[Ffmpeg-devel-irc] ffmpeg-devel.log.20120706

burek burek021 at gmail.com
Sat Jul 7 02:05:03 CEST 2012


[00:11] <saste> ubitux: init2 for the filters which require it
[01:12] <CIA-41> ffmpeg: 03Stefano Sabatini 07master * r3e3e414fe4 10ffmpeg/libavfilter/sink_buffer.c: 
[01:12] <CIA-41> ffmpeg: lavfi/buffersink: fix doxy, remove "video" reference
[01:12] <CIA-41> ffmpeg: The file contains the code for the audio buffersink as well.
[01:58] <durandal_1707> looks like shared build is broken
[02:24] <CIA-41> ffmpeg: 03Paul B Mahol 07master * rc2cbe58cd4 10ffmpeg/libavcodec/paf.c: 
[02:24] <CIA-41> ffmpeg: paf: add forgotten CODEC_CAP_DR1
[02:24] <CIA-41> ffmpeg: Signed-off-by: Paul B Mahol <onemda at gmail.com>
[02:26] <CIA-41> ffmpeg: 03Michael Niedermayer 07master * r636dbe2678 10ffmpeg/tests/fate/ac3.mak: 
[02:26] <CIA-41> ffmpeg: fate: increase fuzzyness for ac3 encode for 3dnow
[02:26] <CIA-41> ffmpeg: Signed-off-by: Michael Niedermayer <michaelni at gmx.at>
[02:26] <CIA-41> ffmpeg: 03Michael Niedermayer 07master * r9862cbd7a4 10ffmpeg/libavcodec/avcodec.h: 
[02:26] <CIA-41> ffmpeg: add side data to indicate the amount of samples to skip at the begin and end
[02:26] <CIA-41> ffmpeg: Reviewed-by: Nicolas George <nicolas.george at normalesup.org>
[02:26] <CIA-41> ffmpeg: Reveiwed-by: Reimar Döffinger <Reimar.Doeffinger at gmx.de>
[02:26] <CIA-41> ffmpeg: Signed-off-by: Michael Niedermayer <michaelni at gmx.at>
[02:26] <CIA-41> ffmpeg: 03Michael Niedermayer 07master * rec08676232 10ffmpeg/libavutil/blowfish.c: 
[02:26] <CIA-41> ffmpeg: Blowfish: restructure/simplify code a bit.
[02:26] <CIA-41> ffmpeg: Very slightly faster (2% or so)
[02:26] <CIA-41> ffmpeg: Signed-off-by: Michael Niedermayer <michaelni at gmx.at>
[02:53] <CIA-41> ffmpeg: 03Nicolas George 07master * r8293a21a9c 10ffmpeg/libavcodec/libavcodec.v: 
[02:53] <CIA-41> ffmpeg: lavc: fix shared build after dnxhd merge.
[02:53] <CIA-41> ffmpeg: Signed-off-by: Nicolas George <nicolas.george at normalesup.org>
[02:53] <CIA-41> ffmpeg: Signed-off-by: Michael Niedermayer <michaelni at gmx.at>
[03:04] <durandal_1707> tiff both decoder and encoder is awfull
[03:05] <Compn> awful quality or code or ?
[03:05] <Compn> its not lossless ?
[03:06] <durandal_1707> awful code
[03:06] <durandal_1707> it is buggy too...
[03:06] <Compn> kostya added support for a ton of tiff variants
[03:07] <durandal_1707> support for various pix fmts is hack
[03:09] <durandal_1707> doing AV_WB/L when tiff supports both le and be
[03:12] <Compn> theres lots of old decoders that need love in ffmpeg
[03:12] <Compn> and old demuxers
[03:13] <CIA-41> ffmpeg: 03Paul B Mahol 07master * r8f239412d8 10ffmpeg/libavcodec/tiff.c: 
[03:13] <CIA-41> ffmpeg: tiffdec: PIX_FMT_GRAY16LE support
[03:13] <CIA-41> ffmpeg: Signed-off-by: Paul B Mahol <onemda at gmail.com>
[03:16] <durandal_1707> michaelni: DR1 works for referenced frames?
