[Ffmpeg-devel-irc] ffmpeg-devel.log.20120828

Wed Aug 29 02:05:03 CEST 2012

[00:54] <llogan> docs state hh:mm:ss[.xxx] for -ss, -t, etc. i assume xxx is milliseconds? why not use ms instead of xxx if that is so?
[00:56] <ubitux> to show there is 3 digits?
[00:56] <ubitux> and yes sure it's milliseconds
[01:05] <llogan> i think people will know it is 3 digits with ms
[01:05] <ubitux> the rest of the pattern use the number of char for the number of expected digits
[01:06] <ubitux> i personally found this more obvious that way that using .ms
[01:06] <ubitux> than*
[01:06] <ubitux> but feel free to start a bikeshed thread on the ml ;)
[01:08] <llogan> lazy gauge is too high right now
[01:24] <saste> michaelni: where are documented the alignment requirements for libswscale?
[01:24] <saste> i mean, the data provided to sws_scale need need always to be aligned to X?
[01:25] <saste> what happens if not, will it crash?
[01:25] <saste> and which is the alignment size?
[01:25] <saste> in libavfilter we always assume 16, but where this value comes from?
[01:26] <ubitux> av_malloc()?
[01:26] <saste> the buffer is not always malloced
[01:26] <ubitux> really?
[01:27] <saste> for example i'm reading rawvideo from a file, and linesize alignment is 1
[01:27] <saste> if alignment is required, then I need to copy the data to an allocated buffer
[01:27] <saste> all this should be documented in swscale docs
[02:10] Action: michaelni wants to awnser saste
[02:10] <michaelni> but no saste here anymore ... so email it will be
[02:13] <ubitux> michaelni: i'm curious about the answer btw
[02:13] <ubitux> (and maybe some other devs?)
[02:15] <michaelni> well swscale should not crash from unaligned input
[02:16] <michaelni> it could be pretty slow though
[02:16] <michaelni> and there could be bugs that cause iz to crash
[02:18] <ubitux> ok
[02:31] <xxthink> does ffmpeg use reviewboard to review the code?
[02:31] <Daemon404> it uses a mailing list.
[03:26] <CIA-56> ffmpeg: 03Michael Niedermayer 07master * re367d90664 10ffmpeg/libavformat/aiffdec.c: 
[03:26] <CIA-56> ffmpeg: aiffdec: support QDM2
[03:26] <CIA-56> ffmpeg: Fixes ticket204
[03:26] <CIA-56> ffmpeg: Signed-off-by: Michael Niedermayer <michaelni at gmx.at>
[04:19] <CIA-56> ffmpeg: 03Michael Niedermayer 07master * rfa6c795c3f 10ffmpeg/libavcodec/allcodecs.c: 
[04:19] <CIA-56> ffmpeg: allcodecs: fix vda order
[04:19] <CIA-56> ffmpeg: Found-by: compn
[04:19] <CIA-56> ffmpeg: Signed-off-by: Michael Niedermayer <michaelni at gmx.at>
[04:19] <CIA-56> ffmpeg: 03Xidorn Quan 07master * r78cca74222 10ffmpeg/ (4 files in 2 dirs): 
[04:19] <CIA-56> ffmpeg: vdadec: implement vda decoder
[04:19] <CIA-56> ffmpeg: Signed-off-by: Michael Niedermayer <michaelni at gmx.at>
[04:19] <CIA-56> ffmpeg: 03Michael Niedermayer 07master * re3b87b355c 10ffmpeg/configure: 
[04:19] <CIA-56> ffmpeg: configure: fix vda order
[04:19] <CIA-56> ffmpeg: Found-by: compn
[04:19] <CIA-56> ffmpeg: Signed-off-by: Michael Niedermayer <michaelni at gmx.at>
[04:40] <ubitux> pengvado: in 566858a7, you added av_always_inline to an already inlined function (lowpass); it seems to cause some build issues
[04:41] <ubitux> gcc3 in particular
[04:41] <ubitux> (see http://fate.ffmpeg.org/)
[04:42] <ubitux> looks like the inline keyword should be dropped but well.
