Subject: Re: [linux-audio-dev] Another Annoying "How Do I Get Started" Question
From: Steve Harris (S.W.Harris_AT_ecs.soton.ac.uk)
Date: Thu Jul 12 2001 - 18:58:33 EEST
On Wed, Jul 11, 2001 at 03:55:20PM -0400, Greg Berchin wrote:
>
> This confuses me. In reading the documentation for the FPU on the Athlon,
> they state that it can perform two pipelined double precision floating
> point multiplies per cycle. They do not mention any changes in this
> performance that is dependent upon the values of the numbers to be
> multiplied.
Are we talking about SIMD instructions here? I've never used them, but
with generic i686 as produced by gcc -O6:
$ applyplugin m.wav /dev/null amp_1181.so amp -89 svf_1214.so svf 2 3000 0.5 0.5
Plugin cycles: 57664916 (163/sample)
$ applyplugin m.wav /dev/null svf_1214.so svf 2 3000 0.5 0.5
Plugin cycles: 1760478710 (4990/sample)
Try it for yourself if you don't believe me. The first one is using an amp
plugin to drop the range of the input smples to what the plugin is
expecting (-1,+1), the second one just leaves it as (-32768,+32768).
m.wav is mono, 16 bit, just 4 seconds of the middle of a random song I picked.
This is on a PIII, not an Athlon though. If someone could try it on an
Athlon that would be cool.
> There were some benchmarks in a recent issue of EE Times magazine, I think,
> in which the GHz Pentium III blew the doors off just about every available
> DSP for FFT computations and other real-world numerical tasks. The only
> places where DSPs have real advantages any more are low-cost and low power
> applications.
Cool. I can't wait for better SSE support in linux then.
- Steve
This archive was generated by hypermail 2b28 : Thu Jul 12 2001 - 18:14:28 EEST