(0x50d88878), C32e(0xa52b8e7a),
+	C32e(0x03898a8f), C32e(0x594a13f8), C32e(0x09929b80), C32e(0x1a233917),
+	C32e(0x651075da), C32e(0xd7845331), C32e(0x84d551c6), C32e(0xd003d3b8),
+	C32e(0x82dc5ec3), C32e(0x29e2cbb0), C32e(0x5ac39977), C32e(0x1e2d3311),
+	C32e(0x7b3d46cb), C32e(0xa8b71ffc), C32e(0x6d0c61d6), C32e(0x2c624e3a)
+};
+
+static const sph_u32 T0dn[] = {
+	C32e(0xf497a5c6), C32e(0x97eb84f8), C32e(0xb0c799ee), C32e(0x8cf78df6),
+	C32e(0x17e50dff), C32e(0xdcb7bdd6), C32e(0xc8a7b1de), C32e(0xfc395491),
+	C32e(0xf0c05060), C32e(0x05040302), C32e(0xe087a9ce), C32e(0x87ac7d56),
+	C32e(0x2bd519e7), C32e(0xa67162b5), C32e(0x319ae64d), C32e(0xb5c39aec),
+	C32e(0xcf05458f), C32e(0xbc3e9d1f), C32e(0xc0094089), C32e(0x92ef87fa),
+	C32e(0x3fc515ef), C32e(0x267febb2), C32e(0x4007c98e), C32e(0x1ded0bfb),
+	C32e(0x2f82ec41), C32e(0xa97d67b3), C32e(0x1cbefd5f), C32e(0x258aea45),
+	C32e(0xda46bf23), C32e(0x02a6f753), C32e(0xa1d396e4), C32e(0xed2d5b9b),
+	C32e(0x5deac275), C32e(0x24d91ce1), C32e(0xe97aae3d), C32e(0xbe986a4c),
+	C32e(0xeed85a6c), C32e(0xc3fc417e), C32e(0x06f102f5), C32e(0xd11d4f83),
+	C32e(0xe4d05c68), C32e(0x07a2f451), C32e(0x5cb934d1), C32e(0x18e908f9),
+	C32e(0xaedf93e2), C32e(0x954d73ab), C32e(0xf5c45362), C32e(0x41543f2a),
+	C32e(0x14100c08), C32e(0xf6315295), C32e(0xaf8c6546), C32e(0xe2215e9d),
+	C32e(0x78602830), C32e(0xf86ea137), C32e(0x11140f0a), C32e(0xc45eb52f),
+	C32e(0x1b1c090e), C32e(0x5a483624), C32e(0xb6369b1b), C32e(0x47a53ddf),
+	C32e(0x6a8126cd), C32e(0xbb9c694e), C32e(0x4cfecd7f), C32e(0xbacf9fea),
+	C32e(0x2d241b12), C32e(0xb93a9e1d), C32e(0x9cb07458), C32e(0x72682e34),
+	C32e(0x776c2d36), C32e(0xcda3b2dc), C32e(0x2973eeb4), C32e(0x16b6fb5b),
+	C32e(0x0153f6a4), C32e(0xd7ec4d76), C32e(0xa37561b7), C32e(0x49face7d),
+	C32e(0x8da47b52), C32e(0x42a13edd), C32e(0x93bc715e), C32e(0xa2269713),
+	C32e(0x0457f5a6), C32e(0xb86968b9), C32e(0x00000000), C32e(0x74992cc1),
+	C32e(0xa0806040), C32e(0x21dd1fe3), C32e(0x43f2c879), C32e(0x2c77edb6),
+	C32e(0xd9b3bed4), C32e(0xca01468d), C32e(0x70ced967), C32e(0xdde44b72),
+	C32e(0x7933de94), C32e(0x672bd498), C32e(0x237be8b0), C32e(0xde114a85),
+	C32e(0xbd6d6bbb), C32e(0x7e912ac5), C32e(0x349ee54f), C32e(0x3ac116ed),
+	C32e(0x5417c586), C32e(0x622fd79a), C32e(0xffcc5566), C32e(0xa7229411),
+	C32e(0x4a0fcf8a), C32e(0x30c910e9), C32e(0x0a080604), C32e(0x98e781fe),
+	C32e(0x0b5bf0a0), C32e(0xccf04478), C32e(0xd54aba25), C32e(0x3e96e34b),
+	C32e(0x0e5ff3a2), C32e(0x19bafe5d), C32e(0x5b1bc080), C32e(0x850a8a05),
+	C32e(0xec7ead3f), C32e(0xdf42bc21), C32e(0xd8e04870), C32e(0x0cf904f1),
+	C32e(0x7ac6df63), C32e(0x58eec177), C32e(0x9f4575af), C32e(0xa5846342),
+	C32e(0x50403020), C32e(0x2ed11ae5), C32e(0x12e10efd), C32e(0xb7656dbf),
+	C32e(0xd4194c81), C32e(0x3c301418), C32e(0x5f4c3526), C32e(0x719d2fc3),
+	C32e(0x3867e1be), C32e(0xfd6aa235), C32e(0x4f0bcc88), C32e(0x4b5c392e),
+	C32e(0xf93d5793), C32e(0x0daaf255), C32e(0x9de382fc), C32e(0xc9f4477a),
+	C32e(0xef8bacc8), C32e(0x326fe7ba), C32e(0x7d642b32), C32e(0xa4d795e6),
+	C32e(0xfb9ba0c0), C32e(0xb3329819), C32e(0x6827d19e), C32e(0x815d7fa3),
+	C32e(0xaa886644), C32e(0x82a87e54), C32e(0xe676ab3b), C32e(0x9e16830b),
+	C32e(0x4503ca8c), C32e(0x7b9529c7), C32e(0x6ed6d36b), C32e(0x44503c28),
+	C32e(0x8b5579a7), C32e(0x3d63e2bc), C32e(0x272c1d16), C32e(0x9a4176ad),
+	C32e(0x4dad3bdb), C32e(0xfac85664), C32e(0xd2e84e74), C32e(0x22281e14),
+	C32e(0x763fdb92), C32e(0x1e180a0c), C32e(0xb4906c48), C32e(0x376be4b8),
+	C32e(0xe7255d9f), C32e(0xb2616ebd), C32e(0x2a86ef43), C32e(0xf193a6c4),
+	C32e(0xe372a839), C32e(0xf762a431), C32e(0x59bd37d3), C32e(0x86ff8bf2),
+	C32e(0x56b132d5), C32e(0xc50d438b), C32e(0xebdc596e), C32e(0xc2afb7da),
+	C32e(0x8f028c01), C32e(0xac7964b1), C32e(0x6d23d29c), C32e(0x3b92e049),
+	C32e(0xc7abb4d8), C32e(0x1543faac), C32e(0x09fd07f3), C32e(0x6f8525cf),
+	C32e(0xea8fafca), C32e(0x89f38ef4), C32e(0x208ee947), C32e(0x28201810),
+	C32e(0x64ded56f), C32e(0x83fb88f0), C32e(0xb1946f4a), C32e(0x96b8725c),
+	C32e(0x6c702438), C32e(0x08aef157), C32e(0x52e6c773), C32e(0xf3355197),
+	C32e(0x658d23cb), C32e(0x84597ca1), C32e(0xbfcb9ce8), C32e(0x637c213e),
+	C32e(0x7c37dd96), C32e(0x7fc2dc61), C32e(0x911a860d), C32e(0x941e850f),
+	C32e(0xabdb90e0), C32e(0xc6f8427c), C32e(0x57e2c471), C32e(0xe583aacc),
+	C32e(0x733bd890), C32e(0x0f0c0506), C32e(0x03f501f7), C32e(0x3638121c),
+	C32e(0xfe9fa3c2), C32e(0xe1d45f6a), C32e(0x1047f9ae), C32e(0x6bd2d069),
+	C32e(0xa82e9117), C32e(0xe8295899), C32e(0x6974273a), C32e(0xd04eb927),
+	C32e(0x48a938d9), C32e(0x35cd13eb), C32e(0xce56b32b), C32e(0x55443322),
+	C32e(0xd6bfbbd2), C32e(0x904970a9), C32e(0x800e8907), C32e(0xf266a733),
+	C32e(0xc15ab62d), C32e(0x6678223c), C32e(0xad2a9215), C32e(0x608920c9),
+	C32e(0xdb154987), C32e(0x1a4fffaa), C32e(0x88a07850), C32e(0x8e517aa5),
+	C32e(0x8a068f03), C32e(0x13b2f859), C32e(0x9b128009), C32e(0x3934171a),
+	C32e(0x75cada65), C32e(0x53b531d7), C32e(0x5113c684), C32e(0xd3bbb8d0),
+	C32e(0x5e1fc382), C32e(0xcb52b029), C32e(0x99b4775a), C32e(0x333c111e),
+	C32e(0x46f6cb7b), C32e(0x1f4bfca8), C32e(0x61dad66d), C32e(0x4e583a2c)
+};
+
+static const sph_u32 T1up[] = {
+	C32e(0xc6c632f4), C32e(0xf8f86f97), C32e(0xeeee5eb0), C32e(0xf6f67a8c),
+	C32e(0xffffe817), C32e(0xd6d60adc), C32e(0xdede16c8), C32e(0x91916dfc),
+	C32e(0x606090f0), C32e(0x02020705), C32e(0xcece2ee0), C32e(0x5656d187),
+	C32e(0xe7e7cc2b), C32e(0xb5b513a6), C32e(0x4d4d7c31), C32e(0xecec59b5),
+	C32e(0x8f8f40cf), C32e(0x1f1fa3bc), C32e(0x898949c0), C32e(0xfafa6892),
+	C32e(0xefefd03f), C32e(0xb2b29426), C32e(0x8e8ece40), C32e(0xfbfbe61d),
+	C32e(0x41416e2f), C32e(0xb3b31aa9), C32e(0x5f5f431c), C32e(0x45456025),
+	C32e(0x2323f9da), C32e(0x53535102), C32e(0xe4e445a1), C32e(0x9b9b76ed),
+	C32e(0x7575285d), C32e(0xe1e1c524), C32e(0x3d3dd4e9), C32e(0x4c4cf2be),
+	C32e(0x6c6c82ee), C32e(0x7e7ebdc3), C32e(0xf5f5f306), C32e(0x838352d1),
+	C32e(0x68688ce4), C32e(0x51515607), C32e(0xd1d18d5c), C32e(0xf9f9e118),
+	C32e(0xe2e24cae), C32e(0xabab3e95), C32e(0x626297f5), C32e(0x2a2a6b41),
+	C32e(0x08081c14), C32e(0x959563f6), C32e(0x4646e9af), C32e(0x9d9d7fe2),
+	C32e(0x30304878), C32e(0x3737cff8), C32e(0x0a0a1b11), C32e(0x2f2febc4),
+	C32e(0x0e0e151b), C32e(0x24247e5a), C32e(0x1b1badb6), C32e(0xdfdf9847),
+	C32e(0xcdcda76a), C32e(0x4e4ef5bb), C32e(0x7f7f334c), C32e(0xeaea50ba),
+	C32e(0x12123f2d), C32e(0x1d1da4b9), C32e(0x5858c49c), C32e(0x34344672),
+	C32e(0x36364177), C32e(0xdcdc11cd), C32e(0xb4b49d29), C32e(0x5b5b4d16),
+	C32e(0xa4a4a501), C32e(0x7676a1d7), C32e(0xb7b714a3), C32e(0x7d7d3449),
+	C32e(0x5252df8d), C32e(0xdddd9f42), C32e(0x5e5ecd93), C32e(0x1313b1a2),
+	C32e(0xa6a6a204), C32e(0xb9b901b8), C32e(0x00000000), C32e(0xc1c1b574),
+	C32e(0x4040e0a0), C32e(0xe3e3c221), C32e(0x79793a43), C32e(0xb6b69a2c),
+	C32e(0xd4d40dd9), C32e(0x8d8d47ca), C32e(0x67671770), C32e(0x7272afdd),
+	C32e(0x9494ed79), C32e(0x9898ff67), C32e(0xb0b09323), C32e(0x85855bde),
+	C32e(0xbbbb06bd), C32e(0xc5c5bb7e), C32e(0x4f4f7b34), C32e(0xededd73a),
+	C32e(0x8686d254), C32e(0x9a9af862), C32e(0x666699ff), C32e(0x1111b6a7),
+	C32e(0x8a8ac04a), C32e(0xe9e9d930), C32e(0x04040e0a), C32e(0xfefe6698),
+	C32e(0xa0a0ab0b), C32e(0x7878b4cc), C32e(0x2525f0d5), C32e(0x4b4b753e),
+	C32e(0xa2a2ac0e), C32e(0x5d5d4419), C32e(0x8080db5b), C32e(0x05058085),
+	C32e(0x3f3fd3ec), C32e(0x2121fedf), C32e(0x7070a8d8), C32e(0xf1f1fd0c),
+	C32e(0x6363197a), C32e(0x77772f58), C32e(0xafaf309f), C32e(0x4242e7a5),
+	C32e(0x20207050), C32e(0xe5e5cb2e), C32e(0xfdfdef12), C32e(0xbfbf08b7),
+	C32e(0x818155d4), C32e(0x1818243c), C32e(0x2626795f), C32e(0xc3c3b271),
+	C32e(0xbebe8638), C32e(0x3535c8fd), C32e(0x8888c74f), C32e(0x2e2e654b),
+	C32e(0x93936af9), C32e(0x5555580d), C32e(0xfcfc619d), C32e(0x7a7ab3c9),
+	C32e(0xc8c827ef), C32e(0xbaba8832), C32e(0x32324f7d), C32e(0xe6e642a4),
+	C32e(0xc0c03bfb), C32e(0x1919aab3), C32e(0x9e9ef668), C32e(0xa3a32281),
+	C32e(0x4444eeaa), C32e(0x5454d682), C32e(0x3b3bdde6), C32e(0x0b0b959e),
+	C32e(0x8c8cc945), C32e(0xc7c7bc7b), C32e(0x6b6b056e), C32e(0x28286c44),
+	C32e(0xa7a72c8b), C32e(0xbcbc813d), C32e(0x16163127), C32e(0xadad379a),
+	C32e(0xdbdb964d), C32e(0x64649efa), C32e(0x7474a6d2), C32e(0x14143622),
+	C32e(0x9292e476), C32e(0x0c0c121e), C32e(0x4848fcb4), C32e(0xb8b88f37),
+	C32e(0x9f9f78e7), C32e(0xbdbd0fb2), C32e(0x4343692a), C32e(0xc4c435f1),
+	C32e(0x3939dae3), C32e(0x3131c6f7), C32e(0xd3d38a59), C32e(0xf2f27486),
+	C32e(0xd5d58356), C32e(0x8b8b4ec5), C32e(0x6e6e85eb), C32e(0xdada18c2),
+	C32e(0x01018e8f), C32e(0xb1b11dac), C32e(0x9c9cf16d), C32e(0x4949723b),
+	C32e(0xd8d81fc7), C32e(0xacacb915), C32e(0xf3f3fa09), C32e(0xcfcfa06f),
+	C32e(0xcaca20ea), C32e(0xf4f47d89), C32e(0x47476720), C32e(0x10103828),
+	C32e(0x6f6f0b64), C32e(0xf0f07383), C32e(0x4a4afbb1), C32e(0x5c5cca96),
+	C32e(0x3838546c), C32e(0x57575f08), C32e(0x73732152), C32e(0x979764f3),
+	C32e(0xcbcbae65), C32e(0xa1a12584), C32e(0xe8e857bf), C32e(0x3e3e5d63),
+	C32e(0x9696ea7c), C32e(0x61611e7f), C32e(0x0d0d9c91), C32e(0x0f0f9b94),
+	C32e(0xe0e04bab), C32e(0x7c7cbac6), C32e(0x71712657), C32e(0xcccc29e5),
+	C32e(0x9090e373), C32e(0x0606090f), C32e(0xf7f7f403), C32e(0x1c1c2a36),
+	C32e(0xc2c23cfe), C32e(0x6a6a8be1), C32e(0xaeaebe10), C32e(0x6969026b),
+	C32e(0x1717bfa8), C32e(0x999971e8), C32e(0x3a3a5369), C32e(0x2727f7d0),
+	C32e(0xd9d99148), C32e(0xebebde35), C32e(0x2b2be5ce), C32e(0x22227755),
+	C32e(0xd2d204d6), C32e(0xa9a93990), C32e(0x07078780), C32e(0x3333c1f2),
+	C32e(0x2d2decc1), C32e(0x3c3c5a66), C32e(0x1515b8ad), C32e(0xc9c9a960),
+	C32e(0x87875cdb), C32e(0xaaaab01a), C32e(0x5050d888), C32e(0xa5a52b8e),
+	C32e(0x0303898a), C32e(0x59594a13), C32e(0x0909929b), C32e(0x1a1a2339),
+	C32e(0x65651075), C32e(0xd7d78453), C32e(0x8484d551), C32e(0xd0d003d3),
+	C32e(0x8282dc5e), C32e(0x2929e2cb), C32e(0x5a5ac399), C32e(0x1e1e2d33),
+	C32e(0x7b7b3d46), C32e(0xa8a8b71f), C32e(0x6d6d0c61), C32e(0x2c2c624e)
+};
+
+static const sph_u32 T1dn[] = {
+	C32e(0xa5f497a5), C32e(0x8497eb84), C32e(0x99b0c799), C32e(0x8d8cf78d),
+	C32e(0x0d17e50d), C32e(0xbddcb7bd), C32e(0xb1c8a7b1), C32e(0x54fc3954),
+	C32e(0x50f0c050), C32e(0x03050403), C32e(0xa9e087a9), C32e(0x7d87ac7d),
+	C32e(0x192bd519), C32e(0x62a67162), C32e(0xe6319ae6), C32e(0x9ab5c39a),
+	C32e(0x45cf0545), C32e(0x9dbc3e9d), C32e(0x40c00940), C32e(0x8792ef87),
+	C32e(0x153fc515), C32e(0xeb267feb), C32e(0xc94007c9), C32e(0x0b1ded0b),
+	C32e(0xec2f82ec), C32e(0x67a97d67), C32e(0xfd1cbefd), C32e(0xea258aea),
+	C32e(0xbfda46bf), C32e(0xf702a6f7), C32e(0x96a1d396), C32e(0x5bed2d5b),
+	C32e(0xc25deac2), C32e(0x1c24d91c), C32e(0xaee97aae), C32e(0x6abe986a),
+	C32e(0x5aeed85a), C32e(0x41c3fc41), C32e(0x0206f102), C32e(0x4fd11d4f),
+	C32e(0x5ce4d05c), C32e(0xf407a2f4), C32e(0x345cb934), C32e(0x0818e908),
+	C32e(0x93aedf93), C32e(0x73954d73), C32e(0x53f5c453), C32e(0x3f41543f),
+	C32e(0x0c14100c), C32e(0x52f63152), C32e(0x65af8c65), C32e(0x5ee2215e),
+	C32e(0x28786028), C32e(0xa1f86ea1), C32e(0x0f11140f), C32e(0xb5c45eb5),
+	C32e(0x091b1c09), C32e(0x365a4836), C32e(0x9bb6369b), C32e(0x3d47a53d),
+	C32e(0x266a8126), C32e(0x69bb9c69), C32e(0xcd4cfecd), C32e(0x9fbacf9f),
+	C32e(0x1b2d241b), C32e(0x9eb93a9e), C32e(0x749cb074), C32e(0x2e72682e),
+	C32e(0x2d776c2d), C32e(0xb2cda3b2), C32e(0xee2973ee), C32e(0xfb16b6fb),
+	C32e(0xf60153f6), C32e(0x4dd7ec4d), C32e(0x61a37561), C32e(0xce49face),
+	C32e(0x7b8da47b), C32e(0x3e42a13e), C32e(0x7193bc71), C32e(0x97a22697),
+	C32e(0xf50457f5), C32e(0x68b86968), C32e(0x00000000), C32e(0x2c74992c),
+	C32e(0x60a08060), C32e(0x1f21dd1f), C32e(0xc843f2c8), C32e(0xed2c77ed),
+	C32e(0xbed9b3be), C32e(0x46ca0146), C32e(0xd970ced9), C32e(0x4bdde44b),
+	C32e(0xde7933de), C32e(0xd4672bd4), C32e(0xe8237be8), C32e(0x4ade114a),
+	C32e(0x6bbd6d6b), C32e(0x2a7e912a), C32e(0xe5349ee5), C32e(0x163ac116),
+	C32e(0xc55417c5), C32e(0xd7622fd7), C32e(0x55ffcc55), C32e(0x94a72294),
+	C32e(0xcf4a0fcf), C32e(0x1030c910), C32e(0x060a0806), C32e(0x8198e781),
+	C32e(0xf00b5bf0), C32e(0x44ccf044), C32e(0xbad54aba), C32e(0xe33e96e3),
+	C32e(0xf30e5ff3), C32e(0xfe19bafe), C32e(0xc05b1bc0), C32e(0x8a850a8a),
+	C32e(0xadec7ead), C32e(0xbcdf42bc), C32e(0x48d8e048), C32e(0x040cf904),
+	C32e(0xdf7ac6df), C32e(0xc158eec1), C32e(0x759f4575), C32e(0x63a58463),
+	C32e(0x30504030), C32e(0x1a2ed11a), C32e(0x0e12e10e), C32e(0x6db7656d),
+	C32e(0x4cd4194c), C32e(0x143c3014), C32e(0x355f4c35), C32e(0x2f719d2f),
+	C32e(0xe13867e1), C32e(0xa2fd6aa2), C32e(0xcc4f0bcc), C32e(0x394b5c39),
+	C32e(0x57f93d57), C32e(0xf20daaf2), C32e(0x829de382), C32e(0x47c9f447),
+	C32e(0xacef8bac), C32e(0xe7326fe7), C32e(0x2b7d642b), C32e(0x95a4d795),
+	C32e(0xa0fb9ba0), C32e(0x98b33298), C32e(0xd16827d1), C32e(0x7f815d7f),
+	C32e(0x66aa8866), C32e(0x7e82a87e), C32e(0xabe676ab), C32e(0x839e1683),
+	C32e(0xca4503ca), C32e(0x297b9529), C32e(0xd36ed6d3), C32e(0x3c44503c),
+	C32e(0x798b5579), C32e(0xe23d63e2), C32e(0x1d272c1d), C32e(0x769a4176),
+	C32e(0x3b4dad3b), C32e(0x56fac856), C32e(0x4ed2e84e), C32e(0x1e22281e),
+	C32e(0xdb763fdb), C32e(0x0a1e180a), C32e(0x6cb4906c), C32e(0xe4376be4),
+	C32e(0x5de7255d), C32e(0x6eb2616e), C32e(0xef2a86ef), C32e(0xa6f193a6),
+	C32e(0xa8e372a8), C32e(0xa4f762a4), C32e(0x3759bd37), C32e(0x8b86ff8b),
+	C32e(0x3256b132), C32e(0x43c50d43), C32e(0x59ebdc59), C32e(0xb7c2afb7),
+	C32e(0x8c8f028c), C32e(0x64ac7964), C32e(0xd26d23d2), C32e(0xe03b92e0),
+	C32e(0xb4c7abb4), C32e(0xfa1543fa), C32e(0x0709fd07), C32e(0x256f8525),
+	C32e(0xafea8faf), C32e(0x8e89f38e), C32e(0xe9208ee9), C32e(0x18282018),
+	C32e(0xd564ded5), C32e(0x8883fb88), C32e(0x6fb1946f), C32e(0x7296b872),
+	C32e(0x246c7024), C32e(0xf108aef1), C32e(0xc752e6c7), C32e(0x51f33551),
+	C32e(0x23658d23), C32e(0x7c84597c), C32e(0x9cbfcb9c), C32e(0x21637c21),
+	C32e(0xdd7c37dd), C32e(0xdc7fc2dc), C32e(0x86911a86), C32e(0x85941e85),
+	C32e(0x90abdb90), C32e(0x42c6f842), C32e(0xc457e2c4), C32e(0xaae583aa),
+	C32e(0xd8733bd8), C32e(0x050f0c05), C32e(0x0103f501), C32e(0x12363812),
+	C32e(0xa3fe9fa3), C32e(0x5fe1d45f), C32e(0xf91047f9), C32e(0xd06bd2d0),
+	C32e(0x91a82e91), C32e(0x58e82958), C32e(0x27697427), C32e(0xb9d04eb9),
+	C32e(0x3848a938), C32e(0x1335cd13), C32e(0xb3ce56b3), C32e(0x33554433),
+	C32e(0xbbd6bfbb), C32e(0x70904970), C32e(0x89800e89), C32e(0xa7f266a7),
+	C32e(0xb6c15ab6), C32e(0x22667822), C32e(0x92ad2a92), C32e(0x20608920),
+	C32e(0x49db1549), C32e(0xff1a4fff), C32e(0x7888a078), C32e(0x7a8e517a),
+	C32e(0x8f8a068f), C32e(0xf813b2f8), C32e(0x809b1280), C32e(0x17393417),
+	C32e(0xda75cada), C32e(0x3153b531), C32e(0xc65113c6), C32e(0xb8d3bbb8),
+	C32e(0xc35e1fc3), C32e(0xb0cb52b0), C32e(0x7799b477), C32e(0x11333c11),
+	C32e(0xcb46f6cb), C32e(0xfc1f4bfc), C32e(0xd661dad6), C32e(0x3a4e583a)
+};
+
+static const sph_u32 T2up[] = {
+	C32e(0xa5c6c632), C32e(0x84f8f86f), C32e(0x99eeee5e), C32e(0x8df6f67a),
+	C32e(0x0dffffe8), C32e(0xbdd6d60a), C32e(0xb1dede16), C32e(0x5491916d),
+	C32e(0x50606090), C32e(0x03020207), C32e(0xa9cece2e), C32e(0x7d5656d1),
+	C32e(0x19e7e7cc), C32e(0x62b5b513), C32e(0xe64d4d7c), C32e(0x9aecec59),
+	C32e(0x458f8f40), C32e(0x9d1f1fa3), C32e(0x40898949), C32e(0x87fafa68),
+	C32e(0x15efefd0), C32e(0xebb2b294), C32e(0xc98e8ece), C32e(0x0bfbfbe6),
+	C32e(0xec41416e), C32e(0x67b3b31a), C32e(0xfd5f5f43), C32e(0xea454560),
+	C32e(0xbf2323f9), C32e(0xf7535351), C32e(0x96e4e445), C32e(0x5b9b9b76),
+	C32e(0xc2757528), C32e(0x1ce1e1c5), C32e(0xae3d3dd4), C32e(0x6a4c4cf2),
+	C32e(0x5a6c6c82), C32e(0x417e7ebd), C32e(0x02f5f5f3), C32e(0x4f838352),
+	C32e(0x5c68688c), C32e(0xf4515156), C32e(0x34d1d18d), C32e(0x08f9f9e1),
+	C32e(0x93e2e24c), C32e(0x73abab3e), C32e(0x53626297), C32e(0x3f2a2a6b),
+	C32e(0x0c08081c), C32e(0x52959563), C32e(0x654646e9), C32e(0x5e9d9d7f),
+	C32e(0x28303048), C32e(0xa13737cf), C32e(0x0f0a0a1b), C32e(0xb52f2feb),
+	C32e(0x090e0e15), C32e(0x3624247e), C32e(0x9b1b1bad), C32e(0x3ddfdf98),
+	C32e(0x26cdcda7), C32e(0x694e4ef5), C32e(0xcd7f7f33), C32e(0x9feaea50),
+	C32e(0x1b12123f), C32e(0x9e1d1da4), C32e(0x745858c4), C32e(0x2e343446),
+	C32e(0x2d363641), C32e(0xb2dcdc11), C32e(0xeeb4b49d), C32e(0xfb5b5b4d),
+	C32e(0xf6a4a4a5), C32e(0x4d7676a1), C32e(0x61b7b714), C32e(0xce7d7d34),
+	C32e(0x7b5252df), C32e(0x3edddd9f), C32e(0x715e5ecd), C32e(0x971313b1),
+	C32e(0xf5a6a6a2), C32e(0x68b9b901), C32e(0x00000000), C32e(0x2cc1c1b5),
+	C32e(0x604040e0), C32e(0x1fe3e3c2), C32e(0xc879793a), C32e(0xedb6b69a),
+	C32e(0xbed4d40d), C32e(0x468d8d47), C32e(0xd9676717), C32e(0x4b7272af),
+	C32e(0xde9494ed), C32e(0xd49898ff), C32e(0xe8b0b093), C32e(0x4a85855b),
+	C32e(0x6bbbbb06), C32e(0x2ac5c5bb), C32e(0xe54f4f7b), C32e(0x16ededd7),
+	C32e(0xc58686d2), C32e(0xd79a9af8), C32e(0x55666699), C32e(0x941111b6),
+	C32e(0xcf8a8ac0), C32e(0x10e9e9d9), C32e(0x0604040e), C32e(0x81fefe66),
+	C32e(0xf0a0a0ab), C32e(0x447878b4), C32e(0xba2525f0), C32e(0xe34b4b75),
+	C32e(0xf3a2a2ac), C32e(0xfe5d5d44), C32e(0xc08080db), C32e(0x8a050580),
+	C32e(0xad3f3fd3), C32e(0xbc2121fe), C32e(0x487070a8), C32e(0x04f1f1fd),
+	C32e(0xdf636319), C32e(0xc177772f), C32e(0x75afaf30), C32e(0x634242e7),
+	C32e(0x30202070), C32e(0x1ae5e5cb), C32e(0x0efdfdef), C32e(0x6dbfbf08),
+	C32e(0x4c818155), C32e(0x14181824), C32e(0x35262679), C32e(0x2fc3c3b2),
+	C32e(0xe1bebe86), C32e(0xa23535c8), C32e(0xcc8888c7), C32e(0x392e2e65),
+	C32e(0x5793936a), C32e(0xf2555558), C32e(0x82fcfc61), C32e(0x477a7ab3),
+	C32e(0xacc8c827), C32e(0xe7baba88), C32e(0x2b32324f), C32e(0x95e6e642),
+	C32e(0xa0c0c03b), C32e(0x981919aa), C32e(0xd19e9ef6), C32e(0x7fa3a322),
+	C32e(0x664444ee), C32e(0x7e5454d6), C32e(0xab3b3bdd), C32e(0x830b0b95),
+	C32e(0xca8c8cc9), C32e(0x29c7c7bc), C32e(0xd36b6b05), C32e(0x3c28286c),
+	C32e(0x79a7a72c), C32e(0xe2bcbc81), C32e(0x1d161631), C32e(0x76adad37),
+	C32e(0x3bdbdb96), C32e(0x5664649e), C32e(0x4e7474a6), C32e(0x1e141436),
+	C32e(0xdb9292e4), C32e(0x0a0c0c12), C32e(0x6c4848fc), C32e(0xe4b8b88f),
+	C32e(0x5d9f9f78), C32e(0x6ebdbd0f), C32e(0xef434369), C32e(0xa6c4c435),
+	C32e(0xa83939da), C32e(0xa43131c6), C32e(0x37d3d38a), C32e(0x8bf2f274),
+	C32e(0x32d5d583), C32e(0x438b8b4e), C32e(0x596e6e85), C32e(0xb7dada18),
+	C32e(0x8c01018e), C32e(0x64b1b11d), C32e(0xd29c9cf1), C32e(0xe0494972),
+	C32e(0xb4d8d81f), C32e(0xfaacacb9), C32e(0x07f3f3fa), C32e(0x25cfcfa0),
+	C32e(0xafcaca20), C32e(0x8ef4f47d), C32e(0xe9474767), C32e(0x18101038),
+	C32e(0xd56f6f0b), C32e(0x88f0f073), C32e(0x6f4a4afb), C32e(0x725c5cca),
+	C32e(0x24383854), C32e(0xf157575f), C32e(0xc7737321), C32e(0x51979764),
+	C32e(0x23cbcbae), C32e(0x7ca1a125), C32e(0x9ce8e857), C32e(0x213e3e5d),
+	C32e(0xdd9696ea), C32e(0xdc61611e), C32e(0x860d0d9c), C32e(0x850f0f9b),
+	C32e(0x90e0e04b), C32e(0x427c7cba), C32e(0xc4717126), C32e(0xaacccc29),
+	C32e(0xd89090e3), C32e(0x05060609), C32e(0x01f7f7f4), C32e(0x121c1c2a),
+	C32e(0xa3c2c23c), C32e(0x5f6a6a8b), C32e(0xf9aeaebe), C32e(0xd0696902),
+	C32e(0x911717bf), C32e(0x58999971), C32e(0x273a3a53), C32e(0xb92727f7),
+	C32e(0x38d9d991), C32e(0x13ebebde), C32e(0xb32b2be5), C32e(0x33222277),
+	C32e(0xbbd2d204), C32e(0x70a9a939), C32e(0x89070787), C32e(0xa73333c1),
+	C32e(0xb62d2dec), C32e(0x223c3c5a), C32e(0x921515b8), C32e(0x20c9c9a9),
+	C32e(0x4987875c), C32e(0xffaaaab0), C32e(0x785050d8), C32e(0x7aa5a52b),
+	C32e(0x8f030389), C32e(0xf859594a), C32e(0x80090992), C32e(0x171a1a23),
+	C32e(0xda656510), C32e(0x31d7d784), C32e(0xc68484d5), C32e(0xb8d0d003),
+	C32e(0xc38282dc), C32e(0xb02929e2), C32e(0x775a5ac3), C32e(0x111e1e2d),
+	C32e(0xcb7b7b3d), C32e(0xfca8a8b7), C32e(0xd66d6d0c), C32e(0x3a2c2c62)
+};
+
+static const sph_u32 T2dn[] = {
+	C32e(0xf4a5f497), C32e(0x978497eb), C32e(0xb099b0c7), C32e(0x8c8d8cf7),
+	C32e(0x170d17e5), C32e(0xdcbddcb7), C32e(0xc8b1c8a7), C32e(0xfc54fc39),
+	C32e(0xf050f0c0), C32e(0x05030504), C32e(0xe0a9e087), C32e(0x877d87ac),
+	C32e(0x2b192bd5), C32e(0xa662a671), C32e(0x31e6319a), C32e(0xb59ab5c3),
+	C32e(0xcf45cf05), C32e(0xbc9dbc3e), C32e(0xc040c009), C32e(0x928792ef),
+	C32e(0x3f153fc5), C32e(0x26eb267f), C32e(0x40c94007), C32e(0x1d0b1ded),
+	C32e(0x2fec2f82), C32e(0xa967a97d), C32e(0x1cfd1cbe), C32e(0x25ea258a),
+	C32e(0xdabfda46), C32e(0x02f702a6), C32e(0xa196a1d3), C32e(0xed5bed2d),
+	C32e(0x5dc25dea), C32e(0x241c24d9), C32e(0xe9aee97a), C32e(0xbe6abe98),
+	C32e(0xee5aeed8), C32e(0xc341c3fc), C32e(0x060206f1), C32e(0xd14fd11d),
+	C32e(0xe45ce4d0), C32e(0x07f407a2), C32e(0x5c345cb9), C32e(0x180818e9),
+	C32e(0xae93aedf), C32e(0x9573954d), C32e(0xf553f5c4), C32e(0x413f4154),
+	C32e(0x140c1410), C32e(0xf652f631), C32e(0xaf65af8c), C32e(0xe25ee221),
+	C32e(0x78287860), C32e(0xf8a1f86e), C32e(0x110f1114), C32e(0xc4b5c45e),
+	C32e(0x1b091b1c), C32e(0x5a365a48), C32e(0xb69bb636), C32e(0x473d47a5),
+	C32e(0x6a266a81), C32e(0xbb69bb9c), C32e(0x4ccd4cfe), C32e(0xba9fbacf),
+	C32e(0x2d1b2d24), C32e(0xb99eb93a), C32e(0x9c749cb0), C32e(0x722e7268),
+	C32e(0x772d776c), C32e(0xcdb2cda3), C32e(0x29ee2973), C32e(0x16fb16b6),
+	C32e(0x01f60153), C32e(0xd74dd7ec), C32e(0xa361a375), C32e(0x49ce49fa),
+	C32e(0x8d7b8da4), C32e(0x423e42a1), C32e(0x937193bc), C32e(0xa297a226),
+	C32e(0x04f50457), C32e(0xb868b869), C32e(0x00000000), C32e(0x742c7499),
+	C32e(0xa060a080), C32e(0x211f21dd), C32e(0x43c843f2), C32e(0x2ced2c77),
+	C32e(0xd9bed9b3), C32e(0xca46ca01), C32e(0x70d970ce), C32e(0xdd4bdde4),
+	C32e(0x79de7933), C32e(0x67d4672b), C32e(0x23e8237b), C32e(0xde4ade11),
+	C32e(0xbd6bbd6d), C32e(0x7e2a7e91), C32e(0x34e5349e), C32e(0x3a163ac1),
+	C32e(0x54c55417), C32e(0x62d7622f), C32e(0xff55ffcc), C32e(0xa794a722),
+	C32e(0x4acf4a0f), C32e(0x301030c9), C32e(0x0a060a08), C32e(0x988198e7),
+	C32e(0x0bf00b5b), C32e(0xcc44ccf0), C32e(0xd5bad54a), C32e(0x3ee33e96),
+	C32e(0x0ef30e5f), C32e(0x19fe19ba), C32e(0x5bc05b1b), C32e(0x858a850a),
+	C32e(0xecadec7e), C32e(0xdfbcdf42), C32e(0xd848d8e0), C32e(0x0c040cf9),
+	C32e(0x7adf7ac6), C32e(0x58c158ee), C32e(0x9f759f45), C32e(0xa563a584),
+	C32e(0x50305040), C32e(0x2e1a2ed1), C32e(0x120e12e1), C32e(0xb76db765),
+	C32e(0xd44cd419), C32e(0x3c143c30), C32e(0x5f355f4c), C32e(0x712f719d),
+	C32e(0x38e13867), C32e(0xfda2fd6a), C32e(0x4fcc4f0b), C32e(0x4b394b5c),
+	C32e(0xf957f93d), C32e(0x0df20daa), C32e(0x9d829de3), C32e(0xc947c9f4),
+	C32e(0xefacef8b), C32e(0x32e7326f), C32e(0x7d2b7d64), C32e(0xa495a4d7),
+	C32e(0xfba0fb9b), C32e(0xb398b332), C32e(0x68d16827), C32e(0x817f815d),
+	C32e(0xaa66aa88), C32e(0x827e82a8), C32e(0xe6abe676), C32e(0x9e839e16),
+	C32e(0x45ca4503), C32e(0x7b297b95), C32e(0x6ed36ed6), C32e(0x443c4450),
+	C32e(0x8b798b55), C32e(0x3de23d63), C32e(0x271d272c), C32e(0x9a769a41),
+	C32e(0x4d3b4dad), C32e(0xfa56fac8), C32e(0xd24ed2e8), C32e(0x221e2228),
+	C32e(0x76db763f), C32e(0x1e0a1e18), C32e(0xb46cb490), C32e(0x37e4376b),
+	C32e(0xe75de725), C32e(0xb26eb261), C32e(0x2aef2a86), C32e(0xf1a6f193),
+	C32e(0xe3a8e372), C32e(0xf7a4f762), C32e(0x593759bd), C32e(0x868b86ff),
+	C32e(0x563256b1), C32e(0xc543c50d), C32e(0xeb59ebdc), C32e(0xc2b7c2af),
+	C32e(0x8f8c8f02), C32e(0xac64ac79), C32e(0x6dd26d23), C32e(0x3be03b92),
+	C32e(0xc7b4c7ab), C32e(0x15fa1543), C32e(0x090709fd), C32e(0x6f256f85),
+	C32e(0xeaafea8f), C32e(0x898e89f3), C32e(0x20e9208e), C32e(0x28182820),
+	C32e(0x64d564de), C32e(0x838883fb), C32e(0xb16fb194), C32e(0x967296b8),
+	C32e(0x6c246c70), C32e(0x08f108ae), C32e(0x52c752e6), C32e(0xf351f335),
+	C32e(0x6523658d), C32e(0x847c8459), C32e(0xbf9cbfcb), C32e(0x6321637c),
+	C32e(0x7cdd7c37), C32e(0x7fdc7fc2), C32e(0x9186911a), C32e(0x9485941e),
+	C32e(0xab90abdb), C32e(0xc642c6f8), C32e(0x57c457e2), C32e(0xe5aae583),
+	C32e(0x73d8733b), C32e(0x0f050f0c), C32e(0x030103f5), C32e(0x36123638),
+	C32e(0xfea3fe9f), C32e(0xe15fe1d4), C32e(0x10f91047), C32e(0x6bd06bd2),
+	C32e(0xa891a82e), C32e(0xe858e829), C32e(0x69276974), C32e(0xd0b9d04e),
+	C32e(0x483848a9), C32e(0x351335cd), C32e(0xceb3ce56), C32e(0x55335544),
+	C32e(0xd6bbd6bf), C32e(0x90709049), C32e(0x8089800e), C32e(0xf2a7f266),
+	C32e(0xc1b6c15a), C32e(0x66226678), C32e(0xad92ad2a), C32e(0x60206089),
+	C32e(0xdb49db15), C32e(0x1aff1a4f), C32e(0x887888a0), C32e(0x8e7a8e51),
+	C32e(0x8a8f8a06), C32e(0x13f813b2), C32e(0x9b809b12), C32e(0x39173934),
+	C32e(0x75da75ca), C32e(0x533153b5), C32e(0x51c65113), C32e(0xd3b8d3bb),
+	C32e(0x5ec35e1f), C32e(0xcbb0cb52), C32e(0x997799b4), C32e(0x3311333c),
+	C32e(0x46cb46f6), C32e(0x1ffc1f4b), C32e(0x61d661da), C32e(0x4e3a4e58)
+};
+
+static const sph_u32 T3up[] = {
+	C32e(0x97a5c6c6), C32e(0xeb84f8f8), C32e(0xc799eeee), C32e(0xf78df6f6),
+	C32e(0xe50dffff), C32e(0xb7bdd6d6), C32e(0xa7b1dede), C32e(0x39549191),
+	C32e(0xc0506060), C32e(0x04030202), C32e(0x87a9cece), C32e(0xac7d5656),
+	C32e(0xd519e7e7), C32e(0x7162b5b5), C32e(0x9ae64d4d), C32e(0xc39aecec),
+	C32e(0x05458f8f), C32e(0x3e9d1f1f), C32e(0x09408989), C32e(0xef87fafa),
+	C32e(0xc515efef), C32e(0x7febb2b2), C32e(0x07c98e8e), C32e(0xed0bfbfb),
+	C32e(0x82ec4141), C32e(0x7d67b3b3), C32e(0xbefd5f5f), C32e(0x8aea4545),
+	C32e(0x46bf2323), C32e(0xa6f75353), C32e(0xd396e4e4), C32e(0x2d5b9b9b),
+	C32e(0xeac27575), C32e(0xd91ce1e1), C32e(0x7aae3d3d), C32e(0x986a4c4c),
+	C32e(0xd85a6c6c), C32e(0xfc417e7e), C32e(0xf102f5f5), C32e(0x1d4f8383),
+	C32e(0xd05c6868), C32e(0xa2f45151), C32e(0xb934d1d1), C32e(0xe908f9f9),
+	C32e(0xdf93e2e2), C32e(0x4d73abab), C32e(0xc4536262), C32e(0x543f2a2a),
+	C32e(0x100c0808), C32e(0x31529595), C32e(0x8c654646), C32e(0x215e9d9d),
+	C32e(0x60283030), C32e(0x6ea13737), C32e(0x140f0a0a), C32e(0x5eb52f2f),
+	C32e(0x1c090e0e), C32e(0x48362424), C32e(0x369b1b1b), C32e(0xa53ddfdf),
+	C32e(0x8126cdcd), C32e(0x9c694e4e), C32e(0xfecd7f7f), C32e(0xcf9feaea),
+	C32e(0x241b1212), C32e(0x3a9e1d1d), C32e(0xb0745858), C32e(0x682e3434),
+	C32e(0x6c2d3636), C32e(0xa3b2dcdc), C32e(0x73eeb4b4), C32e(0xb6fb5b5b),
+	C32e(0x53f6a4a4), C32e(0xec4d7676), C32e(0x7561b7b7), C32e(0xface7d7d),
+	C32e(0xa47b5252), C32e(0xa13edddd), C32e(0xbc715e5e), C32e(0x26971313),
+	C32e(0x57f5a6a6), C32e(0x6968b9b9), C32e(0x00000000), C32e(0x992cc1c1),
+	C32e(0x80604040), C32e(0xdd1fe3e3), C32e(0xf2c87979), C32e(0x77edb6b6),
+	C32e(0xb3bed4d4), C32e(0x01468d8d), C32e(0xced96767), C32e(0xe44b7272),
+	C32e(0x33de9494), C32e(0x2bd49898), C32e(0x7be8b0b0), C32e(0x114a8585),
+	C32e(0x6d6bbbbb), C32e(0x912ac5c5), C32e(0x9ee54f4f), C32e(0xc116eded),
+	C32e(0x17c58686), C32e(0x2fd79a9a), C32e(0xcc556666), C32e(0x22941111),
+	C32e(0x0fcf8a8a), C32e(0xc910e9e9), C32e(0x08060404), C32e(0xe781fefe),
+	C32e(0x5bf0a0a0), C32e(0xf0447878), C32e(0x4aba2525), C32e(0x96e34b4b),
+	C32e(0x5ff3a2a2), C32e(0xbafe5d5d), C32e(0x1bc08080), C32e(0x0a8a0505),
+	C32e(0x7ead3f3f), C32e(0x42bc2121), C32e(0xe0487070), C32e(0xf904f1f1),
+	C32e(0xc6df6363), C32e(0xeec17777), C32e(0x4575afaf), C32e(0x84634242),
+	C32e(0x40302020), C32e(0xd11ae5e5), C32e(0xe10efdfd), C32e(0x656dbfbf),
+	C32e(0x194c8181), C32e(0x30141818), C32e(0x4c352626), C32e(0x9d2fc3c3),
+	C32e(0x67e1bebe), C32e(0x6aa23535), C32e(0x0bcc8888), C32e(0x5c392e2e),
+	C32e(0x3d579393), C32e(0xaaf25555), C32e(0xe382fcfc), C32e(0xf4477a7a),
+	C32e(0x8bacc8c8), C32e(0x6fe7baba), C32e(0x642b3232), C32e(0xd795e6e6),
+	C32e(0x9ba0c0c0), C32e(0x32981919), C32e(0x27d19e9e), C32e(0x5d7fa3a3),
+	C32e(0x88664444), C32e(0xa87e5454), C32e(0x76ab3b3b), C32e(0x16830b0b),
+	C32e(0x03ca8c8c), C32e(0x9529c7c7), C32e(0xd6d36b6b), C32e(0x503c2828),
+	C32e(0x5579a7a7), C32e(0x63e2bcbc), C32e(0x2c1d1616), C32e(0x4176adad),
+	C32e(0xad3bdbdb), C32e(0xc8566464), C32e(0xe84e7474), C32e(0x281e1414),
+	C32e(0x3fdb9292), C32e(0x180a0c0c), C32e(0x906c4848), C32e(0x6be4b8b8),
+	C32e(0x255d9f9f), C32e(0x616ebdbd), C32e(0x86ef4343), C32e(0x93a6c4c4),
+	C32e(0x72a83939), C32e(0x62a43131), C32e(0xbd37d3d3), C32e(0xff8bf2f2),
+	C32e(0xb132d5d5), C32e(0x0d438b8b), C32e(0xdc596e6e), C32e(0xafb7dada),
+	C32e(0x028c0101), C32e(0x7964b1b1), C32e(0x23d29c9c), C32e(0x92e04949),
+	C32e(0xabb4d8d8), C32e(0x43faacac), C32e(0xfd07f3f3), C32e(0x8525cfcf),
+	C32e(0x8fafcaca), C32e(0xf38ef4f4), C32e(0x8ee94747), C32e(0x20181010),
+	C32e(0xded56f6f), C32e(0xfb88f0f0), C32e(0x946f4a4a), C32e(0xb8725c5c),
+	C32e(0x70243838), C32e(0xaef15757), C32e(0xe6c77373), C32e(0x35519797),
+	C32e(0x8d23cbcb), C32e(0x597ca1a1), C32e(0xcb9ce8e8), C32e(0x7c213e3e),
+	C32e(0x37dd9696), C32e(0xc2dc6161), C32e(0x1a860d0d), C32e(0x1e850f0f),
+	C32e(0xdb90e0e0), C32e(0xf8427c7c), C32e(0xe2c47171), C32e(0x83aacccc),
+	C32e(0x3bd89090), C32e(0x0c050606), C32e(0xf501f7f7), C32e(0x38121c1c),
+	C32e(0x9fa3c2c2), C32e(0xd45f6a6a), C32e(0x47f9aeae), C32e(0xd2d06969),
+	C32e(0x2e911717), C32e(0x29589999), C32e(0x74273a3a), C32e(0x4eb92727),
+	C32e(0xa938d9d9), C32e(0xcd13ebeb), C32e(0x56b32b2b), C32e(0x44332222),
+	C32e(0xbfbbd2d2), C32e(0x4970a9a9), C32e(0x0e890707), C32e(0x66a73333),
+	C32e(0x5ab62d2d), C32e(0x78223c3c), C32e(0x2a921515), C32e(0x8920c9c9),
+	C32e(0x15498787), C32e(0x4fffaaaa), C32e(0xa0785050), C32e(0x517aa5a5),
+	C32e(0x068f0303), C32e(0xb2f85959), C32e(0x12800909), C32e(0x34171a1a),
+	C32e(0xcada6565), C32e(0xb531d7d7), C32e(0x13c68484), C32e(0xbbb8d0d0),
+	C32e(0x1fc38282), C32e(0x52b02929), C32e(0xb4775a5a), C32e(0x3c111e1e),
+	C32e(0xf6cb7b7b), C32e(0x4bfca8a8), C32e(0xdad66d6d), C32e(0x583a2c2c)
+};
+
+static const sph_u32 T3dn[] = {
+	C32e(0x32f4a5f4), C32e(0x6f978497), C32e(0x5eb099b0), C32e(0x7a8c8d8c),
+	C32e(0xe8170d17), C32e(0x0adcbddc), C32e(0x16c8b1c8), C32e(0x6dfc54fc),
+	C32e(0x90f050f0), C32e(0x07050305), C32e(0x2ee0a9e0), C32e(0xd1877d87),
+	C32e(0xcc2b192b), C32e(0x13a662a6), C32e(0x7c31e631), C32e(0x59b59ab5),
+	C32e(0x40cf45cf), C32e(0xa3bc9dbc), C32e(0x49c040c0), C32e(0x68928792),
+	C32e(0xd03f153f), C32e(0x9426eb26), C32e(0xce40c940), C32e(0xe61d0b1d),
+	C32e(0x6e2fec2f), C32e(0x1aa967a9), C32e(0x431cfd1c), C32e(0x6025ea25),
+	C32e(0xf9dabfda), C32e(0x5102f702), C32e(0x45a196a1), C32e(0x76ed5bed),
+	C32e(0x285dc25d), C32e(0xc5241c24), C32e(0xd4e9aee9), C32e(0xf2be6abe),
+	C32e(0x82ee5aee), C32e(0xbdc341c3), C32e(0xf3060206), C32e(0x52d14fd1),
+	C32e(0x8ce45ce4), C32e(0x5607f407), C32e(0x8d5c345c), C32e(0xe1180818),
+	C32e(0x4cae93ae), C32e(0x3e957395), C32e(0x97f553f5), C32e(0x6b413f41),
+	C32e(0x1c140c14), C32e(0x63f652f6), C32e(0xe9af65af), C32e(0x7fe25ee2),
+	C32e(0x48782878), C32e(0xcff8a1f8), C32e(0x1b110f11), C32e(0xebc4b5c4),
+	C32e(0x151b091b), C32e(0x7e5a365a), C32e(0xadb69bb6), C32e(0x98473d47),
+	C32e(0xa76a266a), C32e(0xf5bb69bb), C32e(0x334ccd4c), C32e(0x50ba9fba),
+	C32e(0x3f2d1b2d), C32e(0xa4b99eb9), C32e(0xc49c749c), C32e(0x46722e72),
+	C32e(0x41772d77), C32e(0x11cdb2cd), C32e(0x9d29ee29), C32e(0x4d16fb16),
+	C32e(0xa501f601), C32e(0xa1d74dd7), C32e(0x14a361a3), C32e(0x3449ce49),
+	C32e(0xdf8d7b8d), C32e(0x9f423e42), C32e(0xcd937193), C32e(0xb1a297a2),
+	C32e(0xa204f504), C32e(0x01b868b8), C32e(0x00000000), C32e(0xb5742c74),
+	C32e(0xe0a060a0), C32e(0xc2211f21), C32e(0x3a43c843), C32e(0x9a2ced2c),
+	C32e(0x0dd9bed9), C32e(0x47ca46ca), C32e(0x1770d970), C32e(0xafdd4bdd),
+	C32e(0xed79de79), C32e(0xff67d467), C32e(0x9323e823), C32e(0x5bde4ade),
+	C32e(0x06bd6bbd), C32e(0xbb7e2a7e), C32e(0x7b34e534), C32e(0xd73a163a),
+	C32e(0xd254c554), C32e(0xf862d762), C32e(0x99ff55ff), C32e(0xb6a794a7),
+	C32e(0xc04acf4a), C32e(0xd9301030), C32e(0x0e0a060a), C32e(0x66988198),
+	C32e(0xab0bf00b), C32e(0xb4cc44cc), C32e(0xf0d5bad5), C32e(0x753ee33e),
+	C32e(0xac0ef30e), C32e(0x4419fe19), C32e(0xdb5bc05b), C32e(0x80858a85),
+	C32e(0xd3ecadec), C32e(0xfedfbcdf), C32e(0xa8d848d8), C32e(0xfd0c040c),
+	C32e(0x197adf7a), C32e(0x2f58c158), C32e(0x309f759f), C32e(0xe7a563a5),
+	C32e(0x70503050), C32e(0xcb2e1a2e), C32e(0xef120e12), C32e(0x08b76db7),
+	C32e(0x55d44cd4), C32e(0x243c143c), C32e(0x795f355f), C32e(0xb2712f71),
+	C32e(0x8638e138), C32e(0xc8fda2fd), C32e(0xc74fcc4f), C32e(0x654b394b),
+	C32e(0x6af957f9), C32e(0x580df20d), C32e(0x619d829d), C32e(0xb3c947c9),
+	C32e(0x27efacef), C32e(0x8832e732), C32e(0x4f7d2b7d), C32e(0x42a495a4),
+	C32e(0x3bfba0fb), C32e(0xaab398b3), C32e(0xf668d168), C32e(0x22817f81),
+	C32e(0xeeaa66aa), C32e(0xd6827e82), C32e(0xdde6abe6), C32e(0x959e839e),
+	C32e(0xc945ca45), C32e(0xbc7b297b), C32e(0x056ed36e), C32e(0x6c443c44),
+	C32e(0x2c8b798b), C32e(0x813de23d), C32e(0x31271d27), C32e(0x379a769a),
+	C32e(0x964d3b4d), C32e(0x9efa56fa), C32e(0xa6d24ed2), C32e(0x36221e22),
+	C32e(0xe476db76), C32e(0x121e0a1e), C32e(0xfcb46cb4), C32e(0x8f37e437),
+	C32e(0x78e75de7), C32e(0x0fb26eb2), C32e(0x692aef2a), C32e(0x35f1a6f1),
+	C32e(0xdae3a8e3), C32e(0xc6f7a4f7), C32e(0x8a593759), C32e(0x74868b86),
+	C32e(0x83563256), C32e(0x4ec543c5), C32e(0x85eb59eb), C32e(0x18c2b7c2),
+	C32e(0x8e8f8c8f), C32e(0x1dac64ac), C32e(0xf16dd26d), C32e(0x723be03b),
+	C32e(0x1fc7b4c7), C32e(0xb915fa15), C32e(0xfa090709), C32e(0xa06f256f),
+	C32e(0x20eaafea), C32e(0x7d898e89), C32e(0x6720e920), C32e(0x38281828),
+	C32e(0x0b64d564), C32e(0x73838883), C32e(0xfbb16fb1), C32e(0xca967296),
+	C32e(0x546c246c), C32e(0x5f08f108), C32e(0x2152c752), C32e(0x64f351f3),
+	C32e(0xae652365), C32e(0x25847c84), C32e(0x57bf9cbf), C32e(0x5d632163),
+	C32e(0xea7cdd7c), C32e(0x1e7fdc7f), C32e(0x9c918691), C32e(0x9b948594),
+	C32e(0x4bab90ab), C32e(0xbac642c6), C32e(0x2657c457), C32e(0x29e5aae5),
+	C32e(0xe373d873), C32e(0x090f050f), C32e(0xf4030103), C32e(0x2a361236),
+	C32e(0x3cfea3fe), C32e(0x8be15fe1), C32e(0xbe10f910), C32e(0x026bd06b),
+	C32e(0xbfa891a8), C32e(0x71e858e8), C32e(0x53692769), C32e(0xf7d0b9d0),
+	C32e(0x91483848), C32e(0xde351335), C32e(0xe5ceb3ce), C32e(0x77553355),
+	C32e(0x04d6bbd6), C32e(0x39907090), C32e(0x87808980), C32e(0xc1f2a7f2),
+	C32e(0xecc1b6c1), C32e(0x5a662266), C32e(0xb8ad92ad), C32e(0xa9602060),
+	C32e(0x5cdb49db), C32e(0xb01aff1a), C32e(0xd8887888), C32e(0x2b8e7a8e),
+	C32e(0x898a8f8a), C32e(0x4a13f813), C32e(0x929b809b), C32e(0x23391739),
+	C32e(0x1075da75), C32e(0x84533153), C32e(0xd551c651), C32e(0x03d3b8d3),
+	C32e(0xdc5ec35e), C32e(0xe2cbb0cb), C32e(0xc3997799), C32e(0x2d331133),
+	C32e(0x3d46cb46), C32e(0xb71ffc1f), C32e(0x0c61d661), C32e(0x624e3a4e)
+};
+
+#define DECL_STATE_SMALL \
+	sph_u32 H[16];
+
+#define READ_STATE_SMALL(sc)   do { \
+		memcpy(H, (sc)->state.narrow, sizeof H); \
+	} while (0)
+
+#define WRITE_STATE_SMALL(sc)   do { \
+		memcpy((sc)->state.narrow, H, sizeof H); \
+	} while (0)
+
+#define XCAT(x, y)    XCAT_(x, y)
+#define XCAT_(x, y)   x ## y
+
+#define RSTT(d0, d1, a, b0, b1, b2, b3, b4, b5, b6, b7)   do { \
+		t[d0] = T0up[B32_0(a[b0])] \
+			^ T1up[B32_1(a[b1])] \
+			^ T2up[B32_2(a[b2])] \
+			^ T3up[B32_3(a[b3])] \
+			^ T0dn[B32_0(a[b4])] \
+			^ T1dn[B32_1(a[b5])] \
+			^ T2dn[B32_2(a[b6])] \
+			^ T3dn[B32_3(a[b7])]; \
+		t[d1] = T0dn[B32_0(a[b0])] \
+			^ T1dn[B32_1(a[b1])] \
+			^ T2dn[B32_2(a[b2])] \
+			^ T3dn[B32_3(a[b3])] \
+			^ T0up[B32_0(a[b4])] \
+			^ T1up[B32_1(a[b5])] \
+			^ T2up[B32_2(a[b6])] \
+			^ T3up[B32_3(a[b7])]; \
+	} while (0)
+
+#define ROUND_SMALL_P(a, r)   do { \
+		sph_u32 t[16]; \
+		a[0x0] ^= PC32up(0x00, r); \
+		a[0x1] ^= PC32dn(0x00, r); \
+		a[0x2] ^= PC32up(0x10, r); \
+		a[0x3] ^= PC32dn(0x10, r); \
+		a[0x4] ^= PC32up(0x20, r); \
+		a[0x5] ^= PC32dn(0x20, r); \
+		a[0x6] ^= PC32up(0x30, r); \
+		a[0x7] ^= PC32dn(0x30, r); \
+		a[0x8] ^= PC32up(0x40, r); \
+		a[0x9] ^= PC32dn(0x40, r); \
+		a[0xA] ^= PC32up(0x50, r); \
+		a[0xB] ^= PC32dn(0x50, r); \
+		a[0xC] ^= PC32up(0x60, r); \
+		a[0xD] ^= PC32dn(0x60, r); \
+		a[0xE] ^= PC32up(0x70, r); \
+		a[0xF] ^= PC32dn(0x70, r); \
+		RSTT(0x0, 0x1, a, 0x0, 0x2, 0x4, 0x6, 0x9, 0xB, 0xD, 0xF); \
+		RSTT(0x2, 0x3, a, 0x2, 0x4, 0x6, 0x8, 0xB, 0xD, 0xF, 0x1); \
+		RSTT(0x4, 0x5, a, 0x4, 0x6, 0x8, 0xA, 0xD, 0xF, 0x1, 0x3); \
+		RSTT(0x6, 0x7, a, 0x6, 0x8, 0xA, 0xC, 0xF, 0x1, 0x3, 0x5); \
+		RSTT(0x8, 0x9, a, 0x8, 0xA, 0xC, 0xE, 0x1, 0x3, 0x5, 0x7); \
+		RSTT(0xA, 0xB, a, 0xA, 0xC, 0xE, 0x0, 0x3, 0x5, 0x7, 0x9); \
+		RSTT(0xC, 0xD, a, 0xC, 0xE, 0x0, 0x2, 0x5, 0x7, 0x9, 0xB); \
+		RSTT(0xE, 0xF, a, 0xE, 0x0, 0x2, 0x4, 0x7, 0x9, 0xB, 0xD); \
+		memcpy(a, t, sizeof t); \
+	} while (0)
+
+#define ROUND_SMALL_Q(a, r)   do { \
+		sph_u32 t[16]; \
+		a[0x0] ^= QC32up(0x00, r); \
+		a[0x1] ^= QC32dn(0x00, r); \
+		a[0x2] ^= QC32up(0x10, r); \
+		a[0x3] ^= QC32dn(0x10, r); \
+		a[0x4] ^= QC32up(0x20, r); \
+		a[0x5] ^= QC32dn(0x20, r); \
+		a[0x6] ^= QC32up(0x30, r); \
+		a[0x7] ^= QC32dn(0x30, r); \
+		a[0x8] ^= QC32up(0x40, r); \
+		a[0x9] ^= QC32dn(0x40, r); \
+		a[0xA] ^= QC32up(0x50, r); \
+		a[0xB] ^= QC32dn(0x50, r); \
+		a[0xC] ^= QC32up(0x60, r); \
+		a[0xD] ^= QC32dn(0x60, r); \
+		a[0xE] ^= QC32up(0x70, r); \
+		a[0xF] ^= QC32dn(0x70, r); \
+		RSTT(0x0, 0x1, a, 0x2, 0x6, 0xA, 0xE, 0x1, 0x5, 0x9, 0xD); \
+		RSTT(0x2, 0x3, a, 0x4, 0x8, 0xC, 0x0, 0x3, 0x7, 0xB, 0xF); \
+		RSTT(0x4, 0x5, a, 0x6, 0xA, 0xE, 0x2, 0x5, 0x9, 0xD, 0x1); \
+		RSTT(0x6, 0x7, a, 0x8, 0xC, 0x0, 0x4, 0x7, 0xB, 0xF, 0x3); \
+		RSTT(0x8, 0x9, a, 0xA, 0xE, 0x2, 0x6, 0x9, 0xD, 0x1, 0x5); \
+		RSTT(0xA, 0xB, a, 0xC, 0x0, 0x4, 0x8, 0xB, 0xF, 0x3, 0x7); \
+		RSTT(0xC, 0xD, a, 0xE, 0x2, 0x6, 0xA, 0xD, 0x1, 0x5, 0x9); \
+		RSTT(0xE, 0xF, a, 0x0, 0x4, 0x8, 0xC, 0xF, 0x3, 0x7, 0xB); \
+		memcpy(a, t, sizeof t); \
+	} while (0)
+
+#if SPH_SMALL_FOOTPRINT_GROESTL
+
+#define PERM_SMALL_P(a)   do { \
+		int r; \
+		for (r = 0; r < 10; r ++) \
+			ROUND_SMALL_P(a, r); \
+	} while (0)
+
+#define PERM_SMALL_Q(a)   do { \
+		int r; \
+		for (r = 0; r < 10; r ++) \
+			ROUND_SMALL_Q(a, r); \
+	} while (0)
+
+#else
+
+#define PERM_SMALL_P(a)   do { \
+		int r; \
+		for (r = 0; r < 10; r += 2) { \
+			ROUND_SMALL_P(a, r + 0); \
+			ROUND_SMALL_P(a, r + 1); \
+		} \
+	} while (0)
+
+#define PERM_SMALL_Q(a)   do { \
+		int r; \
+		for (r = 0; r < 10; r += 2) { \
+			ROUND_SMALL_Q(a, r + 0); \
+			ROUND_SMALL_Q(a, r + 1); \
+		} \
+	} while (0)
+
+#endif
+
+#define COMPRESS_SMALL   do { \
+		sph_u32 g[16], m[16]; \
+		size_t u; \
+		for (u = 0; u < 16; u ++) { \
+			m[u] = dec32e_aligned(buf + (u << 2)); \
+			g[u] = m[u] ^ H[u]; \
+		} \
+		PERM_SMALL_P(g); \
+		PERM_SMALL_Q(m); \
+		for (u = 0; u < 16; u ++) \
+			H[u] ^= g[u] ^ m[u]; \
+	} while (0)
+
+#define FINAL_SMALL   do { \
+		sph_u32 x[16]; \
+		size_t u; \
+		memcpy(x, H, sizeof x); \
+		PERM_SMALL_P(x); \
+		for (u = 0; u < 16; u ++) \
+			H[u] ^= x[u]; \
+	} while (0)
+
+#define DECL_STATE_BIG \
+	sph_u32 H[32];
+
+#define READ_STATE_BIG(sc)   do { \
+		memcpy(H, (sc)->state.narrow, sizeof H); \
+	} while (0)
+
+#define WRITE_STATE_BIG(sc)   do { \
+		memcpy((sc)->state.narrow, H, sizeof H); \
+	} while (0)
+
+#if SPH_SMALL_FOOTPRINT_GROESTL
+
+#define RBTT(d0, d1, a, b0, b1, b2, b3, b4, b5, b6, b7)   do { \
+		sph_u32 fu2 = T0up[B32_2(a[b2])]; \
+		sph_u32 fd2 = T0dn[B32_2(a[b2])]; \
+		sph_u32 fu3 = T1up[B32_3(a[b3])]; \
+		sph_u32 fd3 = T1dn[B32_3(a[b3])]; \
+		sph_u32 fu6 = T0up[B32_2(a[b6])]; \
+		sph_u32 fd6 = T0dn[B32_2(a[b6])]; \
+		sph_u32 fu7 = T1up[B32_3(a[b7])]; \
+		sph_u32 fd7 = T1dn[B32_3(a[b7])]; \
+		t[d0] = T0up[B32_0(a[b0])] \
+			^ T1up[B32_1(a[b1])] \
+			^ R32u(fu2, fd2) \
+			^ R32u(fu3, fd3) \
+			^ T0dn[B32_0(a[b4])] \
+			^ T1dn[B32_1(a[b5])] \
+			^ R32d(fu6, fd6) \
+			^ R32d(fu7, fd7); \
+		t[d1] = T0dn[B32_0(a[b0])] \
+			^ T1dn[B32_1(a[b1])] \
+			^ R32d(fu2, fd2) \
+			^ R32d(fu3, fd3) \
+			^ T0up[B32_0(a[b4])] \
+			^ T1up[B32_1(a[b5])] \
+			^ R32u(fu6, fd6) \
+			^ R32u(fu7, fd7); \
+	} while (0)
+
+#else
+
+#define RBTT(d0, d1, a, b0, b1, b2, b3, b4, b5, b6, b7)   do { \
+		t[d0] = T0up[B32_0(a[b0])] \
+			^ T1up[B32_1(a[b1])] \
+			^ T2up[B32_2(a[b2])] \
+			^ T3up[B32_3(a[b3])] \
+			^ T0dn[B32_0(a[b4])] \
+			^ T1dn[B32_1(a[b5])] \
+			^ T2dn[B32_2(a[b6])] \
+			^ T3dn[B32_3(a[b7])]; \
+		t[d1] = T0dn[B32_0(a[b0])] \
+			^ T1dn[B32_1(a[b1])] \
+			^ T2dn[B32_2(a[b2])] \
+			^ T3dn[B32_3(a[b3])] \
+			^ T0up[B32_0(a[b4])] \
+			^ T1up[B32_1(a[b5])] \
+			^ T2up[B32_2(a[b6])] \
+			^ T3up[B32_3(a[b7])]; \
+	} while (0)
+
+#endif
+
+#if SPH_SMALL_FOOTPRINT_GROESTL
+
+#define ROUND_BIG_P(a, r)   do { \
+		sph_u32 t[32]; \
+		size_t u; \
+		a[0x00] ^= PC32up(0x00, r); \
+		a[0x01] ^= PC32dn(0x00, r); \
+		a[0x02] ^= PC32up(0x10, r); \
+		a[0x03] ^= PC32dn(0x10, r); \
+		a[0x04] ^= PC32up(0x20, r); \
+		a[0x05] ^= PC32dn(0x20, r); \
+		a[0x06] ^= PC32up(0x30, r); \
+		a[0x07] ^= PC32dn(0x30, r); \
+		a[0x08] ^= PC32up(0x40, r); \
+		a[0x09] ^= PC32dn(0x40, r); \
+		a[0x0A] ^= PC32up(0x50, r); \
+		a[0x0B] ^= PC32dn(0x50, r); \
+		a[0x0C] ^= PC32up(0x60, r); \
+		a[0x0D] ^= PC32dn(0x60, r); \
+		a[0x0E] ^= PC32up(0x70, r); \
+		a[0x0F] ^= PC32dn(0x70, r); \
+		a[0x10] ^= PC32up(0x80, r); \
+		a[0x11] ^= PC32dn(0x80, r); \
+		a[0x12] ^= PC32up(0x90, r); \
+		a[0x13] ^= PC32dn(0x90, r); \
+		a[0x14] ^= PC32up(0xA0, r); \
+		a[0x15] ^= PC32dn(0xA0, r); \
+		a[0x16] ^= PC32up(0xB0, r); \
+		a[0x17] ^= PC32dn(0xB0, r); \
+		a[0x18] ^= PC32up(0xC0, r); \
+		a[0x19] ^= PC32dn(0xC0, r); \
+		a[0x1A] ^= PC32up(0xD0, r); \
+		a[0x1B] ^= PC32dn(0xD0, r); \
+		a[0x1C] ^= PC32up(0xE0, r); \
+		a[0x1D] ^= PC32dn(0xE0, r); \
+		a[0x1E] ^= PC32up(0xF0, r); \
+		a[0x1F] ^= PC32dn(0xF0, r); \
+		for (u = 0; u < 32; u += 8) { \
+			RBTT(u + 0x00, (u + 0x01) & 0x1F, a, \
+				u + 0x00, (u + 0x02) & 0x1F, \
+				(u + 0x04) & 0x1F, (u + 0x06) & 0x1F, \
+				(u + 0x09) & 0x1F, (u + 0x0B) & 0x1F, \
+				(u + 0x0D) & 0x1F, (u + 0x17) & 0x1F); \
+			RBTT(u + 0x02, (u + 0x03) & 0x1F, a, \
+				u + 0x02, (u + 0x04) & 0x1F, \
+				(u + 0x06) & 0x1F, (u + 0x08) & 0x1F, \
+				(u + 0x0B) & 0x1F, (u + 0x0D) & 0x1F, \
+				(u + 0x0F) & 0x1F, (u + 0x19) & 0x1F); \
+			RBTT(u + 0x04, (u + 0x05) & 0x1F, a, \
+				u + 0x04, (u + 0x06) & 0x1F, \
+				(u + 0x08) & 0x1F, (u + 0x0A) & 0x1F, \
+				(u + 0x0D) & 0x1F, (u + 0x0F) & 0x1F, \
+				(u + 0x11) & 0x1F, (u + 0x1B) & 0x1F); \
+			RBTT(u + 0x06, (u + 0x07) & 0x1F, a, \
+				u + 0x06, (u + 0x08) & 0x1F, \
+				(u + 0x0A) & 0x1F, (u + 0x0C) & 0x1F, \
+				(u + 0x0F) & 0x1F, (u + 0x11) & 0x1F, \
+				(u + 0x13) & 0x1F, (u + 0x1D) & 0x1F); \
+		} \
+		memcpy(a, t, sizeof t); \
+	} while (0)
+
+#define ROUND_BIG_Q(a, r)   do { \
+		sph_u32 t[32]; \
+		size_t u; \
+		a[0x00] ^= QC32up(0x00, r); \
+		a[0x01] ^= QC32dn(0x00, r); \
+		a[0x02] ^= QC32up(0x10, r); \
+		a[0x03] ^= QC32dn(0x10, r); \
+		a[0x04] ^= QC32up(0x20, r); \
