[Ffmpeg-devel-irc] ffmpeg-devel.log.20131025

burek burek021 at gmail.com
Sat Oct 26 02:05:02 CEST 2013


[00:20] <Compn> if someone wants to save me from that evansport thread
[00:20] <Compn> i'd be happy
[00:21] <cone-698> ffmpeg.git 03Michael Niedermayer 07master:da54ad9a7208: ffmpeg_filter: Pass command line -threads X to filtergraph
[00:21] <cone-698> ffmpeg.git 03Michael Niedermayer 07master:29ffeef5e73b: avcodec/h264: do not trust last_pic_droppable when marking pictures as done
[01:23] <cone-698> ffmpeg.git 03Michael Niedermayer 07master:4fb14f84926e: avcodec/wmalosslessdec: fix data flushing at the end
[01:23] <llogan> RE: #3045, i wonder if an info message "Use -framerate instead of -r for x11grab" would be good.
[01:45] <michaelni> llogan, dunno but if you think so patch is welcome
[01:51] <BBB> ubitux: cool
[01:51] <BBB> ubitux: don't forget review, I may have comments, or real experts like loren/jason
[01:58] <llogan> michaelni: i can try but it might be laughable
[01:59] Action: llogan sticks to user help/trolling/asking for console outputs
[02:21] <cone-698> ffmpeg.git 03Michael Niedermayer 07master:a443b9732f75: ffmpeg: make stream default if its the only of its kind and no source can be unambigously assigned to it
[07:19] <ubitux> <@BBB> ubitux: don't forget review, I may have comments, or real experts like loren/jason // ofc, i won't push this out of the blue :)
[09:18] <cone-822> ffmpeg.git 03Paul B Mahol 07master:0f15ebe588a7: MAINTAINERS: update
[09:30] <wcpan_> hi, does libswscale support yuv420p to rgb24 conversion on arm?
[09:30] <nevcairiel> it can convert anything with C code, however that may not be very fast
[09:30] <ubitux> there is no arm optim in sws... :(
[09:31] <durandal_1707> s/anything/most things
[09:31] <wcpan_> I got some "No accelerated colorspace conversion found from yuv420p to rgb24"
[09:31] <durandal_1707> that is is ...
[09:31] <nevcairiel> it will still convert
[09:31] <nevcairiel> it'll just be slow
[09:33] <wcpan_> ouch, I see. thank you
[09:46] <durandal_1707> saste: my fork will not have 'ff' in its name
[10:53] <durandal_1707> i would really like to push mergeplanes soon
[10:58] <durandal_1707> should i add vf_dlopen from mpv wrapper?
[10:59] <durandal_1707> which is kind of wrapper around libmpcodecs
[11:42] <durandal_1707> michaelni: will you apply '[PATCH] build: remove pointless condition'
[11:42] <durandal_1707> ?
[11:43] <cone-822> ffmpeg.git 03Michael Niedermayer 07master:3db7ae6a9a25: doc/issue_tracker: theres no substatus in trac
[11:43] <cone-822> ffmpeg.git 03Michael Niedermayer 07master:2a28bad82202: doc/issue_tracker: remove patches as we dont use trac to handle patches
[11:43] <cone-822> ffmpeg.git 03Michael Niedermayer 07master:cdb46f15fff4: doc/issue_tracker: update resolution names
[11:46] <durandal_1707> hah, i can open trac issue with type:sponsoring_request and subject: pizza request!
[11:47] <michaelni> durandal_1707, iam trying to apply but it conflicts with 94d707af3c8d2eb7dde7048929b056a451daabcb
[11:47] <durandal_1707> ah it is already done....
