[FFmpeg-devel] WMA regression still broken / decouple regression rules

Michael Niedermayer michaelni
Tue Sep 2 15:51:43 CEST 2008


On Tue, Sep 02, 2008 at 02:10:03PM +0200, Benjamin Larsson wrote:
> Robert Swain wrote:
> > 2008/9/2 Robert Swain <robert.swain at gmail.com>:
> >   
> >> 2008/9/2 Michael Niedermayer <michaelni at gmx.at>:
> >>     
> >>> On Mon, Sep 01, 2008 at 10:11:14PM -0700, Mike Melanson wrote:
> >>>       
> >>>> Hi,
> >>>>
> >>>> For most of the x86_32 FATE configurations, (all except icc and gcc
> >>>> 2.95.3), the FFmpeg regressions are breaking on the WMA test. It was
> >>>> last working on 14980:

14980 does not work on x86-64

more preissely:

@@ -205,7 +205,7 @@
 353368 ./tests/data/a-flac.flac
 95e54b261530a1bcf6de6fe3b21dc5f6 *./tests/data/flac.vsynth.out.wav
 stddev:    0.00 PSNR:999.99 bytes:  1058444/  1058444
-4435d87463cd6c5407bd88cca241ca56 *./tests/data/a-wmav1.asf
+6f7f3116b801ea641ce14f827de05cce *./tests/data/a-wmav1.asf
 106004 ./tests/data/a-wmav1.asf
 stddev:12251.38 PSNR: 14.56 bytes:  1056812/  1058444
 stddev: 2105.96 PSNR: 29.85 bytes:  1048620/  1058444

when was it last working on x86-64 ?


> >>>>
> >>>>   http://svn.mplayerhq.hu/ffmpeg?view=rev&revision=14980
> >>>>
> >>>> And it was either 14981 (voroshil) or 14982 (aurel) that broke it:
> >>>>
> >>>>   http://svn.mplayerhq.hu/ffmpeg?view=rev&revision=14981
> >>>>   http://svn.mplayerhq.hu/ffmpeg?view=rev&revision=14982
> >>>>
> >>>> This seems to invert the breakage (previously, icc and gcc 2.95.3 were
> >>>> broken on WMA, and I think the x86_64 configs were, as well).
> >>>>         
> >>> When where ALL the regression tests last working on ALL configurations (not
> >>> counting buggy compilers)?
> >>>       
> >> I think they stopped all working when I switched the window init code
> >> in wma.c to use ff_sine_window_init(). This worked fine on most x86_32
> >> but not on ICC or GCC 2.95.3. And I think it was broken (regressions
> >> failed for some wma test) on x86_64 as well. The diff to the
> >> regressions that was applied to fix this seems to have inverted the
> >> situation.
> >>     
> >
> > Oh sorry, it wasn't a diff to the regressions, it was the addition of
> > the intermediate variable 'alpha'.
> >
> > Rob
> 
> Regressions test on float code shouldn't use a md5sum it is bound to be 
> different depending on platform. Instead of trying to patch up a test 
> that worked by luck in one configuration we should work out a generic 
> way to test float codecs.

Well, x86_32-gcc2.95,x86-32-icc x86_64 and ppc match
thus md5 is clearly not different depening on platform currently
its different depending on gcc version on x86-32

Besides, ratecontrol.c also uses floats and our regression tests that use
it never broke and disabling the bitstream tests entirely for ratecontrol
because it shouldnt work is rater radical.



[...]
-- 
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: Digital signature
URL: <http://lists.mplayerhq.hu/pipermail/ffmpeg-devel/attachments/20080902/5f54d892/attachment.pgp>



More information about the ffmpeg-devel mailing list