+		a[0x05] ^= QC32dn(0x20, r); \
+		a[0x06] ^= QC32up(0x30, r); \
+		a[0x07] ^= QC32dn(0x30, r); \
+		a[0x08] ^= QC32up(0x40, r); \
+		a[0x09] ^= QC32dn(0x40, r); \
+		a[0x0A] ^= QC32up(0x50, r); \
+		a[0x0B] ^= QC32dn(0x50, r); \
+		a[0x0C] ^= QC32up(0x60, r); \
+		a[0x0D] ^= QC32dn(0x60, r); \
+		a[0x0E] ^= QC32up(0x70, r); \
+		a[0x0F] ^= QC32dn(0x70, r); \
+		a[0x10] ^= QC32up(0x80, r); \
+		a[0x11] ^= QC32dn(0x80, r); \
+		a[0x12] ^= QC32up(0x90, r); \
+		a[0x13] ^= QC32dn(0x90, r); \
+		a[0x14] ^= QC32up(0xA0, r); \
+		a[0x15] ^= QC32dn(0xA0, r); \
+		a[0x16] ^= QC32up(0xB0, r); \
+		a[0x17] ^= QC32dn(0xB0, r); \
+		a[0x18] ^= QC32up(0xC0, r); \
+		a[0x19] ^= QC32dn(0xC0, r); \
+		a[0x1A] ^= QC32up(0xD0, r); \
+		a[0x1B] ^= QC32dn(0xD0, r); \
+		a[0x1C] ^= QC32up(0xE0, r); \
+		a[0x1D] ^= QC32dn(0xE0, r); \
+		a[0x1E] ^= QC32up(0xF0, r); \
+		a[0x1F] ^= QC32dn(0xF0, r); \
+		for (u = 0; u < 32; u += 8) { \
+			RBTT(u + 0x00, (u + 0x01) & 0x1F, a, \
+				(u + 0x02) & 0x1F, (u + 0x06) & 0x1F, \
+				(u + 0x0A) & 0x1F, (u + 0x16) & 0x1F, \
+				(u + 0x01) & 0x1F, (u + 0x05) & 0x1F, \
+				(u + 0x09) & 0x1F, (u + 0x0D) & 0x1F); \
+			RBTT(u + 0x02, (u + 0x03) & 0x1F, a, \
+				(u + 0x04) & 0x1F, (u + 0x08) & 0x1F, \
+				(u + 0x0C) & 0x1F, (u + 0x18) & 0x1F, \
+				(u + 0x03) & 0x1F, (u + 0x07) & 0x1F, \
+				(u + 0x0B) & 0x1F, (u + 0x0F) & 0x1F); \
+			RBTT(u + 0x04, (u + 0x05) & 0x1F, a, \
+				(u + 0x06) & 0x1F, (u + 0x0A) & 0x1F, \
+				(u + 0x0E) & 0x1F, (u + 0x1A) & 0x1F, \
+				(u + 0x05) & 0x1F, (u + 0x09) & 0x1F, \
+				(u + 0x0D) & 0x1F, (u + 0x11) & 0x1F); \
+			RBTT(u + 0x06, (u + 0x07) & 0x1F, a, \
+				(u + 0x08) & 0x1F, (u + 0x0C) & 0x1F, \
+				(u + 0x10) & 0x1F, (u + 0x1C) & 0x1F, \
+				(u + 0x07) & 0x1F, (u + 0x0B) & 0x1F, \
+				(u + 0x0F) & 0x1F, (u + 0x13) & 0x1F); \
+		} \
+		memcpy(a, t, sizeof t); \
+	} while (0)
+
+#else
+
+#define ROUND_BIG_P(a, r)   do { \
+		sph_u32 t[32]; \
+		a[0x00] ^= PC32up(0x00, r); \
+		a[0x01] ^= PC32dn(0x00, r); \
+		a[0x02] ^= PC32up(0x10, r); \
+		a[0x03] ^= PC32dn(0x10, r); \
+		a[0x04] ^= PC32up(0x20, r); \
+		a[0x05] ^= PC32dn(0x20, r); \
+		a[0x06] ^= PC32up(0x30, r); \
+		a[0x07] ^= PC32dn(0x30, r); \
+		a[0x08] ^= PC32up(0x40, r); \
+		a[0x09] ^= PC32dn(0x40, r); \
+		a[0x0A] ^= PC32up(0x50, r); \
+		a[0x0B] ^= PC32dn(0x50, r); \
+		a[0x0C] ^= PC32up(0x60, r); \
+		a[0x0D] ^= PC32dn(0x60, r); \
+		a[0x0E] ^= PC32up(0x70, r); \
+		a[0x0F] ^= PC32dn(0x70, r); \
+		a[0x10] ^= PC32up(0x80, r); \
+		a[0x11] ^= PC32dn(0x80, r); \
+		a[0x12] ^= PC32up(0x90, r); \
+		a[0x13] ^= PC32dn(0x90, r); \
+		a[0x14] ^= PC32up(0xA0, r); \
+		a[0x15] ^= PC32dn(0xA0, r); \
+		a[0x16] ^= PC32up(0xB0, r); \
+		a[0x17] ^= PC32dn(0xB0, r); \
+		a[0x18] ^= PC32up(0xC0, r); \
+		a[0x19] ^= PC32dn(0xC0, r); \
+		a[0x1A] ^= PC32up(0xD0, r); \
+		a[0x1B] ^= PC32dn(0xD0, r); \
+		a[0x1C] ^= PC32up(0xE0, r); \
+		a[0x1D] ^= PC32dn(0xE0, r); \
+		a[0x1E] ^= PC32up(0xF0, r); \
+		a[0x1F] ^= PC32dn(0xF0, r); \
+		RBTT(0x00, 0x01, a, \
+			0x00, 0x02, 0x04, 0x06, 0x09, 0x0B, 0x0D, 0x17); \
+		RBTT(0x02, 0x03, a, \
+			0x02, 0x04, 0x06, 0x08, 0x0B, 0x0D, 0x0F, 0x19); \
+		RBTT(0x04, 0x05, a, \
+			0x04, 0x06, 0x08, 0x0A, 0x0D, 0x0F, 0x11, 0x1B); \
+		RBTT(0x06, 0x07, a, \
+			0x06, 0x08, 0x0A, 0x0C, 0x0F, 0x11, 0x13, 0x1D); \
+		RBTT(0x08, 0x09, a, \
+			0x08, 0x0A, 0x0C, 0x0E, 0x11, 0x13, 0x15, 0x1F); \
+		RBTT(0x0A, 0x0B, a, \
+			0x0A, 0x0C, 0x0E, 0x10, 0x13, 0x15, 0x17, 0x01); \
+		RBTT(0x0C, 0x0D, a, \
+			0x0C, 0x0E, 0x10, 0x12, 0x15, 0x17, 0x19, 0x03); \
+		RBTT(0x0E, 0x0F, a, \
+			0x0E, 0x10, 0x12, 0x14, 0x17, 0x19, 0x1B, 0x05); \
+		RBTT(0x10, 0x11, a, \
+			0x10, 0x12, 0x14, 0x16, 0x19, 0x1B, 0x1D, 0x07); \
+		RBTT(0x12, 0x13, a, \
+			0x12, 0x14, 0x16, 0x18, 0x1B, 0x1D, 0x1F, 0x09); \
+		RBTT(0x14, 0x15, a, \
+			0x14, 0x16, 0x18, 0x1A, 0x1D, 0x1F, 0x01, 0x0B); \
+		RBTT(0x16, 0x17, a, \
+			0x16, 0x18, 0x1A, 0x1C, 0x1F, 0x01, 0x03, 0x0D); \
+		RBTT(0x18, 0x19, a, \
+			0x18, 0x1A, 0x1C, 0x1E, 0x01, 0x03, 0x05, 0x0F); \
+		RBTT(0x1A, 0x1B, a, \
+			0x1A, 0x1C, 0x1E, 0x00, 0x03, 0x05, 0x07, 0x11); \
+		RBTT(0x1C, 0x1D, a, \
+			0x1C, 0x1E, 0x00, 0x02, 0x05, 0x07, 0x09, 0x13); \
+		RBTT(0x1E, 0x1F, a, \
+			0x1E, 0x00, 0x02, 0x04, 0x07, 0x09, 0x0B, 0x15); \
+		memcpy(a, t, sizeof t); \
+	} while (0)
+
+#define ROUND_BIG_Q(a, r)   do { \
+		sph_u32 t[32]; \
+		a[0x00] ^= QC32up(0x00, r); \
+		a[0x01] ^= QC32dn(0x00, r); \
+		a[0x02] ^= QC32up(0x10, r); \
+		a[0x03] ^= QC32dn(0x10, r); \
+		a[0x04] ^= QC32up(0x20, r); \
+		a[0x05] ^= QC32dn(0x20, r); \
+		a[0x06] ^= QC32up(0x30, r); \
+		a[0x07] ^= QC32dn(0x30, r); \
+		a[0x08] ^= QC32up(0x40, r); \
+		a[0x09] ^= QC32dn(0x40, r); \
+		a[0x0A] ^= QC32up(0x50, r); \
+		a[0x0B] ^= QC32dn(0x50, r); \
+		a[0x0C] ^= QC32up(0x60, r); \
+		a[0x0D] ^= QC32dn(0x60, r); \
+		a[0x0E] ^= QC32up(0x70, r); \
+		a[0x0F] ^= QC32dn(0x70, r); \
+		a[0x10] ^= QC32up(0x80, r); \
+		a[0x11] ^= QC32dn(0x80, r); \
+		a[0x12] ^= QC32up(0x90, r); \
+		a[0x13] ^= QC32dn(0x90, r); \
+		a[0x14] ^= QC32up(0xA0, r); \
+		a[0x15] ^= QC32dn(0xA0, r); \
+		a[0x16] ^= QC32up(0xB0, r); \
+		a[0x17] ^= QC32dn(0xB0, r); \
+		a[0x18] ^= QC32up(0xC0, r); \
+		a[0x19] ^= QC32dn(0xC0, r); \
+		a[0x1A] ^= QC32up(0xD0, r); \
+		a[0x1B] ^= QC32dn(0xD0, r); \
+		a[0x1C] ^= QC32up(0xE0, r); \
+		a[0x1D] ^= QC32dn(0xE0, r); \
+		a[0x1E] ^= QC32up(0xF0, r); \
+		a[0x1F] ^= QC32dn(0xF0, r); \
+		RBTT(0x00, 0x01, a, \
+			0x02, 0x06, 0x0A, 0x16, 0x01, 0x05, 0x09, 0x0D); \
+		RBTT(0x02, 0x03, a, \
+			0x04, 0x08, 0x0C, 0x18, 0x03, 0x07, 0x0B, 0x0F); \
+		RBTT(0x04, 0x05, a, \
+			0x06, 0x0A, 0x0E, 0x1A, 0x05, 0x09, 0x0D, 0x11); \
+		RBTT(0x06, 0x07, a, \
+			0x08, 0x0C, 0x10, 0x1C, 0x07, 0x0B, 0x0F, 0x13); \
+		RBTT(0x08, 0x09, a, \
+			0x0A, 0x0E, 0x12, 0x1E, 0x09, 0x0D, 0x11, 0x15); \
+		RBTT(0x0A, 0x0B, a, \
+			0x0C, 0x10, 0x14, 0x00, 0x0B, 0x0F, 0x13, 0x17); \
+		RBTT(0x0C, 0x0D, a, \
+			0x0E, 0x12, 0x16, 0x02, 0x0D, 0x11, 0x15, 0x19); \
+		RBTT(0x0E, 0x0F, a, \
+			0x10, 0x14, 0x18, 0x04, 0x0F, 0x13, 0x17, 0x1B); \
+		RBTT(0x10, 0x11, a, \
+			0x12, 0x16, 0x1A, 0x06, 0x11, 0x15, 0x19, 0x1D); \
+		RBTT(0x12, 0x13, a, \
+			0x14, 0x18, 0x1C, 0x08, 0x13, 0x17, 0x1B, 0x1F); \
+		RBTT(0x14, 0x15, a, \
+			0x16, 0x1A, 0x1E, 0x0A, 0x15, 0x19, 0x1D, 0x01); \
+		RBTT(0x16, 0x17, a, \
+			0x18, 0x1C, 0x00, 0x0C, 0x17, 0x1B, 0x1F, 0x03); \
+		RBTT(0x18, 0x19, a, \
+			0x1A, 0x1E, 0x02, 0x0E, 0x19, 0x1D, 0x01, 0x05); \
+		RBTT(0x1A, 0x1B, a, \
+			0x1C, 0x00, 0x04, 0x10, 0x1B, 0x1F, 0x03, 0x07); \
+		RBTT(0x1C, 0x1D, a, \
+			0x1E, 0x02, 0x06, 0x12, 0x1D, 0x01, 0x05, 0x09); \
+		RBTT(0x1E, 0x1F, a, \
+			0x00, 0x04, 0x08, 0x14, 0x1F, 0x03, 0x07, 0x0B); \
+		memcpy(a, t, sizeof t); \
+	} while (0)
+
+#endif
+
+#if SPH_SMALL_FOOTPRINT_GROESTL
+
+#define PERM_BIG_P(a)   do { \
+		int r; \
+		for (r = 0; r < 14; r ++) \
+			ROUND_BIG_P(a, r); \
+	} while (0)
+
+#define PERM_BIG_Q(a)   do { \
+		int r; \
+		for (r = 0; r < 14; r ++) \
+			ROUND_BIG_Q(a, r); \
+	} while (0)
+
+#else
+
+#define PERM_BIG_P(a)   do { \
+		int r; \
+		for (r = 0; r < 14; r += 2) { \
+			ROUND_BIG_P(a, r + 0); \
+			ROUND_BIG_P(a, r + 1); \
+		} \
+	} while (0)
+
+#define PERM_BIG_Q(a)   do { \
+		int r; \
+		for (r = 0; r < 14; r += 2) { \
+			ROUND_BIG_Q(a, r + 0); \
+			ROUND_BIG_Q(a, r + 1); \
+		} \
+	} while (0)
+
+#endif
+
+#define COMPRESS_BIG   do { \
+		sph_u32 g[32], m[32]; \
+		size_t u; \
+		for (u = 0; u < 32; u ++) { \
+			m[u] = dec32e_aligned(buf + (u << 2)); \
+			g[u] = m[u] ^ H[u]; \
+		} \
+		PERM_BIG_P(g); \
+		PERM_BIG_Q(m); \
+		for (u = 0; u < 32; u ++) \
+			H[u] ^= g[u] ^ m[u]; \
+	} while (0)
+
+#define FINAL_BIG   do { \
+		sph_u32 x[32]; \
+		size_t u; \
+		memcpy(x, H, sizeof x); \
+		PERM_BIG_P(x); \
+		for (u = 0; u < 32; u ++) \
+			H[u] ^= x[u]; \
+	} while (0)
+
+#endif
+
+static void
+groestl_small_init(sph_groestl_small_context *sc, unsigned out_size)
+{
+	size_t u;
+
+	sc->ptr = 0;
+#if SPH_GROESTL_64
+	for (u = 0; u < 7; u ++)
+		sc->state.wide[u] = 0;
+#if USE_LE
+	sc->state.wide[7] = ((sph_u64)(out_size & 0xFF) << 56)
+		| ((sph_u64)(out_size & 0xFF00) << 40);
+#else
+	sc->state.wide[7] = (sph_u64)out_size;
+#endif
+#else
+	for (u = 0; u < 15; u ++)
+		sc->state.narrow[u] = 0;
+#if USE_LE
+	sc->state.narrow[15] = ((sph_u32)(out_size & 0xFF) << 24)
+		| ((sph_u32)(out_size & 0xFF00) << 8);
+#else
+	sc->state.narrow[15] = (sph_u32)out_size;
+#endif
+#endif
+#if SPH_64
+	sc->count = 0;
+#else
+	sc->count_high = 0;
+	sc->count_low = 0;
+#endif
+}
+
+static void
+groestl_small_core(sph_groestl_small_context *sc, const void *data, size_t len)
+{
+	unsigned char *buf;
+	size_t ptr;
+	DECL_STATE_SMALL
+
+	buf = sc->buf;
+	ptr = sc->ptr;
+	if (len < (sizeof sc->buf) - ptr) {
+		memcpy(buf + ptr, data, len);
+		ptr += len;
+		sc->ptr = ptr;
+		return;
+	}
+
+	READ_STATE_SMALL(sc);
+	while (len > 0) {
+		size_t clen;
+
+		clen = (sizeof sc->buf) - ptr;
+		if (clen > len)
+			clen = len;
+		memcpy(buf + ptr, data, clen);
+		ptr += clen;
+		data = (const unsigned char *)data + clen;
+		len -= clen;
+		if (ptr == sizeof sc->buf) {
+			COMPRESS_SMALL;
+#if SPH_64
+			sc->count ++;
+#else
+			if ((sc->count_low = SPH_T32(sc->count_low + 1)) == 0)
+				sc->count_high = SPH_T32(sc->count_high + 1);
+#endif
+			ptr = 0;
+		}
+	}
+	WRITE_STATE_SMALL(sc);
+	sc->ptr = ptr;
+}
+
+static void
+groestl_small_close(sph_groestl_small_context *sc,
+	unsigned ub, unsigned n, void *dst, size_t out_len)
+{
+	unsigned char *buf;
+	unsigned char pad[72];
+	size_t u, ptr, pad_len;
+#if SPH_64
+	sph_u64 count;
+#else
+	sph_u32 count_high, count_low;
+#endif
+	unsigned z;
+	DECL_STATE_SMALL
+
+	buf = sc->buf;
+	ptr = sc->ptr;
+	z = 0x80 >> n;
+	pad[0] = ((ub & -z) | z) & 0xFF;
+	if (ptr < 56) {
+		pad_len = 64 - ptr;
+#if SPH_64
+		count = SPH_T64(sc->count + 1);
+#else
+		count_low = SPH_T32(sc->count_low + 1);
+		count_high = SPH_T32(sc->count_high);
+		if (count_low == 0)
+			count_high = SPH_T32(count_high + 1);
+#endif
+	} else {
+		pad_len = 128 - ptr;
+#if SPH_64
+		count = SPH_T64(sc->count + 2);
+#else
+		count_low = SPH_T32(sc->count_low + 2);
+		count_high = SPH_T32(sc->count_high);
+		if (count_low <= 1)
+			count_high = SPH_T32(count_high + 1);
+#endif
+	}
+	memset(pad + 1, 0, pad_len - 9);
+#if SPH_64
+	sph_enc64be(pad + pad_len - 8, count);
+#else
+	sph_enc64be(pad + pad_len - 8, count_high);
+	sph_enc64be(pad + pad_len - 4, count_low);
+#endif
+	groestl_small_core(sc, pad, pad_len);
+	READ_STATE_SMALL(sc);
+	FINAL_SMALL;
+#if SPH_GROESTL_64
+	for (u = 0; u < 4; u ++)
+		enc64e(pad + (u << 3), H[u + 4]);
+#else
+	for (u = 0; u < 8; u ++)
+		enc32e(pad + (u << 2), H[u + 8]);
+#endif
+	memcpy(dst, pad + 32 - out_len, out_len);
+	groestl_small_init(sc, (unsigned)out_len << 3);
+}
+
+static void
+groestl_big_init(sph_groestl_big_context *sc, unsigned out_size)
+{
+	size_t u;
+
+	sc->ptr = 0;
+#if SPH_GROESTL_64
+	for (u = 0; u < 15; u ++)
+		sc->state.wide[u] = 0;
+#if USE_LE
+	sc->state.wide[15] = ((sph_u64)(out_size & 0xFF) << 56)
+		| ((sph_u64)(out_size & 0xFF00) << 40);
+#else
+	sc->state.wide[15] = (sph_u64)out_size;
+#endif
+#else
+	for (u = 0; u < 31; u ++)
+		sc->state.narrow[u] = 0;
+#if USE_LE
+	sc->state.narrow[31] = ((sph_u32)(out_size & 0xFF) << 24)
+		| ((sph_u32)(out_size & 0xFF00) << 8);
+#else
+	sc->state.narrow[31] = (sph_u32)out_size;
+#endif
+#endif
+#if SPH_64
+	sc->count = 0;
+#else
+	sc->count_high = 0;
+	sc->count_low = 0;
+#endif
+}
+
+static void
+groestl_big_core(sph_groestl_big_context *sc, const void *data, size_t len)
+{
+	unsigned char *buf;
+	size_t ptr;
+	DECL_STATE_BIG
+
+	buf = sc->buf;
+	ptr = sc->ptr;
+	if (len < (sizeof sc->buf) - ptr) {
+		memcpy(buf + ptr, data, len);
+		ptr += len;
+		sc->ptr = ptr;
+		return;
+	}
+
+	READ_STATE_BIG(sc);
+	while (len > 0) {
+		size_t clen;
+
+		clen = (sizeof sc->buf) - ptr;
+		if (clen > len)
+			clen = len;
+		memcpy(buf + ptr, data, clen);
+		ptr += clen;
+		data = (const unsigned char *)data + clen;
+		len -= clen;
+		if (ptr == sizeof sc->buf) {
+			COMPRESS_BIG;
+#if SPH_64
+			sc->count ++;
+#else
+			if ((sc->count_low = SPH_T32(sc->count_low + 1)) == 0)
+				sc->count_high = SPH_T32(sc->count_high + 1);
+#endif
+			ptr = 0;
+		}
+	}
+	WRITE_STATE_BIG(sc);
+	sc->ptr = ptr;
+}
+
+static void
+groestl_big_close(sph_groestl_big_context *sc,
+	unsigned ub, unsigned n, void *dst, size_t out_len)
+{
+	unsigned char *buf;
+	unsigned char pad[136];
+	size_t ptr, pad_len, u;
+#if SPH_64
+	sph_u64 count;
+#else
+	sph_u32 count_high, count_low;
+#endif
+	unsigned z;
+	DECL_STATE_BIG
+
+	buf = sc->buf;
+	ptr = sc->ptr;
+	z = 0x80 >> n;
+	pad[0] = ((ub & -z) | z) & 0xFF;
+	if (ptr < 120) {
+		pad_len = 128 - ptr;
+#if SPH_64
+		count = SPH_T64(sc->count + 1);
+#else
+		count_low = SPH_T32(sc->count_low + 1);
+		count_high = SPH_T32(sc->count_high);
+		if (count_low == 0)
+			count_high = SPH_T32(count_high + 1);
+#endif
+	} else {
+		pad_len = 256 - ptr;
+#if SPH_64
+		count = SPH_T64(sc->count + 2);
+#else
+		count_low = SPH_T32(sc->count_low + 2);
+		count_high = SPH_T32(sc->count_high);
+		if (count_low <= 1)
+			count_high = SPH_T32(count_high + 1);
+#endif
+	}
+	memset(pad + 1, 0, pad_len - 9);
+#if SPH_64
+	sph_enc64be(pad + pad_len - 8, count);
+#else
+	sph_enc64be(pad + pad_len - 8, count_high);
+	sph_enc64be(pad + pad_len - 4, count_low);
+#endif
+	groestl_big_core(sc, pad, pad_len);
+	READ_STATE_BIG(sc);
+	FINAL_BIG;
+#if SPH_GROESTL_64
+	for (u = 0; u < 8; u ++)
+		enc64e(pad + (u << 3), H[u + 8]);
+#else
+	for (u = 0; u < 16; u ++)
+		enc32e(pad + (u << 2), H[u + 16]);
+#endif
+	memcpy(dst, pad + 64 - out_len, out_len);
+	groestl_big_init(sc, (unsigned)out_len << 3);
+}
+
+/* see sph_groestl.h */
+void
+sph_groestl224_init(void *cc)
+{
+	groestl_small_init(cc, 224);
+}
+
+/* see sph_groestl.h */
+void
+sph_groestl224(void *cc, const void *data, size_t len)
+{
+	groestl_small_core(cc, data, len);
+}
+
+/* see sph_groestl.h */
+void
+sph_groestl224_close(void *cc, void *dst)
+{
+	groestl_small_close(cc, 0, 0, dst, 28);
+}
+
+/* see sph_groestl.h */
+void
+sph_groestl224_addbits_and_close(void *cc, unsigned ub, unsigned n, void *dst)
+{
+	groestl_small_close(cc, ub, n, dst, 28);
+}
+
+/* see sph_groestl.h */
+void
+sph_groestl256_init(void *cc)
+{
+	groestl_small_init(cc, 256);
+}
+
+/* see sph_groestl.h */
+void
+sph_groestl256(void *cc, const void *data, size_t len)
+{
+	groestl_small_core(cc, data, len);
+}
+
+/* see sph_groestl.h */
+void
+sph_groestl256_close(void *cc, void *dst)
+{
+	groestl_small_close(cc, 0, 0, dst, 32);
+}
+
+/* see sph_groestl.h */
+void
+sph_groestl256_addbits_and_close(void *cc, unsigned ub, unsigned n, void *dst)
+{
+	groestl_small_close(cc, ub, n, dst, 32);
+}
+
+/* see sph_groestl.h */
+void
+sph_groestl384_init(void *cc)
+{
+	groestl_big_init(cc, 384);
+}
+
+/* see sph_groestl.h */
+void
+sph_groestl384(void *cc, const void *data, size_t len)
+{
+	groestl_big_core(cc, data, len);
+}
+
+/* see sph_groestl.h */
+void
+sph_groestl384_close(void *cc, void *dst)
+{
+	groestl_big_close(cc, 0, 0, dst, 48);
+}
+
+/* see sph_groestl.h */
+void
+sph_groestl384_addbits_and_close(void *cc, unsigned ub, unsigned n, void *dst)
+{
+	groestl_big_close(cc, ub, n, dst, 48);
+}
+
+/* see sph_groestl.h */
+void
+sph_groestl512_init(void *cc)
+{
+	groestl_big_init(cc, 512);
+}
+
+/* see sph_groestl.h */
+void
+sph_groestl512(void *cc, const void *data, size_t len)
+{
+	groestl_big_core(cc, data, len);
+}
+
+/* see sph_groestl.h */
+void
+sph_groestl512_close(void *cc, void *dst)
+{
+	groestl_big_close(cc, 0, 0, dst, 64);
+}
+
+/* see sph_groestl.h */
+void
+sph_groestl512_addbits_and_close(void *cc, unsigned ub, unsigned n, void *dst)
+{
+	groestl_big_close(cc, ub, n, dst, 64);
+}
+
+#ifdef __cplusplus
+}
+#endif
diff --git a/heavy.cu b/heavy.cu
new file mode 100644
index 0000000000..6ce1a663d2
--- /dev/null
+++ b/heavy.cu
@@ -0,0 +1,416 @@
+#include <string.h>
+#include <openssl/sha.h>
+#include <cuda.h>
+#include "cuda_runtime.h"
+#include "device_launch_parameters.h"
+#include <map>
+
+#ifndef _WIN32
+#include <unistd.h>
+#endif
+
+// include thrust
+#include <thrust/version.h>
+#include <thrust/remove.h>
+#include <thrust/device_vector.h>
+#include <thrust/iterator/constant_iterator.h>
+
+#include "miner.h"
+
+#include "hefty1.h"
+#include "sph_keccak.h"
+#include "sph_blake.h"
+#include "sph_groestl.h"
+
+#include "cuda_hefty1.h"
+#include "cuda_sha256.h"
+#include "cuda_keccak512.h"
+#include "cuda_groestl512.h"
+#include "cuda_blake512.h"
+#include "cuda_combine.h"
+
+extern uint32_t *d_hash2output[8];
+extern uint32_t *d_hash3output[8];
+extern uint32_t *d_hash4output[8];
+extern uint32_t *d_hash5output[8];
+
+#define HEAVYCOIN_BLKHDR_SZ        84
+
+// nonce-array für die threads
+uint32_t *d_nonceVector[8];
+
+/* Combines top 64-bits from each hash into a single hash */
+static void combine_hashes(uint32_t *out, const uint32_t *hash1, const uint32_t *hash2, const uint32_t *hash3, const uint32_t *hash4)
+{
+    const uint32_t *hash[4] = { hash1, hash2, hash3, hash4 };
+    int bits;
+    unsigned int i;
+    uint32_t mask;
+    unsigned int k;
+
+    /* Transpose first 64 bits of each hash into out */
+    memset(out, 0, 32);
+    bits = 0;
+    for (i = 7; i >= 6; i--) {
+        for (mask = 0x80000000; mask; mask >>= 1) {
+            for (k = 0; k < 4; k++) {
+                out[(255 - bits)/32] <<= 1;
+                if ((hash[k][i] & mask) != 0)
+                    out[(255 - bits)/32] |= 1;
+                bits++;
+            }
+        }
+    }
+}
+
+#ifdef _MSC_VER
+#include <intrin.h>
+static uint32_t __inline bitsset( uint32_t x )
+{
+   DWORD r = 0;
+   _BitScanReverse(&r, x);
+   return r;
+}
+#else
+static uint32_t bitsset( uint32_t x )
+{
+    return 31-__builtin_clz(x);
+}
+#endif
+
+// Finde das high bit in einem Multiword-Integer.
+static int findhighbit(const uint32_t *ptarget, int words)
+{
+    int i;
+    int highbit = 0;
+    for (i=words-1; i >= 0; --i)
+    {
+        if (ptarget[i] != 0) {
+            highbit = i*32 + bitsset(ptarget[i])+1;
+                break;
+        }
+    }
+    return highbit;
+}
+
+// Generiere ein Multiword-Integer das die Zahl
+// (2 << highbit) - 1 repräsentiert.
+static void genmask(uint32_t *ptarget, int words, int highbit)
+{
+    int i;
+    for (i=words-1; i >= 0; --i)
+    {
+        if ((i+1)*32 <= highbit)
+            ptarget[i] = 0xffffffff;
+        else if (i*32 > highbit)
+            ptarget[i] = 0x00000000;
+        else
+            ptarget[i] = (1 << (highbit-i*32)) - 1;
+    }
+}
+
+struct check_nonce_for_remove
+{    
+    check_nonce_for_remove(uint64_t target, uint32_t *hashes, uint32_t hashlen, uint32_t startNonce) :
+        m_target(target),
+        m_hashes(hashes),
+        m_hashlen(hashlen),
+        m_startNonce(startNonce) { }
+
+    __device__
+    bool operator()(const uint32_t x)
+    {
+        // Position im Hash Buffer
+        uint32_t hashIndex = x - m_startNonce;
+        // Wert des Hashes (als uint64_t) auslesen.
+        // Steht im 6. und 7. Wort des Hashes (jeder dieser Hashes hat 512 Bits)
+        uint64_t hashValue = *((uint64_t*)(&m_hashes[m_hashlen*hashIndex + 6]));
+        // gegen das Target prüfen. Es dürfen nur Bits aus dem Target gesetzt sein.
+        return (hashValue & m_target) != hashValue;
+    }
+
+    uint64_t  m_target;
+    uint32_t *m_hashes;
+    uint32_t  m_hashlen;
+    uint32_t  m_startNonce;
+};
+
+// Zahl der CUDA Devices im System bestimmen
+extern "C" int cuda_num_devices()
+{
+    int version;
+    cudaError_t err = cudaDriverGetVersion(&version);
+    if (err != cudaSuccess)
+    {
+        applog(LOG_ERR, "Unable to query CUDA driver version! Is an nVidia driver installed?");
+        exit(1);
+    }
+
+    int maj = version / 1000, min = version % 100; // same as in deviceQuery sample
+    if (maj < 5 || (maj == 5 && min < 5))
+    {
+        applog(LOG_ERR, "Driver does not support CUDA %d.%d API! Update your nVidia driver!", 5, 5);
+        exit(1);
+    }
+
+    int GPU_N;
+    err = cudaGetDeviceCount(&GPU_N);
+    if (err != cudaSuccess)
+    {
+        applog(LOG_ERR, "Unable to query number of CUDA devices! Is an nVidia driver installed?");
+        exit(1);
+    }
+    return GPU_N;
+}
+
+// Zeitsynchronisations-Routine von cudaminer mit CPU sleep
+typedef struct { double value[8]; } tsumarray;
+cudaError_t MyStreamSynchronize(cudaStream_t stream, int situation, int thr_id)
+{
+    cudaError_t result = cudaSuccess;
+    if (situation >= 0)
+    {   
+        static std::map<int, tsumarray> tsum;
+
+        double a = 0.95, b = 0.05;
+        if (tsum.find(situation) == tsum.end()) { a = 0.5; b = 0.5; } // faster initial convergence
+
+        double tsync = 0.0;
+        double tsleep = 0.95 * tsum[situation].value[thr_id];
+        if (cudaStreamQuery(stream) == cudaErrorNotReady)
+        {
+            usleep((useconds_t)(1e6*tsleep));
+            struct timeval tv_start, tv_end;
+            gettimeofday(&tv_start, NULL);
+            result = cudaStreamSynchronize(stream);
+            gettimeofday(&tv_end, NULL);
+            tsync = 1e-6 * (tv_end.tv_usec-tv_start.tv_usec) + (tv_end.tv_sec-tv_start.tv_sec);
+        }
+        if (tsync >= 0) tsum[situation].value[thr_id] = a * tsum[situation].value[thr_id] + b * (tsleep+tsync);
+    }
+    else
+        result = cudaStreamSynchronize(stream);
+    return result;
+}
+
+int scanhash_heavy_cpp(int thr_id, uint32_t *pdata,
+ const uint32_t *ptarget, uint32_t max_nonce,
+ unsigned long *hashes_done, uint32_t maxvote);
+
+extern "C"
+int scanhash_heavy(int thr_id, uint32_t *pdata,
+ const uint32_t *ptarget, uint32_t max_nonce,
+ unsigned long *hashes_done, uint32_t maxvote)
+{
+ return scanhash_heavy_cpp(thr_id, pdata,
+  ptarget, max_nonce, hashes_done, maxvote);
+}
+
+int scanhash_heavy_cpp(int thr_id, uint32_t *pdata,
+ const uint32_t *ptarget, uint32_t max_nonce,
+ unsigned long *hashes_done, uint32_t maxvote)
+{
+    // CUDA will process thousands of threads.
+    const int throughput = 4096 * 128;
+
+    int rc = 0;
+    uint32_t *hash = NULL;
+    cudaMallocHost(&hash, throughput*8*sizeof(uint32_t));
+    uint32_t *cpu_nonceVector = NULL;
+    cudaMallocHost(&cpu_nonceVector, throughput*sizeof(uint32_t));
+
+    int nrmCalls[6];
+    memset(nrmCalls, 0, sizeof(int) * 6);
+
+    uint32_t start_nonce = pdata[19];    
+    uint16_t *ext = (uint16_t *)&pdata[20];
+
+    // für jeden Hash ein individuelles Target erstellen basierend
+    // auf dem höchsten Bit, das in ptarget gesetzt ist.
+    int highbit = findhighbit(ptarget, 8);
+    uint32_t target2[2], target3[2], target4[2], target5[2];
+    genmask(target2, 2, highbit/4+(((highbit%4)>3)?1:0) ); // SHA256
+    genmask(target3, 2, highbit/4+(((highbit%4)>2)?1:0) ); // keccak512
+    genmask(target4, 2, highbit/4+(((highbit%4)>1)?1:0) ); // groestl512
+    genmask(target5, 2, highbit/4+(((highbit%4)>0)?1:0) ); // blake512
+
+    static bool init[8] = {0,0,0,0,0,0,0,0};
+    if (!init[thr_id])
+    {
+        hefty_cpu_init(thr_id, throughput);
+        sha256_cpu_init(thr_id, throughput);
+        keccak512_cpu_init(thr_id, throughput);
+        groestl512_cpu_init(thr_id, throughput);
+        blake512_cpu_init(thr_id, throughput);
+        combine_cpu_init(thr_id, throughput);
+        init[thr_id] = true;
+        cudaMalloc(&d_nonceVector[thr_id], sizeof(uint32_t) * throughput);
+    }
+
+
+    if (opt_vote > maxvote) {
+        printf("Warning: Your block reward vote (%hu) exceeds "
+                "the maxvote reported by the pool (%hu).\n",
+                opt_vote, maxvote);
+    }
+
+    if (opt_trust_pool && opt_vote > maxvote) {
+        printf("Warning: Capping block reward vote to maxvote reported by pool.\n");
+        ext[0] = maxvote;
+    }
+    else
+        ext[0] = opt_vote;
+
+    // Setze die Blockdaten
+    hefty_cpu_setBlock(thr_id, throughput, pdata);
+    sha256_cpu_setBlock(pdata);
+    keccak512_cpu_setBlock(pdata);
+    groestl512_cpu_setBlock(pdata);
+    blake512_cpu_setBlock(pdata);
+
+    do {
+        int i;
+
+        ////// Compaction init
+        thrust::device_ptr<uint32_t> devNoncePtr(d_nonceVector[thr_id]);
+        thrust::device_ptr<uint32_t> devNoncePtrEnd((d_nonceVector[thr_id]) + throughput);
+        uint32_t actualNumberOfValuesInNonceVectorGPU = throughput;
+
+        hefty_cpu_hash(thr_id, throughput, pdata[19]);
+        //cudaThreadSynchronize();
+        sha256_cpu_hash(thr_id, throughput, pdata[19]);
+        //cudaThreadSynchronize();
+
+        // Hier ist die längste CPU Wartephase. Deshalb ein strategisches MyStreamSynchronize() hier.
+        MyStreamSynchronize(NULL, 0, thr_id);
+
+        ////// Compaction
+        devNoncePtrEnd = thrust::remove_if(devNoncePtr, devNoncePtrEnd, check_nonce_for_remove(*((uint64_t*)target2), d_hash2output[thr_id], 8, pdata[19]));
+        actualNumberOfValuesInNonceVectorGPU = (uint32_t)(devNoncePtrEnd - devNoncePtr);
+        if(actualNumberOfValuesInNonceVectorGPU == 0)
+            goto emptyNonceVector;
+        
+        keccak512_cpu_hash(thr_id, actualNumberOfValuesInNonceVectorGPU, pdata[19]);
+        //cudaThreadSynchronize();
+
+        ////// Compaction
+        devNoncePtrEnd = thrust::remove_if(devNoncePtr, devNoncePtrEnd, check_nonce_for_remove(*((uint64_t*)target3), d_hash3output[thr_id], 16, pdata[19]));
+        actualNumberOfValuesInNonceVectorGPU = (uint32_t)(devNoncePtrEnd - devNoncePtr);
+        if(actualNumberOfValuesInNonceVectorGPU == 0)
+            goto emptyNonceVector;
+
+        blake512_cpu_hash(thr_id, actualNumberOfValuesInNonceVectorGPU, pdata[19]);
+        //cudaThreadSynchronize();
+
+        ////// Compaction
+        devNoncePtrEnd = thrust::remove_if(devNoncePtr, devNoncePtrEnd, check_nonce_for_remove(*((uint64_t*)target5), d_hash5output[thr_id], 16, pdata[19]));
+        actualNumberOfValuesInNonceVectorGPU = (uint32_t)(devNoncePtrEnd - devNoncePtr);
+        if(actualNumberOfValuesInNonceVectorGPU == 0)
+            goto emptyNonceVector;
+
+        groestl512_cpu_hash(thr_id, actualNumberOfValuesInNonceVectorGPU, pdata[19]);
+        //cudaThreadSynchronize();
+
+        ////// Compaction
+        devNoncePtrEnd = thrust::remove_if(devNoncePtr, devNoncePtrEnd, check_nonce_for_remove(*((uint64_t*)target4), d_hash4output[thr_id], 16, pdata[19]));
+        actualNumberOfValuesInNonceVectorGPU = (uint32_t)(devNoncePtrEnd - devNoncePtr);
+        if(actualNumberOfValuesInNonceVectorGPU == 0)
+            goto emptyNonceVector;
+        
+        // combine
+        combine_cpu_hash(thr_id, actualNumberOfValuesInNonceVectorGPU, pdata[19], hash);
+
+        // Ergebnisse kopieren
+        if(actualNumberOfValuesInNonceVectorGPU > 0)
+        {
+            cudaMemcpy(cpu_nonceVector, d_nonceVector[thr_id], sizeof(uint32_t) * actualNumberOfValuesInNonceVectorGPU, cudaMemcpyDeviceToHost);
+        
+            for (i=0; i<actualNumberOfValuesInNonceVectorGPU;++i)
+            {
+                uint32_t nonce = cpu_nonceVector[i];
+                //uint32_t index = nonce - pdata[19];
+                uint32_t index = i;
+                uint32_t *foundhash = &hash[8*index];
+                if (foundhash[7] <= ptarget[7]) {
+                    if (fulltest(foundhash, ptarget)) {
+                        uint32_t verification[8];
+                        pdata[19] += nonce - pdata[19];
+                        heavycoin_hash((unsigned char *)verification, (const unsigned char *)pdata, HEAVYCOIN_BLKHDR_SZ);
+                        if (memcmp(verification, foundhash, 8*sizeof(uint32_t))) {
+                            applog(LOG_ERR, "hash for nonce=$%08X does not validate on CPU!\n", nonce);
+                        }
+                        else
+                        {
+                            *hashes_done = pdata[19] - start_nonce;
+                            rc = 1;
+                            goto exit;
+                        }
+                    }
+                }
+            }
+        }
+
+emptyNonceVector:
+
+        pdata[19] += throughput;
+
+    } while (pdata[19] < max_nonce && !work_restart[thr_id].restart);
+    *hashes_done = pdata[19] - start_nonce;
+
+exit:
+    cudaFreeHost(cpu_nonceVector);
+    cudaFreeHost(hash);
+    return rc;
+}
+
+void heavycoin_hash(unsigned char* output, const unsigned char* input, int len)
+{
+    unsigned char hash1[32];
+    unsigned char hash2[32];
+    uint32_t hash3[16];
+    uint32_t hash4[16];
+    uint32_t hash5[16];
+    uint32_t *final;
+    SHA256_CTX ctx;
+    sph_keccak512_context keccakCtx;
+    sph_groestl512_context groestlCtx;
+    sph_blake512_context blakeCtx;
+
+    HEFTY1(input, len, hash1);
+
+    /* HEFTY1 is new, so take an extra security measure to eliminate
+     * the possiblity of collisions:
+     *
+     *     Hash(x) = SHA256(x + HEFTY1(x))
+     *
+     * N.B. '+' is concatenation.
+     */
+    SHA256_Init(&ctx);
+    SHA256_Update(&ctx, input, len);
+    SHA256_Update(&ctx, hash1, sizeof(hash1));
+    SHA256_Final(hash2, &ctx);
+
+    /* Additional security: Do not rely on a single cryptographic hash
+     * function.  Instead, combine the outputs of 4 of the most secure
+     * cryptographic hash functions-- SHA256, KECCAK512, GROESTL512
+     * and BLAKE512.
+     */
+
+    sph_keccak512_init(&keccakCtx);
+    sph_keccak512(&keccakCtx, input, len);
+    sph_keccak512(&keccakCtx, hash1, sizeof(hash1));
+    sph_keccak512_close(&keccakCtx, (void *)&hash3);
+
+    sph_groestl512_init(&groestlCtx);
+    sph_groestl512(&groestlCtx, input, len);
+    sph_groestl512(&groestlCtx, hash1, sizeof(hash1));
+    sph_groestl512_close(&groestlCtx, (void *)&hash4);
+
+    sph_blake512_init(&blakeCtx);
+    sph_blake512(&blakeCtx, input, len);
+    sph_blake512(&blakeCtx, (unsigned char *)&hash1, sizeof(hash1));
+    sph_blake512_close(&blakeCtx, (void *)&hash5);
+
+    final = (uint32_t *)output;
+    combine_hashes(final, (uint32_t *)hash2, hash3, hash4, hash5);
+}
diff --git a/hefty1.c b/hefty1.c
new file mode 100644
index 0000000000..ca20db8eb7
--- /dev/null
+++ b/hefty1.c
@@ -0,0 +1,371 @@
+/*
+ * HEFTY1 CPU-only cryptographic hash function
+ *
+ * Copyright (c) 2014, dbcc14 <BM-NBx4AKznJuyem3dArgVY8MGyABpihRy5>
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *
+ * 1. Redistributions of source code must retain the above copyright notice, this
+ *    list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright notice,
+ *    this list of conditions and the following disclaimer in the documentation
+ *    and/or other materials provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
+ * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ * DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
+ * ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ * The views and conclusions contained in the software and documentation are those
+ * of the authors and should not be interpreted as representing official policies,
+ * either expressed or implied, of the FreeBSD Project.
+ */
+
+#include <assert.h>
+#include <string.h>
+
+#include "hefty1.h"
+
+#ifdef WIN32
+#define inline __inline
+#endif
+
+#define Min(A, B) (A <= B ? A : B)
+
+#define RoundFunc(ctx, A, B, C, D, E, F, G, H, W, K)                    \
+    {                                                                   \
+        /* To thwart parallelism, Br modifies itself each time it's     \
+         * called.  This also means that calling it in different        \
+         * orders yeilds different results.  In C the order of          \
+         * evaluation of function arguments and + operands are          \
+         * unspecified (and depends on the compiler), so we must make   \
+         * the order of Br calls explicit.                              \
+         */                                                             \
+        uint32_t brG = Br(ctx, G);                                      \
+        uint32_t tmp1 = Ch(E, Br(ctx, F), brG) + H + W + K;             \
+        uint32_t tmp2 = tmp1 + Sigma1(Br(ctx, E));                      \
+        uint32_t brC = Br(ctx, C);                                      \
+        uint32_t brB = Br(ctx, B);                                      \
+        uint32_t tmp3 = Ma(Br(ctx, A), brB, brC);                       \
+        uint32_t tmp4 = tmp3 + Sigma0(Br(ctx, A));                      \
+        H = G;                                                          \
+        G = F;                                                          \
+        F = E;                                                          \
+        E = D + Br(ctx, tmp2);                                          \
+        D = C;                                                          \
+        C = B;                                                          \
+        B = A;                                                          \
+        A = tmp2 + tmp4;                                                \
+    }                                                                   \
+
+/* Nothing up my sleeve constants */
+const static uint32_t K[64] = {
+    0x428a2f98UL, 0x71374491UL, 0xb5c0fbcfUL, 0xe9b5dba5UL,
+    0x3956c25bUL, 0x59f111f1UL, 0x923f82a4UL, 0xab1c5ed5UL,
+    0xd807aa98UL, 0x12835b01UL, 0x243185beUL, 0x550c7dc3UL,
+    0x72be5d74UL, 0x80deb1feUL, 0x9bdc06a7UL, 0xc19bf174UL,
+    0xe49b69c1UL, 0xefbe4786UL, 0x0fc19dc6UL, 0x240ca1ccUL,
+    0x2de92c6fUL, 0x4a7484aaUL, 0x5cb0a9dcUL, 0x76f988daUL,
+    0x983e5152UL, 0xa831c66dUL, 0xb00327c8UL, 0xbf597fc7UL,
+    0xc6e00bf3UL, 0xd5a79147UL, 0x06ca6351UL, 0x14292967UL,
+    0x27b70a85UL, 0x2e1b2138UL, 0x4d2c6dfcUL, 0x53380d13UL,
+    0x650a7354UL, 0x766a0abbUL, 0x81c2c92eUL, 0x92722c85UL,
+    0xa2bfe8a1UL, 0xa81a664bUL, 0xc24b8b70UL, 0xc76c51a3UL,
+    0xd192e819UL, 0xd6990624UL, 0xf40e3585UL, 0x106aa070UL,
+    0x19a4c116UL, 0x1e376c08UL, 0x2748774cUL, 0x34b0bcb5UL,
+    0x391c0cb3UL, 0x4ed8aa4aUL, 0x5b9cca4fUL, 0x682e6ff3UL,
+    0x748f82eeUL, 0x78a5636fUL, 0x84c87814UL, 0x8cc70208UL,
+    0x90befffaUL, 0xa4506cebUL, 0xbef9a3f7UL, 0xc67178f2UL
+};
+
+/* Initial hash values */
+const static uint32_t H[HEFTY1_STATE_WORDS] = {
+    0x6a09e667UL,
+    0xbb67ae85UL,
+    0x3c6ef372UL,
+    0xa54ff53aUL,
+    0x510e527fUL,
+    0x9b05688cUL,
+    0x1f83d9abUL,
+    0x5be0cd19UL
+};
+
+static inline uint32_t Rr(uint32_t X, uint8_t n)
+{
+    return (X >> n) | (X << (32 - n));
+}
+
+static inline uint32_t Ch(uint32_t E, uint32_t F, uint32_t G)
+{
+    return (E & F) ^ (~E & G);
+}
+
+static inline uint32_t Sigma1(uint32_t E)
+{
+    return Rr(E, 6) ^ Rr(E, 11) ^ Rr(E, 25);
+}
+
+static inline uint32_t sigma1(uint32_t X)
+{
+    return Rr(X, 17) ^ Rr(X, 19) ^ (X >> 10);
+}
+
+static inline uint32_t Ma(uint32_t A, uint32_t B, uint32_t C)
+{
+    return (A & B) ^ (A & C) ^ (B & C);
+}
+
+static inline uint32_t Sigma0(uint32_t A)
+{
+    return Rr(A, 2) ^ Rr(A, 13) ^ Rr(A, 22);
+}
+
+static inline uint32_t sigma0(uint32_t X)
+{
+    return Rr(X, 7) ^ Rr(X, 18) ^ (X >> 3);
+}
+
+static inline uint32_t Reverse32(uint32_t n)
+{
+    #if BYTE_ORDER == LITTLE_ENDIAN
+        return n << 24 | (n & 0x0000ff00) << 8 | (n & 0x00ff0000) >> 8 | n >> 24;
+    #else
+        return n;
+    #endif
+}
+
+static inline uint64_t Reverse64(uint64_t n)
+{
+    #if BYTE_ORDER == LITTLE_ENDIAN
+        uint32_t a = n >> 32;
+        uint32_t b = (n << 32) >> 32;
+
+        return (uint64_t)Reverse32(b) << 32 | Reverse32(a);
+    #else
+        return n;
+    #endif
+}
+
+/* Smoosh byte into nibble */
+static inline uint8_t Smoosh4(uint8_t X)
+{
+    return (X >> 4) ^ (X & 0xf);
+}
+
+/* Smoosh 32-bit word into 2-bits */
+static inline uint8_t Smoosh2(uint32_t X)
+{
+    uint16_t w = (X >> 16) ^ (X & 0xffff);
+    uint8_t n = Smoosh4((w >> 8) ^ (w & 0xff));
+    return (n >> 2) ^ (n & 0x3);
+}
+#include <stdio.h>
+static void Mangle(uint32_t *S)
+{
+    uint8_t r0 = Smoosh4(S[0] >> 24);
+    uint8_t r1 = Smoosh4(S[0] >> 16);
+    uint8_t r2 = Smoosh4(S[0] >> 8);
+    uint8_t r3 = Smoosh4(S[0] & 0xff);
+
+    /* Diffuse */
+    S[1] ^= Rr(S[0], r0);
+    switch (Smoosh2(S[1])) {
+      case 0: S[2] ^= Rr(S[0], 1 + r0); break;
+      case 1: S[2] += Rr(~S[0], 1 + r1); break;
+      case 2: S[2] &= Rr(~S[0], 1 + r2); break;
+      case 3: S[2] ^= Rr(S[0], 1 + r3); break;
+    }
+    switch (Smoosh2(S[1] ^ S[2])) {
+      case 0: S[3] ^= Rr(S[0], 2 + r0); break;
+      case 1: S[3] += Rr(~S[0], 2 + r1); break;
+      case 2: S[3] &= Rr(~S[0], 2 + r2); break;
+      case 3: S[3] ^= Rr(S[0], 2 + r3); break;
+    }
+
+    /* Compress */
+    S[0] ^= (S[1] ^ S[2]) + S[3];
+}
+
+static void Absorb(uint32_t *S, uint32_t X)
+{
+    uint32_t *R = S;
+    R[0] ^= X;
+    Mangle(S);
+}
+
+static uint32_t Squeeze(uint32_t *S)
+{
+    uint32_t Y = S[0];
+    Mangle(S);
+    return Y;
+}
+
+/* Branch, compress and serialize function */
+static inline uint32_t Br(HEFTY1_CTX *ctx, uint32_t X)
+{
+    uint32_t R = Squeeze(ctx->sponge);
+
+    uint8_t r0 = R >> 8;
+    uint8_t r1 = R & 0xff;
+
+    uint32_t Y = 1 << (r0 % 32);
+
+    switch (r1 % 4)
+    {
+    case 0:
+        /* Do nothing */
+        break;
+    case 1:
+        return X & ~Y;
+    case 2:
+        return X | Y;
+    case 3:
+        return X ^ Y;
+    }
+
+    return X;
+}
+
+static void HashBlock(HEFTY1_CTX *ctx)
+{
+    uint32_t A, B, C, D, E, F, G, H;
+    uint32_t W[HEFTY1_BLOCK_BYTES];
+    int t;
+
+    assert(ctx);
+
+    A = ctx->h[0];
+    B = ctx->h[1];
+    C = ctx->h[2];
+    D = ctx->h[3];
+    E = ctx->h[4];
+    F = ctx->h[5];
+    G = ctx->h[6];
+    H = ctx->h[7];
+
+    t = 0;
+    for (; t < 16; t++) {
+        W[t] = Reverse32(((uint32_t *)&ctx->block[0])[t]); /* To host byte order */
+        Absorb(ctx->sponge, W[t] ^ K[t]);
+    }
+
+    for (t = 0; t < 16; t++) {
+        Absorb(ctx->sponge, D ^ H);
+        RoundFunc(ctx, A, B, C, D, E, F, G, H, W[t], K[t]);
+    }
+    for (t = 16; t < 64; t++) {
+        Absorb(ctx->sponge, H + D);
+        W[t] = sigma1(W[t - 2]) + W[t - 7] + sigma0(W[t - 15]) + W[t - 16];
+        RoundFunc(ctx, A, B, C, D, E, F, G, H, W[t], K[t]);
+    }
+
+    ctx->h[0] += A;
+    ctx->h[1] += B;
+    ctx->h[2] += C;
+    ctx->h[3] += D;
+    ctx->h[4] += E;
+    ctx->h[5] += F;
+    ctx->h[6] += G;
+    ctx->h[7] += H;
+
+    A = 0;
+    B = 0;
+    C = 0;
+    D = 0;
+    E = 0;
+    F = 0;
+    G = 0;
+    H = 0;
+
+    memset(W, 0, sizeof(W));
+}
+
+/* Public interface */
+
+void HEFTY1_Init(HEFTY1_CTX *ctx)
+{
+    assert(ctx);
+
+    memcpy(ctx->h, H, sizeof(ctx->h));
+    memset(ctx->block, 0, sizeof(ctx->block));
+    ctx->written = 0;
+    memset(ctx->sponge, 0, sizeof(ctx->sponge));
+}
+
+void HEFTY1_Update(HEFTY1_CTX *ctx, const void *buf, size_t len)
+{
+    uint64_t read;
+    assert(ctx);
+
+    read = 0;
+    while (len) {
+        uint64_t end = ctx->written % HEFTY1_BLOCK_BYTES;
+        uint64_t count = Min(len, HEFTY1_BLOCK_BYTES - end);
+        memcpy(&ctx->block[end], &((unsigned char *)buf)[read], (size_t)count);
+        len -= (size_t)count;
+        read += count;
+        ctx->written += count;
+        if (!(ctx->written % HEFTY1_BLOCK_BYTES))
+            HashBlock(ctx);
+    }
+}
+
+void HEFTY1_Final(unsigned char *digest, HEFTY1_CTX *ctx)
+{
+    uint64_t used;
+    uint64_t *len;
+    int i;
+    assert(digest);
+    assert(ctx);
+
+    /* Pad message (FIPS 180 Section 5.1.1) */
+    used = ctx->written % HEFTY1_BLOCK_BYTES;
+    ctx->block[used++] = 0x80; /* Append 1 to end of message */
+    if (used > HEFTY1_BLOCK_BYTES - 8) {
+        /* We have already written into the last 64bits, so
+         * we must continue into the next block. */
+        memset(&ctx->block[used], 0, HEFTY1_BLOCK_BYTES - (size_t)used);
+        HashBlock(ctx);
+        used = 0; /* Create a new block (below) */
+    }
+
+    /* All remaining bits to zero */
+    memset(&ctx->block[used], 0, HEFTY1_BLOCK_BYTES - 8 - (size_t)used);
+
+    /* The last 64bits encode the length (in network byte order) */
+    len = (uint64_t *)&ctx->block[HEFTY1_BLOCK_BYTES - 8];
+    *len = Reverse64(ctx->written*8);
+
+    HashBlock(ctx);
+
+    /* Convert back to network byte order */
+    i = 0;
+    for (; i < HEFTY1_STATE_WORDS; i++)
+        ctx->h[i] = Reverse32(ctx->h[i]);
+
+    memcpy(digest, ctx->h, sizeof(ctx->h));
+    memset(ctx, 0, sizeof(HEFTY1_CTX));
+}
+
+unsigned char* HEFTY1(const unsigned char *buf, size_t len, unsigned char *digest)
+{
+    HEFTY1_CTX ctx;
+    static unsigned char m[HEFTY1_DIGEST_BYTES];
+
+    if (!digest)
+        digest = m;
+
+    HEFTY1_Init(&ctx);
+    HEFTY1_Update(&ctx, buf, len);
+    HEFTY1_Final(digest, &ctx);
+
+    return digest;
+}
diff --git a/hefty1.h b/hefty1.h
new file mode 100644
index 0000000000..29939e86ff
--- /dev/null
+++ b/hefty1.h
@@ -0,0 +1,66 @@
+/*
+ * HEFTY1 CPU-only cryptographic hash function
+ *
+ * Copyright (c) 2014, dbcc14 <BM-NBx4AKznJuyem3dArgVY8MGyABpihRy5>
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *
+ * 1. Redistributions of source code must retain the above copyright notice, this
+ *    list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright notice,
+ *    this list of conditions and the following disclaimer in the documentation
+ *    and/or other materials provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
+ * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ * DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
+ * ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ * The views and conclusions contained in the software and documentation are those
+ * of the authors and should not be interpreted as representing official policies,
+ * either expressed or implied, of the FreeBSD Project.