[12:24] <Daemon404> michaelni, ping re: null deref patchset
[12:36] <ubitux> huh
[12:37] <ubitux> TRANSPOSE4x4W has a weird output layout
[12:37] <ubitux> i have this input:
[12:37] <ubitux> 27: $st0 = -nan(0xffe90011fff7000b)
[12:37] <ubitux> 26: $st1 = -nan(0xffe90022ffd3001c)
[12:37] <ubitux> 25: $st2 = -nan(0xffe9003cffa10036)
[12:37] <ubitux> 24: $st3 = -nan(0xffe9004dff7d0047)
[12:38] <ubitux> i'm doing a TRANSPOSE4x4W 0, 1, 2, 3, 4
[12:38] <ubitux> the output is correct but the out reg looks shuffled
[12:38] <ubitux> basically 1 is a "disabled" output
[12:39] <ubitux> 0, 2, 3, 4 contains the appropriate output, randomized a little
[12:42] <BBB> ubitux: look at the SWAP macro
[12:42] <BBB> ubitux: it does indeed shuffle them
[12:42] <BBB> ubitux: so m0 is no longer mm0
[12:42] <BBB> or so
[12:42] <BBB> maybe m0 stays m0, but m1 becomes mm4, etc.
[12:43] <BBB> TRANSPOSE4x4W uses BUTTERFLY which uses SWAP
[12:56] <ubitux> ok
[12:56] <ubitux> so i'm not doing something wrong
[12:56] <ubitux> i'll just play with those registers instead
[13:02] <cone-822> ffmpeg.git 03Michael Niedermayer 07master:52297945ecd6: doc/issue_tracker: remove more mentions of patches
[13:02] <cone-822> ffmpeg.git 03Michael Niedermayer 07master:6abb9eb52523: avformat/wavdec: Fix smv packet interleaving
[13:03] <BBB> ubitux: if output is correct and make fate-vp9 passes, you're done, really ;)
[13:03] <ubitux> yeah sure
[13:15] <ubitux> it seems i forgot to reload my code as well, so what i said wasn't exact either, just some shuffling
[13:47] <cone-822> ffmpeg.git 03Derek Buitenhuis 07master:530cd2893e02: tree-test: Don't return restricted exit codes
[13:47] <cone-822> ffmpeg.git 03Derek Buitenhuis 07master:8dc1b7bd228c: base64-test: Remove posibility of returning restricted exit codes
[13:47] <cone-822> ffmpeg.git 03Derek Buitenhuis 07master:5331d2b93fe2: dct-test: Remove possibility of returning a restricted exit code
[13:47] <cone-822> ffmpeg.git 03Derek Buitenhuis 07master:7177df90a043: fft-test: Remove possibility of returning restricted exit code
[13:47] <cone-822> ffmpeg.git 03Michael Niedermayer 07master:d9ad30507233: Merge remote-tracking branch 'qatar/master'
[13:56] <saste> who wants to review my channel_layout patch?
[13:56] <saste> i'll push it soon
[13:57] <durandal_1707> who wants to review my mergeplanes patch?
[13:57] <durandal_1707> i'll push it soon
[13:58] <wm4> is the libswscale filter thing pushed yet?
[14:01] <durandal_1707> saste: i see nothing obviously wrong/incorrect/misleading/evil/... in channel_layout patch
[14:01] <saste> wm4, not if no-one will review it
[14:01] <wm4> well, not like I'm a fan of it, it just blocks my own patch
[14:02] <saste> michaelni, waiting for your review ^^
[14:20] <ubitux> wth
[14:22] <ubitux> TRANSPOSE4x4W is doing some magic like randomizing my registers access
[14:22] Action: ubitux doesn't get it
[14:23] <durandal_1707> can't be
[14:24] <ubitux> so, i have this in my .asm: http://pastie.org/8429562
[14:24] <ubitux> generated output is correct
[14:24] <ubitux> ’ http://pastie.org/8429563
[14:25] <ubitux> now i add a TRANSPOSE4x4W 0, 1, 2, 3, 4 before those pxor
[14:25] <ubitux> this is what happen in the generated asm:
[14:25] <ubitux> http://pastie.org/8429566
[14:26] <ubitux> any idea wth is happening?
[14:26] <ubitux> BBB?
[14:30] <durandal_1707> ubitux: you expect it to be same?
[14:30] <ubitux> yes?
