Thanks, what I was slightly concerned about is that we now have the implicit assumption that the initial value is 0. I mean that's what the vectorizer does for reductions but theoretically, wouldn't we also combine other values into 0 now?
Regards Robin
