[Ffmpeg-devel-irc] ffmpeg-devel.log.20160601

[burek]

[00:17:34 CEST] <Compn> BBB : funny, i dont remember stabbing anyone
[00:18:12 CEST] <Compn> BBB : if you are offended that i remembered an objection to a server admin... i'm sorry i guess
[00:18:27 CEST] <BBB> hes offline, thats the definition of behind someones back"
[00:18:36 CEST] <Compn> kierank is on irc
[00:18:39 CEST] <Compn> offline no clue
[00:19:02 CEST] <BBB> UK probably asleep but who knows
[00:19:05 CEST] <BBB> anyway, be nice
[00:19:19 CEST] <BBB> and more importantly, be respectful
[00:19:21 CEST] <Compn> implying i was being mean
[00:19:23 CEST] <BBB> kierank does good stuff for this project
[00:19:36 CEST] <BBB> except for kierank who objected to it .. because reasons
[00:19:41 CEST] <nevcairiel> bringing the objection up with a comment like "because reasons" already expresses a certain degree of disagreement and is just not necessary
[00:19:45 CEST] <Compn> i cant remember the reasons, i'm sorry
[00:19:46 CEST] <BBB> that is dismissive, disrespectful
[00:19:57 CEST] <BBB> yeah yeah right
[00:20:01 CEST] <BBB> please just be respectful
[00:20:12 CEST] <BBB> thats all Im asking
[00:20:15 CEST] <Compn> your comment "yeah yeah right" is dismissive and disrespectful
[00:20:19 CEST] <Compn> but thats ok , because i dont mind
[00:20:47 CEST] <Compn> i like kierank even if we disagree on things
[00:20:56 CEST] <Compn> he is a valuable part of the project
[00:21:50 CEST] <kierank> What did I do
[00:23:36 CEST] <Compn> no idea
[00:23:39 CEST] <Compn> bbb took offense 
[05:25:36 CEST] <prelude2004c> hey guys.. so i am back and finally got the vpdpau working to compare the outcome of nvenc vs nvtranscoder and right now its already a problem but clearly something is wrong :) 
[05:30:00 CEST] <prelude2004c> so far there is a quality difference
[05:30:04 CEST] <prelude2004c> def. noticible
[05:38:39 CEST] <prelude2004c> ok so far good results
[05:38:56 CEST] <prelude2004c> just have to figure out why the audio/video sync is off
[06:11:01 CEST] <jya> hi.. how do I run a particular fate test? don't want to run all of them (takes too long)
[06:44:55 CEST] <rcombs> jya: make fate-<testname>
[06:45:32 CEST] <jya> rcombs: thanks
[09:13:03 CEST] <ubitux> ff_h264_decode_slice_header() is a nice function :))
[09:25:14 CEST] <nevcairiel> so are you getting it done ubitux? :)
[09:25:33 CEST] <ubitux> yeah, at least the merging conflict parts
[09:25:50 CEST] <ubitux> the last chunks are a bit tricky though
[09:26:56 CEST] <ubitux> i've additionally updated vdpau.c and videotoolbox.c not present in libav, but i'm not sure i'll be able to test easily
[10:13:09 CEST] <nevcairiel> let me know if you need any help, these patches are certainly one of the harder ones to merge
[10:14:29 CEST] <BtbN> prelude2004c, passing the exact same encoder arguments will yield in the exact same quality. So you are just passing the wrong options.
[10:43:38 CEST] <durandal_1707> michaelni: have you read my mail?
[11:17:02 CEST] <ubitux> so is current_sps_id still needed?
[11:17:46 CEST] <nevcairiel> dont think so ubitux
[11:18:06 CEST] <nevcairiel> its only used in one spot, and that check should be replaced by something else
[11:18:34 CEST] <ubitux> it's supposed to be redundant with the check just above
[11:18:52 CEST] <ubitux> can we drop it before the merge of the commit?
[11:19:33 CEST] <ubitux> according to 99b12357f47efd7da070e29d888e9e7646824667
[11:19:39 CEST] <ubitux> "The current_sps_id is not removed as it used in security related code."
