[Mlir-commits] [mlir] [mlir][affine] Handle generic memory ops in loop fusion dependence analysis (PR #224657)

[llvmlistbot at llvm.org]

================

----------------
lipracer wrote:

Looks correct to me now. Only ask: please add a comment on the value-less branch explaining why it returns true (value-less read/write effects target unknown memory, and a may-analysis must not
      under-approximate). Without it this branch reads as arbitrary and is likely to get "simplified" away later, silently reintroducing the under-approximation. Non-blocking otherwise.

https://github.com/llvm/llvm-project/pull/224657