Skip to content

bpo-29782: Consolidate _Py_Bit_Length() - #20739

Merged
vstinner merged 21 commits into
python:masterfrom
niklasf:bpo-29782-msb
Jun 15, 2020
Merged

bpo-29782: Consolidate _Py_Bit_Length()#20739
vstinner merged 21 commits into
python:masterfrom
niklasf:bpo-29782-msb

Conversation

@niklasf

@niklasfniklasf commented Jun 8, 2020

Copy link
Copy Markdown
Contributor

In GH-2866, _Py_Bit_Length() was added to pymath.h for lack of a better
location. GH-20518 added a more appropriate header file for bit utilities. It
also shows how to properly use intrinsics. This allows reconsidering bpo-29782.

  • Move the function to the new header.
  • Changed return type to match __builtin_clzl and reviewed usage.
  • Use intrinsics where available.
  • Pick a (mostly theoretical) fallback implementation suitable for
    inlining.

https://bugs.python.org/issue29782

@vstinnervstinner left a comment

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Maybe add test_bit_length() to _testinternalcapi.c: see test_popcount().

Comment threadInclude/internal/pycore_bitutils.h Outdated
Comment threadInclude/internal/pycore_bitutils.h Outdated
Comment threadInclude/internal/pycore_bitutils.h Outdated
Comment threadModules/mathmodule.c Outdated
@vstinner

Copy link
Copy Markdown
Member

Previous rejected attempt: PR #594.

Comment threadInclude/internal/pycore_bitutils.h Outdated
@vstinner

Copy link
Copy Markdown
Member

cc @mdickinson

@vstinnervstinner left a comment

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I disliked PR #594 because it added #include <intrin.h> to a new *public header file: Include/pyintrinsics.h. This PR is different, all changes are made in the internal C API. We have way more freedom in the internal C API.

Comment threadModules/mathmodule.c Outdated
@vstinner

Copy link
Copy Markdown
Member

The Windows build looks broken:

Run .\python.bat -m test.pythoninfo
Running Release|x64 interpreter...
##[error]Process completed with exit code 1.

@mdickinson
mdickinson self-requested a review June 8, 2020 18:26
niklasf added 4 commits June 8, 2020 21:18
In pythonGH-2866, _Py_Bit_Length() was added to pymath.h for lack of a better
location. pythonGH-20518 added a more appropriate header file for bit utilities. It
also shows how to properly use intrinsics. This allows reconsidering bpo-29782.
* Move the function to the new header.
* Changed return type to match __builtin_clzl and reviewed usage.
* Use intrinsics where available.
* Pick a (mostly theoretical) fallback implementation suitable for
inlining.
Comment threadInclude/internal/pycore_bitutils.h Outdated
Comment threadModules/mathmodule.c Outdated
Comment threadInclude/internal/pycore_bitutils.h Outdated
Comment threadInclude/internal/pycore_bitutils.h Outdated
niklasfand others added 2 commits June 8, 2020 22:03
Co-authored-by: Victor Stinner <vstinner@python.org>
Comment threadInclude/internal/pycore_bitutils.h Outdated
Comment threadInclude/internal/pycore_bitutils.h Outdated
Comment threadInclude/internal/pycore_bitutils.h
@niklasf

Copy link
Copy Markdown
ContributorAuthor

Thanks for reviewing.

Also added unit tests as suggested (in addition to the existing test coverage). I briefly disabled the branch with clz (8e3ce66) to get a CI run with the fallback implementation. The current fallback is identical to the original (except for inlining).

@vstinnervstinner left a comment

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks for the many updates! The code now looks better! New review.

Comment threadInclude/internal/pycore_bitutils.h
Comment threadModules/_testinternalcapi.c Outdated
Comment threadModules/_testinternalcapi.c Outdated
Comment threadObjects/longobject.c Outdated
Comment threadInclude/internal/pycore_bitutils.h Outdated
@vstinner

Copy link
Copy Markdown
Member

Using gcc 10 -O3 -flto, I get the following machine code in test_bit_length():

# input = rax
# if (x != 0)
test rax,rax
je 0x7fffeaa99978 <test_bit_length+504>
# return (int)sizeof(unsigned long) * 8 - __builtin_clzl(x);
# return 0x40 - bsr(rax) ^ 0x3f
bsr rax,rax
mov ecx,0x40
xor rax,0x3f
mov r9d,ecx
sub r9d,eax
# result = r9d

x86 BSR (Bit Scan Reverse) instruction is used as expected.

@vstinnervstinner left a comment

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM!

@mdickinson: Do you want to review this PR as well?

@mdickinson

Copy link
Copy Markdown
Member

@mdickinson: Do you want to review this PR as well?

I'd very much like to take a look, but am short on available time. Can you give me until the weekend? If I've failed to review by the start of next week, go ahead and merge.

@vstinner

Copy link
Copy Markdown
Member

@mdickinson: this PR can wait one week :-)

Comment threadObjects/longobject.c
static int
bit_length_digit(digit x)
{
Py_BUILD_ASSERT(PyLong_SHIFT <= sizeof(unsigned long) * 8);

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It would be nice if C provided some standard way to get the width (in the sense of C99 §6.2.6.2p6) of an integer type, so that we didn't have to use sizeof(unsigned long) * 8, which could fail in the presence of padding bits. But as far as I know it doesn't. Related: https://stackoverflow.com/q/3957252/270986

@mdickinsonmdickinson left a comment

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM!

@vstinner
vstinner merged commit 794e7d1 into python:masterJun 15, 2020
@vstinner

Copy link
Copy Markdown
Member

Thanks @niklasf, I merged your PR.

I removed "(mostly theoretical)" from your commit message:

* Pick a (mostly theoretical) fallback implementation suitable for
inlining.

We support other C compilers, like XLC.

Sign up for freeto join this conversation on GitHub. Already have an account? Sign in to comment

Labels

Projects

None yet

Development

Successfully merging this pull request may close these issues.

5 participants

@niklasf@vstinner@mdickinson@the-knights-who-say-ni@bedevere-bot