[MPlayer-dev-eng] MPEG-1/2 decoding benchmarks

[Rich Felker]

On Tue, Apr 19, 2005 at 03:52:34AM +0200, Diego Biurrun wrote:
> On Mon, Apr 18, 2005 at 04:47:13PM +0200, Michael Niedermayer wrote:
> > 
> > On Monday 18 April 2005 15:06, Diego Biurrun wrote:
> > >
> > > If you have further suggestions or things I should do differently,
> > > speak up.  I _could_ also test on a Pentium MMX 233, but I doubt the
> > 
> > maybe try a different gcc version, i didnt benchmark ffmpeg with 2.95 lately, 
> > but with 3.3 ffmpeg beats libmpeg2 by 10% on many files like 
> > matrixbench_mpeg2.mpg, while on others it lost miserably by a similar amount
> 
> I retested with gcc 3.3.5 on x86 as requested as well as 2.95.4 on
> PPC.
> 
> Result:
> 
> x86:
> libmpeg2 0.4.0 is up to 5% faster than 0.3.1, but gcc 3.3.5 gives up
> to a 5% slowdown compared to 2.95.4.
> 
> For FFmpeg "lavdopts idct=4" no longer is an improvement, but
> "lavdopts fast" is still the quickest.
> 
> When comparing FFmpeg and libmpeg2, libmpeg2 no longer is the clear
> winner, it's faster (7%) for bomb.mpeg, but slower for heist60.mpeg
> (5%) and dvd.mpeg (8%).

Conclusion: libmpeg2 is considerably faster than ffmpeg, but the
suckiness of gcc3 slows down libmpeg2 so much that it's no longer a
clear winner when compiling with gcc3.

Rich