+ */
+
+#ifndef __HEFTY1_H__
+#define __HEFTY1_H__
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#ifndef WIN32
+#include <sys/types.h>
+#endif
+
+#include <inttypes.h>
+
+#define HEFTY1_DIGEST_BYTES 32
+#define HEFTY1_BLOCK_BYTES 64
+#define HEFTY1_STATE_WORDS 8
+#define HEFTY1_SPONGE_WORDS 4
+
+typedef struct HEFTY1_CTX {
+    uint32_t h[HEFTY1_STATE_WORDS];
+    uint8_t  block[HEFTY1_BLOCK_BYTES];
+    uint64_t written;
+    uint32_t sponge[HEFTY1_SPONGE_WORDS];
+} HEFTY1_CTX;
+
+void HEFTY1_Init(HEFTY1_CTX *cxt);
+void HEFTY1_Update(HEFTY1_CTX *cxt, const void *data, size_t len);
+void HEFTY1_Final(unsigned char *digest, HEFTY1_CTX *cxt);
+unsigned char* HEFTY1(const unsigned char *data, size_t len, unsigned char *digest);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* __HEFTY1_H__ */
diff --git a/install-sh b/install-sh
new file mode 100644
index 0000000000..9c04de25c4
--- /dev/null
+++ b/install-sh
@@ -0,0 +1,527 @@
+#!/bin/sh
+# install - install a program, script, or datafile
+
+scriptversion=2011-01-19.21; # UTC
+
+# This originates from X11R5 (mit/util/scripts/install.sh), which was
+# later released in X11R6 (xc/config/util/install.sh) with the
+# following copyright and license.
+#
+# Copyright (C) 1994 X Consortium
+#
+# Permission is hereby granted, free of charge, to any person obtaining a copy
+# of this software and associated documentation files (the "Software"), to
+# deal in the Software without restriction, including without limitation the
+# rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+# sell copies of the Software, and to permit persons to whom the Software is
+# furnished to do so, subject to the following conditions:
+#
+# The above copyright notice and this permission notice shall be included in
+# all copies or substantial portions of the Software.
+#
+# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL THE
+# X CONSORTIUM BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
+# AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNEC-
+# TION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+#
+# Except as contained in this notice, the name of the X Consortium shall not
+# be used in advertising or otherwise to promote the sale, use or other deal-
+# ings in this Software without prior written authorization from the X Consor-
+# tium.
+#
+#
+# FSF changes to this file are in the public domain.
+#
+# Calling this script install-sh is preferred over install.sh, to prevent
+# `make' implicit rules from creating a file called install from it
+# when there is no Makefile.
+#
+# This script is compatible with the BSD install script, but was written
+# from scratch.
+
+nl='
+'
+IFS=" ""	$nl"
+
+# set DOITPROG to echo to test this script
+
+# Don't use :- since 4.3BSD and earlier shells don't like it.
+doit=${DOITPROG-}
+if test -z "$doit"; then
+  doit_exec=exec
+else
+  doit_exec=$doit
+fi
+
+# Put in absolute file names if you don't have them in your path;
+# or use environment vars.
+
+chgrpprog=${CHGRPPROG-chgrp}
+chmodprog=${CHMODPROG-chmod}
+chownprog=${CHOWNPROG-chown}
+cmpprog=${CMPPROG-cmp}
+cpprog=${CPPROG-cp}
+mkdirprog=${MKDIRPROG-mkdir}
+mvprog=${MVPROG-mv}
+rmprog=${RMPROG-rm}
+stripprog=${STRIPPROG-strip}
+
+posix_glob='?'
+initialize_posix_glob='
+  test "$posix_glob" != "?" || {
+    if (set -f) 2>/dev/null; then
+      posix_glob=
+    else
+      posix_glob=:
+    fi
+  }
+'
+
+posix_mkdir=
+
+# Desired mode of installed file.
+mode=0755
+
+chgrpcmd=
+chmodcmd=$chmodprog
+chowncmd=
+mvcmd=$mvprog
+rmcmd="$rmprog -f"
+stripcmd=
+
+src=
+dst=
+dir_arg=
+dst_arg=
+
+copy_on_change=false
+no_target_directory=
+
+usage="\
+Usage: $0 [OPTION]... [-T] SRCFILE DSTFILE
+   or: $0 [OPTION]... SRCFILES... DIRECTORY
+   or: $0 [OPTION]... -t DIRECTORY SRCFILES...
+   or: $0 [OPTION]... -d DIRECTORIES...
+
+In the 1st form, copy SRCFILE to DSTFILE.
+In the 2nd and 3rd, copy all SRCFILES to DIRECTORY.
+In the 4th, create DIRECTORIES.
+
+Options:
+     --help     display this help and exit.
+     --version  display version info and exit.
+
+  -c            (ignored)
+  -C            install only if different (preserve the last data modification time)
+  -d            create directories instead of installing files.
+  -g GROUP      $chgrpprog installed files to GROUP.
+  -m MODE       $chmodprog installed files to MODE.
+  -o USER       $chownprog installed files to USER.
+  -s            $stripprog installed files.
+  -t DIRECTORY  install into DIRECTORY.
+  -T            report an error if DSTFILE is a directory.
+
+Environment variables override the default commands:
+  CHGRPPROG CHMODPROG CHOWNPROG CMPPROG CPPROG MKDIRPROG MVPROG
+  RMPROG STRIPPROG
+"
+
+while test $# -ne 0; do
+  case $1 in
+    -c) ;;
+
+    -C) copy_on_change=true;;
+
+    -d) dir_arg=true;;
+
+    -g) chgrpcmd="$chgrpprog $2"
+	shift;;
+
+    --help) echo "$usage"; exit $?;;
+
+    -m) mode=$2
+	case $mode in
+	  *' '* | *'	'* | *'
+'*	  | *'*'* | *'?'* | *'['*)
+	    echo "$0: invalid mode: $mode" >&2
+	    exit 1;;
+	esac
+	shift;;
+
+    -o) chowncmd="$chownprog $2"
+	shift;;
+
+    -s) stripcmd=$stripprog;;
+
+    -t) dst_arg=$2
+	# Protect names problematic for `test' and other utilities.
+	case $dst_arg in
+	  -* | [=\(\)!]) dst_arg=./$dst_arg;;
+	esac
+	shift;;
+
+    -T) no_target_directory=true;;
+
+    --version) echo "$0 $scriptversion"; exit $?;;
+
+    --)	shift
+	break;;
+
+    -*)	echo "$0: invalid option: $1" >&2
+	exit 1;;
+
+    *)  break;;
+  esac
+  shift
+done
+
+if test $# -ne 0 && test -z "$dir_arg$dst_arg"; then
+  # When -d is used, all remaining arguments are directories to create.
+  # When -t is used, the destination is already specified.
+  # Otherwise, the last argument is the destination.  Remove it from $@.
+  for arg
+  do
+    if test -n "$dst_arg"; then
+      # $@ is not empty: it contains at least $arg.
+      set fnord "$@" "$dst_arg"
+      shift # fnord
+    fi
+    shift # arg
+    dst_arg=$arg
+    # Protect names problematic for `test' and other utilities.
+    case $dst_arg in
+      -* | [=\(\)!]) dst_arg=./$dst_arg;;
+    esac
+  done
+fi
+
+if test $# -eq 0; then
+  if test -z "$dir_arg"; then
+    echo "$0: no input file specified." >&2
+    exit 1
+  fi
+  # It's OK to call `install-sh -d' without argument.
+  # This can happen when creating conditional directories.
+  exit 0
+fi
+
+if test -z "$dir_arg"; then
+  do_exit='(exit $ret); exit $ret'
+  trap "ret=129; $do_exit" 1
+  trap "ret=130; $do_exit" 2
+  trap "ret=141; $do_exit" 13
+  trap "ret=143; $do_exit" 15
+
+  # Set umask so as not to create temps with too-generous modes.
+  # However, 'strip' requires both read and write access to temps.
+  case $mode in
+    # Optimize common cases.
+    *644) cp_umask=133;;
+    *755) cp_umask=22;;
+
+    *[0-7])
+      if test -z "$stripcmd"; then
+	u_plus_rw=
+      else
+	u_plus_rw='% 200'
+      fi
+      cp_umask=`expr '(' 777 - $mode % 1000 ')' $u_plus_rw`;;
+    *)
+      if test -z "$stripcmd"; then
+	u_plus_rw=
+      else
+	u_plus_rw=,u+rw
+      fi
+      cp_umask=$mode$u_plus_rw;;
+  esac
+fi
+
+for src
+do
+  # Protect names problematic for `test' and other utilities.
+  case $src in
+    -* | [=\(\)!]) src=./$src;;
+  esac
+
+  if test -n "$dir_arg"; then
+    dst=$src
+    dstdir=$dst
+    test -d "$dstdir"
+    dstdir_status=$?
+  else
+
+    # Waiting for this to be detected by the "$cpprog $src $dsttmp" command
+    # might cause directories to be created, which would be especially bad
+    # if $src (and thus $dsttmp) contains '*'.
+    if test ! -f "$src" && test ! -d "$src"; then
+      echo "$0: $src does not exist." >&2
+      exit 1
+    fi
+
+    if test -z "$dst_arg"; then
+      echo "$0: no destination specified." >&2
+      exit 1
+    fi
+    dst=$dst_arg
+
+    # If destination is a directory, append the input filename; won't work
+    # if double slashes aren't ignored.
+    if test -d "$dst"; then
+      if test -n "$no_target_directory"; then
+	echo "$0: $dst_arg: Is a directory" >&2
+	exit 1
+      fi
+      dstdir=$dst
+      dst=$dstdir/`basename "$src"`
+      dstdir_status=0
+    else
+      # Prefer dirname, but fall back on a substitute if dirname fails.
+      dstdir=`
+	(dirname "$dst") 2>/dev/null ||
+	expr X"$dst" : 'X\(.*[^/]\)//*[^/][^/]*/*$' \| \
+	     X"$dst" : 'X\(//\)[^/]' \| \
+	     X"$dst" : 'X\(//\)$' \| \
+	     X"$dst" : 'X\(/\)' \| . 2>/dev/null ||
+	echo X"$dst" |
+	    sed '/^X\(.*[^/]\)\/\/*[^/][^/]*\/*$/{
+		   s//\1/
+		   q
+		 }
+		 /^X\(\/\/\)[^/].*/{
+		   s//\1/
+		   q
+		 }
+		 /^X\(\/\/\)$/{
+		   s//\1/
+		   q
+		 }
+		 /^X\(\/\).*/{
+		   s//\1/
+		   q
+		 }
+		 s/.*/./; q'
+      `
+
+      test -d "$dstdir"
+      dstdir_status=$?
+    fi
+  fi
+
+  obsolete_mkdir_used=false
+
+  if test $dstdir_status != 0; then
+    case $posix_mkdir in
+      '')
+	# Create intermediate dirs using mode 755 as modified by the umask.
+	# This is like FreeBSD 'install' as of 1997-10-28.
+	umask=`umask`
+	case $stripcmd.$umask in
+	  # Optimize common cases.
+	  *[2367][2367]) mkdir_umask=$umask;;
+	  .*0[02][02] | .[02][02] | .[02]) mkdir_umask=22;;
+
+	  *[0-7])
+	    mkdir_umask=`expr $umask + 22 \
+	      - $umask % 100 % 40 + $umask % 20 \
+	      - $umask % 10 % 4 + $umask % 2
+	    `;;
+	  *) mkdir_umask=$umask,go-w;;
+	esac
+
+	# With -d, create the new directory with the user-specified mode.
+	# Otherwise, rely on $mkdir_umask.
+	if test -n "$dir_arg"; then
+	  mkdir_mode=-m$mode
+	else
+	  mkdir_mode=
+	fi
+
+	posix_mkdir=false
+	case $umask in
+	  *[123567][0-7][0-7])
+	    # POSIX mkdir -p sets u+wx bits regardless of umask, which
+	    # is incompatible with FreeBSD 'install' when (umask & 300) != 0.
+	    ;;
+	  *)
+	    tmpdir=${TMPDIR-/tmp}/ins$RANDOM-$$
+	    trap 'ret=$?; rmdir "$tmpdir/d" "$tmpdir" 2>/dev/null; exit $ret' 0
+
+	    if (umask $mkdir_umask &&
+		exec $mkdirprog $mkdir_mode -p -- "$tmpdir/d") >/dev/null 2>&1
+	    then
+	      if test -z "$dir_arg" || {
+		   # Check for POSIX incompatibilities with -m.
+		   # HP-UX 11.23 and IRIX 6.5 mkdir -m -p sets group- or
+		   # other-writeable bit of parent directory when it shouldn't.
+		   # FreeBSD 6.1 mkdir -m -p sets mode of existing directory.
+		   ls_ld_tmpdir=`ls -ld "$tmpdir"`
+		   case $ls_ld_tmpdir in
+		     d????-?r-*) different_mode=700;;
+		     d????-?--*) different_mode=755;;
+		     *) false;;
+		   esac &&
+		   $mkdirprog -m$different_mode -p -- "$tmpdir" && {
+		     ls_ld_tmpdir_1=`ls -ld "$tmpdir"`
+		     test "$ls_ld_tmpdir" = "$ls_ld_tmpdir_1"
+		   }
+		 }
+	      then posix_mkdir=:
+	      fi
+	      rmdir "$tmpdir/d" "$tmpdir"
+	    else
+	      # Remove any dirs left behind by ancient mkdir implementations.
+	      rmdir ./$mkdir_mode ./-p ./-- 2>/dev/null
+	    fi
+	    trap '' 0;;
+	esac;;
+    esac
+
+    if
+      $posix_mkdir && (
+	umask $mkdir_umask &&
+	$doit_exec $mkdirprog $mkdir_mode -p -- "$dstdir"
+      )
+    then :
+    else
+
+      # The umask is ridiculous, or mkdir does not conform to POSIX,
+      # or it failed possibly due to a race condition.  Create the
+      # directory the slow way, step by step, checking for races as we go.
+
+      case $dstdir in
+	/*) prefix='/';;
+	[-=\(\)!]*) prefix='./';;
+	*)  prefix='';;
+      esac
+
+      eval "$initialize_posix_glob"
+
+      oIFS=$IFS
+      IFS=/
+      $posix_glob set -f
+      set fnord $dstdir
+      shift
+      $posix_glob set +f
+      IFS=$oIFS
+
+      prefixes=
+
+      for d
+      do
+	test X"$d" = X && continue
+
+	prefix=$prefix$d
+	if test -d "$prefix"; then
+	  prefixes=
+	else
+	  if $posix_mkdir; then
+	    (umask=$mkdir_umask &&
+	     $doit_exec $mkdirprog $mkdir_mode -p -- "$dstdir") && break
+	    # Don't fail if two instances are running concurrently.
+	    test -d "$prefix" || exit 1
+	  else
+	    case $prefix in
+	      *\'*) qprefix=`echo "$prefix" | sed "s/'/'\\\\\\\\''/g"`;;
+	      *) qprefix=$prefix;;
+	    esac
+	    prefixes="$prefixes '$qprefix'"
+	  fi
+	fi
+	prefix=$prefix/
+      done
+
+      if test -n "$prefixes"; then
+	# Don't fail if two instances are running concurrently.
+	(umask $mkdir_umask &&
+	 eval "\$doit_exec \$mkdirprog $prefixes") ||
+	  test -d "$dstdir" || exit 1
+	obsolete_mkdir_used=true
+      fi
+    fi
+  fi
+
+  if test -n "$dir_arg"; then
+    { test -z "$chowncmd" || $doit $chowncmd "$dst"; } &&
+    { test -z "$chgrpcmd" || $doit $chgrpcmd "$dst"; } &&
+    { test "$obsolete_mkdir_used$chowncmd$chgrpcmd" = false ||
+      test -z "$chmodcmd" || $doit $chmodcmd $mode "$dst"; } || exit 1
+  else
+
+    # Make a couple of temp file names in the proper directory.
+    dsttmp=$dstdir/_inst.$$_
+    rmtmp=$dstdir/_rm.$$_
+
+    # Trap to clean up those temp files at exit.
+    trap 'ret=$?; rm -f "$dsttmp" "$rmtmp" && exit $ret' 0
+
+    # Copy the file name to the temp name.
+    (umask $cp_umask && $doit_exec $cpprog "$src" "$dsttmp") &&
+
+    # and set any options; do chmod last to preserve setuid bits.
+    #
+    # If any of these fail, we abort the whole thing.  If we want to
+    # ignore errors from any of these, just make sure not to ignore
+    # errors from the above "$doit $cpprog $src $dsttmp" command.
+    #
+    { test -z "$chowncmd" || $doit $chowncmd "$dsttmp"; } &&
+    { test -z "$chgrpcmd" || $doit $chgrpcmd "$dsttmp"; } &&
+    { test -z "$stripcmd" || $doit $stripcmd "$dsttmp"; } &&
+    { test -z "$chmodcmd" || $doit $chmodcmd $mode "$dsttmp"; } &&
+
+    # If -C, don't bother to copy if it wouldn't change the file.
+    if $copy_on_change &&
+       old=`LC_ALL=C ls -dlL "$dst"	2>/dev/null` &&
+       new=`LC_ALL=C ls -dlL "$dsttmp"	2>/dev/null` &&
+
+       eval "$initialize_posix_glob" &&
+       $posix_glob set -f &&
+       set X $old && old=:$2:$4:$5:$6 &&
+       set X $new && new=:$2:$4:$5:$6 &&
+       $posix_glob set +f &&
+
+       test "$old" = "$new" &&
+       $cmpprog "$dst" "$dsttmp" >/dev/null 2>&1
+    then
+      rm -f "$dsttmp"
+    else
+      # Rename the file to the real destination.
+      $doit $mvcmd -f "$dsttmp" "$dst" 2>/dev/null ||
+
+      # The rename failed, perhaps because mv can't rename something else
+      # to itself, or perhaps because mv is so ancient that it does not
+      # support -f.
+      {
+	# Now remove or move aside any old file at destination location.
+	# We try this two ways since rm can't unlink itself on some
+	# systems and the destination file might be busy for other
+	# reasons.  In this case, the final cleanup might fail but the new
+	# file should still install successfully.
+	{
+	  test ! -f "$dst" ||
+	  $doit $rmcmd -f "$dst" 2>/dev/null ||
+	  { $doit $mvcmd -f "$dst" "$rmtmp" 2>/dev/null &&
+	    { $doit $rmcmd -f "$rmtmp" 2>/dev/null; :; }
+	  } ||
+	  { echo "$0: cannot unlink or rename $dst" >&2
+	    (exit 1); exit 1
+	  }
+	} &&
+
+	# Now rename the file to the real destination.
+	$doit $mvcmd "$dsttmp" "$dst"
+      }
+    fi || exit 1
+
+    trap '' 0
+  fi
+done
+
+# Local variables:
+# eval: (add-hook 'write-file-hooks 'time-stamp)
+# time-stamp-start: "scriptversion="
+# time-stamp-format: "%:y-%02m-%02d.%02H"
+# time-stamp-time-zone: "UTC"
+# time-stamp-end: "; # UTC"
+# End:
diff --git a/keccak.c b/keccak.c
new file mode 100644
index 0000000000..8c90f3a574
--- /dev/null
+++ b/keccak.c
@@ -0,0 +1,1824 @@
+/* $Id: keccak.c 259 2011-07-19 22:11:27Z tp $ */
+/*
+ * Keccak implementation.
+ *
+ * ==========================(LICENSE BEGIN)============================
+ *
+ * Copyright (c) 2007-2010  Projet RNRT SAPHIR
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining
+ * a copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sublicense, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice shall be
+ * included in all copies or substantial portions of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ * IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY
+ * CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * ===========================(LICENSE END)=============================
+ *
+ * @author   Thomas Pornin <thomas.pornin@cryptolog.com>
+ */
+
+#include <stddef.h>
+#include <string.h>
+
+#include "sph_keccak.h"
+
+#ifdef __cplusplus
+extern "C"{
+#endif
+
+/*
+ * Parameters:
+ *
+ *  SPH_KECCAK_64          use a 64-bit type
+ *  SPH_KECCAK_UNROLL      number of loops to unroll (0/undef for full unroll)
+ *  SPH_KECCAK_INTERLEAVE  use bit-interleaving (32-bit type only)
+ *  SPH_KECCAK_NOCOPY      do not copy the state into local variables
+ * 
+ * If there is no usable 64-bit type, the code automatically switches
+ * back to the 32-bit implementation.
+ *
+ * Some tests on an Intel Core2 Q6600 (both 64-bit and 32-bit, 32 kB L1
+ * code cache), a PowerPC (G3, 32 kB L1 code cache), an ARM920T core
+ * (16 kB L1 code cache), and a small MIPS-compatible CPU (Broadcom BCM3302,
+ * 8 kB L1 code cache), seem to show that the following are optimal:
+ *
+ * -- x86, 64-bit: use the 64-bit implementation, unroll 8 rounds,
+ * do not copy the state; unrolling 2, 6 or all rounds also provides
+ * near-optimal performance.
+ * -- x86, 32-bit: use the 32-bit implementation, unroll 6 rounds,
+ * interleave, do not copy the state. Unrolling 1, 2, 4 or 8 rounds
+ * also provides near-optimal performance.
+ * -- PowerPC: use the 64-bit implementation, unroll 8 rounds,
+ * copy the state. Unrolling 4 or 6 rounds is near-optimal.
+ * -- ARM: use the 64-bit implementation, unroll 2 or 4 rounds,
+ * copy the state.
+ * -- MIPS: use the 64-bit implementation, unroll 2 rounds, copy
+ * the state. Unrolling only 1 round is also near-optimal.
+ *
+ * Also, interleaving does not always yield actual improvements when
+ * using a 32-bit implementation; in particular when the architecture
+ * does not offer a native rotation opcode (interleaving replaces one
+ * 64-bit rotation with two 32-bit rotations, which is a gain only if
+ * there is a native 32-bit rotation opcode and not a native 64-bit
+ * rotation opcode; also, interleaving implies a small overhead when
+ * processing input words).
+ *
+ * To sum up:
+ * -- when possible, use the 64-bit code
+ * -- exception: on 32-bit x86, use 32-bit code
+ * -- when using 32-bit code, use interleaving
+ * -- copy the state, except on x86
+ * -- unroll 8 rounds on "big" machine, 2 rounds on "small" machines
+ */
+
+#if SPH_SMALL_FOOTPRINT && !defined SPH_SMALL_FOOTPRINT_KECCAK
+#define SPH_SMALL_FOOTPRINT_KECCAK   1
+#endif
+
+/*
+ * By default, we select the 64-bit implementation if a 64-bit type
+ * is available, unless a 32-bit x86 is detected.
+ */
+#if !defined SPH_KECCAK_64 && SPH_64 \
+	&& !(defined __i386__ || SPH_I386_GCC || SPH_I386_MSVC)
+#define SPH_KECCAK_64   1
+#endif
+
+/*
+ * If using a 32-bit implementation, we prefer to interleave.
+ */
+#if !SPH_KECCAK_64 && !defined SPH_KECCAK_INTERLEAVE
+#define SPH_KECCAK_INTERLEAVE   1
+#endif
+
+/*
+ * Unroll 8 rounds on big systems, 2 rounds on small systems.
+ */
+#ifndef SPH_KECCAK_UNROLL
+#if SPH_SMALL_FOOTPRINT_KECCAK
+#define SPH_KECCAK_UNROLL   2
+#else
+#define SPH_KECCAK_UNROLL   8
+#endif
+#endif
+
+/*
+ * We do not want to copy the state to local variables on x86 (32-bit
+ * and 64-bit alike).
+ */
+#ifndef SPH_KECCAK_NOCOPY
+#if defined __i386__ || defined __x86_64 || SPH_I386_MSVC || SPH_I386_GCC
+#define SPH_KECCAK_NOCOPY   1
+#else
+#define SPH_KECCAK_NOCOPY   0
+#endif
+#endif
+
+#ifdef _MSC_VER
+#pragma warning (disable: 4146)
+#endif
+
+#if SPH_KECCAK_64
+
+static const sph_u64 RC[] = {
+	SPH_C64(0x0000000000000001), SPH_C64(0x0000000000008082),
+	SPH_C64(0x800000000000808A), SPH_C64(0x8000000080008000),
+	SPH_C64(0x000000000000808B), SPH_C64(0x0000000080000001),
+	SPH_C64(0x8000000080008081), SPH_C64(0x8000000000008009),
+	SPH_C64(0x000000000000008A), SPH_C64(0x0000000000000088),
+	SPH_C64(0x0000000080008009), SPH_C64(0x000000008000000A),
+	SPH_C64(0x000000008000808B), SPH_C64(0x800000000000008B),
+	SPH_C64(0x8000000000008089), SPH_C64(0x8000000000008003),
+	SPH_C64(0x8000000000008002), SPH_C64(0x8000000000000080),
+	SPH_C64(0x000000000000800A), SPH_C64(0x800000008000000A),
+	SPH_C64(0x8000000080008081), SPH_C64(0x8000000000008080),
+	SPH_C64(0x0000000080000001), SPH_C64(0x8000000080008008)
+};
+
+#if SPH_KECCAK_NOCOPY
+
+#define a00   (kc->u.wide[ 0])
+#define a10   (kc->u.wide[ 1])
+#define a20   (kc->u.wide[ 2])
+#define a30   (kc->u.wide[ 3])
+#define a40   (kc->u.wide[ 4])
+#define a01   (kc->u.wide[ 5])
+#define a11   (kc->u.wide[ 6])
+#define a21   (kc->u.wide[ 7])
+#define a31   (kc->u.wide[ 8])
+#define a41   (kc->u.wide[ 9])
+#define a02   (kc->u.wide[10])
+#define a12   (kc->u.wide[11])
+#define a22   (kc->u.wide[12])
+#define a32   (kc->u.wide[13])
+#define a42   (kc->u.wide[14])
+#define a03   (kc->u.wide[15])
+#define a13   (kc->u.wide[16])
+#define a23   (kc->u.wide[17])
+#define a33   (kc->u.wide[18])
+#define a43   (kc->u.wide[19])
+#define a04   (kc->u.wide[20])
+#define a14   (kc->u.wide[21])
+#define a24   (kc->u.wide[22])
+#define a34   (kc->u.wide[23])
+#define a44   (kc->u.wide[24])
+
+#define DECL_STATE
+#define READ_STATE(sc)
+#define WRITE_STATE(sc)
+
+#define INPUT_BUF(size)   do { \
+		size_t j; \
+		for (j = 0; j < (size); j += 8) { \
+			kc->u.wide[j >> 3] ^= sph_dec64le_aligned(buf + j); \
+		} \
+	} while (0)
+
+#define INPUT_BUF144   INPUT_BUF(144)
+#define INPUT_BUF136   INPUT_BUF(136)
+#define INPUT_BUF104   INPUT_BUF(104)
+#define INPUT_BUF72    INPUT_BUF(72)
+
+#else
+
+#define DECL_STATE \
+	sph_u64 a00, a01, a02, a03, a04; \
+	sph_u64 a10, a11, a12, a13, a14; \
+	sph_u64 a20, a21, a22, a23, a24; \
+	sph_u64 a30, a31, a32, a33, a34; \
+	sph_u64 a40, a41, a42, a43, a44;
+
+#define READ_STATE(state)   do { \
+		a00 = (state)->u.wide[ 0]; \
+		a10 = (state)->u.wide[ 1]; \
+		a20 = (state)->u.wide[ 2]; \
+		a30 = (state)->u.wide[ 3]; \
+		a40 = (state)->u.wide[ 4]; \
+		a01 = (state)->u.wide[ 5]; \
+		a11 = (state)->u.wide[ 6]; \
+		a21 = (state)->u.wide[ 7]; \
+		a31 = (state)->u.wide[ 8]; \
+		a41 = (state)->u.wide[ 9]; \
+		a02 = (state)->u.wide[10]; \
+		a12 = (state)->u.wide[11]; \
+		a22 = (state)->u.wide[12]; \
+		a32 = (state)->u.wide[13]; \
+		a42 = (state)->u.wide[14]; \
+		a03 = (state)->u.wide[15]; \
+		a13 = (state)->u.wide[16]; \
+		a23 = (state)->u.wide[17]; \
+		a33 = (state)->u.wide[18]; \
+		a43 = (state)->u.wide[19]; \
+		a04 = (state)->u.wide[20]; \
+		a14 = (state)->u.wide[21]; \
+		a24 = (state)->u.wide[22]; \
+		a34 = (state)->u.wide[23]; \
+		a44 = (state)->u.wide[24]; \
+	} while (0)
+
+#define WRITE_STATE(state)   do { \
+		(state)->u.wide[ 0] = a00; \
+		(state)->u.wide[ 1] = a10; \
+		(state)->u.wide[ 2] = a20; \
+		(state)->u.wide[ 3] = a30; \
+		(state)->u.wide[ 4] = a40; \
+		(state)->u.wide[ 5] = a01; \
+		(state)->u.wide[ 6] = a11; \
+		(state)->u.wide[ 7] = a21; \
+		(state)->u.wide[ 8] = a31; \
+		(state)->u.wide[ 9] = a41; \
+		(state)->u.wide[10] = a02; \
+		(state)->u.wide[11] = a12; \
+		(state)->u.wide[12] = a22; \
+		(state)->u.wide[13] = a32; \
+		(state)->u.wide[14] = a42; \
+		(state)->u.wide[15] = a03; \
+		(state)->u.wide[16] = a13; \
+		(state)->u.wide[17] = a23; \
+		(state)->u.wide[18] = a33; \
+		(state)->u.wide[19] = a43; \
+		(state)->u.wide[20] = a04; \
+		(state)->u.wide[21] = a14; \
+		(state)->u.wide[22] = a24; \
+		(state)->u.wide[23] = a34; \
+		(state)->u.wide[24] = a44; \
+	} while (0)
+
+#define INPUT_BUF144   do { \
+		a00 ^= sph_dec64le_aligned(buf +   0); \
+		a10 ^= sph_dec64le_aligned(buf +   8); \
+		a20 ^= sph_dec64le_aligned(buf +  16); \
+		a30 ^= sph_dec64le_aligned(buf +  24); \
+		a40 ^= sph_dec64le_aligned(buf +  32); \
+		a01 ^= sph_dec64le_aligned(buf +  40); \
+		a11 ^= sph_dec64le_aligned(buf +  48); \
+		a21 ^= sph_dec64le_aligned(buf +  56); \
+		a31 ^= sph_dec64le_aligned(buf +  64); \
+		a41 ^= sph_dec64le_aligned(buf +  72); \
+		a02 ^= sph_dec64le_aligned(buf +  80); \
+		a12 ^= sph_dec64le_aligned(buf +  88); \
+		a22 ^= sph_dec64le_aligned(buf +  96); \
+		a32 ^= sph_dec64le_aligned(buf + 104); \
+		a42 ^= sph_dec64le_aligned(buf + 112); \
+		a03 ^= sph_dec64le_aligned(buf + 120); \
+		a13 ^= sph_dec64le_aligned(buf + 128); \
+		a23 ^= sph_dec64le_aligned(buf + 136); \
+	} while (0)
+
+#define INPUT_BUF136   do { \
+		a00 ^= sph_dec64le_aligned(buf +   0); \
+		a10 ^= sph_dec64le_aligned(buf +   8); \
+		a20 ^= sph_dec64le_aligned(buf +  16); \
+		a30 ^= sph_dec64le_aligned(buf +  24); \
+		a40 ^= sph_dec64le_aligned(buf +  32); \
+		a01 ^= sph_dec64le_aligned(buf +  40); \
+		a11 ^= sph_dec64le_aligned(buf +  48); \
+		a21 ^= sph_dec64le_aligned(buf +  56); \
+		a31 ^= sph_dec64le_aligned(buf +  64); \
+		a41 ^= sph_dec64le_aligned(buf +  72); \
+		a02 ^= sph_dec64le_aligned(buf +  80); \
+		a12 ^= sph_dec64le_aligned(buf +  88); \
+		a22 ^= sph_dec64le_aligned(buf +  96); \
+		a32 ^= sph_dec64le_aligned(buf + 104); \
+		a42 ^= sph_dec64le_aligned(buf + 112); \
+		a03 ^= sph_dec64le_aligned(buf + 120); \
+		a13 ^= sph_dec64le_aligned(buf + 128); \
+	} while (0)
+
+#define INPUT_BUF104   do { \
+		a00 ^= sph_dec64le_aligned(buf +   0); \
+		a10 ^= sph_dec64le_aligned(buf +   8); \
+		a20 ^= sph_dec64le_aligned(buf +  16); \
+		a30 ^= sph_dec64le_aligned(buf +  24); \
+		a40 ^= sph_dec64le_aligned(buf +  32); \
+		a01 ^= sph_dec64le_aligned(buf +  40); \
+		a11 ^= sph_dec64le_aligned(buf +  48); \
+		a21 ^= sph_dec64le_aligned(buf +  56); \
+		a31 ^= sph_dec64le_aligned(buf +  64); \
+		a41 ^= sph_dec64le_aligned(buf +  72); \
+		a02 ^= sph_dec64le_aligned(buf +  80); \
+		a12 ^= sph_dec64le_aligned(buf +  88); \
+		a22 ^= sph_dec64le_aligned(buf +  96); \
+	} while (0)
+
+#define INPUT_BUF72   do { \
+		a00 ^= sph_dec64le_aligned(buf +   0); \
+		a10 ^= sph_dec64le_aligned(buf +   8); \
+		a20 ^= sph_dec64le_aligned(buf +  16); \
+		a30 ^= sph_dec64le_aligned(buf +  24); \
+		a40 ^= sph_dec64le_aligned(buf +  32); \
+		a01 ^= sph_dec64le_aligned(buf +  40); \
+		a11 ^= sph_dec64le_aligned(buf +  48); \
+		a21 ^= sph_dec64le_aligned(buf +  56); \
+		a31 ^= sph_dec64le_aligned(buf +  64); \
+	} while (0)
+
+#define INPUT_BUF(lim)   do { \
+		a00 ^= sph_dec64le_aligned(buf +   0); \
+		a10 ^= sph_dec64le_aligned(buf +   8); \
+		a20 ^= sph_dec64le_aligned(buf +  16); \
+		a30 ^= sph_dec64le_aligned(buf +  24); \
+		a40 ^= sph_dec64le_aligned(buf +  32); \
+		a01 ^= sph_dec64le_aligned(buf +  40); \
+		a11 ^= sph_dec64le_aligned(buf +  48); \
+		a21 ^= sph_dec64le_aligned(buf +  56); \
+		a31 ^= sph_dec64le_aligned(buf +  64); \
+		if ((lim) == 72) \
+			break; \
+		a41 ^= sph_dec64le_aligned(buf +  72); \
+		a02 ^= sph_dec64le_aligned(buf +  80); \
+		a12 ^= sph_dec64le_aligned(buf +  88); \
+		a22 ^= sph_dec64le_aligned(buf +  96); \
+		if ((lim) == 104) \
+			break; \
+		a32 ^= sph_dec64le_aligned(buf + 104); \
+		a42 ^= sph_dec64le_aligned(buf + 112); \
+		a03 ^= sph_dec64le_aligned(buf + 120); \
+		a13 ^= sph_dec64le_aligned(buf + 128); \
+		if ((lim) == 136) \
+			break; \
+		a23 ^= sph_dec64le_aligned(buf + 136); \
+	} while (0)
+
+#endif
+
+#define DECL64(x)        sph_u64 x
+#define MOV64(d, s)      (d = s)
+#define XOR64(d, a, b)   (d = a ^ b)
+#define AND64(d, a, b)   (d = a & b)
+#define OR64(d, a, b)    (d = a | b)
+#define NOT64(d, s)      (d = SPH_T64(~s))
+#define ROL64(d, v, n)   (d = SPH_ROTL64(v, n))
+#define XOR64_IOTA       XOR64
+
+#else
+
+static const struct {
+	sph_u32 high, low;
+} RC[] = {
+#if SPH_KECCAK_INTERLEAVE
+	{ SPH_C32(0x00000000), SPH_C32(0x00000001) },
+	{ SPH_C32(0x00000089), SPH_C32(0x00000000) },
+	{ SPH_C32(0x8000008B), SPH_C32(0x00000000) },
+	{ SPH_C32(0x80008080), SPH_C32(0x00000000) },
+	{ SPH_C32(0x0000008B), SPH_C32(0x00000001) },
+	{ SPH_C32(0x00008000), SPH_C32(0x00000001) },
+	{ SPH_C32(0x80008088), SPH_C32(0x00000001) },
+	{ SPH_C32(0x80000082), SPH_C32(0x00000001) },
+	{ SPH_C32(0x0000000B), SPH_C32(0x00000000) },
+	{ SPH_C32(0x0000000A), SPH_C32(0x00000000) },
+	{ SPH_C32(0x00008082), SPH_C32(0x00000001) },
+	{ SPH_C32(0x00008003), SPH_C32(0x00000000) },
+	{ SPH_C32(0x0000808B), SPH_C32(0x00000001) },
+	{ SPH_C32(0x8000000B), SPH_C32(0x00000001) },
+	{ SPH_C32(0x8000008A), SPH_C32(0x00000001) },
+	{ SPH_C32(0x80000081), SPH_C32(0x00000001) },
+	{ SPH_C32(0x80000081), SPH_C32(0x00000000) },
+	{ SPH_C32(0x80000008), SPH_C32(0x00000000) },
+	{ SPH_C32(0x00000083), SPH_C32(0x00000000) },
+	{ SPH_C32(0x80008003), SPH_C32(0x00000000) },
+	{ SPH_C32(0x80008088), SPH_C32(0x00000001) },
+	{ SPH_C32(0x80000088), SPH_C32(0x00000000) },
+	{ SPH_C32(0x00008000), SPH_C32(0x00000001) },
+	{ SPH_C32(0x80008082), SPH_C32(0x00000000) }
+#else
+	{ SPH_C32(0x00000000), SPH_C32(0x00000001) },
+	{ SPH_C32(0x00000000), SPH_C32(0x00008082) },
+	{ SPH_C32(0x80000000), SPH_C32(0x0000808A) },
+	{ SPH_C32(0x80000000), SPH_C32(0x80008000) },
+	{ SPH_C32(0x00000000), SPH_C32(0x0000808B) },
+	{ SPH_C32(0x00000000), SPH_C32(0x80000001) },
+	{ SPH_C32(0x80000000), SPH_C32(0x80008081) },
+	{ SPH_C32(0x80000000), SPH_C32(0x00008009) },
+	{ SPH_C32(0x00000000), SPH_C32(0x0000008A) },
+	{ SPH_C32(0x00000000), SPH_C32(0x00000088) },
+	{ SPH_C32(0x00000000), SPH_C32(0x80008009) },
+	{ SPH_C32(0x00000000), SPH_C32(0x8000000A) },
+	{ SPH_C32(0x00000000), SPH_C32(0x8000808B) },
+	{ SPH_C32(0x80000000), SPH_C32(0x0000008B) },
+	{ SPH_C32(0x80000000), SPH_C32(0x00008089) },
+	{ SPH_C32(0x80000000), SPH_C32(0x00008003) },
+	{ SPH_C32(0x80000000), SPH_C32(0x00008002) },
+	{ SPH_C32(0x80000000), SPH_C32(0x00000080) },
+	{ SPH_C32(0x00000000), SPH_C32(0x0000800A) },
+	{ SPH_C32(0x80000000), SPH_C32(0x8000000A) },
+	{ SPH_C32(0x80000000), SPH_C32(0x80008081) },
+	{ SPH_C32(0x80000000), SPH_C32(0x00008080) },
+	{ SPH_C32(0x00000000), SPH_C32(0x80000001) },
+	{ SPH_C32(0x80000000), SPH_C32(0x80008008) }
+#endif
+};
+
+#if SPH_KECCAK_INTERLEAVE
+
+#define INTERLEAVE(xl, xh)   do { \
+		sph_u32 l, h, t; \
+		l = (xl); h = (xh); \
+		t = (l ^ (l >> 1)) & SPH_C32(0x22222222); l ^= t ^ (t << 1); \
+		t = (h ^ (h >> 1)) & SPH_C32(0x22222222); h ^= t ^ (t << 1); \
+		t = (l ^ (l >> 2)) & SPH_C32(0x0C0C0C0C); l ^= t ^ (t << 2); \
+		t = (h ^ (h >> 2)) & SPH_C32(0x0C0C0C0C); h ^= t ^ (t << 2); \
+		t = (l ^ (l >> 4)) & SPH_C32(0x00F000F0); l ^= t ^ (t << 4); \
+		t = (h ^ (h >> 4)) & SPH_C32(0x00F000F0); h ^= t ^ (t << 4); \
+		t = (l ^ (l >> 8)) & SPH_C32(0x0000FF00); l ^= t ^ (t << 8); \
+		t = (h ^ (h >> 8)) & SPH_C32(0x0000FF00); h ^= t ^ (t << 8); \
+		t = (l ^ SPH_T32(h << 16)) & SPH_C32(0xFFFF0000); \
+		l ^= t; h ^= t >> 16; \
+		(xl) = l; (xh) = h; \
+	} while (0)
+
+#define UNINTERLEAVE(xl, xh)   do { \
+		sph_u32 l, h, t; \
+		l = (xl); h = (xh); \
+		t = (l ^ SPH_T32(h << 16)) & SPH_C32(0xFFFF0000); \
+		l ^= t; h ^= t >> 16; \
+		t = (l ^ (l >> 8)) & SPH_C32(0x0000FF00); l ^= t ^ (t << 8); \
+		t = (h ^ (h >> 8)) & SPH_C32(0x0000FF00); h ^= t ^ (t << 8); \
+		t = (l ^ (l >> 4)) & SPH_C32(0x00F000F0); l ^= t ^ (t << 4); \
+		t = (h ^ (h >> 4)) & SPH_C32(0x00F000F0); h ^= t ^ (t << 4); \
+		t = (l ^ (l >> 2)) & SPH_C32(0x0C0C0C0C); l ^= t ^ (t << 2); \
+		t = (h ^ (h >> 2)) & SPH_C32(0x0C0C0C0C); h ^= t ^ (t << 2); \
+		t = (l ^ (l >> 1)) & SPH_C32(0x22222222); l ^= t ^ (t << 1); \
+		t = (h ^ (h >> 1)) & SPH_C32(0x22222222); h ^= t ^ (t << 1); \
+		(xl) = l; (xh) = h; \
+	} while (0)
+
+#else
+
+#define INTERLEAVE(l, h)
+#define UNINTERLEAVE(l, h)
+
+#endif
+
+#if SPH_KECCAK_NOCOPY
+
+#define a00l   (kc->u.narrow[2 *  0 + 0])
+#define a00h   (kc->u.narrow[2 *  0 + 1])
+#define a10l   (kc->u.narrow[2 *  1 + 0])
+#define a10h   (kc->u.narrow[2 *  1 + 1])
+#define a20l   (kc->u.narrow[2 *  2 + 0])
+#define a20h   (kc->u.narrow[2 *  2 + 1])
+#define a30l   (kc->u.narrow[2 *  3 + 0])
+#define a30h   (kc->u.narrow[2 *  3 + 1])
+#define a40l   (kc->u.narrow[2 *  4 + 0])
+#define a40h   (kc->u.narrow[2 *  4 + 1])
+#define a01l   (kc->u.narrow[2 *  5 + 0])
+#define a01h   (kc->u.narrow[2 *  5 + 1])
+#define a11l   (kc->u.narrow[2 *  6 + 0])
+#define a11h   (kc->u.narrow[2 *  6 + 1])
+#define a21l   (kc->u.narrow[2 *  7 + 0])
+#define a21h   (kc->u.narrow[2 *  7 + 1])
+#define a31l   (kc->u.narrow[2 *  8 + 0])
+#define a31h   (kc->u.narrow[2 *  8 + 1])
+#define a41l   (kc->u.narrow[2 *  9 + 0])
+#define a41h   (kc->u.narrow[2 *  9 + 1])
+#define a02l   (kc->u.narrow[2 * 10 + 0])
+#define a02h   (kc->u.narrow[2 * 10 + 1])
+#define a12l   (kc->u.narrow[2 * 11 + 0])
+#define a12h   (kc->u.narrow[2 * 11 + 1])
+#define a22l   (kc->u.narrow[2 * 12 + 0])
+#define a22h   (kc->u.narrow[2 * 12 + 1])
+#define a32l   (kc->u.narrow[2 * 13 + 0])
+#define a32h   (kc->u.narrow[2 * 13 + 1])
+#define a42l   (kc->u.narrow[2 * 14 + 0])
+#define a42h   (kc->u.narrow[2 * 14 + 1])
+#define a03l   (kc->u.narrow[2 * 15 + 0])
+#define a03h   (kc->u.narrow[2 * 15 + 1])
+#define a13l   (kc->u.narrow[2 * 16 + 0])
+#define a13h   (kc->u.narrow[2 * 16 + 1])
+#define a23l   (kc->u.narrow[2 * 17 + 0])
+#define a23h   (kc->u.narrow[2 * 17 + 1])
+#define a33l   (kc->u.narrow[2 * 18 + 0])
+#define a33h   (kc->u.narrow[2 * 18 + 1])
+#define a43l   (kc->u.narrow[2 * 19 + 0])
+#define a43h   (kc->u.narrow[2 * 19 + 1])
+#define a04l   (kc->u.narrow[2 * 20 + 0])
+#define a04h   (kc->u.narrow[2 * 20 + 1])
+#define a14l   (kc->u.narrow[2 * 21 + 0])
+#define a14h   (kc->u.narrow[2 * 21 + 1])
+#define a24l   (kc->u.narrow[2 * 22 + 0])
+#define a24h   (kc->u.narrow[2 * 22 + 1])
+#define a34l   (kc->u.narrow[2 * 23 + 0])
+#define a34h   (kc->u.narrow[2 * 23 + 1])
+#define a44l   (kc->u.narrow[2 * 24 + 0])
+#define a44h   (kc->u.narrow[2 * 24 + 1])
+
+#define DECL_STATE
+#define READ_STATE(state)
+#define WRITE_STATE(state)
+
+#define INPUT_BUF(size)   do { \
+		size_t j; \
+		for (j = 0; j < (size); j += 8) { \
+			sph_u32 tl, th; \
+			tl = sph_dec32le_aligned(buf + j + 0); \
+			th = sph_dec32le_aligned(buf + j + 4); \
+			INTERLEAVE(tl, th); \
+			kc->u.narrow[(j >> 2) + 0] ^= tl; \
+			kc->u.narrow[(j >> 2) + 1] ^= th; \
+		} \
+	} while (0)
+
+#define INPUT_BUF144   INPUT_BUF(144)
+#define INPUT_BUF136   INPUT_BUF(136)
+#define INPUT_BUF104   INPUT_BUF(104)
+#define INPUT_BUF72    INPUT_BUF(72)
+
+#else
+
+#define DECL_STATE \
+	sph_u32 a00l, a00h, a01l, a01h, a02l, a02h, a03l, a03h, a04l, a04h; \
+	sph_u32 a10l, a10h, a11l, a11h, a12l, a12h, a13l, a13h, a14l, a14h; \
+	sph_u32 a20l, a20h, a21l, a21h, a22l, a22h, a23l, a23h, a24l, a24h; \
+	sph_u32 a30l, a30h, a31l, a31h, a32l, a32h, a33l, a33h, a34l, a34h; \
+	sph_u32 a40l, a40h, a41l, a41h, a42l, a42h, a43l, a43h, a44l, a44h;
+
+#define READ_STATE(state)   do { \
+		a00l = (state)->u.narrow[2 *  0 + 0]; \
+		a00h = (state)->u.narrow[2 *  0 + 1]; \
+		a10l = (state)->u.narrow[2 *  1 + 0]; \
+		a10h = (state)->u.narrow[2 *  1 + 1]; \
+		a20l = (state)->u.narrow[2 *  2 + 0]; \
+		a20h = (state)->u.narrow[2 *  2 + 1]; \
+		a30l = (state)->u.narrow[2 *  3 + 0]; \
+		a30h = (state)->u.narrow[2 *  3 + 1]; \
+		a40l = (state)->u.narrow[2 *  4 + 0]; \
+		a40h = (state)->u.narrow[2 *  4 + 1]; \
+		a01l = (state)->u.narrow[2 *  5 + 0]; \
+		a01h = (state)->u.narrow[2 *  5 + 1]; \
+		a11l = (state)->u.narrow[2 *  6 + 0]; \
+		a11h = (state)->u.narrow[2 *  6 + 1]; \
+		a21l = (state)->u.narrow[2 *  7 + 0]; \
+		a21h = (state)->u.narrow[2 *  7 + 1]; \
+		a31l = (state)->u.narrow[2 *  8 + 0]; \
+		a31h = (state)->u.narrow[2 *  8 + 1]; \
+		a41l = (state)->u.narrow[2 *  9 + 0]; \
+		a41h = (state)->u.narrow[2 *  9 + 1]; \
+		a02l = (state)->u.narrow[2 * 10 + 0]; \
+		a02h = (state)->u.narrow[2 * 10 + 1]; \
+		a12l = (state)->u.narrow[2 * 11 + 0]; \
+		a12h = (state)->u.narrow[2 * 11 + 1]; \
+		a22l = (state)->u.narrow[2 * 12 + 0]; \
+		a22h = (state)->u.narrow[2 * 12 + 1]; \
+		a32l = (state)->u.narrow[2 * 13 + 0]; \
+		a32h = (state)->u.narrow[2 * 13 + 1]; \
+		a42l = (state)->u.narrow[2 * 14 + 0]; \
+		a42h = (state)->u.narrow[2 * 14 + 1]; \
+		a03l = (state)->u.narrow[2 * 15 + 0]; \
+		a03h = (state)->u.narrow[2 * 15 + 1]; \
+		a13l = (state)->u.narrow[2 * 16 + 0]; \
+		a13h = (state)->u.narrow[2 * 16 + 1]; \
+		a23l = (state)->u.narrow[2 * 17 + 0]; \
+		a23h = (state)->u.narrow[2 * 17 + 1]; \
+		a33l = (state)->u.narrow[2 * 18 + 0]; \
+		a33h = (state)->u.narrow[2 * 18 + 1]; \
+		a43l = (state)->u.narrow[2 * 19 + 0]; \
+		a43h = (state)->u.narrow[2 * 19 + 1]; \
+		a04l = (state)->u.narrow[2 * 20 + 0]; \
+		a04h = (state)->u.narrow[2 * 20 + 1]; \
+		a14l = (state)->u.narrow[2 * 21 + 0]; \
+		a14h = (state)->u.narrow[2 * 21 + 1]; \
+		a24l = (state)->u.narrow[2 * 22 + 0]; \
+		a24h = (state)->u.narrow[2 * 22 + 1]; \
+		a34l = (state)->u.narrow[2 * 23 + 0]; \
+		a34h = (state)->u.narrow[2 * 23 + 1]; \
+		a44l = (state)->u.narrow[2 * 24 + 0]; \
+		a44h = (state)->u.narrow[2 * 24 + 1]; \
+	} while (0)
+
+#define WRITE_STATE(state)   do { \
+		(state)->u.narrow[2 *  0 + 0] = a00l; \
+		(state)->u.narrow[2 *  0 + 1] = a00h; \
+		(state)->u.narrow[2 *  1 + 0] = a10l; \
+		(state)->u.narrow[2 *  1 + 1] = a10h; \
+		(state)->u.narrow[2 *  2 + 0] = a20l; \
+		(state)->u.narrow[2 *  2 + 1] = a20h; \
+		(state)->u.narrow[2 *  3 + 0] = a30l; \
+		(state)->u.narrow[2 *  3 + 1] = a30h; \
+		(state)->u.narrow[2 *  4 + 0] = a40l; \
+		(state)->u.narrow[2 *  4 + 1] = a40h; \
+		(state)->u.narrow[2 *  5 + 0] = a01l; \
+		(state)->u.narrow[2 *  5 + 1] = a01h; \
+		(state)->u.narrow[2 *  6 + 0] = a11l; \
+		(state)->u.narrow[2 *  6 + 1] = a11h; \
+		(state)->u.narrow[2 *  7 + 0] = a21l; \
+		(state)->u.narrow[2 *  7 + 1] = a21h; \
+		(state)->u.narrow[2 *  8 + 0] = a31l; \
+		(state)->u.narrow[2 *  8 + 1] = a31h; \
+		(state)->u.narrow[2 *  9 + 0] = a41l; \
+		(state)->u.narrow[2 *  9 + 1] = a41h; \
+		(state)->u.narrow[2 * 10 + 0] = a02l; \
+		(state)->u.narrow[2 * 10 + 1] = a02h; \
+		(state)->u.narrow[2 * 11 + 0] = a12l; \
+		(state)->u.narrow[2 * 11 + 1] = a12h; \
+		(state)->u.narrow[2 * 12 + 0] = a22l; \
+		(state)->u.narrow[2 * 12 + 1] = a22h; \
+		(state)->u.narrow[2 * 13 + 0] = a32l; \
+		(state)->u.narrow[2 * 13 + 1] = a32h; \
+		(state)->u.narrow[2 * 14 + 0] = a42l; \
+		(state)->u.narrow[2 * 14 + 1] = a42h; \
+		(state)->u.narrow[2 * 15 + 0] = a03l; \
+		(state)->u.narrow[2 * 15 + 1] = a03h; \
+		(state)->u.narrow[2 * 16 + 0] = a13l; \
+		(state)->u.narrow[2 * 16 + 1] = a13h; \
+		(state)->u.narrow[2 * 17 + 0] = a23l; \
+		(state)->u.narrow[2 * 17 + 1] = a23h; \
+		(state)->u.narrow[2 * 18 + 0] = a33l; \
+		(state)->u.narrow[2 * 18 + 1] = a33h; \
+		(state)->u.narrow[2 * 19 + 0] = a43l; \
+		(state)->u.narrow[2 * 19 + 1] = a43h; \
+		(state)->u.narrow[2 * 20 + 0] = a04l; \
+		(state)->u.narrow[2 * 20 + 1] = a04h; \
+		(state)->u.narrow[2 * 21 + 0] = a14l; \
+		(state)->u.narrow[2 * 21 + 1] = a14h; \
+		(state)->u.narrow[2 * 22 + 0] = a24l; \
+		(state)->u.narrow[2 * 22 + 1] = a24h; \
+		(state)->u.narrow[2 * 23 + 0] = a34l; \
+		(state)->u.narrow[2 * 23 + 1] = a34h; \
+		(state)->u.narrow[2 * 24 + 0] = a44l; \
+		(state)->u.narrow[2 * 24 + 1] = a44h; \
+	} while (0)
+
+#define READ64(d, off)   do { \
+		sph_u32 tl, th; \
+		tl = sph_dec32le_aligned(buf + (off)); \
+		th = sph_dec32le_aligned(buf + (off) + 4); \
+		INTERLEAVE(tl, th); \
+		d ## l ^= tl; \
+		d ## h ^= th; \
+	} while (0)
+
+#define INPUT_BUF144   do { \
+		READ64(a00,   0); \
+		READ64(a10,   8); \
+		READ64(a20,  16); \
+		READ64(a30,  24); \
+		READ64(a40,  32); \
+		READ64(a01,  40); \
+		READ64(a11,  48); \
+		READ64(a21,  56); \
+		READ64(a31,  64); \
+		READ64(a41,  72); \
+		READ64(a02,  80); \
+		READ64(a12,  88); \
+		READ64(a22,  96); \
+		READ64(a32, 104); \
+		READ64(a42, 112); \
+		READ64(a03, 120); \
+		READ64(a13, 128); \
+		READ64(a23, 136); \
+	} while (0)
+
+#define INPUT_BUF136   do { \
+		READ64(a00,   0); \
+		READ64(a10,   8); \
+		READ64(a20,  16); \
+		READ64(a30,  24); \
+		READ64(a40,  32); \
+		READ64(a01,  40); \
+		READ64(a11,  48); \
+		READ64(a21,  56); \
+		READ64(a31,  64); \
+		READ64(a41,  72); \
+		READ64(a02,  80); \
+		READ64(a12,  88); \
+		READ64(a22,  96); \
+		READ64(a32, 104); \
+		READ64(a42, 112); \
+		READ64(a03, 120); \
+		READ64(a13, 128); \
+	} while (0)
+
+#define INPUT_BUF104   do { \
+		READ64(a00,   0); \
+		READ64(a10,   8); \
+		READ64(a20,  16); \
+		READ64(a30,  24); \
+		READ64(a40,  32); \
+		READ64(a01,  40); \
+		READ64(a11,  48); \
+		READ64(a21,  56); \
+		READ64(a31,  64); \
+		READ64(a41,  72); \
+		READ64(a02,  80); \
+		READ64(a12,  88); \
+		READ64(a22,  96); \
+	} while (0)
+
+#define INPUT_BUF72   do { \
+		READ64(a00,   0); \
+		READ64(a10,   8); \
+		READ64(a20,  16); \
+		READ64(a30,  24); \
+		READ64(a40,  32); \
+		READ64(a01,  40); \
+		READ64(a11,  48); \
+		READ64(a21,  56); \
+		READ64(a31,  64); \
+	} while (0)
+
+#define INPUT_BUF(lim)   do { \
+		READ64(a00,   0); \
+		READ64(a10,   8); \
+		READ64(a20,  16); \
+		READ64(a30,  24); \
+		READ64(a40,  32); \
+		READ64(a01,  40); \
+		READ64(a11,  48); \
+		READ64(a21,  56); \
+		READ64(a31,  64); \
+		if ((lim) == 72) \
+			break; \
+		READ64(a41,  72); \
+		READ64(a02,  80); \
+		READ64(a12,  88); \
+		READ64(a22,  96); \
+		if ((lim) == 104) \
+			break; \
+		READ64(a32, 104); \
+		READ64(a42, 112); \
+		READ64(a03, 120); \
+		READ64(a13, 128); \
+		if ((lim) == 136) \
+			break; \
+		READ64(a23, 136); \
+	} while (0)
+
+#endif
+
+#define DECL64(x)        sph_u64 x ## l, x ## h
+#define MOV64(d, s)      (d ## l = s ## l, d ## h = s ## h)
+#define XOR64(d, a, b)   (d ## l = a ## l ^ b ## l, d ## h = a ## h ^ b ## h)
+#define AND64(d, a, b)   (d ## l = a ## l & b ## l, d ## h = a ## h & b ## h)
+#define OR64(d, a, b)    (d ## l = a ## l | b ## l, d ## h = a ## h | b ## h)
+#define NOT64(d, s)      (d ## l = SPH_T32(~s ## l), d ## h = SPH_T32(~s ## h))
+#define ROL64(d, v, n)   ROL64_ ## n(d, v)
+
+#if SPH_KECCAK_INTERLEAVE
+
+#define ROL64_odd1(d, v)   do { \
+		sph_u32 tmp; \
+		tmp = v ## l; \
+		d ## l = SPH_T32(v ## h << 1) | (v ## h >> 31); \
+		d ## h = tmp; \
+	} while (0)
+
+#define ROL64_odd63(d, v)   do { \
+		sph_u32 tmp; \
+		tmp = SPH_T32(v ## l << 31) | (v ## l >> 1); \
+		d ## l = v ## h; \
+		d ## h = tmp; \
+	} while (0)
+
+#define ROL64_odd(d, v, n)   do { \
+		sph_u32 tmp; \
+		tmp = SPH_T32(v ## l << (n - 1)) | (v ## l >> (33 - n)); \
+		d ## l = SPH_T32(v ## h << n) | (v ## h >> (32 - n)); \
+		d ## h = tmp; \
+	} while (0)
+
+#define ROL64_even(d, v, n)   do { \
+		d ## l = SPH_T32(v ## l << n) | (v ## l >> (32 - n)); \
+		d ## h = SPH_T32(v ## h << n) | (v ## h >> (32 - n)); \
+	} while (0)
+
+#define ROL64_0(d, v)
+#define ROL64_1(d, v)    ROL64_odd1(d, v)
+#define ROL64_2(d, v)    ROL64_even(d, v,  1)
+#define ROL64_3(d, v)    ROL64_odd( d, v,  2)
+#define ROL64_4(d, v)    ROL64_even(d, v,  2)
+#define ROL64_5(d, v)    ROL64_odd( d, v,  3)
+#define ROL64_6(d, v)    ROL64_even(d, v,  3)
+#define ROL64_7(d, v)    ROL64_odd( d, v,  4)
+#define ROL64_8(d, v)    ROL64_even(d, v,  4)
+#define ROL64_9(d, v)    ROL64_odd( d, v,  5)
+#define ROL64_10(d, v)   ROL64_even(d, v,  5)
+#define ROL64_11(d, v)   ROL64_odd( d, v,  6)
+#define ROL64_12(d, v)   ROL64_even(d, v,  6)
+#define ROL64_13(d, v)   ROL64_odd( d, v,  7)
+#define ROL64_14(d, v)   ROL64_even(d, v,  7)
+#define ROL64_15(d, v)   ROL64_odd( d, v,  8)
+#define ROL64_16(d, v)   ROL64_even(d, v,  8)
+#define ROL64_17(d, v)   ROL64_odd( d, v,  9)
+#define ROL64_18(d, v)   ROL64_even(d, v,  9)
+#define ROL64_19(d, v)   ROL64_odd( d, v, 10)
+#define ROL64_20(d, v)   ROL64_even(d, v, 10)
+#define ROL64_21(d, v)   ROL64_odd( d, v, 11)
+#define ROL64_22(d, v)   ROL64_even(d, v, 11)
+#define ROL64_23(d, v)   ROL64_odd( d, v, 12)
+#define ROL64_24(d, v)   ROL64_even(d, v, 12)
+#define ROL64_25(d, v)   ROL64_odd( d, v, 13)
+#define ROL64_26(d, v)   ROL64_even(d, v, 13)
+#define ROL64_27(d, v)   ROL64_odd( d, v, 14)
+#define ROL64_28(d, v)   ROL64_even(d, v, 14)
+#define ROL64_29(d, v)   ROL64_odd( d, v, 15)
+#define ROL64_30(d, v)   ROL64_even(d, v, 15)
+#define ROL64_31(d, v)   ROL64_odd( d, v, 16)
+#define ROL64_32(d, v)   ROL64_even(d, v, 16)
+#define ROL64_33(d, v)   ROL64_odd( d, v, 17)
+#define ROL64_34(d, v)   ROL64_even(d, v, 17)
+#define ROL64_35(d, v)   ROL64_odd( d, v, 18)
+#define ROL64_36(d, v)   ROL64_even(d, v, 18)
+#define ROL64_37(d, v)   ROL64_odd( d, v, 19)
+#define ROL64_38(d, v)   ROL64_even(d, v, 19)
+#define ROL64_39(d, v)   ROL64_odd( d, v, 20)
+#define ROL64_40(d, v)   ROL64_even(d, v, 20)
+#define ROL64_41(d, v)   ROL64_odd( d, v, 21)
+#define ROL64_42(d, v)   ROL64_even(d, v, 21)
+#define ROL64_43(d, v)   ROL64_odd( d, v, 22)
+#define ROL64_44(d, v)   ROL64_even(d, v, 22)
+#define ROL64_45(d, v)   ROL64_odd( d, v, 23)
+#define ROL64_46(d, v)   ROL64_even(d, v, 23)
+#define ROL64_47(d, v)   ROL64_odd( d, v, 24)
+#define ROL64_48(d, v)   ROL64_even(d, v, 24)
+#define ROL64_49(d, v)   ROL64_odd( d, v, 25)
+#define ROL64_50(d, v)   ROL64_even(d, v, 25)
+#define ROL64_51(d, v)   ROL64_odd( d, v, 26)
+#define ROL64_52(d, v)   ROL64_even(d, v, 26)
+#define ROL64_53(d, v)   ROL64_odd( d, v, 27)
+#define ROL64_54(d, v)   ROL64_even(d, v, 27)
+#define ROL64_55(d, v)   ROL64_odd( d, v, 28)
+#define ROL64_56(d, v)   ROL64_even(d, v, 28)
+#define ROL64_57(d, v)   ROL64_odd( d, v, 29)
+#define ROL64_58(d, v)   ROL64_even(d, v, 29)
+#define ROL64_59(d, v)   ROL64_odd( d, v, 30)
+#define ROL64_60(d, v)   ROL64_even(d, v, 30)
+#define ROL64_61(d, v)   ROL64_odd( d, v, 31)
+#define ROL64_62(d, v)   ROL64_even(d, v, 31)
+#define ROL64_63(d, v)   ROL64_odd63(d, v)
+
+#else
+
+#define ROL64_small(d, v, n)   do { \
+		sph_u32 tmp; \
+		tmp = SPH_T32(v ## l << n) | (v ## h >> (32 - n)); \
+		d ## h = SPH_T32(v ## h << n) | (v ## l >> (32 - n)); \
+		d ## l = tmp; \
+	} while (0)
+
+#define ROL64_0(d, v)    0
+#define ROL64_1(d, v)    ROL64_small(d, v, 1)
+#define ROL64_2(d, v)    ROL64_small(d, v, 2)
+#define ROL64_3(d, v)    ROL64_small(d, v, 3)
+#define ROL64_4(d, v)    ROL64_small(d, v, 4)
+#define ROL64_5(d, v)    ROL64_small(d, v, 5)
+#define ROL64_6(d, v)    ROL64_small(d, v, 6)
+#define ROL64_7(d, v)    ROL64_small(d, v, 7)
+#define ROL64_8(d, v)    ROL64_small(d, v, 8)
+#define ROL64_9(d, v)    ROL64_small(d, v, 9)
+#define ROL64_10(d, v)   ROL64_small(d, v, 10)
+#define ROL64_11(d, v)   ROL64_small(d, v, 11)
+#define ROL64_12(d, v)   ROL64_small(d, v, 12)
+#define ROL64_13(d, v)   ROL64_small(d, v, 13)
+#define ROL64_14(d, v)   ROL64_small(d, v, 14)
+#define ROL64_15(d, v)   ROL64_small(d, v, 15)
+#define ROL64_16(d, v)   ROL64_small(d, v, 16)
+#define ROL64_17(d, v)   ROL64_small(d, v, 17)
+#define ROL64_18(d, v)   ROL64_small(d, v, 18)
+#define ROL64_19(d, v)   ROL64_small(d, v, 19)
+#define ROL64_20(d, v)   ROL64_small(d, v, 20)
+#define ROL64_21(d, v)   ROL64_small(d, v, 21)
+#define ROL64_22(d, v)   ROL64_small(d, v, 22)
+#define ROL64_23(d, v)   ROL64_small(d, v, 23)
+#define ROL64_24(d, v)   ROL64_small(d, v, 24)
+#define ROL64_25(d, v)   ROL64_small(d, v, 25)
+#define ROL64_26(d, v)   ROL64_small(d, v, 26)
+#define ROL64_27(d, v)   ROL64_small(d, v, 27)
+#define ROL64_28(d, v)   ROL64_small(d, v, 28)
+#define ROL64_29(d, v)   ROL64_small(d, v, 29)
+#define ROL64_30(d, v)   ROL64_small(d, v, 30)
+#define ROL64_31(d, v)   ROL64_small(d, v, 31)
+
+#define ROL64_32(d, v)   do { \
+		sph_u32 tmp; \
+		tmp = v ## l; \
+		d ## l = v ## h; \
+		d ## h = tmp; \
+	} while (0)
+
+#define ROL64_big(d, v, n)   do { \
+		sph_u32 trl, trh; \
+		ROL64_small(tr, v, n); \
+		d ## h = trl; \
+		d ## l = trh; \
+	} while (0)
+
+#define ROL64_33(d, v)   ROL64_big(d, v, 1)
+#define ROL64_34(d, v)   ROL64_big(d, v, 2)
+#define ROL64_35(d, v)   ROL64_big(d, v, 3)
+#define ROL64_36(d, v)   ROL64_big(d, v, 4)
+#define ROL64_37(d, v)   ROL64_big(d, v, 5)
+#define ROL64_38(d, v)   ROL64_big(d, v, 6)
+#define ROL64_39(d, v)   ROL64_big(d, v, 7)
+#define ROL64_40(d, v)   ROL64_big(d, v, 8)
+#define ROL64_41(d, v)   ROL64_big(d, v, 9)
+#define ROL64_42(d, v)   ROL64_big(d, v, 10)
+#define ROL64_43(d, v)   ROL64_big(d, v, 11)
+#define ROL64_44(d, v)   ROL64_big(d, v, 12)
+#define ROL64_45(d, v)   ROL64_big(d, v, 13)
+#define ROL64_46(d, v)   ROL64_big(d, v, 14)
+#define ROL64_47(d, v)   ROL64_big(d, v, 15)
+#define ROL64_48(d, v)   ROL64_big(d, v, 16)
+#define ROL64_49(d, v)   ROL64_big(d, v, 17)
+#define ROL64_50(d, v)   ROL64_big(d, v, 18)
+#define ROL64_51(d, v)   ROL64_big(d, v, 19)
+#define ROL64_52(d, v)   ROL64_big(d, v, 20)
+#define ROL64_53(d, v)   ROL64_big(d, v, 21)
+#define ROL64_54(d, v)   ROL64_big(d, v, 22)
+#define ROL64_55(d, v)   ROL64_big(d, v, 23)
+#define ROL64_56(d, v)   ROL64_big(d, v, 24)
+#define ROL64_57(d, v)   ROL64_big(d, v, 25)
+#define ROL64_58(d, v)   ROL64_big(d, v, 26)
+#define ROL64_59(d, v)   ROL64_big(d, v, 27)
+#define ROL64_60(d, v)   ROL64_big(d, v, 28)
+#define ROL64_61(d, v)   ROL64_big(d, v, 29)
+#define ROL64_62(d, v)   ROL64_big(d, v, 30)
+#define ROL64_63(d, v)   ROL64_big(d, v, 31)
+
+#endif
+
+#define XOR64_IOTA(d, s, k) \
+	(d ## l = s ## l ^ k.low, d ## h = s ## h ^ k.high)
+
+#endif
+
+#define TH_ELT(t, c0, c1, c2, c3, c4, d0, d1, d2, d3, d4)   do { \
+		DECL64(tt0); \
+		DECL64(tt1); \
+		DECL64(tt2); \
+		DECL64(tt3); \
+		XOR64(tt0, d0, d1); \
+		XOR64(tt1, d2, d3); \
+		XOR64(tt0, tt0, d4); \
+		XOR64(tt0, tt0, tt1); \
+		ROL64(tt0, tt0, 1); \
+		XOR64(tt2, c0, c1); \
+		XOR64(tt3, c2, c3); \
+		XOR64(tt0, tt0, c4); \
+		XOR64(tt2, tt2, tt3); \
+		XOR64(t, tt0, tt2); \
+	} while (0)
+
+#define THETA(b00, b01, b02, b03, b04, b10, b11, b12, b13, b14, \
+	b20, b21, b22, b23, b24, b30, b31, b32, b33, b34, \
+	b40, b41, b42, b43, b44) \
+	do { \
+		DECL64(t0); \
+		DECL64(t1); \
+		DECL64(t2); \
+		DECL64(t3); \
+		DECL64(t4); \
+		TH_ELT(t0, b40, b41, b42, b43, b44, b10, b11, b12, b13, b14); \
+		TH_ELT(t1, b00, b01, b02, b03, b04, b20, b21, b22, b23, b24); \
+		TH_ELT(t2, b10, b11, b12, b13, b14, b30, b31, b32, b33, b34); \
+		TH_ELT(t3, b20, b21, b22, b23, b24, b40, b41, b42, b43, b44); \
+		TH_ELT(t4, b30, b31, b32, b33, b34, b00, b01, b02, b03, b04); \
+		XOR64(b00, b00, t0); \
+		XOR64(b01, b01, t0); \
+		XOR64(b02, b02, t0); \
+		XOR64(b03, b03, t0); \
+		XOR64(b04, b04, t0); \
+		XOR64(b10, b10, t1); \
+		XOR64(b11, b11, t1); \
+		XOR64(b12, b12, t1); \
+		XOR64(b13, b13, t1); \
+		XOR64(b14, b14, t1); \
+		XOR64(b20, b20, t2); \
+		XOR64(b21, b21, t2); \
+		XOR64(b22, b22, t2); \
+		XOR64(b23, b23, t2); \
+		XOR64(b24, b24, t2); \
+		XOR64(b30, b30, t3); \
+		XOR64(b31, b31, t3); \
+		XOR64(b32, b32, t3); \
+		XOR64(b33, b33, t3); \
+		XOR64(b34, b34, t3); \
+		XOR64(b40, b40, t4); \
+		XOR64(b41, b41, t4); \
+		XOR64(b42, b42, t4); \
+		XOR64(b43, b43, t4); \
+		XOR64(b44, b44, t4); \
+	} while (0)
+
+#define RHO(b00, b01, b02, b03, b04, b10, b11, b12, b13, b14, \
+	b20, b21, b22, b23, b24, b30, b31, b32, b33, b34, \
+	b40, b41, b42, b43, b44) \
+	do { \
+		/* ROL64(b00, b00,  0); */ \
+		ROL64(b01, b01, 36); \
+		ROL64(b02, b02,  3); \
+		ROL64(b03, b03, 41); \
+		ROL64(b04, b04, 18); \
+		ROL64(b10, b10,  1); \
+		ROL64(b11, b11, 44); \
+		ROL64(b12, b12, 10); \
+		ROL64(b13, b13, 45); \
+		ROL64(b14, b14,  2); \
+		ROL64(b20, b20, 62); \
+		ROL64(b21, b21,  6); \
+		ROL64(b22, b22, 43); \
+		ROL64(b23, b23, 15); \
+		ROL64(b24, b24, 61); \
+		ROL64(b30, b30, 28); \
+		ROL64(b31, b31, 55); \
+		ROL64(b32, b32, 25); \
+		ROL64(b33, b33, 21); \
+		ROL64(b34, b34, 56); \
+		ROL64(b40, b40, 27); \
+		ROL64(b41, b41, 20); \
+		ROL64(b42, b42, 39); \
+		ROL64(b43, b43,  8); \
+		ROL64(b44, b44, 14); \
+	} while (0)
+
+/*
+ * The KHI macro integrates the "lane complement" optimization. On input,
+ * some words are complemented:
+ *    a00 a01 a02 a04 a13 a20 a21 a22 a30 a33 a34 a43
+ * On output, the following words are complemented:
+ *    a04 a10 a20 a22 a23 a31
+ *
+ * The (implicit) permutation and the theta expansion will bring back
+ * the input mask for the next round.
