[Mlir-commits] [mlir] [mlir][vector] Deal with special patterns when emulating masked load/store (PR #75587)

[Jakub Kuderski]

kuhar wrote:

> > What happens when you run `--canonicalize --cse` after masked load/store emulation? Is this enough to fold all the masking away? If not, what is left?
> 
> There is no change after applying `--canonicalize --cse`.

I mean with the current lowering without the changes from this PR. When I wrote `vector.gather` lowering I found that we didn't need any special cases for constant inputs as everything would just fold away: https://github.com/llvm/llvm-project/blob/1b4a0794b0c7244a0b74ac0a2239fb60a62f9c70/mlir/test/Dialect/Vector/vector-gather-lowering.mlir#L130-L153

https://github.com/llvm/llvm-project/pull/75587