topperc wrote: > The original patch only supported ordered fadd. Why does this to support more than just vp.reduce.fadd? Oh I guess its because of the scalar input that the regular reduce intrinsics don't have? https://github.com/llvm/llvm-project/pull/131974