[Mlir-commits] [mlir] [mlir] [tensor] Add patterns to remove whole slicing of tensors (PR #107046)

[Mehdi Amini]

joker-eph wrote:

@matthias-springer  can you comment here?

To me this is a problem of coupling that would indicate that the bufferization needs a "prepare" phase independent of the canonicalization, but it does not make sense to keep useless redundant information in the IR just because "maybe somehow that affect this one pass".

https://github.com/llvm/llvm-project/pull/107046