[14:31] <ubitux> the thing is, beginning of the function is matching with the code
[14:31] <ubitux> it's only post transpose that the output gets somehow fucked up
[14:32] <ubitux> i'm certainly missing something really stupid
[14:32] <BBB> ubitux: SWAP
[14:32] <BBB> ubitux: check the SWAP macro in x86inc.asm
[14:33] <BBB> ubitux: specifically, check how TRANSPOSE4x4W calls BUTTERFLY (or one of its variants), and that calls SWAP
[14:33] <ubitux> oh shit
[14:33] <BBB> it's expected, and it's ok
[14:33] <ubitux> yeah ok i see
[14:33] <BBB> SWAP is totally awesome
[14:33] <BBB> it's like the best thing about x86inc.asm
[14:33] <ubitux> so the yasm code is still consistent
[14:33] <BBB> perhaps only second to it allowing the same code for mmx, sse and avx
[14:34] <ubitux> it doesn't swap, it just redefines the registers?
[14:35] <BBB> right, it's a swap only in code-sense
[14:35] <ubitux> that's completely insane haha
[14:35] <BBB> it's a swap without having to emit an instruction
[14:35] <ubitux> i see
[14:35] <BBB> = insanely fast
[14:35] <ubitux> haha sure yeah
[14:36] <BBB> like "//" being a speed operator in c/c++
[14:36] <ubitux> i like this, but it doesn't help debugging with gdb :D
[14:36] <BBB> put before any line of code you'd like to speed up
[14:36] <BBB> swap is similarly cheaty
[14:37] <BBB> gotta go to work, bbl
[14:37] <BBB> send me email if you need immediate help
[14:38] <ubitux> no it's ok now thank you
[15:26] <saste> durandal_1707, i'm going to review mergeplanes
[15:29] <durandal_1707> that would be 5th review or so...
[15:45] <cone-822> ffmpeg.git 03Stefano Sabatini 07master:a3505136f966: doc/utils: document channel layout specification
[15:45] <cone-822> ffmpeg.git 03Stefano Sabatini 07master:b4991292aa83: doc: add a few links to the Channel Layout syntax section
[16:53] <durandal_1707> what a brain damage, libav disables crc checking by default
[16:53] <nevcairiel> why would crc be on by default
[16:53] <nevcairiel> it wastes cpu and users dont care
[16:53] <durandal_1707> it changes defaults
[16:54] <nevcairiel> isnt it off by default here too
[16:54] Action: Daemon404 fails to see a reason to keep it enabled by default
[16:58] <durandal_1707> nevcairiel: if it is off by default why would libav change default to off?
[16:58] <nevcairiel> maybe it was still on in whatever branch they have of the decoder?
[16:59] <durandal_1707> you are breaking stuff all the time
[16:59] <Daemon404> how DARE they break their own WIP patch
[16:59] <Daemon404> that you chose to merge early
[16:59] <Daemon404> what assholes!
[17:00] <durandal_1707> lol, no ffmpeg did not merged that crc thing....
[17:02] <cone-822> ffmpeg.git 03Michael Niedermayer 07master:f5dae4894d10: configure: abort if threading without atomics is used
[17:47] <cone-822> ffmpeg.git 03Michael Niedermayer 07master:cbb7f56dfeb1: metasound & twinvqdec: return the number of read bytes from read_bitstream
[17:47] <cone-822> ffmpeg.git 03Michael Niedermayer 07master:2e7a1fdd6d51: avcodec/twinvq: return the actual number of consumed bytes
[18:01] <Daemon404> [jpeg2000 @ 0x1b32400] unsupported marker 0xFF5F at pos 0x10B
[18:01] <Daemon404> [jpeg2000 @ 0x1b32400] Support for 4 components is not implemented. Update your FFmpeg version to the newest one from Git. If the problem still occurs, it means that your file has a feature which has not been implemented.
[18:01] <Daemon404> [jpeg2000 @ 0x1b32400] If you want to help, upload a sample of this file to ftp://upload.ffmpeg.org/MPlayer/incoming/ and contact the ffmpeg-devel mailing list.
