Repository navigation
Synchronize GQA reduction scratch before reuse - #3600
Open
1sgtpepper wants to merge 2 commits into
Open
1sgtpepper wants to merge 2 commits into
1sgtpepper wants to merge 2 commits into
Conversation
Signed-off-by: 1sgtpepper <cynejarviszarceno@gmail.com>
Signed-off-by: 1sgtpepper <cynejarviszarceno@gmail.com>
|
This PR has been labeled |
This branch has not been deployed
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Fixes #3597.
Complete all
cta_reducescratch reads before the next reduction reuses the storage. The existing barrier orders publication before reads; a second arrival on that same 128-thread barrier orders reads before reuse.Tests
Build
93_blackwell_low_latency_gqa_cta_reducewithCUTLASS_NVCC_ARCHS=100a, then run:On B200/CUDA 13.1.1, the repeated-reduction test passes numerically with zero racecheck hazards. Restoring the original header produces wrong values and eight hazards. CMake build.
This fixes scratch reuse independently of the mailbox initialization race in #3601. With both fixes, the contiguous and paged example checks pass. Another 16 contiguous regression cases pass under racecheck and memcheck.