+ */
+
+#define KHI_XO(d, a, b, c)   do { \
+		DECL64(kt); \
+		OR64(kt, b, c); \
+		XOR64(d, a, kt); \
+	} while (0)
+
+#define KHI_XA(d, a, b, c)   do { \
+		DECL64(kt); \
+		AND64(kt, b, c); \
+		XOR64(d, a, kt); \
+	} while (0)
+
+#define KHI(b00, b01, b02, b03, b04, b10, b11, b12, b13, b14, \
+	b20, b21, b22, b23, b24, b30, b31, b32, b33, b34, \
+	b40, b41, b42, b43, b44) \
+	do { \
+		DECL64(c0); \
+		DECL64(c1); \
+		DECL64(c2); \
+		DECL64(c3); \
+		DECL64(c4); \
+		DECL64(bnn); \
+		NOT64(bnn, b20); \
+		KHI_XO(c0, b00, b10, b20); \
+		KHI_XO(c1, b10, bnn, b30); \
+		KHI_XA(c2, b20, b30, b40); \
+		KHI_XO(c3, b30, b40, b00); \
+		KHI_XA(c4, b40, b00, b10); \
+		MOV64(b00, c0); \
+		MOV64(b10, c1); \
+		MOV64(b20, c2); \
+		MOV64(b30, c3); \
+		MOV64(b40, c4); \
+		NOT64(bnn, b41); \
+		KHI_XO(c0, b01, b11, b21); \
+		KHI_XA(c1, b11, b21, b31); \
+		KHI_XO(c2, b21, b31, bnn); \
+		KHI_XO(c3, b31, b41, b01); \
+		KHI_XA(c4, b41, b01, b11); \
+		MOV64(b01, c0); \
+		MOV64(b11, c1); \
+		MOV64(b21, c2); \
+		MOV64(b31, c3); \
+		MOV64(b41, c4); \
+		NOT64(bnn, b32); \
+		KHI_XO(c0, b02, b12, b22); \
+		KHI_XA(c1, b12, b22, b32); \
+		KHI_XA(c2, b22, bnn, b42); \
+		KHI_XO(c3, bnn, b42, b02); \
+		KHI_XA(c4, b42, b02, b12); \
+		MOV64(b02, c0); \
+		MOV64(b12, c1); \
+		MOV64(b22, c2); \
+		MOV64(b32, c3); \
+		MOV64(b42, c4); \
+		NOT64(bnn, b33); \
+		KHI_XA(c0, b03, b13, b23); \
+		KHI_XO(c1, b13, b23, b33); \
+		KHI_XO(c2, b23, bnn, b43); \
+		KHI_XA(c3, bnn, b43, b03); \
+		KHI_XO(c4, b43, b03, b13); \
+		MOV64(b03, c0); \
+		MOV64(b13, c1); \
+		MOV64(b23, c2); \
+		MOV64(b33, c3); \
+		MOV64(b43, c4); \
+		NOT64(bnn, b14); \
+		KHI_XA(c0, b04, bnn, b24); \
+		KHI_XO(c1, bnn, b24, b34); \
+		KHI_XA(c2, b24, b34, b44); \
+		KHI_XO(c3, b34, b44, b04); \
+		KHI_XA(c4, b44, b04, b14); \
+		MOV64(b04, c0); \
+		MOV64(b14, c1); \
+		MOV64(b24, c2); \
+		MOV64(b34, c3); \
+		MOV64(b44, c4); \
+	} while (0)
+
+#define IOTA(r)   XOR64_IOTA(a00, a00, r)
+
+#define P0    a00, a01, a02, a03, a04, a10, a11, a12, a13, a14, a20, a21, \
+              a22, a23, a24, a30, a31, a32, a33, a34, a40, a41, a42, a43, a44
+#define P1    a00, a30, a10, a40, a20, a11, a41, a21, a01, a31, a22, a02, \
+              a32, a12, a42, a33, a13, a43, a23, a03, a44, a24, a04, a34, a14
+#define P2    a00, a33, a11, a44, a22, a41, a24, a02, a30, a13, a32, a10, \
+              a43, a21, a04, a23, a01, a34, a12, a40, a14, a42, a20, a03, a31
+#define P3    a00, a23, a41, a14, a32, a24, a42, a10, a33, a01, a43, a11, \
+              a34, a02, a20, a12, a30, a03, a21, a44, a31, a04, a22, a40, a13
+#define P4    a00, a12, a24, a31, a43, a42, a04, a11, a23, a30, a34, a41, \
+              a03, a10, a22, a21, a33, a40, a02, a14, a13, a20, a32, a44, a01
+#define P5    a00, a21, a42, a13, a34, a04, a20, a41, a12, a33, a03, a24, \
+              a40, a11, a32, a02, a23, a44, a10, a31, a01, a22, a43, a14, a30
+#define P6    a00, a02, a04, a01, a03, a20, a22, a24, a21, a23, a40, a42, \
+              a44, a41, a43, a10, a12, a14, a11, a13, a30, a32, a34, a31, a33
+#define P7    a00, a10, a20, a30, a40, a22, a32, a42, a02, a12, a44, a04, \
+              a14, a24, a34, a11, a21, a31, a41, a01, a33, a43, a03, a13, a23
+#define P8    a00, a11, a22, a33, a44, a32, a43, a04, a10, a21, a14, a20, \
+              a31, a42, a03, a41, a02, a13, a24, a30, a23, a34, a40, a01, a12
+#define P9    a00, a41, a32, a23, a14, a43, a34, a20, a11, a02, a31, a22, \
+              a13, a04, a40, a24, a10, a01, a42, a33, a12, a03, a44, a30, a21
+#define P10   a00, a24, a43, a12, a31, a34, a03, a22, a41, a10, a13, a32, \
+              a01, a20, a44, a42, a11, a30, a04, a23, a21, a40, a14, a33, a02
+#define P11   a00, a42, a34, a21, a13, a03, a40, a32, a24, a11, a01, a43, \
+              a30, a22, a14, a04, a41, a33, a20, a12, a02, a44, a31, a23, a10
+#define P12   a00, a04, a03, a02, a01, a40, a44, a43, a42, a41, a30, a34, \
+              a33, a32, a31, a20, a24, a23, a22, a21, a10, a14, a13, a12, a11
+#define P13   a00, a20, a40, a10, a30, a44, a14, a34, a04, a24, a33, a03, \
+              a23, a43, a13, a22, a42, a12, a32, a02, a11, a31, a01, a21, a41
+#define P14   a00, a22, a44, a11, a33, a14, a31, a03, a20, a42, a23, a40, \
+              a12, a34, a01, a32, a04, a21, a43, a10, a41, a13, a30, a02, a24
+#define P15   a00, a32, a14, a41, a23, a31, a13, a40, a22, a04, a12, a44, \
+              a21, a03, a30, a43, a20, a02, a34, a11, a24, a01, a33, a10, a42
+#define P16   a00, a43, a31, a24, a12, a13, a01, a44, a32, a20, a21, a14, \
+              a02, a40, a33, a34, a22, a10, a03, a41, a42, a30, a23, a11, a04
+#define P17   a00, a34, a13, a42, a21, a01, a30, a14, a43, a22, a02, a31, \
+              a10, a44, a23, a03, a32, a11, a40, a24, a04, a33, a12, a41, a20
+#define P18   a00, a03, a01, a04, a02, a30, a33, a31, a34, a32, a10, a13, \
+              a11, a14, a12, a40, a43, a41, a44, a42, a20, a23, a21, a24, a22
+#define P19   a00, a40, a30, a20, a10, a33, a23, a13, a03, a43, a11, a01, \
+              a41, a31, a21, a44, a34, a24, a14, a04, a22, a12, a02, a42, a32
+#define P20   a00, a44, a33, a22, a11, a23, a12, a01, a40, a34, a41, a30, \
+              a24, a13, a02, a14, a03, a42, a31, a20, a32, a21, a10, a04, a43
+#define P21   a00, a14, a23, a32, a41, a12, a21, a30, a44, a03, a24, a33, \
+              a42, a01, a10, a31, a40, a04, a13, a22, a43, a02, a11, a20, a34
+#define P22   a00, a31, a12, a43, a24, a21, a02, a33, a14, a40, a42, a23, \
+              a04, a30, a11, a13, a44, a20, a01, a32, a34, a10, a41, a22, a03
+#define P23   a00, a13, a21, a34, a42, a02, a10, a23, a31, a44, a04, a12, \
+              a20, a33, a41, a01, a14, a22, a30, a43, a03, a11, a24, a32, a40
+
+#define P1_TO_P0   do { \
+		DECL64(t); \
+		MOV64(t, a01); \
+		MOV64(a01, a30); \
+		MOV64(a30, a33); \
+		MOV64(a33, a23); \
+		MOV64(a23, a12); \
+		MOV64(a12, a21); \
+		MOV64(a21, a02); \
+		MOV64(a02, a10); \
+		MOV64(a10, a11); \
+		MOV64(a11, a41); \
+		MOV64(a41, a24); \
+		MOV64(a24, a42); \
+		MOV64(a42, a04); \
+		MOV64(a04, a20); \
+		MOV64(a20, a22); \
+		MOV64(a22, a32); \
+		MOV64(a32, a43); \
+		MOV64(a43, a34); \
+		MOV64(a34, a03); \
+		MOV64(a03, a40); \
+		MOV64(a40, a44); \
+		MOV64(a44, a14); \
+		MOV64(a14, a31); \
+		MOV64(a31, a13); \
+		MOV64(a13, t); \
+	} while (0)
+
+#define P2_TO_P0   do { \
+		DECL64(t); \
+		MOV64(t, a01); \
+		MOV64(a01, a33); \
+		MOV64(a33, a12); \
+		MOV64(a12, a02); \
+		MOV64(a02, a11); \
+		MOV64(a11, a24); \
+		MOV64(a24, a04); \
+		MOV64(a04, a22); \
+		MOV64(a22, a43); \
+		MOV64(a43, a03); \
+		MOV64(a03, a44); \
+		MOV64(a44, a31); \
+		MOV64(a31, t); \
+		MOV64(t, a10); \
+		MOV64(a10, a41); \
+		MOV64(a41, a42); \
+		MOV64(a42, a20); \
+		MOV64(a20, a32); \
+		MOV64(a32, a34); \
+		MOV64(a34, a40); \
+		MOV64(a40, a14); \
+		MOV64(a14, a13); \
+		MOV64(a13, a30); \
+		MOV64(a30, a23); \
+		MOV64(a23, a21); \
+		MOV64(a21, t); \
+	} while (0)
+
+#define P4_TO_P0   do { \
+		DECL64(t); \
+		MOV64(t, a01); \
+		MOV64(a01, a12); \
+		MOV64(a12, a11); \
+		MOV64(a11, a04); \
+		MOV64(a04, a43); \
+		MOV64(a43, a44); \
+		MOV64(a44, t); \
+		MOV64(t, a02); \
+		MOV64(a02, a24); \
+		MOV64(a24, a22); \
+		MOV64(a22, a03); \
+		MOV64(a03, a31); \
+		MOV64(a31, a33); \
+		MOV64(a33, t); \
+		MOV64(t, a10); \
+		MOV64(a10, a42); \
+		MOV64(a42, a32); \
+		MOV64(a32, a40); \
+		MOV64(a40, a13); \
+		MOV64(a13, a23); \
+		MOV64(a23, t); \
+		MOV64(t, a14); \
+		MOV64(a14, a30); \
+		MOV64(a30, a21); \
+		MOV64(a21, a41); \
+		MOV64(a41, a20); \
+		MOV64(a20, a34); \
+		MOV64(a34, t); \
+	} while (0)
+
+#define P6_TO_P0   do { \
+		DECL64(t); \
+		MOV64(t, a01); \
+		MOV64(a01, a02); \
+		MOV64(a02, a04); \
+		MOV64(a04, a03); \
+		MOV64(a03, t); \
+		MOV64(t, a10); \
+		MOV64(a10, a20); \
+		MOV64(a20, a40); \
+		MOV64(a40, a30); \
+		MOV64(a30, t); \
+		MOV64(t, a11); \
+		MOV64(a11, a22); \
+		MOV64(a22, a44); \
+		MOV64(a44, a33); \
+		MOV64(a33, t); \
+		MOV64(t, a12); \
+		MOV64(a12, a24); \
+		MOV64(a24, a43); \
+		MOV64(a43, a31); \
+		MOV64(a31, t); \
+		MOV64(t, a13); \
+		MOV64(a13, a21); \
+		MOV64(a21, a42); \
+		MOV64(a42, a34); \
+		MOV64(a34, t); \
+		MOV64(t, a14); \
+		MOV64(a14, a23); \
+		MOV64(a23, a41); \
+		MOV64(a41, a32); \
+		MOV64(a32, t); \
+	} while (0)
+
+#define P8_TO_P0   do { \
+		DECL64(t); \
+		MOV64(t, a01); \
+		MOV64(a01, a11); \
+		MOV64(a11, a43); \
+		MOV64(a43, t); \
+		MOV64(t, a02); \
+		MOV64(a02, a22); \
+		MOV64(a22, a31); \
+		MOV64(a31, t); \
+		MOV64(t, a03); \
+		MOV64(a03, a33); \
+		MOV64(a33, a24); \
+		MOV64(a24, t); \
+		MOV64(t, a04); \
+		MOV64(a04, a44); \
+		MOV64(a44, a12); \
+		MOV64(a12, t); \
+		MOV64(t, a10); \
+		MOV64(a10, a32); \
+		MOV64(a32, a13); \
+		MOV64(a13, t); \
+		MOV64(t, a14); \
+		MOV64(a14, a21); \
+		MOV64(a21, a20); \
+		MOV64(a20, t); \
+		MOV64(t, a23); \
+		MOV64(a23, a42); \
+		MOV64(a42, a40); \
+		MOV64(a40, t); \
+		MOV64(t, a30); \
+		MOV64(a30, a41); \
+		MOV64(a41, a34); \
+		MOV64(a34, t); \
+	} while (0)
+
+#define P12_TO_P0   do { \
+		DECL64(t); \
+		MOV64(t, a01); \
+		MOV64(a01, a04); \
+		MOV64(a04, t); \
+		MOV64(t, a02); \
+		MOV64(a02, a03); \
+		MOV64(a03, t); \
+		MOV64(t, a10); \
+		MOV64(a10, a40); \
+		MOV64(a40, t); \
+		MOV64(t, a11); \
+		MOV64(a11, a44); \
+		MOV64(a44, t); \
+		MOV64(t, a12); \
+		MOV64(a12, a43); \
+		MOV64(a43, t); \
+		MOV64(t, a13); \
+		MOV64(a13, a42); \
+		MOV64(a42, t); \
+		MOV64(t, a14); \
+		MOV64(a14, a41); \
+		MOV64(a41, t); \
+		MOV64(t, a20); \
+		MOV64(a20, a30); \
+		MOV64(a30, t); \
+		MOV64(t, a21); \
+		MOV64(a21, a34); \
+		MOV64(a34, t); \
+		MOV64(t, a22); \
+		MOV64(a22, a33); \
+		MOV64(a33, t); \
+		MOV64(t, a23); \
+		MOV64(a23, a32); \
+		MOV64(a32, t); \
+		MOV64(t, a24); \
+		MOV64(a24, a31); \
+		MOV64(a31, t); \
+	} while (0)
+
+#define LPAR   (
+#define RPAR   )
+
+#define KF_ELT(r, s, k)   do { \
+		THETA LPAR P ## r RPAR; \
+		RHO LPAR P ## r RPAR; \
+		KHI LPAR P ## s RPAR; \
+		IOTA(k); \
+	} while (0)
+
+#define DO(x)   x
+
+#define KECCAK_F_1600   DO(KECCAK_F_1600_)
+
+#if SPH_KECCAK_UNROLL == 1
+
+#define KECCAK_F_1600_   do { \
+		int j; \
+		for (j = 0; j < 24; j ++) { \
+			KF_ELT( 0,  1, RC[j + 0]); \
+			P1_TO_P0; \
+		} \
+	} while (0)
+
+#elif SPH_KECCAK_UNROLL == 2
+
+#define KECCAK_F_1600_   do { \
+		int j; \
+		for (j = 0; j < 24; j += 2) { \
+			KF_ELT( 0,  1, RC[j + 0]); \
+			KF_ELT( 1,  2, RC[j + 1]); \
+			P2_TO_P0; \
+		} \
+	} while (0)
+
+#elif SPH_KECCAK_UNROLL == 4
+
+#define KECCAK_F_1600_   do { \
+		int j; \
+		for (j = 0; j < 24; j += 4) { \
+			KF_ELT( 0,  1, RC[j + 0]); \
+			KF_ELT( 1,  2, RC[j + 1]); \
+			KF_ELT( 2,  3, RC[j + 2]); \
+			KF_ELT( 3,  4, RC[j + 3]); \
+			P4_TO_P0; \
+		} \
+	} while (0)
+
+#elif SPH_KECCAK_UNROLL == 6
+
+#define KECCAK_F_1600_   do { \
+		int j; \
+		for (j = 0; j < 24; j += 6) { \
+			KF_ELT( 0,  1, RC[j + 0]); \
+			KF_ELT( 1,  2, RC[j + 1]); \
+			KF_ELT( 2,  3, RC[j + 2]); \
+			KF_ELT( 3,  4, RC[j + 3]); \
+			KF_ELT( 4,  5, RC[j + 4]); \
+			KF_ELT( 5,  6, RC[j + 5]); \
+			P6_TO_P0; \
+		} \
+	} while (0)
+
+#elif SPH_KECCAK_UNROLL == 8
+
+#define KECCAK_F_1600_   do { \
+		int j; \
+		for (j = 0; j < 24; j += 8) { \
+			KF_ELT( 0,  1, RC[j + 0]); \
+			KF_ELT( 1,  2, RC[j + 1]); \
+			KF_ELT( 2,  3, RC[j + 2]); \
+			KF_ELT( 3,  4, RC[j + 3]); \
+			KF_ELT( 4,  5, RC[j + 4]); \
+			KF_ELT( 5,  6, RC[j + 5]); \
+			KF_ELT( 6,  7, RC[j + 6]); \
+			KF_ELT( 7,  8, RC[j + 7]); \
+			P8_TO_P0; \
+		} \
+	} while (0)
+
+#elif SPH_KECCAK_UNROLL == 12
+
+#define KECCAK_F_1600_   do { \
+		int j; \
+		for (j = 0; j < 24; j += 12) { \
+			KF_ELT( 0,  1, RC[j +  0]); \
+			KF_ELT( 1,  2, RC[j +  1]); \
+			KF_ELT( 2,  3, RC[j +  2]); \
+			KF_ELT( 3,  4, RC[j +  3]); \
+			KF_ELT( 4,  5, RC[j +  4]); \
+			KF_ELT( 5,  6, RC[j +  5]); \
+			KF_ELT( 6,  7, RC[j +  6]); \
+			KF_ELT( 7,  8, RC[j +  7]); \
+			KF_ELT( 8,  9, RC[j +  8]); \
+			KF_ELT( 9, 10, RC[j +  9]); \
+			KF_ELT(10, 11, RC[j + 10]); \
+			KF_ELT(11, 12, RC[j + 11]); \
+			P12_TO_P0; \
+		} \
+	} while (0)
+
+#elif SPH_KECCAK_UNROLL == 0
+
+#define KECCAK_F_1600_   do { \
+		KF_ELT( 0,  1, RC[ 0]); \
+		KF_ELT( 1,  2, RC[ 1]); \
+		KF_ELT( 2,  3, RC[ 2]); \
+		KF_ELT( 3,  4, RC[ 3]); \
+		KF_ELT( 4,  5, RC[ 4]); \
+		KF_ELT( 5,  6, RC[ 5]); \
+		KF_ELT( 6,  7, RC[ 6]); \
+		KF_ELT( 7,  8, RC[ 7]); \
+		KF_ELT( 8,  9, RC[ 8]); \
+		KF_ELT( 9, 10, RC[ 9]); \
+		KF_ELT(10, 11, RC[10]); \
+		KF_ELT(11, 12, RC[11]); \
+		KF_ELT(12, 13, RC[12]); \
+		KF_ELT(13, 14, RC[13]); \
+		KF_ELT(14, 15, RC[14]); \
+		KF_ELT(15, 16, RC[15]); \
+		KF_ELT(16, 17, RC[16]); \
+		KF_ELT(17, 18, RC[17]); \
+		KF_ELT(18, 19, RC[18]); \
+		KF_ELT(19, 20, RC[19]); \
+		KF_ELT(20, 21, RC[20]); \
+		KF_ELT(21, 22, RC[21]); \
+		KF_ELT(22, 23, RC[22]); \
+		KF_ELT(23,  0, RC[23]); \
+	} while (0)
+
+#else
+
+#error Unimplemented unroll count for Keccak.
+
+#endif
+
+static void
+keccak_init(sph_keccak_context *kc, unsigned out_size)
+{
+	int i;
+
+#if SPH_KECCAK_64
+	for (i = 0; i < 25; i ++)
+		kc->u.wide[i] = 0;
+	/*
+	 * Initialization for the "lane complement".
+	 */
+	kc->u.wide[ 1] = SPH_C64(0xFFFFFFFFFFFFFFFF);
+	kc->u.wide[ 2] = SPH_C64(0xFFFFFFFFFFFFFFFF);
+	kc->u.wide[ 8] = SPH_C64(0xFFFFFFFFFFFFFFFF);
+	kc->u.wide[12] = SPH_C64(0xFFFFFFFFFFFFFFFF);
+	kc->u.wide[17] = SPH_C64(0xFFFFFFFFFFFFFFFF);
+	kc->u.wide[20] = SPH_C64(0xFFFFFFFFFFFFFFFF);
+#else
+
+	for (i = 0; i < 50; i ++)
+		kc->u.narrow[i] = 0;
+	/*
+	 * Initialization for the "lane complement".
+	 * Note: since we set to all-one full 64-bit words,
+	 * interleaving (if applicable) is a no-op.
+	 */
+	kc->u.narrow[ 2] = SPH_C32(0xFFFFFFFF);
+	kc->u.narrow[ 3] = SPH_C32(0xFFFFFFFF);
+	kc->u.narrow[ 4] = SPH_C32(0xFFFFFFFF);
+	kc->u.narrow[ 5] = SPH_C32(0xFFFFFFFF);
+	kc->u.narrow[16] = SPH_C32(0xFFFFFFFF);
+	kc->u.narrow[17] = SPH_C32(0xFFFFFFFF);
+	kc->u.narrow[24] = SPH_C32(0xFFFFFFFF);
+	kc->u.narrow[25] = SPH_C32(0xFFFFFFFF);
+	kc->u.narrow[34] = SPH_C32(0xFFFFFFFF);
+	kc->u.narrow[35] = SPH_C32(0xFFFFFFFF);
+	kc->u.narrow[40] = SPH_C32(0xFFFFFFFF);
+	kc->u.narrow[41] = SPH_C32(0xFFFFFFFF);
+#endif
+	kc->ptr = 0;
+	kc->lim = 200 - (out_size >> 2);
+}
+
+static void
+keccak_core(sph_keccak_context *kc, const void *data, size_t len, size_t lim)
+{
+	unsigned char *buf;
+	size_t ptr;
+	DECL_STATE
+
+	buf = kc->buf;
+	ptr = kc->ptr;
+
+	if (len < (lim - ptr)) {
+		memcpy(buf + ptr, data, len);
+		kc->ptr = ptr + len;
+		return;
+	}
+
+	READ_STATE(kc);
+	while (len > 0) {
+		size_t clen;
+
+		clen = (lim - ptr);
+		if (clen > len)
+			clen = len;
+		memcpy(buf + ptr, data, clen);
+		ptr += clen;
+		data = (const unsigned char *)data + clen;
+		len -= clen;
+		if (ptr == lim) {
+			INPUT_BUF(lim);
+			KECCAK_F_1600;
+			ptr = 0;
+		}
+	}
+	WRITE_STATE(kc);
+	kc->ptr = ptr;
+}
+
+#if SPH_KECCAK_64
+
+#define DEFCLOSE(d, lim) \
+	static void keccak_close ## d( \
+		sph_keccak_context *kc, unsigned ub, unsigned n, void *dst) \
+	{ \
+		unsigned eb; \
+		union { \
+			unsigned char tmp[lim + 1]; \
+			sph_u64 dummy;   /* for alignment */ \
+		} u; \
+		size_t j; \
+ \
+		eb = (0x100 | (ub & 0xFF)) >> (8 - n); \
+		if (kc->ptr == (lim - 1)) { \
+			if (n == 7) { \
+				u.tmp[0] = eb; \
+				memset(u.tmp + 1, 0, lim - 1); \
+				u.tmp[lim] = 0x80; \
+				j = 1 + lim; \
+			} else { \
+				u.tmp[0] = eb | 0x80; \
+				j = 1; \
+			} \
+		} else { \
+			j = lim - kc->ptr; \
+			u.tmp[0] = eb; \
+			memset(u.tmp + 1, 0, j - 2); \
+			u.tmp[j - 1] = 0x80; \
+		} \
+		keccak_core(kc, u.tmp, j, lim); \
+		/* Finalize the "lane complement" */ \
+		kc->u.wide[ 1] = ~kc->u.wide[ 1]; \
+		kc->u.wide[ 2] = ~kc->u.wide[ 2]; \
+		kc->u.wide[ 8] = ~kc->u.wide[ 8]; \
+		kc->u.wide[12] = ~kc->u.wide[12]; \
+		kc->u.wide[17] = ~kc->u.wide[17]; \
+		kc->u.wide[20] = ~kc->u.wide[20]; \
+		for (j = 0; j < d; j += 8) \
+			sph_enc64le_aligned(u.tmp + j, kc->u.wide[j >> 3]); \
+		memcpy(dst, u.tmp, d); \
+		keccak_init(kc, (unsigned)d << 3); \
+	} \
+
+#else
+
+#define DEFCLOSE(d, lim) \
+	static void keccak_close ## d( \
+		sph_keccak_context *kc, unsigned ub, unsigned n, void *dst) \
+	{ \
+		unsigned eb; \
+		union { \
+			unsigned char tmp[lim + 1]; \
+			sph_u64 dummy;   /* for alignment */ \
+		} u; \
+		size_t j; \
+ \
+		eb = (0x100 | (ub & 0xFF)) >> (8 - n); \
+		if (kc->ptr == (lim - 1)) { \
+			if (n == 7) { \
+				u.tmp[0] = eb; \
+				memset(u.tmp + 1, 0, lim - 1); \
+				u.tmp[lim] = 0x80; \
+				j = 1 + lim; \
+			} else { \
+				u.tmp[0] = eb | 0x80; \
+				j = 1; \
+			} \
+		} else { \
+			j = lim - kc->ptr; \
+			u.tmp[0] = eb; \
+			memset(u.tmp + 1, 0, j - 2); \
+			u.tmp[j - 1] = 0x80; \
+		} \
+		keccak_core(kc, u.tmp, j, lim); \
+		/* Finalize the "lane complement" */ \
+		kc->u.narrow[ 2] = ~kc->u.narrow[ 2]; \
+		kc->u.narrow[ 3] = ~kc->u.narrow[ 3]; \
+		kc->u.narrow[ 4] = ~kc->u.narrow[ 4]; \
+		kc->u.narrow[ 5] = ~kc->u.narrow[ 5]; \
+		kc->u.narrow[16] = ~kc->u.narrow[16]; \
+		kc->u.narrow[17] = ~kc->u.narrow[17]; \
+		kc->u.narrow[24] = ~kc->u.narrow[24]; \
+		kc->u.narrow[25] = ~kc->u.narrow[25]; \
+		kc->u.narrow[34] = ~kc->u.narrow[34]; \
+		kc->u.narrow[35] = ~kc->u.narrow[35]; \
+		kc->u.narrow[40] = ~kc->u.narrow[40]; \
+		kc->u.narrow[41] = ~kc->u.narrow[41]; \
+		/* un-interleave */ \
+		for (j = 0; j < 50; j += 2) \
+			UNINTERLEAVE(kc->u.narrow[j], kc->u.narrow[j + 1]); \
+		for (j = 0; j < d; j += 4) \
+			sph_enc32le_aligned(u.tmp + j, kc->u.narrow[j >> 2]); \
+		memcpy(dst, u.tmp, d); \
+		keccak_init(kc, (unsigned)d << 3); \
+	} \
+
+#endif
+
+DEFCLOSE(28, 144)
+DEFCLOSE(32, 136)
+DEFCLOSE(48, 104)
+DEFCLOSE(64, 72)
+
+/* see sph_keccak.h */
+void
+sph_keccak224_init(void *cc)
+{
+	keccak_init(cc, 224);
+}
+
+/* see sph_keccak.h */
+void
+sph_keccak224(void *cc, const void *data, size_t len)
+{
+	keccak_core(cc, data, len, 144);
+}
+
+/* see sph_keccak.h */
+void
+sph_keccak224_close(void *cc, void *dst)
+{
+	sph_keccak224_addbits_and_close(cc, 0, 0, dst);
+}
+
+/* see sph_keccak.h */
+void
+sph_keccak224_addbits_and_close(void *cc, unsigned ub, unsigned n, void *dst)
+{
+	keccak_close28(cc, ub, n, dst);
+}
+
+/* see sph_keccak.h */
+void
+sph_keccak256_init(void *cc)
+{
+	keccak_init(cc, 256);
+}
+
+/* see sph_keccak.h */
+void
+sph_keccak256(void *cc, const void *data, size_t len)
+{
+	keccak_core(cc, data, len, 136);
+}
+
+/* see sph_keccak.h */
+void
+sph_keccak256_close(void *cc, void *dst)
+{
+	sph_keccak256_addbits_and_close(cc, 0, 0, dst);
+}
+
+/* see sph_keccak.h */
+void
+sph_keccak256_addbits_and_close(void *cc, unsigned ub, unsigned n, void *dst)
+{
+	keccak_close32(cc, ub, n, dst);
+}
+
+/* see sph_keccak.h */
+void
+sph_keccak384_init(void *cc)
+{
+	keccak_init(cc, 384);
+}
+
+/* see sph_keccak.h */
+void
+sph_keccak384(void *cc, const void *data, size_t len)
+{
+	keccak_core(cc, data, len, 104);
+}
+
+/* see sph_keccak.h */
+void
+sph_keccak384_close(void *cc, void *dst)
+{
+	sph_keccak384_addbits_and_close(cc, 0, 0, dst);
+}
+
+/* see sph_keccak.h */
+void
+sph_keccak384_addbits_and_close(void *cc, unsigned ub, unsigned n, void *dst)
+{
+	keccak_close48(cc, ub, n, dst);
+}
+
+/* see sph_keccak.h */
+void
+sph_keccak512_init(void *cc)
+{
+	keccak_init(cc, 512);
+}
+
+/* see sph_keccak.h */
+void
+sph_keccak512(void *cc, const void *data, size_t len)
+{
+	keccak_core(cc, data, len, 72);
+}
+
+/* see sph_keccak.h */
+void
+sph_keccak512_close(void *cc, void *dst)
+{
+	sph_keccak512_addbits_and_close(cc, 0, 0, dst);
+}
+
+/* see sph_keccak.h */
+void
+sph_keccak512_addbits_and_close(void *cc, unsigned ub, unsigned n, void *dst)
+{
+	keccak_close64(cc, ub, n, dst);
+}
+
+
+#ifdef __cplusplus
+}
+#endif
diff --git a/miner.h b/miner.h
new file mode 100644
index 0000000000..0a743e7e18
--- /dev/null
+++ b/miner.h
@@ -0,0 +1,315 @@
+#ifndef __MINER_H__
+#define __MINER_H__
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#include "cpuminer-config.h"
+
+#include <stdbool.h>
+#include <inttypes.h>
+#include <sys/time.h>
+#include <pthread.h>
+#include <jansson.h>
+#include <curl/curl.h>
+
+#ifdef WIN32
+#define snprintf(...) _snprintf(__VA_ARGS__)
+#define strdup(x) _strdup(x)
+#define strncasecmp(x,y,z) _strnicmp(x,y,z)
+#define strcasecmp(x,y) _stricmp(x,y)
+typedef int ssize_t;
+#endif
+
+#ifdef STDC_HEADERS
+# include <stdlib.h>
+# include <stddef.h>
+#else
+# ifdef HAVE_STDLIB_H
+#  include <stdlib.h>
+# endif
+#endif
+#ifdef HAVE_ALLOCA_H
+# include <alloca.h>
+#elif !defined alloca
+# ifdef __GNUC__
+#  define alloca __builtin_alloca
+# elif defined _AIX
+#  define alloca __alloca
+# elif defined _MSC_VER
+#  include <malloc.h>
+#  define alloca _alloca
+# elif !defined HAVE_ALLOCA
+#  ifdef  __cplusplus
+extern "C"
+#  endif
+void *alloca (size_t);
+# endif
+#endif
+
+#ifdef HAVE_SYSLOG_H
+#include <syslog.h>
+#else
+enum {
+	LOG_ERR,
+	LOG_WARNING,
+	LOG_NOTICE,
+	LOG_INFO,
+	LOG_DEBUG,
+};
+#endif
+
+#undef unlikely
+#undef likely
+#if defined(__GNUC__) && (__GNUC__ > 2) && defined(__OPTIMIZE__)
+#define unlikely(expr) (__builtin_expect(!!(expr), 0))
+#define likely(expr) (__builtin_expect(!!(expr), 1))
+#else
+#define unlikely(expr) (expr)
+#define likely(expr) (expr)
+#endif
+
+#ifndef ARRAY_SIZE
+#define ARRAY_SIZE(arr) (sizeof(arr) / sizeof((arr)[0]))
+#endif
+
+#if ((__GNUC__ > 4) || (__GNUC__ == 4 && __GNUC_MINOR__ >= 3))
+#define WANT_BUILTIN_BSWAP
+#else
+#define bswap_32(x) ((((x) << 24) & 0xff000000u) | (((x) << 8) & 0x00ff0000u) \
+                   | (((x) >> 8) & 0x0000ff00u) | (((x) >> 24) & 0x000000ffu))
+#endif
+
+static inline uint32_t swab32(uint32_t v)
+{
+#ifdef WANT_BUILTIN_BSWAP
+	return __builtin_bswap32(v);
+#else
+	return bswap_32(v);
+#endif
+}
+
+#ifdef HAVE_SYS_ENDIAN_H
+#include <sys/endian.h>
+#endif
+
+#if !HAVE_DECL_BE32DEC
+static inline uint32_t be32dec(const void *pp)
+{
+	const uint8_t *p = (uint8_t const *)pp;
+	return ((uint32_t)(p[3]) + ((uint32_t)(p[2]) << 8) +
+	    ((uint32_t)(p[1]) << 16) + ((uint32_t)(p[0]) << 24));
+}
+#endif
+
+#if !HAVE_DECL_LE32DEC
+static inline uint32_t le32dec(const void *pp)
+{
+	const uint8_t *p = (uint8_t const *)pp;
+	return ((uint32_t)(p[0]) + ((uint32_t)(p[1]) << 8) +
+	    ((uint32_t)(p[2]) << 16) + ((uint32_t)(p[3]) << 24));
+}
+#endif
+
+#if !HAVE_DECL_BE32ENC
+static inline void be32enc(void *pp, uint32_t x)
+{
+	uint8_t *p = (uint8_t *)pp;
+	p[3] = x & 0xff;
+	p[2] = (x >> 8) & 0xff;
+	p[1] = (x >> 16) & 0xff;
+	p[0] = (x >> 24) & 0xff;
+}
+#endif
+
+#if !HAVE_DECL_LE32ENC
+static inline void le32enc(void *pp, uint32_t x)
+{
+	uint8_t *p = (uint8_t *)pp;
+	p[0] = x & 0xff;
+	p[1] = (x >> 8) & 0xff;
+	p[2] = (x >> 16) & 0xff;
+	p[3] = (x >> 24) & 0xff;
+}
+#endif
+
+#if !HAVE_DECL_BE16DEC
+static inline uint16_t be16dec(const void *pp)
+{
+	const uint8_t *p = (uint8_t const *)pp;
+	return ((uint16_t)(p[1]) + ((uint16_t)(p[0]) << 8));
+}
+#endif
+
+#if !HAVE_DECL_BE16ENC
+static inline void be16enc(void *pp, uint16_t x)
+{
+	uint8_t *p = (uint8_t *)pp;
+	p[1] = x & 0xff;
+	p[0] = (x >> 8) & 0xff;
+}
+#endif
+
+#if !HAVE_DECL_LE16DEC
+static inline uint16_t le16dec(const void *pp)
+{
+	const uint8_t *p = (uint8_t const *)pp;
+	return ((uint16_t)(p[0]) + ((uint16_t)(p[1]) << 8));
+}
+#endif
+
+#if !HAVE_DECL_LE16ENC
+static inline void le16enc(void *pp, uint16_t x)
+{
+	uint8_t *p = (uint8_t *)pp;
+	p[0] = x & 0xff;
+	p[1] = (x >> 8) & 0xff;
+}
+#endif
+
+#if JANSSON_MAJOR_VERSION >= 2
+#define JSON_LOADS(str, err_ptr) json_loads((str), 0, (err_ptr))
+#else
+#define JSON_LOADS(str, err_ptr) json_loads((str), (err_ptr))
+#endif
+
+#define USER_AGENT PACKAGE_NAME "/" PACKAGE_VERSION
+
+void sha256_init(uint32_t *state);
+void sha256_transform(uint32_t *state, const uint32_t *block, int swap);
+void sha256d(unsigned char *hash, const unsigned char *data, int len);
+
+#if defined(__ARM_NEON__) || defined(__i386__) || defined(__x86_64__)
+#define HAVE_SHA256_4WAY 0
+int sha256_use_4way();
+void sha256_init_4way(uint32_t *state);
+void sha256_transform_4way(uint32_t *state, const uint32_t *block, int swap);
+#endif
+
+#if defined(__x86_64__) && defined(USE_AVX2)
+#define HAVE_SHA256_8WAY 0
+int sha256_use_8way();
+void sha256_init_8way(uint32_t *state);
+void sha256_transform_8way(uint32_t *state, const uint32_t *block, int swap);
+#endif
+
+extern int scanhash_sha256d(int thr_id, uint32_t *pdata,
+	const uint32_t *ptarget, uint32_t max_nonce, unsigned long *hashes_done);
+
+extern unsigned char *scrypt_buffer_alloc();
+
+extern int scanhash_scrypt(int thr_id, uint32_t *pdata,
+	unsigned char *scratchbuf, const uint32_t *ptarget,
+	uint32_t max_nonce, unsigned long *hashes_done);
+
+extern int scanhash_heavy(int thr_id, uint32_t *pdata,
+	const uint32_t *ptarget, uint32_t max_nonce,
+	unsigned long *hashes_done, uint32_t maxvote);
+
+extern int scanhash_fugue256(int thr_id, uint32_t *pdata,
+	const uint32_t *ptarget, uint32_t max_nonce,
+	unsigned long *hashes_done);
+
+extern void heavycoin_hash(unsigned char* output, const unsigned char* input, int len);
+
+struct thr_info {
+	int		id;
+	pthread_t	pth;
+	struct thread_q	*q;
+};
+
+struct work_restart {
+	volatile unsigned long	restart;
+	char			padding[128 - sizeof(unsigned long)];
+};
+
+extern bool opt_debug;
+extern bool opt_protocol;
+extern int opt_timeout;
+extern bool want_longpoll;
+extern bool have_longpoll;
+extern bool want_stratum;
+extern bool have_stratum;
+extern char *opt_cert;
+extern char *opt_proxy;
+extern long opt_proxy_type;
+extern bool use_syslog;
+extern pthread_mutex_t applog_lock;
+extern struct thr_info *thr_info;
+extern int longpoll_thr_id;
+extern int stratum_thr_id;
+extern struct work_restart *work_restart;
+extern bool opt_trust_pool;
+extern uint16_t opt_vote;
+
+extern void applog(int prio, const char *fmt, ...);
+extern json_t *json_rpc_call(CURL *curl, const char *url, const char *userpass,
+	const char *rpc_req, bool, bool, int *);
+extern char *bin2hex(const unsigned char *p, size_t len);
+extern bool hex2bin(unsigned char *p, const char *hexstr, size_t len);
+extern int timeval_subtract(struct timeval *result, struct timeval *x,
+	struct timeval *y);
+extern bool fulltest(const uint32_t *hash, const uint32_t *target);
+extern void diff_to_target(uint32_t *target, double diff);
+
+struct stratum_job {
+	char *job_id;
+	unsigned char prevhash[32];
+	size_t coinbase_size;
+	unsigned char *coinbase;
+	unsigned char *xnonce2;
+	int merkle_count;
+	unsigned char **merkle;
+	unsigned char version[4];
+	unsigned char nbits[4];
+	unsigned char ntime[4];
+	bool clean;
+	unsigned char nreward[2];
+	double diff;
+};
+
+struct stratum_ctx {
+	char *url;
+
+	CURL *curl;
+	char *curl_url;
+	char curl_err_str[CURL_ERROR_SIZE];
+	curl_socket_t sock;
+	size_t sockbuf_size;
+	char *sockbuf;
+	pthread_mutex_t sock_lock;
+
+	double next_diff;
+
+	char *session_id;
+	size_t xnonce1_size;
+	unsigned char *xnonce1;
+	size_t xnonce2_size;
+	struct stratum_job job;
+	pthread_mutex_t work_lock;
+};
+
+bool stratum_socket_full(struct stratum_ctx *sctx, int timeout);
+bool stratum_send_line(struct stratum_ctx *sctx, char *s);
+char *stratum_recv_line(struct stratum_ctx *sctx);
+bool stratum_connect(struct stratum_ctx *sctx, const char *url);
+void stratum_disconnect(struct stratum_ctx *sctx);
+bool stratum_subscribe(struct stratum_ctx *sctx);
+bool stratum_authorize(struct stratum_ctx *sctx, const char *user, const char *pass);
+bool stratum_handle_method(struct stratum_ctx *sctx, const char *s);
+
+struct thread_q;
+
+extern struct thread_q *tq_new(void);
+extern void tq_free(struct thread_q *tq);
+extern bool tq_push(struct thread_q *tq, void *data);
+extern void *tq_pop(struct thread_q *tq, const struct timespec *abstime);
+extern void tq_freeze(struct thread_q *tq);
+extern void tq_thaw(struct thread_q *tq);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* __MINER_H__ */
diff --git a/missing b/missing
new file mode 100644
index 0000000000..1c8ff7049d
--- /dev/null
+++ b/missing
@@ -0,0 +1,367 @@
+#! /bin/sh
+# Common stub for a few missing GNU programs while installing.
+
+scriptversion=2006-05-10.23
+
+# Copyright (C) 1996, 1997, 1999, 2000, 2002, 2003, 2004, 2005, 2006
+#   Free Software Foundation, Inc.
+# Originally by Fran,cois Pinard <pinard@iro.umontreal.ca>, 1996.
+
+# This program is free software; you can redistribute it and/or modify
+# it under the terms of the GNU General Public License as published by
+# the Free Software Foundation; either version 2, or (at your option)
+# any later version.
+
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+# GNU General Public License for more details.
+
+# You should have received a copy of the GNU General Public License
+# along with this program; if not, write to the Free Software
+# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
+# 02110-1301, USA.
+
+# As a special exception to the GNU General Public License, if you
+# distribute this file as part of a program that contains a
+# configuration script generated by Autoconf, you may include it under
+# the same distribution terms that you use for the rest of that program.
+
+if test $# -eq 0; then
+  echo 1>&2 "Try \`$0 --help' for more information"
+  exit 1
+fi
+
+run=:
+sed_output='s/.* --output[ =]\([^ ]*\).*/\1/p'
+sed_minuso='s/.* -o \([^ ]*\).*/\1/p'
+
+# In the cases where this matters, `missing' is being run in the
+# srcdir already.
+if test -f configure.ac; then
+  configure_ac=configure.ac
+else
+  configure_ac=configure.in
+fi
+
+msg="missing on your system"
+
+case $1 in
+--run)
+  # Try to run requested program, and just exit if it succeeds.
+  run=
+  shift
+  "$@" && exit 0
+  # Exit code 63 means version mismatch.  This often happens
+  # when the user try to use an ancient version of a tool on
+  # a file that requires a minimum version.  In this case we
+  # we should proceed has if the program had been absent, or
+  # if --run hadn't been passed.
+  if test $? = 63; then
+    run=:
+    msg="probably too old"
+  fi
+  ;;
+
+  -h|--h|--he|--hel|--help)
+    echo "\
+$0 [OPTION]... PROGRAM [ARGUMENT]...
+
+Handle \`PROGRAM [ARGUMENT]...' for when PROGRAM is missing, or return an
+error status if there is no known handling for PROGRAM.
+
+Options:
+  -h, --help      display this help and exit
+  -v, --version   output version information and exit
+  --run           try to run the given command, and emulate it if it fails
+
+Supported PROGRAM values:
+  aclocal      touch file \`aclocal.m4'
+  autoconf     touch file \`configure'
+  autoheader   touch file \`config.h.in'
+  autom4te     touch the output file, or create a stub one
+  automake     touch all \`Makefile.in' files
+  bison        create \`y.tab.[ch]', if possible, from existing .[ch]
+  flex         create \`lex.yy.c', if possible, from existing .c
+  help2man     touch the output file
+  lex          create \`lex.yy.c', if possible, from existing .c
+  makeinfo     touch the output file
+  tar          try tar, gnutar, gtar, then tar without non-portable flags
+  yacc         create \`y.tab.[ch]', if possible, from existing .[ch]
+
+Send bug reports to <bug-automake@gnu.org>."
+    exit $?
+    ;;
+
+  -v|--v|--ve|--ver|--vers|--versi|--versio|--version)
+    echo "missing $scriptversion (GNU Automake)"
+    exit $?
+    ;;
+
+  -*)
+    echo 1>&2 "$0: Unknown \`$1' option"
+    echo 1>&2 "Try \`$0 --help' for more information"
+    exit 1
+    ;;
+
+esac
+
+# Now exit if we have it, but it failed.  Also exit now if we
+# don't have it and --version was passed (most likely to detect
+# the program).
+case $1 in
+  lex|yacc)
+    # Not GNU programs, they don't have --version.
+    ;;
+
+  tar)
+    if test -n "$run"; then
+       echo 1>&2 "ERROR: \`tar' requires --run"
+       exit 1
+    elif test "x$2" = "x--version" || test "x$2" = "x--help"; then
+       exit 1
+    fi
+    ;;
+
+  *)
+    if test -z "$run" && ($1 --version) > /dev/null 2>&1; then
+       # We have it, but it failed.
+       exit 1
+    elif test "x$2" = "x--version" || test "x$2" = "x--help"; then
+       # Could not run --version or --help.  This is probably someone
+       # running `$TOOL --version' or `$TOOL --help' to check whether
+       # $TOOL exists and not knowing $TOOL uses missing.
+       exit 1
+    fi
+    ;;
+esac
+
+# If it does not exist, or fails to run (possibly an outdated version),
+# try to emulate it.
+case $1 in
+  aclocal*)
+    echo 1>&2 "\
+WARNING: \`$1' is $msg.  You should only need it if
+         you modified \`acinclude.m4' or \`${configure_ac}'.  You might want
+         to install the \`Automake' and \`Perl' packages.  Grab them from
+         any GNU archive site."
+    touch aclocal.m4
+    ;;
+
+  autoconf)
+    echo 1>&2 "\
+WARNING: \`$1' is $msg.  You should only need it if
+         you modified \`${configure_ac}'.  You might want to install the
+         \`Autoconf' and \`GNU m4' packages.  Grab them from any GNU
+         archive site."
+    touch configure
+    ;;
+
+  autoheader)
+    echo 1>&2 "\
+WARNING: \`$1' is $msg.  You should only need it if
+         you modified \`acconfig.h' or \`${configure_ac}'.  You might want
+         to install the \`Autoconf' and \`GNU m4' packages.  Grab them
+         from any GNU archive site."
