[Mlir-commits] [mlir] [mlir][linalg] Fix EraseIdentityLinalgOp on fill-like ops (PR #130000)

[Andrzej Warzyński]

banach-space wrote:

This looks pretty straightforward and overall LGTM. That said, it's not my area of expertise, so an extra pair of eyes would be helpful.

I've left a few minor suggestions inline. Also - could you update the outdated comment above `EraseIdentityLinalgOp`?

> /// Remove any linalg operation (on tensors) that are just copying

And regarding:

> This check already exists for linalg ops with pure tensor semantics.

Is there a negative test for the tensor case?

https://github.com/llvm/llvm-project/pull/130000