[Ffmpeg-devel-irc] ffmpeg-devel.log.20170423

[burek]

[00:24:08 CEST] <philipl> jamrial: https://github.com/philipl/FFmpeg/commit/b8a9e2f34e409d81a28355f41cd8ff0b955ddf68
[00:24:14 CEST] <philipl> Can send an actual pull request if you want it
[00:24:29 CEST] <jamrial> no, that's good
[01:32:57 CEST] <cone-450> ffmpeg 03Anton Khirnov 07master:549d0bdca53a: decode: be more explicit about storing the last packet properties
[01:32:58 CEST] <cone-450> ffmpeg 03Anton Khirnov 07master:061a0c14bb57: decode: restructure the core decoding code
[01:32:59 CEST] <cone-450> ffmpeg 03James Almer 07master:1fd76277708c: Merge commit '549d0bdca53af7a6e0c612ab4b03baecf3a5878f'
[01:33:00 CEST] <cone-450> ffmpeg 03James Almer 07master:bddb2343b6e5: Merge commit '061a0c14bb5767bca72e3a7227ca400de439ba09'
[01:33:01 CEST] <cone-450> ffmpeg 03Philip Langdale 07master:3148387086ad: avcodec/crystalhd: Adapt to new new decode API
[02:13:33 CEST] <kierank> how the hell do i set interlaced field order for a sample
[02:13:39 CEST] <kierank> raw sample
[02:38:18 CEST] <atomnuker> -field_order tt
[02:38:40 CEST] <kierank> ah it's this vf_scale thing where it ignores is
[02:38:42 CEST] <kierank> it
[02:38:43 CEST] <kierank> grr
[03:57:15 CEST] <cone-450> ffmpeg 03Anton Khirnov 07master:972c71e9cb63: lavc: add support for filtering packets before decoding
[03:57:16 CEST] <cone-450> ffmpeg 03James Almer 07master:844a115cd3cd: Revert "avcodec/bsf: Forbid packet without payload in av_bsf_send_packet"
[03:57:17 CEST] <cone-450> ffmpeg 03Jan Sebechlebsky 07master:f92e1af84420: avcodec/bsf: Check for packet payload when setting BSF EOF flag.
[03:57:18 CEST] <cone-450> ffmpeg 03James Almer 07master:79778bb9b0ae: Merge commit '972c71e9cb63e24f57ee481e413199c7d88a8813'
[04:04:42 CEST] <cone-450> ffmpeg 03Anton Khirnov 07master:8fb4210ad878: qsvdec_h2645: switch to the new generic filtering mechanism
[04:04:43 CEST] <cone-450> ffmpeg 03James Almer 07master:58ed9deec868: Merge commit '8fb4210ad8785c01fccf2fc59af6a6fa2892b6b2'
[04:28:24 CEST] <cone-450> ffmpeg 03Anton Khirnov 07master:03a80925effc: lavc: add a bitstream filter for splitting VP9 superframes
[04:28:25 CEST] <cone-450> ffmpeg 03James Almer 07master:fdeab95a823b: Merge commit '03a80925effc2698d21dc0b00290eecf42dd9e68'
[04:31:34 CEST] <jamrial> ubitux: next commit doesn't apply cleanly since it basically implements vp9_parser but using the above bitstream filter
[04:31:45 CEST] <jamrial> not sure what you and BBB talked/thought about it, so i'll leave it to you
[04:32:44 CEST] <jamrial> but at least the decode api commits have been dealt with :D
[04:34:18 CEST] <jamrial> philipl: the autobsf for decoders was merged, if you wanted to look at getting crystalhd to use it
[04:34:32 CEST] <jamrial> see the qsvdec commit for an example
[04:39:53 CEST] <philipl> jamrial: what exactly does autobsf do?
[04:39:58 CEST] <philipl> it handles things like annexb?
