[FFmpeg-devel] FATE Errors

Sat May 28 11:57:18 EEST 2022

> -----Original Message-----
> From: ffmpeg-devel <ffmpeg-devel-bounces at ffmpeg.org> On Behalf Of
> Paul B Mahol
> Sent: Saturday, May 28, 2022 10:50 AM
> To: FFmpeg development discussions and patches <ffmpeg-
> devel at ffmpeg.org>
> Subject: Re: [FFmpeg-devel] FATE Errors
> 
> On Sat, May 28, 2022 at 10:41 AM Soft Works <softworkz at hotmail.com>
> wrote:
> 
> >
> >
> > > -----Original Message-----
> > > From: ffmpeg-devel <ffmpeg-devel-bounces at ffmpeg.org> On Behalf Of
> > > Paul B Mahol
> > > Sent: Saturday, May 28, 2022 10:33 AM
> > > To: FFmpeg development discussions and patches <ffmpeg-
> > > devel at ffmpeg.org>
> > > Subject: Re: [FFmpeg-devel] FATE Errors
> > >
> > > On Sat, May 28, 2022 at 10:12 AM Soft Works
> <softworkz at hotmail.com>
> > > wrote:
> > >
> > > >
> > > >
> > > > > -----Original Message-----
> > > > > From: ffmpeg-devel <ffmpeg-devel-bounces at ffmpeg.org> On
> Behalf Of
> > > > > Paul B Mahol
> > > > > Sent: Saturday, May 28, 2022 10:04 AM
> > > > > To: FFmpeg development discussions and patches <ffmpeg-
> > > > > devel at ffmpeg.org>
> > > > > Subject: Re: [FFmpeg-devel] FATE Errors
> > > > >
> > > > > On Sat, May 28, 2022 at 9:47 AM Soft Works
> > > <softworkz at hotmail.com>
> > > > > wrote:
> > > > >
> > > > > >
> > > > > >
> > > > > > > -----Original Message-----
> > > > > > > From: ffmpeg-devel <ffmpeg-devel-bounces at ffmpeg.org> On
> > > Behalf Of
> > > > > > > Paul B Mahol
> > > > > > > Sent: Saturday, May 28, 2022 9:35 AM
> > > > > > > To: FFmpeg development discussions and patches <ffmpeg-
> > > > > > > devel at ffmpeg.org>
> > > > > > > Subject: Re: [FFmpeg-devel] FATE Errors
> > > > > > >
> > > > > > > On Sat, May 28, 2022 at 9:26 AM Soft Works
> > > > > <softworkz at hotmail.com>
> > > > > > > wrote:
> > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > > > -----Original Message-----
> > > > > > > > > From: ffmpeg-devel <ffmpeg-devel-bounces at ffmpeg.org>
> On
> > > > > Behalf Of
> > > > > > > > > Paul B Mahol
> > > > > > > > > Sent: Saturday, May 28, 2022 9:26 AM
> > > > > > > > > To: FFmpeg development discussions and patches
> <ffmpeg-
> > > > > > > > > devel at ffmpeg.org>
> > > > > > > > > Subject: Re: [FFmpeg-devel] FATE Errors
> > > > > > > > >
> > > > > > > > > On Fri, May 27, 2022 at 11:11 PM Soft Works
> > > > > > > <softworkz at hotmail.com>
> > > > > > > > > wrote:
> > > > > > > > >
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > > > -----Original Message-----
> > > > > > > > > > > From: ffmpeg-devel <ffmpeg-devel-
> bounces at ffmpeg.org>
> > > On
> > > > > > > Behalf Of
> > > > > > > > > > > Andreas Rheinhardt
> > > > > > > > > > > Sent: Saturday, April 30, 2022 10:02 PM
> > > > > > > > > > > To: ffmpeg-devel at ffmpeg.org
> > > > > > > > > > > Subject: Re: [FFmpeg-devel] FATE Errors
> > > > > > > > > > >
> > > > > > > > > > > James Almer:
> > > > > > > > > > > >
> > > > > > > > > > > >
> > > > > > > > > > > > On 4/30/2022 4:06 PM, Andreas Rheinhardt wrote:
> > > > > > > > > > > >> Soft Works:
> > > > > > > > > > > >>> Hi,
> > > > > > > > > > > >>>
> > > > > > > > > > > >>> is it a known issue that the current head of
> the
> > > > > master
> > > > > > > > > branch
> > > > > > > > > > > has
