[Ffmpeg-devel-irc] ffmpeg-devel.log.20140113

burek burek021 at gmail.com
Tue Jan 14 02:05:02 CET 2014

[01:31] <cone-846> ffmpeg.git 03Michael Niedermayer 07master:7b41cbac7fc0: avcodec/huffman: extend ff_huff_gen_len_table() to allow >8bit
[01:31] <cone-846> ffmpeg.git 03Michael Niedermayer 07master:7a8318fbb1f5: avcodec/huffman: raise the input capability of ff_huff_gen_len_table() from 8 to 14bit
[04:08] <llogan> ubitux: Trivia: The movie "The Fifth Estate" contains a reference to ffmpeg at ~1h15s.
[05:15] <cone-821> ffmpeg.git 03Luca Barbato 07master:48a5b155433e: hevc: Reject impossible dependent tile
[05:15] <cone-821> ffmpeg.git 03Luca Barbato 07master:aead772b5814: hevc: Bound check slice_qp
[05:15] <cone-821> ffmpeg.git 03Michael Niedermayer 07master:22bfb4be284c: avcodec/hevc: Check entry point arrays for malloc failure
[07:29] <ubitux> llogan :)
[10:58] <funman> I wonder why mraulet added the temporary (from the draft) HEVC transport stream ID to the demuxer but not the muxer
[10:58] <funman> it's a bit useless if noone generates ts files with that id
[10:58] <funman> https://github.com/OpenHEVC/openHEVC/blob/hevc/libavformat/mpegts.c oh well he did
[10:59] <funman> michaelni: any reason against picking that up ?
[11:00] <funman> woops mpegts.c is the demux
[11:00] <JEEB> just make sure that's in the latest draft too
[11:00] <JEEB> I don't think it's changed, but just make sure
[11:00] <JEEB> unless you have access to the documents
[11:00] <funman> well i don't even have the draft
[11:40] <ubitux> arh
[11:40] <ubitux> it seems best_effort_timestamp might be broken in case of non-flushing
[11:41] <ubitux> i need to confirm this, but if you reach the end, don't flush, seek to start, decode
[11:41] <ubitux> the best_effort_timestamp might be broken (with a ending ts info)
[11:47] <wm4> best_effort_timestamp is in general not very reliable
[11:48] <ubitux> what should i use? :P
[11:49] <wm4> write your own complicated custom code
[11:49] <wm4> for users, best_effort_timestamp isn't an option anyway, because libav doesn't have it
[11:51] <nevcairiel> by that logic ffmpeg may as well stop developing features
[11:51] <wm4> if it's something as basic as this, then yes
[11:51] <wm4> and it does
[11:52] <nevcairiel> by my logic, once you have enough compelling features, people will choose the right lib that works for them
[11:54] <durandal_1707> huh, why cehoys replies to msgs i do not see?
[11:54] <wm4> durandal_1707: spam filter?
[11:56] <durandal_1707> no, i think cehoyos picked wrong pill
[11:57] <durandal_1707> wm4: oops you are right, fucking retarded spam filter
[12:02] <ubitux> anyway it seems to do some flushing post seek
[12:02] <ubitux> that's really weird, but my code might be wrong
[12:44] <wm4> there should be a ffmpeg consulting company
[12:45] <BBB> ubitux: yeah we beat them by a good 15% now
[12:45] <ubitux> yeaah
[12:45] <ubitux> BBB: not in 32-bit and ARM though i suppose
[12:45] <ubitux> BBB: with threading?
[12:46] <ubitux> also we don't have any avx yet
[12:46] <ubitux> i think they do
[12:53] <pross-au> BBB: about?
[13:00] <pross-au> disregard. think i solved it
[13:19] <Rodeo> Hi! Is anyone here going to FOSDEM and still looking to share accommodations (e.g. hotel)?
[13:19] <BBB> ubitux: w/o threading, w/ a lot more
[13:20] <BBB> ubitux: depending on thread type basically
[13:20] <BBB> tey have some avx2 in the encoder, I didn't look to see if they have it in the decoder
[13:20] <ubitux> Rodeo: i don't know if some ff dev developers are going to fosdem
[13:20] <ubitux> BBB: i was think of avx
[13:20] <BBB> oh
[13:20] <ubitux> not even avx2
[13:21] <Rodeo> neither do I, which is why I'm asking ;-)
[13:21] <ubitux> thinking*
[13:21] <BBB> it's trivially simple to add avx
[13:21] <BBB> we already made the macros the correct way
[13:21] <BBB> just add some INIT_XMM avx and we're done
[13:21] <BBB> kierank has a test machine he volunteered for us to use for testing
[13:21] <ubitux> well, will that save some unpacking?