[11:20:52 CEST] <nevcairiel> that makes no sense to me, the two checks should be identical
[11:21:24 CEST] <ubitux> michaelni ^
[11:22:00 CEST] <nevcairiel> it doesnt really block anything, so you could keep it for now and consider dropping it later
[11:22:08 CEST] <nevcairiel> its impact on the merge should be trivial at best
[11:35:16 CEST] <durandal_1707> anyone knows how init_vlc works?
[11:37:24 CEST] <iive> what exactly do you need? I've looked at it long ago.
[11:39:52 CEST] <kierank> yes
[12:01:26 CEST] <durandal_1707> I have codecs which uses vlc, two tables, one is uint8t of 256 and another one uint16_t I need to figure out how it was generated 
[12:05:34 CEST] <prelude2004c> DHE , no luck on the wrap issue
[12:05:38 CEST] <prelude2004c> still failed
[12:08:28 CEST] <ubitux> https://github.com/so-fancy/diff-so-fancy why no builtin :(
[12:13:11 CEST] <DHE> prelude2004c: ah geez...
[12:14:24 CEST] <DHE> prelude2004c: did you get a message saying "Stream timestamps wrapping" (requires default loglevel)
[12:54:35 CEST] <BtbN> Why does the qsvdec register itself as hwaccel? Doesn't it behave like a normal decoder?
[12:55:05 CEST] <BtbN> hm, it registers as both
[12:55:18 CEST] <BtbN> but doesn't define any functions for the hwaccel
[12:55:22 CEST] <nevcairiel> it needs a hwaccel for the qsv pixel format
[12:55:32 CEST] <nevcairiel> its just for weird api needs
[12:55:48 CEST] <BtbN> So just defining the empty AVHWAccel struct does that?
[12:56:12 CEST] <nevcairiel> it allows you to use get_format to select a pixel format then
[12:56:16 CEST] <nevcairiel> for decoders
[12:57:36 CEST] <BtbN> strange stuff
[12:57:55 CEST] <BtbN> Do I need that for a potential cuvid decoder, or is it fine if I just allways output FMT_CUDA frames?
[12:58:17 CEST] <BtbN> And leave pulling it to RAM to the automatic download filter
[12:58:20 CEST] <iive> why not just call get_format() directly?
[13:01:37 CEST] <BtbN> Well, the QSV decoder predates the automatic hw upload/download stuff in lavf
[13:02:06 CEST] <BtbN> So I'll just hardcode it to output CUDA frames, and the filter can then take care of downloading it. Anything else would just duplicate code anyway.
[13:02:12 CEST] <nevcairiel> decoders that output nothing but an opaque format is probably not a good idea
[13:02:43 CEST] <nevcairiel> not everything goes through ffmpeg.c and lavfi, making that a mandatory requirement to use it seems silly
[13:03:06 CEST] <BtbN> I'd have to duplicate the entirety of vf_hwdownload_cuda for that.
[13:06:42 CEST] <BtbN> hm, wait. The actual upload/download stuff is in lavu. So it could potentialy be used.
[13:27:48 CEST] <Franciman> how can I filter AVStreams of audio?
[13:39:23 CEST] <durandal_1707> kierank: I'm attempting to write decoder for sheervideo
[13:45:21 CEST] <durandal_1707> basic stuff is already on wiki, just need to figure how to create vlc tables from their static tables
[13:49:12 CEST] <ubitux> michaelni: do you remember what edabbfbaf2c9bc478edc6169cd7714da68ba70b5 might fix?
[13:57:49 CEST] <ubitux> i might be reintroducing an issue in the merge related to this
[13:58:55 CEST] <michaelni> ubitux, the sps code was full of race conditions, i dont remember that particular one
[13:59:53 CEST] <michaelni> is there some branch or patch i could test before its pushed ?