[05:32] <CIA-56> ffmpeg: 03Michael Niedermayer 07master * r4f92d31aa4 10ffmpeg/libavfilter/vf_hqdn3d.c: 
[05:32] <CIA-56> ffmpeg: vf_hqdn3d: remove duplicate inline
[05:32] <CIA-56> ffmpeg: This should fix compilation with gcc 3
[05:32] <CIA-56> ffmpeg: Found-by: ubitux
[05:32] <CIA-56> ffmpeg: Signed-off-by: Michael Niedermayer <michaelni at gmx.at>
[08:56] <randomguy123> hi ffmpeg. Is there any way to mark raw yuv input to be interlaced? (I need to do using avformat api).
[08:57] <randomguy123> there was a thread about that on ffmpeg-users, and it was aborted in the middle
[08:58] <randomguy123> also, I have interlaced mpeg2 video. If I do ffmpeg -i ntsc.mpg ntsc.mp4 then resulting video is also interlaced (that is, playback is messed up on PC), but in media info it's listed as progressive;
[08:59] <randomguy123> basically, ffmpeg either should report it as interlaced or deinterlace it
[09:46] <ubitux> oh it's fun
[09:46] <ubitux> i was able to trigger the race for cama2_vtc_b in valgrind, but only once
[09:47] <ubitux> maybe some particular mem align or something
[09:48] <ubitux> the output was flooded, but here is a sample: http://pastie.org/4602183
[09:49] <ubitux> it looks similar to what Daemon404 spotted with the intel thing
[09:49] <ubitux> some stuff in cabac
[09:50] <ubitux> Skyler: maybe you would be interested? :p
[09:53] <Skyler> ?
[09:54] <ubitux> iirc you said you were kind of familiar with cabac
[09:56] <ubitux> (you're aware there is a race in the cama2_vtc_b test, right?)
[09:57] <Skyler> no, I really am not
[09:57] <Skyler> since it's only in mvd, I highly doubt it's related to cabac, it sounds far more likely related to motion vectors
[09:57] <ubitux> ok well then at times, the fate threads instance fail on two particular test
[09:58] <ubitux> and the h264 cama2_vtc_b is the most common of the two
[09:58] <Skyler> slice or frame?
[09:59] <ubitux> thread type is set to frame+slice :p
[09:59] <Skyler> what does that mean?  which gets used?
[09:59] <ubitux> dunno, how can i check?
[09:59] <Skyler> um...   see which one causes the errors?   it's kind of impossible to debug without at least knowing what kind of threading is being used
[10:00] <ubitux> the error is pretty rare
[10:00] <ubitux> and hardly reproducible
[10:00] <Skyler> erm, do a quick printf to see which actually gets used?
[10:00] <ubitux> i just happen to trigger it by chance
[10:03] <ubitux> looks like it's frame
[10:06] <Skyler> sorry, about to go to sleep, might look tomorrow
[10:06] <ubitux> sure no worry, thanks for your concern :)
[12:27] <durandal_1707> is vlc givin money for new audio decoders?
[12:45] <saste> durandal_1707: http://wiki.videolan.org/Bounties
[12:45] <saste> but he's not on irc anymore
[15:09] <CIA-56> ffmpeg: 03Derek Buitenhuis 07master * r61e262f6d7 10ffmpeg/libavcodec/vp56.h: (log message trimmed)
[15:09] <CIA-56> ffmpeg: vp56: Don't use DECLARE_ALIGN on a typedef name
[15:09] <CIA-56> ffmpeg: Instead, use it on the first member, since by definition, if
[15:09] <CIA-56> ffmpeg: any member is aligned, the whole struct must be, in order to
[15:09] <CIA-56> ffmpeg: maintain that alignment.
