dcaballe wrote: I don't have a strong concern here but I'm curious about the motivation to prioritize a 0-D vector over a scalar. Is this mostly to preserve a 0-D tensor input? https://github.com/llvm/llvm-project/pull/116069