[04:40:16 CEST] <jamrial> yes, look at 8fb4210ad878
[04:40:35 CEST] <jamrial> basically, you can get rid of all the bsf code in crystalhd.c for h264
[04:41:05 CEST] <philipl> cool beans
[05:02:57 CEST] <philipl> And now packed bframe mpeg4 doesn't work again. *sigh*
[08:27:38 CEST] <chatter29> hey guys
[10:14:10 CEST] <cone-327> ffmpeg 03Muhammad Faiz 07master:31f61b0d4f8d: avcodec: do not use AVFrame accessor
[10:14:10 CEST] <cone-327> ffmpeg 03Muhammad Faiz 07master:9f030ee00f04: avformat: do not use AVFrame accessor
[10:14:10 CEST] <cone-327> ffmpeg 03Muhammad Faiz 07master:8103c5952236: avdevice: do not use AVFrame accessor
[10:14:10 CEST] <cone-327> ffmpeg 03Muhammad Faiz 07master:6af050d7d0c3: avfilter: do not use AVFrame accessor
[10:14:10 CEST] <cone-327> ffmpeg 03Muhammad Faiz 07master:8893c943a9bd: ff*: do not use AVFrame accessor
[10:14:11 CEST] <cone-327> ffmpeg 03Muhammad Faiz 07master:327a1c0dee05: examples: do not use AVFrame accessor
[10:14:11 CEST] <cone-327> ffmpeg 03Muhammad Faiz 07master:cdd3048134b8: tests: do not use AVFrame accessor
[12:07:06 CEST] <cone-327> ffmpeg 03Paul B Mahol 07master:ac30754a148d: avcodec/dnxhd_parser: fix parsing interlaced video, simplify code
[13:26:14 CEST] <ubitux> BBB: ping
[13:28:23 CEST] <BBB> ubitux: hi
[13:28:41 CEST] <ubitux> BBB: hi :)
[13:28:48 CEST] <ubitux> BBB: we're at a vp9 merge
[13:28:59 CEST] <ubitux> libav used a bsf for the superframes
[13:29:13 CEST] <ubitux> while we do this in a parser
[13:29:28 CEST] <BBB> Im fine with the bsf as long as its auto-inserted and frame-threading speed isnt affected
[13:29:33 CEST] <ubitux> we need to unify this, probably by porting our vp9 parser optimizations into the bsf
[13:29:33 CEST] <BBB> (I told that to anton)
[13:29:52 CEST] <ubitux> do you mind looking into this?
[13:30:05 CEST] <ubitux> their bsf is currently in our tree
[13:30:14 CEST] <ubitux> libavcodec/vp9_superframe_split_bsf.c
[13:30:24 CEST] <ubitux> it probably needs the switch optimization
[13:30:36 CEST] <BBB> how do I use it?
[13:30:44 CEST] <ubitux> see fa1749dd34
[13:30:46 CEST] <BBB> is there a cli option to use bsfs on input?
[13:31:09 CEST] <ubitux> try ff_vp9_decoder.bsfs = "vp9_superframe_split"
[13:31:28 CEST] <ubitux> maybe -bsf:v vp9_superframe_split as input option?
[13:31:40 CEST] <BBB> ok
[13:31:50 CEST] <BBB> and is there a mechanism to auto-insert them on the input-side?
[13:32:05 CEST] <BBB> (like auto-bsf in muxers, but for decoders)
[13:32:30 CEST] <ubitux> 13:31 <@ubitux> try ff_vp9_decoder.bsfs = "vp9_superframe_split"
[13:32:35 CEST] <ubitux> this doesn't do it?
[13:33:06 CEST] <BBB> I havent tested :)
[13:33:08 CEST] <BBB> Ill try
[13:33:22 CEST] <BBB> how much time do I have to work on this?
[13:33:27 CEST] <BBB> are parsers entirely going away?
[13:34:22 CEST] <ubitux> asap is the best, otherwise we can noop the commit and add it to the long queue of doc/libav-merge.txt
[13:35:04 CEST] <ubitux> no idea about the parser, but a large rework is probably needed/wanted/wished/fantasmed for a while
[13:39:15 CEST] <BBB> but is there a plan on parsers?
[13:39:28 CEST] <BBB> like, do they still have a use case that is not covered by the fantastic new input bsfs?
[13:40:43 CEST] <ubitux> no idea
[13:41:07 CEST] <ubitux> i don't see anything relevant related to this in their log
[13:45:10 CEST] <durandal_1707> ubitux: have you seen #6303?
[13:45:48 CEST] <ubitux> i didn't write movtextenc.c
[13:46:35 CEST] <ubitux> i think philipl may be more fluent with it
[13:47:39 CEST] <BBB> ubitux: maybe parsers should just go?
[13:47:54 CEST] <ubitux> probably yes
[13:48:27 CEST] <ubitux> but i don't think bsf can do everything parsers (should) do
[13:48:38 CEST] <ubitux> typically, you want to be able to probe without decoding
[13:48:53 CEST] <ubitux> to prevent huge slowdowns
[13:53:38 CEST] <durandal_1707> what? you want to kill all parsers?
[13:54:08 CEST] <nevcairiel> this concept of bsf can replace parsers that do nothing but bitstream splitting
[13:54:18 CEST] <nevcairiel> however, i do believe that avformat should do this task before outputting packets
[13:54:24 CEST] <nevcairiel> for compat with any other decoders
[14:07:51 CEST] <wm4> parsers are needed to extract codec parameters
[14:07:57 CEST] <wm4> BSFs can't do that
[14:11:07 CEST] <BBB> nevcairiel: there seem to be varying opinions on that
[14:11:25 CEST] <nevcairiel> well i use avformat with other decoders
[14:11:28 CEST] <nevcairiel> so its a fixed requirements
[14:11:33 CEST] <BBB> nevcairiel: for example, I agree with that, but several people object to that in cases where the original (pre-parsed) packet boundaries reflect what is expected by a spec
[14:11:45 CEST] <BBB> e.g. should vp9 split superframe or frame boundaries?
