[FFmpeg-devel] Proposed patch

[Roman Shaposhnick]

On Sat, 2007-08-04 at 02:40 +0200, Michael Niedermayer wrote:
> Hi
> 
> On Thu, Aug 02, 2007 at 08:22:02PM -0700, Roman Shaposhnik wrote:
> > Guys,
> > 
> > two questions: 
> >    1. Does an attached patch look ok?
> 
> this changes one of the most speed critical things in lavc so this needs
> extensive benchmarks, x86+ppc mpeg4 decoding at minimum
> (more non x86 archs would be very welcome too)

Hi Michael, 

here's what I did: I added START_TIMER/STOP_TIMER around
avcodec_decode_video and experimented with MPEG4 recording
of Lord of the Rings. I compared the # of dezicycles before
and after the patch. On x86 the change seems to be a wash +-0.5%.

On SPARC I plan to use Sun's Performance Analyzer to get
the same result (sorry, I'm currently swamped with LinuxWorld
otherwise I would have had the results already).

Now, the questions that I have are:

   * does this seem to be enough to justify the patch or do you 
     want me to run a specific kind of benchmarking?
   * has the ppc question been fully answered (again, I don't have PPC)?

Thanks,
Roman.