[14:01:18 CEST] <ubitux> sure, it's not working anyway yet
[14:01:41 CEST] <ubitux> it's in the branch merge-libav on my github
[14:02:10 CEST] <ubitux> michaelni: so i'm looking at merging this chunk currently: http://sprunge.us/GMFC
[14:02:38 CEST] <ubitux> (full commit hash is 3176217)
[14:03:34 CEST] <ubitux> and since edabbfba removed the copy to *sps copy, i'm wondering if it's going to have an influence on this diff
[14:03:52 CEST] <ubitux> (i have many other questions on the whole diff, but taking it step by step)
[14:04:13 CEST] <michaelni> i dont want to be involved in the merges
[14:05:02 CEST] <ubitux> well you wanted to test at least, and that's your code, but sure ok
[14:05:24 CEST] <michaelni> i can & will test but i dont really remember the code
[14:05:38 CEST] <michaelni> just that it worked fine and that i couldnt cleanly fix it because of the merges
[14:06:50 CEST] <michaelni> as a clean fix would have requires more substantial changes
[14:07:12 CEST] <michaelni> ubitux, src/libavcodec/dxva2_h264.c:174:46: error: pps undeclared (first use in this function)
[14:07:41 CEST] <ubitux> michaelni: will fix, it was a blind for these files
[14:07:59 CEST] <ubitux> it's not usable yet btw, it will simply not work
[14:08:17 CEST] <michaelni> then testing it isnt easy ;)
[14:09:05 CEST] <ubitux> michaelni: compilation should be fixed
[14:10:25 CEST] <michaelni> ubitux, src/libavcodec/dxva2_h264.c:288:51: error: request for member init_qp in something not a structure or union
[14:11:03 CEST] <ubitux> fixed
[14:16:41 CEST] <michaelni> ubitux, build works, fate doesnt
[14:16:54 CEST] <michaelni> make: *** [fate-filter-aresample] Error 139
[14:17:18 CEST] <ubitux> yeah, i know
[14:17:19 CEST] <BtbN> If cuvid tells me "// Presentation timestamp (10MHz clock)", is "cupkt.timestamp = av_rescale_q(avpkt->pts, avctx->time_base, (AVRational){1, 10000000});" the correct interpretation of that?
[14:18:31 CEST] <nevcairiel> i still think that trying to support cuvid at this point in time is just wasted effort, nvidia hasnt improved it in years, and it has collected a whole bunch of bugs over the years
[14:19:06 CEST] <michaelni> ubitux, ok, please ping me once you need/want me to test 
[14:19:31 CEST] <michaelni> ubitux, you can also ask me questions but i mist likely dont remember when its years ago
[14:19:35 CEST] <ubitux> michaelni: yes, thank you
[14:19:37 CEST] <ubitux> sure
[14:20:06 CEST] <ubitux> i'm reviewing the whole diff to find the most obvious issues, then will proceed to fix the bugs it introduced
[14:20:19 CEST] <ubitux> i took note of every sensible part and the related tickets for testing
[14:20:30 CEST] <ubitux> most likely i did some stuff wrong
[14:20:53 CEST] <BtbN> nevcairiel, it claims to support HEVC10, so I'd guess it got at least some kind of update?
[14:21:23 CEST] <BtbN> They never updated the doc aparently, but the headers have it all.
[14:21:28 CEST] <nevcairiel> it can claim all sorts of things, last time i checked the only surface format it defined was nv12, unless they updated it recently
[14:21:56 CEST] <nevcairiel> (and nv12 is 8-bit only)
[14:22:16 CEST] <BtbN> Even if not, Nvidia seeing cuvid being used could be some kind of motiviation to fix it up a bit.
[14:22:48 CEST] <nevcairiel> i've supported it for years, and been in contact with cuvid to fix the issues, they didnt care
[14:22:57 CEST] <nevcairiel> in contact with nvidia*
[14:24:06 CEST] <BtbN> hm, i think I have the parameters in the wrong order? it's av_rescale(a, b, c) -> a*b/c, so it's want (pts, target_timebase, source_timebase)?
