Enforce 128B alignment for TMA shared memory allocation. #3023
Conversation
This file contains 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
zasdfgbnm
approved these changes
Sep 26, 2024
There was a problem hiding this comment.
This change looks good to me. But before you merge, could you move your changes in csrc/device_lower/pass/allocation.cpp, as well as initializeMbarrier and invalidateMbarrier, from #2833 to this PR? Because these are the things that makes TMA buffer no longer aligned well. I checked them in #2833 and they look good to me as well. Also, a nit pick: Should we move initializeMbarrier and invalidateMbarrier to allocation.cpp or as an ir_utils?
|
!build |
zasdfgbnm
approved these changes
Sep 26, 2024
|
I moved |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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Outdated suggestions cannot be applied.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.
From https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#table-alignment-multi-dim-tma, the shared memory address must be 128B aligned.
This PR enforces this alignment for all TMA shared memory allocations to avoid
CUDA_ERROR_MISALIGNED_ADDRESSerrors.