[15:09] <CIA-56> ffmpeg: Fixes compilation with some finicky compilers, like a mix of libclang/msvc
[15:09] <CIA-56> ffmpeg: Idea for fix from Måns Rullgård.
[15:16] <Black0ut> Hello
[15:20] <CIA-56> ffmpeg: 03Nicolas George 07master * rfb6fb0dedb 10ffmpeg/tools/make_chlayout_test: 
[15:20] <CIA-56> ffmpeg: tools: add make_chlayout_test perl script.
[15:20] <CIA-56> ffmpeg: This script uses the flite source to produce files
[15:20] <CIA-56> ffmpeg: suitable to test channels order and layout.
[15:20] <CIA-56> ffmpeg: 03Nicolas George 07master * r786dadc014 10ffmpeg/ (4 files in 2 dirs): (log message trimmed)
[15:20] <CIA-56> ffmpeg: cmdutils: add -layouts option.
[15:20] <CIA-56> ffmpeg: Extract of the output:
[15:20] <CIA-56> ffmpeg: | Individual channels:
[15:20] <CIA-56> ffmpeg: | NAME DESCRIPTION
[15:20] <CIA-56> ffmpeg: | FL front left
[15:20] <CIA-56> ffmpeg: | FR front right
[16:01] <Compn> hey, ciabot is back to longer commit logs :)
[16:01] <Compn> yay
[16:33] <CIA-56> ffmpeg: 03Martin Storsjö 07master * rb42be2d56a 10ffmpeg/avprobe.c: 
[16:33] <CIA-56> ffmpeg: avprobe: Get rid of ugly casts in the options table
[16:33] <CIA-56> ffmpeg: This also makes sure the function pointers are stored via the same
[16:33] <CIA-56> ffmpeg: union member as they are accessed via.
[16:33] <CIA-56> ffmpeg: Signed-off-by: Martin Storsjö <martin at martin.st>
[16:33] <CIA-56> ffmpeg: 03Kostya Shishkov 07master * r49df339f4b 10ffmpeg/libavcodec/mss1.c: 
[16:33] <CIA-56> ffmpeg: mss1: merge decode_intra() and decode_inter()
[16:33] <CIA-56> ffmpeg: They have most of their code in common.
[16:33] <CIA-56> ffmpeg: 03Martin Storsjö 07master * rb441a4517b 10ffmpeg/libavfilter/vf_hqdn3d.c: 
[16:33] <CIA-56> ffmpeg: vf_hqdn3d: Remove a duplicate inline declaration
[16:33] <CIA-56> ffmpeg: Compilation seems to fail on GCC 3.4 due to this duplicate
[16:33] <CIA-56> ffmpeg: declaration.
[16:33] <CIA-56> ffmpeg: Signed-off-by: Martin Storsjö <martin at martin.st>
[16:33] <CIA-56> ffmpeg: Signed-off-by: Mans Rullgard <mans at mansr.com>
[16:33] <CIA-56> ffmpeg: 03Diego Biurrun 07master * r5549854335 10ffmpeg/libavcodec/ (Makefile celp_math.c celp_math.h lsp.c): celp_math: Move ff_cos() to the only place it is used
[16:33] <CIA-56> ffmpeg: 03Michael Niedermayer 07master * r416d2f7a12 10ffmpeg/: (log message trimmed)
[16:33] <CIA-56> ffmpeg: Merge remote-tracking branch 'qatar/master'
[16:33] <CIA-56> ffmpeg: * qatar/master:
[16:33] <CIA-56> ffmpeg:  vc1: export some functions
[16:33] <CIA-56> ffmpeg:  configure: use HOSTCC_C/O in check_host_cc
[16:33] <CIA-56> ffmpeg:  configure: use AS_O setting in check_as
[16:33] <CIA-56> ffmpeg:  configure: use LD_O setting in check_ld()
[16:33] <CIA-56> ffmpeg: 03Brad Smith 07master * rd5683aba5a 10ffmpeg/configure: 
[16:33] <CIA-56> ffmpeg: configure: Fix shared library creation for OpenBSD
[16:34] <CIA-56> ffmpeg: OpenBSD does not expect or require symlinked shared libraries with
[16:34] <CIA-56> ffmpeg: just the major version in the name.
