[ODE] Some SSE in Quick step

[Aras Pranckevicius]

> AFAIK, nobody's actually ever verified that D3DX's mathematical
> manipulations are slower than ODE's built-in.  Which is part of why a
> referred to this as a "non-argument".

I've measured D3DX vs. plain C++ (VC6) vs. Nebula2' "somewhat SSE optimized" 
math classes (using intrinsics, not asm code). I've measured matrix*matrix 
operation. IIRC, results were what D3DX loses by a small amount to both C++ 
and Nebula2 code (it was on AthlonXP, so SSE probably doesn't do much over 
C++).

I also remember VirtualDub's rants about how SSE intrinsics suck on VC - 
http://www.virtualdub.org/oldnews :)

Yes, D3D software vertex shaders rock, but D3DX math functions for small 
things (eg. single matrix*matrix) aren't That Good. Probably they would be 
good if I were transforming 1000 vectors by a matrix in a single call, but I 
guess that's not that needed in ODE...


Aras Pranckevicius aka NeARAZ
http://www.gim.ktu.lt/nesnausk/nearaz/