+    files=`sed -n 's/^[ ]*A[CM]_CONFIG_HEADER(\([^)]*\)).*/\1/p' ${configure_ac}`
+    test -z "$files" && files="config.h"
+    touch_files=
+    for f in $files; do
+      case $f in
+      *:*) touch_files="$touch_files "`echo "$f" |
+				       sed -e 's/^[^:]*://' -e 's/:.*//'`;;
+      *) touch_files="$touch_files $f.in";;
+      esac
+    done
+    touch $touch_files
+    ;;
+
+  automake*)
+    echo 1>&2 "\
+WARNING: \`$1' is $msg.  You should only need it if
+         you modified \`Makefile.am', \`acinclude.m4' or \`${configure_ac}'.
+         You might want to install the \`Automake' and \`Perl' packages.
+         Grab them from any GNU archive site."
+    find . -type f -name Makefile.am -print |
+	   sed 's/\.am$/.in/' |
+	   while read f; do touch "$f"; done
+    ;;
+
+  autom4te)
+    echo 1>&2 "\
+WARNING: \`$1' is needed, but is $msg.
+         You might have modified some files without having the
+         proper tools for further handling them.
+         You can get \`$1' as part of \`Autoconf' from any GNU
+         archive site."
+
+    file=`echo "$*" | sed -n "$sed_output"`
+    test -z "$file" && file=`echo "$*" | sed -n "$sed_minuso"`
+    if test -f "$file"; then
+	touch $file
+    else
+	test -z "$file" || exec >$file
+	echo "#! /bin/sh"
+	echo "# Created by GNU Automake missing as a replacement of"
+	echo "#  $ $@"
+	echo "exit 0"
+	chmod +x $file
+	exit 1
+    fi
+    ;;
+
+  bison|yacc)
+    echo 1>&2 "\
+WARNING: \`$1' $msg.  You should only need it if
+         you modified a \`.y' file.  You may need the \`Bison' package
+         in order for those modifications to take effect.  You can get
+         \`Bison' from any GNU archive site."
+    rm -f y.tab.c y.tab.h
+    if test $# -ne 1; then
+        eval LASTARG="\${$#}"
+	case $LASTARG in
+	*.y)
+	    SRCFILE=`echo "$LASTARG" | sed 's/y$/c/'`
+	    if test -f "$SRCFILE"; then
+	         cp "$SRCFILE" y.tab.c
+	    fi
+	    SRCFILE=`echo "$LASTARG" | sed 's/y$/h/'`
+	    if test -f "$SRCFILE"; then
+	         cp "$SRCFILE" y.tab.h
+	    fi
+	  ;;
+	esac
+    fi
+    if test ! -f y.tab.h; then
+	echo >y.tab.h
+    fi
+    if test ! -f y.tab.c; then
+	echo 'main() { return 0; }' >y.tab.c
+    fi
+    ;;
+
+  lex|flex)
+    echo 1>&2 "\
+WARNING: \`$1' is $msg.  You should only need it if
+         you modified a \`.l' file.  You may need the \`Flex' package
+         in order for those modifications to take effect.  You can get
+         \`Flex' from any GNU archive site."
+    rm -f lex.yy.c
+    if test $# -ne 1; then
+        eval LASTARG="\${$#}"
+	case $LASTARG in
+	*.l)
+	    SRCFILE=`echo "$LASTARG" | sed 's/l$/c/'`
+	    if test -f "$SRCFILE"; then
+	         cp "$SRCFILE" lex.yy.c
+	    fi
+	  ;;
+	esac
+    fi
+    if test ! -f lex.yy.c; then
+	echo 'main() { return 0; }' >lex.yy.c
+    fi
+    ;;
+
+  help2man)
+    echo 1>&2 "\
+WARNING: \`$1' is $msg.  You should only need it if
+	 you modified a dependency of a manual page.  You may need the
+	 \`Help2man' package in order for those modifications to take
+	 effect.  You can get \`Help2man' from any GNU archive site."
+
+    file=`echo "$*" | sed -n "$sed_output"`
+    test -z "$file" && file=`echo "$*" | sed -n "$sed_minuso"`
+    if test -f "$file"; then
+	touch $file
+    else
+	test -z "$file" || exec >$file
+	echo ".ab help2man is required to generate this page"
+	exit 1
+    fi
+    ;;
+
+  makeinfo)
+    echo 1>&2 "\
+WARNING: \`$1' is $msg.  You should only need it if
+         you modified a \`.texi' or \`.texinfo' file, or any other file
+         indirectly affecting the aspect of the manual.  The spurious
+         call might also be the consequence of using a buggy \`make' (AIX,
+         DU, IRIX).  You might want to install the \`Texinfo' package or
+         the \`GNU make' package.  Grab either from any GNU archive site."
+    # The file to touch is that specified with -o ...
+    file=`echo "$*" | sed -n "$sed_output"`
+    test -z "$file" && file=`echo "$*" | sed -n "$sed_minuso"`
+    if test -z "$file"; then
+      # ... or it is the one specified with @setfilename ...
+      infile=`echo "$*" | sed 's/.* \([^ ]*\) *$/\1/'`
+      file=`sed -n '
+	/^@setfilename/{
+	  s/.* \([^ ]*\) *$/\1/
+	  p
+	  q
+	}' $infile`
+      # ... or it is derived from the source name (dir/f.texi becomes f.info)
+      test -z "$file" && file=`echo "$infile" | sed 's,.*/,,;s,.[^.]*$,,'`.info
+    fi
+    # If the file does not exist, the user really needs makeinfo;
+    # let's fail without touching anything.
+    test -f $file || exit 1
+    touch $file
+    ;;
+
+  tar)
+    shift
+
+    # We have already tried tar in the generic part.
+    # Look for gnutar/gtar before invocation to avoid ugly error
+    # messages.
+    if (gnutar --version > /dev/null 2>&1); then
+       gnutar "$@" && exit 0
+    fi
+    if (gtar --version > /dev/null 2>&1); then
+       gtar "$@" && exit 0
+    fi
+    firstarg="$1"
+    if shift; then
+	case $firstarg in
+	*o*)
+	    firstarg=`echo "$firstarg" | sed s/o//`
+	    tar "$firstarg" "$@" && exit 0
+	    ;;
+	esac
+	case $firstarg in
+	*h*)
+	    firstarg=`echo "$firstarg" | sed s/h//`
+	    tar "$firstarg" "$@" && exit 0
+	    ;;
+	esac
+    fi
+
+    echo 1>&2 "\
+WARNING: I can't seem to be able to run \`tar' with the given arguments.
+         You may want to install GNU tar or Free paxutils, or check the
+         command line arguments."
+    exit 1
+    ;;
+
+  *)
+    echo 1>&2 "\
+WARNING: \`$1' is needed, and is $msg.
+         You might have modified some files without having the
+         proper tools for further handling them.  Check the \`README' file,
+         it often tells you about the needed prerequisites for installing
+         this package.  You may also peek at any GNU archive site, in case
+         some other package would contain this missing \`$1' program."
+    exit 1
+    ;;
+esac
+
+exit 0
+
+# Local variables:
+# eval: (add-hook 'write-file-hooks 'time-stamp)
+# time-stamp-start: "scriptversion="
+# time-stamp-format: "%:y-%02m-%02d.%02H"
+# time-stamp-end: "$"
+# End:
diff --git a/scrypt.c b/scrypt.c
new file mode 100644
index 0000000000..5efd0e23b6
--- /dev/null
+++ b/scrypt.c
@@ -0,0 +1,756 @@
+/*
+ * Copyright 2009 Colin Percival, 2011 ArtForz, 2011-2013 pooler
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
+ * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
+ * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+ * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
+ * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
+ * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
+ * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ *
+ * This file was originally written by Colin Percival as part of the Tarsnap
+ * online backup system.
+ */
+
+#include "cpuminer-config.h"
+#include "miner.h"
+
+#include <stdlib.h>
+#include <string.h>
+#include <inttypes.h>
+
+static const uint32_t keypad[12] = {
+	0x80000000, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0x00000280
+};
+static const uint32_t innerpad[11] = {
+	0x80000000, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0x000004a0
+};
+static const uint32_t outerpad[8] = {
+	0x80000000, 0, 0, 0, 0, 0, 0, 0x00000300
+};
+static const uint32_t finalblk[16] = {
+	0x00000001, 0x80000000, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0x00000620
+};
+
+static inline void HMAC_SHA256_80_init(const uint32_t *key,
+	uint32_t *tstate, uint32_t *ostate)
+{
+	uint32_t ihash[8];
+	uint32_t pad[16];
+	int i;
+
+	/* tstate is assumed to contain the midstate of key */
+	memcpy(pad, key + 16, 16);
+	memcpy(pad + 4, keypad, 48);
+	sha256_transform(tstate, pad, 0);
+	memcpy(ihash, tstate, 32);
+
+	sha256_init(ostate);
+	for (i = 0; i < 8; i++)
+		pad[i] = ihash[i] ^ 0x5c5c5c5c;
+	for (; i < 16; i++)
+		pad[i] = 0x5c5c5c5c;
+	sha256_transform(ostate, pad, 0);
+
+	sha256_init(tstate);
+	for (i = 0; i < 8; i++)
+		pad[i] = ihash[i] ^ 0x36363636;
+	for (; i < 16; i++)
+		pad[i] = 0x36363636;
+	sha256_transform(tstate, pad, 0);
+}
+
+static inline void PBKDF2_SHA256_80_128(const uint32_t *tstate,
+	const uint32_t *ostate, const uint32_t *salt, uint32_t *output)
+{
+	uint32_t istate[8], ostate2[8];
+	uint32_t ibuf[16], obuf[16];
+	int i, j;
+
+	memcpy(istate, tstate, 32);
+	sha256_transform(istate, salt, 0);
+	
+	memcpy(ibuf, salt + 16, 16);
+	memcpy(ibuf + 5, innerpad, 44);
+	memcpy(obuf + 8, outerpad, 32);
+
+	for (i = 0; i < 4; i++) {
+		memcpy(obuf, istate, 32);
+		ibuf[4] = i + 1;
+		sha256_transform(obuf, ibuf, 0);
+
+		memcpy(ostate2, ostate, 32);
+		sha256_transform(ostate2, obuf, 0);
+		for (j = 0; j < 8; j++)
+			output[8 * i + j] = swab32(ostate2[j]);
+	}
+}
+
+static inline void PBKDF2_SHA256_128_32(uint32_t *tstate, uint32_t *ostate,
+	const uint32_t *salt, uint32_t *output)
+{
+	uint32_t buf[16];
+	int i;
+	
+	sha256_transform(tstate, salt, 1);
+	sha256_transform(tstate, salt + 16, 1);
+	sha256_transform(tstate, finalblk, 0);
+	memcpy(buf, tstate, 32);
+	memcpy(buf + 8, outerpad, 32);
+
+	sha256_transform(ostate, buf, 0);
+	for (i = 0; i < 8; i++)
+		output[i] = swab32(ostate[i]);
+}
+
+
+#if HAVE_SHA256_4WAY
+
+static const uint32_t keypad_4way[4 * 12] = {
+	0x80000000, 0x80000000, 0x80000000, 0x80000000,
+	0x00000000, 0x00000000, 0x00000000, 0x00000000,
+	0x00000000, 0x00000000, 0x00000000, 0x00000000,
+	0x00000000, 0x00000000, 0x00000000, 0x00000000,
+	0x00000000, 0x00000000, 0x00000000, 0x00000000,
+	0x00000000, 0x00000000, 0x00000000, 0x00000000,
+	0x00000000, 0x00000000, 0x00000000, 0x00000000,
+	0x00000000, 0x00000000, 0x00000000, 0x00000000,
+	0x00000000, 0x00000000, 0x00000000, 0x00000000,
+	0x00000000, 0x00000000, 0x00000000, 0x00000000,
+	0x00000000, 0x00000000, 0x00000000, 0x00000000,
+	0x00000280, 0x00000280, 0x00000280, 0x00000280
+};
+static const uint32_t innerpad_4way[4 * 11] = {
+	0x80000000, 0x80000000, 0x80000000, 0x80000000,
+	0x00000000, 0x00000000, 0x00000000, 0x00000000,
+	0x00000000, 0x00000000, 0x00000000, 0x00000000,
+	0x00000000, 0x00000000, 0x00000000, 0x00000000,
+	0x00000000, 0x00000000, 0x00000000, 0x00000000,
+	0x00000000, 0x00000000, 0x00000000, 0x00000000,
+	0x00000000, 0x00000000, 0x00000000, 0x00000000,
+	0x00000000, 0x00000000, 0x00000000, 0x00000000,
+	0x00000000, 0x00000000, 0x00000000, 0x00000000,
+	0x00000000, 0x00000000, 0x00000000, 0x00000000,
+	0x000004a0, 0x000004a0, 0x000004a0, 0x000004a0
+};
+static const uint32_t outerpad_4way[4 * 8] = {
+	0x80000000, 0x80000000, 0x80000000, 0x80000000,
+	0x00000000, 0x00000000, 0x00000000, 0x00000000,
+	0x00000000, 0x00000000, 0x00000000, 0x00000000,
+	0x00000000, 0x00000000, 0x00000000, 0x00000000,
+	0x00000000, 0x00000000, 0x00000000, 0x00000000,
+	0x00000000, 0x00000000, 0x00000000, 0x00000000,
+	0x00000000, 0x00000000, 0x00000000, 0x00000000,
+	0x00000300, 0x00000300, 0x00000300, 0x00000300
+};
+static const uint32_t finalblk_4way[4 * 16] __attribute__((aligned(16))) = {
+	0x00000001, 0x00000001, 0x00000001, 0x00000001,
+	0x80000000, 0x80000000, 0x80000000, 0x80000000,
+	0x00000000, 0x00000000, 0x00000000, 0x00000000,
+	0x00000000, 0x00000000, 0x00000000, 0x00000000,
+	0x00000000, 0x00000000, 0x00000000, 0x00000000,
+	0x00000000, 0x00000000, 0x00000000, 0x00000000,
+	0x00000000, 0x00000000, 0x00000000, 0x00000000,
+	0x00000000, 0x00000000, 0x00000000, 0x00000000,
+	0x00000000, 0x00000000, 0x00000000, 0x00000000,
+	0x00000000, 0x00000000, 0x00000000, 0x00000000,
+	0x00000000, 0x00000000, 0x00000000, 0x00000000,
+	0x00000000, 0x00000000, 0x00000000, 0x00000000,
+	0x00000000, 0x00000000, 0x00000000, 0x00000000,
+	0x00000000, 0x00000000, 0x00000000, 0x00000000,
+	0x00000000, 0x00000000, 0x00000000, 0x00000000,
+	0x00000620, 0x00000620, 0x00000620, 0x00000620
+};
+
+static inline void HMAC_SHA256_80_init_4way(const uint32_t *key,
+	uint32_t *tstate, uint32_t *ostate)
+{
+	uint32_t ihash[4 * 8] __attribute__((aligned(16)));
+	uint32_t pad[4 * 16] __attribute__((aligned(16)));
+	int i;
+
+	/* tstate is assumed to contain the midstate of key */
+	memcpy(pad, key + 4 * 16, 4 * 16);
+	memcpy(pad + 4 * 4, keypad_4way, 4 * 48);
+	sha256_transform_4way(tstate, pad, 0);
+	memcpy(ihash, tstate, 4 * 32);
+
+	sha256_init_4way(ostate);
+	for (i = 0; i < 4 * 8; i++)
+		pad[i] = ihash[i] ^ 0x5c5c5c5c;
+	for (; i < 4 * 16; i++)
+		pad[i] = 0x5c5c5c5c;
+	sha256_transform_4way(ostate, pad, 0);
+
+	sha256_init_4way(tstate);
+	for (i = 0; i < 4 * 8; i++)
+		pad[i] = ihash[i] ^ 0x36363636;
+	for (; i < 4 * 16; i++)
+		pad[i] = 0x36363636;
+	sha256_transform_4way(tstate, pad, 0);
+}
+
+static inline void PBKDF2_SHA256_80_128_4way(const uint32_t *tstate,
+	const uint32_t *ostate, const uint32_t *salt, uint32_t *output)
+{
+	uint32_t istate[4 * 8] __attribute__((aligned(16)));
+	uint32_t ostate2[4 * 8] __attribute__((aligned(16)));
+	uint32_t ibuf[4 * 16] __attribute__((aligned(16)));
+	uint32_t obuf[4 * 16] __attribute__((aligned(16)));
+	int i, j;
+
+	memcpy(istate, tstate, 4 * 32);
+	sha256_transform_4way(istate, salt, 0);
+	
+	memcpy(ibuf, salt + 4 * 16, 4 * 16);
+	memcpy(ibuf + 4 * 5, innerpad_4way, 4 * 44);
+	memcpy(obuf + 4 * 8, outerpad_4way, 4 * 32);
+
+	for (i = 0; i < 4; i++) {
+		memcpy(obuf, istate, 4 * 32);
+		ibuf[4 * 4 + 0] = i + 1;
+		ibuf[4 * 4 + 1] = i + 1;
+		ibuf[4 * 4 + 2] = i + 1;
+		ibuf[4 * 4 + 3] = i + 1;
+		sha256_transform_4way(obuf, ibuf, 0);
+
+		memcpy(ostate2, ostate, 4 * 32);
+		sha256_transform_4way(ostate2, obuf, 0);
+		for (j = 0; j < 4 * 8; j++)
+			output[4 * 8 * i + j] = swab32(ostate2[j]);
+	}
+}
+
+static inline void PBKDF2_SHA256_128_32_4way(uint32_t *tstate,
+	uint32_t *ostate, const uint32_t *salt, uint32_t *output)
+{
+	uint32_t buf[4 * 16] __attribute__((aligned(16)));
+	int i;
+	
+	sha256_transform_4way(tstate, salt, 1);
+	sha256_transform_4way(tstate, salt + 4 * 16, 1);
+	sha256_transform_4way(tstate, finalblk_4way, 0);
+	memcpy(buf, tstate, 4 * 32);
+	memcpy(buf + 4 * 8, outerpad_4way, 4 * 32);
+
+	sha256_transform_4way(ostate, buf, 0);
+	for (i = 0; i < 4 * 8; i++)
+		output[i] = swab32(ostate[i]);
+}
+
+#endif /* HAVE_SHA256_4WAY */
+
+
+#if HAVE_SHA256_8WAY
+
+static const uint32_t finalblk_8way[8 * 16] __attribute__((aligned(32))) = {
+	0x00000001, 0x00000001, 0x00000001, 0x00000001, 0x00000001, 0x00000001, 0x00000001, 0x00000001,
+	0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000,
+	0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000,
+	0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000,
+	0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000,
+	0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000,
+	0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000,
+	0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000,
+	0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000,
+	0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000,
+	0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000,
+	0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000,
+	0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000,
+	0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000,
+	0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000,
+	0x00000620, 0x00000620, 0x00000620, 0x00000620, 0x00000620, 0x00000620, 0x00000620, 0x00000620
+};
+
+static inline void HMAC_SHA256_80_init_8way(const uint32_t *key,
+	uint32_t *tstate, uint32_t *ostate)
+{
+	uint32_t ihash[8 * 8] __attribute__((aligned(32)));
+	uint32_t pad[8 * 16] __attribute__((aligned(32)));
+	int i;
+	
+	/* tstate is assumed to contain the midstate of key */
+	memcpy(pad, key + 8 * 16, 8 * 16);
+	for (i = 0; i < 8; i++)
+		pad[8 * 4 + i] = 0x80000000;
+	memset(pad + 8 * 5, 0x00, 8 * 40);
+	for (i = 0; i < 8; i++)
+		pad[8 * 15 + i] = 0x00000280;
+	sha256_transform_8way(tstate, pad, 0);
+	memcpy(ihash, tstate, 8 * 32);
+	
+	sha256_init_8way(ostate);
+	for (i = 0; i < 8 * 8; i++)
+		pad[i] = ihash[i] ^ 0x5c5c5c5c;
+	for (; i < 8 * 16; i++)
+		pad[i] = 0x5c5c5c5c;
+	sha256_transform_8way(ostate, pad, 0);
+	
+	sha256_init_8way(tstate);
+	for (i = 0; i < 8 * 8; i++)
+		pad[i] = ihash[i] ^ 0x36363636;
+	for (; i < 8 * 16; i++)
+		pad[i] = 0x36363636;
+	sha256_transform_8way(tstate, pad, 0);
+}
+
+static inline void PBKDF2_SHA256_80_128_8way(const uint32_t *tstate,
+	const uint32_t *ostate, const uint32_t *salt, uint32_t *output)
+{
+	uint32_t istate[8 * 8] __attribute__((aligned(32)));
+	uint32_t ostate2[8 * 8] __attribute__((aligned(32)));
+	uint32_t ibuf[8 * 16] __attribute__((aligned(32)));
+	uint32_t obuf[8 * 16] __attribute__((aligned(32)));
+	int i, j;
+	
+	memcpy(istate, tstate, 8 * 32);
+	sha256_transform_8way(istate, salt, 0);
+	
+	memcpy(ibuf, salt + 8 * 16, 8 * 16);
+	for (i = 0; i < 8; i++)
+		ibuf[8 * 5 + i] = 0x80000000;
+	memset(ibuf + 8 * 6, 0x00, 8 * 36);
+	for (i = 0; i < 8; i++)
+		ibuf[8 * 15 + i] = 0x000004a0;
+	
+	for (i = 0; i < 8; i++)
+		obuf[8 * 8 + i] = 0x80000000;
+	memset(obuf + 8 * 9, 0x00, 8 * 24);
+	for (i = 0; i < 8; i++)
+		obuf[8 * 15 + i] = 0x00000300;
+	
+	for (i = 0; i < 4; i++) {
+		memcpy(obuf, istate, 8 * 32);
+		ibuf[8 * 4 + 0] = i + 1;
+		ibuf[8 * 4 + 1] = i + 1;
+		ibuf[8 * 4 + 2] = i + 1;
+		ibuf[8 * 4 + 3] = i + 1;
+		ibuf[8 * 4 + 4] = i + 1;
+		ibuf[8 * 4 + 5] = i + 1;
+		ibuf[8 * 4 + 6] = i + 1;
+		ibuf[8 * 4 + 7] = i + 1;
+		sha256_transform_8way(obuf, ibuf, 0);
+		
+		memcpy(ostate2, ostate, 8 * 32);
+		sha256_transform_8way(ostate2, obuf, 0);
+		for (j = 0; j < 8 * 8; j++)
+			output[8 * 8 * i + j] = swab32(ostate2[j]);
+	}
+}
+
+static inline void PBKDF2_SHA256_128_32_8way(uint32_t *tstate,
+	uint32_t *ostate, const uint32_t *salt, uint32_t *output)
+{
+	uint32_t buf[8 * 16] __attribute__((aligned(32)));
+	int i;
+	
+	sha256_transform_8way(tstate, salt, 1);
+	sha256_transform_8way(tstate, salt + 8 * 16, 1);
+	sha256_transform_8way(tstate, finalblk_8way, 0);
+	
+	memcpy(buf, tstate, 8 * 32);
+	for (i = 0; i < 8; i++)
+		buf[8 * 8 + i] = 0x80000000;
+	memset(buf + 8 * 9, 0x00, 8 * 24);
+	for (i = 0; i < 8; i++)
+		buf[8 * 15 + i] = 0x00000300;
+	sha256_transform_8way(ostate, buf, 0);
+	
+	for (i = 0; i < 8 * 8; i++)
+		output[i] = swab32(ostate[i]);
+}
+
+#endif /* HAVE_SHA256_8WAY */
+
+
+#if defined(__x86_64__)
+
+#define SCRYPT_MAX_WAYS 1
+#define HAVE_SCRYPT_3WAY 0
+#define scrypt_best_throughput() 1
+static void scrypt_core(uint32_t *X, uint32_t *V);
+void scrypt_core_3way(uint32_t *X, uint32_t *V);
+#if defined(USE_AVX2)
+#undef SCRYPT_MAX_WAYS
+#define SCRYPT_MAX_WAYS 21
+#define HAVE_SCRYPT_6WAY 0
+void scrypt_core_6way(uint32_t *X, uint32_t *V);
+#endif
+
+#elif defined(__i386__)
+
+#define SCRYPT_MAX_WAYS 1
+#define scrypt_best_throughput() 1
+static void scrypt_core(uint32_t *X, uint32_t *V);
+
+#elif defined(__arm__) && defined(__APCS_32__)
+
+static void scrypt_core(uint32_t *X, uint32_t *V);
+#if defined(__ARM_NEON__)
+#undef HAVE_SHA256_4WAY
+#define SCRYPT_MAX_WAYS 1
+#define HAVE_SCRYPT_3WAY 0
+#define scrypt_best_throughput() 1
+void scrypt_core_3way(uint32_t *X, uint32_t *V);
+#endif
+
+#endif
+
+static inline void xor_salsa8(uint32_t B[16], const uint32_t Bx[16])
+{
+	uint32_t x00,x01,x02,x03,x04,x05,x06,x07,x08,x09,x10,x11,x12,x13,x14,x15;
+	int i;
+
+	x00 = (B[ 0] ^= Bx[ 0]);
+	x01 = (B[ 1] ^= Bx[ 1]);
+	x02 = (B[ 2] ^= Bx[ 2]);
+	x03 = (B[ 3] ^= Bx[ 3]);
+	x04 = (B[ 4] ^= Bx[ 4]);
+	x05 = (B[ 5] ^= Bx[ 5]);
+	x06 = (B[ 6] ^= Bx[ 6]);
+	x07 = (B[ 7] ^= Bx[ 7]);
+	x08 = (B[ 8] ^= Bx[ 8]);
+	x09 = (B[ 9] ^= Bx[ 9]);
+	x10 = (B[10] ^= Bx[10]);
+	x11 = (B[11] ^= Bx[11]);
+	x12 = (B[12] ^= Bx[12]);
+	x13 = (B[13] ^= Bx[13]);
+	x14 = (B[14] ^= Bx[14]);
+	x15 = (B[15] ^= Bx[15]);
+	for (i = 0; i < 8; i += 2) {
+#define R(a, b) (((a) << (b)) | ((a) >> (32 - (b))))
+		/* Operate on columns. */
+		x04 ^= R(x00+x12, 7);	x09 ^= R(x05+x01, 7);
+		x14 ^= R(x10+x06, 7);	x03 ^= R(x15+x11, 7);
+		
+		x08 ^= R(x04+x00, 9);	x13 ^= R(x09+x05, 9);
+		x02 ^= R(x14+x10, 9);	x07 ^= R(x03+x15, 9);
+		
+		x12 ^= R(x08+x04,13);	x01 ^= R(x13+x09,13);
+		x06 ^= R(x02+x14,13);	x11 ^= R(x07+x03,13);
+		
+		x00 ^= R(x12+x08,18);	x05 ^= R(x01+x13,18);
+		x10 ^= R(x06+x02,18);	x15 ^= R(x11+x07,18);
+		
+		/* Operate on rows. */
+		x01 ^= R(x00+x03, 7);	x06 ^= R(x05+x04, 7);
+		x11 ^= R(x10+x09, 7);	x12 ^= R(x15+x14, 7);
+		
+		x02 ^= R(x01+x00, 9);	x07 ^= R(x06+x05, 9);
+		x08 ^= R(x11+x10, 9);	x13 ^= R(x12+x15, 9);
+		
+		x03 ^= R(x02+x01,13);	x04 ^= R(x07+x06,13);
+		x09 ^= R(x08+x11,13);	x14 ^= R(x13+x12,13);
+		
+		x00 ^= R(x03+x02,18);	x05 ^= R(x04+x07,18);
+		x10 ^= R(x09+x08,18);	x15 ^= R(x14+x13,18);
+#undef R
+	}
+	B[ 0] += x00;
+	B[ 1] += x01;
+	B[ 2] += x02;
+	B[ 3] += x03;
+	B[ 4] += x04;
+	B[ 5] += x05;
+	B[ 6] += x06;
+	B[ 7] += x07;
+	B[ 8] += x08;
+	B[ 9] += x09;
+	B[10] += x10;
+	B[11] += x11;
+	B[12] += x12;
+	B[13] += x13;
+	B[14] += x14;
+	B[15] += x15;
+}
+
+static inline void scrypt_core(uint32_t *X, uint32_t *V)
+{
+	uint32_t i, j, k;
+	
+	for (i = 0; i < 1024; i++) {
+		memcpy(&V[i * 32], X, 128);
+		xor_salsa8(&X[0], &X[16]);
+		xor_salsa8(&X[16], &X[0]);
+	}
+	for (i = 0; i < 1024; i++) {
+		j = 32 * (X[16] & 1023);
+		for (k = 0; k < 32; k++)
+			X[k] ^= V[j + k];
+		xor_salsa8(&X[0], &X[16]);
+		xor_salsa8(&X[16], &X[0]);
+	}
+}
+
+#ifndef SCRYPT_MAX_WAYS
+#define SCRYPT_MAX_WAYS 1
+#define scrypt_best_throughput() 1
+#endif
+
+#define SCRYPT_BUFFER_SIZE (SCRYPT_MAX_WAYS * 131072 + 63)
+
+unsigned char *scrypt_buffer_alloc()
+{
+	return malloc(SCRYPT_BUFFER_SIZE);
+}
+
+static void scrypt_1024_1_1_256(const uint32_t *input, uint32_t *output,
+	uint32_t *midstate, unsigned char *scratchpad)
+{
+	uint32_t tstate[8], ostate[8];
+	uint32_t X[32];
+	uint32_t *V;
+	
+	V = (uint32_t *)(((uintptr_t)(scratchpad) + 63) & ~ (uintptr_t)(63));
+
+	memcpy(tstate, midstate, 32);
+	HMAC_SHA256_80_init(input, tstate, ostate);
+	PBKDF2_SHA256_80_128(tstate, ostate, input, X);
+
+	scrypt_core(X, V);
+
+	PBKDF2_SHA256_128_32(tstate, ostate, X, output);
+}
+
+#if HAVE_SHA256_4WAY
+static void scrypt_1024_1_1_256_4way(const uint32_t *input,
+	uint32_t *output, uint32_t *midstate, unsigned char *scratchpad)
+{
+	uint32_t tstate[4 * 8] __attribute__((aligned(128)));
+	uint32_t ostate[4 * 8] __attribute__((aligned(128)));
+	uint32_t W[4 * 32] __attribute__((aligned(128)));
+	uint32_t X[4 * 32] __attribute__((aligned(128)));
+	uint32_t *V;
+	int i, k;
+	
+	V = (uint32_t *)(((uintptr_t)(scratchpad) + 63) & ~ (uintptr_t)(63));
+
+	for (i = 0; i < 20; i++)
+		for (k = 0; k < 4; k++)
+			W[4 * i + k] = input[k * 20 + i];
+	for (i = 0; i < 8; i++)
+		for (k = 0; k < 4; k++)
+			tstate[4 * i + k] = midstate[i];
+	HMAC_SHA256_80_init_4way(W, tstate, ostate);
+	PBKDF2_SHA256_80_128_4way(tstate, ostate, W, W);
+	for (i = 0; i < 32; i++)
+		for (k = 0; k < 4; k++)
+			X[k * 32 + i] = W[4 * i + k];
+	scrypt_core(X + 0 * 32, V);
+	scrypt_core(X + 1 * 32, V);
+	scrypt_core(X + 2 * 32, V);
+	scrypt_core(X + 3 * 32, V);
+	for (i = 0; i < 32; i++)
+		for (k = 0; k < 4; k++)
+			W[4 * i + k] = X[k * 32 + i];
+	PBKDF2_SHA256_128_32_4way(tstate, ostate, W, W);
+	for (i = 0; i < 8; i++)
+		for (k = 0; k < 4; k++)
+			output[k * 8 + i] = W[4 * i + k];
+}
+#endif /* HAVE_SHA256_4WAY */
+
+#if HAVE_SCRYPT_3WAY
+
+static void scrypt_1024_1_1_256_3way(const uint32_t *input,
+	uint32_t *output, uint32_t *midstate, unsigned char *scratchpad)
+{
+	uint32_t tstate[3 * 8], ostate[3 * 8];
+	uint32_t X[3 * 32] __attribute__((aligned(64)));
+	uint32_t *V;
+	
+	V = (uint32_t *)(((uintptr_t)(scratchpad) + 63) & ~ (uintptr_t)(63));
+
+	memcpy(tstate +  0, midstate, 32);
+	memcpy(tstate +  8, midstate, 32);
+	memcpy(tstate + 16, midstate, 32);
+	HMAC_SHA256_80_init(input +  0, tstate +  0, ostate +  0);
+	HMAC_SHA256_80_init(input + 20, tstate +  8, ostate +  8);
+	HMAC_SHA256_80_init(input + 40, tstate + 16, ostate + 16);
+	PBKDF2_SHA256_80_128(tstate +  0, ostate +  0, input +  0, X +  0);
+	PBKDF2_SHA256_80_128(tstate +  8, ostate +  8, input + 20, X + 32);
+	PBKDF2_SHA256_80_128(tstate + 16, ostate + 16, input + 40, X + 64);
+
+	scrypt_core_3way(X, V);
+
+	PBKDF2_SHA256_128_32(tstate +  0, ostate +  0, X +  0, output +  0);
+	PBKDF2_SHA256_128_32(tstate +  8, ostate +  8, X + 32, output +  8);
+	PBKDF2_SHA256_128_32(tstate + 16, ostate + 16, X + 64, output + 16);
+}
+
+#if HAVE_SHA256_4WAY
+static void scrypt_1024_1_1_256_12way(const uint32_t *input,
+	uint32_t *output, uint32_t *midstate, unsigned char *scratchpad)
+{
+	uint32_t tstate[12 * 8] __attribute__((aligned(128)));
+	uint32_t ostate[12 * 8] __attribute__((aligned(128)));
+	uint32_t W[12 * 32] __attribute__((aligned(128)));
+	uint32_t X[12 * 32] __attribute__((aligned(128)));
+	uint32_t *V;
+	int i, j, k;
+	
+	V = (uint32_t *)(((uintptr_t)(scratchpad) + 63) & ~ (uintptr_t)(63));
+
+	for (j = 0; j < 3; j++)
+		for (i = 0; i < 20; i++)
+			for (k = 0; k < 4; k++)
+				W[128 * j + 4 * i + k] = input[80 * j + k * 20 + i];
+	for (j = 0; j < 3; j++)
+		for (i = 0; i < 8; i++)
+			for (k = 0; k < 4; k++)
+				tstate[32 * j + 4 * i + k] = midstate[i];
+	HMAC_SHA256_80_init_4way(W +   0, tstate +  0, ostate +  0);
+	HMAC_SHA256_80_init_4way(W + 128, tstate + 32, ostate + 32);
+	HMAC_SHA256_80_init_4way(W + 256, tstate + 64, ostate + 64);
+	PBKDF2_SHA256_80_128_4way(tstate +  0, ostate +  0, W +   0, W +   0);
+	PBKDF2_SHA256_80_128_4way(tstate + 32, ostate + 32, W + 128, W + 128);
+	PBKDF2_SHA256_80_128_4way(tstate + 64, ostate + 64, W + 256, W + 256);
+	for (j = 0; j < 3; j++)
+		for (i = 0; i < 32; i++)
+			for (k = 0; k < 4; k++)
+				X[128 * j + k * 32 + i] = W[128 * j + 4 * i + k];
+	scrypt_core_3way(X + 0 * 96, V);
+	scrypt_core_3way(X + 1 * 96, V);
+	scrypt_core_3way(X + 2 * 96, V);
+	scrypt_core_3way(X + 3 * 96, V);
+	for (j = 0; j < 3; j++)
+		for (i = 0; i < 32; i++)
+			for (k = 0; k < 4; k++)
+				W[128 * j + 4 * i + k] = X[128 * j + k * 32 + i];
+	PBKDF2_SHA256_128_32_4way(tstate +  0, ostate +  0, W +   0, W +   0);
+	PBKDF2_SHA256_128_32_4way(tstate + 32, ostate + 32, W + 128, W + 128);
+	PBKDF2_SHA256_128_32_4way(tstate + 64, ostate + 64, W + 256, W + 256);
+	for (j = 0; j < 3; j++)
+		for (i = 0; i < 8; i++)
+			for (k = 0; k < 4; k++)
+				output[32 * j + k * 8 + i] = W[128 * j + 4 * i + k];
+}
+#endif /* HAVE_SHA256_4WAY */
+
+#endif /* HAVE_SCRYPT_3WAY */
+
+#if HAVE_SCRYPT_6WAY
+static void scrypt_1024_1_1_256_24way(const uint32_t *input,
+	uint32_t *output, uint32_t *midstate, unsigned char *scratchpad)
+{
+	uint32_t tstate[24 * 8] __attribute__((aligned(128)));
+	uint32_t ostate[24 * 8] __attribute__((aligned(128)));
+	uint32_t W[24 * 32] __attribute__((aligned(128)));
+	uint32_t X[24 * 32] __attribute__((aligned(128)));
+	uint32_t *V;
+	int i, j, k;
+	
+	V = (uint32_t *)(((uintptr_t)(scratchpad) + 63) & ~ (uintptr_t)(63));
+	
+	for (j = 0; j < 3; j++) 
+		for (i = 0; i < 20; i++)
+			for (k = 0; k < 8; k++)
+				W[8 * 32 * j + 8 * i + k] = input[8 * 20 * j + k * 20 + i];
+	for (j = 0; j < 3; j++)
+		for (i = 0; i < 8; i++)
+			for (k = 0; k < 8; k++)
+				tstate[8 * 8 * j + 8 * i + k] = midstate[i];
+	HMAC_SHA256_80_init_8way(W +   0, tstate +   0, ostate +   0);
+	HMAC_SHA256_80_init_8way(W + 256, tstate +  64, ostate +  64);
+	HMAC_SHA256_80_init_8way(W + 512, tstate + 128, ostate + 128);
+	PBKDF2_SHA256_80_128_8way(tstate +   0, ostate +   0, W +   0, W +   0);
+	PBKDF2_SHA256_80_128_8way(tstate +  64, ostate +  64, W + 256, W + 256);
+	PBKDF2_SHA256_80_128_8way(tstate + 128, ostate + 128, W + 512, W + 512);
+	for (j = 0; j < 3; j++)
+		for (i = 0; i < 32; i++)
+			for (k = 0; k < 8; k++)
+				X[8 * 32 * j + k * 32 + i] = W[8 * 32 * j + 8 * i + k];
+	scrypt_core_6way(X + 0 * 32, V);
+	scrypt_core_6way(X + 6 * 32, V);
+	scrypt_core_6way(X + 12 * 32, V);
+	scrypt_core_6way(X + 18 * 32, V);
+	for (j = 0; j < 3; j++)
+		for (i = 0; i < 32; i++)
+			for (k = 0; k < 8; k++)
+				W[8 * 32 * j + 8 * i + k] = X[8 * 32 * j + k * 32 + i];
+	PBKDF2_SHA256_128_32_8way(tstate +   0, ostate +   0, W +   0, W +   0);
+	PBKDF2_SHA256_128_32_8way(tstate +  64, ostate +  64, W + 256, W + 256);
+	PBKDF2_SHA256_128_32_8way(tstate + 128, ostate + 128, W + 512, W + 512);
+	for (j = 0; j < 3; j++)
+		for (i = 0; i < 8; i++)
+			for (k = 0; k < 8; k++)
+				output[8 * 8 * j + k * 8 + i] = W[8 * 32 * j + 8 * i + k];
+}
+#endif /* HAVE_SCRYPT_6WAY */
+
+int scanhash_scrypt(int thr_id, uint32_t *pdata,
+	unsigned char *scratchbuf, const uint32_t *ptarget,
+	uint32_t max_nonce, unsigned long *hashes_done)
+{
+	uint32_t data[SCRYPT_MAX_WAYS * 20], hash[SCRYPT_MAX_WAYS * 8];
+	uint32_t midstate[8];
+	uint32_t n = pdata[19] - 1;
+	const uint32_t Htarg = ptarget[7];
+	int throughput = scrypt_best_throughput();
+	int i;
+	
+#if HAVE_SHA256_4WAY
+	if (sha256_use_4way())
+		throughput *= 4;
+#endif
+	
+	for (i = 0; i < throughput; i++)
+		memcpy(data + i * 20, pdata, 80);
+	
+	sha256_init(midstate);
+	sha256_transform(midstate, data, 0);
+	
+	do {
+		for (i = 0; i < throughput; i++)
+			data[i * 20 + 19] = ++n;
+		
+#if defined(HAVE_SHA256_4WAY)
+		if (throughput == 4)
+			scrypt_1024_1_1_256_4way(data, hash, midstate, scratchbuf);
+		else
+#endif
+#if defined(HAVE_SCRYPT_3WAY) && defined(HAVE_SHA256_4WAY)
+		if (throughput == 12)
+			scrypt_1024_1_1_256_12way(data, hash, midstate, scratchbuf);
+		else
+#endif
+#if defined(HAVE_SCRYPT_6WAY)
+		if (throughput == 24)
+			scrypt_1024_1_1_256_24way(data, hash, midstate, scratchbuf);
+		else
+#endif
+#if defined(HAVE_SCRYPT_3WAY)
+		if (throughput == 3)
+			scrypt_1024_1_1_256_3way(data, hash, midstate, scratchbuf);
+		else
+#endif
+		scrypt_1024_1_1_256(data, hash, midstate, scratchbuf);
+		
+		for (i = 0; i < throughput; i++) {
+			if (hash[i * 8 + 7] <= Htarg && fulltest(hash + i * 8, ptarget)) {
+				*hashes_done = n - pdata[19] + 1;
+				pdata[19] = data[i * 20 + 19];
+				return 1;
+			}
+		}
+	} while (n < max_nonce && !work_restart[thr_id].restart);
+	
+	*hashes_done = n - pdata[19] + 1;
+	pdata[19] = n;
+	return 0;
+}
diff --git a/sha2.c b/sha2.c
new file mode 100644
index 0000000000..4bfdcc7d0f
--- /dev/null
+++ b/sha2.c
@@ -0,0 +1,630 @@
+/*
+ * Copyright 2011 ArtForz
+ * Copyright 2011-2013 pooler
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License as published by the Free
+ * Software Foundation; either version 2 of the License, or (at your option)
+ * any later version.  See COPYING for more details.
+ */
+
+#include "cpuminer-config.h"
+#include "miner.h"
+
+#include <string.h>
+#include <inttypes.h>
+
+#if defined(__arm__) && defined(__APCS_32__)
+#define EXTERN_SHA256
+#endif
+
+static const uint32_t sha256_h[8] = {
+	0x6a09e667, 0xbb67ae85, 0x3c6ef372, 0xa54ff53a,
+	0x510e527f, 0x9b05688c, 0x1f83d9ab, 0x5be0cd19
+};
+
+static const uint32_t sha256_k[64] = {
+	0x428a2f98, 0x71374491, 0xb5c0fbcf, 0xe9b5dba5,
+	0x3956c25b, 0x59f111f1, 0x923f82a4, 0xab1c5ed5,
+	0xd807aa98, 0x12835b01, 0x243185be, 0x550c7dc3,
+	0x72be5d74, 0x80deb1fe, 0x9bdc06a7, 0xc19bf174,
+	0xe49b69c1, 0xefbe4786, 0x0fc19dc6, 0x240ca1cc,
+	0x2de92c6f, 0x4a7484aa, 0x5cb0a9dc, 0x76f988da,
+	0x983e5152, 0xa831c66d, 0xb00327c8, 0xbf597fc7,
+	0xc6e00bf3, 0xd5a79147, 0x06ca6351, 0x14292967,
+	0x27b70a85, 0x2e1b2138, 0x4d2c6dfc, 0x53380d13,
+	0x650a7354, 0x766a0abb, 0x81c2c92e, 0x92722c85,
+	0xa2bfe8a1, 0xa81a664b, 0xc24b8b70, 0xc76c51a3,
+	0xd192e819, 0xd6990624, 0xf40e3585, 0x106aa070,
+	0x19a4c116, 0x1e376c08, 0x2748774c, 0x34b0bcb5,
+	0x391c0cb3, 0x4ed8aa4a, 0x5b9cca4f, 0x682e6ff3,
+	0x748f82ee, 0x78a5636f, 0x84c87814, 0x8cc70208,
+	0x90befffa, 0xa4506ceb, 0xbef9a3f7, 0xc67178f2
+};
+
+void sha256_init(uint32_t *state)
+{
+	memcpy(state, sha256_h, 32);
+}
+
+/* Elementary functions used by SHA256 */
+#define Ch(x, y, z)     ((x & (y ^ z)) ^ z)
+#define Maj(x, y, z)    ((x & (y | z)) | (y & z))
+#define ROTR(x, n)      ((x >> n) | (x << (32 - n)))
+#define S0(x)           (ROTR(x, 2) ^ ROTR(x, 13) ^ ROTR(x, 22))
+#define S1(x)           (ROTR(x, 6) ^ ROTR(x, 11) ^ ROTR(x, 25))
+#define s0(x)           (ROTR(x, 7) ^ ROTR(x, 18) ^ (x >> 3))
+#define s1(x)           (ROTR(x, 17) ^ ROTR(x, 19) ^ (x >> 10))
+
+/* SHA256 round function */
+#define RND(a, b, c, d, e, f, g, h, k) \
+	do { \
+		t0 = h + S1(e) + Ch(e, f, g) + k; \
+		t1 = S0(a) + Maj(a, b, c); \
+		d += t0; \
+		h  = t0 + t1; \
+	} while (0)
+
+/* Adjusted round function for rotating state */
+#define RNDr(S, W, i) \
+	RND(S[(64 - i) % 8], S[(65 - i) % 8], \
+	    S[(66 - i) % 8], S[(67 - i) % 8], \
+	    S[(68 - i) % 8], S[(69 - i) % 8], \
+	    S[(70 - i) % 8], S[(71 - i) % 8], \
+	    W[i] + sha256_k[i])
+
+#ifndef EXTERN_SHA256
+
+/*
+ * SHA256 block compression function.  The 256-bit state is transformed via
+ * the 512-bit input block to produce a new state.
+ */
+void sha256_transform(uint32_t *state, const uint32_t *block, int swap)
+{
+	uint32_t W[64];
+	uint32_t S[8];
+	uint32_t t0, t1;
+	int i;
+
+	/* 1. Prepare message schedule W. */
+	if (swap) {
+		for (i = 0; i < 16; i++)
+			W[i] = swab32(block[i]);
+	} else
+		memcpy(W, block, 64);
+	for (i = 16; i < 64; i += 2) {
+		W[i]   = s1(W[i - 2]) + W[i - 7] + s0(W[i - 15]) + W[i - 16];
+		W[i+1] = s1(W[i - 1]) + W[i - 6] + s0(W[i - 14]) + W[i - 15];
+	}
+
+	/* 2. Initialize working variables. */
+	memcpy(S, state, 32);
+
+	/* 3. Mix. */
+	RNDr(S, W,  0);
+	RNDr(S, W,  1);
+	RNDr(S, W,  2);
+	RNDr(S, W,  3);
+	RNDr(S, W,  4);
+	RNDr(S, W,  5);
+	RNDr(S, W,  6);
+	RNDr(S, W,  7);
+	RNDr(S, W,  8);
+	RNDr(S, W,  9);
+	RNDr(S, W, 10);
+	RNDr(S, W, 11);
+	RNDr(S, W, 12);
+	RNDr(S, W, 13);
+	RNDr(S, W, 14);
+	RNDr(S, W, 15);
+	RNDr(S, W, 16);
+	RNDr(S, W, 17);
+	RNDr(S, W, 18);
+	RNDr(S, W, 19);
+	RNDr(S, W, 20);
+	RNDr(S, W, 21);
+	RNDr(S, W, 22);
+	RNDr(S, W, 23);
+	RNDr(S, W, 24);
+	RNDr(S, W, 25);
+	RNDr(S, W, 26);
+	RNDr(S, W, 27);
+	RNDr(S, W, 28);
+	RNDr(S, W, 29);
+	RNDr(S, W, 30);
+	RNDr(S, W, 31);
+	RNDr(S, W, 32);
+	RNDr(S, W, 33);
+	RNDr(S, W, 34);
+	RNDr(S, W, 35);
+	RNDr(S, W, 36);
+	RNDr(S, W, 37);
+	RNDr(S, W, 38);
+	RNDr(S, W, 39);
+	RNDr(S, W, 40);
+	RNDr(S, W, 41);
+	RNDr(S, W, 42);
+	RNDr(S, W, 43);
+	RNDr(S, W, 44);
+	RNDr(S, W, 45);
+	RNDr(S, W, 46);
+	RNDr(S, W, 47);
+	RNDr(S, W, 48);
+	RNDr(S, W, 49);
+	RNDr(S, W, 50);
+	RNDr(S, W, 51);
+	RNDr(S, W, 52);
+	RNDr(S, W, 53);
+	RNDr(S, W, 54);
+	RNDr(S, W, 55);
+	RNDr(S, W, 56);
+	RNDr(S, W, 57);
+	RNDr(S, W, 58);
+	RNDr(S, W, 59);
+	RNDr(S, W, 60);
+	RNDr(S, W, 61);
+	RNDr(S, W, 62);
+	RNDr(S, W, 63);
+
+	/* 4. Mix local working variables into global state */
+	for (i = 0; i < 8; i++)
+		state[i] += S[i];
+}
+
+#endif /* EXTERN_SHA256 */
+
+
+static const uint32_t sha256d_hash1[16] = {
+	0x00000000, 0x00000000, 0x00000000, 0x00000000,
+	0x00000000, 0x00000000, 0x00000000, 0x00000000,
+	0x80000000, 0x00000000, 0x00000000, 0x00000000,
+	0x00000000, 0x00000000, 0x00000000, 0x00000100
+};
+
+static void sha256d_80_swap(uint32_t *hash, const uint32_t *data)
+{
+	uint32_t S[16];
+	int i;
+
+	sha256_init(S);
+	sha256_transform(S, data, 0);
+	sha256_transform(S, data + 16, 0);
+	memcpy(S + 8, sha256d_hash1 + 8, 32);
+	sha256_init(hash);
+	sha256_transform(hash, S, 0);
+	for (i = 0; i < 8; i++)
+		hash[i] = swab32(hash[i]);
+}
+
+void sha256d(unsigned char *hash, const unsigned char *data, int len)
+{
+	uint32_t S[16], T[16];
+	int i, r;
+
+	sha256_init(S);
+	for (r = len; r > -9; r -= 64) {
+		if (r < 64)
+			memset(T, 0, 64);
+		memcpy(T, data + len - r, r > 64 ? 64 : (r < 0 ? 0 : r));
+		if (r >= 0 && r < 64)
+			((unsigned char *)T)[r] = 0x80;
+		for (i = 0; i < 16; i++)
+			T[i] = be32dec(T + i);
+		if (r < 56)
+			T[15] = 8 * len;
+		sha256_transform(S, T, 0);
+	}
+	memcpy(S + 8, sha256d_hash1 + 8, 32);
+	sha256_init(T);
+	sha256_transform(T, S, 0);
+	for (i = 0; i < 8; i++)
+		be32enc((uint32_t *)hash + i, T[i]);
+}
+
+static inline void sha256d_preextend(uint32_t *W)
+{
+	W[16] = s1(W[14]) + W[ 9] + s0(W[ 1]) + W[ 0];
+	W[17] = s1(W[15]) + W[10] + s0(W[ 2]) + W[ 1];
+	W[18] = s1(W[16]) + W[11]             + W[ 2];
+	W[19] = s1(W[17]) + W[12] + s0(W[ 4]);
+	W[20] =             W[13] + s0(W[ 5]) + W[ 4];
+	W[21] =             W[14] + s0(W[ 6]) + W[ 5];
+	W[22] =             W[15] + s0(W[ 7]) + W[ 6];
+	W[23] =             W[16] + s0(W[ 8]) + W[ 7];
+	W[24] =             W[17] + s0(W[ 9]) + W[ 8];
+	W[25] =                     s0(W[10]) + W[ 9];
+	W[26] =                     s0(W[11]) + W[10];
+	W[27] =                     s0(W[12]) + W[11];
+	W[28] =                     s0(W[13]) + W[12];
+	W[29] =                     s0(W[14]) + W[13];
+	W[30] =                     s0(W[15]) + W[14];
+	W[31] =                     s0(W[16]) + W[15];
+}
+
+static inline void sha256d_prehash(uint32_t *S, const uint32_t *W)
+{
+	uint32_t t0, t1;
+	RNDr(S, W, 0);
+	RNDr(S, W, 1);
+	RNDr(S, W, 2);
+}
+
+#ifdef EXTERN_SHA256
+
+void sha256d_ms(uint32_t *hash, uint32_t *W,
+	const uint32_t *midstate, const uint32_t *prehash);
+
+#else
+
+static inline void sha256d_ms(uint32_t *hash, uint32_t *W,
+	const uint32_t *midstate, const uint32_t *prehash)
+{
+	uint32_t S[64];
+	uint32_t t0, t1;
+	int i;
+
+	S[18] = W[18];
+	S[19] = W[19];
+	S[20] = W[20];
+	S[22] = W[22];
+	S[23] = W[23];
+	S[24] = W[24];
+	S[30] = W[30];
+	S[31] = W[31];
+
+	W[18] += s0(W[3]);
+	W[19] += W[3];
+	W[20] += s1(W[18]);
+	W[21]  = s1(W[19]);
+	W[22] += s1(W[20]);
+	W[23] += s1(W[21]);
+	W[24] += s1(W[22]);
+	W[25]  = s1(W[23]) + W[18];
+	W[26]  = s1(W[24]) + W[19];
+	W[27]  = s1(W[25]) + W[20];
+	W[28]  = s1(W[26]) + W[21];
+	W[29]  = s1(W[27]) + W[22];
+	W[30] += s1(W[28]) + W[23];
+	W[31] += s1(W[29]) + W[24];
+	for (i = 32; i < 64; i += 2) {
+		W[i]   = s1(W[i - 2]) + W[i - 7] + s0(W[i - 15]) + W[i - 16];
+		W[i+1] = s1(W[i - 1]) + W[i - 6] + s0(W[i - 14]) + W[i - 15];
+	}
+
+	memcpy(S, prehash, 32);
+
+	RNDr(S, W,  3);
+	RNDr(S, W,  4);
+	RNDr(S, W,  5);
+	RNDr(S, W,  6);
+	RNDr(S, W,  7);
+	RNDr(S, W,  8);
+	RNDr(S, W,  9);
+	RNDr(S, W, 10);
+	RNDr(S, W, 11);
+	RNDr(S, W, 12);
+	RNDr(S, W, 13);
+	RNDr(S, W, 14);
+	RNDr(S, W, 15);
+	RNDr(S, W, 16);
+	RNDr(S, W, 17);
+	RNDr(S, W, 18);
+	RNDr(S, W, 19);
+	RNDr(S, W, 20);
+	RNDr(S, W, 21);
+	RNDr(S, W, 22);
+	RNDr(S, W, 23);
+	RNDr(S, W, 24);
+	RNDr(S, W, 25);
+	RNDr(S, W, 26);
+	RNDr(S, W, 27);
+	RNDr(S, W, 28);
+	RNDr(S, W, 29);
+	RNDr(S, W, 30);
+	RNDr(S, W, 31);
+	RNDr(S, W, 32);
+	RNDr(S, W, 33);
+	RNDr(S, W, 34);
+	RNDr(S, W, 35);
+	RNDr(S, W, 36);
+	RNDr(S, W, 37);
+	RNDr(S, W, 38);
+	RNDr(S, W, 39);
+	RNDr(S, W, 40);
+	RNDr(S, W, 41);
+	RNDr(S, W, 42);
+	RNDr(S, W, 43);
+	RNDr(S, W, 44);
+	RNDr(S, W, 45);
+	RNDr(S, W, 46);
+	RNDr(S, W, 47);
+	RNDr(S, W, 48);
+	RNDr(S, W, 49);
+	RNDr(S, W, 50);
+	RNDr(S, W, 51);
+	RNDr(S, W, 52);
+	RNDr(S, W, 53);
+	RNDr(S, W, 54);
+	RNDr(S, W, 55);
+	RNDr(S, W, 56);
+	RNDr(S, W, 57);
+	RNDr(S, W, 58);
+	RNDr(S, W, 59);
+	RNDr(S, W, 60);
+	RNDr(S, W, 61);
+	RNDr(S, W, 62);
+	RNDr(S, W, 63);
+
+	for (i = 0; i < 8; i++)
+		S[i] += midstate[i];
+	
+	W[18] = S[18];
+	W[19] = S[19];
+	W[20] = S[20];
+	W[22] = S[22];
+	W[23] = S[23];
+	W[24] = S[24];
+	W[30] = S[30];
+	W[31] = S[31];
+	
+	memcpy(S + 8, sha256d_hash1 + 8, 32);
+	S[16] = s1(sha256d_hash1[14]) + sha256d_hash1[ 9] + s0(S[ 1]) + S[ 0];
+	S[17] = s1(sha256d_hash1[15]) + sha256d_hash1[10] + s0(S[ 2]) + S[ 1];
+	S[18] = s1(S[16]) + sha256d_hash1[11] + s0(S[ 3]) + S[ 2];
+	S[19] = s1(S[17]) + sha256d_hash1[12] + s0(S[ 4]) + S[ 3];
+	S[20] = s1(S[18]) + sha256d_hash1[13] + s0(S[ 5]) + S[ 4];
+	S[21] = s1(S[19]) + sha256d_hash1[14] + s0(S[ 6]) + S[ 5];
+	S[22] = s1(S[20]) + sha256d_hash1[15] + s0(S[ 7]) + S[ 6];
+	S[23] = s1(S[21]) + S[16] + s0(sha256d_hash1[ 8]) + S[ 7];
+	S[24] = s1(S[22]) + S[17] + s0(sha256d_hash1[ 9]) + sha256d_hash1[ 8];
+	S[25] = s1(S[23]) + S[18] + s0(sha256d_hash1[10]) + sha256d_hash1[ 9];
+	S[26] = s1(S[24]) + S[19] + s0(sha256d_hash1[11]) + sha256d_hash1[10];
+	S[27] = s1(S[25]) + S[20] + s0(sha256d_hash1[12]) + sha256d_hash1[11];
+	S[28] = s1(S[26]) + S[21] + s0(sha256d_hash1[13]) + sha256d_hash1[12];
+	S[29] = s1(S[27]) + S[22] + s0(sha256d_hash1[14]) + sha256d_hash1[13];
+	S[30] = s1(S[28]) + S[23] + s0(sha256d_hash1[15]) + sha256d_hash1[14];
+	S[31] = s1(S[29]) + S[24] + s0(S[16])             + sha256d_hash1[15];
+	for (i = 32; i < 60; i += 2) {
+		S[i]   = s1(S[i - 2]) + S[i - 7] + s0(S[i - 15]) + S[i - 16];
+		S[i+1] = s1(S[i - 1]) + S[i - 6] + s0(S[i - 14]) + S[i - 15];
+	}
+	S[60] = s1(S[58]) + S[53] + s0(S[45]) + S[44];
+
+	sha256_init(hash);
+
+	RNDr(hash, S,  0);
+	RNDr(hash, S,  1);
+	RNDr(hash, S,  2);
+	RNDr(hash, S,  3);
+	RNDr(hash, S,  4);
+	RNDr(hash, S,  5);
+	RNDr(hash, S,  6);
+	RNDr(hash, S,  7);
+	RNDr(hash, S,  8);
+	RNDr(hash, S,  9);
+	RNDr(hash, S, 10);
+	RNDr(hash, S, 11);
+	RNDr(hash, S, 12);
+	RNDr(hash, S, 13);
+	RNDr(hash, S, 14);
+	RNDr(hash, S, 15);
+	RNDr(hash, S, 16);
+	RNDr(hash, S, 17);
+	RNDr(hash, S, 18);
+	RNDr(hash, S, 19);
+	RNDr(hash, S, 20);
+	RNDr(hash, S, 21);
+	RNDr(hash, S, 22);
+	RNDr(hash, S, 23);
+	RNDr(hash, S, 24);
+	RNDr(hash, S, 25);
+	RNDr(hash, S, 26);
+	RNDr(hash, S, 27);
+	RNDr(hash, S, 28);
+	RNDr(hash, S, 29);
+	RNDr(hash, S, 30);
+	RNDr(hash, S, 31);
+	RNDr(hash, S, 32);
+	RNDr(hash, S, 33);
+	RNDr(hash, S, 34);
+	RNDr(hash, S, 35);
+	RNDr(hash, S, 36);
+	RNDr(hash, S, 37);
+	RNDr(hash, S, 38);
+	RNDr(hash, S, 39);
+	RNDr(hash, S, 40);
+	RNDr(hash, S, 41);
+	RNDr(hash, S, 42);
+	RNDr(hash, S, 43);
+	RNDr(hash, S, 44);
+	RNDr(hash, S, 45);
+	RNDr(hash, S, 46);
+	RNDr(hash, S, 47);
+	RNDr(hash, S, 48);
+	RNDr(hash, S, 49);
+	RNDr(hash, S, 50);
+	RNDr(hash, S, 51);
+	RNDr(hash, S, 52);
+	RNDr(hash, S, 53);
+	RNDr(hash, S, 54);
+	RNDr(hash, S, 55);
+	RNDr(hash, S, 56);
+	
+	hash[2] += hash[6] + S1(hash[3]) + Ch(hash[3], hash[4], hash[5])
+	         + S[57] + sha256_k[57];
+	hash[1] += hash[5] + S1(hash[2]) + Ch(hash[2], hash[3], hash[4])
+	         + S[58] + sha256_k[58];
+	hash[0] += hash[4] + S1(hash[1]) + Ch(hash[1], hash[2], hash[3])
+	         + S[59] + sha256_k[59];
+	hash[7] += hash[3] + S1(hash[0]) + Ch(hash[0], hash[1], hash[2])
+	         + S[60] + sha256_k[60]
+	         + sha256_h[7];
+}
+
+#endif /* EXTERN_SHA256 */
+
+#if HAVE_SHA256_4WAY
+
+void sha256d_ms_4way(uint32_t *hash,  uint32_t *data,
+	const uint32_t *midstate, const uint32_t *prehash);
+
+static inline int scanhash_sha256d_4way(int thr_id, uint32_t *pdata,
+	const uint32_t *ptarget, uint32_t max_nonce, unsigned long *hashes_done)
+{
+	uint32_t data[4 * 64] __attribute__((aligned(128)));
+	uint32_t hash[4 * 8] __attribute__((aligned(32)));
+	uint32_t midstate[4 * 8] __attribute__((aligned(32)));
+	uint32_t prehash[4 * 8] __attribute__((aligned(32)));
+	uint32_t n = pdata[19] - 1;
+	const uint32_t first_nonce = pdata[19];
+	const uint32_t Htarg = ptarget[7];
+	int i, j;
+	
+	memcpy(data, pdata + 16, 64);
+	sha256d_preextend(data);
+	for (i = 31; i >= 0; i--)
+		for (j = 0; j < 4; j++)
+			data[i * 4 + j] = data[i];
+	
+	sha256_init(midstate);
+	sha256_transform(midstate, pdata, 0);
+	memcpy(prehash, midstate, 32);
+	sha256d_prehash(prehash, pdata + 16);
+	for (i = 7; i >= 0; i--) {
+		for (j = 0; j < 4; j++) {
+			midstate[i * 4 + j] = midstate[i];
+			prehash[i * 4 + j] = prehash[i];
+		}
+	}
+	
+	do {
+		for (i = 0; i < 4; i++)
+			data[4 * 3 + i] = ++n;
+		
+		sha256d_ms_4way(hash, data, midstate, prehash);
+		
+		for (i = 0; i < 4; i++) {
+			if (swab32(hash[4 * 7 + i]) <= Htarg) {
+				pdata[19] = data[4 * 3 + i];
+				sha256d_80_swap(hash, pdata);
+				if (fulltest(hash, ptarget)) {
+					*hashes_done = n - first_nonce + 1;
+					return 1;
+				}
+			}
+		}
+	} while (n < max_nonce && !work_restart[thr_id].restart);
+	
+	*hashes_done = n - first_nonce + 1;
+	pdata[19] = n;
+	return 0;
+}
+
+#endif /* HAVE_SHA256_4WAY */
+
+#if HAVE_SHA256_8WAY
+
+void sha256d_ms_8way(uint32_t *hash,  uint32_t *data,
+	const uint32_t *midstate, const uint32_t *prehash);
+
+static inline int scanhash_sha256d_8way(int thr_id, uint32_t *pdata,
+	const uint32_t *ptarget, uint32_t max_nonce, unsigned long *hashes_done)
+{
+	uint32_t data[8 * 64] __attribute__((aligned(128)));
+	uint32_t hash[8 * 8] __attribute__((aligned(32)));
+	uint32_t midstate[8 * 8] __attribute__((aligned(32)));
+	uint32_t prehash[8 * 8] __attribute__((aligned(32)));
+	uint32_t n = pdata[19] - 1;
+	const uint32_t first_nonce = pdata[19];
+	const uint32_t Htarg = ptarget[7];
+	int i, j;
+	
+	memcpy(data, pdata + 16, 64);
+	sha256d_preextend(data);
+	for (i = 31; i >= 0; i--)
+		for (j = 0; j < 8; j++)
+			data[i * 8 + j] = data[i];
+	
+	sha256_init(midstate);
+	sha256_transform(midstate, pdata, 0);
+	memcpy(prehash, midstate, 32);
+	sha256d_prehash(prehash, pdata + 16);
+	for (i = 7; i >= 0; i--) {
+		for (j = 0; j < 8; j++) {
+			midstate[i * 8 + j] = midstate[i];
+			prehash[i * 8 + j] = prehash[i];
+		}
+	}
+	
+	do {
+		for (i = 0; i < 8; i++)
+			data[8 * 3 + i] = ++n;
+		
+		sha256d_ms_8way(hash, data, midstate, prehash);
+		
+		for (i = 0; i < 8; i++) {
+			if (swab32(hash[8 * 7 + i]) <= Htarg) {
+				pdata[19] = data[8 * 3 + i];
+				sha256d_80_swap(hash, pdata);
+				if (fulltest(hash, ptarget)) {
+					*hashes_done = n - first_nonce + 1;
+					return 1;
+				}
+			}
+		}
+	} while (n < max_nonce && !work_restart[thr_id].restart);
+	
+	*hashes_done = n - first_nonce + 1;
+	pdata[19] = n;
+	return 0;
+}
+
+#endif /* HAVE_SHA256_8WAY */
+
+int scanhash_sha256d(int thr_id, uint32_t *pdata, const uint32_t *ptarget,
+	uint32_t max_nonce, unsigned long *hashes_done)
+{
+	uint32_t data[64] /* __attribute__((aligned(128))) */;
+	uint32_t hash[8] /* __attribute__((aligned(32))) */;
+	uint32_t midstate[8] /* __attribute__((aligned(32))) */;
+	uint32_t prehash[8] /* __attribute__((aligned(32))) */;
+	uint32_t n = pdata[19] - 1;
+	const uint32_t first_nonce = pdata[19];
+	const uint32_t Htarg = ptarget[7];
+	
+#if HAVE_SHA256_8WAY
+	if (sha256_use_8way())
+		return scanhash_sha256d_8way(thr_id, pdata, ptarget,
+			max_nonce, hashes_done);
+#endif
+#if HAVE_SHA256_4WAY
+	if (sha256_use_4way())
+		return scanhash_sha256d_4way(thr_id, pdata, ptarget,
+			max_nonce, hashes_done);
+#endif
+	
+	memcpy(data, pdata + 16, 64);
+	sha256d_preextend(data);
+	
+	sha256_init(midstate);
+	sha256_transform(midstate, pdata, 0);
+	memcpy(prehash, midstate, 32);
+	sha256d_prehash(prehash, pdata + 16);
+	
+	do {
+		data[3] = ++n;
+		sha256d_ms(hash, data, midstate, prehash);
+		if (swab32(hash[7]) <= Htarg) {
+			pdata[19] = data[3];
+			sha256d_80_swap(hash, pdata);
+			if (fulltest(hash, ptarget)) {
+				*hashes_done = n - first_nonce + 1;
+				return 1;
+			}
+		}
+	} while (n < max_nonce && !work_restart[thr_id].restart);
+	
+	*hashes_done = n - first_nonce + 1;
+	pdata[19] = n;
+	return 0;
+}
diff --git a/sph_blake.h b/sph_blake.h
new file mode 100644
index 0000000000..0fc4295dac
--- /dev/null
+++ b/sph_blake.h
@@ -0,0 +1,327 @@
+/* $Id: sph_blake.h 252 2011-06-07 17:55:14Z tp $ */
+/**
+ * BLAKE interface. BLAKE is a family of functions which differ by their
+ * output size; this implementation defines BLAKE for output sizes 224,
+ * 256, 384 and 512 bits. This implementation conforms to the "third
+ * round" specification.
