[Ffmpeg-devel] [BUG] regression tests failure in r7867

[Michael Niedermayer]

Hi

On Sat, Feb 10, 2007 at 01:31:05AM +0100, Guillaume POIRIER wrote:
> Hi,
> 
> On 2/9/07, Michael Niedermayer <michaelni at gmx.at> wrote:
> >Hi
> >
> >On Fri, Feb 09, 2007 at 11:20:27PM +0100, Guillaume POIRIER wrote:
> >> Hi,
> >>
> >> On 2/7/07, Baptiste Coudurier <baptiste.coudurier at smartjog.com> wrote:
> >> >Hi
> >> >
> >> >M?ns Rullg?rd wrote:
> >> >> Benoit Fouet said:
> >> >> All the tests (except ffserver) passed last night on my x86-64.  That
> >> >was before
> >> >> the pixfmt checkin though, which is why I suggested it might be the
> >> >culprit.
> >> >>
> >> >
> >> >No problem here, I just fixed some libav problems.
> >>
> >> Just tonight, on PPC, with r7903, the regression tests are as follow:
> >>
> >> ../ffmpeg_g -y -flags +bitexact -dct fastint -idct simple -y -i
> >> ./data/a-vorbis.asf -f wav ./data/out.wav
> >> FFmpeg version SVN-r7903, Copyright (c) 2000-2007 Fabrice Bellard, et al.
> >>  configuration: --enable-gpl --cpu=g5
> >>  libavutil version: 49.3.0
> >>  libavcodec version: 51.32.0
> >>  libavformat version: 51.8.0
> >>  built on Feb  9 2007 23:10:26, gcc: 4.0.1 (Apple Computer, Inc. build 
> >5363)
> >> size=    1032kB time=6.0 bitrate=1411.3kbits/s
> >> --- ./data/ffmpeg.regression    2007-02-09 23:15:12.000000000 +0100
> >> +++ /Users/guillaumepoirier/Prgm/ffmpeg/tests/ffmpeg.regression.ref
> >> 2007-02-09 23:08:20.000000000 +0100
> >> @@ -190,15 +190,15 @@
> >> stddev: 33.31 PSNR:65.87 bytes:1040384
> >> 768305cec342e01b7a9d38e971ef0174 *./data/a-wmav1.asf
> >>    99596 ./data/a-wmav1.asf
> >> -da5422f83dc05390942f32a1b944c1e5 *./data/out.wav
> >> +2ab193df22bb473e61015e56ac11faf3 *./data/out.wav
> >> stddev:12251.50 PSNR:14.56 bytes:1056768
> >> stddev:2106.00 PSNR:29.85 bytes:1048576
> >> c706c7cf9db10df7a7353a40feddcfdd *./data/a-wmav2.asf
> >>    99602 ./data/a-wmav2.asf
> >> -f42e82b11484f3cebdec207fe5c0b61f *./data/out.wav
> >> +8d05d345bc14153f3d0da95fdf34e084 *./data/out.wav
> >
> >well, the wma codec uses floats so its very possible to be a simple 
> >rounding
> >difference, if you could upload one of the out.wav (that is one generated
> >by decoding a-wmav1/2.asf) i could take a look ...
> 
> I uploaded a-wmav2.asf-out.wav incoming of samples.mplayerhq.hu

this one contains a few +-1 errors, likely caused by different rounding
someone could try to replace a few float by double and see if that makes 
the difference go away ...
besides that disabling the md5 check on out.wav for these should solve
the problem


> 
> 
> >> stddev:12255.92 PSNR:14.55 bytes:1056768
> >> stddev:2099.31 PSNR:29.88 bytes:1048576
> >> -46de10949d472b2ff2622ca220ec5318 *./data/a-vorbis.asf
> >> +ef725cb9c13b9437a41428eabadeba52 *./data/a-vorbis.asf
> >>    35686 ./data/a-vorbis.asf
> >> -3ea5225c7ec10fefea600c150f326bdb *./data/out.wav
> >> -stddev:3984.67 PSNR:24.31 bytes:1056768
> >> +e46368ef818f2ec3abcc3f8390d14df5 *./data/out.wav
> >> +stddev:3882.46 PSNR:24.54 bytes:1056768
> >
> >this is more then just a rounding difference, first step here is to check
> >how vorbis encoded on x86 looks when decoded on ppc and the other way
> >around so we kno if its the encoder or decoder after that tracking that
> >down may or may not be easy ...
> 
> I uploaded a-vorbis-PPC.asf and a-vorbis-x86-64.asf (the name says it
> all which machine generated them).
> 
> I can't hear a difference between both samples, so I don't know what
> should be concluded out of that... Maybe I should try with a "natural"
> sample (i.e. music, rather than the computer-generated sine wave).
> Am I right.

all vorbis files decode to identical output so the decoder is clearly
broken on non x86, the difference is too large for mere rounding
also a-vorbis-PPC.asf and a-vorbis-x86-64.asf differ only in 2 bytes

someone should investigate what these bytes are, my feeling tells me
they might be some uninitalized vars or something like that, volunteers?

[...]

-- 
Michael     GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB

Republics decline into democracies and democracies degenerate into
despotisms. -- Aristotle
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: not available
URL: <http://lists.mplayerhq.hu/pipermail/ffmpeg-devel/attachments/20070212/9013e520/attachment.pgp>