[llvm] [KnownBits] Make nuw and nsw support in computeForAddSub optimal (PR #83382)

[via llvm-commits]

goldsteinn wrote:

> > > This seems to add a good bit of compile-time overhead: http://llvm-compile-time-tracker.com/compare.php?from=2a67c28abe8cfde47c5058abbeb4b5ff9a393192&to=e383a7e50bf0b4303931ef119c856dcf53f3dd9e&stat=instructions%3Au
> > 
> > 
> > Wow, is the implementation just that in-efficient, or do you think its changing control flow decisions in some key places?
> 
> Pretty sure this is either the implementation being slow, or the change at https://github.com/llvm/llvm-project/pull/83382/files#diff-4cc32f30c79b4c8161eac82916c70c1d56e75b0dd7d6e56bbb76f8b16e20b32dR361 (which might result in more KnownBits calculations now -- dunno whether that case is common or not).

Okay, Ill investigate and see if I can make the impact a bit more reasonable.

https://github.com/llvm/llvm-project/pull/83382