Transposed matmul block packing #944
Merged
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
Enables pack-matmul driver to handle transpose matmul variants. PyTorch nn.Linear layer by default transposes matrix B (weights). This change allows accelerating also these workload by using upstream block packing capability to handle multiple matmul variants. After packing, matmuls are represented by the same generic format accepted by the current XSMM lowering. A small performance hit compared to standard linalg.matmul is due to a linalg.transpose of a matmul_tranpose_a|b input matrix not being lowered to XSMM.
|
In general, our infrastructure can't fully handle matmul_transpose. However, by prepacking matmuls, which we do anyway, the transpose variants can be converted to a form compatible with the rest of our pipeline. |
rengolin
reviewed
Jul 17, 2024
rengolin
approved these changes
Jul 18, 2024
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Outdated suggestions cannot be applied.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.
Enables pack-matmul driver to handle transpose matmul variants.
PyTorch nn.Linear layer by default transposes matrix B (weights). This change allows accelerating also these workload by using upstream block packing capability to handle multiple matmul variants. After packing, matmuls are represented by the same generic format accepted by the current XSMM lowering.
A small performance hit compared to standard linalg.matmul is due to a linalg.transpose of a matmul_tranpose_a|b input matrix not being lowered to XSMM.