[PATCH] D22377: [SCEV] trip count calculation for loops with unknown stride

[Eli Friedman via llvm-commits]

eli.friedman added a comment.

Re: the side-effect issue: You're trying to prove the loop will eventually either exit or execute undefined behavior.  loopHasNoAbnormalExits proves that if the loop exits, it will exit via a visible edge (so it can't throw an exception or kill the program).  There's a gap between the two: loopHasNoAbnormalExits doesn't care whether there's a nested infinite loop.  In practice, this becomes an issue with atomic operations.


https://reviews.llvm.org/D22377