Fix SM120 blockscaled FP32 epilogue store layout#3345
Open
NVIDIA-JerryChen wants to merge 1 commit into
Open
Conversation
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Both kernels used
StMatrix8x8x16bOp(..., 2)to build the C atom passed tomake_tiled_copy_C_atom. ForFloat32outputs, the actual register-to-SMEM copy atom isCopyUniversalOp, but the C TV layout is still derived from this stmatrix C atom. The x2 layout does not cover the full C fragment for this epilogue path, causing incomplete SMEM writes before the final TMA store. Changing the C atom toStMatrix8x8x16bOp(..., 4)provides a C TV layout that covers the full C fragment and fixes the observed FP32 output mismatch. In fact, unless there is a special case, we do not use x2; by default, we always use x4. Therefore, I think x2 came from a typo.