Skip to content

fix using __AMDGCN_WAVEFRONT_SIZE - #307

Open
psychocoderHPC wants to merge 1 commit into
alpaka-group:devfrom
psychocoderHPC:fix-hipWavefrontIssue
Open

fix using __AMDGCN_WAVEFRONT_SIZE#307
psychocoderHPC wants to merge 1 commit into
alpaka-group:devfrom
psychocoderHPC:fix-hipWavefrontIssue

Conversation

@psychocoderHPC

@psychocoderHPC psychocoderHPC commented Jul 10, 2026

Copy link
Copy Markdown
Member

fix #299

  • Refactor the code to use alpaka's warp size function

  • fix bug that Scatter::getAvailableSlotsAccelerator() was not
    mentioning the requirement to be called as a collective method.

  • merge update alpaka version  #306 first (first 2 commits)

@psychocoderHPC psychocoderHPC added this to the 3.0.0 milestone Jul 10, 2026
sharedResult = 0u;

// wait that all shared memory is initialized
alpaka::syncBlockThreads(acc);

Copy link
Copy Markdown
Member Author

Choose a reason for hiding this comment

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

This synchronization is the reason why the old implementation already enforced that this method should be called with all threads in the thread block.

@chillenzer

Copy link
Copy Markdown
Contributor

Please rebase.

@chillenzer

Copy link
Copy Markdown
Contributor

While you're at it, it would be great to split off the formatting things from the functional changes. But not blocking... only if time permits.

fix alpaka-group#299

- refactor the code to use alpaka's warp size function
- fix bug that `Scatter::getAvailableSlotsAccelerator()` was not mentioning the requirement to be called as collective method.
@psychocoderHPC
psychocoderHPC force-pushed the fix-hipWavefrontIssue branch from e5cdefe to cb3e733 Compare July 14, 2026 12:17
@psychocoderHPC

Copy link
Copy Markdown
Member Author

I removed the formatting changes. I forgot to run pre-commit install in this repository, so alpaka configs were used, and the changes were introduced.

Sign up for free to 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.

Wavefront size macro is not available since ROCm 7.2

2 participants