[03:17] <michaelni> sure
[03:17] <durandal_1707> michaelni: well thing again when reviewing next patch.... :)
[03:17] <durandal_1707> s/thing/think
[03:20] <durandal_1707> michaelni: libav seems to not have this flac failure on its machines with valgrind enabled
[03:21] <michaelni> hmm
[03:26] <durandal_1707> these inline asm is causing clang to spam screen each time i compile ffmpeg
[03:29] <durandal_1707> and big failure of tif decoder it that it does not support yuv formats
[04:04] <CIA-41> ffmpeg: 03Michael Niedermayer 07master * r894319e010 10ffmpeg/libavcodec/lpc.c: 
[04:04] <CIA-41> ffmpeg: lpc: use default number of passes when lpc_passes is invalid
[04:04] <CIA-41> ffmpeg: Should fix valgrind failure (uninitialized vars)
[04:04] <CIA-41> ffmpeg: Signed-off-by: Michael Niedermayer <michaelni at gmx.at>
[04:06] <durandal_1707> tiff encoder is hopefully fixed now
[04:20] <CIA-41> ffmpeg: 03Paul B Mahol 07master * r6936111d66 10ffmpeg/libavcodec/bmv.c: 
[04:20] <CIA-41> ffmpeg: bmv: add CODEC_CAP_DR1
[04:20] <CIA-41> ffmpeg: Signed-off-by: Paul B Mahol <onemda at gmail.com>
[04:50] <durandal_1707> michaelni: of course 16bit tiff works i tested it with feh (uses imlib)
[06:15] <Jumpyshoes> michaelni: if you have time http://ffmpeg.org/pipermail/ffmpeg-devel/2012-July/127180.html
[06:34] <ohsix> http://awesomium.com/ is this one of those situations where i should expect to find the source for libav/ffmpeg somewhere?
[06:36] <Compn> you try contacting them ? :p
[06:36] <Compn> most of these frontends just use stock binaries
[06:36] <Compn> :P
[06:38] <Compn> http://code.google.com/p/browsernode/source/browse/trunk/awesomium/?r=4
[06:38] <Compn> is that what you want ohsix ?
[06:41] <ohsix> didn't even google, was poking around their website
[06:42] <ohsix> that name sort of looks like a coincidence, maybe :o
[06:42] <ohsix> or it uses awesomeium
[06:42] <Compn> could be
[06:42] <Compn> i honestly dont know
[06:42] <Compn> half the internet uses ffmpeg now
[06:43] <ohsix> "powered by the Chromium port of the WebKit rendering engine via the awesomium browser embedding library."
[06:43] <Compn> the other half pays china to code up 'proprietary' tools which ... use ffmpeg :P
[06:43] <ohsix> i wanna figure out how to sell .net bindings to open source stuff for megabucks
[06:52] <ohsix> there's a certain irony in just embedding the stuff into a 2d surface too, the browser can do tons of crap now
[08:37] <michaelni> Hi Jumpyshoes, 
[08:38] <michaelni> thats a pretty big patch :)
[08:39] <michaelni> did you benchmark if it has some negative speed effect on the non threaded or frame threaded case ?
[08:40] <michaelni> the addition of pause_hint() could be factored in a seperate patch and probably it could be used in other places too
[08:42] <Jumpyshoes> michaelni: around 0.5% decrease on non-threaded and frame threaded. the majority of the speed decrease is because of the first patch, as decoding an mb row is no longer inlined
[08:52] <michaelni> can that speedloss be avoided easy with some inline or always_inline ?
[08:52] <Jumpyshoes> michaelni: no, the function needs to not be inlined since it is called by execute2
[08:56] <michaelni> yes for the slices case but could some alternative codepath be added or would this be messy?
[08:57] <Jumpyshoes> michaelni: i can only think of messy options, which requires code duplication
[08:59] <michaelni> hmm, then its maybe not worth it 
[09:05] <highgod> Hi all.I am a new comer.I have a question.How can I know the two pkts are the same?Can I use pos flag?
[09:06] <michaelni> Jumpyshoes, did you check its faster with ff_thread_sleep() than without ?
