Merge branch 'migrate-to-maven'

AUTHORS

@@ -5,4 +5,6 @@ Mikio L. Braun
 Additional Programming and Contributions:
 
 Johannes Schaback
-Jan Saputra Müller
+Jan Saputra Müller (exponential matrix multiplication, decomposition)
+Matthias L. Jugel (packaging)
+Nicolas Oury (generalized eigenvectors)

BUILDING_ATLAS

@@ -42,27 +42,29 @@ git clone git://github.com/mikiobraun/jblas.git
 
 The general steps are as follows:
 
-* Compile lapack. Lapack uses blas for its low-level computations. ATLAS also
-  implements a subset of Lapack, but not all routines. But you can build ATLAS
-  to include the Lapack routines it does not implement, and this is what we're
-  going to do.
+* Compile lapack. Lapack uses blas for its low-level
+  computations. ATLAS also implements a subset of Lapack, but not all
+  routines. But you can build ATLAS to include the Lapack routines it
+  does not implement, and this is what we're going to do.
 
 * Compile ATLAS.
 
 * Compile the native part of jblas with "make".
 
-* Compile the java part of jblas with "ant".
+* Compile the java part of jblas with "ant" or maven.
 
-The ATLAS build process is a bit, well, unusual, coming with compiled (not scripted)
-configuration scripts. ATLAS has prebuilt configurations which result in much
-faster compilation times. If it doesn't find such a profile for your computer,
-it will perform a large number of timing experiments to find the best combination
-(hence the name "Automatically Tuned...").
+The ATLAS build process is a bit, well, unusual, coming with compiled
+(not scripted) configuration scripts. ATLAS has prebuilt
+configurations which result in much faster compilation times. If it
+doesn't find such a profile for your computer, it will perform a large
+number of timing experiments to find the best combination (hence the
+name "Automatically Tuned...").
 
-Finally, some older processors don't support the SSE3 command set. Some tweaking
-is necessary to compile ATLAS for SSE2, namely configuring as for SSE3, and
-then going into the file "Make.inc", and replacing SSE2 by SSE3 or removing -DATL_SSE3
-from the compiler directives.
+Finally, some older processors don't support the SSE3 command
+set. Some tweaking is necessary to compile ATLAS for SSE2, namely
+configuring as for SSE3, and then going into the file "Make.inc", and
+replacing SSE2 by SSE3 or removing -DATL_SSE3 from the compiler
+directives.
 
 For reference, here is a list of the main ATLAS options.
 
@@ -129,7 +131,8 @@ ATLAS Operating System Options: (-O)
 LAPACK
 ------
 
-Copy make.inc.LINUX from INSTALL directory, update options in the file like this:
+Copy make.inc.LINUX from INSTALL directory, update options in the file
+like this:
 
 FORTRAN  = gfortran 
 OPTS     = -O2 -fPIC
@@ -218,15 +221,17 @@ It's important to set the 32bit flag. Otherwise, it seems to construct
 5. Windows XP (32bit)
 =====================
 
-I'm describing here using cygwin. For some reason, the gcc4 version in cygwin
-does not support compiling to mingw, but using the cygwin version leads
-to some address spaces clashes with Java (at least, the last time I checked),
-leading to a crash as soon as the cygwin compiled code is loaded.
+I'm describing here using cygwin. For some reason, the gcc4 version in
+cygwin does not support compiling to mingw, but using the cygwin
+version leads to some address spaces clashes with Java (at least, the
+last time I checked), leading to a crash as soon as the cygwin
+compiled code is loaded.
 
-What all of this means is that you should use the old gcc3 "legacy" compilers.
+What all of this means is that you should use the old gcc3 "legacy"
+compilers.
 
-Also watch out for directory names with spaces in them, ATLAS does not like
-them.
+Also watch out for directory names with spaces in them, ATLAS does not
+like them.
 
