As with all performance techniques, benchmarking is the only way to establish whether loop unrolling actually optimizes your heuristic.<p>For example, no manual loop unrolling strategies were able to improve my Cisco password decryptor. It's worth experimenting, and then falling back to the normal way of coding when tests reveal no advantage.<p>Like all good optimization techniques, the compiler gets better and better at applying the right ones for the right source code.