[Mlir-commits] [mlir] [mlir][vector] Pad the target shape when unrolling masks (PR #234448)

[Federico Bruzzone]

https://github.com/FedericoBruzzone commented:

Thanks for looking into this!

Since this issue is already being addressed in #217514, I'd prefer to avoid having two independent PRs targeting the same issue without coordinating the proposed solutions.

The two approaches appear to differ: #217514 returns a match failure for rank mismatches, whereas this PR pads the target shape and also handles the `constant_mask` normalization case.

Could we first compare the two approaches in the existing PR and agree on the desired fix?

CC: @amanyagami 

https://github.com/llvm/llvm-project/pull/234448