> > > > > > > > > > > >>> FATE errors?
> > > > > > > > > > > >>>
> > > > > > > > > > > >>> I get the same locally as well as on the
> > > automated
> > > > > GitHub
> > > > > > > > > build.
> > > > > > > > > > > >>> One is this:
> > > > > > > > > > > >>>
> > > > > > > > > > > >>> --- ./tests/ref/vsynth/vsynth2-mpeg2-422
> 2022-04-
> > > 30
> > > > > > > > > > > >>> 14:23:44.330424058 +0000
> > > > > > > > > > > >>> +++ tests/data/fate/vsynth2-mpeg2-422 2022-
> 04-30
> > > > > > > > > > > 14:38:41.071678201
> > > > > > > > > > > >>> +0000
> > > > > > > > > > > >>> @@ -1,4 +1,4 @@
> > > > > > > > > > > >>> -b2fa9b73c3547191ecc01b8163abd4e5
> > > > > > > > > > > >>> *tests/data/fate/vsynth2-mpeg2-422.mpeg2video
> > > > > > > > > > > >>> -379164 tests/data/fate/vsynth2-mpeg2-
> > > 422.mpeg2video
> > > > > > > > > > > >>> -704f6a96f93c2409219bd48b74169041
> > > > > > > > > > > >>> *tests/data/fate/vsynth2-mpeg2-
> 422.out.rawvideo
> > > > > > > > > > > >>> -stddev: 4.17 PSNR: 35.73 MAXDIFF: 70 bytes:
> > > 7603200/
> > > > > > > 7603200
> > > > > > > > > > > >>> +8f6d565723ccf879ab2b5aa910b7ce21
> > > > > > > > > > > >>> *tests/data/fate/vsynth2-mpeg2-422.mpeg2video
> > > > > > > > > > > >>> +380544 tests/data/fate/vsynth2-mpeg2-
> > > 422.mpeg2video
> > > > > > > > > > > >>> +0797fddea4835687dedddebbbe98fa8f
> > > > > > > > > > > >>> *tests/data/fate/vsynth2-mpeg2-
> 422.out.rawvideo
> > > > > > > > > > > >>> +stddev: 4.16 PSNR: 35.73 MAXDIFF: 75 bytes:
> > > 7603200/
> > > > > > > 7603200
> > > > > > > > > > > >>> Test vsynth2-mpeg2-422 failed. Look at
> > > > > > > > > > > >>> tests/data/fate/vsynth2-mpeg2-422.err for
> > > details.
> > > > > > > > > > > >>> make: *** [tests/Makefile:277: fate-vsynth2-
> > > mpeg2-
> > > > > 422]
> > > > > > > Error
> > > > > > > > > 1
> > > > > > > > > > > >>>
> > > > > > > > > > > >>>
> > > > > > > > > > > >>> Is anybody seeing the same?
> > > > > > > > > > > >>>
> > > > > > > > > > > >>> Thanks,
> > > > > > > > > > > >>> sw
> > > > > > > > > > > >>
> > > > > > > > > > > >> http://fate.ffmpeg.org/ doesn't show recent
> > > > > regressions
> > > > > > > and
> > > > > > > > > FATE
> > > > > > > > > > > is fine
> > > > > > > > > > > >> for me locally. I recently made changes to
> FATE
> > > > > (namely
> > > > > > > the
> > > > > > > > > test
> > > > > > > > > > > >> requirements, vcodec.mak (where the vsynth-
> tests
> > > > > reside)
> > > > > > > among
> > > > > > > > > the
> > > > > > > > > > > files
> > > > > > > > > > > >> affected), so I am interested in whether the
> > > failing
> > > > > tests
> > > > > > > are
> > > > > > > > > > > >> concentrated on the files recently changed by
> me
> > > (it
> > > > > would
> > > > > > > > > > > obviously not
> > > > > > > > > > > >> haved pushed them if I knew them to cause
> issues;
> > > also
> > > > > > > > > patchwork
> > > > > > > > > > > was
> > > > > > > > > > > >> fine).
> > > > > > > > > > > >> Are these issues reproducible? Can you bisect
> > > them?
> > > > > > > > > > > >>
> > > > > > > > > > > >> - Andreas
> > > > > > > > > > > >
> > > > > > > > > > > > This may be the alignment issue introduced in
> lavfi
> > > in
> > > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > >
> > > > > > >
> > > > >
> > >