[14:24:51 CEST] <nevcairiel> note that av_rescale and av_rescale_q are a bit confusing, as it can appear like they want the params in a different order
[14:25:08 CEST] <BtbN> yeah, that's exactly what I'm confused about right now
[14:26:11 CEST] <nevcairiel> av_rescale_q wants (timestamp, source, target), for av_rescale, its a bit weirder
[14:26:25 CEST] <BtbN> I'm using the q variant
[14:26:52 CEST] <BtbN> So "cupkt.timestamp = av_rescale_q(avpkt->pts, avctx->time_base, (AVRational){1, 10000000});" should be correct then for a 10MHz clock
[14:36:59 CEST] <KGB> [13FFV1] 15michaelni pushed 1 new commit to 06master: 02https://git.io/vrNOl
[14:36:59 CEST] <KGB> 13FFV1/06master 14c5e8bc1 15Jérôme Martinez: Remove slice_count reference...
[14:39:02 CEST] <kierank> Any questions about dnxhd, avid guy on stage
[14:41:20 CEST] <kierank> or dnxhr
[15:01:13 CEST] <cone-788> ffmpeg 03Piotr Bandurski 07master:4c7d3e827c80: avformat/isom: add MagicYUV FourCCs
[15:05:09 CEST] <durandal_1707> lol I got funny mail from Papua New Guinea
[15:06:40 CEST] <BtbN> they need you to receive and send on a few million dollars?
[15:10:15 CEST] <durandal_1707> no, develop codec to fight piracy
[15:12:56 CEST] <durandal_1707> hardware only for DVD it must not play on ms Windows
[15:20:09 CEST] <BBB> kierank: ?
[15:20:14 CEST] <BBB> kierank: some conference?
[15:20:17 CEST] <kierank> yeah about hdr
[15:20:21 CEST] <BBB> which one?
[15:20:34 CEST] <BBB> I feel so ignorant sometimes :D
[15:22:53 CEST] <BtbN> Hm, the presence of my cuvid decoder somehow breaks the matroska demuxer, it can't find a pixel format, and returns -1 oO
[15:23:31 CEST] <nevcairiel> make sure its in the list far below all other d ecoders
[15:25:30 CEST] <BtbN> But why does the matroska demuxer care?
[15:28:15 CEST] <durandal_1707> spammer
[15:30:59 CEST] <kierank> BBB: some ebu hdr workshop
[15:32:59 CEST] <ubitux> nevcairiel: https://github.com/ubitux/FFmpeg/commit/ac6dfcd34337adca99990c1bd8e41f18922ec103
[15:33:06 CEST] <ubitux> can you check if you see something obviously wrong?
[15:33:46 CEST] <ubitux> (it breaks the decode somewhere, didn't figure out where yet)
[15:35:46 CEST] <BtbN> Hm, I'm not sure I understand how this works. The pix_fmts in struct AVCodec are the supported output pixel formats, right?
[15:38:15 CEST] <BtbN> So I put only the CUDA format in there, and aparently that breaks stuff, as the demuxers check the decoder for support of the format in the streams. Hm
[15:40:34 CEST] <iive> huh?
[15:40:48 CEST] <iive> that kind of sounds wrong.
[15:42:24 CEST] <BtbN> https://github.com/BtbN/FFmpeg/blob/master/libavcodec/cuvid.c#L250 leads to this when trying to use it: https://bpaste.net/show/bcd5e6e873e8
[15:43:48 CEST] <BtbN> So I guess it compares the pixel format the container reports against the ones the decoder lists as supported?
[16:34:25 CEST] <cone-788> ffmpeg 03Neil Birkbeck 07master:785038c92cc7: lavu: Adding ARIB STD-B67 (hybrid log-gamma) enum value and transfer function.
[16:34:26 CEST] <cone-788> ffmpeg 03Michael Niedermayer 07master:77ba4a784302: library.mak: Put -Umain back for tests
[16:42:45 CEST] <prelude2004c> Hello everyone.. DHE, no i didn't have loglevel up at all
[16:53:59 CEST] <prelude2004c> DHE, do you think what you did at all affects audio offset ? i changed things up to nvenc like you guys wanted and vdpau and it is working.. but my audio is like drifting around and not in sync with video
[16:54:06 CEST] <prelude2004c> code is simple now .
