[Ffmpeg-devel-irc] ffmpeg-devel.log.20120708

[burek]

[00:27] <CIA-41> ffmpeg: 03Hendrik Leppkes 07master * rd039b6074c 10ffmpeg/libavcodec/aacdec.c: 
[00:27] <CIA-41> ffmpeg: latm: fix initialization on some streams when no extradata is available
[00:27] <CIA-41> ffmpeg: Signed-off-by: Michael Niedermayer <michaelni at gmx.at>
[00:27] <CIA-41> ffmpeg: 03Michael Niedermayer 07master * racdf9a6ca5 10ffmpeg/libavcodec/aacdec.c: 
[00:27] <CIA-41> ffmpeg: aacdec: correct "audio config changed" warning message
[00:27] <CIA-41> ffmpeg: Idea-by: Reimar
[00:27] <CIA-41> ffmpeg: Signed-off-by: Michael Niedermayer <michaelni at gmx.at>
[00:36] <CIA-41> ffmpeg: 03Paul B Mahol 07master * r2d42008757 10ffmpeg/libavcodec/sanm.c: 
[00:36] <CIA-41> ffmpeg: sanm: make use of dsputil
[00:36] <CIA-41> ffmpeg: About 27% faster decoding.
[00:36] <CIA-41> ffmpeg: Signed-off-by: Paul B Mahol <onemda at gmail.com>
[01:24] <CIA-41> ffmpeg: 03Stefano Sabatini 07master * rabb0a9982f 10ffmpeg/libavfilter/ (8 files): 
[01:24] <CIA-41> ffmpeg: lavfi: do not pass opaque field to init functions
[01:24] <CIA-41> ffmpeg: Fix signature mismatch warnings.
[03:50] <CIA-41> ffmpeg: 03Michael Niedermayer 07master * re145afb4b0 10ffmpeg/libavfilter/vf_tinterlace.c: 
[03:50] <CIA-41> ffmpeg: vf_tinterlace: use av_assert
[03:50] <CIA-41> ffmpeg: Signed-off-by: Michael Niedermayer <michaelni at gmx.at>
[04:18] <durandal_1707> michaelni: so in dr1 case linesize of frame should not be modified?
[04:21] <durandal_1707> heh yop is changing linesize too, so i guess that is cause of breakdown...
[04:22] <durandal_1707> this should be probably documented
[09:26] <jesk> does it make a difference to move mkv-with-h264-and-ac3 to mp4-with-h264-aac in one step
[09:26] <jesk> or to move the h264 first in one file and then transcode the ac3 to aac in one file and then mux them together to the mp4
[09:31] <jesk> is one over the other more "reliable"?
[14:14] <highgod> Hi,Does anyone do any reaserch on mpeg4-part2 using dxva2 using ffmpeg?
[14:15] <highgod> I want to add mpeg4-part2 hardware accelerator to ffmpeg.But I am a newcommer on this field
[14:27] <michaelni> highgod, best look at the existing hwaccel code
[14:27] <michaelni> to see how API/ABI works
[14:27] <michaelni> then just implement your hwaccel like others where
[14:36] <highgod> OK,I know.Thank you.I want to know that witch kind of hwaccel is most similar with mpeg4-part2?
[14:37] <nevcairiel> there is already a vaapi implementation for mpeg4, using that to extend it to dxva should not be all that hard
[14:41] <highgod> Ok,Thanks.
[14:44] <Compn> highgod : vlc has done some dxva2 stuff with ffmpeg, you might want to ask them 
[14:44] <Compn> #videolan that is
[14:47] <highgod> OK,Thanks
[14:58] <highgod> Hi,I want to ask a question again.I have add hwaccel to decode wmv3 and VC1 in ffmpeg.exe.But the duration of the lock operation is different between two video format.The duration of lock operation for wmv3 is 12ms,but VC1 just cost 2ms.
[14:59] <highgod> the lock operation function is called like this,"IDirect3DSurface9_LockRect( d3d, &lock, NULL,D3DLOCK_READONLY)"
[15:04] <nevcairiel> When locking takes longer it means the decoder is still writing to that surface
[15:06] <nevcairiel> in my experience, its better to store 4 surfaces in a buffer and then lock them with a 4-frame delay, that way you get the best pipelining of the hardware decoder
[15:06] <Compn> highgod : are you going to post your patches soon? :)
[15:10] <highgod> Yes,maybe,But for wmv3,the problem has not been fixed
[15:13] <highgod> nevcairiel,the lock operation is called after the hwaccel.Do you mean that the write operation is still running after the decode function finish?
