[llvm] [InstCombine] restrict InstCombineNegator for sub(y,and(lshr(x,C),1)) (PR #222075)

[Nikita Popov via llvm-commits]

nikic wrote:

If we don't canonicalize in this direction (which I agree does not look profitable), would it make sense to have a transform for the reverse direction?

https://github.com/llvm/llvm-project/pull/222075