[llvm] [LV] Generate scalar VPInstruction::Not if possible. (PR #152628)

[Luke Lau via llvm-commits]

lukel97 wrote:

> Hmm, can you wait a day or so before landing this patch? I'm worried about the potential codegen impact of this PR and would like to see the difference in generated output for some loops.

That would be good to check on AArch64. I just checked RISC-V and there was no diff on llvm-test-suite. I think VectorCombine might have already been scalarizing this.

https://github.com/llvm/llvm-project/pull/152628