[llvm] [SCEV] Strip incorrect code in proving RHS > Start in howManyLT (PR #222878)

[Ramkumar Ramachandra via llvm-commits]

artagnon wrote:

> (With that in mind, I think the change you were going for in #222939 is in principle possible, you just need to be careful about the predicate checking. I think you could convert Pred=GT, FoundPred=GE to use FoundPred=GT instead, but you can't convert Pred=GE, FoundPred=GT to use Pred=GT instead.)

Thanks again for the guidance! I will continue working away at the problem, and turn this into a draft for now: yes, I missed the implication subtlety :(

https://github.com/llvm/llvm-project/pull/222878