[PATCH] D81004: [MemorySSA] Skip phi translation for phis in cycles (WIP).

Wed Jun 10 17:47:25 PDT 2020

fhahn added a comment.

In D81004#2086387 <https://reviews.llvm.org/D81004#2086387>, @efriedma wrote:

> It looks like this is a regression from https://reviews.llvm.org/D79068 .

I am not sure, I think the patch above fixes an issue with translating a phi to a wrong address. I think the issue in this patch is slightly different: the PHI is translated correctly, but the result is used in an unsound way.

> Yes, when we're doing PHI translation in MemorySSA, we have to consider if there are cycles.  Not sure what the algorithm needs to look like off the top of my head.

Great, I just wanted to make sure I am not missing anything obvious before digging too deep.

Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D81004/new/

https://reviews.llvm.org/D81004