Skip to content

cmake : fix compile assumptions for power9/etc #2777

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.

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

Merged
merged 1 commit into from
Feb 5, 2025

Conversation

midnightmagic
Copy link
Contributor

A regression was introduced in the move to CMake and with the
patch for POWER10 support.

This patch corrects CMake build logic to correctly detect and define
-cpu=power9 during build, which turns back on all the VSX codepaths
for POWER9 machines.

Results especially with -DGGML_BLAS=1 restore whisper.cpp performance
to faster-than-realtime transcription with the medium-q5 model on a
Talos II underclocked to 3.3GHz.

@ggerganov
Copy link
Member

Could you rebase on latest master - I think the CI builds should work now.

README.md Outdated
@@ -14,7 +14,7 @@ High-performance inference of [OpenAI's Whisper](https://github.com/openai/whisp
- Plain C/C++ implementation without dependencies
- Apple Silicon first-class citizen - optimized via ARM NEON, Accelerate framework, Metal and [Core ML](#core-ml-support)
- AVX intrinsics support for x86 architectures
- VSX intrinsics support for POWER architectures
- VSX intrinsics support for POWER architectures(#power-vsx-intrinsics)
Copy link
Contributor

Choose a reason for hiding this comment

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

Does this link work without [...] to indicate the link text?

Copy link
Contributor Author

Choose a reason for hiding this comment

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

It does not; corrected.

@fitzsim
Copy link
Contributor

fitzsim commented Feb 4, 2025

Without this patch on my system jfk.wav with ggml-base.en.bin took 7362.89 ms which represents a regression from the Makefile build. With this patch, performance is restored to 2614.46 ms. I reviewed the patch over email with @midnightmagic. I think it looks good, other than the Markdown question I asked as a patch comment.

* Add small comment re: VSX to readme
@midnightmagic midnightmagic force-pushed the fix-power9-cmake-logic branch from 931ffce to 9581255 Compare February 5, 2025 01:37
@midnightmagic
Copy link
Contributor Author

As requested—force-pushed a rebase on current master as of a few minutes ago.

Copy link
Member

@ggerganov ggerganov left a comment

Choose a reason for hiding this comment

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

Thanks, we can merge if the CI is green.

@ggerganov ggerganov merged commit 46d07b9 into ggml-org:master Feb 5, 2025
43 checks passed
joelvaneenwyk pushed a commit to joelvaneenwyk/whisper-cpp that referenced this pull request Mar 3, 2025
* Add small comment re: VSX to readme

Co-authored-by: midnight <[email protected]>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants