Add rewrite for matmul when only one of the inputs has batched dimensions #558
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Outdated suggestions cannot be applied.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.
Description
This rewrites replaces a batched matmul by a core matmul by raveling the batched dimensions of the batched input, doing a 2D matmul and then unravelling the batched dimensions of the output.
This sidesteps the Blockwise Dot operation and allows specialization into BLAS routines.
The idea was taken from these two discussions:
numpy/numpy#7569
numpy/numpy#8957
Related Issue
Checklist
Type of change