+ *
+ * ==========================(LICENSE BEGIN)============================
+ *
+ * Copyright (c) 2007-2010  Projet RNRT SAPHIR
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining
+ * a copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sublicense, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice shall be
+ * included in all copies or substantial portions of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ * IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY
+ * CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * ===========================(LICENSE END)=============================
+ *
+ * @file     sph_blake.h
+ * @author   Thomas Pornin <thomas.pornin@cryptolog.com>
+ */
+
+#ifndef SPH_BLAKE_H__
+#define SPH_BLAKE_H__
+
+#ifdef __cplusplus
+extern "C"{
+#endif
+
+#include <stddef.h>
+#include "sph_types.h"
+
+/**
+ * Output size (in bits) for BLAKE-224.
+ */
+#define SPH_SIZE_blake224   224
+
+/**
+ * Output size (in bits) for BLAKE-256.
+ */
+#define SPH_SIZE_blake256   256
+
+#if SPH_64
+
+/**
+ * Output size (in bits) for BLAKE-384.
+ */
+#define SPH_SIZE_blake384   384
+
+/**
+ * Output size (in bits) for BLAKE-512.
+ */
+#define SPH_SIZE_blake512   512
+
+#endif
+
+/**
+ * This structure is a context for BLAKE-224 and BLAKE-256 computations:
+ * it contains the intermediate values and some data from the last
+ * entered block. Once a BLAKE computation has been performed, the
+ * context can be reused for another computation.
+ *
+ * The contents of this structure are private. A running BLAKE
+ * computation can be cloned by copying the context (e.g. with a simple
+ * <code>memcpy()</code>).
+ */
+typedef struct {
+#ifndef DOXYGEN_IGNORE
+	unsigned char buf[64];    /* first field, for alignment */
+	size_t ptr;
+	sph_u32 H[8];
+	sph_u32 S[4];
+	sph_u32 T0, T1;
+#endif
+} sph_blake_small_context;
+
+/**
+ * This structure is a context for BLAKE-224 computations. It is
+ * identical to the common <code>sph_blake_small_context</code>.
+ */
+typedef sph_blake_small_context sph_blake224_context;
+
+/**
+ * This structure is a context for BLAKE-256 computations. It is
+ * identical to the common <code>sph_blake_small_context</code>.
+ */
+typedef sph_blake_small_context sph_blake256_context;
+
+#if SPH_64
+
+/**
+ * This structure is a context for BLAKE-384 and BLAKE-512 computations:
+ * it contains the intermediate values and some data from the last
+ * entered block. Once a BLAKE computation has been performed, the
+ * context can be reused for another computation.
+ *
+ * The contents of this structure are private. A running BLAKE
+ * computation can be cloned by copying the context (e.g. with a simple
+ * <code>memcpy()</code>).
+ */
+typedef struct {
+#ifndef DOXYGEN_IGNORE
+	unsigned char buf[128];    /* first field, for alignment */
+	size_t ptr;
+	sph_u64 H[8];
+	sph_u64 S[4];
+	sph_u64 T0, T1;
+#endif
+} sph_blake_big_context;
+
+/**
+ * This structure is a context for BLAKE-384 computations. It is
+ * identical to the common <code>sph_blake_small_context</code>.
+ */
+typedef sph_blake_big_context sph_blake384_context;
+
+/**
+ * This structure is a context for BLAKE-512 computations. It is
+ * identical to the common <code>sph_blake_small_context</code>.
+ */
+typedef sph_blake_big_context sph_blake512_context;
+
+#endif
+
+/**
+ * Initialize a BLAKE-224 context. This process performs no memory allocation.
+ *
+ * @param cc   the BLAKE-224 context (pointer to a
+ *             <code>sph_blake224_context</code>)
+ */
+void sph_blake224_init(void *cc);
+
+/**
+ * Process some data bytes. It is acceptable that <code>len</code> is zero
+ * (in which case this function does nothing).
+ *
+ * @param cc     the BLAKE-224 context
+ * @param data   the input data
+ * @param len    the input data length (in bytes)
+ */
+void sph_blake224(void *cc, const void *data, size_t len);
+
+/**
+ * Terminate the current BLAKE-224 computation and output the result into
+ * the provided buffer. The destination buffer must be wide enough to
+ * accomodate the result (28 bytes). The context is automatically
+ * reinitialized.
+ *
+ * @param cc    the BLAKE-224 context
+ * @param dst   the destination buffer
+ */
+void sph_blake224_close(void *cc, void *dst);
+
+/**
+ * Add a few additional bits (0 to 7) to the current computation, then
+ * terminate it and output the result in the provided buffer, which must
+ * be wide enough to accomodate the result (28 bytes). If bit number i
+ * in <code>ub</code> has value 2^i, then the extra bits are those
+ * numbered 7 downto 8-n (this is the big-endian convention at the byte
+ * level). The context is automatically reinitialized.
+ *
+ * @param cc    the BLAKE-224 context
+ * @param ub    the extra bits
+ * @param n     the number of extra bits (0 to 7)
+ * @param dst   the destination buffer
+ */
+void sph_blake224_addbits_and_close(
+	void *cc, unsigned ub, unsigned n, void *dst);
+
+/**
+ * Initialize a BLAKE-256 context. This process performs no memory allocation.
+ *
+ * @param cc   the BLAKE-256 context (pointer to a
+ *             <code>sph_blake256_context</code>)
+ */
+void sph_blake256_init(void *cc);
+
+/**
+ * Process some data bytes. It is acceptable that <code>len</code> is zero
+ * (in which case this function does nothing).
+ *
+ * @param cc     the BLAKE-256 context
+ * @param data   the input data
+ * @param len    the input data length (in bytes)
+ */
+void sph_blake256(void *cc, const void *data, size_t len);
+
+/**
+ * Terminate the current BLAKE-256 computation and output the result into
+ * the provided buffer. The destination buffer must be wide enough to
+ * accomodate the result (32 bytes). The context is automatically
+ * reinitialized.
+ *
+ * @param cc    the BLAKE-256 context
+ * @param dst   the destination buffer
+ */
+void sph_blake256_close(void *cc, void *dst);
+
+/**
+ * Add a few additional bits (0 to 7) to the current computation, then
+ * terminate it and output the result in the provided buffer, which must
+ * be wide enough to accomodate the result (32 bytes). If bit number i
+ * in <code>ub</code> has value 2^i, then the extra bits are those
+ * numbered 7 downto 8-n (this is the big-endian convention at the byte
+ * level). The context is automatically reinitialized.
+ *
+ * @param cc    the BLAKE-256 context
+ * @param ub    the extra bits
+ * @param n     the number of extra bits (0 to 7)
+ * @param dst   the destination buffer
+ */
+void sph_blake256_addbits_and_close(
+	void *cc, unsigned ub, unsigned n, void *dst);
+
+#if SPH_64
+
+/**
+ * Initialize a BLAKE-384 context. This process performs no memory allocation.
+ *
+ * @param cc   the BLAKE-384 context (pointer to a
+ *             <code>sph_blake384_context</code>)
+ */
+void sph_blake384_init(void *cc);
+
+/**
+ * Process some data bytes. It is acceptable that <code>len</code> is zero
+ * (in which case this function does nothing).
+ *
+ * @param cc     the BLAKE-384 context
+ * @param data   the input data
+ * @param len    the input data length (in bytes)
+ */
+void sph_blake384(void *cc, const void *data, size_t len);
+
+/**
+ * Terminate the current BLAKE-384 computation and output the result into
+ * the provided buffer. The destination buffer must be wide enough to
+ * accomodate the result (48 bytes). The context is automatically
+ * reinitialized.
+ *
+ * @param cc    the BLAKE-384 context
+ * @param dst   the destination buffer
+ */
+void sph_blake384_close(void *cc, void *dst);
+
+/**
+ * Add a few additional bits (0 to 7) to the current computation, then
+ * terminate it and output the result in the provided buffer, which must
+ * be wide enough to accomodate the result (48 bytes). If bit number i
+ * in <code>ub</code> has value 2^i, then the extra bits are those
+ * numbered 7 downto 8-n (this is the big-endian convention at the byte
+ * level). The context is automatically reinitialized.
+ *
+ * @param cc    the BLAKE-384 context
+ * @param ub    the extra bits
+ * @param n     the number of extra bits (0 to 7)
+ * @param dst   the destination buffer
+ */
+void sph_blake384_addbits_and_close(
+	void *cc, unsigned ub, unsigned n, void *dst);
+
+/**
+ * Initialize a BLAKE-512 context. This process performs no memory allocation.
+ *
+ * @param cc   the BLAKE-512 context (pointer to a
+ *             <code>sph_blake512_context</code>)
+ */
+void sph_blake512_init(void *cc);
+
+/**
+ * Process some data bytes. It is acceptable that <code>len</code> is zero
+ * (in which case this function does nothing).
+ *
+ * @param cc     the BLAKE-512 context
+ * @param data   the input data
+ * @param len    the input data length (in bytes)
+ */
+void sph_blake512(void *cc, const void *data, size_t len);
+
+/**
+ * Terminate the current BLAKE-512 computation and output the result into
+ * the provided buffer. The destination buffer must be wide enough to
+ * accomodate the result (64 bytes). The context is automatically
+ * reinitialized.
+ *
+ * @param cc    the BLAKE-512 context
+ * @param dst   the destination buffer
+ */
+void sph_blake512_close(void *cc, void *dst);
+
+/**
+ * Add a few additional bits (0 to 7) to the current computation, then
+ * terminate it and output the result in the provided buffer, which must
+ * be wide enough to accomodate the result (64 bytes). If bit number i
+ * in <code>ub</code> has value 2^i, then the extra bits are those
+ * numbered 7 downto 8-n (this is the big-endian convention at the byte
+ * level). The context is automatically reinitialized.
+ *
+ * @param cc    the BLAKE-512 context
+ * @param ub    the extra bits
+ * @param n     the number of extra bits (0 to 7)
+ * @param dst   the destination buffer
+ */
+void sph_blake512_addbits_and_close(
+	void *cc, unsigned ub, unsigned n, void *dst);
+
+#endif
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif
diff --git a/sph_fugue.h b/sph_fugue.h
new file mode 100644
index 0000000000..c8ff395e2e
--- /dev/null
+++ b/sph_fugue.h
@@ -0,0 +1,81 @@
+#ifndef SPH_FUGUE_H__
+#define SPH_FUGUE_H__
+
+#include <stddef.h>
+#include "sph_types.h"
+
+#ifdef __cplusplus
+extern "C"{
+#endif
+
+#define SPH_SIZE_fugue224   224
+
+#define SPH_SIZE_fugue256   256
+
+#define SPH_SIZE_fugue384   384
+
+#define SPH_SIZE_fugue512   512
+
+typedef struct {
+#ifndef DOXYGEN_IGNORE
+	sph_u32 partial;
+	unsigned partial_len;
+	unsigned round_shift;
+	sph_u32 S[36];
+#if SPH_64
+	sph_u64 bit_count;
+#else
+	sph_u32 bit_count_high, bit_count_low;
+#endif
+#endif
+} sph_fugue_context;
+
+typedef sph_fugue_context sph_fugue224_context;
+
+typedef sph_fugue_context sph_fugue256_context;
+
+typedef sph_fugue_context sph_fugue384_context;
+
+typedef sph_fugue_context sph_fugue512_context;
+
+void sph_fugue224_init(void *cc);
+
+void sph_fugue224(void *cc, const void *data, size_t len);
+
+void sph_fugue224_close(void *cc, void *dst);
+
+void sph_fugue224_addbits_and_close(
+	void *cc, unsigned ub, unsigned n, void *dst);
+
+void sph_fugue256_init(void *cc);
+
+void sph_fugue256(void *cc, const void *data, size_t len);
+
+void sph_fugue256_close(void *cc, void *dst);
+
+void sph_fugue256_addbits_and_close(
+	void *cc, unsigned ub, unsigned n, void *dst);
+
+void sph_fugue384_init(void *cc);
+
+void sph_fugue384(void *cc, const void *data, size_t len);
+
+void sph_fugue384_close(void *cc, void *dst);
+
+void sph_fugue384_addbits_and_close(
+	void *cc, unsigned ub, unsigned n, void *dst);
+
+void sph_fugue512_init(void *cc);
+
+void sph_fugue512(void *cc, const void *data, size_t len);
+
+void sph_fugue512_close(void *cc, void *dst);
+
+void sph_fugue512_addbits_and_close(
+	void *cc, unsigned ub, unsigned n, void *dst);
+
+#ifdef __cplusplus
+}
+#endif	
+	
+#endif
diff --git a/sph_groestl.h b/sph_groestl.h
new file mode 100644
index 0000000000..a9974319c4
--- /dev/null
+++ b/sph_groestl.h
@@ -0,0 +1,329 @@
+/* $Id: sph_groestl.h 216 2010-06-08 09:46:57Z tp $ */
+/**
+ * Groestl interface. This code implements Groestl with the recommended
+ * parameters for SHA-3, with outputs of 224, 256, 384 and 512 bits.
+ *
+ * ==========================(LICENSE BEGIN)============================
+ *
+ * Copyright (c) 2007-2010  Projet RNRT SAPHIR
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining
+ * a copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sublicense, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice shall be
+ * included in all copies or substantial portions of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ * IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY
+ * CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * ===========================(LICENSE END)=============================
+ *
+ * @file     sph_groestl.h
+ * @author   Thomas Pornin <thomas.pornin@cryptolog.com>
+ */
+
+#ifndef SPH_GROESTL_H__
+#define SPH_GROESTL_H__
+
+#ifdef __cplusplus
+extern "C"{
+#endif
+
+#include <stddef.h>
+#include "sph_types.h"
+
+/**
+ * Output size (in bits) for Groestl-224.
+ */
+#define SPH_SIZE_groestl224   224
+
+/**
+ * Output size (in bits) for Groestl-256.
+ */
+#define SPH_SIZE_groestl256   256
+
+/**
+ * Output size (in bits) for Groestl-384.
+ */
+#define SPH_SIZE_groestl384   384
+
+/**
+ * Output size (in bits) for Groestl-512.
+ */
+#define SPH_SIZE_groestl512   512
+
+/**
+ * This structure is a context for Groestl-224 and Groestl-256 computations:
+ * it contains the intermediate values and some data from the last
+ * entered block. Once a Groestl computation has been performed, the
+ * context can be reused for another computation.
+ *
+ * The contents of this structure are private. A running Groestl
+ * computation can be cloned by copying the context (e.g. with a simple
+ * <code>memcpy()</code>).
+ */
+typedef struct {
+#ifndef DOXYGEN_IGNORE
+	unsigned char buf[64];    /* first field, for alignment */
+	size_t ptr;
+	union {
+#if SPH_64
+		sph_u64 wide[8];
+#endif
+		sph_u32 narrow[16];
+	} state;
+#if SPH_64
+	sph_u64 count;
+#else
+	sph_u32 count_high, count_low;
+#endif
+#endif
+} sph_groestl_small_context;
+
+/**
+ * This structure is a context for Groestl-224 computations. It is
+ * identical to the common <code>sph_groestl_small_context</code>.
+ */
+typedef sph_groestl_small_context sph_groestl224_context;
+
+/**
+ * This structure is a context for Groestl-256 computations. It is
+ * identical to the common <code>sph_groestl_small_context</code>.
+ */
+typedef sph_groestl_small_context sph_groestl256_context;
+
+/**
+ * This structure is a context for Groestl-384 and Groestl-512 computations:
+ * it contains the intermediate values and some data from the last
+ * entered block. Once a Groestl computation has been performed, the
+ * context can be reused for another computation.
+ *
+ * The contents of this structure are private. A running Groestl
+ * computation can be cloned by copying the context (e.g. with a simple
+ * <code>memcpy()</code>).
+ */
+typedef struct {
+#ifndef DOXYGEN_IGNORE
+	unsigned char buf[128];    /* first field, for alignment */
+	size_t ptr;
+	union {
+#if SPH_64
+		sph_u64 wide[16];
+#endif
+		sph_u32 narrow[32];
+	} state;
+#if SPH_64
+	sph_u64 count;
+#else
+	sph_u32 count_high, count_low;
+#endif
+#endif
+} sph_groestl_big_context;
+
+/**
+ * This structure is a context for Groestl-384 computations. It is
+ * identical to the common <code>sph_groestl_small_context</code>.
+ */
+typedef sph_groestl_big_context sph_groestl384_context;
+
+/**
+ * This structure is a context for Groestl-512 computations. It is
+ * identical to the common <code>sph_groestl_small_context</code>.
+ */
+typedef sph_groestl_big_context sph_groestl512_context;
+
+/**
+ * Initialize a Groestl-224 context. This process performs no memory allocation.
+ *
+ * @param cc   the Groestl-224 context (pointer to a
+ *             <code>sph_groestl224_context</code>)
+ */
+void sph_groestl224_init(void *cc);
+
+/**
+ * Process some data bytes. It is acceptable that <code>len</code> is zero
+ * (in which case this function does nothing).
+ *
+ * @param cc     the Groestl-224 context
+ * @param data   the input data
+ * @param len    the input data length (in bytes)
+ */
+void sph_groestl224(void *cc, const void *data, size_t len);
+
+/**
+ * Terminate the current Groestl-224 computation and output the result into
+ * the provided buffer. The destination buffer must be wide enough to
+ * accomodate the result (28 bytes). The context is automatically
+ * reinitialized.
+ *
+ * @param cc    the Groestl-224 context
+ * @param dst   the destination buffer
+ */
+void sph_groestl224_close(void *cc, void *dst);
+
+/**
+ * Add a few additional bits (0 to 7) to the current computation, then
+ * terminate it and output the result in the provided buffer, which must
+ * be wide enough to accomodate the result (28 bytes). If bit number i
+ * in <code>ub</code> has value 2^i, then the extra bits are those
+ * numbered 7 downto 8-n (this is the big-endian convention at the byte
+ * level). The context is automatically reinitialized.
+ *
+ * @param cc    the Groestl-224 context
+ * @param ub    the extra bits
+ * @param n     the number of extra bits (0 to 7)
+ * @param dst   the destination buffer
+ */
+void sph_groestl224_addbits_and_close(
+	void *cc, unsigned ub, unsigned n, void *dst);
+
+/**
+ * Initialize a Groestl-256 context. This process performs no memory allocation.
+ *
+ * @param cc   the Groestl-256 context (pointer to a
+ *             <code>sph_groestl256_context</code>)
+ */
+void sph_groestl256_init(void *cc);
+
+/**
+ * Process some data bytes. It is acceptable that <code>len</code> is zero
+ * (in which case this function does nothing).
+ *
+ * @param cc     the Groestl-256 context
+ * @param data   the input data
+ * @param len    the input data length (in bytes)
+ */
+void sph_groestl256(void *cc, const void *data, size_t len);
+
+/**
+ * Terminate the current Groestl-256 computation and output the result into
+ * the provided buffer. The destination buffer must be wide enough to
+ * accomodate the result (32 bytes). The context is automatically
+ * reinitialized.
+ *
+ * @param cc    the Groestl-256 context
+ * @param dst   the destination buffer
+ */
+void sph_groestl256_close(void *cc, void *dst);
+
+/**
+ * Add a few additional bits (0 to 7) to the current computation, then
+ * terminate it and output the result in the provided buffer, which must
+ * be wide enough to accomodate the result (32 bytes). If bit number i
+ * in <code>ub</code> has value 2^i, then the extra bits are those
+ * numbered 7 downto 8-n (this is the big-endian convention at the byte
+ * level). The context is automatically reinitialized.
+ *
+ * @param cc    the Groestl-256 context
+ * @param ub    the extra bits
+ * @param n     the number of extra bits (0 to 7)
+ * @param dst   the destination buffer
+ */
+void sph_groestl256_addbits_and_close(
+	void *cc, unsigned ub, unsigned n, void *dst);
+
+/**
+ * Initialize a Groestl-384 context. This process performs no memory allocation.
+ *
+ * @param cc   the Groestl-384 context (pointer to a
+ *             <code>sph_groestl384_context</code>)
+ */
+void sph_groestl384_init(void *cc);
+
+/**
+ * Process some data bytes. It is acceptable that <code>len</code> is zero
+ * (in which case this function does nothing).
+ *
+ * @param cc     the Groestl-384 context
+ * @param data   the input data
+ * @param len    the input data length (in bytes)
+ */
+void sph_groestl384(void *cc, const void *data, size_t len);
+
+/**
+ * Terminate the current Groestl-384 computation and output the result into
+ * the provided buffer. The destination buffer must be wide enough to
+ * accomodate the result (48 bytes). The context is automatically
+ * reinitialized.
+ *
+ * @param cc    the Groestl-384 context
+ * @param dst   the destination buffer
+ */
+void sph_groestl384_close(void *cc, void *dst);
+
+/**
+ * Add a few additional bits (0 to 7) to the current computation, then
+ * terminate it and output the result in the provided buffer, which must
+ * be wide enough to accomodate the result (48 bytes). If bit number i
+ * in <code>ub</code> has value 2^i, then the extra bits are those
+ * numbered 7 downto 8-n (this is the big-endian convention at the byte
+ * level). The context is automatically reinitialized.
+ *
+ * @param cc    the Groestl-384 context
+ * @param ub    the extra bits
+ * @param n     the number of extra bits (0 to 7)
+ * @param dst   the destination buffer
+ */
+void sph_groestl384_addbits_and_close(
+	void *cc, unsigned ub, unsigned n, void *dst);
+
+/**
+ * Initialize a Groestl-512 context. This process performs no memory allocation.
+ *
+ * @param cc   the Groestl-512 context (pointer to a
+ *             <code>sph_groestl512_context</code>)
+ */
+void sph_groestl512_init(void *cc);
+
+/**
+ * Process some data bytes. It is acceptable that <code>len</code> is zero
+ * (in which case this function does nothing).
+ *
+ * @param cc     the Groestl-512 context
+ * @param data   the input data
+ * @param len    the input data length (in bytes)
+ */
+void sph_groestl512(void *cc, const void *data, size_t len);
+
+/**
+ * Terminate the current Groestl-512 computation and output the result into
+ * the provided buffer. The destination buffer must be wide enough to
+ * accomodate the result (64 bytes). The context is automatically
+ * reinitialized.
+ *
+ * @param cc    the Groestl-512 context
+ * @param dst   the destination buffer
+ */
+void sph_groestl512_close(void *cc, void *dst);
+
+/**
+ * Add a few additional bits (0 to 7) to the current computation, then
+ * terminate it and output the result in the provided buffer, which must
+ * be wide enough to accomodate the result (64 bytes). If bit number i
+ * in <code>ub</code> has value 2^i, then the extra bits are those
+ * numbered 7 downto 8-n (this is the big-endian convention at the byte
+ * level). The context is automatically reinitialized.
+ *
+ * @param cc    the Groestl-512 context
+ * @param ub    the extra bits
+ * @param n     the number of extra bits (0 to 7)
+ * @param dst   the destination buffer
+ */
+void sph_groestl512_addbits_and_close(
+	void *cc, unsigned ub, unsigned n, void *dst);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif
diff --git a/sph_keccak.h b/sph_keccak.h
new file mode 100644
index 0000000000..87605983a9
--- /dev/null
+++ b/sph_keccak.h
@@ -0,0 +1,293 @@
+/* $Id: sph_keccak.h 216 2010-06-08 09:46:57Z tp $ */
+/**
+ * Keccak interface. This is the interface for Keccak with the
+ * recommended parameters for SHA-3, with output lengths 224, 256,
+ * 384 and 512 bits.
+ *
+ * ==========================(LICENSE BEGIN)============================
+ *
+ * Copyright (c) 2007-2010  Projet RNRT SAPHIR
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining
+ * a copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sublicense, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice shall be
+ * included in all copies or substantial portions of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ * IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY
+ * CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * ===========================(LICENSE END)=============================
+ *
+ * @file     sph_keccak.h
+ * @author   Thomas Pornin <thomas.pornin@cryptolog.com>
+ */
+
+#ifndef SPH_KECCAK_H__
+#define SPH_KECCAK_H__
+
+#ifdef __cplusplus
+extern "C"{
+#endif
+
+#include <stddef.h>
+#include "sph_types.h"
+
+/**
+ * Output size (in bits) for Keccak-224.
+ */
+#define SPH_SIZE_keccak224   224
+
+/**
+ * Output size (in bits) for Keccak-256.
+ */
+#define SPH_SIZE_keccak256   256
+
+/**
+ * Output size (in bits) for Keccak-384.
+ */
+#define SPH_SIZE_keccak384   384
+
+/**
+ * Output size (in bits) for Keccak-512.
+ */
+#define SPH_SIZE_keccak512   512
+
+/**
+ * This structure is a context for Keccak computations: it contains the
+ * intermediate values and some data from the last entered block. Once a
+ * Keccak computation has been performed, the context can be reused for
+ * another computation.
+ *
+ * The contents of this structure are private. A running Keccak computation
+ * can be cloned by copying the context (e.g. with a simple
+ * <code>memcpy()</code>).
+ */
+typedef struct {
+#ifndef DOXYGEN_IGNORE
+	unsigned char buf[144];    /* first field, for alignment */
+	size_t ptr, lim;
+	union {
+#if SPH_64
+		sph_u64 wide[25];
+#endif
+		sph_u32 narrow[50];
+	} u;
+#endif
+} sph_keccak_context;
+
+/**
+ * Type for a Keccak-224 context (identical to the common context).
+ */
+typedef sph_keccak_context sph_keccak224_context;
+
+/**
+ * Type for a Keccak-256 context (identical to the common context).
+ */
+typedef sph_keccak_context sph_keccak256_context;
+
+/**
+ * Type for a Keccak-384 context (identical to the common context).
+ */
+typedef sph_keccak_context sph_keccak384_context;
+
+/**
+ * Type for a Keccak-512 context (identical to the common context).
+ */
+typedef sph_keccak_context sph_keccak512_context;
+
+/**
+ * Initialize a Keccak-224 context. This process performs no memory allocation.
+ *
+ * @param cc   the Keccak-224 context (pointer to a
+ *             <code>sph_keccak224_context</code>)
+ */
+void sph_keccak224_init(void *cc);
+
+/**
+ * Process some data bytes. It is acceptable that <code>len</code> is zero
+ * (in which case this function does nothing).
+ *
+ * @param cc     the Keccak-224 context
+ * @param data   the input data
+ * @param len    the input data length (in bytes)
+ */
+void sph_keccak224(void *cc, const void *data, size_t len);
+
+/**
+ * Terminate the current Keccak-224 computation and output the result into
+ * the provided buffer. The destination buffer must be wide enough to
+ * accomodate the result (28 bytes). The context is automatically
+ * reinitialized.
+ *
+ * @param cc    the Keccak-224 context
+ * @param dst   the destination buffer
+ */
+void sph_keccak224_close(void *cc, void *dst);
+
+/**
+ * Add a few additional bits (0 to 7) to the current computation, then
+ * terminate it and output the result in the provided buffer, which must
+ * be wide enough to accomodate the result (28 bytes). If bit number i
+ * in <code>ub</code> has value 2^i, then the extra bits are those
+ * numbered 7 downto 8-n (this is the big-endian convention at the byte
+ * level). The context is automatically reinitialized.
+ *
+ * @param cc    the Keccak-224 context
+ * @param ub    the extra bits
+ * @param n     the number of extra bits (0 to 7)
+ * @param dst   the destination buffer
+ */
+void sph_keccak224_addbits_and_close(
+	void *cc, unsigned ub, unsigned n, void *dst);
+
+/**
+ * Initialize a Keccak-256 context. This process performs no memory allocation.
+ *
+ * @param cc   the Keccak-256 context (pointer to a
+ *             <code>sph_keccak256_context</code>)
+ */
+void sph_keccak256_init(void *cc);
+
+/**
+ * Process some data bytes. It is acceptable that <code>len</code> is zero
+ * (in which case this function does nothing).
+ *
+ * @param cc     the Keccak-256 context
+ * @param data   the input data
+ * @param len    the input data length (in bytes)
+ */
+void sph_keccak256(void *cc, const void *data, size_t len);
+
+/**
+ * Terminate the current Keccak-256 computation and output the result into
+ * the provided buffer. The destination buffer must be wide enough to
+ * accomodate the result (32 bytes). The context is automatically
+ * reinitialized.
+ *
+ * @param cc    the Keccak-256 context
+ * @param dst   the destination buffer
+ */
+void sph_keccak256_close(void *cc, void *dst);
+
+/**
+ * Add a few additional bits (0 to 7) to the current computation, then
+ * terminate it and output the result in the provided buffer, which must
+ * be wide enough to accomodate the result (32 bytes). If bit number i
+ * in <code>ub</code> has value 2^i, then the extra bits are those
+ * numbered 7 downto 8-n (this is the big-endian convention at the byte
+ * level). The context is automatically reinitialized.
+ *
+ * @param cc    the Keccak-256 context
+ * @param ub    the extra bits
+ * @param n     the number of extra bits (0 to 7)
+ * @param dst   the destination buffer
+ */
+void sph_keccak256_addbits_and_close(
+	void *cc, unsigned ub, unsigned n, void *dst);
+
+/**
+ * Initialize a Keccak-384 context. This process performs no memory allocation.
+ *
+ * @param cc   the Keccak-384 context (pointer to a
+ *             <code>sph_keccak384_context</code>)
+ */
+void sph_keccak384_init(void *cc);
+
+/**
+ * Process some data bytes. It is acceptable that <code>len</code> is zero
+ * (in which case this function does nothing).
+ *
+ * @param cc     the Keccak-384 context
+ * @param data   the input data
+ * @param len    the input data length (in bytes)
+ */
+void sph_keccak384(void *cc, const void *data, size_t len);
+
+/**
+ * Terminate the current Keccak-384 computation and output the result into
+ * the provided buffer. The destination buffer must be wide enough to
+ * accomodate the result (48 bytes). The context is automatically
+ * reinitialized.
+ *
+ * @param cc    the Keccak-384 context
+ * @param dst   the destination buffer
+ */
+void sph_keccak384_close(void *cc, void *dst);
+
+/**
+ * Add a few additional bits (0 to 7) to the current computation, then
+ * terminate it and output the result in the provided buffer, which must
+ * be wide enough to accomodate the result (48 bytes). If bit number i
+ * in <code>ub</code> has value 2^i, then the extra bits are those
+ * numbered 7 downto 8-n (this is the big-endian convention at the byte
+ * level). The context is automatically reinitialized.
+ *
+ * @param cc    the Keccak-384 context
+ * @param ub    the extra bits
+ * @param n     the number of extra bits (0 to 7)
+ * @param dst   the destination buffer
+ */
+void sph_keccak384_addbits_and_close(
+	void *cc, unsigned ub, unsigned n, void *dst);
+
+/**
+ * Initialize a Keccak-512 context. This process performs no memory allocation.
+ *
+ * @param cc   the Keccak-512 context (pointer to a
+ *             <code>sph_keccak512_context</code>)
+ */
+void sph_keccak512_init(void *cc);
+
+/**
+ * Process some data bytes. It is acceptable that <code>len</code> is zero
+ * (in which case this function does nothing).
+ *
+ * @param cc     the Keccak-512 context
+ * @param data   the input data
+ * @param len    the input data length (in bytes)
+ */
+void sph_keccak512(void *cc, const void *data, size_t len);
+
+/**
+ * Terminate the current Keccak-512 computation and output the result into
+ * the provided buffer. The destination buffer must be wide enough to
+ * accomodate the result (64 bytes). The context is automatically
+ * reinitialized.
+ *
+ * @param cc    the Keccak-512 context
+ * @param dst   the destination buffer
+ */
+void sph_keccak512_close(void *cc, void *dst);
+
+/**
+ * Add a few additional bits (0 to 7) to the current computation, then
+ * terminate it and output the result in the provided buffer, which must
+ * be wide enough to accomodate the result (64 bytes). If bit number i
+ * in <code>ub</code> has value 2^i, then the extra bits are those
+ * numbered 7 downto 8-n (this is the big-endian convention at the byte
+ * level). The context is automatically reinitialized.
+ *
+ * @param cc    the Keccak-512 context
+ * @param ub    the extra bits
+ * @param n     the number of extra bits (0 to 7)
+ * @param dst   the destination buffer
+ */
+void sph_keccak512_addbits_and_close(
+	void *cc, unsigned ub, unsigned n, void *dst);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif
diff --git a/sph_types.h b/sph_types.h
new file mode 100644
index 0000000000..054c96f2f2
--- /dev/null
+++ b/sph_types.h
@@ -0,0 +1,1976 @@
+/* $Id: sph_types.h 260 2011-07-21 01:02:38Z tp $ */
+/**
+ * Basic type definitions.
+ *
+ * This header file defines the generic integer types that will be used
+ * for the implementation of hash functions; it also contains helper
+ * functions which encode and decode multi-byte integer values, using
+ * either little-endian or big-endian conventions.
+ *
+ * This file contains a compile-time test on the size of a byte
+ * (the <code>unsigned char</code> C type). If bytes are not octets,
+ * i.e. if they do not have a size of exactly 8 bits, then compilation
+ * is aborted. Architectures where bytes are not octets are relatively
+ * rare, even in the embedded devices market. We forbid non-octet bytes
+ * because there is no clear convention on how octet streams are encoded
+ * on such systems.
+ *
+ * ==========================(LICENSE BEGIN)============================
+ *
+ * Copyright (c) 2007-2010  Projet RNRT SAPHIR
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining
+ * a copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sublicense, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice shall be
+ * included in all copies or substantial portions of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ * IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY
+ * CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * ===========================(LICENSE END)=============================
+ *
+ * @file     sph_types.h
+ * @author   Thomas Pornin <thomas.pornin@cryptolog.com>
+ */
+
+#ifndef SPH_TYPES_H__
+#define SPH_TYPES_H__
+
+#include <limits.h>
+
+/*
+ * All our I/O functions are defined over octet streams. We do not know
+ * how to handle input data if bytes are not octets.
+ */
+#if CHAR_BIT != 8
+#error This code requires 8-bit bytes
+#endif
+
+/* ============= BEGIN documentation block for Doxygen ============ */
+
+#ifdef DOXYGEN_IGNORE
+
+/** @mainpage sphlib C code documentation
+ *
+ * @section overview Overview
+ *
+ * <code>sphlib</code> is a library which contains implementations of
+ * various cryptographic hash functions. These pages have been generated
+ * with <a href="http://www.doxygen.org/index.html">doxygen</a> and
+ * document the API for the C implementations.
+ *
+ * The API is described in appropriate header files, which are available
+ * in the "Files" section. Each hash function family has its own header,
+ * whose name begins with <code>"sph_"</code> and contains the family
+ * name. For instance, the API for the RIPEMD hash functions is available
+ * in the header file <code>sph_ripemd.h</code>.
+ *
+ * @section principles API structure and conventions
+ *
+ * @subsection io Input/output conventions
+ *
+ * In all generality, hash functions operate over strings of bits.
+ * Individual bits are rarely encountered in C programming or actual
+ * communication protocols; most protocols converge on the ubiquitous
+ * "octet" which is a group of eight bits. Data is thus expressed as a
+ * stream of octets. The C programming language contains the notion of a
+ * "byte", which is a data unit managed under the type <code>"unsigned
+ * char"</code>. The C standard prescribes that a byte should hold at
+ * least eight bits, but possibly more. Most modern architectures, even
+ * in the embedded world, feature eight-bit bytes, i.e. map bytes to
+ * octets.
+ *
+ * Nevertheless, for some of the implemented hash functions, an extra
+ * API has been added, which allows the input of arbitrary sequences of
+ * bits: when the computation is about to be closed, 1 to 7 extra bits
+ * can be added. The functions for which this API is implemented include
+ * the SHA-2 functions and all SHA-3 candidates.
+ *
+ * <code>sphlib</code> defines hash function which may hash octet streams,
+ * i.e. streams of bits where the number of bits is a multiple of eight.
+ * The data input functions in the <code>sphlib</code> API expect data
+ * as anonymous pointers (<code>"const void *"</code>) with a length
+ * (of type <code>"size_t"</code>) which gives the input data chunk length
+ * in bytes. A byte is assumed to be an octet; the <code>sph_types.h</code>
+ * header contains a compile-time test which prevents compilation on
+ * architectures where this property is not met.
+ *
+ * The hash function output is also converted into bytes. All currently
+ * implemented hash functions have an output width which is a multiple of
+ * eight, and this is likely to remain true for new designs.
+ *
+ * Most hash functions internally convert input data into 32-bit of 64-bit
+ * words, using either little-endian or big-endian conversion. The hash
+ * output also often consists of such words, which are encoded into output
+ * bytes with a similar endianness convention. Some hash functions have
+ * been only loosely specified on that subject; when necessary,
+ * <code>sphlib</code> has been tested against published "reference"
+ * implementations in order to use the same conventions.
+ *
+ * @subsection shortname Function short name
+ *
+ * Each implemented hash function has a "short name" which is used
+ * internally to derive the identifiers for the functions and context
+ * structures which the function uses. For instance, MD5 has the short
+ * name <code>"md5"</code>. Short names are listed in the next section,
+ * for the implemented hash functions. In subsequent sections, the
+ * short name will be assumed to be <code>"XXX"</code>: replace with the
+ * actual hash function name to get the C identifier.
+ *
+ * Note: some functions within the same family share the same core
+ * elements, such as update function or context structure. Correspondingly,
+ * some of the defined types or functions may actually be macros which
+ * transparently evaluate to another type or function name.
+ *
+ * @subsection context Context structure
+ *
+ * Each implemented hash fonction has its own context structure, available
+ * under the type name <code>"sph_XXX_context"</code> for the hash function
+ * with short name <code>"XXX"</code>. This structure holds all needed
+ * state for a running hash computation.
+ *
+ * The contents of these structures are meant to be opaque, and private
+ * to the implementation. However, these contents are specified in the
+ * header files so that application code which uses <code>sphlib</code>
+ * may access the size of those structures.
+ *
+ * The caller is responsible for allocating the context structure,
+ * whether by dynamic allocation (<code>malloc()</code> or equivalent),
+ * static allocation (a global permanent variable), as an automatic
+ * variable ("on the stack"), or by any other mean which ensures proper
+ * structure alignment. <code>sphlib</code> code performs no dynamic
+ * allocation by itself.
+ *
+ * The context must be initialized before use, using the
+ * <code>sph_XXX_init()</code> function. This function sets the context
+ * state to proper initial values for hashing.
+ *
+ * Since all state data is contained within the context structure,
+ * <code>sphlib</code> is thread-safe and reentrant: several hash
+ * computations may be performed in parallel, provided that they do not
+ * operate on the same context. Moreover, a running computation can be
+ * cloned by copying the context (with a simple <code>memcpy()</code>):
+ * the context and its clone are then independant and may be updated
+ * with new data and/or closed without interfering with each other.
+ * Similarly, a context structure can be moved in memory at will:
+ * context structures contain no pointer, in particular no pointer to
+ * themselves.
+ *
+ * @subsection dataio Data input
+ *
+ * Hashed data is input with the <code>sph_XXX()</code> fonction, which
+ * takes as parameters a pointer to the context, a pointer to the data
+ * to hash, and the number of data bytes to hash. The context is updated
+ * with the new data.
+ *
+ * Data can be input in one or several calls, with arbitrary input lengths.
+ * However, it is best, performance wise, to input data by relatively big
+ * chunks (say a few kilobytes), because this allows <code>sphlib</code> to
+ * optimize things and avoid internal copying.
+ *
+ * When all data has been input, the context can be closed with
+ * <code>sph_XXX_close()</code>. The hash output is computed and written
+ * into the provided buffer. The caller must take care to provide a
+ * buffer of appropriate length; e.g., when using SHA-1, the output is
+ * a 20-byte word, therefore the output buffer must be at least 20-byte
+ * long.
+ *
+ * For some hash functions, the <code>sph_XXX_addbits_and_close()</code>
+ * function can be used instead of <code>sph_XXX_close()</code>. This
+ * function can take a few extra <strong>bits</strong> to be added at
+ * the end of the input message. This allows hashing messages with a
+ * bit length which is not a multiple of 8. The extra bits are provided
+ * as an unsigned integer value, and a bit count. The bit count must be
+ * between 0 and 7, inclusive. The extra bits are provided as bits 7 to
+ * 0 (bits of numerical value 128, 64, 32... downto 0), in that order.
+ * For instance, to add three bits of value 1, 1 and 0, the unsigned
+ * integer will have value 192 (1*128 + 1*64 + 0*32) and the bit count
+ * will be 3.
+ *
+ * The <code>SPH_SIZE_XXX</code> macro is defined for each hash function;
+ * it evaluates to the function output size, expressed in bits. For instance,
+ * <code>SPH_SIZE_sha1</code> evaluates to <code>160</code>.
+ *
+ * When closed, the context is automatically reinitialized and can be
+ * immediately used for another computation. It is not necessary to call
+ * <code>sph_XXX_init()</code> after a close. Note that
+ * <code>sph_XXX_init()</code> can still be called to "reset" a context,
+ * i.e. forget previously input data, and get back to the initial state.
+ *
+ * @subsection alignment Data alignment
+ *
+ * "Alignment" is a property of data, which is said to be "properly
+ * aligned" when its emplacement in memory is such that the data can
+ * be optimally read by full words. This depends on the type of access;
+ * basically, some hash functions will read data by 32-bit or 64-bit
+ * words. <code>sphlib</code> does not mandate such alignment for input
+ * data, but using aligned data can substantially improve performance.
+ *
+ * As a rule, it is best to input data by chunks whose length (in bytes)
+ * is a multiple of eight, and which begins at "generally aligned"
+ * addresses, such as the base address returned by a call to
+ * <code>malloc()</code>.
+ *
+ * @section functions Implemented functions
+ *
+ * We give here the list of implemented functions. They are grouped by
+ * family; to each family corresponds a specific header file. Each
+ * individual function has its associated "short name". Please refer to
+ * the documentation for that header file to get details on the hash
+ * function denomination and provenance.
+ *
+ * Note: the functions marked with a '(64)' in the list below are
+ * available only if the C compiler provides an integer type of length
+ * 64 bits or more. Such a type is mandatory in the latest C standard
+ * (ISO 9899:1999, aka "C99") and is present in several older compilers
+ * as well, so chances are that such a type is available.
+ *
+ * - HAVAL family: file <code>sph_haval.h</code>
+ *   - HAVAL-128/3 (128-bit, 3 passes): short name: <code>haval128_3</code>
+ *   - HAVAL-128/4 (128-bit, 4 passes): short name: <code>haval128_4</code>
+ *   - HAVAL-128/5 (128-bit, 5 passes): short name: <code>haval128_5</code>
+ *   - HAVAL-160/3 (160-bit, 3 passes): short name: <code>haval160_3</code>
+ *   - HAVAL-160/4 (160-bit, 4 passes): short name: <code>haval160_4</code>
+ *   - HAVAL-160/5 (160-bit, 5 passes): short name: <code>haval160_5</code>
+ *   - HAVAL-192/3 (192-bit, 3 passes): short name: <code>haval192_3</code>
+ *   - HAVAL-192/4 (192-bit, 4 passes): short name: <code>haval192_4</code>
+ *   - HAVAL-192/5 (192-bit, 5 passes): short name: <code>haval192_5</code>
+ *   - HAVAL-224/3 (224-bit, 3 passes): short name: <code>haval224_3</code>
+ *   - HAVAL-224/4 (224-bit, 4 passes): short name: <code>haval224_4</code>
+ *   - HAVAL-224/5 (224-bit, 5 passes): short name: <code>haval224_5</code>
+ *   - HAVAL-256/3 (256-bit, 3 passes): short name: <code>haval256_3</code>
+ *   - HAVAL-256/4 (256-bit, 4 passes): short name: <code>haval256_4</code>
+ *   - HAVAL-256/5 (256-bit, 5 passes): short name: <code>haval256_5</code>
+ * - MD2: file <code>sph_md2.h</code>, short name: <code>md2</code>
+ * - MD4: file <code>sph_md4.h</code>, short name: <code>md4</code>
+ * - MD5: file <code>sph_md5.h</code>, short name: <code>md5</code>
+ * - PANAMA: file <code>sph_panama.h</code>, short name: <code>panama</code>
+ * - RadioGatun family: file <code>sph_radiogatun.h</code>
+ *   - RadioGatun[32]: short name: <code>radiogatun32</code>
+ *   - RadioGatun[64]: short name: <code>radiogatun64</code> (64)
+ * - RIPEMD family: file <code>sph_ripemd.h</code>
+ *   - RIPEMD: short name: <code>ripemd</code>
+ *   - RIPEMD-128: short name: <code>ripemd128</code>
+ *   - RIPEMD-160: short name: <code>ripemd160</code>
+ * - SHA-0: file <code>sph_sha0.h</code>, short name: <code>sha0</code>
+ * - SHA-1: file <code>sph_sha1.h</code>, short name: <code>sha1</code>
+ * - SHA-2 family, 32-bit hashes: file <code>sph_sha2.h</code>
+ *   - SHA-224: short name: <code>sha224</code>
+ *   - SHA-256: short name: <code>sha256</code>
+ *   - SHA-384: short name: <code>sha384</code> (64)
+ *   - SHA-512: short name: <code>sha512</code> (64)
+ * - Tiger family: file <code>sph_tiger.h</code>
+ *   - Tiger: short name: <code>tiger</code> (64)
+ *   - Tiger2: short name: <code>tiger2</code> (64)
+ * - WHIRLPOOL family: file <code>sph_whirlpool.h</code>
+ *   - WHIRLPOOL-0: short name: <code>whirlpool0</code> (64)
+ *   - WHIRLPOOL-1: short name: <code>whirlpool1</code> (64)
+ *   - WHIRLPOOL: short name: <code>whirlpool</code> (64)
+ *
+ * The fourteen second-round SHA-3 candidates are also implemented;
+ * when applicable, the implementations follow the "final" specifications
+ * as published for the third round of the SHA-3 competition (BLAKE,
+ * Groestl, JH, Keccak and Skein have been tweaked for third round).
+ *
+ * - BLAKE family: file <code>sph_blake.h</code>
+ *   - BLAKE-224: short name: <code>blake224</code>
+ *   - BLAKE-256: short name: <code>blake256</code>
+ *   - BLAKE-384: short name: <code>blake384</code>
+ *   - BLAKE-512: short name: <code>blake512</code>
+ * - BMW (Blue Midnight Wish) family: file <code>sph_bmw.h</code>
+ *   - BMW-224: short name: <code>bmw224</code>
+ *   - BMW-256: short name: <code>bmw256</code>
+ *   - BMW-384: short name: <code>bmw384</code> (64)
+ *   - BMW-512: short name: <code>bmw512</code> (64)
+ * - CubeHash family: file <code>sph_cubehash.h</code> (specified as
+ *   CubeHash16/32 in the CubeHash specification)
+ *   - CubeHash-224: short name: <code>cubehash224</code>
+ *   - CubeHash-256: short name: <code>cubehash256</code>
+ *   - CubeHash-384: short name: <code>cubehash384</code>
+ *   - CubeHash-512: short name: <code>cubehash512</code>
+ * - ECHO family: file <code>sph_echo.h</code>
+ *   - ECHO-224: short name: <code>echo224</code>
+ *   - ECHO-256: short name: <code>echo256</code>
+ *   - ECHO-384: short name: <code>echo384</code>
+ *   - ECHO-512: short name: <code>echo512</code>
+ * - Fugue family: file <code>sph_fugue.h</code>
+ *   - Fugue-224: short name: <code>fugue224</code>
+ *   - Fugue-256: short name: <code>fugue256</code>
+ *   - Fugue-384: short name: <code>fugue384</code>
+ *   - Fugue-512: short name: <code>fugue512</code>
+ * - Groestl family: file <code>sph_groestl.h</code>
+ *   - Groestl-224: short name: <code>groestl224</code>
+ *   - Groestl-256: short name: <code>groestl256</code>
+ *   - Groestl-384: short name: <code>groestl384</code>
+ *   - Groestl-512: short name: <code>groestl512</code>
+ * - Hamsi family: file <code>sph_hamsi.h</code>
+ *   - Hamsi-224: short name: <code>hamsi224</code>
+ *   - Hamsi-256: short name: <code>hamsi256</code>
+ *   - Hamsi-384: short name: <code>hamsi384</code>
+ *   - Hamsi-512: short name: <code>hamsi512</code>
+ * - JH family: file <code>sph_jh.h</code>
+ *   - JH-224: short name: <code>jh224</code>
+ *   - JH-256: short name: <code>jh256</code>
+ *   - JH-384: short name: <code>jh384</code>
+ *   - JH-512: short name: <code>jh512</code>
+ * - Keccak family: file <code>sph_keccak.h</code>
+ *   - Keccak-224: short name: <code>keccak224</code>
+ *   - Keccak-256: short name: <code>keccak256</code>
+ *   - Keccak-384: short name: <code>keccak384</code>
+ *   - Keccak-512: short name: <code>keccak512</code>
+ * - Luffa family: file <code>sph_luffa.h</code>
+ *   - Luffa-224: short name: <code>luffa224</code>
+ *   - Luffa-256: short name: <code>luffa256</code>
+ *   - Luffa-384: short name: <code>luffa384</code>
+ *   - Luffa-512: short name: <code>luffa512</code>
+ * - Shabal family: file <code>sph_shabal.h</code>
+ *   - Shabal-192: short name: <code>shabal192</code>
+ *   - Shabal-224: short name: <code>shabal224</code>
+ *   - Shabal-256: short name: <code>shabal256</code>
+ *   - Shabal-384: short name: <code>shabal384</code>
+ *   - Shabal-512: short name: <code>shabal512</code>
+ * - SHAvite-3 family: file <code>sph_shavite.h</code>
+ *   - SHAvite-224 (nominally "SHAvite-3 with 224-bit output"):
+ *     short name: <code>shabal224</code>
+ *   - SHAvite-256 (nominally "SHAvite-3 with 256-bit output"):
+ *     short name: <code>shabal256</code>
+ *   - SHAvite-384 (nominally "SHAvite-3 with 384-bit output"):
+ *     short name: <code>shabal384</code>
+ *   - SHAvite-512 (nominally "SHAvite-3 with 512-bit output"):
+ *     short name: <code>shabal512</code>
+ * - SIMD family: file <code>sph_simd.h</code>
+ *   - SIMD-224: short name: <code>simd224</code>
+ *   - SIMD-256: short name: <code>simd256</code>
+ *   - SIMD-384: short name: <code>simd384</code>
+ *   - SIMD-512: short name: <code>simd512</code>
+ * - Skein family: file <code>sph_skein.h</code>
+ *   - Skein-224 (nominally specified as Skein-512-224): short name:
+ *     <code>skein224</code> (64)
+ *   - Skein-256 (nominally specified as Skein-512-256): short name:
+ *     <code>skein256</code> (64)
+ *   - Skein-384 (nominally specified as Skein-512-384): short name:
+ *     <code>skein384</code> (64)
+ *   - Skein-512 (nominally specified as Skein-512-512): short name:
+ *     <code>skein512</code> (64)
+ *
+ * For the second-round SHA-3 candidates, the functions are as specified
+ * for round 2, i.e. with the "tweaks" that some candidates added
+ * between round 1 and round 2. Also, some of the submitted packages for
+ * round 2 contained errors, in the specification, reference code, or
+ * both. <code>sphlib</code> implements the corrected versions.
+ */
+
+/** @hideinitializer
+ * Unsigned integer type whose length is at least 32 bits; on most
+ * architectures, it will have a width of exactly 32 bits. Unsigned C
+ * types implement arithmetics modulo a power of 2; use the
+ * <code>SPH_T32()</code> macro to ensure that the value is truncated
+ * to exactly 32 bits. Unless otherwise specified, all macros and
+ * functions which accept <code>sph_u32</code> values assume that these
+ * values fit on 32 bits, i.e. do not exceed 2^32-1, even on architectures
+ * where <code>sph_u32</code> is larger than that.
+ */
+typedef __arch_dependant__ sph_u32;
+
+/** @hideinitializer
+ * Signed integer type corresponding to <code>sph_u32</code>; it has
+ * width 32 bits or more.
+ */
+typedef __arch_dependant__ sph_s32;
+
+/** @hideinitializer
+ * Unsigned integer type whose length is at least 64 bits; on most
+ * architectures which feature such a type, it will have a width of
+ * exactly 64 bits. C99-compliant platform will have this type; it
+ * is also defined when the GNU compiler (gcc) is used, and on
+ * platforms where <code>unsigned long</code> is large enough. If this
+ * type is not available, then some hash functions which depends on
+ * a 64-bit type will not be available (most notably SHA-384, SHA-512,
+ * Tiger and WHIRLPOOL).
+ */
+typedef __arch_dependant__ sph_u64;
+
+/** @hideinitializer
+ * Signed integer type corresponding to <code>sph_u64</code>; it has
+ * width 64 bits or more.
+ */
+typedef __arch_dependant__ sph_s64;
+
+/**
+ * This macro expands the token <code>x</code> into a suitable
+ * constant expression of type <code>sph_u32</code>. Depending on
+ * how this type is defined, a suffix such as <code>UL</code> may
+ * be appended to the argument.
+ *
+ * @param x   the token to expand into a suitable constant expression
+ */
+#define SPH_C32(x)
+
+/**
+ * Truncate a 32-bit value to exactly 32 bits. On most systems, this is
+ * a no-op, recognized as such by the compiler.
+ *
+ * @param x   the value to truncate (of type <code>sph_u32</code>)
+ */
+#define SPH_T32(x)
+
+/**
+ * Rotate a 32-bit value by a number of bits to the left. The rotate
+ * count must reside between 1 and 31. This macro assumes that its
+ * first argument fits in 32 bits (no extra bit allowed on machines where
+ * <code>sph_u32</code> is wider); both arguments may be evaluated
+ * several times.
+ *
+ * @param x   the value to rotate (of type <code>sph_u32</code>)
+ * @param n   the rotation count (between 1 and 31, inclusive)
+ */
+#define SPH_ROTL32(x, n)
+
+/**
+ * Rotate a 32-bit value by a number of bits to the left. The rotate
+ * count must reside between 1 and 31. This macro assumes that its
+ * first argument fits in 32 bits (no extra bit allowed on machines where
+ * <code>sph_u32</code> is wider); both arguments may be evaluated
+ * several times.
+ *
+ * @param x   the value to rotate (of type <code>sph_u32</code>)
+ * @param n   the rotation count (between 1 and 31, inclusive)
+ */
+#define SPH_ROTR32(x, n)
+
+/**
+ * This macro is defined on systems for which a 64-bit type has been
+ * detected, and is used for <code>sph_u64</code>.
+ */
+#define SPH_64
+
+/**
+ * This macro is defined on systems for the "native" integer size is
+ * 64 bits (64-bit values fit in one register).
+ */
+#define SPH_64_TRUE
+
+/**
+ * This macro expands the token <code>x</code> into a suitable
+ * constant expression of type <code>sph_u64</code>. Depending on
+ * how this type is defined, a suffix such as <code>ULL</code> may
+ * be appended to the argument. This macro is defined only if a
+ * 64-bit type was detected and used for <code>sph_u64</code>.
+ *
+ * @param x   the token to expand into a suitable constant expression
+ */
+#define SPH_C64(x)
+
+/**
+ * Truncate a 64-bit value to exactly 64 bits. On most systems, this is
+ * a no-op, recognized as such by the compiler. This macro is defined only
+ * if a 64-bit type was detected and used for <code>sph_u64</code>.
+ *
+ * @param x   the value to truncate (of type <code>sph_u64</code>)
+ */
+#define SPH_T64(x)
+
+/**
+ * Rotate a 64-bit value by a number of bits to the left. The rotate
+ * count must reside between 1 and 63. This macro assumes that its
+ * first argument fits in 64 bits (no extra bit allowed on machines where
+ * <code>sph_u64</code> is wider); both arguments may be evaluated
+ * several times. This macro is defined only if a 64-bit type was detected
+ * and used for <code>sph_u64</code>.
+ *
+ * @param x   the value to rotate (of type <code>sph_u64</code>)
+ * @param n   the rotation count (between 1 and 63, inclusive)
+ */
+#define SPH_ROTL64(x, n)
+
+/**
+ * Rotate a 64-bit value by a number of bits to the left. The rotate
+ * count must reside between 1 and 63. This macro assumes that its
+ * first argument fits in 64 bits (no extra bit allowed on machines where
+ * <code>sph_u64</code> is wider); both arguments may be evaluated
+ * several times. This macro is defined only if a 64-bit type was detected
+ * and used for <code>sph_u64</code>.
+ *
+ * @param x   the value to rotate (of type <code>sph_u64</code>)
+ * @param n   the rotation count (between 1 and 63, inclusive)
+ */
+#define SPH_ROTR64(x, n)
+
+/**
+ * This macro evaluates to <code>inline</code> or an equivalent construction,
+ * if available on the compilation platform, or to nothing otherwise. This
+ * is used to declare inline functions, for which the compiler should
+ * endeavour to include the code directly in the caller. Inline functions
+ * are typically defined in header files as replacement for macros.
+ */
+#define SPH_INLINE
+
+/**
+ * This macro is defined if the platform has been detected as using
+ * little-endian convention. This implies that the <code>sph_u32</code>
+ * type (and the <code>sph_u64</code> type also, if it is defined) has
+ * an exact width (i.e. exactly 32-bit, respectively 64-bit).
+ */
+#define SPH_LITTLE_ENDIAN
+
+/**
+ * This macro is defined if the platform has been detected as using
+ * big-endian convention. This implies that the <code>sph_u32</code>
+ * type (and the <code>sph_u64</code> type also, if it is defined) has
+ * an exact width (i.e. exactly 32-bit, respectively 64-bit).
+ */
+#define SPH_BIG_ENDIAN
+
+/**
+ * This macro is defined if 32-bit words (and 64-bit words, if defined)
+ * can be read from and written to memory efficiently in little-endian
+ * convention. This is the case for little-endian platforms, and also
+ * for the big-endian platforms which have special little-endian access
+ * opcodes (e.g. Ultrasparc).
+ */
+#define SPH_LITTLE_FAST
+
+/**
+ * This macro is defined if 32-bit words (and 64-bit words, if defined)
+ * can be read from and written to memory efficiently in big-endian
+ * convention. This is the case for little-endian platforms, and also
+ * for the little-endian platforms which have special big-endian access
+ * opcodes.
