[Mlir-commits] [mlir] [mlir][scf] Add simple LICM pattern for `scf.while` (PR #76370)

[Ivan Butygin]

Hardcode84 wrote:

> Making LICM part of the scf.for canonical form is on the other hand making scf.for opinionated in terms of representing an overly abstract loop form, reducing its applicability. 

Side comment: Isn't this argument applies to this entire concept of uplifting `scf.while` to `scf.for` as part of canonicalization?

As I said previously, I can live with the LICM being separate pass, the big limitation of this approach is that Upstream currently missing a facility to run sequence a passes until fixed point is reached.

https://github.com/llvm/llvm-project/pull/76370