[14:11:52 CEST] <BBB> ffvp9 expects frame, muxers expect superframe
[14:11:55 CEST] <BBB> which is correct?
[14:12:03 CEST] <nevcairiel> the vp9 case is a bit special
[14:12:07 CEST] <BBB> same for packed divx
[14:12:22 CEST] <BBB> it is splitting, and it affects the answer
[14:12:40 CEST] <BBB> this concept of bsf can replace parsers that do nothing but bitstream splitting
[14:12:51 CEST] <BBB> however, i do believe that avformat should do this task before outputting packets
[14:12:53 CEST] <BBB> right?
[15:27:44 CEST] <cone-905> ffmpeg 03Nicolas George 07master:0dfb4d4b438b: ffmpeg: do not report EOF on filters as an error.
[17:18:49 CEST] <philipl> durandal_1707, ubitux: Well, 6303 is an attempt to do something stupid - you can't put movtext inside anything except an mp4 container.
[17:18:55 CEST] <philipl> It obviously shouldn't crash, but...
[17:19:42 CEST] <philipl> Still, it crashes when using an mp4 container. Will investigate.
[17:19:44 CEST] <durandal_1707> it crashes if you pick null output
[17:20:17 CEST] <durandal_1707> it tries to write to null pointer
[17:21:34 CEST] <philipl> yeah. I'll poke at it today.
[17:40:24 CEST] <cone-905> ffmpeg 03Paul B Mahol 07master:0699722ad050: avfilter/vf_maskedclamp: limit overshot and undershot to UINT16_MAX
[17:55:44 CEST] <cone-905> ffmpeg 03James Almer 07master:16c88465a1c1: avcodec/decode: also update consumed bytes on last_pkt_props->size
[18:23:30 CEST] <cone-905> ffmpeg 03Paul B Mahol 07master:9d1f9ba582aa: avfilter/vf_maskedclamp: fix bug when copying >8bit plane(s)
[18:23:31 CEST] <cone-905> ffmpeg 03Paul B Mahol 07master:710c97d5f6a8: avfilter/vf_premultiply: add planes option
[18:34:35 CEST] <cone-905> ffmpeg 03Paul B Mahol 07master:f4218d93efae: avfilter/vf_maskedmerge: fix bug when copying >8bit plane(s)
[18:46:15 CEST] <cone-905> ffmpeg 03Philip Langdale 07master:41b0561dc724: avcodec/crystalhd: Switch to the new generic filtering mechanism
[18:47:45 CEST] <jamrial> philipl: nice
[19:11:07 CEST] <durandal_1707> BtbN: do you mind if i write lumakey filter?
[19:11:54 CEST] <cone-905> ffmpeg 03Derek Buitenhuis 07master:8e6b9ef4733b: webm_dash_manifest_demuxer: Fix UB in cue timestamp string code and make it actually work
[19:11:55 CEST] <cone-905> ffmpeg 03Derek Buitenhuis 07master:6ba1c9bf7e8a: webm_dash_manifest_demuxer: Fix initialization range for files with cues at the front
[19:12:35 CEST] <philipl> durandal_1707: That ass sample seems mangled.
[19:13:17 CEST] <philipl> The encoder blows up because it's seeing a close tag before any opens, and I can fix that not to blow up, but it's not valid
[19:13:58 CEST] <philipl> converting to srt produces a very different file
[19:14:04 CEST] <philipl> srt and back, rather.
[19:46:35 CEST] <cone-905> ffmpeg 03Philip Langdale 07master:f95c81ce1045: avcodec/movtextenc: Ignore unmatched closing style tags
[23:24:35 CEST] <jamrial> wm4: can you check why fate-h264-attachment-631 is failing when you use four or more threads?
[23:25:05 CEST] <jamrial> it's dropping one frame for every four decoding threads
[23:25:27 CEST] <jamrial> i ran out of ideas and things to try with this
[23:38:41 CEST] <wm4> what caused it?
[23:53:33 CEST] <jamrial> wm4: the new new api merge i think
[23:53:44 CEST] <jamrial> it's frame threads only, not slice
[23:56:09 CEST] <wm4> maybe I can take a look tomorrow, not like I know the code much better though
[23:56:58 CEST] <jamrial> thanks
[00:00:00 CEST] --- Mon Apr 24 2017