+ */
+#define SPH_BIG_FAST
+
+/**
+ * On some platforms, this macro is defined to an unsigned integer type
+ * into which pointer values may be cast. The resulting value can then
+ * be tested for being a multiple of 2, 4 or 8, indicating an aligned
+ * pointer for, respectively, 16-bit, 32-bit or 64-bit memory accesses.
+ */
+#define SPH_UPTR
+
+/**
+ * When defined, this macro indicates that unaligned memory accesses
+ * are possible with only a minor penalty, and thus should be prefered
+ * over strategies which first copy data to an aligned buffer.
+ */
+#define SPH_UNALIGNED
+
+/**
+ * Byte-swap a 32-bit word (i.e. <code>0x12345678</code> becomes
+ * <code>0x78563412</code>). This is an inline function which resorts
+ * to inline assembly on some platforms, for better performance.
+ *
+ * @param x   the 32-bit value to byte-swap
+ * @return  the byte-swapped value
+ */
+static inline sph_u32 sph_bswap32(sph_u32 x);
+
+/**
+ * Byte-swap a 64-bit word. This is an inline function which resorts
+ * to inline assembly on some platforms, for better performance. This
+ * function is defined only if a suitable 64-bit type was found for
+ * <code>sph_u64</code>
+ *
+ * @param x   the 64-bit value to byte-swap
+ * @return  the byte-swapped value
+ */
+static inline sph_u64 sph_bswap64(sph_u64 x);
+
+/**
+ * Decode a 16-bit unsigned value from memory, in little-endian convention
+ * (least significant byte comes first).
+ *
+ * @param src   the source address
+ * @return  the decoded value
+ */
+static inline unsigned sph_dec16le(const void *src);
+
+/**
+ * Encode a 16-bit unsigned value into memory, in little-endian convention
+ * (least significant byte comes first).
+ *
+ * @param dst   the destination buffer
+ * @param val   the value to encode
+ */
+static inline void sph_enc16le(void *dst, unsigned val);
+
+/**
+ * Decode a 16-bit unsigned value from memory, in big-endian convention
+ * (most significant byte comes first).
+ *
+ * @param src   the source address
+ * @return  the decoded value
+ */
+static inline unsigned sph_dec16be(const void *src);
+
+/**
+ * Encode a 16-bit unsigned value into memory, in big-endian convention
+ * (most significant byte comes first).
+ *
+ * @param dst   the destination buffer
+ * @param val   the value to encode
+ */
+static inline void sph_enc16be(void *dst, unsigned val);
+
+/**
+ * Decode a 32-bit unsigned value from memory, in little-endian convention
+ * (least significant byte comes first).
+ *
+ * @param src   the source address
+ * @return  the decoded value
+ */
+static inline sph_u32 sph_dec32le(const void *src);
+
+/**
+ * Decode a 32-bit unsigned value from memory, in little-endian convention
+ * (least significant byte comes first). This function assumes that the
+ * source address is suitably aligned for a direct access, if the platform
+ * supports such things; it can thus be marginally faster than the generic
+ * <code>sph_dec32le()</code> function.
+ *
+ * @param src   the source address
+ * @return  the decoded value
+ */
+static inline sph_u32 sph_dec32le_aligned(const void *src);
+
+/**
+ * Encode a 32-bit unsigned value into memory, in little-endian convention
+ * (least significant byte comes first).
+ *
+ * @param dst   the destination buffer
+ * @param val   the value to encode
+ */
+static inline void sph_enc32le(void *dst, sph_u32 val);
+
+/**
+ * Encode a 32-bit unsigned value into memory, in little-endian convention
+ * (least significant byte comes first). This function assumes that the
+ * destination address is suitably aligned for a direct access, if the
+ * platform supports such things; it can thus be marginally faster than
+ * the generic <code>sph_enc32le()</code> function.
+ *
+ * @param dst   the destination buffer
+ * @param val   the value to encode
+ */
+static inline void sph_enc32le_aligned(void *dst, sph_u32 val);
+
+/**
+ * Decode a 32-bit unsigned value from memory, in big-endian convention
+ * (most significant byte comes first).
+ *
+ * @param src   the source address
+ * @return  the decoded value
+ */
+static inline sph_u32 sph_dec32be(const void *src);
+
+/**
+ * Decode a 32-bit unsigned value from memory, in big-endian convention
+ * (most significant byte comes first). This function assumes that the
+ * source address is suitably aligned for a direct access, if the platform
+ * supports such things; it can thus be marginally faster than the generic
+ * <code>sph_dec32be()</code> function.
+ *
+ * @param src   the source address
+ * @return  the decoded value
+ */
+static inline sph_u32 sph_dec32be_aligned(const void *src);
+
+/**
+ * Encode a 32-bit unsigned value into memory, in big-endian convention
+ * (most significant byte comes first).
+ *
+ * @param dst   the destination buffer
+ * @param val   the value to encode
+ */
+static inline void sph_enc32be(void *dst, sph_u32 val);
+
+/**
+ * Encode a 32-bit unsigned value into memory, in big-endian convention
+ * (most significant byte comes first). This function assumes that the
+ * destination address is suitably aligned for a direct access, if the
+ * platform supports such things; it can thus be marginally faster than
+ * the generic <code>sph_enc32be()</code> function.
+ *
+ * @param dst   the destination buffer
+ * @param val   the value to encode
+ */
+static inline void sph_enc32be_aligned(void *dst, sph_u32 val);
+
+/**
+ * Decode a 64-bit unsigned value from memory, in little-endian convention
+ * (least significant byte comes first). This function is defined only
+ * if a suitable 64-bit type was detected and used for <code>sph_u64</code>.
+ *
+ * @param src   the source address
+ * @return  the decoded value
+ */
+static inline sph_u64 sph_dec64le(const void *src);
+
+/**
+ * Decode a 64-bit unsigned value from memory, in little-endian convention
+ * (least significant byte comes first). This function assumes that the
+ * source address is suitably aligned for a direct access, if the platform
+ * supports such things; it can thus be marginally faster than the generic
+ * <code>sph_dec64le()</code> function. This function is defined only
+ * if a suitable 64-bit type was detected and used for <code>sph_u64</code>.
+ *
+ * @param src   the source address
+ * @return  the decoded value
+ */
+static inline sph_u64 sph_dec64le_aligned(const void *src);
+
+/**
+ * Encode a 64-bit unsigned value into memory, in little-endian convention
+ * (least significant byte comes first). This function is defined only
+ * if a suitable 64-bit type was detected and used for <code>sph_u64</code>.
+ *
+ * @param dst   the destination buffer
+ * @param val   the value to encode
+ */
+static inline void sph_enc64le(void *dst, sph_u64 val);
+
+/**
+ * Encode a 64-bit unsigned value into memory, in little-endian convention
+ * (least significant byte comes first). This function assumes that the
+ * destination address is suitably aligned for a direct access, if the
+ * platform supports such things; it can thus be marginally faster than
+ * the generic <code>sph_enc64le()</code> function. This function is defined
+ * only if a suitable 64-bit type was detected and used for
+ * <code>sph_u64</code>.
+ *
+ * @param dst   the destination buffer
+ * @param val   the value to encode
+ */
+static inline void sph_enc64le_aligned(void *dst, sph_u64 val);
+
+/**
+ * Decode a 64-bit unsigned value from memory, in big-endian convention
+ * (most significant byte comes first). This function is defined only
+ * if a suitable 64-bit type was detected and used for <code>sph_u64</code>.
+ *
+ * @param src   the source address
+ * @return  the decoded value
+ */
+static inline sph_u64 sph_dec64be(const void *src);
+
+/**
+ * Decode a 64-bit unsigned value from memory, in big-endian convention
+ * (most significant byte comes first). This function assumes that the
+ * source address is suitably aligned for a direct access, if the platform
+ * supports such things; it can thus be marginally faster than the generic
+ * <code>sph_dec64be()</code> function. This function is defined only
+ * if a suitable 64-bit type was detected and used for <code>sph_u64</code>.
+ *
+ * @param src   the source address
+ * @return  the decoded value
+ */
+static inline sph_u64 sph_dec64be_aligned(const void *src);
+
+/**
+ * Encode a 64-bit unsigned value into memory, in big-endian convention
+ * (most significant byte comes first). This function is defined only
+ * if a suitable 64-bit type was detected and used for <code>sph_u64</code>.
+ *
+ * @param dst   the destination buffer
+ * @param val   the value to encode
+ */
+static inline void sph_enc64be(void *dst, sph_u64 val);
+
+/**
+ * Encode a 64-bit unsigned value into memory, in big-endian convention
+ * (most significant byte comes first). This function assumes that the
+ * destination address is suitably aligned for a direct access, if the
+ * platform supports such things; it can thus be marginally faster than
+ * the generic <code>sph_enc64be()</code> function. This function is defined
+ * only if a suitable 64-bit type was detected and used for
+ * <code>sph_u64</code>.
+ *
+ * @param dst   the destination buffer
+ * @param val   the value to encode
+ */
+static inline void sph_enc64be_aligned(void *dst, sph_u64 val);
+
+#endif
+
+/* ============== END documentation block for Doxygen ============= */
+
+#ifndef DOXYGEN_IGNORE
+
+/*
+ * We want to define the types "sph_u32" and "sph_u64" which hold
+ * unsigned values of at least, respectively, 32 and 64 bits. These
+ * tests should select appropriate types for most platforms. The
+ * macro "SPH_64" is defined if the 64-bit is supported.
+ */
+
+#undef SPH_64
+#undef SPH_64_TRUE
+
+#if defined __STDC__ && __STDC_VERSION__ >= 199901L
+
+/*
+ * On C99 implementations, we can use <stdint.h> to get an exact 64-bit
+ * type, if any, or otherwise use a wider type (which must exist, for
+ * C99 conformance).
+ */
+
+#include <stdint.h>
+
+#ifdef UINT32_MAX
+typedef uint32_t sph_u32;
+typedef int32_t sph_s32;
+#else
+typedef uint_fast32_t sph_u32;
+typedef int_fast32_t sph_s32;
+#endif
+#if !SPH_NO_64
+#ifdef UINT64_MAX
+typedef uint64_t sph_u64;
+typedef int64_t sph_s64;
+#else
+typedef uint_fast64_t sph_u64;
+typedef int_fast64_t sph_s64;
+#endif
+#endif
+
+#define SPH_C32(x)    ((sph_u32)(x))
+#if !SPH_NO_64
+#define SPH_C64(x)    ((sph_u64)(x))
+#define SPH_64  1
+#endif
+
+#else
+
+/*
+ * On non-C99 systems, we use "unsigned int" if it is wide enough,
+ * "unsigned long" otherwise. This supports all "reasonable" architectures.
+ * We have to be cautious: pre-C99 preprocessors handle constants
+ * differently in '#if' expressions. Hence the shifts to test UINT_MAX.
+ */
+
+#if ((UINT_MAX >> 11) >> 11) >= 0x3FF
+
+typedef unsigned int sph_u32;
+typedef int sph_s32;
+
+#define SPH_C32(x)    ((sph_u32)(x ## U))
+
+#else
+
+typedef unsigned long sph_u32;
+typedef long sph_s32;
+
+#define SPH_C32(x)    ((sph_u32)(x ## UL))
+
+#endif
+
+#if !SPH_NO_64
+
+/*
+ * We want a 64-bit type. We use "unsigned long" if it is wide enough (as
+ * is common on 64-bit architectures such as AMD64, Alpha or Sparcv9),
+ * "unsigned long long" otherwise, if available. We use ULLONG_MAX to
+ * test whether "unsigned long long" is available; we also know that
+ * gcc features this type, even if the libc header do not know it.
+ */
+
+#if ((ULONG_MAX >> 31) >> 31) >= 3
+
+typedef unsigned long sph_u64;
+typedef long sph_s64;
+
+#define SPH_C64(x)    ((sph_u64)(x ## UL))
+
+#define SPH_64  1
+
+#elif ((ULLONG_MAX >> 31) >> 31) >= 3 || defined __GNUC__
+
+typedef unsigned long long sph_u64;
+typedef long long sph_s64;
+
+#define SPH_C64(x)    ((sph_u64)(x ## ULL))
+
+#define SPH_64  1
+
+#else
+
+/*
+ * No 64-bit type...
+ */
+
+#endif
+
+#endif
+
+#endif
+
+/*
+ * If the "unsigned long" type has length 64 bits or more, then this is
+ * a "true" 64-bit architectures. This is also true with Visual C on
+ * amd64, even though the "long" type is limited to 32 bits.
+ */
+#if SPH_64 && (((ULONG_MAX >> 31) >> 31) >= 3 || defined _M_X64)
+#define SPH_64_TRUE   1
+#endif
+
+/*
+ * Implementation note: some processors have specific opcodes to perform
+ * a rotation. Recent versions of gcc recognize the expression above and
+ * use the relevant opcodes, when appropriate.
+ */
+
+#define SPH_T32(x)    ((x) & SPH_C32(0xFFFFFFFF))
+#define SPH_ROTL32(x, n)   SPH_T32(((x) << (n)) | ((x) >> (32 - (n))))
+#define SPH_ROTR32(x, n)   SPH_ROTL32(x, (32 - (n)))
+
+#if SPH_64
+
+#define SPH_T64(x)    ((x) & SPH_C64(0xFFFFFFFFFFFFFFFF))
+#define SPH_ROTL64(x, n)   SPH_T64(((x) << (n)) | ((x) >> (64 - (n))))
+#define SPH_ROTR64(x, n)   SPH_ROTL64(x, (64 - (n)))
+
+#endif
+
+#ifndef DOXYGEN_IGNORE
+/*
+ * Define SPH_INLINE to be an "inline" qualifier, if available. We define
+ * some small macro-like functions which benefit greatly from being inlined.
+ */
+#if (defined __STDC__ && __STDC_VERSION__ >= 199901L) || defined __GNUC__
+#define SPH_INLINE inline
+#elif defined _MSC_VER
+#define SPH_INLINE __inline
+#else
+#define SPH_INLINE
+#endif
+#endif
+
+/*
+ * We define some macros which qualify the architecture. These macros
+ * may be explicit set externally (e.g. as compiler parameters). The
+ * code below sets those macros if they are not already defined.
+ *
+ * Most macros are boolean, thus evaluate to either zero or non-zero.
+ * The SPH_UPTR macro is special, in that it evaluates to a C type,
+ * or is not defined.
+ *
+ * SPH_UPTR             if defined: unsigned type to cast pointers into
+ *
+ * SPH_UNALIGNED        non-zero if unaligned accesses are efficient
+ * SPH_LITTLE_ENDIAN    non-zero if architecture is known to be little-endian
+ * SPH_BIG_ENDIAN       non-zero if architecture is known to be big-endian
+ * SPH_LITTLE_FAST      non-zero if little-endian decoding is fast
+ * SPH_BIG_FAST         non-zero if big-endian decoding is fast
+ *
+ * If SPH_UPTR is defined, then encoding and decoding of 32-bit and 64-bit
+ * values will try to be "smart". Either SPH_LITTLE_ENDIAN or SPH_BIG_ENDIAN
+ * _must_ be non-zero in those situations. The 32-bit and 64-bit types
+ * _must_ also have an exact width.
+ *
+ * SPH_SPARCV9_GCC_32   UltraSPARC-compatible with gcc, 32-bit mode
+ * SPH_SPARCV9_GCC_64   UltraSPARC-compatible with gcc, 64-bit mode
+ * SPH_SPARCV9_GCC      UltraSPARC-compatible with gcc
+ * SPH_I386_GCC         x86-compatible (32-bit) with gcc
+ * SPH_I386_MSVC        x86-compatible (32-bit) with Microsoft Visual C
+ * SPH_AMD64_GCC        x86-compatible (64-bit) with gcc
+ * SPH_AMD64_MSVC       x86-compatible (64-bit) with Microsoft Visual C
+ * SPH_PPC32_GCC        PowerPC, 32-bit, with gcc
+ * SPH_PPC64_GCC        PowerPC, 64-bit, with gcc
+ *
+ * TODO: enhance automatic detection, for more architectures and compilers.
+ * Endianness is the most important. SPH_UNALIGNED and SPH_UPTR help with
+ * some very fast functions (e.g. MD4) when using unaligned input data.
+ * The CPU-specific-with-GCC macros are useful only for inline assembly,
+ * normally restrained to this header file.
+ */
+
+/*
+ * 32-bit x86, aka "i386 compatible".
+ */
+#if defined __i386__ || defined _M_IX86
+
+#define SPH_DETECT_UNALIGNED         1
+#define SPH_DETECT_LITTLE_ENDIAN     1
+#define SPH_DETECT_UPTR              sph_u32
+#ifdef __GNUC__
+#define SPH_DETECT_I386_GCC          1
+#endif
+#ifdef _MSC_VER
+#define SPH_DETECT_I386_MSVC         1
+#endif
+
+/*
+ * 64-bit x86, hereafter known as "amd64".
+ */
+#elif defined __x86_64 || defined _M_X64
+
+#define SPH_DETECT_UNALIGNED         1
+#define SPH_DETECT_LITTLE_ENDIAN     1
+#define SPH_DETECT_UPTR              sph_u64
+#ifdef __GNUC__
+#define SPH_DETECT_AMD64_GCC         1
+#endif
+#ifdef _MSC_VER
+#define SPH_DETECT_AMD64_MSVC        1
+#endif
+
+/*
+ * 64-bit Sparc architecture (implies v9).
+ */
+#elif ((defined __sparc__ || defined __sparc) && defined __arch64__) \
+	|| defined __sparcv9
+
+#define SPH_DETECT_BIG_ENDIAN        1
+#define SPH_DETECT_UPTR              sph_u64
+#ifdef __GNUC__
+#define SPH_DETECT_SPARCV9_GCC_64    1
+#define SPH_DETECT_LITTLE_FAST       1
+#endif
+
+/*
+ * 32-bit Sparc.
+ */
+#elif (defined __sparc__ || defined __sparc) \
+	&& !(defined __sparcv9 || defined __arch64__)
+
+#define SPH_DETECT_BIG_ENDIAN        1
+#define SPH_DETECT_UPTR              sph_u32
+#if defined __GNUC__ && defined __sparc_v9__
+#define SPH_DETECT_SPARCV9_GCC_32    1
+#define SPH_DETECT_LITTLE_FAST       1
+#endif
+
+/*
+ * ARM, little-endian.
+ */
+#elif defined __arm__ && __ARMEL__
+
+#define SPH_DETECT_LITTLE_ENDIAN     1
+
+/*
+ * MIPS, little-endian.
+ */
+#elif MIPSEL || _MIPSEL || __MIPSEL || __MIPSEL__
+
+#define SPH_DETECT_LITTLE_ENDIAN     1
+
+/*
+ * MIPS, big-endian.
+ */
+#elif MIPSEB || _MIPSEB || __MIPSEB || __MIPSEB__
+
+#define SPH_DETECT_BIG_ENDIAN        1
+
+/*
+ * PowerPC.
+ */
+#elif defined __powerpc__ || defined __POWERPC__ || defined __ppc__ \
+	|| defined _ARCH_PPC
+
+/*
+ * Note: we do not declare cross-endian access to be "fast": even if
+ * using inline assembly, implementation should still assume that
+ * keeping the decoded word in a temporary is faster than decoding
+ * it again.
+ */
+#if defined __GNUC__
+#if SPH_64_TRUE
+#define SPH_DETECT_PPC64_GCC         1
+#else
+#define SPH_DETECT_PPC32_GCC         1
+#endif
+#endif
+
+#if defined __BIG_ENDIAN__ || defined _BIG_ENDIAN
+#define SPH_DETECT_BIG_ENDIAN        1
+#elif defined __LITTLE_ENDIAN__ || defined _LITTLE_ENDIAN
+#define SPH_DETECT_LITTLE_ENDIAN     1
+#endif
+
+/*
+ * Itanium, 64-bit.
+ */
+#elif defined __ia64 || defined __ia64__ \
+	|| defined __itanium__ || defined _M_IA64
+
+#if defined __BIG_ENDIAN__ || defined _BIG_ENDIAN
+#define SPH_DETECT_BIG_ENDIAN        1
+#else
+#define SPH_DETECT_LITTLE_ENDIAN     1
+#endif
+#if defined __LP64__ || defined _LP64
+#define SPH_DETECT_UPTR              sph_u64
+#else
+#define SPH_DETECT_UPTR              sph_u32
+#endif
+
+#endif
+
+#if defined SPH_DETECT_SPARCV9_GCC_32 || defined SPH_DETECT_SPARCV9_GCC_64
+#define SPH_DETECT_SPARCV9_GCC       1
+#endif
+
+#if defined SPH_DETECT_UNALIGNED && !defined SPH_UNALIGNED
+#define SPH_UNALIGNED         SPH_DETECT_UNALIGNED
+#endif
+#if defined SPH_DETECT_UPTR && !defined SPH_UPTR
+#define SPH_UPTR              SPH_DETECT_UPTR
+#endif
+#if defined SPH_DETECT_LITTLE_ENDIAN && !defined SPH_LITTLE_ENDIAN
+#define SPH_LITTLE_ENDIAN     SPH_DETECT_LITTLE_ENDIAN
+#endif
+#if defined SPH_DETECT_BIG_ENDIAN && !defined SPH_BIG_ENDIAN
+#define SPH_BIG_ENDIAN        SPH_DETECT_BIG_ENDIAN
+#endif
+#if defined SPH_DETECT_LITTLE_FAST && !defined SPH_LITTLE_FAST
+#define SPH_LITTLE_FAST       SPH_DETECT_LITTLE_FAST
+#endif
+#if defined SPH_DETECT_BIG_FAST && !defined SPH_BIG_FAST
+#define SPH_BIG_FAST    SPH_DETECT_BIG_FAST
+#endif
+#if defined SPH_DETECT_SPARCV9_GCC_32 && !defined SPH_SPARCV9_GCC_32
+#define SPH_SPARCV9_GCC_32    SPH_DETECT_SPARCV9_GCC_32
+#endif
+#if defined SPH_DETECT_SPARCV9_GCC_64 && !defined SPH_SPARCV9_GCC_64
+#define SPH_SPARCV9_GCC_64    SPH_DETECT_SPARCV9_GCC_64
+#endif
+#if defined SPH_DETECT_SPARCV9_GCC && !defined SPH_SPARCV9_GCC
+#define SPH_SPARCV9_GCC       SPH_DETECT_SPARCV9_GCC
+#endif
+#if defined SPH_DETECT_I386_GCC && !defined SPH_I386_GCC
+#define SPH_I386_GCC          SPH_DETECT_I386_GCC
+#endif
+#if defined SPH_DETECT_I386_MSVC && !defined SPH_I386_MSVC
+#define SPH_I386_MSVC         SPH_DETECT_I386_MSVC
+#endif
+#if defined SPH_DETECT_AMD64_GCC && !defined SPH_AMD64_GCC
+#define SPH_AMD64_GCC         SPH_DETECT_AMD64_GCC
+#endif
+#if defined SPH_DETECT_AMD64_MSVC && !defined SPH_AMD64_MSVC
+#define SPH_AMD64_MSVC        SPH_DETECT_AMD64_MSVC
+#endif
+#if defined SPH_DETECT_PPC32_GCC && !defined SPH_PPC32_GCC
+#define SPH_PPC32_GCC         SPH_DETECT_PPC32_GCC
+#endif
+#if defined SPH_DETECT_PPC64_GCC && !defined SPH_PPC64_GCC
+#define SPH_PPC64_GCC         SPH_DETECT_PPC64_GCC
+#endif
+
+#if SPH_LITTLE_ENDIAN && !defined SPH_LITTLE_FAST
+#define SPH_LITTLE_FAST              1
+#endif
+#if SPH_BIG_ENDIAN && !defined SPH_BIG_FAST
+#define SPH_BIG_FAST                 1
+#endif
+
+#if defined SPH_UPTR && !(SPH_LITTLE_ENDIAN || SPH_BIG_ENDIAN)
+#error SPH_UPTR defined, but endianness is not known.
+#endif
+
+#if SPH_I386_GCC && !SPH_NO_ASM
+
+/*
+ * On x86 32-bit, with gcc, we use the bswapl opcode to byte-swap 32-bit
+ * values.
+ */
+
+static SPH_INLINE sph_u32
+sph_bswap32(sph_u32 x)
+{
+	__asm__ __volatile__ ("bswapl %0" : "=r" (x) : "0" (x));
+	return x;
+}
+
+#if SPH_64
+
+static SPH_INLINE sph_u64
+sph_bswap64(sph_u64 x)
+{
+	return ((sph_u64)sph_bswap32((sph_u32)x) << 32)
+		| (sph_u64)sph_bswap32((sph_u32)(x >> 32));
+}
+
+#endif
+
+#elif SPH_AMD64_GCC && !SPH_NO_ASM
+
+/*
+ * On x86 64-bit, with gcc, we use the bswapl opcode to byte-swap 32-bit
+ * and 64-bit values.
+ */
+
+static SPH_INLINE sph_u32
+sph_bswap32(sph_u32 x)
+{
+	__asm__ __volatile__ ("bswapl %0" : "=r" (x) : "0" (x));
+	return x;
+}
+
+#if SPH_64
+
+static SPH_INLINE sph_u64
+sph_bswap64(sph_u64 x)
+{
+	__asm__ __volatile__ ("bswapq %0" : "=r" (x) : "0" (x));
+	return x;
+}
+
+#endif
+
+/*
+ * Disabled code. Apparently, Microsoft Visual C 2005 is smart enough
+ * to generate proper opcodes for endianness swapping with the pure C
+ * implementation below.
+ *
+
+#elif SPH_I386_MSVC && !SPH_NO_ASM
+
+static __inline sph_u32 __declspec(naked) __fastcall
+sph_bswap32(sph_u32 x)
+{
+	__asm {
+		bswap  ecx
+		mov    eax,ecx
+		ret
+	}
+}
+
+#if SPH_64
+
+static SPH_INLINE sph_u64
+sph_bswap64(sph_u64 x)
+{
+	return ((sph_u64)sph_bswap32((sph_u32)x) << 32)
+		| (sph_u64)sph_bswap32((sph_u32)(x >> 32));
+}
+
+#endif
+
+ *
+ * [end of disabled code]
+ */
+
+#else
+
+static SPH_INLINE sph_u32
+sph_bswap32(sph_u32 x)
+{
+	x = SPH_T32((x << 16) | (x >> 16));
+	x = ((x & SPH_C32(0xFF00FF00)) >> 8)
+		| ((x & SPH_C32(0x00FF00FF)) << 8);
+	return x;
+}
+
+#if SPH_64
+
+/**
+ * Byte-swap a 64-bit value.
+ *
+ * @param x   the input value
+ * @return  the byte-swapped value
+ */
+static SPH_INLINE sph_u64
+sph_bswap64(sph_u64 x)
+{
+	x = SPH_T64((x << 32) | (x >> 32));
+	x = ((x & SPH_C64(0xFFFF0000FFFF0000)) >> 16)
+		| ((x & SPH_C64(0x0000FFFF0000FFFF)) << 16);
+	x = ((x & SPH_C64(0xFF00FF00FF00FF00)) >> 8)
+		| ((x & SPH_C64(0x00FF00FF00FF00FF)) << 8);
+	return x;
+}
+
+#endif
+
+#endif
+
+#if SPH_SPARCV9_GCC && !SPH_NO_ASM
+
+/*
+ * On UltraSPARC systems, native ordering is big-endian, but it is
+ * possible to perform little-endian read accesses by specifying the
+ * address space 0x88 (ASI_PRIMARY_LITTLE). Basically, either we use
+ * the opcode "lda [%reg]0x88,%dst", where %reg is the register which
+ * contains the source address and %dst is the destination register,
+ * or we use "lda [%reg+imm]%asi,%dst", which uses the %asi register
+ * to get the address space name. The latter format is better since it
+ * combines an addition and the actual access in a single opcode; but
+ * it requires the setting (and subsequent resetting) of %asi, which is
+ * slow. Some operations (i.e. MD5 compression function) combine many
+ * successive little-endian read accesses, which may share the same
+ * %asi setting. The macros below contain the appropriate inline
+ * assembly.
+ */
+
+#define SPH_SPARCV9_SET_ASI   \
+	sph_u32 sph_sparcv9_asi; \
+	__asm__ __volatile__ ( \
+		"rd %%asi,%0\n\twr %%g0,0x88,%%asi" : "=r" (sph_sparcv9_asi));
+
+#define SPH_SPARCV9_RESET_ASI  \
+	__asm__ __volatile__ ("wr %%g0,%0,%%asi" : : "r" (sph_sparcv9_asi));
+
+#define SPH_SPARCV9_DEC32LE(base, idx)   ({ \
+		sph_u32 sph_sparcv9_tmp; \
+		__asm__ __volatile__ ("lda [%1+" #idx "*4]%%asi,%0" \
+			: "=r" (sph_sparcv9_tmp) : "r" (base)); \
+		sph_sparcv9_tmp; \
+	})
+
+#endif
+
+static SPH_INLINE void
+sph_enc16be(void *dst, unsigned val)
+{
+	((unsigned char *)dst)[0] = (val >> 8);
+	((unsigned char *)dst)[1] = val;
+}
+
+static SPH_INLINE unsigned
+sph_dec16be(const void *src)
+{
+	return ((unsigned)(((const unsigned char *)src)[0]) << 8)
+		| (unsigned)(((const unsigned char *)src)[1]);
+}
+
+static SPH_INLINE void
+sph_enc16le(void *dst, unsigned val)
+{
+	((unsigned char *)dst)[0] = val;
+	((unsigned char *)dst)[1] = val >> 8;
+}
+
+static SPH_INLINE unsigned
+sph_dec16le(const void *src)
+{
+	return (unsigned)(((const unsigned char *)src)[0])
+		| ((unsigned)(((const unsigned char *)src)[1]) << 8);
+}
+
+/**
+ * Encode a 32-bit value into the provided buffer (big endian convention).
+ *
+ * @param dst   the destination buffer
+ * @param val   the 32-bit value to encode
+ */
+static SPH_INLINE void
+sph_enc32be(void *dst, sph_u32 val)
+{
+#if defined SPH_UPTR
+#if SPH_UNALIGNED
+#if SPH_LITTLE_ENDIAN
+	val = sph_bswap32(val);
+#endif
+	*(sph_u32 *)dst = val;
+#else
+	if (((SPH_UPTR)dst & 3) == 0) {
+#if SPH_LITTLE_ENDIAN
+		val = sph_bswap32(val);
+#endif
+		*(sph_u32 *)dst = val;
+	} else {
+		((unsigned char *)dst)[0] = (val >> 24);
+		((unsigned char *)dst)[1] = (val >> 16);
+		((unsigned char *)dst)[2] = (val >> 8);
+		((unsigned char *)dst)[3] = val;
+	}
+#endif
+#else
+	((unsigned char *)dst)[0] = (val >> 24);
+	((unsigned char *)dst)[1] = (val >> 16);
+	((unsigned char *)dst)[2] = (val >> 8);
+	((unsigned char *)dst)[3] = val;
+#endif
+}
+
+/**
+ * Encode a 32-bit value into the provided buffer (big endian convention).
+ * The destination buffer must be properly aligned.
+ *
+ * @param dst   the destination buffer (32-bit aligned)
+ * @param val   the value to encode
+ */
+static SPH_INLINE void
+sph_enc32be_aligned(void *dst, sph_u32 val)
+{
+#if SPH_LITTLE_ENDIAN
+	*(sph_u32 *)dst = sph_bswap32(val);
+#elif SPH_BIG_ENDIAN
+	*(sph_u32 *)dst = val;
+#else
+	((unsigned char *)dst)[0] = (val >> 24);
+	((unsigned char *)dst)[1] = (val >> 16);
+	((unsigned char *)dst)[2] = (val >> 8);
+	((unsigned char *)dst)[3] = val;
+#endif
+}
+
+/**
+ * Decode a 32-bit value from the provided buffer (big endian convention).
+ *
+ * @param src   the source buffer
+ * @return  the decoded value
+ */
+static SPH_INLINE sph_u32
+sph_dec32be(const void *src)
+{
+#if defined SPH_UPTR
+#if SPH_UNALIGNED
+#if SPH_LITTLE_ENDIAN
+	return sph_bswap32(*(const sph_u32 *)src);
+#else
+	return *(const sph_u32 *)src;
+#endif
+#else
+	if (((SPH_UPTR)src & 3) == 0) {
+#if SPH_LITTLE_ENDIAN
+		return sph_bswap32(*(const sph_u32 *)src);
+#else
+		return *(const sph_u32 *)src;
+#endif
+	} else {
+		return ((sph_u32)(((const unsigned char *)src)[0]) << 24)
+			| ((sph_u32)(((const unsigned char *)src)[1]) << 16)
+			| ((sph_u32)(((const unsigned char *)src)[2]) << 8)
+			| (sph_u32)(((const unsigned char *)src)[3]);
+	}
+#endif
+#else
+	return ((sph_u32)(((const unsigned char *)src)[0]) << 24)
+		| ((sph_u32)(((const unsigned char *)src)[1]) << 16)
+		| ((sph_u32)(((const unsigned char *)src)[2]) << 8)
+		| (sph_u32)(((const unsigned char *)src)[3]);
+#endif
+}
+
+/**
+ * Decode a 32-bit value from the provided buffer (big endian convention).
+ * The source buffer must be properly aligned.
+ *
+ * @param src   the source buffer (32-bit aligned)
+ * @return  the decoded value
+ */
+static SPH_INLINE sph_u32
+sph_dec32be_aligned(const void *src)
+{
+#if SPH_LITTLE_ENDIAN
+	return sph_bswap32(*(const sph_u32 *)src);
+#elif SPH_BIG_ENDIAN
+	return *(const sph_u32 *)src;
+#else
+	return ((sph_u32)(((const unsigned char *)src)[0]) << 24)
+		| ((sph_u32)(((const unsigned char *)src)[1]) << 16)
+		| ((sph_u32)(((const unsigned char *)src)[2]) << 8)
+		| (sph_u32)(((const unsigned char *)src)[3]);
+#endif
+}
+
+/**
+ * Encode a 32-bit value into the provided buffer (little endian convention).
+ *
+ * @param dst   the destination buffer
+ * @param val   the 32-bit value to encode
+ */
+static SPH_INLINE void
+sph_enc32le(void *dst, sph_u32 val)
+{
+#if defined SPH_UPTR
+#if SPH_UNALIGNED
+#if SPH_BIG_ENDIAN
+	val = sph_bswap32(val);
+#endif
+	*(sph_u32 *)dst = val;
+#else
+	if (((SPH_UPTR)dst & 3) == 0) {
+#if SPH_BIG_ENDIAN
+		val = sph_bswap32(val);
+#endif
+		*(sph_u32 *)dst = val;
+	} else {
+		((unsigned char *)dst)[0] = val;
+		((unsigned char *)dst)[1] = (val >> 8);
+		((unsigned char *)dst)[2] = (val >> 16);
+		((unsigned char *)dst)[3] = (val >> 24);
+	}
+#endif
+#else
+	((unsigned char *)dst)[0] = val;
+	((unsigned char *)dst)[1] = (val >> 8);
+	((unsigned char *)dst)[2] = (val >> 16);
+	((unsigned char *)dst)[3] = (val >> 24);
+#endif
+}
+
+/**
+ * Encode a 32-bit value into the provided buffer (little endian convention).
+ * The destination buffer must be properly aligned.
+ *
+ * @param dst   the destination buffer (32-bit aligned)
+ * @param val   the value to encode
+ */
+static SPH_INLINE void
+sph_enc32le_aligned(void *dst, sph_u32 val)
+{
+#if SPH_LITTLE_ENDIAN
+	*(sph_u32 *)dst = val;
+#elif SPH_BIG_ENDIAN
+	*(sph_u32 *)dst = sph_bswap32(val);
+#else
+	((unsigned char *)dst)[0] = val;
+	((unsigned char *)dst)[1] = (val >> 8);
+	((unsigned char *)dst)[2] = (val >> 16);
+	((unsigned char *)dst)[3] = (val >> 24);
+#endif
+}
+
+/**
+ * Decode a 32-bit value from the provided buffer (little endian convention).
+ *
+ * @param src   the source buffer
+ * @return  the decoded value
+ */
+static SPH_INLINE sph_u32
+sph_dec32le(const void *src)
+{
+#if defined SPH_UPTR
+#if SPH_UNALIGNED
+#if SPH_BIG_ENDIAN
+	return sph_bswap32(*(const sph_u32 *)src);
+#else
+	return *(const sph_u32 *)src;
+#endif
+#else
+	if (((SPH_UPTR)src & 3) == 0) {
+#if SPH_BIG_ENDIAN
+#if SPH_SPARCV9_GCC && !SPH_NO_ASM
+		sph_u32 tmp;
+
+		/*
+		 * "__volatile__" is needed here because without it,
+		 * gcc-3.4.3 miscompiles the code and performs the
+		 * access before the test on the address, thus triggering
+		 * a bus error...
+		 */
+		__asm__ __volatile__ (
+			"lda [%1]0x88,%0" : "=r" (tmp) : "r" (src));
+		return tmp;
+/*
+ * On PowerPC, this turns out not to be worth the effort: the inline
+ * assembly makes GCC optimizer uncomfortable, which tends to nullify
+ * the decoding gains.
+ *
+ * For most hash functions, using this inline assembly trick changes
+ * hashing speed by less than 5% and often _reduces_ it. The biggest
+ * gains are for MD4 (+11%) and CubeHash (+30%). For all others, it is
+ * less then 10%. The speed gain on CubeHash is probably due to the
+ * chronic shortage of registers that CubeHash endures; for the other
+ * functions, the generic code appears to be efficient enough already.
+ *
+#elif (SPH_PPC32_GCC || SPH_PPC64_GCC) && !SPH_NO_ASM
+		sph_u32 tmp;
+
+		__asm__ __volatile__ (
+			"lwbrx %0,0,%1" : "=r" (tmp) : "r" (src));
+		return tmp;
+ */
+#else
+		return sph_bswap32(*(const sph_u32 *)src);
+#endif
+#else
+		return *(const sph_u32 *)src;
+#endif
+	} else {
+		return (sph_u32)(((const unsigned char *)src)[0])
+			| ((sph_u32)(((const unsigned char *)src)[1]) << 8)
+			| ((sph_u32)(((const unsigned char *)src)[2]) << 16)
+			| ((sph_u32)(((const unsigned char *)src)[3]) << 24);
+	}
+#endif
+#else
+	return (sph_u32)(((const unsigned char *)src)[0])
+		| ((sph_u32)(((const unsigned char *)src)[1]) << 8)
+		| ((sph_u32)(((const unsigned char *)src)[2]) << 16)
+		| ((sph_u32)(((const unsigned char *)src)[3]) << 24);
+#endif
+}
+
+/**
+ * Decode a 32-bit value from the provided buffer (little endian convention).
+ * The source buffer must be properly aligned.
+ *
+ * @param src   the source buffer (32-bit aligned)
+ * @return  the decoded value
+ */
+static SPH_INLINE sph_u32
+sph_dec32le_aligned(const void *src)
+{
+#if SPH_LITTLE_ENDIAN
+	return *(const sph_u32 *)src;
+#elif SPH_BIG_ENDIAN
+#if SPH_SPARCV9_GCC && !SPH_NO_ASM
+	sph_u32 tmp;
+
+	__asm__ __volatile__ ("lda [%1]0x88,%0" : "=r" (tmp) : "r" (src));
+	return tmp;
+/*
+ * Not worth it generally.
+ *
+#elif (SPH_PPC32_GCC || SPH_PPC64_GCC) && !SPH_NO_ASM
+	sph_u32 tmp;
+
+	__asm__ __volatile__ ("lwbrx %0,0,%1" : "=r" (tmp) : "r" (src));
+	return tmp;
+ */
+#else
+	return sph_bswap32(*(const sph_u32 *)src);
+#endif
+#else
+	return (sph_u32)(((const unsigned char *)src)[0])
+		| ((sph_u32)(((const unsigned char *)src)[1]) << 8)
+		| ((sph_u32)(((const unsigned char *)src)[2]) << 16)
+		| ((sph_u32)(((const unsigned char *)src)[3]) << 24);
+#endif
+}
+
+#if SPH_64
+
+/**
+ * Encode a 64-bit value into the provided buffer (big endian convention).
+ *
+ * @param dst   the destination buffer
+ * @param val   the 64-bit value to encode
+ */
+static SPH_INLINE void
+sph_enc64be(void *dst, sph_u64 val)
+{
+#if defined SPH_UPTR
+#if SPH_UNALIGNED
+#if SPH_LITTLE_ENDIAN
+	val = sph_bswap64(val);
+#endif
+	*(sph_u64 *)dst = val;
+#else
+	if (((SPH_UPTR)dst & 7) == 0) {
+#if SPH_LITTLE_ENDIAN
+		val = sph_bswap64(val);
+#endif
+		*(sph_u64 *)dst = val;
+	} else {
+		((unsigned char *)dst)[0] = (val >> 56);
+		((unsigned char *)dst)[1] = (val >> 48);
+		((unsigned char *)dst)[2] = (val >> 40);
+		((unsigned char *)dst)[3] = (val >> 32);
+		((unsigned char *)dst)[4] = (val >> 24);
+		((unsigned char *)dst)[5] = (val >> 16);
+		((unsigned char *)dst)[6] = (val >> 8);
+		((unsigned char *)dst)[7] = val;
+	}
+#endif
+#else
+	((unsigned char *)dst)[0] = (val >> 56);
+	((unsigned char *)dst)[1] = (val >> 48);
+	((unsigned char *)dst)[2] = (val >> 40);
+	((unsigned char *)dst)[3] = (val >> 32);
+	((unsigned char *)dst)[4] = (val >> 24);
+	((unsigned char *)dst)[5] = (val >> 16);
+	((unsigned char *)dst)[6] = (val >> 8);
+	((unsigned char *)dst)[7] = val;
+#endif
+}
+
+/**
+ * Encode a 64-bit value into the provided buffer (big endian convention).
+ * The destination buffer must be properly aligned.
+ *
+ * @param dst   the destination buffer (64-bit aligned)
+ * @param val   the value to encode
+ */
+static SPH_INLINE void
+sph_enc64be_aligned(void *dst, sph_u64 val)
+{
+#if SPH_LITTLE_ENDIAN
+	*(sph_u64 *)dst = sph_bswap64(val);
+#elif SPH_BIG_ENDIAN
+	*(sph_u64 *)dst = val;
+#else
+	((unsigned char *)dst)[0] = (val >> 56);
+	((unsigned char *)dst)[1] = (val >> 48);
+	((unsigned char *)dst)[2] = (val >> 40);
+	((unsigned char *)dst)[3] = (val >> 32);
+	((unsigned char *)dst)[4] = (val >> 24);
+	((unsigned char *)dst)[5] = (val >> 16);
+	((unsigned char *)dst)[6] = (val >> 8);
+	((unsigned char *)dst)[7] = val;
+#endif
+}
+
+/**
+ * Decode a 64-bit value from the provided buffer (big endian convention).
+ *
+ * @param src   the source buffer
+ * @return  the decoded value
+ */
+static SPH_INLINE sph_u64
+sph_dec64be(const void *src)
+{
+#if defined SPH_UPTR
+#if SPH_UNALIGNED
+#if SPH_LITTLE_ENDIAN
+	return sph_bswap64(*(const sph_u64 *)src);
+#else
+	return *(const sph_u64 *)src;
+#endif
+#else
+	if (((SPH_UPTR)src & 7) == 0) {
+#if SPH_LITTLE_ENDIAN
+		return sph_bswap64(*(const sph_u64 *)src);
+#else
+		return *(const sph_u64 *)src;
+#endif
+	} else {
+		return ((sph_u64)(((const unsigned char *)src)[0]) << 56)
+			| ((sph_u64)(((const unsigned char *)src)[1]) << 48)
+			| ((sph_u64)(((const unsigned char *)src)[2]) << 40)
+			| ((sph_u64)(((const unsigned char *)src)[3]) << 32)
+			| ((sph_u64)(((const unsigned char *)src)[4]) << 24)
+			| ((sph_u64)(((const unsigned char *)src)[5]) << 16)
+			| ((sph_u64)(((const unsigned char *)src)[6]) << 8)
+			| (sph_u64)(((const unsigned char *)src)[7]);
+	}
+#endif
+#else
+	return ((sph_u64)(((const unsigned char *)src)[0]) << 56)
+		| ((sph_u64)(((const unsigned char *)src)[1]) << 48)
+		| ((sph_u64)(((const unsigned char *)src)[2]) << 40)
+		| ((sph_u64)(((const unsigned char *)src)[3]) << 32)
+		| ((sph_u64)(((const unsigned char *)src)[4]) << 24)
+		| ((sph_u64)(((const unsigned char *)src)[5]) << 16)
+		| ((sph_u64)(((const unsigned char *)src)[6]) << 8)
+		| (sph_u64)(((const unsigned char *)src)[7]);
+#endif
+}
+
+/**
+ * Decode a 64-bit value from the provided buffer (big endian convention).
+ * The source buffer must be properly aligned.
+ *
+ * @param src   the source buffer (64-bit aligned)
+ * @return  the decoded value
+ */
+static SPH_INLINE sph_u64
+sph_dec64be_aligned(const void *src)
+{
+#if SPH_LITTLE_ENDIAN
+	return sph_bswap64(*(const sph_u64 *)src);
+#elif SPH_BIG_ENDIAN
+	return *(const sph_u64 *)src;
+#else
+	return ((sph_u64)(((const unsigned char *)src)[0]) << 56)
+		| ((sph_u64)(((const unsigned char *)src)[1]) << 48)
+		| ((sph_u64)(((const unsigned char *)src)[2]) << 40)
+		| ((sph_u64)(((const unsigned char *)src)[3]) << 32)
+		| ((sph_u64)(((const unsigned char *)src)[4]) << 24)
+		| ((sph_u64)(((const unsigned char *)src)[5]) << 16)
+		| ((sph_u64)(((const unsigned char *)src)[6]) << 8)
+		| (sph_u64)(((const unsigned char *)src)[7]);
+#endif
+}
+
+/**
+ * Encode a 64-bit value into the provided buffer (little endian convention).
+ *
+ * @param dst   the destination buffer
+ * @param val   the 64-bit value to encode
+ */
+static SPH_INLINE void
+sph_enc64le(void *dst, sph_u64 val)
+{
+#if defined SPH_UPTR
+#if SPH_UNALIGNED
+#if SPH_BIG_ENDIAN
+	val = sph_bswap64(val);
+#endif
+	*(sph_u64 *)dst = val;
+#else
+	if (((SPH_UPTR)dst & 7) == 0) {
+#if SPH_BIG_ENDIAN
+		val = sph_bswap64(val);
+#endif
+		*(sph_u64 *)dst = val;
+	} else {
+		((unsigned char *)dst)[0] = val;
+		((unsigned char *)dst)[1] = (val >> 8);
+		((unsigned char *)dst)[2] = (val >> 16);
+		((unsigned char *)dst)[3] = (val >> 24);
+		((unsigned char *)dst)[4] = (val >> 32);
+		((unsigned char *)dst)[5] = (val >> 40);
+		((unsigned char *)dst)[6] = (val >> 48);
+		((unsigned char *)dst)[7] = (val >> 56);
+	}
+#endif
+#else
+	((unsigned char *)dst)[0] = val;
+	((unsigned char *)dst)[1] = (val >> 8);
+	((unsigned char *)dst)[2] = (val >> 16);
+	((unsigned char *)dst)[3] = (val >> 24);
+	((unsigned char *)dst)[4] = (val >> 32);
+	((unsigned char *)dst)[5] = (val >> 40);
+	((unsigned char *)dst)[6] = (val >> 48);
+	((unsigned char *)dst)[7] = (val >> 56);
+#endif
+}
+
+/**
+ * Encode a 64-bit value into the provided buffer (little endian convention).
+ * The destination buffer must be properly aligned.
+ *
+ * @param dst   the destination buffer (64-bit aligned)
+ * @param val   the value to encode
+ */
+static SPH_INLINE void
+sph_enc64le_aligned(void *dst, sph_u64 val)
+{
+#if SPH_LITTLE_ENDIAN
+	*(sph_u64 *)dst = val;
+#elif SPH_BIG_ENDIAN
+	*(sph_u64 *)dst = sph_bswap64(val);
+#else
+	((unsigned char *)dst)[0] = val;
+	((unsigned char *)dst)[1] = (val >> 8);
+	((unsigned char *)dst)[2] = (val >> 16);
+	((unsigned char *)dst)[3] = (val >> 24);
+	((unsigned char *)dst)[4] = (val >> 32);
+	((unsigned char *)dst)[5] = (val >> 40);
+	((unsigned char *)dst)[6] = (val >> 48);
+	((unsigned char *)dst)[7] = (val >> 56);
+#endif
+}
+
+/**
+ * Decode a 64-bit value from the provided buffer (little endian convention).
+ *
+ * @param src   the source buffer
+ * @return  the decoded value
+ */
+static SPH_INLINE sph_u64
+sph_dec64le(const void *src)
+{
+#if defined SPH_UPTR
+#if SPH_UNALIGNED
+#if SPH_BIG_ENDIAN
+	return sph_bswap64(*(const sph_u64 *)src);
+#else
+	return *(const sph_u64 *)src;
+#endif
+#else
+	if (((SPH_UPTR)src & 7) == 0) {
+#if SPH_BIG_ENDIAN
+#if SPH_SPARCV9_GCC_64 && !SPH_NO_ASM
+		sph_u64 tmp;
+
+		__asm__ __volatile__ (
+			"ldxa [%1]0x88,%0" : "=r" (tmp) : "r" (src));
+		return tmp;
+/*
+ * Not worth it generally.
+ *
+#elif SPH_PPC32_GCC && !SPH_NO_ASM
+		return (sph_u64)sph_dec32le_aligned(src)
+			| ((sph_u64)sph_dec32le_aligned(
+				(const char *)src + 4) << 32);
+#elif SPH_PPC64_GCC && !SPH_NO_ASM
+		sph_u64 tmp;
+
+		__asm__ __volatile__ (
+			"ldbrx %0,0,%1" : "=r" (tmp) : "r" (src));
+		return tmp;
+ */
+#else
+		return sph_bswap64(*(const sph_u64 *)src);
+#endif
+#else
+		return *(const sph_u64 *)src;
+#endif
+	} else {
+		return (sph_u64)(((const unsigned char *)src)[0])
+			| ((sph_u64)(((const unsigned char *)src)[1]) << 8)
+			| ((sph_u64)(((const unsigned char *)src)[2]) << 16)
+			| ((sph_u64)(((const unsigned char *)src)[3]) << 24)
+			| ((sph_u64)(((const unsigned char *)src)[4]) << 32)
+			| ((sph_u64)(((const unsigned char *)src)[5]) << 40)
+			| ((sph_u64)(((const unsigned char *)src)[6]) << 48)
+			| ((sph_u64)(((const unsigned char *)src)[7]) << 56);
+	}
+#endif
+#else
+	return (sph_u64)(((const unsigned char *)src)[0])
+		| ((sph_u64)(((const unsigned char *)src)[1]) << 8)
+		| ((sph_u64)(((const unsigned char *)src)[2]) << 16)
+		| ((sph_u64)(((const unsigned char *)src)[3]) << 24)
+		| ((sph_u64)(((const unsigned char *)src)[4]) << 32)
+		| ((sph_u64)(((const unsigned char *)src)[5]) << 40)
+		| ((sph_u64)(((const unsigned char *)src)[6]) << 48)
+		| ((sph_u64)(((const unsigned char *)src)[7]) << 56);
+#endif
+}
+
+/**
+ * Decode a 64-bit value from the provided buffer (little endian convention).
+ * The source buffer must be properly aligned.
+ *
+ * @param src   the source buffer (64-bit aligned)
+ * @return  the decoded value
+ */
+static SPH_INLINE sph_u64
+sph_dec64le_aligned(const void *src)
+{
+#if SPH_LITTLE_ENDIAN
+	return *(const sph_u64 *)src;
+#elif SPH_BIG_ENDIAN
+#if SPH_SPARCV9_GCC_64 && !SPH_NO_ASM
+	sph_u64 tmp;
+
+	__asm__ __volatile__ ("ldxa [%1]0x88,%0" : "=r" (tmp) : "r" (src));
+	return tmp;
+/*
+ * Not worth it generally.
+ *
+#elif SPH_PPC32_GCC && !SPH_NO_ASM
+	return (sph_u64)sph_dec32le_aligned(src)
+		| ((sph_u64)sph_dec32le_aligned((const char *)src + 4) << 32);
+#elif SPH_PPC64_GCC && !SPH_NO_ASM
+	sph_u64 tmp;
+
+	__asm__ __volatile__ ("ldbrx %0,0,%1" : "=r" (tmp) : "r" (src));
+	return tmp;
+ */
+#else
+	return sph_bswap64(*(const sph_u64 *)src);
+#endif
+#else
+	return (sph_u64)(((const unsigned char *)src)[0])
+		| ((sph_u64)(((const unsigned char *)src)[1]) << 8)
+		| ((sph_u64)(((const unsigned char *)src)[2]) << 16)
+		| ((sph_u64)(((const unsigned char *)src)[3]) << 24)
+		| ((sph_u64)(((const unsigned char *)src)[4]) << 32)
+		| ((sph_u64)(((const unsigned char *)src)[5]) << 40)
+		| ((sph_u64)(((const unsigned char *)src)[6]) << 48)
+		| ((sph_u64)(((const unsigned char *)src)[7]) << 56);
+#endif
+}
+
+#endif
+
+#endif /* Doxygen excluded block */
+
+#endif
diff --git a/uint256.h b/uint256.h
new file mode 100644
index 0000000000..2a252c94f3
--- /dev/null
+++ b/uint256.h
@@ -0,0 +1,784 @@
+// Copyright (c) 2009-2010 Satoshi Nakamoto
+// Copyright (c) 2009-2012 The Bitcoin developers
+// Distributed under the MIT/X11 software license, see the accompanying
+// file COPYING or http://www.opensource.org/licenses/mit-license.php.
+#ifndef BITCOIN_UINT256_H
+#define BITCOIN_UINT256_H
+
+#include <limits.h>
+#include <stdio.h>
+#include <string.h>
+#include <inttypes.h>
+#include <string>
+#include <vector>
+
+typedef long long  int64;
+typedef unsigned long long  uint64;
+
+
+inline int Testuint256AdHoc(std::vector<std::string> vArg);
+
+
+
+/** Base class without constructors for uint256 and uint160.
+ * This makes the compiler let you use it in a union.
+ */
+template<unsigned int BITS>
+class base_uint
+{
+protected:
+    enum { WIDTH=BITS/32 };
+    uint32_t pn[WIDTH];
+public:
+
+    bool operator!() const
+    {
+        for (int i = 0; i < WIDTH; i++)
+            if (pn[i] != 0)
+                return false;
+        return true;
+    }
+
+    const base_uint operator~() const
+    {
+        base_uint ret;
+        for (int i = 0; i < WIDTH; i++)
+            ret.pn[i] = ~pn[i];
+        return ret;
+    }
+
+    const base_uint operator-() const
+    {
+        base_uint ret;
+        for (int i = 0; i < WIDTH; i++)
+            ret.pn[i] = ~pn[i];
+        ret++;
+        return ret;
+    }
+
+    double getdouble() const
+    {
+        double ret = 0.0;
+        double fact = 1.0;
+        for (int i = 0; i < WIDTH; i++) {
+            ret += fact * pn[i];
+            fact *= 4294967296.0;
+        }
+        return ret;
+    }
+
+    base_uint& operator=(uint64 b)
+    {
+        pn[0] = (unsigned int)b;
+        pn[1] = (unsigned int)(b >> 32);
+        for (int i = 2; i < WIDTH; i++)
+            pn[i] = 0;
+        return *this;
+    }
+
+    base_uint& operator^=(const base_uint& b)
+    {
+        for (int i = 0; i < WIDTH; i++)
+            pn[i] ^= b.pn[i];
+        return *this;
+    }
+
+    base_uint& operator&=(const base_uint& b)
+    {
+        for (int i = 0; i < WIDTH; i++)
+            pn[i] &= b.pn[i];
+        return *this;
+    }
+
+    base_uint& operator|=(const base_uint& b)
+    {
+        for (int i = 0; i < WIDTH; i++)
+            pn[i] |= b.pn[i];
+        return *this;
+    }
+
+    base_uint& operator^=(uint64 b)
+    {
+        pn[0] ^= (unsigned int)b;
+        pn[1] ^= (unsigned int)(b >> 32);
+        return *this;
+    }
+
+    base_uint& operator|=(uint64 b)
+    {
+        pn[0] |= (unsigned int)b;
+        pn[1] |= (unsigned int)(b >> 32);
+        return *this;
+    }
+
+    base_uint& operator<<=(unsigned int shift)
+    {
+        base_uint a(*this);
+        for (int i = 0; i < WIDTH; i++)
+            pn[i] = 0;
+        int k = shift / 32;
+        shift = shift % 32;
+        for (int i = 0; i < WIDTH; i++)
+        {
+            if (i+k+1 < WIDTH && shift != 0)
+                pn[i+k+1] |= (a.pn[i] >> (32-shift));
+            if (i+k < WIDTH)
+                pn[i+k] |= (a.pn[i] << shift);
+        }
+        return *this;
+    }
+
+    base_uint& operator>>=(unsigned int shift)
+    {
+        base_uint a(*this);
+        for (int i = 0; i < WIDTH; i++)
+            pn[i] = 0;
+        int k = shift / 32;
+        shift = shift % 32;
+        for (int i = 0; i < WIDTH; i++)
+        {
+            if (i-k-1 >= 0 && shift != 0)
+                pn[i-k-1] |= (a.pn[i] << (32-shift));
+            if (i-k >= 0)
+                pn[i-k] |= (a.pn[i] >> shift);
+        }
+        return *this;
+    }
+
+    base_uint& operator+=(const base_uint& b)
+    {
+        uint64 carry = 0;
+        for (int i = 0; i < WIDTH; i++)
+        {
+            uint64 n = carry + pn[i] + b.pn[i];
+            pn[i] = n & 0xffffffff;
+            carry = n >> 32;
+        }
+        return *this;
+    }
+
+    base_uint& operator-=(const base_uint& b)
+    {
+        *this += -b;
+        return *this;
+    }
+
+    base_uint& operator+=(uint64 b64)
+    {
+        base_uint b;
+        b = b64;
+        *this += b;
+        return *this;
+    }
+
+    base_uint& operator-=(uint64 b64)
+    {
+        base_uint b;
+        b = b64;
+        *this += -b;
+        return *this;
+    }
+
+
+    base_uint& operator++()
+    {
+        // prefix operator
+        int i = 0;
+        while (++pn[i] == 0 && i < WIDTH-1)
+            i++;
+        return *this;
+    }
+
+    const base_uint operator++(int)
+    {
+        // postfix operator
+        const base_uint ret = *this;
+        ++(*this);
+        return ret;
+    }
+
+    base_uint& operator--()
+    {
+        // prefix operator
+        int i = 0;
+        while (--pn[i] == -1 && i < WIDTH-1)
+            i++;
+        return *this;
+    }
+
+    const base_uint operator--(int)
+    {
+        // postfix operator
+        const base_uint ret = *this;
+        --(*this);
+        return ret;
+    }
+
+
+    friend inline bool operator<(const base_uint& a, const base_uint& b)
+    {
+        for (int i = base_uint::WIDTH-1; i >= 0; i--)
+        {
+            if (a.pn[i] < b.pn[i])
+                return true;
+            else if (a.pn[i] > b.pn[i])
+                return false;
+        }
+        return false;
+    }
+
+    friend inline bool operator<=(const base_uint& a, const base_uint& b)
+    {
+        for (int i = base_uint::WIDTH-1; i >= 0; i--)
+        {
+            if (a.pn[i] < b.pn[i])
+                return true;
+            else if (a.pn[i] > b.pn[i])
+                return false;
+        }
+        return true;
+    }
+
+    friend inline bool operator>(const base_uint& a, const base_uint& b)
+    {
+        for (int i = base_uint::WIDTH-1; i >= 0; i--)
+        {
+            if (a.pn[i] > b.pn[i])
+                return true;
+            else if (a.pn[i] < b.pn[i])
+                return false;
+        }
+        return false;
+    }
+
+    friend inline bool operator>=(const base_uint& a, const base_uint& b)
+    {
+        for (int i = base_uint::WIDTH-1; i >= 0; i--)
+        {
+            if (a.pn[i] > b.pn[i])
+                return true;
+            else if (a.pn[i] < b.pn[i])
+                return false;
+        }
+        return true;
+    }
+
+    friend inline bool operator==(const base_uint& a, const base_uint& b)
+    {
+        for (int i = 0; i < base_uint::WIDTH; i++)
+            if (a.pn[i] != b.pn[i])
+                return false;
+        return true;
+    }
+
+    friend inline bool operator==(const base_uint& a, uint64 b)
+    {
+        if (a.pn[0] != (unsigned int)b)
+            return false;
+        if (a.pn[1] != (unsigned int)(b >> 32))
+            return false;
+        for (int i = 2; i < base_uint::WIDTH; i++)
+            if (a.pn[i] != 0)
+                return false;
+        return true;
+    }
+
+    friend inline bool operator!=(const base_uint& a, const base_uint& b)
+    {
+        return (!(a == b));
+    }
+
+    friend inline bool operator!=(const base_uint& a, uint64 b)
+    {
+        return (!(a == b));
+    }
+
+
+
+    std::string GetHex() const
+    {
+        char psz[sizeof(pn)*2 + 1];
+        for (unsigned int i = 0; i < sizeof(pn); i++)
+            sprintf(psz + i*2, "%02x", ((unsigned char*)pn)[sizeof(pn) - i - 1]);
+        return std::string(psz, psz + sizeof(pn)*2);
+    }
+
+    void SetHex(const char* psz)
+    {
+        for (int i = 0; i < WIDTH; i++)
+            pn[i] = 0;
+
+        // skip leading spaces
+        while (isspace(*psz))
+            psz++;
+
+        // skip 0x
+        if (psz[0] == '0' && tolower(psz[1]) == 'x')
+            psz += 2;
+
+        // hex string to uint
+        static const unsigned char phexdigit[256] = { 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, 0,1,2,3,4,5,6,7,8,9,0,0,0,0,0,0, 0,0xa,0xb,0xc,0xd,0xe,0xf,0,0,0,0,0,0,0,0,0, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, 0,0xa,0xb,0xc,0xd,0xe,0xf,0,0,0,0,0,0,0,0,0 };
+        const char* pbegin = psz;
+        while (phexdigit[(unsigned char)*psz] || *psz == '0')
+            psz++;
+        psz--;
+        unsigned char* p1 = (unsigned char*)pn;
+        unsigned char* pend = p1 + WIDTH * 4;
+        while (psz >= pbegin && p1 < pend)
+        {
+            *p1 = phexdigit[(unsigned char)*psz--];
+            if (psz >= pbegin)
+            {
+                *p1 |= (phexdigit[(unsigned char)*psz--] << 4);
+                p1++;
+            }
+        }
+    }
+
+    void SetHex(const std::string& str)
+    {
+        SetHex(str.c_str());
+    }
+
+    std::string ToString() const
+    {
+        return (GetHex());
+    }
+
+    unsigned char* begin()
+    {
+        return (unsigned char*)&pn[0];
+    }
+
+    unsigned char* end()
+    {
+        return (unsigned char*)&pn[WIDTH];
+    }
+
+    const unsigned char* begin() const
+    {
+        return (unsigned char*)&pn[0];
+    }
+
+    const unsigned char* end() const
+    {
+        return (unsigned char*)&pn[WIDTH];
+    }
+
+    unsigned int size() const
+    {
+        return sizeof(pn);
+    }
+
+    uint64 Get64(int n=0) const
+    {
+        return pn[2*n] | (uint64)pn[2*n+1] << 32;
+    }
+
+//    unsigned int GetSerializeSize(int nType=0, int nVersion=PROTOCOL_VERSION) const
+    unsigned int GetSerializeSize(int nType, int nVersion) const
+    {
+        return sizeof(pn);
+    }
+
+    template<typename Stream>
+//    void Serialize(Stream& s, int nType=0, int nVersion=PROTOCOL_VERSION) const
+    void Serialize(Stream& s, int nType, int nVersion) const
+    {
+        s.write((char*)pn, sizeof(pn));
+    }
+
+    template<typename Stream>
+//    void Unserialize(Stream& s, int nType=0, int nVersion=PROTOCOL_VERSION)
+    void Unserialize(Stream& s, int nType, int nVersion)
+    {
+        s.read((char*)pn, sizeof(pn));
+    }
+
+
+    friend class uint160;
+    friend class uint256;
+    friend inline int Testuint256AdHoc(std::vector<std::string> vArg);
+};
+
+typedef base_uint<160> base_uint160;
+typedef base_uint<256> base_uint256;
+
+
+
+//
+// uint160 and uint256 could be implemented as templates, but to keep
+// compile errors and debugging cleaner, they're copy and pasted.
