[PATCH] D103834: [SCEV] Properly guard reasoning about infinite loops being UB on mustprogress

[Philip Reames via Phabricator via llvm-commits]

reames added a comment.

JFYI, C and C++ apparently have different rules here.  For c++ your example function is mustprogress, for C only the loop is.  (Based on a quick skim of checkIfLoopMustProgress clang's CodeGenFunction.h)  I have no idea if the C rules could be more aggressive here or not, you'd be best taking that up with a clang developer.

It does look like checking the loop mustprogress flag would be sufficient to address the specific regression seen here.  I'll see if that's easy to implement quickly.

For the weakness being discussed, feel free to file a bug.  I'm currently working through a collection of missing cases in SCEV's trip count logic and having an example to come back to would help.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D103834/new/

https://reviews.llvm.org/D103834