On Tue, 5 Oct 1999, Greg Maxwell wrote:
> Just saw 'GOGO' on freshmeat.net, aparently it's an mp3 encoder based on
> Lame 3.23beta with MMX, SSE(KNI), and 3Dnow support. It's supposed to be
> 2x as fast as Lame 3.23...
GOGO's is damn speedy, but I'm not sure if it's the direciton LAME should
be going, or if there's any work that they've done that we can use
About gogo:
- japanese comments
- chock full of asm routines. In fact, they have simultaneously
included multiple implementations of each critical routine in each of the
processor flavours they are supporting (code is a little confusing)
- i'm not sure they've optimized any routines, but "just"
rewritten them in asm (which is a pretty huge task)
- I remember doing some timing comparisons between gogo, lame and
xing. lame slowest, then gogo, then xing the quickest, iirc. gogo only
became faster than xing when the psychoacoustic model was turned off
(-nopsy, equivalent to lame's -f)
This tells me something:- xing's psychoacoustic modelling must use
something pretty different from the ISO code.
2 related questions:
- anyone tried implementing psy model 1 with layer3?
- anyone heard any more about a new fraunhofer codec?
later
mike
--
MP3 ENCODER mailing list ( http://geek.rcc.se/mp3encoder/ )
