In case you haven't read through the existing threads; I've optimized several areas of PQ by converting the existing floating point routines to fixed point. If you rewrite those routines in ASM and use floating point it will still be slower than the fixed point C routines. It takes somewhere around 10 times longer to perform a floating point operation compared to an integer operation.<br>So, I believe far more performance increase can be achieved through fixed point conversion compared to rewriting in ASM. Of course the ultimate goal is to have fixed point ASM code.

<br>I am somewhere in the middle of a very large fixed point conversion, where I convert all the vertices of the models to fixed point when they are loaded, and all math from then on is pure fixed point. I'd say its roughly half done. If you take a look at quakedef.h, you'll see several definitions that toggle my fixed point optimizations. USE_PQ_OPT3 toggles the optimization I was working on last (of course it is not usable yet). Search for those other defines to see what I've already optimized.<br><br>Dan East