[MPlayer-users] mp3lib/ffmp3 performance issues on via c7

[ACS]

For a PII, mp3lib uses the mmx optimized assembler routines.
These routines run at about 50...55% of the cpu load compared to compiled 
i386 C source. (recently tested this myself on mpg123 on P3. mpg123 uses 
similar libraries as mp3lib)

There's a drawback of mmx optimization:
Instead of floating points, mmx asm optimization uses some 16 bit integer 
arithmetic so LSB accuracy is lost for 16 bit out!
For accuracys sake, I'd be glad not to use this mmx "optimized" code.

adrian bacon