[16:34] <CIA-56> ffmpeg: Signed-off-by: Diego Biurrun <diego at biurrun.de>
[16:34] <CIA-56> ffmpeg: 03Derek Buitenhuis 07master * r5395d882b4 10ffmpeg/libavcodec/vp56.h: (log message trimmed)
[16:34] <CIA-56> ffmpeg: vp56: Don't use DECLARE_ALIGN on a typedef name
[16:34] <CIA-56> ffmpeg: Instead, use it on the first member, since by definition, if
[16:34] <CIA-56> ffmpeg: any member is aligned, the whole struct must be, in order to
[16:34] <CIA-56> ffmpeg: maintain that alignment.
[16:34] <CIA-56> ffmpeg: Fixes compilation with some finicky compilers.
[16:34] <CIA-56> ffmpeg: Idea for fix from Måns Rullgård.
[16:34] <CIA-56> (9 lines omitted)
[17:59] <CIA-56> ffmpeg: 03Ronald S. Bultje 07master * rba53720280 10ffmpeg/ (6 files in 3 dirs): 
[17:59] <CIA-56> ffmpeg: lavu: add snprintf(), vsnprint() and strtod() replacements for MS runtime.
[17:59] <CIA-56> ffmpeg: Signed-off-by: Michael Niedermayer <michaelni at gmx.at>
[18:26] <saste> michaelni: regarding smartblur, I think i'll commit the version with GPL license, since the patch should be ready from the technical POV
[18:27] <saste> let me know if you want to relicense it, I think both me and nyuhu don't know if you agreed about relicensing it or not
[18:28] <CIA-56> ffmpeg: 03Xidorn Quan 07master * r1ea77a52c0 10ffmpeg/MAINTAINERS: 
[18:28] <CIA-56> ffmpeg: add myself as vda decoder maintainer
[18:28] <CIA-56> ffmpeg: Signed-off-by: Michael Niedermayer <michaelni at gmx.at>
[18:29] <michaelni> saste, i dont remember being asked to relicense smartblur, i just remember agreeing for vf_hue
[18:30] <michaelni> but i really think its wiser if we leave things GPL and wait for some company who wants it LGPL and ask for some donation in exchange
[18:30] <saste> michaelni: fine with me
[18:30] <nyuhu> good
[18:30] <Compn> they can sponsor it for $1k :)
[18:31] <saste> nyuhu: please update the license and repost it
[18:31] <nyuhu> okay
[18:35] <ubitux> oh av_realloc doesn't give an aligned buffer, that sucks :)
[18:36] <michaelni> its hard to implement an efficient aligned realloc() when the underlaying libc realloc doesnt align
[18:36] <ubitux> yeah i figured that out
[18:36] <ubitux> i wonder if that's not causing other issues than the one nicolas spotted
[18:37] <Compn> start building against dalias' project...
[18:37] <Compn> uclibc whatnot
[18:37] <Daemon404> i build ffmpeg statically with uclibc
[18:37] <Daemon404> fwiw
[18:37] <Daemon404> regularily
[18:37] <nyuhu> saste : then I should add gpl to the dependencies in the configure right ?
[18:37] <Daemon404> i.e. static uclibc
[18:37] <ubitux> nyuhu: yes
[18:37] <michaelni> Daemon404, you could add a uclibc fate client
[18:38] <Compn> er musl > http://www.etalabs.net/musl/
[18:38] <Compn> whatnot
[18:38] <Daemon404> michaelni, perhaps
[18:38] <Daemon404> i dont have infinite hardware tho
[18:38] <Daemon404> ;)
[18:39] <Daemon404> http://japland.org/tools/linux/poky-uclibc-x86_64-x86_64-static-toolchain-1.2%2bsnapshot-20120507.tar.xz
[18:39] <Daemon404> my old toolchain is sitll floating around
[18:47] <Daemon404> looks like make fate-h264 with inspxe-cl is gonna take close to an entire day
[18:47] <Daemon404> slow much?
