[Mlir-commits] [mlir] [mlir][vector] Add `SwapShapeCastOfTranspose` canonicalizer pattern (PR #100933)

[Diego Caballero]

dcaballe wrote:

Isn't this pointing at an implementation gap in the drop unit dimension transformations? If we apply drop unit dimensions here, we should get a shape_cast before the transpose, which should cancel the shape_cast after the transpose?

https://github.com/llvm/llvm-project/pull/100933