I don't believe these tests show much. You need to be doing something realistic inside the loop, and not repeating it using num_iterations, that will probably b

I don't believe these tests show much. You need to be doing something realistic inside the loop, and not repeating it using num_iterations, that will probably be optimized out into an unrolled memcpy. Too simplistic to be an accurate test. Modern CPUs will precache memory forwards as well as in reverse.