[18:55] <CIA-56> ffmpeg: 03Nicolas George 07master * r18217bb0f5 10ffmpeg/ffmpeg.c: 
[18:55] <CIA-56> ffmpeg: ffmpeg: report max time of unfinished streams in stats.
[18:55] <CIA-56> ffmpeg: Eliminating finished streams avoids the progress stopping
[18:55] <CIA-56> ffmpeg: with the first stream.
[18:55] <CIA-56> ffmpeg: Using the max instead of the min avoids the progress stopping
[18:55] <CIA-56> ffmpeg: with gaps in sparse streams (subtitles).
[18:55] <CIA-56> ffmpeg: Negligible change for normal circumstances.
[19:49] <philipl> ubitux. michaelni: Any opinions on my last patch?
[19:49] <philipl> It's my previous patch with a new description :-) But it seems a reasonable thing to do to unbreak subtitle transcoding out of mkv.
[19:54] CTCP re:  from CIA-56 (CIA-56!cia at 198.71.88.9) to #ffmpeg-devel
[19:54] <CIA-56> ffmpeg: 03Jemy Tran 07master * r316afee708 10ffmpeg/ (7 files in 3 dirs): 
[19:54] <CIA-56> ffmpeg: lavfi: add smartblur filter
[19:54] <CIA-56> ffmpeg: This is a port of the MPlayer smartblur filter (libmpcodecs/vf_smartblur.c)
[19:54] <CIA-56> ffmpeg: by Michael Niedermayer.
[19:54] <CIA-56> ffmpeg: Signed-off-by: Stefano Sabatini <stefasab at gmail.com>
[20:00] <michaelni> philipl, ill leave subtitle stuff to ubitux, if he is happy with it, so am i
[20:01] Action: ubitux is almost always happy
[20:01] <ubitux> philipl: give me a few hours, i'll look at it tonight
[20:02] <philipl> Thanks.
[20:10] <CIA-56> ffmpeg: 03Stefano Sabatini 07master * r74419fcf45 10ffmpeg/doc/examples/ (Makefile scaling_video.c): 
[20:10] <CIA-56> ffmpeg: examples: add scaling_video example
[20:10] <CIA-56> ffmpeg: This example should be useful to show the basic functionality of the
[20:10] <CIA-56> ffmpeg: libswscale API.
[20:10] <CIA-56> ffmpeg: More advanced features (scaling options etc., colorspace tweaking) may be
[20:10] <CIA-56> ffmpeg: added later.
[20:10] <CIA-56> ffmpeg: 03Stefano Sabatini 07master * r2cbffc6706 10ffmpeg/doc/examples/Makefile: 
[20:10] <CIA-56> ffmpeg: examples/Makefile: remove duplicated -f switch
[20:10] <CIA-56> ffmpeg: $(RM) already includes that flag.
[20:10] <CIA-56> ffmpeg: Spotted-by: ubitux
[20:16] <ubitux> saste: btw i think you forgot to add the scaling example binary in the .gitignore
[20:16] <ubitux> sorry to realize that now
[20:21] <saste> ubitux: I always ignore .gitignore
[20:21] <saste> feel free to push
[20:29] <CIA-56> ffmpeg: 03Clément BSsch 07master * rc149586884 10ffmpeg/.gitignore: gitignore: add scaling_video example.
[20:39] <ubitux> philipl: what's the difference between duration and lace_duration?
[20:39] <ubitux> it looks weird to me that we use lace_duration for the convergence_duration and duration for the pkt duration
[20:39] <ubitux> (btw the indent is 4 spaces, not 2)
[20:40] <nevcairiel> i would guess duration is the package duration, and lace_duration the duration of the individual laces, if lacing is used
[20:40] <nevcairiel> doubtful subs use lacing
[20:42] <philipl> ubitux: Hmm. That's actually a change that came in after I originally wrote mine.
