[Mlir-commits] [mlir] Added Memory Effect to assume_alignment (PR #139521)

[Shay Kleiman]

shay-kl wrote:

@MaheshRavishankar 
So you're saying that implicitly if an operation has no memory effects and no results then it makes sense for it to never be optimized away?
Would it be a better idea to explicitly define a trait for this type of operation or should we just deduct that they're an annotation and shouldn't be removed?

https://github.com/llvm/llvm-project/pull/139521