Skip to content

Support compress&expand in avx512vbmi2 #1102

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Merged
serge-sans-paille merged 5 commits into from
Apr 3, 2025
Merged

Support compress&expand in avx512vbmi2 #1102

serge-sans-paille merged 5 commits into from
Apr 3, 2025

Conversation

@junparser
Copy link
Contributor

@junparser junparser commented Mar 28, 2025

No description provided.

@junparser
Copy link
Contributor Author

junparser commented Mar 28, 2025

@serge-sans-paille [Linux build / gcc 11 - avx512vnni (pull_request)](https://github.com/xtensor-stack/xsimd/actions/runs/14120454206/job/39559599836?pr=1102) looks weird, I can't reproduce it with gcc&gcc12, what's the cflags used here?

@junparser
Copy link
Contributor Author

junparser commented Apr 2, 2025

@serge-sans-paille [Linux build / gcc 11 - avx512vnni (pull_request)](https://github.com/xtensor-stack/xsimd/actions/runs/14120454206/job/39559599836?pr=1102) looks weird, I can't reproduce it with gcc&gcc12, what's the cflags used here?

this caused by _mm512_shuffle_epi8 for batch swizzle in avx512bw which is incorrect. use _mm512_permutexvar_epi8 in avx512vbmi instead.

@junparser
Copy link
Contributor Author

junparser commented Apr 2, 2025

@serge-sans-paille , CMake Error at benchmark/CMakeLists.txt:12 (cmake_minimum_required): Compatibility with CMake < 3.5 has been removed from CMake. , seems we hit cmake issue. would you please take a look?

@serge-sans-paille
Copy link
Contributor

serge-sans-paille commented Apr 2, 2025

@junparser #1104 should do the trick

@junparser
Copy link
Contributor Author

junparser commented Apr 2, 2025

@junparser #1104 should do the trick

Also, after this commit, would you like to add one more gcc 11 - avx512vbmi2 in ci?

@serge-sans-paille
Copy link
Contributor

serge-sans-paille commented Apr 2, 2025

gcc 11 - avx512vbmi2

#1105

@serge-sans-paille
Copy link
Contributor

serge-sans-paille commented Apr 2, 2025

@junparser
Copy link
Contributor Author

junparser commented Apr 3, 2025 •
edited
Loading

Thanks, now this pr passes avx512vbmi2. Also is it possible to add something like [Linux build / clang18/20 - avx512vbmi2 in ci?

serge-sans-paille
serge-sans-paille reviewed Apr 3, 2025
View reviewed changes
include/xsimd/config/xsimd_arch.hpp

using all_x86_architectures = arch_list<
avx512vnni<avx512vbmi>, avx512vbmi, avx512ifma, avx512pf, avx512vnni<avx512bw>, avx512bw, avx512er, avx512dq, avx512cd, avx512f,
avx512vnni<avx512vbmi2>, avx512vbmi2, avx512vbmi, avx512ifma, avx512pf, avx512vnni<avx512bw>, avx512bw, avx512er, avx512dq, avx512cd, avx512f,
Copy link
Contributor

@serge-sans-paille serge-sans-paille Apr 3, 2025

Choose a reason for hiding this comment

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

Can you motivate this change? Why would composition between avx512 VNNI be more legit with VMBI2 rather than VBMI?

Copy link
Contributor Author

@junparser junparser Apr 3, 2025 •
edited
Loading

Choose a reason for hiding this comment

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

I made this change based on https://en.wikichip.org/wiki/x86/avx512_vnni. The table shows that all of the arch have vbmi2 as well as vnni.

serge-sans-paille
serge-sans-paille reviewed Apr 3, 2025
View reviewed changes
@serge-sans-paille serge-sans-paille merged commit 4b182f4 into xtensor-stack:master Apr 3, 2025
60 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@serge-sans-paille serge-sans-paille serge-sans-paille left review comments

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

2 participants

@junparser @serge-sans-paille