[Mlir-commits] [mlir] [MLIR][Affine] Hoist memory accesses with disjoint access ranges (PR #228969)

[Jiaqi He]

heturing wrote:

I also noticed a FIXME in the affine loop-invariant code motion implementation saying that aliasing between memrefs is not checked yet. Without such alias checking, the optimization can be unsound in some cases. I'm happy to work on this as a follow-up once this PR is merged, since it would require modifying part of the code in this PR.

https://github.com/llvm/llvm-project/pull/228969