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.

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

[ROCm] Move device libs path initialization into CompileToHsaco #8506

Conversation

draganmladjenovic
Copy link
Contributor

This fixes sporadic crashes in multithreaded_compilation_test_gpu.

This fixes sporadic crashes in multithreaded_compilation_test_gpu.
@github-actions github-actions bot added the kokoro:force-run Forces CI to rerun label Jan 16, 2024
@kokoro-team kokoro-team removed the kokoro:force-run Forces CI to rerun label Jan 16, 2024
Copy link
Member

@penpornk penpornk left a comment

Choose a reason for hiding this comment

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

Thank you for the fix!

copybara-service bot pushed a commit to tensorflow/tensorflow that referenced this pull request Jan 17, 2024
…saco

Imported from GitHub PR openxla/xla#8506

This fixes sporadic crashes in multithreaded_compilation_test_gpu.
Copybara import of the project:

--
a6fc9ada24d551f35e5f01bafb2cadbcf848f41b by Dragan Mladjenovic <Dragan.Mladjenovic@amd.com>:

[ROCm] Move device libs path initialization into CompileToHsaco

This fixes sporadic crashes in multithreaded_compilation_test_gpu.

Merging this change closes #8506

PiperOrigin-RevId: 599134851
copybara-service bot pushed a commit that referenced this pull request Jan 17, 2024
Use a backend-specific flag to allow overriding the MLIR pass pipeline when
compiling to CPU via the new runtime pass. This is helpful to debug the
pipeline, in particular the performance of the generated code, without having
to recompile XLA every time. The feature is marked as experimental, hidden deep
inside debug options and obviously disabled in absence of the option.

Also update the signatures of `create_compilation_pipeline` and
`create_specialization_pipeline` in `JitOptions` to return a status. This
allows for a more graceful failure when MLIR pass pipeline cannot be
constructed. CPU pipeline constructor was already aborting with a fatal error,
update that to returning internal error. This makes the JIT friendlier to
library users such as JAX.

FUTURE_COPYBARA_INTEGRATE_REVIEW=#8506 from ROCmSoftwarePlatform:ci_multithreaded_compilation_test_crash a6fc9ad
PiperOrigin-RevId: 598624926
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.

5 participants