[13:21] <BBB> mova mostly
[13:21] <BBB> avx adds three-operand instructions
[13:21] <ubitux> yeah right, but that's not the interesting thing
[13:21] <BBB> oh, you'd be surprised
[13:22] <BBB> we use it all over the place
[13:22] <kierank> I have an avx2 machine but nowhere to colocate it :(
[13:22] <BBB> ubitux: example, for prores simd, it helped in the order of 10% or so, iirc
[13:22] <ubitux> BBB: yeah but wouldn't the main benefit in stuff like loop filter where you don't unpack twice
[13:23] <BBB> you mean we punpcklbw on the same data twice?
[13:23] <BBB> or unpack = load-from-memory?
[13:23] <ubitux> yes the punpck[lh]bw/repack
[13:24] <BBB> where do we do it twice?
[13:24] <BBB> in filter6/14?
[13:24] <ubitux> yes
[13:24] <BBB> hm... maybe we can merge these two
[13:25] <ubitux> would that mean using half of 256b regs for the previous code?
[13:26] <BBB> I hope not
[13:26] <BBB> rather, for avx2, we would do 32bytes all at once
[13:26] <BBB> we have 32 registers anyway
[13:26] <ubitux> ah
[13:27] <ubitux> also, i guess for itxfm with 16x16 and 32x32 having 256b reg would help
[13:27] <ubitux> and i'm assuming you will have to write some code avx specific
[13:27] <ubitux> no?
[13:49] <BBB> ubitux: absolutely
[13:49] <BBB> ubitux: it's not that trivial to convert from sseX to avx2, but it can be done I guess
[13:49] <BBB> 32x256b regs certainly gives incentive
[13:49] <ubitux> again, i'm thinking avx :p
[13:50] <ubitux> avx is 256b and avx2 512 right?
[13:50] <BBB> oh wait the 32 was for the generation after that
[13:50] <BBB> nah I'm confusing two things, 32 is for 512b regs, the one after avx/2
[13:50] <BBB> iirc
[13:50] <BBB> avx2 is integer form of avx, like sse2 is integer form of sse
[13:52] <ubitux> oh so, since i have here a cpu with avx, it has absolutely no use for integers except the mova thing?
[13:52] <ubitux> (and eventually float ops?)
[13:52] <BBB> yes
[13:53] <ubitux> mmh ok
[14:02] <michaelni> funman, i cherry picked all things that mraulet asked me to pick, if theres more thats ready and should be picked
[14:03] <michaelni> just say what / which hash
[14:13] <cone-123> ffmpeg.git 03Luca Barbato 07master:c798a6fedc3d: vc1: Factorize out chroma MC
[14:13] <cone-123> ffmpeg.git 03Michael Niedermayer 07master:1df53ff22de2: Merge commit 'c798a6fedc3dd9ed3ac1fb2d95819af58705a87e'
[14:14] <michaelni> funman / and others, also if theres some hevc repo/tree that can just be git merged thats welcome as well
[14:24] <cone-123> ffmpeg.git 03Michael Niedermayer 07master:6d98959c8a5a: vc1: Add avg_no_rnd_vc1_chroma_mc4_c()
[14:24] <cone-123> ffmpeg.git 03Michael Niedermayer 07master:58eb9b929998: Merge commit '6d98959c8a5a749592a878b9523b31eccea5d102'
[14:33] <funman> michaelni: there doesn't seem to be more, I'll ask mraulet when I see him
[14:36] <cone-123> ffmpeg.git 03Tim Walker 07master:49b9badcaab5: vc1: Enable the interlaced B-frame codepath.
