[PATCH] D153953: Revert "[AMDGPU] Mark mbcnt as convergent"

[Jay Foad via Phabricator via llvm-commits]

foad added a comment.

In D153953#4458134 <https://reviews.llvm.org/D153953#4458134>, @sameerds wrote:

> @pravinjagtap @arsenm ... reverting the mbcnt intrinsic affects tests that were added for atomic optimizations. In particular, the mbcnt is now being moved across/into/out of control flow because it is no longer convergent. I eyeballed one example and it seemed okay to me, but a more thorough check will be useful.

They are just being moved from before the loop to after the loop. This is fine. It is even a bit weird that the atomic optimizer pass emits them before the loop in the first place.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D153953/new/

https://reviews.llvm.org/D153953