[PATCH] D104445: [SimplifyCFGPass] Tail-merging function-terminating blocks

[Matt Arsenault via Phabricator via llvm-commits]

arsenm added a comment.

In D104445#2827485 <https://reviews.llvm.org/D104445#2827485>, @rnk wrote:

> GPU targets, wasm, and Windows EH preparation may need to undo this type of region-breaking transform late in the backend, and if they do, the only thing this transform is doing is throwing away debug info.

For AMDGPU we generally need to converge to a single return block anyway. We end up trying to merge divergent return-like things into a single exit point (e.g. AMDGPUUnifyDivergentExitNodes) like this patch is doing


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D104445/new/

https://reviews.llvm.org/D104445