cuda: keep the PTX L2 prefetch hints off HIP and MUSA - #168
Merged
Conversation
The prefetch.global.L2 asm and __cvta_generic_to_global added in #135 were guarded by __CUDA_ARCH__ alone, which hipcc also defines, so the ROCm and Windows HIP release builds fail on mmq.cuh and mmvq.cu. Exclude HIP and MUSA from the two hint blocks; the surrounding code is unchanged.
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
What
Two guard changes, no other code:
around the two PTX L2 prefetch hints added in #135: the Q1_0 J128 tile hint in
mmq.cuhand the decode weight prefetch inmmvq.cu.Why
hipcc defines
__CUDA_ARCH__for device code too, so the HIP builds try to compileprefetch.global.L2inline PTX and the CUDA-only__cvta_generic_to_global. Both ROCm jobs of the release workflow fail on it (run 34291069282):For HIP and MUSA this restores the pre-#135 code at those two spots; CUDA is unchanged.
Verification
test-backend-ops test -b CUDA0 -o MUL_MAT1283/1283.