 LAPACK
 ------
@@ -245,8 +250,31 @@ compiler explicitly to "g77" (and the bit numbers to 32):
 64-bit
 ------
 
-Unfortunately, there is no support for 64bit under Windows right now,
-mainly because there is no 64bit mingw compiler in cygwin. There
-exists the mingw-w64 bit project
-(http://sourceforge.net/projects/mingw-w64/), but I didn't managed to
-compile ATLAS using those.
+Unfortunately, it is currently not possible to compile ATLAS for 64bit
+under windows, mainly because cygwin is 32bit. There exists the
+ming-w64 bit project (http://sourceforge.net/projects/mingw-w64/),
+which provides 64bit compiler for the MINGW environment (a subset of
+the UNIX- like environment cygwin provides, unfortunately insufficient
+to build ATLAS).
+
+What you can do, though,
+
+Pure LAPACK-build on Windows 64 bit
+-----------------------------------
+
+Install the x86_64-w64-mingw32 compilers (just about everything except
+for ada and objective C) ;)
+
+In order to run fortran files, you might need to put
+
+x86_64-w64-mingw32/sys-root/mingw/bin/
+
+into your PATH.
+
+Then, go into lapack-lite, and edit make.inc such that gfortran is
+replaced by x86_64-w64-mingw32-gfortran everywhere. You should also
+include the -funroll-all-loops -O3 options.
+
+Compile using "make -k blaslib lapacklib" as usual, and rename the
+results to libblas.a and liblapack.a. Then, do a normal lapack build.
+

INSTALL

@@ -1,5 +1,6 @@
 INSTALL Guide 
 
+Version 1.2.0, January 7, 2011
 Version 1.0, December 22, 2009
 Version 0.3, September 17, 2009
 Version 0.2, May 8, 2009
@@ -22,6 +23,9 @@ If you are only adding code on the Java side, then you can work with
 the supplied build.xml ant script. Simply type "ant jar" to compile
 the new jar file.
 
+Starting with version 1.2.0, you can also use maven to compile the project.
+Both alternatives exist in parallel.
+
 If you want to access further functions from LAPACK, you need to
 recompile the native part. This procedure is a bit more involved, see
 below.
@@ -33,8 +37,12 @@ COMPILING JBLAS (JAVA PART)
 Prerequisites: 
   - java JDK
   - ant
+  - maven
   - ruby
 
+ANT BUILD
+---------
+
 Running "ant clean jar" should do the job ;)
 
 "ant jar" collects all static JNI libraries. Other build targets are:
@@ -54,14 +62,20 @@ Running "ant clean jar" should do the job ;)
   Build a jar file without the JNI and ATLAS libraries. Both must be
   installed in the usual library locations.
 
+MAVEN BUILD
+-----------
+
+Just the usual "mvn clean install" builds everything and installs the jar
+file in your local repository.
 
 INSTALLING JBLAS
+----------------
 
-1. Add "jblas-1.0.jar" to your CLASSPATH
+1. Add "jblas-1.2.0.jar" to your CLASSPATH
 
 2. By default, the dynamic libraries are extracted from the jar file
    to the default temporary directory and loaded from there. If you
-   don't want to do this, extract the dynamic library from jblas-0.2.jar
+   don't want to do this, extract the dynamic library from jblas-1.2.0.jar
    and copy it somewhere where it can be found.
 
    For Linux, use LD_LIBRARY_PATH, for Windows, PATH
@@ -130,6 +144,12 @@ The configure script understands the following options:
    Searches for static libraries only. Resulting jblas will not depend on
    shared LAPACK/BLAS/ATLAS libraries. Default for Windows XP.
 
