[PATCH] D153537: [AMDGPU] Do not wait for vscnt on function entry and return

[Jay Foad via Phabricator via llvm-commits]

foad added a comment.

>> It is only used to resolve memory dependencies, and that is handled by SIMemoryLegalizer.
>
> The idea was the stores on the caller side don't need to wait to finish because the prolog wait will take care of it after the latency of the jump. Is the memory legalizer not taking advantage of this?

I don't see how it can be taking advantage of that, because it only considers each load or store in isolation.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D153537/new/

https://reviews.llvm.org/D153537