Fix gemm bugs for sm_80
NVIDIA_GPU
target
#470
Merged
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Outdated suggestions cannot be applied.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.
For
sm_80
, global memory data isround_to_tf32
in any case even if a non-joint matrix version of_gemm
is selected. This rounding causes test failings. This patch splits the class for loading global memory values forgemm_local
andgemm_local_joint_matrix
fixing the issue.Moreover, this patch prevents symmetric operands and
double
datatype to be processed bygemm_local_joint_matrix
since it does not support them.