[iiwusynth-devel] Re: SSE

[Josh Green]

On Sun, 2002-10-27 at 21:56, M. Nentwig wrote:
> Hello,
> 
> I found some material on SSE:
> 
> http://x86.ddj.com/articles/sse_pt1/
> http://x86.ddj.com/articles/sse_pt2/
> http://x86.ddj.com/articles/sse_pt3/simd3.htm
> http://courses.ece.uiuc.edu/ece291/archive/spr2002/lecture/lecture18_6pps.pdf
> 
> and the gcc manual
> http://gcc.gnu.org/onlinedocs/gcc-3.2/gcc/X86-Built-in-Functions.html#X86%20Built-in%20Functions
> 
> It can really process four floats at the same time. Now think about the
> 4th order interpolation routine (a0*s[0]+a1*s[1]+a2*s[2]+a3*s[3])...
> There was one line that worries me, though. The article says, that the
> operating system has to support SSE for context switching.
> 

Cool!

I tried the gcc switches including the "-march=pentium3" but I did not
get any speed increase (according to profiling output when the synth is
closed). This was kind of disappointing :( Ohh well, it seemed too easy
anyways.

Time to code by hand.. I'm pretty sure you have nothing to worry about
in regards to Linux. I know of one program in particular that uses SSE
(mplayer) on my system. Also, "cat /proc/cpuinfo" shows "SSE" in the
flags section.

> Cheers
> 
> Markus

Lates.
        Josh