fhahn wrote: > Returning invalid for a scalarized intrinsic is fine FWIW, but can you update the PR title/description to say "Fix crash costing cttz.elts with zvl32b" Sure, updated thanks https://github.com/llvm/llvm-project/pull/226181