[llvm] [SCEV] Introduce IncrementNSUW, replacing NSSW (PR #219559)

[Ramkumar Ramachandra via llvm-commits]

artagnon wrote:

> TBH I don't really get this change. NUUW, NUSW, NSSW are all reasonable wrapping behaviors, but I have a hard time understanding how NSUW would be useful (except for the case where it happens to be equivalent to neither NUUW or NSSW).

Hm, my initial idea was that NSSW is probably too strong a predicate to work with _in practice_, as proving things about (implications, that it is redundant etc.) is harder, but maybe this justification is unsound? I will try updating the patch with NSSW to eliminate the regressions to see if there are any improvements we get?

https://github.com/llvm/llvm-project/pull/219559