Optimising code for Altivec G4 code is real fruity

The IBM page is dedicated to optimising for AltiVec, which the geeky among you will know is the performance engine included in Apple's G4 processors. Products like Adobe Photoshop has AltiVec optimisations in them which allow them to perform better on G4, clock for clock, than the previous generation G3 processor. Think of it as a fruity, and probably more useful, SSE.

However, one can't help but laugh at the silliness of the IBM example code given. IBM spends a page with some complex, untidy code, and then says that 'Because the code runs in the system idle loop, it ends up running the most (and thus getting the largest speedup) when the system is fairly quiet.' How helpful!

Tacking on to this theory, IBM then go on to demonstrate how you can make the spinning beach-ball of death (the 'system busy, wait a couple of minutes' OSX annoyance) more processor friendly by converting RGB pixel values. The page might as well be Microsoft showing how to make the BSOD process more memory efficient.

There is a lesson to learn from the example code, however, and the lesson is that unrolling code, whilst horrendously ugly from a code quality point of view, can have some serious performance benefits on Macs that are still relevant today, since the newest G5 chips still support Altivec. If you've got any interest in learning more about the technology, you should head over to IBM and check out their developer resources pages.