[Ffmpeg-devel-irc] ffmpeg-devel.log.20161022

burek burek021 at gmail.com
Sun Oct 23 03:05:02 EEST 2016


[00:05:18 CEST] <nevcairiel> jkqxz: did you test the vc1 decoding? it seems to rely on the parser setting coded_width/height and pix_fmt, but i dont think the vc1 parser does that (but it could be made to do that)
[00:07:25 CEST] <nevcairiel> unless it does that secretly somewhere that i didnt spot yet
[00:24:44 CEST] <jkqxz> nevcairiel:  Once, it was a very cursory "see one thing work and then ignore it because I don't care" test.
[00:24:50 CEST] <jkqxz> Let me try again.
[00:34:38 CEST] <jkqxz> Ah.  From my command history, I used -hwaccel (rather than -c:v) to test that, which silently does nothing (decodes on CPU, which of course succeeds).
[00:34:45 CEST] <jkqxz> Indeed it fails to get the pix_fmt.
[00:41:08 CEST] <RiCON> jkqxz: trying with mpv on an ivy cpu, mpeg2 and h264 don't work, they do without the patch
[00:42:06 CEST] <jkqxz> (MPEG-2 might not work for the same reason.)
[00:42:54 CEST] <jkqxz> (That I totally failed at testing it, I mean.)
[00:43:08 CEST] <jkqxz> How does it fail?  What media SDK version?
[00:43:21 CEST] <RiCON> i'm using mfx_dispatch
[00:46:16 CEST] <RiCON> https://i.fsbn.eu/old-mpeg2.txt https://i.fsbn.eu/new-mpeg2.txt https://i.fsbn.eu/old-h264.txt https://i.fsbn.eu/new-h264.txt
[00:48:07 CEST] <jkqxz> The parser for H.264 has messed up somehow.
[00:49:50 CEST] <jkqxz> Not sure about MPEG-2.
[00:51:35 CEST] <jkqxz> MPEG-2 does work for me when I run it properly.
[00:52:16 CEST] <RiCON> i'll check with just ffmpeg
[00:52:31 CEST] <RiCON> -hwaccel qsv or -c:v qsv_?
[00:52:37 CEST] <jkqxz> The latter.
[00:58:26 CEST] <RiCON> https://i.fsbn.eu/ffmpeg-new-mpeg2.txt https://i.fsbn.eu/ffmpeg-new-h264.txt
[01:36:52 CEST] <cone-212> ffmpeg 03Michael Niedermayer 07release/3.1:2fece989f824: doc/examples/demuxing_decoding: Drop AVFrame->pts use
[01:36:53 CEST] <cone-212> ffmpeg 03Michael Niedermayer 07release/3.1:de487cb765ca: avcodec/utils: Clear MMX state before returning from avcodec_default_execute*()
[01:36:54 CEST] <cone-212> ffmpeg 03Michael Niedermayer 07release/3.1:6456a7416e8f: avcodec/mpegvideo_enc: Clear mmx state in ff_mpv_reallocate_putbitbuffer()
[01:36:55 CEST] <cone-212> ffmpeg 03Michael Niedermayer 07release/3.1:9e6586ceb2f2: avformat/mxfdec: Check size to avoid integer overflow in mxf_read_utf16_string()
[01:41:10 CEST] <RiCON> jkqxz: seems to work with the patch
[01:41:39 CEST] <RiCON> the mpeg2 still no
[01:42:25 CEST] <RiCON> right, the diff just changed h264
[01:47:19 CEST] <jkqxz> I don't think that problem is related to qsv, then.  Your file is annex B but parsed as AVCC, maybe?
[01:47:47 CEST] <jkqxz> Does libav/avconv work?
[01:50:56 CEST] <cone-212> ffmpeg 03Michael Niedermayer 07release/3.1:2a5c41e3e4a7: Chagelog: update
[01:52:00 CEST] <RiCON> i don't have it compiled to check
[02:13:29 CEST] <cone-212> ffmpeg 03Andreas Cadhalpun 07n3.1.5:HEAD: doc: fix spelling errors
[02:19:08 CEST] <kierank> Good to see the scte-35 stuff got pushed in spite of all the objections
[02:19:16 CEST] <kierank> Such maturity
[02:23:12 CEST] <RiCON> jkqxz: mpeg-2 seems to not work as well, but h264 works
[02:24:13 CEST] <RiCON> https://i.fsbn.eu/mpeg2-libav.txt
[03:18:59 CEST] <cone-212> ffmpeg 03Kagami Hiiragi 07master:41da4f8cb3a7: lavc/libvpxenc: fix -auto-alt-ref option type
[03:57:22 CEST] <cone-212> ffmpeg 03Carl Eugen Hoyos 07master:6969bed12c6f: lavf/rtpdec_g726: Map mime type G726 to g726le.
[10:19:18 CEST] <nevcairiel> jkqxz: the h264 stuff is an interesting problem - mostly coming from the fact that the libav h264 parser actually can't parse avcc h264 streams, so it also has no attempts to handle it - but ours can, so it then tries to handle it, and gets confused in qsv because it gets passed through a bsf first :d
