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

[VectorDistribution] Allow f16 vector.multi_reduction distribution #19710

Open
wants to merge 1 commit into
base: main
Choose a base branch
from

Conversation

Groverkss
Copy link
Contributor

The elemBitwidth == maxBitsPerShuffle condition was added when we did not have packed shuffle implemented. Since we now go through gpu.subgroup_reduce, which implements packed shuffle, we only need to check if elemBitwidth <= maxBitsPerShuffle.

Copy link
Member

@kuhar kuhar left a comment

Choose a reason for hiding this comment

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

Should we also add a test for int8? I think that should also be supported now.

Copy link
Contributor

@manupak manupak left a comment

Choose a reason for hiding this comment

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

LGTM!

@Groverkss
Copy link
Contributor Author

Groverkss commented Jan 15, 2025

Should we also add a test for int8? I think that should also be supported now.

I don't think this test actually checks if gpu.subgroup_reduce lowers properly. So we could really throw anything to it. We are just relying on gpu.subgroup_reduce tests to check if int8 works for now. I'd prefer not adding it right now because we don't really use these lowerings for anything other than attention where this reduction will either be fp16/fp32. As we move the warp reduction pipeline, i'll add more tests.

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