bpo-29782: Use __builtin_clzl for bits_in_digit if available #594

niklasf · 2017-03-10T10:58:58Z

No description provided.

the-knights-who-say-ni · 2017-03-10T10:58:59Z

Hello, and thanks for your contribution!

I'm a bot set up to make sure that the project can legally accept your contribution by verifying you have signed the PSF contributor agreement (CLA).

Unfortunately our records indicate you have not signed the CLA. For legal reasons we need you to sign this before we can look at your contribution. Please follow these steps to rectify the issue:

Sign the PSF contributor agreement. The "bugs.python.org username" requested by the form is the "Login name" field in "Your Details" at b.p.o
Wait at least one US business day and then check the "Contributor form received entry under "Your Details" on bugs.python.org to see if your account has been marked as having signed the CLA (the delay is due to a person having to manually check your signed CLA)
Reply here saying you have completed the above steps

Thanks again to your contribution and we look forward to looking at it!

mention-bot · 2017-03-10T10:59:03Z

@niklasf, thanks for your PR! By analyzing the history of the files in this pull request, we identified @mdickinson, @serhiy-storchaka, @tim-one, @brettcannon and @benjaminp to be potential reviewers.

mdickinson

Some comments left on the bpo issue. In brief:

configuration decisions (deciding whether __builtin_clz is available and does the right thing) should be moved into the configuration machinery.
the code currently assumes that digit is the same as unsigned long. That's not always the case.
once the above issues are addressed, we still need to weigh any speed advantage against the costs in terms of maintainability, portability, readability.

mdickinson · 2017-03-10T18:36:52Z

the code currently assumes that digit is the same as unsigned long

Ah, no. I see. This should work whenever digit is no larger than unsigned long, thanks to the implicit type conversion. However, that's probably wasteful in the majority of cases; on many (most?) machines, we should be able to use clz rather than clzl, and it may be faster. And while it's probably true that on all current platforms digit is no larger than unsigned long, that's not an assumption we should hardwire in unnecessarily.

niklasf · 2017-03-10T22:56:29Z

Making an extra version for 16bit ints does not seem to be worth it, so I settled for unsigned long.
Factored out so that mathmodule.c can also take advantage of the function.
I might need help with the configuration process. Checking __GNUC__ does not seem to be uncommon, but is this the place to do it? Note that __GNUC__ is also defined by clang.
I created a new source/header pair which might not be the way to go. Should the header contents go to pyport.h as you suggested? Where to put the C fallback?

vstinner · 2017-03-27T15:30:47Z

Python/pyintrinsics.c

+    5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5
+};
+
+int bit_length(unsigned long d) {


Please rename the function to _Py_bit_length().

Why the result is signed? Why not unsigned int?

vstinner · 2017-03-27T15:31:26Z

Include/pyintrinsics.h

+ * count_leading_zero_bits(x)
+ */
+
+#if defined(__GNUC__)


Is the builtin available in old GCC versions? Any idea when the function was added?

gcc 3.4. I made the guard more precise. It also works for clang.

Btw. while looking how this is usually done in Python I found a typo that would affect gcc 3.x: https://github.com/python/cpython/blob/master/Include/pymacro.h#L56 should say __GNU*C*_MINOR__. I'll make a seperate PR. #878.

vstinner · 2017-03-27T15:31:59Z

Include/pyintrinsics.h

+#if defined(__GNUC__)
+#define HAVE_BIT_LENGTH
+static inline int bit_length(unsigned long d) {
+    return d ? (8 * sizeof(unsigned long) - __builtin_clzl(d)) : 0;


would it be possible to avoid the conditional code?

