[llvm] [InstCombine] Fold (trunc X to i1) & !iszero(X & Pow2)) -> (X & (1 | Pow2)) == (1 | Pow2) (PR #119196)

[Yingwei Zheng via llvm-commits]

https://github.com/dtcxzyw commented:

Can you adjust `foldLogOpOfMaskedICmps` to convert `trunc X to i1` back into `(X & 1) != 0`?

BTW, I skimmed through the comments in https://github.com/llvm/llvm-project/pull/84628. It seems to cause many performance regressions. It would be better to revert #84628 and find another way to unblock https://github.com/llvm/llvm-project/pull/83829.
cc @nikic @YanWQ-monad @DianQK @goldsteinn 


https://github.com/llvm/llvm-project/pull/119196