[MPlayer-users] Re: bug in configuration file

Nilmoni Deb ndeb at ece.cmu.edu
Sat Jan 12 06:07:02 CET 2002

I am afraid the site in question reports something more fundamental 
that just the use of Atlas. In fact it reveals less optimized fpu stack
handling and fetch scheduler as compared to gcc-2.95.x .
Also I posted it to raise questions about gcc-2.9six, not gcc-3.0.x.
The improvement that u see in your case is probably due to the
march=athlon option enabled (which is not there for gcc-2.95.x). 
All the article in question reveals is that gcc-3.0.x could have been
*better* than what it is by simply retaining the good features of 2.95.x.

- Nil

Nilmoni Deb <ndeb at ece.cmu.edu> wrote:
> [Automatic answer: RTFM (read DOCS, FAQ), also read
> The page http://www.bero.org/gcc296.html claims that
> gcc 2.9six "generates better, more optimized code.".
> From the analysis in http://www.cs.utk.edu/~rwhaley/ATLAS/gcc30.html
> it seems quite opposite, i.e., gcc-2.95.x is better.

It is known, that ATLAS suffers from the use of gcc>2.95. But using
gcc-3.0.2 or 3.0.3, I can report a reproducible speed gain of
20% in transcode (and maybe mencoder, didn't yet measure this program)
using opendivx output if compiled with the proper -march and -mcpu
setting for my Athlon. So you can't generalize the speed issues some
programs seem to have nor can you generalize the speed gain other programs
seem to have. 

S --

More information about the MPlayer-users mailing list