[09:07] <Jumpyshoes> michaelni: yes, it's faster with, on both windows and linux
[09:07] <michaelni> ok
[09:07] <Jumpyshoes> michaelni: on the HD clips i tested. i will test low-res clips and per reimar's suggestion
[09:08] <michaelni> ok, also i suspect reimar will do a more complete review than what he posted :)
[09:08] <michaelni> i also should/will once iam more awake ;)
[09:08] <Jumpyshoes> sure
[09:08] <Jumpyshoes> might take me some time to get my hands on good low-res clips
[09:10] <michaelni> highgod, in what use case do you end with possible equal?
[09:10] <michaelni> ... packets
[09:11] <michaelni> pos compare will work for some demuxers but not all
[09:13] <highgod> I add a dxva2 accelerator for ffmpeg in my test project,and I use a thread to decode,the primary thread just read the decoded frame.
[09:14] <highgod> but sometimes,the "best_effort_timestamp" between two neighbouring frame are the same 
[09:15] <highgod> So I doubt maybe the pkt I pass to the subthread is wrong
[09:16] <michaelni> is the best_effort_timestamp in lets say ffplay when playing the same video also the same for these frames ?
[09:16] <highgod> Maybe pass the same pkt to the subthread.But sometimes the code run correct
[09:17] <michaelni> also make sure you call av_dup_packet() before putting a packet into some que
[09:19] <highgod> Thanks,Let me have a look
[09:41] <ubitux> do we support nasm?
[09:43] <ubitux> (just to know if i should add a FATE box for this)
[09:46] <ubitux> well i'm adding it, we'll see.
[10:05] <ubitux> seems not to work :)
[10:05] <ubitux> (http://fate.ffmpeg.org/log.cgi?time=20120706075408&log=compile&slot=x86_64-archlinux-gcc-nasm)
[10:14] <ubitux> saste: "Write a frequency domain filter using the FFT implementation in libavcodec"; what do you mean? if you're talking about the showspectrum filter, i'm working on it :p
[10:17] <saste> ubitux: no it's a separate thing
[10:17] <ubitux> so just transform the samples?
[10:17] <saste> you apply the FQ transform, apply a filter on it, and convert back to space domain
[10:18] <saste> it is used to remove some kind of artifacts, or to remove some frequencies
[10:18] <saste> so there are many interesting applications which can employ the FD transform
[10:19] <ubitux> so DFT, low pass, and IDFT?
[10:19] <saste> yes
[10:19] <ubitux> ok, maybe mentioning a "low pass filter using FFT" would make more sense
[10:21] <ohsix> you can do a low pass with an add :]
[10:21] <ubitux> saste: btw, in other ideas, there is https://ffmpeg.org/trac/ffmpeg/ticket/1430
[10:22] <ubitux> and in filters than "make sense to port", an ivtc would be nice, as well as improving FPS one to do motion estimation, but those might be pretty hard to do
[10:22] <saste> ubitux: I'll just add a link to all the lavfi tickets on trac
[10:23] <ubitux> oh and yet another one: metadata injection
[10:23] <saste> or feel free to add it yourself
[10:23] <ubitux> yeah i really need to ask for an account
[10:26] <ohsix> adding all that stuff won't get people writing their filters for it instead of avisynth :p
[10:28] <ubitux> well, that's a first step
[10:30] <ohsix> i kind of wish i could have used ffmpeg last time i had to encode something
[10:46] <ubitux> saste: btw, the introduction looks a bit aggressive to me
[10:46] <ubitux> "Before posting to the FFmpeg mailing list, make sure you read and understand our netiquette guidelines, especially avoid top-posting and thread-hijacking (note that if you don't understand one of those terms, make sure to have understood them before writing your first post). You should also familiar with the programs diff, patch and git. You have to learn these basics on your own before you start, we will not teach them to ...
[10:46] <ubitux> ... you during the application process."
[10:47] <ubitux> (btw s/familiar/be familiar/)
[10:47] <ubitux> i think we could propose help on IRC for this
[10:48] <ubitux> this text could actually scare ppl and prevent them from asking that kind of question
[10:51] <saste> ubitux: yes i had the same impression, i just copy&pasted the text from the last year
[10:51] <saste> but i'm going to "soften" it
[10:51] <ubitux> thanks :)
[11:09] <ubitux> < ohsix> i kind of wish i could have used ffmpeg last time i had to encode something // what stopped you?
