On Sun, 26 Jul 1998, Yann Guidon wrote:
> > What do you think, should an MP3 encoder benefit from using MMX
> > instructions? Are there plans to add MMX support to 8hz-mp3? Is there
> > anyone working on it now - does he need help? I don't know MMX _yet_ but
> > I'll take the opportunity to learn if there is need.
>
> I think MMX is 1) purely INTEL dependent 2) purely integer dependent
> and 3) purely algorithm dependent.
> 8HzMP3 will lose portability, uses FP and will require too much algorithm
> revamping.
You don't have to lose protability - hide the changes behind an #ifdef. It
would be handy for those with Intel but without MMX anyway.
The idea just came to me after I saw XingPlayer for Windows compress a
4-minute song in two minutes. 128kbps/44kHz, of course. That was on a
PII/266, but it's astonishing nevertheless. Then I read somewhere that it
gains 33% of the speed by using MMX instructions.
> but who knows... maybe with MMX-2 ? :-P
Or K6-2 3Dnow! as Oliver suggested? In any case, using _any_ of these that
would help sounds as a good idea to me.
---
Alex Shnitman
[EMAIL PROTECTED], [EMAIL PROTECTED]
http://alexsh.home.ml.org
