[Mlir-commits] [mlir] [mlir][vector] Fix mask rank extension for TransferWriteNonPermutationLowering (PR #144893)

[Dhairyashil R G]

dhairyashilRG wrote:

@Groverkss, are you still planning to land this? I ran into the same bug while working on #227821: a masked `vector.transfer_write` whose map skips a middle dim, like `(d0, d1, d2) -> (d0, d2)`, fails in `-convert-vector-to-scf` with a mask type mismatch.

I ported this patch to current main and it fixes my case. It needs two small updates: the op creation calls have to move to the `Op::create(builder, ...)` form, and the new test's `CHECK-LABEL` has to end in `_and_transpose` to match the function name.

I'm happy to rebase it, or to send it as a new PR if you prefer.

https://github.com/llvm/llvm-project/pull/144893