02-03-2015, 09:38 PM
I've just run it under cache-grind and it reports a significant number of L1 misses in the code with repetitions and none in the code without. Looks like the a_Same array just pushes the working data over the size of the L1.