[14:36] <cone-123> ffmpeg.git 03Michael Niedermayer 07master:a067445b45fb: Merge commit '49b9badcaab513b65b79be371b1a9c1012e482ea'
[14:37] <michaelni> funman, ok, thanks
[14:43] <cone-123> ffmpeg.git 03Martin Storsjö 07master:99e20125232d: x86/arm: Add clobber tests to libavresample
[14:43] <cone-123> ffmpeg.git 03Michael Niedermayer 07master:9ace154186d8: Merge commit '99e20125232d40ea239a5b54d26bbfe526506114'
[15:14] <cone-123> ffmpeg.git 03Luca Barbato 07release/1.1:9aa22918c258: prores: Error out only on surely incomplete ac_coeffs
[15:14] <cone-123> ffmpeg.git 03Michael Niedermayer 07release/1.1:9f47f95e7087: Merge remote-tracking branch 'qatar/release/9' into release/1.1
[15:21] <mischi> hi.
[15:22] <mischi> I want to suggest to add UltraStar Deluxe, a Karaoke game, to the list of projects using ffmpeg.
[15:23] <mischi> With 10.000 downloads per week from sourceforge, i think it is relevant.
[15:23] <av500> sourceforge still exists?
[15:23] <mischi> hehe
[15:23] <mischi> similar as Pascal, the language used for USDX.
[15:24] <mischi> As such USDX has header translations for using ffmpeg libs from Pascal.
[15:24] <kierank> av500: nowhere has better mirrors than sourceforge
[15:30] <cone-123> ffmpeg.git 03Michael Niedermayer 07release/1.1:1017b5914cec: update for 1.1.8
[15:32] <michaelni> mischi, send patch, see git://ffmpeg.org/ffmpeg-web
[15:47] <cone-123> ffmpeg.git 03Michael Niedermayer 07fatal: ambiguous argument 'refs/tags/n1.1.8': unknown revision or path not in the working tree.
[15:47] <cone-123> Use '--' to separate paths from revisions
[15:47] <cone-123> refs/tags/n1.1.8:HEAD: Merge commit '99e20125232d40ea239a5b54d26bbfe526506114'
[15:50] <Daemon404> who was it that run the vlc irc bots? funman?
[15:51] <ubitux> thresh?
[15:52] <Daemon404> ah right.
[15:52] <Daemon404> it really should be a one line fix
[15:52] <Daemon404> i see that -- error a lot
[16:17] <mischi> michaelni: ok.
[17:29] <cone-123> ffmpeg.git 03Michael Niedermayer 07release/2.1:6fb0f37defbf: avcodec/h264: fix code that blindly dereferences NULL DPB
[17:29] <cone-123> ffmpeg.git 03Michael Niedermayer 07release/2.1:050d8d727ad6: avformat/rmdec: when reading audio blocks, dont leave holes when reading fails
[17:29] <cone-123> ffmpeg.git 03Michael Niedermayer 07release/2.1:4f3b0afaa236: avformat/mxfdec: check avio_read(UID) result
[17:29] <cone-123> ffmpeg.git 03Michael Niedermayer 07release/2.1:763a808c96bb: avformat/flvdec: initialize context before reading from it
[17:29] <cone-123> ffmpeg.git 03Michael Niedermayer 07release/2.1:6da213ce208d: avformat/matroskadec: check generic audio deinterleaver sub_packet_size against frame_size
[17:29] <cone-123> ffmpeg.git 03Michael Niedermayer 07release/2.1:aaaafc300df6: avcodec/wmalosslessdec: Pass on error code from decode_tilehdr()
[17:29] <cone-123> ffmpeg.git 03Michael Niedermayer 07release/2.1:cd753ee36b0f: avcodec/wmalosslessdec: deallocate uninitialized frame on decode_tilehdr() failure
[17:29] <cone-123> ffmpeg.git 03Michael Niedermayer 07release/2.1:57824d1ba617: avcodec/wmalosslessdec: shrink output on error so no uninitialized data is returned
[17:29] <cone-123> ffmpeg.git 03Clément BSsch 07release/2.1:b6af89be6a2b: avformat/pjsdec: dont increase pointer when its already at the end in read_ts()
[17:29] <cone-123> ffmpeg.git 03Michael Niedermayer 07release/2.1:b3c3dc54a595: avcodec/alac: only set *got_frame_ptr when all channels have been decoded
