[Mlir-commits] [mlir] [mlir][Vector] Add canonicalization for extract_strided_slice(create_mask) (PR #146745)

[Kunwar Grover]

Groverkss wrote:

> LG, thanks! Just some minor comments.
> 
> I'm curious: what is the current lowering of this without the canonicalization? A `vector.shuffle` extracting the slice from the mask?

Yes. Not only that, in a lot of cases you could actually fold the mask with the maskedload into a load, but the extract_strided_slice was blocking it.

https://github.com/llvm/llvm-project/pull/146745