https://github.com/artagnon approved this pull request. > On AArch64, this leads to additional vectorization in a few cases in practice. Why is this marked as an NFCI? Looks good otherwise, thanks! https://github.com/llvm/llvm-project/pull/172915