Skip to content
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

[CPU] Optimize forked onednn binary size via removing gemm legacy above avx2. #19867

Closed
wants to merge 17 commits into from
Closed

[CPU] Optimize forked onednn binary size via removing gemm legacy above avx2. #19867

wants to merge 17 commits into from

Conversation

luweizhou2016
Copy link
Contributor

@luweizhou2016 luweizhou2016 commented Sep 15, 2023 •
edited
Loading

Details:

  • Fix RNN brgemm limitation
  • Fix reshape+FC limitation on inner product
  • Remove legacy gemm on ISAs above avx2
  • Enforce jit_gemm avx512 fall back jit_gemm avx2

Tickets:

@luweizhou2016 luweizhou2016 requested review from a team as code owners September 15, 2023 07:52
@github-actions github-actions bot added category: CPU OpenVINO CPU plugin category: build OpenVINO cmake script / infra labels Sep 15, 2023
@luweizhou2016 luweizhou2016 changed the title Luwei/reduce fork [TEST]Luwei/reduce fork Sep 15, 2023
@luweizhou2016 luweizhou2016 requested a review from a team as a code owner September 18, 2023 03:16
@github-actions github-actions bot added the category: ONNX FE OpenVINO ONNX FrontEnd label Sep 18, 2023
@github-actions github-actions bot removed the category: ONNX FE OpenVINO ONNX FrontEnd label Nov 2, 2023
@luweizhou2016 luweizhou2016 changed the title [TEST]Luwei/reduce fork Luwei/reduce fork Nov 2, 2023
@luweizhou2016 luweizhou2016 changed the title Luwei/reduce fork [CPU] Optimize forked onednn binary size via removing gemm legacy above avx2. Nov 2, 2023
@luweizhou2016 luweizhou2016 mentioned this pull request Nov 2, 2023
Merged
@luweizhou2016
Copy link
Contributor Author

ONEDNN PR openvinotoolkit/oneDNN#219

@luweizhou2016 luweizhou2016 added this to the 2023.3 milestone Nov 2, 2023
@luweizhou2016
Copy link
Contributor Author

Seems dien function regression hit CI test with ENALBE_LTO=ON.

dmitry-gorokhov
dmitry-gorokhov reviewed Nov 2, 2023
View reviewed changes
src/plugins/intel_cpu/thirdparty/CMakeLists.txt Outdated
@@ -13,6 +13,8 @@ endif()

if(ENABLE_LTO)
set(CMAKE_INTERPROCEDURAL_OPTIMIZATION_RELEASE ON)
#Only build gemm driver for AVX2 and below ISAs
set(ONEDNN_ENABLE_GEMM_KERNELS_ISA "AVX2" CACHE STRING "" FORCE)
Copy link
Contributor

@dmitry-gorokhov dmitry-gorokhov Nov 2, 2023 •
edited
Loading

Choose a reason for hiding this comment

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

We need to disable avx512 gemm in any case, not only ENABLE_LTO build

@dmitry-gorokhov
Copy link
Contributor

@EgorDuplensky please review as well

@luweizhou2016 luweizhou2016 force-pushed the luwei/reduce_fork branch 2 times, most recently from 4525bf0 to 340522e Compare November 15, 2023 08:17
@luweizhou2016
Copy link
Contributor Author

Only have 3MB gain. Will not merge after dicussing.

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

@dmitry-gorokhov dmitry-gorokhov dmitry-gorokhov left review comments

Assignees

@dmitry-gorokhov dmitry-gorokhov

@EgorDuplensky EgorDuplensky

Labels
category: build OpenVINO cmake script / infra category: CPU OpenVINO CPU plugin
Projects
None yet
Milestone
2023.3
Development

Successfully merging this pull request may close these issues.
3 participants
@luweizhou2016 @dmitry-gorokhov @EgorDuplensky