[Ffmpeg-devel] [patch] libavcodec/mpeg12.c:find_start_code() off by one

[Aurelien Jacobs]

On Thu, 12 Jan 2006 17:11:16 +0100
Michael Niedermayer <michaelni at gmx.at> wrote:

> Hi
> 
> On Thu, Jan 12, 2006 at 03:31:55PM +0100, Diego Biurrun wrote:
> > On Wed, Jan 11, 2006 at 02:39:27PM +0100, Diego Biurrun wrote:
> > > On Wed, Jan 11, 2006 at 01:18:48PM +0000, Benjamin Larsson wrote:
> > > > 
> > > > Well current regressions test fail for me, that is with and without
> > > > this patch and with current cvs on a amd64 box.
> > > 
> > > Hmm, do the regression tests work under AMD64 at all?  The last time I
> > > tested they failed on PPC.  I'll report later.
> > 
> > Yes, they fail on PPC on a vanilla source tree, this is the end of the
> > output of 'make test':
> > 
> > ../ffmpeg_g -y -flags +bitexact -dct fastint -idct simple -y -i
> > ./data/a-adpcm_yam.wav -f wav ./data/out.wav
> > FFmpeg version CVS, Copyright (c) 2000-2004 Fabrice Bellard
> >   configuration:  --enable-gpl 
> >   libavcodec version: 51.0.0
> >   libavformat version: 50.0.0
> >   built on Jan 12 2006 15:14:39, gcc: 4.0.3 20060104 (prerelease)
> > (Debian 4.0.2-6)
> > size=    1032kB time=6.0 bitrate=1411.3kbits/s    
> > --- ./data/ffmpeg.regression    2006-01-12 15:22:01.000000000 +0100
> > +++ /home/diego/src/ffmpeg/tests/ffmpeg.regression.ref  2006-01-08
> > 22:32:12.000000000 +0100
> > @@ -119,12 +119,12 @@
> >  2653642 ./data/a-ffv1.avi
> >  799d3db687f6cdd7a837ec156efc171f *./data/out.yuv
> >  stddev:  0.00 PSNR:99.99 bytes:7602176
> > -472a891a0d07e050943d62504240c50f *./data/a-snow.avi
> > -1362158 ./data/a-snow.avi
> > -946ff31ac244e913452ccac77e27be1b *./data/out.yuv
> > -stddev:  3.00 PSNR:38.56 bytes:7602176
> > -745a5ad480b5e2a2319be85764b89012 *./data/a-snow53.avi
> > -3540248 ./data/a-snow53.avi
> > +14fa515bf25a47fc2ad3d18b726e8e31 *./data/a-snow.avi
> > +1197138 ./data/a-snow.avi
> > +e7c746171b092266b0cf55bb5de2a40a *./data/out.yuv
> > +stddev:  2.89 PSNR:38.87 bytes:7602176
> > +11fd61ee7e67ef7a7b2a3df973691305 *./data/a-snow53.avi
> > +3533710 ./data/a-snow53.avi
> >  799d3db687f6cdd7a837ec156efc171f *./data/out.yuv
> >  stddev:  0.00 PSNR:99.99 bytes:7602176
> >  e1da20e3f52f4d2aa18e9486986161fc *./data/a-dv.dv
> > 
> > Regression test: Error.
> > make[1]: *** [codectest] Error 1
> > make[1]: Leaving directory `/home/diego/src/ffmpeg/tests'
> > make: *** [test] Error 2
> > 
> > The other tests fail at some point at well.  So are the regression test
> > i386-only?
> 
> no they where working fine on non x86 sometime ago but people apparently just 
> ignore it if they fail instead of reporting ... 
> maybe someone could do a binary search to find out when they broke (just
> test a-snow53.avi might be a good idea too to speed this up)

Well. I just did it on amd64. The snow regression test failure was
introduced by the following commit :

  Sat,  7 Jan 2006 16:53:24 +0100 (CET)
  regression test for iterative ME & wavelet cmp functions

Wich essentially contain the following:

-do_ffmpeg $file -y -strict -2 -f pgmyuv -i $raw_src -an -vcodec snow -qscale 2 $file
+do_ffmpeg $file -y -strict -2 -f pgmyuv -i $raw_src -an -vcodec snow -qscale 2 -flags +qpel -me iter -dia_size 2 -cmp 12 -subcmp 12 $file

Aurel