[20:42] <philipl> Previously the matroskadec code was using duration
[20:42] <philipl> Let's try and find out what change did that
[20:43] <nevcairiel> btw, in your patch you write pkt->duration unconditionally, you might as well remove the conditional block 3 lines down then
[20:44] <philipl> at a glance I think I need to update my diff to use lace_duration.
[20:44] <ubitux> saste: any idea why the build error is not triggered by FATE?
[20:45] <philipl> Yeah, it comes from Moritz's fix for 1270.
[20:45] <philipl> So I need to update my diff to reflect that.
[20:45] <saste> ubitux: because people usually compile with optimizations on
[20:45] <ubitux> there is at least one --disable-optimization box
[20:46] <ubitux> (i count 2 of them)
[20:46] <ubitux> and it builds fine here with --disable-optimizations
[20:46] <saste> ubitux: static?
[20:46] <ubitux> yes, just --disable-optimizations
[20:47] <saste> umh... maybe --disable-mmx is needed as well
[20:51] <ubitux> saste: indeed, and with the --disable-mmx flag only as well
[20:51] <ubitux> i wonder if i should add a few boxes for this
[20:52] <ubitux> it's strange that --disable-optimizations doesn't do --disable-mmx
[20:52] <nevcairiel> --disable-asm?
[20:52] <ubitux> we have some --disable-asm boxes and they build fine as well
[20:54] <Daemon404> does this involve any inline asm?
[20:54] <Daemon404> or just yasm?
[20:59] <ubitux> there are some inline asm, but the definition seems to exist even if without inline asm
[21:01] <ubitux> btw, we could also nag diego to fix it since he's playing with that stuff at the moment
[21:01] <ubitux> but i'm afraid he will not appreciate if i ask something to him again :p
[21:03] <ubitux> ah seems Daemon404 did it
[21:04] <saste> ubitux: well I'll wait a few days before pushing my possibly naive fix, possibly diego will come with a better patch
[21:05] <ubitux> saste: if you push it, please specify that's only a problem with --disable-mmx
[21:05] <ubitux> ok
[21:05] <ubitux> saste: it seems --disable-optimizations is also broken on libav side
[21:05] <ubitux> so well...
[21:05] <ubitux> :)
[21:55] <nyuhu> saste : concerning the dynamic expression evaluation in filters, is it fine to use the current_pts attribute in AVFilterLink to get the current time ?
[21:56] <saste> nyuhu: uhm i didn't add that field so can't say out of my mind
[21:56] <saste> nyuhu: I always used cur_buf->pts
[21:57] <nyuhu> ok, Im fine with it
[21:57] <saste> nyuhu: "out of my mind" => from the top of my head
[21:57] <saste> i don't think the two expressions are equivalent ;-)
[21:57] <nyuhu> huhu
[22:16] <Daemon404> an amazing 9 h264 fate tests have been run in 5 hrs
[22:16] <Daemon404> nope. not slow at all.
[22:36] <CIA-56> ffmpeg: 03Michael Niedermayer 07master * r6c06b566d2 10ffmpeg/libavformat/rtpenc_xiph.c: 
[22:36] <CIA-56> ffmpeg: rtpenc_xiph: switch to av_assert
[22:36] <CIA-56> ffmpeg: Signed-off-by: Michael Niedermayer <michaelni at gmx.at>
[23:56] <CIA-56> ffmpeg: 03Nicolas George 07master * ra9b1a151d2 10ffmpeg/ffmpeg.c: 
[23:56] <CIA-56> ffmpeg: ffmpeg: count finished streams for last stats line.
[23:56] <CIA-56> ffmpeg: Otherwise, all streams are finished and the time is nonsensical.
[00:00] --- Wed Aug 29 2012