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

[purnima shrivastava]

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

----------------
purnima-nlp wrote:

Thanks for pointing this out. I agree that value-less memory effects are not represented by `getEffectedValues`, so this could theoretically under-approximate accesses for such ops.

Since this is currently theoretical and the PR is focused on handling generic operand-associated memory accesses without crashing, I’ve left that behavior unchanged here.

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