[15:13] <nevcairiel> yes
[15:13] <nevcairiel> the hardware works async
[15:18] <highgod> Oh,I never think of that.Thanks.I will try it.But for VC1,the speed is very fast.Is wmv3 cost more time than VC1 on writing operation?
[15:19] <nevcairiel> in my experience i got around 20% more speed when using the surface buffering
[15:19] <nevcairiel> went from 50 to consistently > 60 on my old gpu
[15:30] <highgod> oh,4 surfaces.I want to add them to FFmpeg.exe.If I do like this,Maybe I will modify the decode-encode process of ffmpeg.
[17:07] <CIA-41> ffmpeg: 03Nicolas George 07master * rbba819bd8f 10ffmpeg/tests/fate/flac.mak: fate: FLAC tests require SAMPLES.
[17:10] <CIA-41> ffmpeg: 03Stefano Sabatini 07master * r7f08bfae6e 10ffmpeg/ (doc/muxers.texi libavformat/segment.c libavformat/version.h): 
[17:10] <CIA-41> ffmpeg: lavf/segment: add segment_list_type option, extend format for the segment list file
[17:10] <CIA-41> ffmpeg: Add list extended format which specifies in the list file the start and
[17:10] <CIA-41> ffmpeg: ending time for each segment. This is required to make it available this
[17:10] <CIA-41> ffmpeg: information to external tools, avoiding the need to perform file analysis
[17:10] <CIA-41> ffmpeg: in the output segments.
[17:10] <CIA-41> ffmpeg: 03Stefano Sabatini 07master * r0a7b8effbb 10ffmpeg/doc/muxers.texi: doc/muxers: sort options for the segment muxer
[17:10] <CIA-41> ffmpeg: 03Stefano Sabatini 07master * rb010d9b586 10ffmpeg/libavformat/nutenc.c: lavf/nutenc: provide meaningful error message and error code in case of invalid pts
[17:41] <CIA-41> ffmpeg: 03Stefano Sabatini 07master * r8dc31f8da8 10ffmpeg/ (doc/muxers.texi libavformat/segment.c libavformat/version.h): 
[17:41] <CIA-41> ffmpeg: lavf/segment: make use of av_parse_time() when parsing the -segment_time value
[17:41] <CIA-41> ffmpeg: Increase flexibility/consistency.
[17:41] <CIA-41> ffmpeg: Also rename recording_time field to time, for enhanced
[17:41] <CIA-41> ffmpeg: consistency/readability.
[17:41] <CIA-41> ffmpeg: 03Stefano Sabatini 07master * r3f345290dc 10ffmpeg/libavformat/segment.c: lavf/segment: sort options
[18:00] <CIA-41> ffmpeg: 03Michael Niedermayer 07master * r99622f6678 10ffmpeg/libavcodec/mpegaudioenc.c: 
[18:00] <CIA-41> ffmpeg: mpegaudioenc: use av_assert
[18:00] <CIA-41> ffmpeg: Signed-off-by: Michael Niedermayer <michaelni at gmx.at>
[18:00] <CIA-41> ffmpeg: 03Michael Niedermayer 07master * ra6a085a928 10ffmpeg/libavdevice/v4l2.c: 
[18:00] <CIA-41> ffmpeg: v4l2: use av_assert
[18:00] <CIA-41> ffmpeg: Signed-off-by: Michael Niedermayer <michaelni at gmx.at>
[19:00] <ubitux> michaelni: what is this function doing? http://git.videolan.org/?p=ffmpeg.git;a=blob;f=libpostproc/postprocess_internal.h;hb=HEAD#l78 ; it looks like it's just a return (a&256) ? !(a>>31) : 0
[19:00] <ubitux> but i don't understand the meaning
[19:01] <burek> sign of a number?
[19:02] <ubitux> yes, a>>31 is the bit sign
[19:03] <ubitux> but the function looks weird, it's not really clipping
[19:03] <Tjoppen> it seems to return -1 or -2
[19:03] <Tjoppen> git blame and see if there's an explanation in the log
[19:05] <ubitux> 134eb1e50483bc503468796b5a1f31509398e785
[19:09] <ubitux> maybe the point was to have something like av_clip_uint8(), but it doesn't do the same thing
[19:15] <michaelni> should be uint8 clip
[19:16] <michaelni> when is it not ?
[19:16] <jesk> so yes or no? :P
[19:17] <jesk> pull backlog
[19:17] <ubitux> michaelni: well, first the if won't trigger 1024 for instance
[19:17] <jesk> ubitux, do you have an idea? you always have!
