Philipp Hack

CC @chr1sj0nes.

CC @cheshire, @lorenrose1013.

Thanks for the review. Can you PTAL?

@SandSnip3r it would help if PR #56995 was merged before resolving the conflicts in the rewrite test.

Extension of XLA pattern matching: #57648.

Ideally, this would be merged without further delay. In that case, I would separately change the patterns.

> Let me know if you cant reproduce these. All tests pass for me. Can you provide more information on the failures?

I see the expected HLO on A100 and V100. What GPU do you run the tests on?

> EDIT: I do not see the issue on my local 3090. The *F16Padded tests verify the padding and slicing that is applied to run GEMMs based on operands with...

@reedwm This is largely independent of the requirement of transposing A but not B. The column-major/row-major layout returned by GemmConfig::For doesn’t fully describe the configuration of a GEMM. Not considering...