[llvm] [X86] Handle shifts + and in `LowerSELECTWithCmpZero` (PR #107910)

[via llvm-commits]

goldsteinn wrote:

> Thanks @goldsteinn - I'm trying to find profitable uses for these folds on CMOV targets - any ideas?

I think probably most of the profitable cases are already handled in the middle end i.e `foldSelectICmpAndBinOp` or `foldSelectICmpEq`.

If we are generating these in the backend then the cases we handle in the middle end are probably roughly what is profitable even with CMOV.


https://github.com/llvm/llvm-project/pull/107910