Just add a fixme about a possibly faster implementation of some atomic loads on some...