[PATCH] D79855: [AMDGPU] Strengthen export cluster ordering

[Jay Foad via Phabricator via llvm-commits]

foad accepted this revision.
foad added a comment.
This revision is now accepted and ready to land.

This is OK, and I can confirm it fixes some Vulkan CTS failures we were seeing.

For bonus marks:

1. removeExportDependencies iterates the preds of every node. I think it could be a lot faster if it only iterated the preds and succs of export nodes.
2. I still wonder if we could get the barriers right in the first place, perhaps by pretending that each kind of export instruction writes to a fake address space and teaching alias analysis that those address spaces are disjoint.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D79855/new/

https://reviews.llvm.org/D79855