[PATCH] D112693: [SimplifyCFG] Fix miscompile in tryWidenCondBranchToCondBranch. PR52290

[Philip Reames via Phabricator via llvm-commits]

reames added a comment.

Max, the wideable condition intrinsic is carefully specified to not require any particular item down the widened path.  Your explanation is misleading at best.

I suspect you do have a bug here - the test case sure seems to imply that - but it's not the lack of a deoptimizing call down the untaken path of the widenable condition.  I'd almost suspect a bug in your frontend lowering to this if it weren't for the infinite loop in the test case.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D112693/new/

https://reviews.llvm.org/D112693