[Mlir-commits] [mlir] [mlir][vector] Add mask elimination transform (PR #99314)

[Diego Caballero]

dcaballe wrote:

What I'm suggesting here is:
  1. Move lightweight mask removal patterns to the op canonicalization. We want to remove trivial masks as part of the canonicalization. Create a `populateLightweightMaskRemovalPatterns` for them.
  2. Keep the heavyweight mask removal patterns in the pass and also add the patterns in `populateLightweightMaskRemovalPatterns` for completeness.

In that way, we would have the trivial ones as part of canonicalization and when more heavyweight analysis is needed to remove the masks, we will use the pass. Does it make more sense now or am I missing something?

https://github.com/llvm/llvm-project/pull/99314