+--ptatlas
+   Uses multithreaded versions of the ATLAS libraries. AFAIK, these should
+   be compiled on the target computer to have the right number of threads.
+   This means you have to recompile ATLAS, but it is certainly worth
+   the effort, performance improvements are really huge!
+
 --help
    print help
 

Makefile

@@ -32,7 +32,7 @@
 # OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 ## --- END LICENSE BLOCK ---
 
-VERSION=1.1.1
+VERSION=1.2.0
 
 ######################################################################
 #
@@ -54,11 +54,18 @@ PACKAGE=org.jblas
 # generate path from package name
 PACKAGE_PATH=$(subst .,/,$(PACKAGE))
 
-LIB_PATH=native-libs/$(LINKAGE_TYPE)/$(OS_NAME)/$(OS_ARCH)
-FULL_LIB_PATH=native-libs/$(LINKAGE_TYPE)/$(OS_NAME)/$(OS_ARCH_WITH_FLAVOR)
+SRC=src/main
+SRC_JAVA=$(SRC)/java
+SRC_C=$(SRC)/c
+RESOURCES=$(SRC)/resources
 
-GENERATED_SOURCES=src/$(PACKAGE_PATH)/NativeBlas.java native/NativeBlas.c
-GENERATED_HEADERS=include/org_jblas_NativeBlas.h include/org_jblas_util_ArchFlavor.h
+TARGET_C=target/c
+
+LIB_PATH=$(RESOURCES)/lib/$(LINKAGE_TYPE)/$(OS_NAME)/$(OS_ARCH)
+FULL_LIB_PATH=$(RESOURCES)/lib/$(LINKAGE_TYPE)/$(OS_NAME)/$(OS_ARCH_WITH_FLAVOR)
+
+GENERATED_SOURCES=$(SRC_JAVA)/$(PACKAGE_PATH)/NativeBlas.java $(SRC_C)/NativeBlas.c
+GENERATED_HEADERS=$(SRC_C)/org_jblas_NativeBlas.h $(SRC_C)/org_jblas_util_ArchFlavor.h
 SHARED_LIBS=$(FULL_LIB_PATH)/$(LIB)jblas.$(SO) $(LIB_PATH)/$(LIB)jblas_arch_flavor.$(SO) 
 
 #######################################################################
@@ -91,7 +98,7 @@ generate-wrapper: $(GENERATED_SOURCES) $(GENERATED_HEADERS)
 
 # Clean all object files
 clean:
-	rm -f native/*.o native/*.$(SO) $(LIB_PATH)/*.$(SO) $(FULL_LIB_PATH)/*.$(SO) src/$(PACKAGE_PATH)/NativeBlas.java generated-sources
+	rm -f native/*.o native/*.$(SO) $(LIB_PATH)/*.$(SO) $(FULL_LIB_PATH)/*.$(SO) $(SRC_JAVA)/$(PACKAGE_PATH)/NativeBlas.java $(SRC_C)/NativeBlas.c generated-sources
 
 # Full clean, including information extracted from the fortranwrappers.
 # You will need the original fortran sources in order to rebuild
@@ -110,7 +117,7 @@ endif
 generated-sources: \
   scripts/fortranwrapper.rb scripts/fortran/types.rb \
   scripts/fortran/java.rb scripts/java-class.java scripts/java-impl.c \
-  src/org/jblas/util/ArchFlavor.java #src/org/jblas/NativeBlas.java 
+  $(SRC_JAVA)/org/jblas/util/ArchFlavor.java #src/org/jblas/NativeBlas.java
 	$(RUBY) scripts/fortranwrapper.rb --complexcc $(CCC) $(PACKAGE) NativeBlas \
 	$(BLAS)/[sdcz]copy.f \
 	$(BLAS)/[sdcz]swap.f \
@@ -137,20 +144,23 @@ generated-sources: \
 	ant javah
 	touch $@
 
-native/NativeBlas.c: generated-sources
+$(SRC_C)/NativeBlas.c: generated-sources
+
+$(TARGET_C):
+	mkdir $(TARGET_C)
 
-native/NativeBlas.o: native/NativeBlas.c
-	$(CC) $(CFLAGS) $(INCDIRS) -c native/NativeBlas.c -o $@
+$(TARGET_C)/NativeBlas.o: $(SRC_C)/NativeBlas.c $(TARGET_C)
+	$(CC) $(CFLAGS) $(INCDIRS) -c $(SRC_C)/NativeBlas.c -o $@
 