[11:10] <saste> ubitux: init_opaque?
[11:10] <ubitux> saste: sure ok
[11:11] <saste> also, i want to keep args, since it is possible to support *both* interfaces in a filter
[11:11] <ubitux> i really dislike the funcname_<num> for an alternate version, it makes think it's a new/better version and deprecate the previous one
[11:11] <ubitux> saste: ok ok
[11:11] <saste> for example opaque interface for easy generic binary data passing, and args for commandline/scripting testing
[11:40] <ubitux> saste: nit+: mention that it is an "alternative" init function in the doxy please :)
[11:40] <saste> ok
[11:41] <ubitux> thank you
[11:51] <CIA-41> ffmpeg: 03Georg Lippitsch 07master * r60270eb482 10ffmpeg/doc/indevs.texi: 
[11:51] <CIA-41> ffmpeg: doc/indevs/iec61883: document requirements for iec61883 device
[11:51] <CIA-41> ffmpeg: Signed-off-by: Stefano Sabatini <stefasab at gmail.com>
[11:51] <CIA-41> ffmpeg: 03Michael Niedermayer 07master * re0d6b9094c 10ffmpeg/configure: 
[11:51] <CIA-41> ffmpeg: configure: cleanup libiec61883 code, allow enabling via command line
[11:51] <CIA-41> ffmpeg: Signed-off-by: Michael Niedermayer <michaelni at gmx.at>
[11:51] <CIA-41> ffmpeg: Signed-off-by: Stefano Sabatini <stefasab at gmail.com>
[11:55] <ubitux> i added some libs to the shared box, it might fail, but don't worry too much
[11:56] <ubitux> btw, i added a debug box, valgrind+threads, and the failing nasm mentioned this morning
[14:54] <ubitux> huh, what happened to my shared box.
[16:24] <bash`> hi all
[16:25] <bash`> any idea why I get "error: URL_WRONLY undeclared" even if I include libavformat/avformat.h?
[16:25] <bash`>  this is the source: https://github.com/pculture/miro/blob/Miro-5.0/tv/linux/miro-segmenter.c
[16:25] <bash`> I'm using ffmpeg 0.11
[17:08] <michaelni> ubitux, problems with your fate client ?
[17:08] <ubitux> no it's ok i solved it
[17:08] <ubitux> the configure failed but removed the entry on fate.ffmpeg so i was surprise
[17:08] <ubitux> +d
[17:08] <ubitux> michaelni: btw, should i keep the NASM box?
[17:09] <michaelni> my cubox extries disappeared too :(
[17:10] <michaelni> ubitux, i think there where 2 people or so that used nasm,
[17:10] <michaelni> they might be interrested to fix it
[17:10] <michaelni> one was a OS2 user IIRC the other possibly reimar
[17:11] <michaelni> would have to check to be sure i dont mix it up
[17:11] <ubitux> ah the korean guy
[17:12] <ubitux> also the yasm 0.6 box has a lot of warnings
[17:12] <ubitux> some unrecognized cpu or sth iirc
[17:13] <CIA-41> ffmpeg: 03Lou Logan 07master * rb22ecbc6a5 10ffmpeg/ (12 files in 5 dirs): 
[17:13] <CIA-41> ffmpeg: cosmetics: various spelling fixes
[17:13] <CIA-41> ffmpeg: Signed-off-by: Michael Niedermayer <michaelni at gmx.at>
[17:44] <CIA-41> ffmpeg: 03Pavel Koshevoy 07master * r665100c2f1 10ffmpeg/libswscale/ppc/yuv2rgb_altivec.c: (log message trimmed)
[17:44] <CIA-41> ffmpeg: Fix destination row striding in altivec_yuv2_bgr24.
[17:44] <CIA-41> ffmpeg: Misaligned row artifacts showed up when a 624x352 frame was converted
[17:44] <CIA-41> ffmpeg: to BGR24 format. When advancing to the next row the destination linesize
[17:44] <CIA-41> ffmpeg: was added to the last output pointer position which was not linesize aligned,
[17:44] <CIA-41> ffmpeg: resulting in a distorted picture.