Unfortunately I don't think so. __builtin_clzl(0) is undefined or rather has different behavior on various processors.

vstinner · 2017-03-27T15:33:40Z

Include/pyintrinsics.h

+}
+#elif defined(_MSC_VER)
+#define HAVE_BIT_LENGTH
+#include <intrin.h>


Do we need the following pragma?
#pragma intrinsic(_BitScanReverse)

niklasf · 2017-03-28T16:14:12Z

@Haypo Thanks for the review. I updated the PR based on your comments.

vstinner · 2017-04-26T21:50:19Z

http://bugs.python.org/issue29782 has been closed as REJECTED.

the-knights-who-say-ni added the CLA not signed label Mar 10, 2017

serhiy-storchaka assigned mdickinson Mar 10, 2017

serhiy-storchaka added the performance Performance or resource usage label Mar 10, 2017

mdickinson suggested changes Mar 10, 2017

View reviewed changes

niklasf force-pushed the bpo-29782-clzl branch from 4efa595 to bf26e37 Compare March 10, 2017 19:43

the-knights-who-say-ni added CLA signed and removed CLA not signed labels Mar 10, 2017

niklasf force-pushed the bpo-29782-clzl branch 2 times, most recently from 21c4a4b to bff329a Compare March 10, 2017 22:32

niklasf force-pushed the bpo-29782-clzl branch from bff329a to 3cbc250 Compare March 10, 2017 22:57

niklasf mentioned this pull request Mar 22, 2017

bpo-29882: Add an efficient popcount method for integers #771

Merged

vstinner requested changes Mar 27, 2017

View reviewed changes

niklasf force-pushed the bpo-29782-clzl branch from 3cbc250 to 7365797 Compare March 28, 2017 15:53

bpo-29782: Use __builtin_clzl for bits_in_digit if available

f06f612

niklasf force-pushed the bpo-29782-clzl branch from 7365797 to f06f612 Compare March 28, 2017 16:06

vstinner closed this Apr 26, 2017

vstinner mentioned this pull request Jan 16, 2020

bpo-31031: Unify duplicate bits_in_digit and bit_length #2866

Merged

mdickinson mentioned this pull request Apr 22, 2020

Count leading zero bits in one CPU tick #19633

Closed

vstinner mentioned this pull request Jun 8, 2020

bpo-29782: Consolidate _Py_Bit_Length() #20739

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

bpo-29782: Use __builtin_clzl for bits_in_digit if available #594

bpo-29782: Use __builtin_clzl for bits_in_digit if available #594

niklasf commented Mar 10, 2017

the-knights-who-say-ni commented Mar 10, 2017

mention-bot commented Mar 10, 2017

mdickinson left a comment •

edited

Loading

mdickinson commented Mar 10, 2017

niklasf commented Mar 10, 2017 •

edited

Loading

vstinner Mar 27, 2017

vstinner Mar 27, 2017

niklasf Mar 28, 2017 •

edited

Loading

vstinner Mar 27, 2017

niklasf Mar 28, 2017

vstinner Mar 27, 2017

niklasf commented Mar 28, 2017

vstinner commented Apr 26, 2017

bpo-29782: Use __builtin_clzl for bits_in_digit if available #594

bpo-29782: Use __builtin_clzl for bits_in_digit if available #594

Conversation

niklasf commented Mar 10, 2017

the-knights-who-say-ni commented Mar 10, 2017

mention-bot commented Mar 10, 2017

mdickinson left a comment • edited Loading

Choose a reason for hiding this comment

mdickinson commented Mar 10, 2017

niklasf commented Mar 10, 2017 • edited Loading

vstinner Mar 27, 2017

Choose a reason for hiding this comment

vstinner Mar 27, 2017

Choose a reason for hiding this comment

niklasf Mar 28, 2017 • edited Loading

Choose a reason for hiding this comment

vstinner Mar 27, 2017

Choose a reason for hiding this comment

niklasf Mar 28, 2017

Choose a reason for hiding this comment

vstinner Mar 27, 2017

Choose a reason for hiding this comment

niklasf commented Mar 28, 2017

vstinner commented Apr 26, 2017

mdickinson left a comment •

edited

Loading

niklasf commented Mar 10, 2017 •

edited

Loading

niklasf Mar 28, 2017 •

edited

Loading