[17:29] <cone-123> ffmpeg.git 03Michael Niedermayer 07release/2.1:706dca18d0a8: avcodec/hevc: use av_mallocz() for allocating tab_ipm
[17:29] <cone-123> ffmpeg.git 03Luca Barbato 07release/2.1:738a2a04b6d8: hevc: Clip the pixel before shifting
[17:29] <cone-123> ffmpeg.git 03Guillaume Martres 07release/2.1:ea21b7b68cfc: hevc: clip pixels when transquant bypass is used
[17:29] <cone-123> ffmpeg.git 03Guillaume Martres 07release/2.1:e43805d40188: hevc: remove useless clip in FUNC(sao_band_filter)()
[17:29] <cone-123> ffmpeg.git 03Luca Barbato 07release/2.1:39545c54826c: hevc: Reject impossible dependent tile
[17:29] <cone-123> ffmpeg.git 03Luca Barbato 07release/2.1:d63476347a48: hevc: Bound check slice_qp
[17:29] <cone-123> ffmpeg.git 03Michael Niedermayer 07release/2.1:a0aa5c34a964: avcodec/hevc: Check entry point arrays for malloc failure
[17:29] <cone-123> ffmpeg.git 03Carl Eugen Hoyos 07release/2.1:b336daa952d9: Fix a crash on oom when decoding hevc. (cherry picked from commit 5ab1efb9d0dc65e748a0291b67915e35578b302e)
[17:29] <cone-123> ffmpeg.git 03Michael Niedermayer 07release/2.1:29353dd3f815: Update for 2.1.2
[17:52] <cone-123> ffmpeg.git 03Michael Niedermayer 07fatal: ambiguous argument 'refs/tags/n2.1.2': unknown revision or path not in the working tree.
[17:52] <cone-123> Use '--' to separate paths from revisions
[17:52] <cone-123> refs/tags/n2.1.2:HEAD: Merge commit '99e20125232d40ea239a5b54d26bbfe526506114'
[18:55] <Daemon404> 42
[19:05] <gnafu> A likely story.
[19:37] <llogan> BBB: any opinion on http://ffmpeg.org/pipermail/ffmpeg-devel/2014-January/153152.html ?
[20:37] <llogan> Anssi: anything upcoming for HLS in the near future?
[20:39] <Anssi> llogan: well, I'm hoping I get my patchset in shape in the next few weeks...
[20:47] <llogan> Anssi: not sure how i spaced that out...
[20:48] <llogan> my english sucks. i meant that I'm not sure how i forgot that patch set.
[20:50] <Anssi> mostly the id3 timestamp stuff needs some thinking, I may go back to the approach I had in v1 patchset... another thing is if a cleaner approach to subtitles is feasible
[21:34] <cone-644> ffmpeg.git 03Anton Khirnov 07master:c1868e7ee7b0: lavf: simplify handling of offset in av_probe_input_buuffer()
[21:34] <cone-644> ffmpeg.git 03Michael Niedermayer 07master:ececbedaa86c: Merge commit 'c1868e7ee7b07b40a0fe15f50df89fe499a01a50'
[21:40] <cone-644> ffmpeg.git 03Anton Khirnov 07master:8b76362836f3: lavf: use a fixed width type
[21:40] <cone-644> ffmpeg.git 03Michael Niedermayer 07master:220f0927b86f: Merge commit '8b76362836f3c373c3aadc544522edcbef16dd5f'
[22:35] <cone-644> ffmpeg.git 03Anton Khirnov 07master:2115a3597457: lavf: make av_probe_input_buffer more robust
[22:35] <cone-644> ffmpeg.git 03Michael Niedermayer 07master:cdce0e8a506c: Merge remote-tracking branch 'qatar/master'
[22:35] <cone-644> ffmpeg.git 03Michael Niedermayer 07master:6a2064820b52: avformat/utils/av_probe_input_buffer2: Fix pd.buf_size
[22:35] <cone-644> ffmpeg.git 03Michael Niedermayer 07master:8c3b026a0eeb: avformat/utils/av_probe_input_buffer2: fix offset check
[22:35] <cone-644> ffmpeg.git 03Michael Niedermayer 07master:05886c9d4edd: avformat/utils/av_probe_input_buffer2: fix buffer passed to ffio_rewind_with_probe_data()
[23:47] <ubitux> BBB: i have 4 new samples to crash the decoder
[23:47] <ubitux> interested?
[00:00] --- Tue Jan 14 2014

More information about the Ffmpeg-devel-irc mailing list