iree
iree copied to clipboard
Transpose narrow matmuls so that the narrow dimension is M
Narrow-N and narrow-M cases of matmuls are entirely similar. There is no need to write data-tiling logic and ukernels for all combinations of the two narrow dimensions.
We have standardized on only implementing data-tiling logic and ukernels for the narrow-M case. The idea all along was to reduce narrow-N cases to that by transposition. So this issue is about finally doing that.