Persistent AG + GEMM kernels w/ HipKittens & UserBuffers - #707
Open
alextmagro wants to merge 4 commits into
Open
Persistent AG + GEMM kernels w/ HipKittens & UserBuffers#707alextmagro wants to merge 4 commits into
alextmagro wants to merge 4 commits into
Conversation
alextmagro
requested review from
ipanfilo,
wangye805 and
wenchenvincent
as code owners
August 18, 2026 20:44
![github-actions[bot]](https://avatars.githubusercontent.com/in/15368?s=60&v=4){kind=small}
wangye805
requested changes
Aug 20, 2026
wangye805
requested changes
Aug 20, 2026
wangye805
approved these changes
Aug 21, 2026
aris134
self-requested a review
August 21, 2026 18:24
aris134
reviewed
Aug 21, 2026
wenchenvincent
approved these changes
Aug 22, 2026
aris134
approved these changes
Aug 22, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Introduces persistent AG + GEMM kernels for the TN and NN layout -- 1.247x and 1.304x speedup on average across 24 training sizes.
Win is entirely in overlap -- GEMM speeds are 0.885x and 0.923x compared to hipBLASLt, so optimization opportunity available there. FSDP overlap is considered robust, but only tested with an intranode simulation.