[16:54:31 CEST] <prelude2004c> ${ffmpeg} -hwaccel vdpau -threads 1 -i "$stream" $mapping -c:s copy -c:v nvenc -2pass 1 -s 1280x720 -b:v 1800k -preset llhq -minrate 1k -maxrate 2200k -g 180 -r 30 -af "aresample=async=1000,volume=10dB" $audio -avoid_negative_ts 1 -frame_drop_threshold 1.0 -dts_delta_threshold 0 -hls_time 6 -break_non_keyframes 1 -hls_flags delete_segments -start_number $start -hls_list_size 5 ${HLS_PATH_FILES}/${CHN}/${CHN}2M.m3u8 
[16:54:31 CEST] <prelude2004c> 2>/var/log/channels/${CHN}/720p.txt;sleep 5;done
[16:56:21 CEST] <DHE> prelude2004c: if you're using my patch, delete avoid_negative_ts and dts_delta_threshold
[16:56:46 CEST] <DHE> also break_non_keyframe is not applicable
[16:58:39 CEST] <prelude2004c> ok... reason i was using that is because when i do pvr on one of the STB , the segments always start at a negative value which the box coudl not handle so i added that but sure.. for now i will remove
[16:58:46 CEST] <prelude2004c> is this for audio problem ?
[17:00:58 CEST] <prelude2004c> DHE , new stream could be found here : http://tv.zazeen.com/cbcvancouverhd1150-vdpau/cbcvancouverhd1150-vdpau.m3u8
[17:01:35 CEST] <prelude2004c> cool seems in sync now which is good
[17:02:16 CEST] <DHE> doesn't play
[17:02:32 CEST] <DHE> no response to connection attempt
[17:02:35 CEST] <DHE> but anyway
[17:03:13 CEST] <prelude2004c> really? ok try http://67.55.3.103/...
[17:03:18 CEST] <prelude2004c> replace the domain with that IP
[17:03:34 CEST] <DHE> that works...
[17:04:13 CEST] <DHE> audio sync seems okay
[17:05:20 CEST] <DHE> and can you elaborate on the negative value issue?
[17:05:22 CEST] <Illya> michaelni: have you changed anything with your emailing lately? They seem to have stopped rendering properly for me
[17:05:45 CEST] <prelude2004c> this mean anything to you ? Past duration 0.999657 too largeN/ .. i keep getting those
[17:05:51 CEST] <prelude2004c> probably not important i guess due to the live stream
[17:06:18 CEST] <michaelni> Illya, no clue dont really remember anything, what exactly doesnt display corerctly in which way ?
[17:06:43 CEST] <prelude2004c> finally the only last thing that was supposed to work with nvenc ffmpeg is the closed caption .. should be coming through but it is not it seems.. at least vlc is not showing it to me 
[17:07:58 CEST] <Illya> michaelni: hmm, looks like it's all of your emails. Suddenly as well, it must be an issue on my end.  (example) https://usercontent.irccloud-cdn.com/file/BbCmsSL4/Screen%20Shot%202016-06-01%20at%2016.07.03.png
[17:08:12 CEST] <prelude2004c> oh it seems the GIT version doesn't contain close caption data like version 3-0-1 did
[17:08:15 CEST] <prelude2004c> odd.. 
[17:09:05 CEST] <Illya> michaelni: however, when I view source the content is there. I was just wondering if you had changed anything, if not (as you said), I'll look into it further
[17:09:42 CEST] <michaelni> i only remember the regular security related updates
[17:09:47 CEST] <prelude2004c> odd how version 3.0.1 and 3.0.2 have closed caption and yet the git version does not 
[17:10:12 CEST] <prelude2004c> at least on ffprobe they do not show up. 
