[Ffmpeg-devel-irc] ffmpeg-devel.log.20170129

burek burek021 at gmail.com
Mon Jan 30 03:05:04 EET 2017

[01:24:26 CET] <cone-381> ffmpeg 03Andreas Cadhalpun 07master:e3f13d3a8727: 4xm: prevent overflow during block alignment calculation
[01:24:27 CET] <cone-381> ffmpeg 03Andreas Cadhalpun 07master:8812d047bc85: electronicarts: prevent overflow during block alignment calculation
[01:24:28 CET] <cone-381> ffmpeg 03Andreas Cadhalpun 07master:169c1cfa9280: pvfdec: prevent overflow during block alignment calculation
[01:24:30 CET] <cone-381> ffmpeg 03Andreas Cadhalpun 07master:9ec8790ac478: boadec: prevent overflow during block alignment calculation
[13:31:37 CET] <cone-988> ffmpeg 03Paul B Mahol 07master:c6f7f33eec8e: avfilter/vf_remap: add . at end of long description
[15:20:17 CET] <Polochon_street> wm4: hi! I'm trying to make the fate test as you advised, with for example a flac file with id3v2 tags; do you think it should be on the libavformat.mak test file or more on the flac.mak test file?
[15:21:02 CET] <durandal_1707> whatever you prefer
[15:21:26 CET] <wm4> yeah
[15:22:25 CET] <Polochon_street> okay thanks :)
[16:13:11 CET] <cone-988> ffmpeg 03Michael Niedermayer 07master:bbd4d9230407: doc/examples/decoder_targeted: Disable error concealment after 20 frames
[16:14:52 CET] <wm4> michaelni: are you really, really, really sure our examples should contain such a hack?
[16:17:39 CET] <michaelni> that code is used by oss-fuzz
[16:18:34 CET] <michaelni> maybe it should be in tools and not examples
[16:18:56 CET] <wm4> ...
[16:19:19 CET] <wm4> I hope you know that this code is very likely to be copied into random new projects using libavcodec
[16:19:25 CET] <wm4> because that's what the examples are for
[16:19:33 CET] <wm4> so should they really contain something this specific
[16:20:18 CET] <Shiz> ^agree
[16:20:25 CET] <michaelni> if someone wants to move it to tools, i support that
[16:32:07 CET] <atomnuker> michaelni, wm4: patch sent
[17:14:49 CET] <cone-988> ffmpeg 03Rostislav Pehlivanov 07master:e05d2dd86abc: doc/examples/decoder_targeted: move to tools/target_dec_fuzzer.c
[17:16:41 CET] <t_than> BBB, I'd like to ask you a question about the VP9 GSoC project
[17:17:00 CET] <t_than> seems interesting and I'm thinking about applying
[17:17:36 CET] <BBB> ask away
[17:18:01 CET] <t_than> to be sure, when you mention "tile threading support", that is VP9 tiles decoded using pthreads?
[17:18:33 CET] <BBB> indirectly, yes
[17:18:44 CET] <BBB> but I want it using the slice task api in libavcodec
[17:18:56 CET] <BBB> so that its directly available in user applications and tools
[17:19:21 CET] <BBB> compare h264/vp8 -thread_type frame or -thread_type slice
[17:19:42 CET] <BBB> for vp9, -thread_type frame works, but -thread_type slice does not (I want that to work using the tiling feature)
[17:20:57 CET] <BBB> t_than: check how some decoders use avctx->execute2() as part of their internal frame decoding loop
[17:21:33 CET] <BBB> it splits the frame in multiple independent units and decodes each (using slice threading api and pthread) independently
[17:21:40 CET] <BBB> I want that invoked for tiles also
[17:23:11 CET] <t_than> ok, I'll check them out. I've just started out peeking the code and I'll be coming back for more info when I'll have some things figured out
[17:24:02 CET] <BBB> okay
[17:24:15 CET] <t_than> many thanks for the help :)
[17:25:14 CET] <BBB> the main loop is in vp9_decode_frame line 4127-4254
[17:25:20 CET] <BBB> youll notice tile_row and tile_col
[17:25:30 CET] <BBB> tile rows cannot be threaded, but tile cols can
[17:25:42 CET] <BBB> so the idea is to split the decoding into independent tile col units
[17:25:51 CET] <BBB> and use avctx->execute() to decode each unit independently
[17:26:14 CET] <BBB> note that the loop filter crosses tile col boundaries so you need some pthread_cond logic to make sure theres no race conditions there
[17:26:46 CET] <BBB> (e.g. do the loopfilter as one separate thread that waits for all tile cols for that y to be finished before it does its thing)
[17:27:42 CET] <BBB> I can provide some tile col samples once you get to the testing part (not all files have tile cols enabled)
[17:27:52 CET] <BBB> but youtube is a good source for clips also, all hd is tiled vp9 nowadays
[17:28:27 CET] <t_than> that's a useful info
[17:29:40 CET] <t_than> thanks again, I hope we'll talk again soon
[18:27:13 CET] <atomnuker> BBB you can't do that, there's no way to synchronize with tile threading like that
[18:27:27 CET] <BBB> atomnuker: there is, vp8 does it
[18:27:53 CET] <atomnuker> like, it runs the loop filter as part of avctx->execute2?
[18:28:13 CET] <atomnuker> how does it synchronize, do the functions for frame threading still work with slice therading?
[18:39:35 CET] <Polochon_street> wm4: for the id3v2 tests, would a static test over a flac file with id3v2 tags + vorbis comments enough? Like, check if the decoded artist tag is "Artist" and not "Artist;Artist", for example
[18:42:12 CET] <wm4> Polochon_street: yeah
[18:54:01 CET] <cone-988> ffmpeg 03Nicolas George 07master:383057f8e744: lavfi: make ff_framequeue_skip_samples() more useful.
[20:21:28 CET] <philipl> I got bored and decided to revist support for P016 input in swscale. I cannot for the life of me understand what it's doing. I patterned my change on the P010 code but it's not getting called. Something else is going on and some code path is being taken that is dropping all the chroma information (I just get monochrome output)
[20:23:50 CET] <Polochon_street> wm4: for the id3v2 test, something like this? http://sprunge.us/OLaH (this is really a first version, just to check if we're thinking about the same thing)
[20:25:14 CET] <wm4> Polochon_street: normally we'd check ffprobe output or so... I'm not sure if this case warrants adding a new test program, as it'd be relatively unusual
[20:26:14 CET] <Polochon_street> wm4: oh, okay, nevermind then, I'll use ffprobe output
[20:26:37 CET] <wm4> there are possibly other tests that check metadata support or so
[20:33:55 CET] <Polochon_street> well, I'm trying to look for them right now, but I can't find anything. Maybe in audio-filter.mak?
[20:34:09 CET] <Polochon_street> audio.mak*
[22:07:32 CET] <cone-988> ffmpeg 03Matthieu Bouron 07master:2ae827883244: lavc/mjpegdec: consume SOS data even if the frame is discarded
[22:30:58 CET] <ePirat> Good evening
[22:39:39 CET] <durandal_1707> gonna push scc muxer
[23:46:01 CET] <cone-988> ffmpeg 03Muhammad Faiz 07master:c4a3526b57dc: avfilter/showcqt: make minimum timeclamp option lower
[00:00:00 CET] --- Mon Jan 30 2017

More information about the Ffmpeg-devel-irc mailing list