Originally posted by The Baron
Not for a benchmark it doesn't.
If you replace a part of a benchmark with your own creation in order to improve your scores for the benchmark, no matter how crappily the benchmark is coded for whatever architecture you're using, you're corrupting the benchmark. Plain and simple.
For a game? Sure. I'm ALL FOR IT. But for a benchmark? No. Otherwise, you'd have to be dependent on a driver team to optimize for every game on the planet in order to get similar performance.
This is exactly why I have stated why you cannot trust a synthetic benchmark for accurate determination of performance, It's supposed to be Generic, And it's obviously not.
Whether or not these optimisations are "cheating" or not, I could really care less,
In Both cases, The optimisations requiire specific coding for the specific aplication in this case,