[Mlir-commits] [mlir] [MLIR][GPU] Added FoldSelfCopy pattern to gpu memcpy (PR #221216)

[llvmlistbot at llvm.org]

Edward-EJ wrote:

> A wait is pessimistic here I believe. Why can't we just propagate the async dependencies?

I thought that if there are several async dependencies, trying to replace the memcpy with them would lead to a mismatch between the value and the number of replacements. 

>  FYI the PR description becomes the commit message candidate in this project, so please write as such.

Thank you for the feedback. I will update the PR description accordingly. 

https://github.com/llvm/llvm-project/pull/221216