[Mlir-commits] [mlir] [mlir][vector] Remove MatrixMultiplyOp and FlatTransposeOp from Vector dialect (PR #144307)

[Andrzej Warzyński]

banach-space wrote:

Thanks for the feedback and sorry for the delay replying!

The [2nd commit](https://github.com/llvm/llvm-project/pull/144307/commits/1ad2ef4bbf6047e0eb6b70f889a6884be89cf2cd) reverts the dependency of Vector Transforms on the LLVM dialect (i.e. restore the original state). The updated/new transformations have been **moved** to the `CovertVectorToLLVM` pass. So:
* Use `--convert-vector-to-llvm='vector-contract-lowering=matmul'` for `vector.contract` → `llvm.intr.matrix.multiply`,
* Use `--convert-vector-to-llvm='vector-transpose-lowering=flat'` for `vector.transpose` → `llvm.intr.matrix.transpose.

Hopefully this makes sense, thanks!

https://github.com/llvm/llvm-project/pull/144307