[19:17] <ubitux> then it is just doing a "not" on the sign
[19:17] <michaelni> ubitux, where is it feeded with 1024 ?
[19:18] <ubitux> i don't know jesk 
[19:18] <jesk> really? :-(
[19:18] <ubitux> michaelni: oh no idea, the function might expect some very specific input then
[19:19] <ubitux> also, something like 300 will end up to 0, right?
[19:19] <ubitux> i mean 1 sorry
[19:19] <ubitux> shouldn't it be 0xff instead?
[19:20] <ubitux> mmh it returns -1 sorry
[19:22] <ubitux> mmh and the implicit convert will lead to 0xff
[19:27] <jesk> Invalid EBML number size tag 0x0f at pos
[19:28] <jesk> its a nightmare
[19:48] <michaelni> jesk, EBML is a nightmare ?
[19:56] <jesk> no converting my movie archive 
[19:56] <jesk> :P
[19:57] <jesk> for every file i have to do scientific research
[19:59] <jesk> michaelni, do you know what EBML is?
[20:00] <CIA-41> ffmpeg: 03Nicolas George 07master * r65e65e3ce2 10ffmpeg/libavfilter/vf_settb.c: settb: cosmetic: reindent vf_settb structure.
[20:00] <CIA-41> ffmpeg: 03Nicolas George 07master * rba856c0be5 10ffmpeg/ (6 files in 3 dirs): lavfi: implement asettb filter.
[20:00] <CIA-41> ffmpeg: 03Nicolas George 07master * r299387ebf4 10ffmpeg/libavfilter/ (Makefile vf_settb.c): vf_settb: rename the file as f_settb.
[20:01] <jesk> ah, it's mkv specific meta language format
[20:01] <jesk> seems to work good
[20:04] <iive> jesk: binary xml
[20:08] <dan_> so are ac3 audio packets always 32ms?
[20:09] <jesk> iive: yes, thanks
[20:09] <dan_> because in a stream i'm working with, there's a case where it seems one is  split into 1000/1880 instead of the usual 2880 90khz clock ticks O_o
[20:10] <dan_> there's also a whole bunch with PTS difference of 2879 instead of 2880... what's up with that?
[22:30] <CIA-41> ffmpeg: 03Michael Niedermayer 07master * rd8ce478c43 10ffmpeg/libavformat/utils.c: 
[22:30] <CIA-41> ffmpeg: lavf: limit "Adjusting PTS forward" code to video streams
[22:30] <CIA-41> ffmpeg: It can end up slightly messing up a series of valid timestamps
[22:30] <CIA-41> ffmpeg: after a invalid one.
[22:30] <CIA-41> ffmpeg: Signed-off-by: Michael Niedermayer <michaelni at gmx.at>
[22:39] <CIA-41> ffmpeg: 03Anton Khirnov 07master * rdba5b06ead 10ffmpeg/libavformat/flacdec.c: 
[22:39] <CIA-41> ffmpeg: flacdec: don't create an attached picture stream until we have all information.
[22:39] <CIA-41> ffmpeg: This way we don't end with an invalid stream if parsing the picture
[22:39] <CIA-41> ffmpeg: fails.
[22:39] <CIA-41> ffmpeg: 03Anton Khirnov 07master * rb7d3a9a015 10ffmpeg/libavformat/flacdec.c: 
[22:39] <CIA-41> ffmpeg: flacdec: be less strict when parsing attached pictures.
[22:39] <CIA-41> ffmpeg: Only return an error if memory allocation fails or error recognition is
[22:39] <CIA-41> ffmpeg: set to explode. Otherwise just print an error message and continue
[22:39] <CIA-41> ffmpeg: reading the file.
[22:39] <CIA-41> ffmpeg: 03Ronald S. Bultje 07master * r07b287020c 10ffmpeg/ (configure libavutil/x86/timer.h): x86/timer: implement an intrinsic-based version for rdtsc (AV_READ_TIME).
[22:39] <CIA-41> ffmpeg: 03Anton Khirnov 07master * r25b51b2c44 10ffmpeg/libavformat/id3v2.c: id3v2: add a mimetype for bmp pictures.
[22:39] <CIA-41> ffmpeg: 03Ronald S. Bultje 07master * r66a297975d 10ffmpeg/cmdutils.c: (log message trimmed)
[22:39] <CIA-41> ffmpeg: cmutils: include shellapi.h on Win32 (for CommandLineToArgvW).