> https://git.videolan.org/?p=ffmpeg.git;a=commitdiff;h=17a59a634c39b00
> > > > > > > > > > > a680c6ebbaea58db95594d13d
> > > > > > > > > > > > assuming it was not fixed.
> > > > > > > > > > > > I think it only affected targets where
> > > > > av_cpu_max_align()
> > > > > > > > > returned
> > > > > > > > > > > 64.
> > > > > > > > > > >
> > > > > > > > > > > You are completely right: Just making
> > > av_cpu_max_align
> > > > > return
> > > > > > > 64
> > > > > > > > > > > allows
> > > > > > > > > > > to reproduce the issue. And it has nothing to do
> with
> > > my
> > > > > > > recent
> > > > > > > > > FATE
> > > > > > > > > > > patches (545e87f49dc9fa5b880e330fc4e1854df68cc0f1
> > > would
> > > > > be
> > > > > > > the
> > > > > > > > > only
> > > > > > > > > > > contender for changes).
> > > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > > I found that the FATE errors can be avoided by
> using
> > > > > > > > > >
> > > > > > > > > > ./configure --disable-avx512
> > > > > > > > > >
> > > > > > > > > > Yet, I hope it will be fixed at some time..
> > > > > > > > > >
> > > > > > > > >
> > > > > > > > > I hope you will post a fix for it soon.
> > > > > > > >
> > > > > > > > That depends on whether we want alignment values >
> 32bit..?
> > > > > > > >
> > > > > > >
> > > > > > > Nope, its about mpeg video encoder in lavc that does
> funny
> > > things
> > > > > > > from old
> > > > > > > days when linesize alignment was hardcoded.
> > > > > >
> > > > > > Adding this
> > > > > >
> > > > > >    align = FFMIN(align, 32);
> > > > > >
> > > > > > to ff_default_get_video_buffer2() resolves the issue.
> > > > > >
> > > > >
> > > > > Maybe it resolves it for you but that does not mean in any
> way
> > > that
> > > > > that is
> > > > > valid or correct solution.
> > > > >
> > > > >
> > > > > >
> > > > > > That's why I had written:
> > > > > >
> > > > > > > That depends on whether we want alignment values >
> 32bit..?
> > > > > >
> > > > >
> > > > > 32bit is 4 integers. so I failed to follow whatever you tried
> to
> > > > > communicate.
> > > >
> > > > Sorry for being unclear. What I meant to ask is this:
> > > >
> > > > In cases where av_cpu_max_align() returns something > 32 - for
> > > > example 64, do we have to use that max-align value of 64 for
> > > alignment
> > > > in video.c and framepool,c or could we just limit this to 32?
> > > > (hence the FFMIN).
> > > >
> > > > Limiting the alignment to 32 appears to fix the issue, meaning
> that
> > > > the three fate tests mentioned above are producing the expected
> ref
> > > > output and don't fail anymore.
> > > >
> > >
> > > That is hack.
> >
> > Only when 64bit alignment would be required or beneficial in terms
> of
> > performance (e.g. AVX512 operations).
> 
> 
> AVX512 operations are not using 64bit alignment.

They don't require, but I thought they would benefit from it.
You may know better than I do, I had just read this article:

https://www.intel.com/content/www/us/en/developer/articles/technical/data-alignment-to-assist-vectorization.html

> > Which is the reason for my question. If yes, then this would be
> > surely a bad idea.
> >
> 
> It is not Yes/No answer here.

Got you, thanks. I have found a very targeted workaround now.
Will submit in a moment.

sw