[ODE] Some SSE in Quick step

[tbp]

Russ Smith wrote:
> * it's only really efficient on big matrices ... it's better suited as a
> computational engine for, say, matlab. using it to multiply little 4x4
> matrices together is a huge overkill.
> 
You'll get the same symptom from any "external" lib regarding SSE 
speedups, be it D3D or ATLAS.

> * to get maximum efficiency it must be configured for each machine it
> runs on, and the configuration process takes maybe half an hour. a
> 'generic' atlas may perform that well on some machines.
The maximum efficiency metric is hard to come by :)

I came to the same conclusion as you for my own stuff. It's too 
huge/clunky and i couldn't extract the specific routines i was looking 
for without embarking a whole string of cruft in the process.
And that's a shame.