[17:44] <CIA-41> ffmpeg: Signed-off-by: Pavel Koshevoy <pavel at apple.aragog.com>
[18:49] <CIA-41> ffmpeg: 03Nicolas George 07master * r5997285d3a 10ffmpeg/ffmpeg.c: 
[18:49] <CIA-41> ffmpeg: ffmpeg: poll filters even after -t limit.
[18:49] <CIA-41> ffmpeg: If not, frames can still arrive to the sink and accumulate.
[18:49] <CIA-41> ffmpeg: The frames past recording time will be ignored in do_*_out.
[20:44] <ubitux> is there any source filter re-using a buffer ref?
[20:45] <ubitux> (and modifying it each time)
[21:57] <CIA-41> ffmpeg: 03Paul B Mahol 07master * r82c4abbfe1 10ffmpeg/libavcodec/tiffenc.c: 
[21:57] <CIA-41> ffmpeg: tiffenc: use av_pix_fmt_descriptors to set bpp_tab[]
[21:57] <CIA-41> ffmpeg: Signed-off-by: Paul B Mahol <onemda at gmail.com>
[21:57] <CIA-41> ffmpeg: 03Paul B Mahol 07master * re100a259be 10ffmpeg/libavcodec/tiffenc.c: 
[21:57] <CIA-41> ffmpeg: tiffenc: simplify calculation of bpp_tab_size
[21:57] <CIA-41> ffmpeg: Signed-off-by: Paul B Mahol <onemda at gmail.com>
[21:57] <CIA-41> ffmpeg: 03Paul B Mahol 07master * r1806858fdd 10ffmpeg/libavcodec/tiffenc.c: 
[21:57] <CIA-41> ffmpeg: tiffenc: merge common code
[21:57] <CIA-41> ffmpeg: Signed-off-by: Paul B Mahol <onemda at gmail.com>
[21:57] <CIA-41> ffmpeg: 03Paul B Mahol 07master * ra6da14ec3c 10ffmpeg/libavcodec/tiffenc.c: 
[21:57] <CIA-41> ffmpeg: tiffenc: PIX_FMT_GRAY8A support
[21:57] <CIA-41> ffmpeg: Signed-off-by: Paul B Mahol <onemda at gmail.com>
[21:57] <CIA-41> ffmpeg: 03Paul B Mahol 07master * r146d29c744 10ffmpeg/libavcodec/tiffenc.c: 
[21:57] <CIA-41> ffmpeg: tiffenc: use av_get_bits_per_pixel() to get bpp
[21:57] <CIA-41> ffmpeg: Signed-off-by: Paul B Mahol <onemda at gmail.com>
[21:57] <CIA-41> ffmpeg: 03Paul B Mahol 07master * rdf548c59f7 10ffmpeg/libavcodec/tiffenc.c: 
[21:57] <CIA-41> ffmpeg: tiffenc: PIX_FMT_GRAY16LE support
[21:57] <CIA-41> ffmpeg: Signed-off-by: Paul B Mahol <onemda at gmail.com>
[22:33] <RobertNagy> what does a negative linesize mean?
[22:33] <Skyler> The frame's ordered backwards in memory.
[22:34] <RobertNagy> backwards?
[22:35] <durandal_1707> top to bottom - bottom to top
[22:49] <RobertNagy> I'm getting corrupted frames directly from avcodec_decode_video2 when reading from a dshow input
[22:50] <RobertNagy> is tehre anything that differs a dshow input from regular file or ip input?
[22:50] <burek> youtube uses ffmpeg? :))) cool :)
[22:51] <Skyler> Most websites that do any sort of video processing do
[22:52] <nevcairiel> that reminds me, my company wants some sort of mini-youtube on their intranet for employees to upload videos, i'm looking forward to build that for them, finally a fun project ;)
[22:53] <ubitux> i want to do that for myself since a while now
[22:53] <ubitux> with the ffprobe json, scene cut filter etc :)
[22:53] <Compn> there are some youtube clone projects
[22:53] <Compn> so it is possible :)
[00:00] --- Sat Jul  7 2012


More information about the Ffmpeg-devel-irc mailing list