[22:39] <CIA-41> ffmpeg: This is required for CommandLineToArgvW. Normally, shellapi.h is included
[22:39] <CIA-41> ffmpeg: implicitly by windows.h, but if we define WIN32_LEAN_AND_MEAN (or some of
[22:39] <CIA-41> ffmpeg: the other earlier headers have included windows.h with that option),
[22:39] <CIA-41> ffmpeg: windows.h doesn't include this one.
[22:39] <CIA-41> ffmpeg: Thus explicitly include the headers we really need, for clarity and
[22:39] <CIA-41> ffmpeg: 03Michael Niedermayer 07master * rca1f2b3e10 10ffmpeg/: (log message trimmed)
[22:39] <CIA-41> ffmpeg: Merge remote-tracking branch 'qatar/master'
[22:39] <CIA-41> ffmpeg: * qatar/master:
[22:39] <CIA-41> ffmpeg:  cmutils: include shellapi.h on Win32 (for CommandLineToArgvW).
[22:39] <CIA-41> ffmpeg:  x86/timer: implement an intrinsic-based version for rdtsc (AV_READ_TIME).
[22:39] <CIA-41> ffmpeg:  id3v2: add a mimetype for bmp pictures.
[22:39] <CIA-41> ffmpeg:  flacdec: be less strict when parsing attached pictures.
[22:50] <ktdreyer> Hi, at RPM Fusion we are working on getting all our multimedia apps to build against ffmpeg 0.11. Xbmc fails with an error: https://bugzilla.rpmfusion.org/2396 and I was hoping an ffmpeg dev could point me in the right direction?
[22:51] <ktdreyer> I found the ffmpeg commit where AVFormatContext changed, but I need some guidance for how to patch XBMC to match this.
[23:06] <CIA-41> ffmpeg: 03Paul B Mahol 07master * r26fa84105b 10ffmpeg/libavformat/paf.c: 
[23:06] <CIA-41> ffmpeg: paf: properly flag video packets
[23:06] <CIA-41> ffmpeg: Signed-off-by: Paul B Mahol <onemda at gmail.com>
[23:06] <CIA-41> ffmpeg: 03Paul B Mahol 07master * r30659a2233 10ffmpeg/libavformat/smush.c: 
[23:06] <CIA-41> ffmpeg: smush: properly flag audio packets
[23:06] <CIA-41> ffmpeg: Signed-off-by: Paul B Mahol <onemda at gmail.com>
[23:20] <michaelni> ktdreyer, why is xbmc using these these internal variables ?
[23:21] <JEEB> I think they officially still have their own copy of ffmpeg in their code repo? So getting it to build with system libraries can lead to fun stuff? I could be incorrect tho
[23:23] <nevcairiel> yeah they apply custom patches, last i checked
[23:25] <nevcairiel> their ffmpeg is also generally a year old or so =P
[23:25] <durandal_1707> nevcairiel: i do not think so
[23:25] <michaelni> ktdreyer, looking at XBMC sources i see #ifs aLIBAVFORMAT_VERSION_INT < AV_VERSION_INT(54,0,0) round the failing lines
[23:26] <michaelni> so this maybe is working in latest xbmc, but iam just guessing here
[23:27] <nevcairiel> https://github.com/xbmc/xbmc/tree/master/lib/ffmpeg/patches .. quite the list of patches, actually :)
[23:28] <ubitux> wow
[23:28] <nevcairiel> some of those are official patches they backported to their ancient ffmpeg version, though
[23:29] <JEEB> I think some people just fork xbmc and try to make it use the libraries "normally"
[23:29] <JEEB> marcan did that at least
[23:32] <durandal_1707> michaelni: i uploaded something in incoming/paf 
[23:32] <Compn> ktdreyer : btw, we are open to applying xbmc patches, if you need some internal function we can reinstitute it
[23:32] <Compn> also we can review patches if you clean and submit them against master git :)
[23:33] <durandal_1707> Compn: he is working for RPM Fusion and not for xbmc
[23:33] <Compn> oh 
[23:33] <Compn> so distro hacks on top of project hacks :)
[23:33] <Compn> fun
[23:34] <Compn> ktdreyer / durandal_1707: well its in your best interest to get xbmc to send their patches to us :)
[23:46] <michaelni> ktdreyer, maybe simply exporting ff_read_frame_flush() from ffmpeg (add entry to libavformat.v) and using that instead av_... in xbmc would make rthe whole  DllAvFormat.c unneeded
[23:47] <michaelni> but someone from xbmc should probably comment on this idea
[23:48] <michaelni> ... if it works we certainly could export ff_read_frame_flush() from official ffmpeg
[00:00] --- Mon Jul  9 2012