[ODE] Optimizing for SIMD

[Gene Ruebsamen]

Hello,

I wonder how much SIMD operations will help speed up ODE.

Unless I am mistaken, 3DNow! and SSE only support single-precision floating
point operations.  Thus, unless you are willing to take a hit in accuracy,
your performance may not benefit too much; although, I am sure there are
some areas of ODE where the SIMD instructions would benefit greatly..

Check out http://libsimd.sourceforge.net/ .  Looks like someone is working
on a portable SIMD library.  Maybe we could use this in ODE...

Any thoughts?

Gene Ruebsamen

> -----Original Message-----
> From: ode-admin@q12.org [mailto:ode-admin@q12.org]On Behalf Of Eli Curtz
> Sent: Monday, June 10, 2002 1:46 PM
> To: ODE group
> Subject: [ODE] Optimizing for SIMD
>
>
> Hi all,
>
> This may be insane, but I just got the PS2 Linux kit and was idly
> considering porting ODE to it. Doing this "right" would involve
> implementing portions on the vector unit of the PS2. Has anybody out
> there thought about trying to optimize ODE for Altivec, SSE, or
> anything similar? I'm a little bit scared of the LCP solver, <famous
> last words> but it's only a few hundred lines of code, how hard could
> it be? </famous last words>
>
> - eli
>
> --
> Eli Curtz
> Sycophant to the Stars
> eli@nuprometheus.com
> _______________________________________________
> ODE mailing list
> ODE@q12.org
> http://q12.org/mailman/listinfo/ode
>