[llvm] [SCEV] Prove no-self-wrap from negative power of two step (PR #101416)

[Nikita Popov via llvm-commits]

https://github.com/nikic commented:

> Not really a fan of the conditional negate logic here, but it seemed like a bit of overkill to have isKnownToBeAPossibleNegativePowerOfTwo. Any better ideas on how to structure this?

I think it would be okay to add an OrNegative parameter to the function. Computing getNegativeSCEV() may be expensive if it's not just negating a constant.

https://github.com/llvm/llvm-project/pull/101416