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

[Pankaj Chawla via llvm-commits]

pankajchawla added a comment.

Please ignore my previous email. My analysis was incorrect. The stride cannot be negative here as it will lead to UB.

@Reid
You mentioned that this is one of the TU which is changed due to the patch. Are you sure this one is causing the miscompilation?
Does the issue go away if we disable optimizations on this TU?
Did you run opt-bisect on it? Does it point to any pass? If not, is it possible to run it?

As the change is in an analysis, the issue can be in any transformation using this analysis rather than the analysis itself. Any help from your side in narrowing down the issue is greatly appreciated.

I will keep looking into the generated IR for this file in the meantime.

Thanks,
Pankaj


Repository:
  rL LLVM

https://reviews.llvm.org/D22377