[10:21:25 CEST] <nevcairiel> an easy solution would be to either strip or convert the extradata in the internal avctx in ff_qsv_process_data
[10:56:32 CEST] <jkqxz> The altenative being to add more complexity to the parser to identify which type a stream is independent of the extradata?
[11:08:10 CEST] <nevcairiel> you need the extradata to parse avc
[11:10:00 CEST] <nevcairiel> the actual decoder has some hackery to try to detect annexb streams with an avc extradata header because that sometimes happens in broken files (usually not the other w ay around though since that would be much harder)
[11:10:09 CEST] <jkqxz> Yes, but it could spot that later packets are actually annex B.
[11:10:12 CEST] <nevcairiel> i suppose the parser could adapt that, but meh
[11:16:03 CEST] <jkqxz> Hmm, ok.  You first solution with stripping the extradata is probably the right one - there is no actual use for it there (the current qsv decoder never touches it at all).
[11:16:27 CEST] <nevcairiel> the parser reads it of course, but annexb extradata is supposed to be in the stream anyway
[11:17:39 CEST] <nevcairiel> a more complete solution would be to put the pre-converted extradata somehow in QSVContext inside qsvdec_h2645 and then use that for the internal avctx, the bsf produces it anyway in the codecpar
[11:18:52 CEST] <nevcairiel> that way it would make sure nothing is lost
[11:20:01 CEST] <nevcairiel> that would be much easier if parsers were updated to codecpar yet
[11:26:43 CEST] <jkqxz> I'm not seeing what you are suggesting is then done with the extradata?
[11:27:12 CEST] <nevcairiel> use that for the internal avctx so it contains annexb extradata and the parser can read it without getting confused
[11:32:21 CEST] <jkqxz> Which shouldn't make a difference against not having it at all because it's already in the annex B stream, but yes I suppose it is more correct.
[14:28:12 CEST] <cone-061> ffmpeg 03Michael Niedermayer 07master:70dc6bbf1bf0: avcodec/svq1enc: Clear MMX state after svq1_encode_plane()
[14:28:13 CEST] <cone-061> ffmpeg 03Michael Niedermayer 07master:493ad519ddee: avcodec/cavsdec: Clear MMX state after MB decode loop
[14:28:14 CEST] <cone-061> ffmpeg 03Michael Niedermayer 07master:966c5c7bb8ba: avcodec/utils: Move emms_c() before memory allocation functions in avcodec_encode_video2()
[14:28:15 CEST] <cone-061> ffmpeg 03Michael Niedermayer 07master:de0cd0ffc9e4: avcodec/mpegvideo_enc: Add missing emms_c() to clear MMX state after SIMD use
[14:28:16 CEST] <cone-061> ffmpeg 03Michael Niedermayer 07master:2c1d38d1e1c2: avcodec/snowenc: Clear MMX state after edge drawing and picture encode
[14:28:17 CEST] <cone-061> ffmpeg 03Michael Niedermayer 07master:f5495c970cf2: avutil/avassert: Add av_assertX_fpu()
[15:06:32 CEST] <Matador> weee QSV talk.. keep it goin
[19:24:13 CEST] <cone-061> ffmpeg 03Andreas Cadhalpun 07master:178eebd79e5b: mpegts: handle AVMEDIA_TYPE_UNKNOWN correctly
[20:52:28 CEST] <cone-061> ffmpeg 03James Almer 07master:cc71fa319fd7: avformat/matroskaenc: write DisplayWidth and DisplayHeight elements only if they differ from PixelWidth and PixelHeight
[23:35:13 CEST] <philipl> BtbN: https://github.com/philipl/FFmpeg/commit/1b1615481b18ec7cee64eadbe1608e71d5e31eb6
[23:35:23 CEST] <philipl> Tested with a TV stream with resolution changes
[23:54:43 CEST] <BtbN> should be fine, feel free to push (without using cvdl for master, of course)
[23:54:57 CEST] <philipl> of course.
[00:00:00 CEST] --- Sun Oct 23 2016


More information about the Ffmpeg-devel-irc mailing list