-native/jblas_arch_flavor.o: generated-sources
-	$(CC) $(CFLAGS) $(INCDIRS) -c native/jblas_arch_flavor.c -o $@
+$(TARGET_C)/jblas_arch_flavor.o: generated-sources
+	$(CC) $(CFLAGS) $(INCDIRS) -c $(SRC_C)/jblas_arch_flavor.c -o $@
 
 # Move the compile library to the machine specific directory.
-$(FULL_LIB_PATH)/$(LIB)jblas.$(SO) : native/NativeBlas.$(SO)
+$(FULL_LIB_PATH)/$(LIB)jblas.$(SO) : $(TARGET_C)/NativeBlas.$(SO)
 	mkdir -p $(FULL_LIB_PATH)
 	mv "$<" "$@"
 
-$(LIB_PATH)/$(LIB)jblas_arch_flavor.$(SO): native/jblas_arch_flavor.$(SO)
+$(LIB_PATH)/$(LIB)jblas_arch_flavor.$(SO): $(SRC_C)/jblas_arch_flavor.$(SO)
 	mkdir -p $(LIB_PATH)
 	mv "$<" "$@"
 

README

@@ -1,6 +1,7 @@
 jblas is a matrix library for Java which uses existing high
 performance BLAS and LAPACK libraries like ATLAS.
 
+Version 1.2.0, January 7, 2011
 Version 1.1.1
 Version 1.1, August 16, 2010
 Version 1.0.2, February 26, 2010
@@ -16,8 +17,8 @@ Homepage: http://jblas.org
 
 INSTALL
 
-In principle, all you need is the jblas-1.1,jar in your
-classpath. jblas-1.1.jar will then automagically extract your platform
+In principle, all you need is the jblas-1.2.0,jar in your
+classpath. jblas-1.2.0.jar will then automagically extract your platform
 dependent native library to a tempfile and load it from there. You can
 also put that file somewhere in your load path ($LD_LIBRARY_PATH for
 Linux, %PATH for Windows).
@@ -39,22 +40,23 @@ with either plain lapack, or ATLAS
 sources for BLAS and LAPACK, available, for example from
 http://www.netlib.org/lapack/lapack-lite-3.1.1.tgz.
 
-If you still want to build the source your own, see INSTALL for further
-details.
+If you still want to build the source your own, see INSTALL for
+further details.
 
 
 HOW TO GET STARTED
 
-Have a look at javadoc/index.html and javadoc/org/jblas/DoubleMatrix.html
+Have a look at javadoc/index.html and
+javadoc/org/jblas/DoubleMatrix.html
 
 If you want to validate your installation and get some performance
-numbers, try "java -server -jar jblas-1.1.jar", or drop the server
+numbers, try "java -server -jar jblas-1.2.0.jar", or drop the server
 in case, you only have the "client" JVM installed.
 
 
 LICENSE
 
-jblas is distributed under a BSD-style license. See the file LICENSE
+jblas is distributed under a BSD-style license. See the file COPYING
 for more information.
 
 

RELEASE_NOTES

@@ -1,3 +1,29 @@
+Release 1.2.0 - January 7, 2011
+
+- Added Generalized Eigenvalues for symmetric matrices (which you
+  need, for example, for Canoncial Correlation Analysis). See
+  org.jblas.Eigen.
+
+- Added pure-LAPACK build to Windows 64. Not as nearly as fast as
+  ATLAS, but still 50% over pure Java implementation (for example, for
+  DGEMM).
+
+- Changed build system to maven. Actually, the build.xml from the ant
+  file still exists and can also be used, but now you also have a
+  pom.xml.
+
+- As a side effect of the change to maven, test suites are now
+  actually run during the build ;)
+
+Some bugs:
+
+- Fortran functions returning floats (e.g. SDOT) returned 0.0. Fixed
+  now.
+
+- Bug in some complex numbers routines.
+
+---------------------------------------------------------------------
+
 Release 1.1.1
 
 Mostly bug fixes