+//
+
+
+
+//////////////////////////////////////////////////////////////////////////////
+//
+// uint160
+//
+
+/** 160-bit unsigned integer */
+class uint160 : public base_uint160
+{
+public:
+    typedef base_uint160 basetype;
+
+    uint160()
+    {
+        for (int i = 0; i < WIDTH; i++)
+            pn[i] = 0;
+    }
+
+    uint160(const basetype& b)
+    {
+        for (int i = 0; i < WIDTH; i++)
+            pn[i] = b.pn[i];
+    }
+
+    uint160& operator=(const basetype& b)
+    {
+        for (int i = 0; i < WIDTH; i++)
+            pn[i] = b.pn[i];
+        return *this;
+    }
+
+    uint160(uint64 b)
+    {
+        pn[0] = (unsigned int)b;
+        pn[1] = (unsigned int)(b >> 32);
+        for (int i = 2; i < WIDTH; i++)
+            pn[i] = 0;
+    }
+
+    uint160& operator=(uint64 b)
+    {
+        pn[0] = (unsigned int)b;
+        pn[1] = (unsigned int)(b >> 32);
+        for (int i = 2; i < WIDTH; i++)
+            pn[i] = 0;
+        return *this;
+    }
+
+    explicit uint160(const std::string& str)
+    {
+        SetHex(str);
+    }
+
+    explicit uint160(const std::vector<unsigned char>& vch)
+    {
+        if (vch.size() == sizeof(pn))
+            memcpy(pn, &vch[0], sizeof(pn));
+        else
+            *this = 0;
+    }
+};
+
+inline bool operator==(const uint160& a, uint64 b)                           { return (base_uint160)a == b; }
+inline bool operator!=(const uint160& a, uint64 b)                           { return (base_uint160)a != b; }
+inline const uint160 operator<<(const base_uint160& a, unsigned int shift)   { return uint160(a) <<= shift; }
+inline const uint160 operator>>(const base_uint160& a, unsigned int shift)   { return uint160(a) >>= shift; }
+inline const uint160 operator<<(const uint160& a, unsigned int shift)        { return uint160(a) <<= shift; }
+inline const uint160 operator>>(const uint160& a, unsigned int shift)        { return uint160(a) >>= shift; }
+
+inline const uint160 operator^(const base_uint160& a, const base_uint160& b) { return uint160(a) ^= b; }
+inline const uint160 operator&(const base_uint160& a, const base_uint160& b) { return uint160(a) &= b; }
+inline const uint160 operator|(const base_uint160& a, const base_uint160& b) { return uint160(a) |= b; }
+inline const uint160 operator+(const base_uint160& a, const base_uint160& b) { return uint160(a) += b; }
+inline const uint160 operator-(const base_uint160& a, const base_uint160& b) { return uint160(a) -= b; }
+
+inline bool operator<(const base_uint160& a, const uint160& b)          { return (base_uint160)a <  (base_uint160)b; }
+inline bool operator<=(const base_uint160& a, const uint160& b)         { return (base_uint160)a <= (base_uint160)b; }
+inline bool operator>(const base_uint160& a, const uint160& b)          { return (base_uint160)a >  (base_uint160)b; }
+inline bool operator>=(const base_uint160& a, const uint160& b)         { return (base_uint160)a >= (base_uint160)b; }
+inline bool operator==(const base_uint160& a, const uint160& b)         { return (base_uint160)a == (base_uint160)b; }
+inline bool operator!=(const base_uint160& a, const uint160& b)         { return (base_uint160)a != (base_uint160)b; }
+inline const uint160 operator^(const base_uint160& a, const uint160& b) { return (base_uint160)a ^  (base_uint160)b; }
+inline const uint160 operator&(const base_uint160& a, const uint160& b) { return (base_uint160)a &  (base_uint160)b; }
+inline const uint160 operator|(const base_uint160& a, const uint160& b) { return (base_uint160)a |  (base_uint160)b; }
+inline const uint160 operator+(const base_uint160& a, const uint160& b) { return (base_uint160)a +  (base_uint160)b; }
+inline const uint160 operator-(const base_uint160& a, const uint160& b) { return (base_uint160)a -  (base_uint160)b; }
+
+inline bool operator<(const uint160& a, const base_uint160& b)          { return (base_uint160)a <  (base_uint160)b; }
+inline bool operator<=(const uint160& a, const base_uint160& b)         { return (base_uint160)a <= (base_uint160)b; }
+inline bool operator>(const uint160& a, const base_uint160& b)          { return (base_uint160)a >  (base_uint160)b; }
+inline bool operator>=(const uint160& a, const base_uint160& b)         { return (base_uint160)a >= (base_uint160)b; }
+inline bool operator==(const uint160& a, const base_uint160& b)         { return (base_uint160)a == (base_uint160)b; }
+inline bool operator!=(const uint160& a, const base_uint160& b)         { return (base_uint160)a != (base_uint160)b; }
+inline const uint160 operator^(const uint160& a, const base_uint160& b) { return (base_uint160)a ^  (base_uint160)b; }
+inline const uint160 operator&(const uint160& a, const base_uint160& b) { return (base_uint160)a &  (base_uint160)b; }
+inline const uint160 operator|(const uint160& a, const base_uint160& b) { return (base_uint160)a |  (base_uint160)b; }
+inline const uint160 operator+(const uint160& a, const base_uint160& b) { return (base_uint160)a +  (base_uint160)b; }
+inline const uint160 operator-(const uint160& a, const base_uint160& b) { return (base_uint160)a -  (base_uint160)b; }
+
+inline bool operator<(const uint160& a, const uint160& b)               { return (base_uint160)a <  (base_uint160)b; }
+inline bool operator<=(const uint160& a, const uint160& b)              { return (base_uint160)a <= (base_uint160)b; }
+inline bool operator>(const uint160& a, const uint160& b)               { return (base_uint160)a >  (base_uint160)b; }
+inline bool operator>=(const uint160& a, const uint160& b)              { return (base_uint160)a >= (base_uint160)b; }
+inline bool operator==(const uint160& a, const uint160& b)              { return (base_uint160)a == (base_uint160)b; }
+inline bool operator!=(const uint160& a, const uint160& b)              { return (base_uint160)a != (base_uint160)b; }
+inline const uint160 operator^(const uint160& a, const uint160& b)      { return (base_uint160)a ^  (base_uint160)b; }
+inline const uint160 operator&(const uint160& a, const uint160& b)      { return (base_uint160)a &  (base_uint160)b; }
+inline const uint160 operator|(const uint160& a, const uint160& b)      { return (base_uint160)a |  (base_uint160)b; }
+inline const uint160 operator+(const uint160& a, const uint160& b)      { return (base_uint160)a +  (base_uint160)b; }
+inline const uint160 operator-(const uint160& a, const uint160& b)      { return (base_uint160)a -  (base_uint160)b; }
+
+
+
+
+
+
+//////////////////////////////////////////////////////////////////////////////
+//
+// uint256
+//
+
+/** 256-bit unsigned integer */
+class uint256 : public base_uint256
+{
+public:
+    typedef base_uint256 basetype;
+
+    uint256()
+    {
+        for (int i = 0; i < WIDTH; i++)
+            pn[i] = 0;
+    }
+
+    uint256(const basetype& b)
+    {
+        for (int i = 0; i < WIDTH; i++)
+            pn[i] = b.pn[i];
+    }
+
+    uint256& operator=(const basetype& b)
+    {
+        for (int i = 0; i < WIDTH; i++)
+            pn[i] = b.pn[i];
+        return *this;
+    }
+
+    uint256(uint64 b)
+    {
+        pn[0] = (unsigned int)b;
+        pn[1] = (unsigned int)(b >> 32);
+        for (int i = 2; i < WIDTH; i++)
+            pn[i] = 0;
+    }
+
+    uint256& operator=(uint64 b)
+    {
+        pn[0] = (unsigned int)b;
+        pn[1] = (unsigned int)(b >> 32);
+        for (int i = 2; i < WIDTH; i++)
+            pn[i] = 0;
+        return *this;
+    }
+
+    explicit uint256(const std::string& str)
+    {
+        SetHex(str);
+    }
+
+    explicit uint256(const std::vector<unsigned char>& vch)
+    {
+        if (vch.size() == sizeof(pn))
+            memcpy(pn, &vch[0], sizeof(pn));
+        else
+            *this = 0;
+    }
+};
+
+inline bool operator==(const uint256& a, uint64 b)                           { return (base_uint256)a == b; }
+inline bool operator!=(const uint256& a, uint64 b)                           { return (base_uint256)a != b; }
+inline const uint256 operator<<(const base_uint256& a, unsigned int shift)   { return uint256(a) <<= shift; }
+inline const uint256 operator>>(const base_uint256& a, unsigned int shift)   { return uint256(a) >>= shift; }
+inline const uint256 operator<<(const uint256& a, unsigned int shift)        { return uint256(a) <<= shift; }
+inline const uint256 operator>>(const uint256& a, unsigned int shift)        { return uint256(a) >>= shift; }
+
+inline const uint256 operator^(const base_uint256& a, const base_uint256& b) { return uint256(a) ^= b; }
+inline const uint256 operator&(const base_uint256& a, const base_uint256& b) { return uint256(a) &= b; }
+inline const uint256 operator|(const base_uint256& a, const base_uint256& b) { return uint256(a) |= b; }
+inline const uint256 operator+(const base_uint256& a, const base_uint256& b) { return uint256(a) += b; }
+inline const uint256 operator-(const base_uint256& a, const base_uint256& b) { return uint256(a) -= b; }
+
+inline bool operator<(const base_uint256& a, const uint256& b)          { return (base_uint256)a <  (base_uint256)b; }
+inline bool operator<=(const base_uint256& a, const uint256& b)         { return (base_uint256)a <= (base_uint256)b; }
+inline bool operator>(const base_uint256& a, const uint256& b)          { return (base_uint256)a >  (base_uint256)b; }
+inline bool operator>=(const base_uint256& a, const uint256& b)         { return (base_uint256)a >= (base_uint256)b; }
+inline bool operator==(const base_uint256& a, const uint256& b)         { return (base_uint256)a == (base_uint256)b; }
+inline bool operator!=(const base_uint256& a, const uint256& b)         { return (base_uint256)a != (base_uint256)b; }
+inline const uint256 operator^(const base_uint256& a, const uint256& b) { return (base_uint256)a ^  (base_uint256)b; }
+inline const uint256 operator&(const base_uint256& a, const uint256& b) { return (base_uint256)a &  (base_uint256)b; }
+inline const uint256 operator|(const base_uint256& a, const uint256& b) { return (base_uint256)a |  (base_uint256)b; }
+inline const uint256 operator+(const base_uint256& a, const uint256& b) { return (base_uint256)a +  (base_uint256)b; }
+inline const uint256 operator-(const base_uint256& a, const uint256& b) { return (base_uint256)a -  (base_uint256)b; }
+
+inline bool operator<(const uint256& a, const base_uint256& b)          { return (base_uint256)a <  (base_uint256)b; }
+inline bool operator<=(const uint256& a, const base_uint256& b)         { return (base_uint256)a <= (base_uint256)b; }
+inline bool operator>(const uint256& a, const base_uint256& b)          { return (base_uint256)a >  (base_uint256)b; }
+inline bool operator>=(const uint256& a, const base_uint256& b)         { return (base_uint256)a >= (base_uint256)b; }
+inline bool operator==(const uint256& a, const base_uint256& b)         { return (base_uint256)a == (base_uint256)b; }
+inline bool operator!=(const uint256& a, const base_uint256& b)         { return (base_uint256)a != (base_uint256)b; }
+inline const uint256 operator^(const uint256& a, const base_uint256& b) { return (base_uint256)a ^  (base_uint256)b; }
+inline const uint256 operator&(const uint256& a, const base_uint256& b) { return (base_uint256)a &  (base_uint256)b; }
+inline const uint256 operator|(const uint256& a, const base_uint256& b) { return (base_uint256)a |  (base_uint256)b; }
+inline const uint256 operator+(const uint256& a, const base_uint256& b) { return (base_uint256)a +  (base_uint256)b; }
+inline const uint256 operator-(const uint256& a, const base_uint256& b) { return (base_uint256)a -  (base_uint256)b; }
+
+inline bool operator<(const uint256& a, const uint256& b)               { return (base_uint256)a <  (base_uint256)b; }
+inline bool operator<=(const uint256& a, const uint256& b)              { return (base_uint256)a <= (base_uint256)b; }
+inline bool operator>(const uint256& a, const uint256& b)               { return (base_uint256)a >  (base_uint256)b; }
+inline bool operator>=(const uint256& a, const uint256& b)              { return (base_uint256)a >= (base_uint256)b; }
+inline bool operator==(const uint256& a, const uint256& b)              { return (base_uint256)a == (base_uint256)b; }
+inline bool operator!=(const uint256& a, const uint256& b)              { return (base_uint256)a != (base_uint256)b; }
+inline const uint256 operator^(const uint256& a, const uint256& b)      { return (base_uint256)a ^  (base_uint256)b; }
+inline const uint256 operator&(const uint256& a, const uint256& b)      { return (base_uint256)a &  (base_uint256)b; }
+inline const uint256 operator|(const uint256& a, const uint256& b)      { return (base_uint256)a |  (base_uint256)b; }
+inline const uint256 operator+(const uint256& a, const uint256& b)      { return (base_uint256)a +  (base_uint256)b; }
+inline const uint256 operator-(const uint256& a, const uint256& b)      { return (base_uint256)a -  (base_uint256)b; }
+
+
+
+
+
+
+
+
+
+
+#ifdef TEST_UINT256
+
+inline int Testuint256AdHoc(std::vector<std::string> vArg)
+{
+    uint256 g(0);
+
+
+    printf("%s\n", g.ToString().c_str());
+    g--;  printf("g--\n");
+    printf("%s\n", g.ToString().c_str());
+    g--;  printf("g--\n");
+    printf("%s\n", g.ToString().c_str());
+    g++;  printf("g++\n");
+    printf("%s\n", g.ToString().c_str());
+    g++;  printf("g++\n");
+    printf("%s\n", g.ToString().c_str());
+    g++;  printf("g++\n");
+    printf("%s\n", g.ToString().c_str());
+    g++;  printf("g++\n");
+    printf("%s\n", g.ToString().c_str());
+
+
+
+    uint256 a(7);
+    printf("a=7\n");
+    printf("%s\n", a.ToString().c_str());
+
+    uint256 b;
+    printf("b undefined\n");
+    printf("%s\n", b.ToString().c_str());
+    int c = 3;
+
+    a = c;
+    a.pn[3] = 15;
+    printf("%s\n", a.ToString().c_str());
+    uint256 k(c);
+
+    a = 5;
+    a.pn[3] = 15;
+    printf("%s\n", a.ToString().c_str());
+    b = 1;
+    b <<= 52;
+
+    a |= b;
+
+    a ^= 0x500;
+
+    printf("a %s\n", a.ToString().c_str());
+
+    a = a | b | (uint256)0x1000;
+
+
+    printf("a %s\n", a.ToString().c_str());
+    printf("b %s\n", b.ToString().c_str());
+
+    a = 0xfffffffe;
+    a.pn[4] = 9;
+
+    printf("%s\n", a.ToString().c_str());
+    a++;
+    printf("%s\n", a.ToString().c_str());
+    a++;
+    printf("%s\n", a.ToString().c_str());
+    a++;
+    printf("%s\n", a.ToString().c_str());
+    a++;
+    printf("%s\n", a.ToString().c_str());
+
+    a--;
+    printf("%s\n", a.ToString().c_str());
+    a--;
+    printf("%s\n", a.ToString().c_str());
+    a--;
+    printf("%s\n", a.ToString().c_str());
+    uint256 d = a--;
+    printf("%s\n", d.ToString().c_str());
+    printf("%s\n", a.ToString().c_str());
+    a--;
+    printf("%s\n", a.ToString().c_str());
+    a--;
+    printf("%s\n", a.ToString().c_str());
+
+    d = a;
+
+    printf("%s\n", d.ToString().c_str());
+    for (int i = uint256::WIDTH-1; i >= 0; i--) printf("%08x", d.pn[i]); printf("\n");
+
+    uint256 neg = d;
+    neg = ~neg;
+    printf("%s\n", neg.ToString().c_str());
+
+
+    uint256 e = uint256("0xABCDEF123abcdef12345678909832180000011111111");
+    printf("\n");
+    printf("%s\n", e.ToString().c_str());
+
+
+    printf("\n");
+    uint256 x1 = uint256("0xABCDEF123abcdef12345678909832180000011111111");
+    uint256 x2;
+    printf("%s\n", x1.ToString().c_str());
+    for (int i = 0; i < 270; i += 4)
+    {
+        x2 = x1 << i;
+        printf("%s\n", x2.ToString().c_str());
+    }
+
+    printf("\n");
+    printf("%s\n", x1.ToString().c_str());
+    for (int i = 0; i < 270; i += 4)
+    {
+        x2 = x1;
+        x2 >>= i;
+        printf("%s\n", x2.ToString().c_str());
+    }
+
+
+    for (int i = 0; i < 100; i++)
+    {
+        uint256 k = (~uint256(0) >> i);
+        printf("%s\n", k.ToString().c_str());
+    }
+
+    for (int i = 0; i < 100; i++)
+    {
+        uint256 k = (~uint256(0) << i);
+        printf("%s\n", k.ToString().c_str());
+    }
+
+    return (0);
+}
+
+#endif
+
+#endif
diff --git a/util.c b/util.c
new file mode 100644
index 0000000000..3f392bf76e
--- /dev/null
+++ b/util.c
@@ -0,0 +1,1316 @@
+/*
+ * Copyright 2010 Jeff Garzik
+ * Copyright 2012-2014 pooler
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License as published by the Free
+ * Software Foundation; either version 2 of the License, or (at your option)
+ * any later version.  See COPYING for more details.
+ */
+
+#define _GNU_SOURCE
+#include "cpuminer-config.h"
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <ctype.h>
+#include <stdarg.h>
+#include <string.h>
+#include <stdbool.h>
+#include <inttypes.h>
+#include <unistd.h>
+#include <jansson.h>
+#include <curl/curl.h>
+#include <time.h>
+#if defined(WIN32)
+#include <winsock2.h>
+#include <mstcpip.h>
+#else
+#include <errno.h>
+#include <sys/socket.h>
+#include <netinet/in.h>
+#include <netinet/tcp.h>
+#endif
+#include "compat.h"
+#include "miner.h"
+#include "elist.h"
+
+struct data_buffer {
+	void		*buf;
+	size_t		len;
+};
+
+struct upload_buffer {
+	const void	*buf;
+	size_t		len;
+	size_t		pos;
+};
+
+struct header_info {
+	char		*lp_path;
+	char		*reason;
+	char		*stratum_url;
+};
+
+struct tq_ent {
+	void			*data;
+	struct list_head	q_node;
+};
+
+struct thread_q {
+	struct list_head	q;
+
+	bool frozen;
+
+	pthread_mutex_t		mutex;
+	pthread_cond_t		cond;
+};
+
+void applog(int prio, const char *fmt, ...)
+{
+	va_list ap;
+
+	va_start(ap, fmt);
+
+#ifdef HAVE_SYSLOG_H
+	if (use_syslog) {
+		va_list ap2;
+		char *buf;
+		int len;
+		
+		va_copy(ap2, ap);
+		len = vsnprintf(NULL, 0, fmt, ap2) + 1;
+		va_end(ap2);
+		buf = alloca(len);
+		if (vsnprintf(buf, len, fmt, ap) >= 0)
+			syslog(prio, "%s", buf);
+	}
+#else
+	if (0) {}
+#endif
+	else {
+		char *f;
+		int len;
+		time_t now;
+		struct tm tm, *tm_p;
+
+		time(&now);
+
+		pthread_mutex_lock(&applog_lock);
+		tm_p = localtime(&now);
+		memcpy(&tm, tm_p, sizeof(tm));
+		pthread_mutex_unlock(&applog_lock);
+
+		len = (int)(40 + strlen(fmt) + 2);
+		f = (char*)alloca(len);
+		sprintf(f, "[%d-%02d-%02d %02d:%02d:%02d] %s\n",
+			tm.tm_year + 1900,
+			tm.tm_mon + 1,
+			tm.tm_mday,
+			tm.tm_hour,
+			tm.tm_min,
+			tm.tm_sec,
+			fmt);
+		pthread_mutex_lock(&applog_lock);
+		vfprintf(stderr, f, ap);	/* atomic write to stderr */
+		fflush(stderr);
+		pthread_mutex_unlock(&applog_lock);
+	}
+	va_end(ap);
+}
+
+static void databuf_free(struct data_buffer *db)
+{
+	if (!db)
+		return;
+
+	free(db->buf);
+
+	memset(db, 0, sizeof(*db));
+}
+
+static size_t all_data_cb(const void *ptr, size_t size, size_t nmemb,
+			  void *user_data)
+{
+	struct data_buffer *db = (struct data_buffer *)user_data;
+	size_t len = size * nmemb;
+	size_t oldlen, newlen;
+	void *newmem;
+	static const unsigned char zero = 0;
+
+	oldlen = db->len;
+	newlen = oldlen + len;
+
+	newmem = realloc(db->buf, newlen + 1);
+	if (!newmem)
+		return 0;
+
+	db->buf = newmem;
+	db->len = newlen;
+	memcpy((char*)db->buf + oldlen, ptr, len);
+	memcpy((char*)db->buf + newlen, &zero, 1);	/* null terminate */
+
+	return len;
+}
+
+static size_t upload_data_cb(void *ptr, size_t size, size_t nmemb,
+			     void *user_data)
+{
+	struct upload_buffer *ub = (struct upload_buffer *)user_data;
+	unsigned int len = (unsigned int)(size * nmemb);
+
+	if (len > ub->len - ub->pos)
+		len = (unsigned int)(ub->len - ub->pos);
+
+	if (len) {
+		memcpy(ptr, (char*)ub->buf + ub->pos, len);
+		ub->pos += len;
+	}
+
+	return len;
+}
+
+#if LIBCURL_VERSION_NUM >= 0x071200
+static int seek_data_cb(void *user_data, curl_off_t offset, int origin)
+{
+	struct upload_buffer *ub = (struct upload_buffer *)user_data;
+	
+	switch (origin) {
+	case SEEK_SET:
+		ub->pos = (size_t)offset;
+		break;
+	case SEEK_CUR:
+		ub->pos += (size_t)offset;
+		break;
+	case SEEK_END:
+		ub->pos = ub->len + (size_t)offset;
+		break;
+	default:
+		return 1; /* CURL_SEEKFUNC_FAIL */
+	}
+
+	return 0; /* CURL_SEEKFUNC_OK */
+}
+#endif
+
+static size_t resp_hdr_cb(void *ptr, size_t size, size_t nmemb, void *user_data)
+{
+	struct header_info *hi = (struct header_info *)user_data;
+	size_t remlen, slen, ptrlen = size * nmemb;
+	char *rem, *val = NULL, *key = NULL;
+	void *tmp;
+
+	val = (char*)calloc(1, ptrlen);
+	key = (char*)calloc(1, ptrlen);
+	if (!key || !val)
+		goto out;
+
+	tmp = memchr(ptr, ':', ptrlen);
+	if (!tmp || (tmp == ptr))	/* skip empty keys / blanks */
+		goto out;
+	slen = (size_t)((char*)tmp - (char*)ptr);
+	if ((slen + 1) == ptrlen)	/* skip key w/ no value */
+		goto out;
+	memcpy(key, ptr, slen);		/* store & nul term key */
+	key[slen] = 0;
+
+	rem = (char*)ptr + slen + 1;		/* trim value's leading whitespace */
+	remlen = ptrlen - slen - 1;
+	while ((remlen > 0) && (isspace(*rem))) {
+		remlen--;
+		rem++;
+	}
+
+	memcpy(val, rem, remlen);	/* store value, trim trailing ws */
+	val[remlen] = 0;
+	while ((*val) && (isspace(val[strlen(val) - 1]))) {
+		val[strlen(val) - 1] = 0;
+	}
+	if (!*val)			/* skip blank value */
+		goto out;
+
+	if (!strcasecmp("X-Long-Polling", key)) {
+		hi->lp_path = val;	/* steal memory reference */
+		val = NULL;
+	}
+
+	if (!strcasecmp("X-Reject-Reason", key)) {
+		hi->reason = val;	/* steal memory reference */
+		val = NULL;
+	}
+
+	if (!strcasecmp("X-Stratum", key)) {
+		hi->stratum_url = val;	/* steal memory reference */
+		val = NULL;
+	}
+
+out:
+	free(key);
+	free(val);
+	return ptrlen;
+}
+
+#if LIBCURL_VERSION_NUM >= 0x070f06
+static int sockopt_keepalive_cb(void *userdata, curl_socket_t fd,
+	curlsocktype purpose)
+{
+	int keepalive = 1;
+	int tcp_keepcnt = 3;
+	int tcp_keepidle = 50;
+	int tcp_keepintvl = 50;
+#ifdef WIN32
+	DWORD outputBytes;
+#endif
+
+#ifndef WIN32	
+	if (unlikely(setsockopt(fd, SOL_SOCKET, SO_KEEPALIVE, &keepalive,
+		sizeof(keepalive))))
+		return 1;
+#ifdef __linux
+	if (unlikely(setsockopt(fd, SOL_TCP, TCP_KEEPCNT,
+		&tcp_keepcnt, sizeof(tcp_keepcnt))))
+		return 1;
+	if (unlikely(setsockopt(fd, SOL_TCP, TCP_KEEPIDLE,
+		&tcp_keepidle, sizeof(tcp_keepidle))))
+		return 1;
+	if (unlikely(setsockopt(fd, SOL_TCP, TCP_KEEPINTVL,
+		&tcp_keepintvl, sizeof(tcp_keepintvl))))
+		return 1;
+#endif /* __linux */
+#ifdef __APPLE_CC__
+	if (unlikely(setsockopt(fd, IPPROTO_TCP, TCP_KEEPALIVE,
+		&tcp_keepintvl, sizeof(tcp_keepintvl))))
+		return 1;
+#endif /* __APPLE_CC__ */
+#else /* WIN32 */
+	struct tcp_keepalive vals;
+	vals.onoff = 1;
+	vals.keepalivetime = tcp_keepidle * 1000;
+	vals.keepaliveinterval = tcp_keepintvl * 1000;	
+	if (unlikely(WSAIoctl(fd, SIO_KEEPALIVE_VALS, &vals, sizeof(vals),
+		NULL, 0, &outputBytes, NULL, NULL)))
+		return 1;
+#endif /* WIN32 */
+
+	return 0;
+}
+#endif
+
+json_t *json_rpc_call(CURL *curl, const char *url,
+		      const char *userpass, const char *rpc_req,
+		      bool longpoll_scan, bool longpoll, int *curl_err)
+{
+	json_t *val, *err_val, *res_val;
+	int rc;
+	struct data_buffer all_data = {0};
+	struct upload_buffer upload_data;
+	json_error_t err;
+	struct curl_slist *headers = NULL;
+	char len_hdr[64];
+	char curl_err_str[CURL_ERROR_SIZE];
+	long timeout = longpoll ? opt_timeout : 30;
+	struct header_info hi = {0};
+	bool lp_scanning = longpoll_scan && !have_longpoll;
+
+	/* it is assumed that 'curl' is freshly [re]initialized at this pt */
+
+	if (opt_protocol)
+		curl_easy_setopt(curl, CURLOPT_VERBOSE, 1);
+	curl_easy_setopt(curl, CURLOPT_URL, url);
+	if (opt_cert)
+		curl_easy_setopt(curl, CURLOPT_CAINFO, opt_cert);
+	curl_easy_setopt(curl, CURLOPT_ENCODING, "");
+	curl_easy_setopt(curl, CURLOPT_FAILONERROR, 1);
+	curl_easy_setopt(curl, CURLOPT_NOSIGNAL, 1);
+	curl_easy_setopt(curl, CURLOPT_TCP_NODELAY, 1);
+	curl_easy_setopt(curl, CURLOPT_WRITEFUNCTION, all_data_cb);
+	curl_easy_setopt(curl, CURLOPT_WRITEDATA, &all_data);
+	curl_easy_setopt(curl, CURLOPT_READFUNCTION, upload_data_cb);
+	curl_easy_setopt(curl, CURLOPT_READDATA, &upload_data);
+#if LIBCURL_VERSION_NUM >= 0x071200
+	curl_easy_setopt(curl, CURLOPT_SEEKFUNCTION, &seek_data_cb);
+	curl_easy_setopt(curl, CURLOPT_SEEKDATA, &upload_data);
+#endif
+	curl_easy_setopt(curl, CURLOPT_ERRORBUFFER, curl_err_str);
+	curl_easy_setopt(curl, CURLOPT_FOLLOWLOCATION, 1);
+	curl_easy_setopt(curl, CURLOPT_TIMEOUT, timeout);
+	curl_easy_setopt(curl, CURLOPT_HEADERFUNCTION, resp_hdr_cb);
+	curl_easy_setopt(curl, CURLOPT_HEADERDATA, &hi);
+	if (opt_proxy) {
+		curl_easy_setopt(curl, CURLOPT_PROXY, opt_proxy);
+		curl_easy_setopt(curl, CURLOPT_PROXYTYPE, opt_proxy_type);
+	}
+	if (userpass) {
+		curl_easy_setopt(curl, CURLOPT_USERPWD, userpass);
+		curl_easy_setopt(curl, CURLOPT_HTTPAUTH, CURLAUTH_BASIC);
+	}
+#if LIBCURL_VERSION_NUM >= 0x070f06
+	if (longpoll)
+		curl_easy_setopt(curl, CURLOPT_SOCKOPTFUNCTION, sockopt_keepalive_cb);
+#endif
+	curl_easy_setopt(curl, CURLOPT_POST, 1);
+
+	if (opt_protocol)
+		applog(LOG_DEBUG, "JSON protocol request:\n%s\n", rpc_req);
+
+	upload_data.buf = rpc_req;
+	upload_data.len = strlen(rpc_req);
+	upload_data.pos = 0;
+	sprintf(len_hdr, "Content-Length: %lu",
+		(unsigned long) upload_data.len);
+
+	headers = curl_slist_append(headers, "Content-Type: application/json");
+	headers = curl_slist_append(headers, len_hdr);
+	headers = curl_slist_append(headers, "User-Agent: " USER_AGENT);
+	headers = curl_slist_append(headers, "X-Mining-Extensions: midstate");
+	headers = curl_slist_append(headers, "Accept:"); /* disable Accept hdr*/
+	headers = curl_slist_append(headers, "Expect:"); /* disable Expect hdr*/
+
+	curl_easy_setopt(curl, CURLOPT_HTTPHEADER, headers);
+
+	rc = curl_easy_perform(curl);
+	if (curl_err != NULL)
+		*curl_err = rc;
+	if (rc) {
+		if (!(longpoll && rc == CURLE_OPERATION_TIMEDOUT))
+			applog(LOG_ERR, "HTTP request failed: %s", curl_err_str);
+		goto err_out;
+	}
+
+	/* If X-Stratum was found, activate Stratum */
+	if (want_stratum && hi.stratum_url &&
+	    !strncasecmp(hi.stratum_url, "stratum+tcp://", 14) &&
+	    !(opt_proxy && opt_proxy_type == CURLPROXY_HTTP)) {
+		have_stratum = true;
+		tq_push(thr_info[stratum_thr_id].q, hi.stratum_url);
+		hi.stratum_url = NULL;
+	}
+
+	/* If X-Long-Polling was found, activate long polling */
+	if (lp_scanning && hi.lp_path && !have_stratum) {
+		have_longpoll = true;
+		tq_push(thr_info[longpoll_thr_id].q, hi.lp_path);
+		hi.lp_path = NULL;
+	}
+
+	if (!all_data.buf) {
+		applog(LOG_ERR, "Empty data received in json_rpc_call.");
+		goto err_out;
+	}
+
+	val = JSON_LOADS((const char*)all_data.buf, &err);
+	if (!val) {
+		applog(LOG_ERR, "JSON decode failed(%d): %s", err.line, err.text);
+		goto err_out;
+	}
+
+	if (opt_protocol) {
+		char *s = json_dumps(val, JSON_INDENT(3));
+		applog(LOG_DEBUG, "JSON protocol response:\n%s", s);
+		free(s);
+	}
+
+	/* JSON-RPC valid response returns a non-null 'result',
+	 * and a null 'error'. */
+	res_val = json_object_get(val, "result");
+	err_val = json_object_get(val, "error");
+
+	if (!res_val || json_is_null(res_val) ||
+	    (err_val && !json_is_null(err_val))) {
+		char *s;
+
+		if (err_val)
+			s = json_dumps(err_val, JSON_INDENT(3));
+		else
+			s = strdup("(unknown reason)");
+
+		applog(LOG_ERR, "JSON-RPC call failed: %s", s);
+
+		free(s);
+
+		goto err_out;
+	}
+
+	if (hi.reason)
+		json_object_set_new(val, "reject-reason", json_string(hi.reason));
+
+	databuf_free(&all_data);
+	curl_slist_free_all(headers);
+	curl_easy_reset(curl);
+	return val;
+
+err_out:
+	free(hi.lp_path);
+	free(hi.reason);
+	free(hi.stratum_url);
+	databuf_free(&all_data);
+	curl_slist_free_all(headers);
+	curl_easy_reset(curl);
+	return NULL;
+}
+
+char *bin2hex(const unsigned char *p, size_t len)
+{
+	unsigned int i;
+	char *s = (char*)malloc((len * 2) + 1);
+	if (!s)
+		return NULL;
+
+	for (i = 0; i < len; i++)
+		sprintf(s + (i * 2), "%02x", (unsigned int) p[i]);
+
+	return s;
+}
+
+bool hex2bin(unsigned char *p, const char *hexstr, size_t len)
+{
+	char hex_byte[3];
+	char *ep;
+
+	hex_byte[2] = '\0';
+
+	while (*hexstr && len) {
+		if (!hexstr[1]) {
+			applog(LOG_ERR, "hex2bin str truncated");
+			return false;
+		}
+		hex_byte[0] = hexstr[0];
+		hex_byte[1] = hexstr[1];
+		*p = (unsigned char) strtol(hex_byte, &ep, 16);
+		if (*ep) {
+			applog(LOG_ERR, "hex2bin failed on '%s'", hex_byte);
+			return false;
+		}
+		p++;
+		hexstr += 2;
+		len--;
+	}
+
+	return (len == 0 && *hexstr == 0) ? true : false;
+}
+
+/* Subtract the `struct timeval' values X and Y,
+   storing the result in RESULT.
+   Return 1 if the difference is negative, otherwise 0.  */
+int timeval_subtract(struct timeval *result, struct timeval *x,
+	struct timeval *y)
+{
+	/* Perform the carry for the later subtraction by updating Y. */
+	if (x->tv_usec < y->tv_usec) {
+		int nsec = (y->tv_usec - x->tv_usec) / 1000000 + 1;
+		y->tv_usec -= 1000000 * nsec;
+		y->tv_sec += nsec;
+	}
+	if (x->tv_usec - y->tv_usec > 1000000) {
+		int nsec = (x->tv_usec - y->tv_usec) / 1000000;
+		y->tv_usec += 1000000 * nsec;
+		y->tv_sec -= nsec;
+	}
+
+	/* Compute the time remaining to wait.
+	 * `tv_usec' is certainly positive. */
+	result->tv_sec = x->tv_sec - y->tv_sec;
+	result->tv_usec = x->tv_usec - y->tv_usec;
+
+	/* Return 1 if result is negative. */
+	return x->tv_sec < y->tv_sec;
+}
+
+bool fulltest(const uint32_t *hash, const uint32_t *target)
+{
+	int i;
+	bool rc = true;
+	
+	for (i = 7; i >= 0; i--) {
+		if (hash[i] > target[i]) {
+			rc = false;
+			break;
+		}
+		if (hash[i] < target[i]) {
+			rc = true;
+			break;
+		}
+	}
+
+	if (opt_debug) {
+		uint32_t hash_be[8], target_be[8];
+		char *hash_str, *target_str;
+		
+		for (i = 0; i < 8; i++) {
+			be32enc(hash_be + i, hash[7 - i]);
+			be32enc(target_be + i, target[7 - i]);
+		}
+		hash_str = bin2hex((unsigned char *)hash_be, 32);
+		target_str = bin2hex((unsigned char *)target_be, 32);
+
+		applog(LOG_DEBUG, "DEBUG: %s\nHash:   %s\nTarget: %s",
+			rc ? "hash <= target"
+			   : "hash > target (false positive)",
+			hash_str,
+			target_str);
+
+		free(hash_str);
+		free(target_str);
+	}
+
+	return rc;
+}
+
+void diff_to_target(uint32_t *target, double diff)
+{
+	uint64_t m;
+	int k;
+	
+	for (k = 6; k > 0 && diff > 1.0; k--)
+		diff /= 4294967296.0;
+	m = (uint64_t)(4294901760.0 / diff);
+	if (m == 0 && k == 6)
+		memset(target, 0xff, 32);
+	else {
+		memset(target, 0, 32);
+		target[k] = (uint32_t)m;
+		target[k + 1] = (uint32_t)(m >> 32);
+	}
+}
+
+#ifdef WIN32
+#define socket_blocks() (WSAGetLastError() == WSAEWOULDBLOCK)
+#else
+#define socket_blocks() (errno == EAGAIN || errno == EWOULDBLOCK)
+#endif
+
+static bool send_line(curl_socket_t sock, char *s)
+{
+	ssize_t len, sent = 0;
+	
+	len = (ssize_t)strlen(s);
+	s[len++] = '\n';
+
+	while (len > 0) {
+		struct timeval timeout = {0, 0};
+		ssize_t n;
+		fd_set wd;
+
+		FD_ZERO(&wd);
+		FD_SET(sock, &wd);
+		if (select((int)sock + 1, NULL, &wd, NULL, &timeout) < 1)
+			return false;
+		n = send(sock, s + sent, len, 0);
+		if (n < 0) {
+			if (!socket_blocks())
+				return false;
+			n = 0;
+		}
+		sent += n;
+		len -= n;
+	}
+
+	return true;
+}
+
+bool stratum_send_line(struct stratum_ctx *sctx, char *s)
+{
+	bool ret = false;
+
+	if (opt_protocol)
+		applog(LOG_DEBUG, "> %s", s);
+
+	pthread_mutex_lock(&sctx->sock_lock);
+	ret = send_line(sctx->sock, s);
+	pthread_mutex_unlock(&sctx->sock_lock);
+
+	return ret;
+}
+
+static bool socket_full(curl_socket_t sock, int timeout)
+{
+	struct timeval tv;
+	fd_set rd;
+
+	FD_ZERO(&rd);
+	FD_SET(sock, &rd);
+	tv.tv_sec = timeout;
+	tv.tv_usec = 0;
+	if (select((int)sock + 1, &rd, NULL, NULL, &tv) > 0)
+		return true;
+	return false;
+}
+
+bool stratum_socket_full(struct stratum_ctx *sctx, int timeout)
+{
+	return strlen(sctx->sockbuf) || socket_full(sctx->sock, timeout);
+}
+
+#define RBUFSIZE 2048
+#define RECVSIZE (RBUFSIZE - 4)
+
+static void stratum_buffer_append(struct stratum_ctx *sctx, const char *s)
+{
+	size_t old, snew;
+
+	old = strlen(sctx->sockbuf);
+	snew = old + strlen(s) + 1;
+	if (snew >= sctx->sockbuf_size) {
+		sctx->sockbuf_size = snew + (RBUFSIZE - (snew % RBUFSIZE));
+		sctx->sockbuf = (char*)realloc(sctx->sockbuf, sctx->sockbuf_size);
+	}
+	strcpy(sctx->sockbuf + old, s);
+}
+
+char *stratum_recv_line(struct stratum_ctx *sctx)
+{
+	ssize_t len, buflen;
+	char *tok, *sret = NULL;
+
+	if (!strstr(sctx->sockbuf, "\n")) {
+		bool ret = true;
+		time_t rstart;
+
+		time(&rstart);
+		if (!socket_full(sctx->sock, 60)) {
+			applog(LOG_ERR, "stratum_recv_line timed out");
+			goto out;
+		}
+		do {
+			char s[RBUFSIZE];
+			ssize_t n;
+
+			memset(s, 0, RBUFSIZE);
+			n = recv(sctx->sock, s, RECVSIZE, 0);
+			if (!n) {
+				ret = false;
+				break;
+			}
+			if (n < 0) {
+				if (!socket_blocks() || !socket_full(sctx->sock, 1)) {
+					ret = false;
+					break;
+				}
+			} else
+				stratum_buffer_append(sctx, s);
+		} while (time(NULL) - rstart < 60 && !strstr(sctx->sockbuf, "\n"));
+
+		if (!ret) {
+			applog(LOG_ERR, "stratum_recv_line failed");
+			goto out;
+		}
+	}
+
+	buflen = (ssize_t)strlen(sctx->sockbuf);
+	tok = strtok(sctx->sockbuf, "\n");
+	if (!tok) {
+		applog(LOG_ERR, "stratum_recv_line failed to parse a newline-terminated string");
+		goto out;
+	}
+	sret = strdup(tok);
+	len = (ssize_t)strlen(sret);
+
+	if (buflen > len + 1)
+		memmove(sctx->sockbuf, sctx->sockbuf + len + 1, buflen - len + 1);
+	else
+		sctx->sockbuf[0] = '\0';
+
+out:
+	if (sret && opt_protocol)
+		applog(LOG_DEBUG, "< %s", sret);
+	return sret;
+}
+
+#if LIBCURL_VERSION_NUM >= 0x071101
+static curl_socket_t opensocket_grab_cb(void *clientp, curlsocktype purpose,
+	struct curl_sockaddr *addr)
+{
+	curl_socket_t *sock = (curl_socket_t *)clientp;
+	*sock = socket(addr->family, addr->socktype, addr->protocol);
+	return *sock;
+}
+#endif
+
+bool stratum_connect(struct stratum_ctx *sctx, const char *url)
+{
+	CURL *curl;
+	int rc;
+
+	pthread_mutex_lock(&sctx->sock_lock);
+	if (sctx->curl)
+		curl_easy_cleanup(sctx->curl);
+	sctx->curl = curl_easy_init();
+	if (!sctx->curl) {
+		applog(LOG_ERR, "CURL initialization failed");
+		pthread_mutex_unlock(&sctx->sock_lock);
+		return false;
+	}
+	curl = sctx->curl;
+	if (!sctx->sockbuf) {
+		sctx->sockbuf = (char*)calloc(RBUFSIZE, 1);
+		sctx->sockbuf_size = RBUFSIZE;
+	}
+	sctx->sockbuf[0] = '\0';
+	pthread_mutex_unlock(&sctx->sock_lock);
+
+	if (url != sctx->url) {
+		free(sctx->url);
+		sctx->url = strdup(url);
+	}
+	free(sctx->curl_url);
+	sctx->curl_url = (char*)malloc(strlen(url));
+	sprintf(sctx->curl_url, "http%s", strstr(url, "://"));
+
+	if (opt_protocol)
+		curl_easy_setopt(curl, CURLOPT_VERBOSE, 1);
+	curl_easy_setopt(curl, CURLOPT_URL, sctx->curl_url);
+	curl_easy_setopt(curl, CURLOPT_FRESH_CONNECT, 1);
+	curl_easy_setopt(curl, CURLOPT_CONNECTTIMEOUT, 30);
+	curl_easy_setopt(curl, CURLOPT_ERRORBUFFER, sctx->curl_err_str);
+	curl_easy_setopt(curl, CURLOPT_NOSIGNAL, 1);
+	curl_easy_setopt(curl, CURLOPT_TCP_NODELAY, 1);
+	if (opt_proxy && opt_proxy_type != CURLPROXY_HTTP) {
+		curl_easy_setopt(curl, CURLOPT_PROXY, opt_proxy);
+		curl_easy_setopt(curl, CURLOPT_PROXYTYPE, opt_proxy_type);
+	} else if (getenv("http_proxy")) {
+		if (getenv("all_proxy"))
+			curl_easy_setopt(curl, CURLOPT_PROXY, getenv("all_proxy"));
+		else if (getenv("ALL_PROXY"))
+			curl_easy_setopt(curl, CURLOPT_PROXY, getenv("ALL_PROXY"));
+		else
+			curl_easy_setopt(curl, CURLOPT_PROXY, "");
+	}
+#if LIBCURL_VERSION_NUM >= 0x070f06
+	curl_easy_setopt(curl, CURLOPT_SOCKOPTFUNCTION, sockopt_keepalive_cb);
+#endif
+#if LIBCURL_VERSION_NUM >= 0x071101
+	curl_easy_setopt(curl, CURLOPT_OPENSOCKETFUNCTION, opensocket_grab_cb);
+	curl_easy_setopt(curl, CURLOPT_OPENSOCKETDATA, &sctx->sock);
+#endif
+	curl_easy_setopt(curl, CURLOPT_CONNECT_ONLY, 1);
+
+	rc = curl_easy_perform(curl);
+	if (rc) {
+		applog(LOG_ERR, "Stratum connection failed: %s", sctx->curl_err_str);
+		curl_easy_cleanup(curl);
+		sctx->curl = NULL;
+		return false;
+	}
+
+#if LIBCURL_VERSION_NUM < 0x071101
+	/* CURLINFO_LASTSOCKET is broken on Win64; only use it as a last resort */
+	curl_easy_getinfo(curl, CURLINFO_LASTSOCKET, (long *)&sctx->sock);
+#endif
+
+	return true;
+}
+
+void stratum_disconnect(struct stratum_ctx *sctx)
+{
+	pthread_mutex_lock(&sctx->sock_lock);
+	if (sctx->curl) {
+		curl_easy_cleanup(sctx->curl);
+		sctx->curl = NULL;
+		sctx->sockbuf[0] = '\0';
+	}
+	pthread_mutex_unlock(&sctx->sock_lock);
+}
+
+static const char *get_stratum_session_id(json_t *val)
+{
+	json_t *arr_val;
+	int i, n;
+
+	arr_val = json_array_get(val, 0);
+	if (!arr_val || !json_is_array(arr_val))
+		return NULL;
+	n = json_array_size(arr_val);
+	for (i = 0; i < n; i++) {
+		const char *notify;
+		json_t *arr = json_array_get(arr_val, i);
+
+		if (!arr || !json_is_array(arr))
+			break;
+		notify = json_string_value(json_array_get(arr, 0));
+		if (!notify)
+			continue;
+		if (!strcasecmp(notify, "mining.notify"))
+			return json_string_value(json_array_get(arr, 1));
+	}
+	return NULL;
+}
+
+bool stratum_subscribe(struct stratum_ctx *sctx)
+{
+	char *s, *sret = NULL;
+	const char *sid, *xnonce1;
+	int xn2_size;
+	json_t *val = NULL, *res_val, *err_val;
+	json_error_t err;
+	bool ret = false, retry = false;
+
+start:
+	s = (char*)malloc(128 + (sctx->session_id ? strlen(sctx->session_id) : 0));
+	if (retry)
+		sprintf(s, "{\"id\": 1, \"method\": \"mining.subscribe\", \"params\": []}");
+	else if (sctx->session_id)
+		sprintf(s, "{\"id\": 1, \"method\": \"mining.subscribe\", \"params\": [\"" USER_AGENT "\", \"%s\"]}", sctx->session_id);
+	else
+		sprintf(s, "{\"id\": 1, \"method\": \"mining.subscribe\", \"params\": [\"" USER_AGENT "\"]}");
+
+	if (!stratum_send_line(sctx, s))
+		goto out;
+
+	if (!socket_full(sctx->sock, 30)) {
+		applog(LOG_ERR, "stratum_subscribe timed out");
+		goto out;
+	}
+
+	sret = stratum_recv_line(sctx);
+	if (!sret)
+		goto out;
+
+	val = JSON_LOADS(sret, &err);
+	free(sret);
+	if (!val) {
+		applog(LOG_ERR, "JSON decode failed(%d): %s", err.line, err.text);
+		goto out;
+	}
+
+	res_val = json_object_get(val, "result");
+	err_val = json_object_get(val, "error");
+
+	if (!res_val || json_is_null(res_val) ||
+	    (err_val && !json_is_null(err_val))) {
+		if (opt_debug || retry) {
+			free(s);
+			if (err_val)
+				s = json_dumps(err_val, JSON_INDENT(3));
+			else
+				s = strdup("(unknown reason)");
+			applog(LOG_ERR, "JSON-RPC call failed: %s", s);
+		}
+		goto out;
+	}
+
+	sid = get_stratum_session_id(res_val);
+	if (opt_debug && !sid)
+		applog(LOG_DEBUG, "Failed to get Stratum session id");
+	xnonce1 = json_string_value(json_array_get(res_val, 1));
+	if (!xnonce1) {
+		applog(LOG_ERR, "Failed to get extranonce1");
+		goto out;
+	}
+	xn2_size = json_integer_value(json_array_get(res_val, 2));
+	if (!xn2_size) {
+		applog(LOG_ERR, "Failed to get extranonce2_size");
+		goto out;
+	}
+
+	pthread_mutex_lock(&sctx->work_lock);
+	free(sctx->session_id);
+	free(sctx->xnonce1);
+	sctx->session_id = sid ? strdup(sid) : NULL;
+	sctx->xnonce1_size = strlen(xnonce1) / 2;
+	sctx->xnonce1 = (unsigned char*)malloc(sctx->xnonce1_size);
+	hex2bin(sctx->xnonce1, xnonce1, sctx->xnonce1_size);
+	sctx->xnonce2_size = xn2_size;
+	sctx->next_diff = 1.0;
+	pthread_mutex_unlock(&sctx->work_lock);
+
+	if (opt_debug && sid)
+		applog(LOG_DEBUG, "Stratum session id: %s", sctx->session_id);
+
+	ret = true;
+
+out:
+	free(s);
+	if (val)
+		json_decref(val);
+
+	if (!ret) {
+		if (sret && !retry) {
+			retry = true;
+			goto start;
+		}
+	}
+
+	return ret;
+}
+
+bool stratum_authorize(struct stratum_ctx *sctx, const char *user, const char *pass)
+{
+	json_t *val = NULL, *res_val, *err_val;
+	char *s, *sret;
+	json_error_t err;
+	bool ret = false;
+
+	s = (char*)malloc(80 + strlen(user) + strlen(pass));
+	sprintf(s, "{\"id\": 2, \"method\": \"mining.authorize\", \"params\": [\"%s\", \"%s\"]}",
+	        user, pass);
+
+	if (!stratum_send_line(sctx, s))
+		goto out;
+
+	while (1) {
+		sret = stratum_recv_line(sctx);
+		if (!sret)
+			goto out;
+		if (!stratum_handle_method(sctx, sret))
+			break;
+		free(sret);
+	}
+
+	val = JSON_LOADS(sret, &err);
+	free(sret);
+	if (!val) {
+		applog(LOG_ERR, "JSON decode failed(%d): %s", err.line, err.text);
+		goto out;
+	}
+
+	res_val = json_object_get(val, "result");
+	err_val = json_object_get(val, "error");
+
+	if (!res_val || json_is_false(res_val) ||
+	    (err_val && !json_is_null(err_val)))  {
+		applog(LOG_ERR, "Stratum authentication failed");
+		goto out;
+	}
+
+	ret = true;
+
+out:
+	free(s);
+	if (val)
+		json_decref(val);
+
+	return ret;
+}
+
+static bool stratum_notify(struct stratum_ctx *sctx, json_t *params)
+{
+	const char *job_id, *prevhash, *coinb1, *coinb2, *version, *nbits, *ntime, *nreward;
+	size_t coinb1_size, coinb2_size;
+	bool clean, ret = false;
+	int merkle_count, i;
+	json_t *merkle_arr;
+	unsigned char **merkle;
+
+	job_id = json_string_value(json_array_get(params, 0));
+	prevhash = json_string_value(json_array_get(params, 1));
+	coinb1 = json_string_value(json_array_get(params, 2));
+	coinb2 = json_string_value(json_array_get(params, 3));
+	merkle_arr = json_array_get(params, 4);
+	if (!merkle_arr || !json_is_array(merkle_arr))
+		goto out;
+	merkle_count = json_array_size(merkle_arr);
+	version = json_string_value(json_array_get(params, 5));
+	nbits = json_string_value(json_array_get(params, 6));
+	ntime = json_string_value(json_array_get(params, 7));
+	clean = json_is_true(json_array_get(params, 8));
+	nreward = json_string_value(json_array_get(params, 9));
+
+	if (!job_id || !prevhash || !coinb1 || !coinb2 || !version || !nbits || !ntime ||
+	    strlen(prevhash) != 64 || strlen(version) != 8 ||
+	    strlen(nbits) != 8 || strlen(ntime) != 8 || strlen(nreward) != 4) {
+		applog(LOG_ERR, "Stratum notify: invalid parameters");
+		goto out;
+	}
+	merkle = (unsigned char**)malloc(merkle_count * sizeof(char *));
+	for (i = 0; i < merkle_count; i++) {
+		const char *s = json_string_value(json_array_get(merkle_arr, i));
+		if (!s || strlen(s) != 64) {
+			while (i--)
+				free(merkle[i]);
+			free(merkle);
+			applog(LOG_ERR, "Stratum notify: invalid Merkle branch");
+			goto out;
+		}
+		merkle[i] = (unsigned char*)malloc(32);
+		hex2bin(merkle[i], s, 32);
+	}
+
+	pthread_mutex_lock(&sctx->work_lock);
+
+	coinb1_size = strlen(coinb1) / 2;
+	coinb2_size = strlen(coinb2) / 2;
+	sctx->job.coinbase_size = coinb1_size + sctx->xnonce1_size +
+	                          sctx->xnonce2_size + coinb2_size;
+	sctx->job.coinbase = (unsigned char*)realloc(sctx->job.coinbase, sctx->job.coinbase_size);
+	sctx->job.xnonce2 = sctx->job.coinbase + coinb1_size + sctx->xnonce1_size;
+	hex2bin(sctx->job.coinbase, coinb1, coinb1_size);
+	memcpy(sctx->job.coinbase + coinb1_size, sctx->xnonce1, sctx->xnonce1_size);
+	if (!sctx->job.job_id || strcmp(sctx->job.job_id, job_id))
+		memset(sctx->job.xnonce2, 0, sctx->xnonce2_size);
+	hex2bin(sctx->job.xnonce2 + sctx->xnonce2_size, coinb2, coinb2_size);
+
+	free(sctx->job.job_id);
+	sctx->job.job_id = strdup(job_id);
+	hex2bin(sctx->job.prevhash, prevhash, 32);
+
+	for (i = 0; i < sctx->job.merkle_count; i++)
+		free(sctx->job.merkle[i]);
+	free(sctx->job.merkle);
+	sctx->job.merkle = merkle;
+	sctx->job.merkle_count = merkle_count;
+
+	hex2bin(sctx->job.version, version, 4);
+	hex2bin(sctx->job.nbits, nbits, 4);
+	hex2bin(sctx->job.ntime, ntime, 4);
+	hex2bin(sctx->job.nreward, nreward, 2);
+	sctx->job.clean = clean;
+
+	sctx->job.diff = sctx->next_diff;
+
+	pthread_mutex_unlock(&sctx->work_lock);
+
+	ret = true;
+
+out:
+	return ret;
+}
+
+static bool stratum_set_difficulty(struct stratum_ctx *sctx, json_t *params)
+{
+	double diff;
+
+	diff = json_number_value(json_array_get(params, 0));
+	if (diff == 0)
+		return false;
+
+	pthread_mutex_lock(&sctx->work_lock);
+	sctx->next_diff = diff;
+	pthread_mutex_unlock(&sctx->work_lock);
+
+	if (opt_debug)
+		applog(LOG_DEBUG, "Stratum difficulty set to %g", diff);
+
+	return true;
+}
+
+static bool stratum_reconnect(struct stratum_ctx *sctx, json_t *params)
+{
+	json_t *port_val;
+	const char *host;
+	int port;
+
+	host = json_string_value(json_array_get(params, 0));
+	port_val = json_array_get(params, 1);
+	if (json_is_string(port_val))
+		port = atoi(json_string_value(port_val));
+	else
+		port = json_integer_value(port_val);
+	if (!host || !port)
+		return false;
+	
+	free(sctx->url);
+	sctx->url = (char*)malloc(32 + strlen(host));
+	sprintf(sctx->url, "stratum+tcp://%s:%d", host, port);
+
+	applog(LOG_NOTICE, "Server requested reconnection to %s", sctx->url);
+
+	stratum_disconnect(sctx);
+
+	return true;
+}
+
+static bool stratum_get_version(struct stratum_ctx *sctx, json_t *id)
+{
+	char *s;
+	json_t *val;
+	bool ret;
+	
+	if (!id || json_is_null(id))
+		return false;
+
+	val = json_object();
+	json_object_set(val, "id", id);
+	json_object_set_new(val, "error", json_null());
+	json_object_set_new(val, "result", json_string(USER_AGENT));
+	s = json_dumps(val, 0);
+	ret = stratum_send_line(sctx, s);
+	json_decref(val);
+	free(s);
+
+	return ret;
+}
+
+static bool stratum_show_message(struct stratum_ctx *sctx, json_t *id, json_t *params)
+{
+	char *s;
+	json_t *val;
+	bool ret;
+
+	val = json_array_get(params, 0);
+	if (val)
+		applog(LOG_NOTICE, "MESSAGE FROM SERVER: %s", json_string_value(val));
+	
+	if (!id || json_is_null(id))
+		return true;
+
+	val = json_object();
+	json_object_set(val, "id", id);
+	json_object_set_new(val, "error", json_null());
+	json_object_set_new(val, "result", json_true());
+	s = json_dumps(val, 0);
+	ret = stratum_send_line(sctx, s);
+	json_decref(val);
+	free(s);
+
+	return ret;
+}
+
+bool stratum_handle_method(struct stratum_ctx *sctx, const char *s)
+{
+	json_t *val, *id, *params;
+	json_error_t err;
+	const char *method;
+	bool ret = false;
+
+	val = JSON_LOADS(s, &err);
+	if (!val) {
+		applog(LOG_ERR, "JSON decode failed(%d): %s", err.line, err.text);
+		goto out;
+	}
+
+	method = json_string_value(json_object_get(val, "method"));
+	if (!method)
+		goto out;
+	id = json_object_get(val, "id");
+	params = json_object_get(val, "params");
+
+	if (!strcasecmp(method, "mining.notify")) {
+		ret = stratum_notify(sctx, params);
+		goto out;
+	}
+	if (!strcasecmp(method, "mining.set_difficulty")) {
+		ret = stratum_set_difficulty(sctx, params);
+		goto out;
+	}
+	if (!strcasecmp(method, "client.reconnect")) {
+		ret = stratum_reconnect(sctx, params);
+		goto out;
+	}
+	if (!strcasecmp(method, "client.get_version")) {
+		ret = stratum_get_version(sctx, id);
+		goto out;
+	}
+	if (!strcasecmp(method, "client.show_message")) {
+		ret = stratum_show_message(sctx, id, params);
+		goto out;
+	}
+
+out:
+	if (val)
+		json_decref(val);
+
+	return ret;
+}
+
+struct thread_q *tq_new(void)
+{
+	struct thread_q *tq;
+
+	tq = (struct thread_q *)calloc(1, sizeof(*tq));
+	if (!tq)
+		return NULL;
+
+	INIT_LIST_HEAD(&tq->q);
+	pthread_mutex_init(&tq->mutex, NULL);
+	pthread_cond_init(&tq->cond, NULL);
+
+	return tq;
+}
+
+void tq_free(struct thread_q *tq)
+{
+	struct tq_ent *ent, *iter;
+
+	if (!tq)
+		return;
+
+	list_for_each_entry_safe(ent, iter, &tq->q, q_node, struct tq_ent, struct tq_ent) {
+		list_del(&ent->q_node);
+		free(ent);
+	}
+
+	pthread_cond_destroy(&tq->cond);
+	pthread_mutex_destroy(&tq->mutex);
+
+	memset(tq, 0, sizeof(*tq));	/* poison */
+	free(tq);
+}
+
+static void tq_freezethaw(struct thread_q *tq, bool frozen)
+{
+	pthread_mutex_lock(&tq->mutex);
+
+	tq->frozen = frozen;
+
+	pthread_cond_signal(&tq->cond);
+	pthread_mutex_unlock(&tq->mutex);
+}
+
+void tq_freeze(struct thread_q *tq)
+{
+	tq_freezethaw(tq, true);
+}
+
+void tq_thaw(struct thread_q *tq)
+{
+	tq_freezethaw(tq, false);
+}
+
+bool tq_push(struct thread_q *tq, void *data)
+{
+	struct tq_ent *ent;
+	bool rc = true;
+
+	ent = (struct tq_ent *)calloc(1, sizeof(*ent));
+	if (!ent)
+		return false;
+
+	ent->data = data;
+	INIT_LIST_HEAD(&ent->q_node);
+
+	pthread_mutex_lock(&tq->mutex);
+
+	if (!tq->frozen) {
+		list_add_tail(&ent->q_node, &tq->q);
+	} else {
+		free(ent);
+		rc = false;
+	}
+
+	pthread_cond_signal(&tq->cond);
+	pthread_mutex_unlock(&tq->mutex);
+
+	return rc;
+}
+
+void *tq_pop(struct thread_q *tq, const struct timespec *abstime)
+{
+	struct tq_ent *ent;
+	void *rval = NULL;
+	int rc;
+
+	pthread_mutex_lock(&tq->mutex);
+
+	if (!list_empty(&tq->q))
+		goto pop;
+
+	if (abstime)
+		rc = pthread_cond_timedwait(&tq->cond, &tq->mutex, abstime);
+	else
+		rc = pthread_cond_wait(&tq->cond, &tq->mutex);
+	if (rc)
+		goto out;
+	if (list_empty(&tq->q))
+		goto out;
+
+pop:
+	ent = list_entry(tq->q.next, struct tq_ent, q_node);
+	rval = ent->data;
+
+	list_del(&ent->q_node);
+	free(ent);
+
+out:
+	pthread_mutex_unlock(&tq->mutex);
+	return rval;
+}