Hell, if anything on a short sequence the naive (one loop at a time) approach might be faster if it keeps the instruction and data caches more coherent.