Add __floatunsi{sd}f for IEEE754.
Unlike the quad versions, these don't use floating point.