f66e235a13
guru out there can find a way to take advantage of little-endianness to make this computation more efficient. (I am certain that it can be done, but haven't managed to make it work myself.)