[17:12:51 CEST] <Illya> michaelni: looks like emails from you aren't being b64 encoded which seems to be tripping up the parser afaict
[17:13:01 CEST] <DHE> prelude2004c: I don't think nvenc has it yet. libx264 does with the -a53cc option. and my plan for gpu-based work got shot down, so I don't think I'll be doing it myself, unfortunately
[17:13:52 CEST] <prelude2004c> weird part is.. i had it working before at some point.. and i can't remember how i did it 
[17:14:03 CEST] <prelude2004c> but for now it seems the git version doesn't pull the closed caption stuff
[17:14:20 CEST] <prelude2004c> i am going to pull it with the 3.0.1 version to see the diff.
[17:16:08 CEST] <ubitux> ok i found one of the problem in the merge 
[17:16:13 CEST] <ubitux> almost there.
[17:17:28 CEST] <michaelni> Illya, and that was different in the past ?
[17:17:36 CEST] <prelude2004c> ok so i can confirm that the git version is different than the 3.0.1 version ... eg..( 3.0.1 ) >  Stream #0:4[0x1511]: Video: h264 (High) ([27][0][0][0] / 0x001B), yuv420p(tv, bt709), 1280x720 [SAR 1:1 DAR 16:9], Closed Captions, 59.94 fps, 59.94 tbr, 90k tbn ... &  ( GIT version ) >  Stream #0:4[0x1511]: Video: h264 (High) ([27][0][0][0] / 0x001B), yuv420p(tv, bt709), 1280x720 [SAR 1:1 DAR 16:9], 59.94 fps, 59.94 tbr, 90k tbn
[17:17:53 CEST] <prelude2004c> the git does not get closed caption data. clearly something unique there 
[17:33:49 CEST] <Illya> michaelni: no, doesn't seem like it. Must have been because of an update to thunderbird or something
[18:33:31 CEST] <cone-788> ffmpeg 03Michael Niedermayer 07master:003a9e4854d6: avutil/frame: Document avcodec_get_frame_class() and the option naming system
[18:33:32 CEST] <cone-788> ffmpeg 03Michael Niedermayer 07master:b8426b865edd: avcodec: Document where the AVOption names can be found for the fields of AVCodecContext
[18:33:33 CEST] <cone-788> ffmpeg 03Michael Niedermayer 07master:d1efdd54f2bf: avformat: Document where the AVOption names can be found for the fields of AVFormatContext
[19:34:11 CEST] <prelude2004c> DHE or BtbN , do you guys know why version 3.0.1 sees the closed caption source data and the git version does not ? same source same program ID
[19:34:42 CEST] <prelude2004c> i am trying to run 3.0.1 to test the outcome but that has a problem with vdpau ... so i can't test it.. so both have issues but different :) 
[19:35:08 CEST] <prelude2004c> DHE's code was for git version so i wanted to use that but somehow somethign is unique about 3.0.1 where closed caption data is seen on the transport stream
[19:35:21 CEST] <durandal_1707> michaelni: do you know idea how to do init vlc for those two tables used in sheervideo?
[19:38:06 CEST] <michaelni> durandal_1707, whats in the tables ?
[19:39:46 CEST] <michaelni> init_vlc takes bits, len and codes, whatever is in the tables need to be converted to that
[19:39:55 CEST] <durandal_1707> michaelni: first one of 256 size have bunch of numbers 3, 
[19:40:38 CEST] <michaelni> how could i know from that what to give init_vlc ?
[19:41:11 CEST] <durandal_1707> hmm, so I need reverse math now?
[19:41:42 CEST] <durandal_1707> I will post tables later
[19:42:23 CEST] <michaelni> assuming the tables represent a static vlc code the tables would need to be converted into what init_vlc need
[19:42:25 CEST] <durandal_1707> table is probably result of their init vlc...
[19:43:06 CEST] <michaelni> aha, then that has to be "reverted" yes
[19:43:41 CEST] <michaelni> if you want to use init_vlc and related vlc functions
[00:00:00 CEST] --- Thu Jun  2 2016