[18:01] <Daemon404> [jpeg2000 @ 0x1b32400] error during processing marker segment ff90
[18:02] Action: Daemon404 wonders why this dcp file has alpha
[18:02] <Daemon404> :(
[18:02] <j-b> Daemon404: you have a DCP player?
[18:02] <Daemon404> yes it's called vlc
[18:02] <cbsrobot>  maybe it's used for projection mapping
[18:02] Action: Daemon404 runs
[18:02] <Daemon404> either way
[18:02] <ubitux> BBB: we will probably be able to re-use some of the mul/sum/round macro
[18:02] <Daemon404> no image is outout
[18:02] <Daemon404> output*
[18:03] <Daemon404> http://ftp.nluug.nl/pub/graphics/blender/demo/movies/ToS/ToS_4k_DCP.zip <-- playing with this
[18:03] <ubitux> BBB: http://pastie.org/8430133 so this is incomplete but it works
[18:03] <Daemon404> tears of steal
[18:03] <ubitux> BBB: only 1d
[18:03] <ubitux> i should be done "soon"©
[18:03] <durandal_1707> Daemon404: use other decoder?
[18:03] <Daemon404> "other" ?
[18:04] <Daemon404> the openjpeg one?
[18:04] <cbsrobot> Daemon404: fortunately it takes only 2 min to download
[18:04] <Daemon404> lol
[18:04] <durandal_1707> michaelni: that block_align change breaks metasount in asf
[18:04] <durandal_1707> *metasound
[18:05] <j-b> Daemon404: ok, sorry.
[18:05] <Daemon404> j-b, btw my friend says he saw some cinemas in either iceland or norway using vlc 
[18:05] <durandal_1707> for 3D?
[18:06] <Daemon404> no?
[18:06] <durandal_1707> pro cinemas?
[18:06] <j-b> Daemon404: yes, I'm not surprised.
[18:07] <j-b> Daemon404: hence the work on DCP for VLC
[18:07] <mateo`>  Daemon404: vlc can decrypt/play DCP packages and decode jpeg2000 real time now ? :D
[18:07] <j-b> we're not yet there
[18:07] <j-b> mateo`: big machine?
[18:07] Action: cbsrobot saw a 5d cinema in a little village in sardegna a few weeks ago !
[18:08] <cbsrobot> well, let's say aonly the ad said it was 5d
[18:08] <Daemon404> what are the 4th and 5th dimensions?
[18:08] <durandal_1707> 3D + 8ch audio?
[18:10] <durandal_1707> but not anaglyph
[18:10] <nevcairiel> Daemon404: i've seen movement and smell
[18:11] <Daemon404> i was hoping "time" and "space"
[18:11] <durandal_1707> smell?
[18:11] <Daemon404> i.e. free LSD
[18:13] <durandal_1707> sight + hearing + taste + smell + touch + 6th sense
[18:13] <durandal_1707> pain + balance are extra
[18:15] <saste> cbsrobot, then you was in calagonone
[18:19] <durandal_1707> but jpeg2000dec.c code supports 4th component
[18:26] <wm4> the alpha dimension?
[18:29] <durandal_1707> whant to find out? download 13G zip file
[18:29] <Daemon404> i have a 100 mb cut too
[18:30] <Daemon404> http://chromashift.org/sample.mxf
[18:42] <cbsrobot> saste: exactly !
[18:43] <saste> cbsrobot, it was easy :)
[18:44] <cbsrobot> saste: I was intreaged to go there, but I ended up eating pizza at the local pizza delivery shop
[18:45] <saste> cbsrobot, basically as everyone else does
[18:45] <cbsrobot> I hope not (for the cinema owner)
[18:46] <durandal_1707> Daemon404: looks like it is bacause of unsupported marker
[18:46] <cbsrobot> but the logo looked like a bad copy of a 3d software
[18:46] <cbsrobot> else I might have gone there & !
[18:46] <cbsrobot> *logo of a 3d software
[18:47] <durandal_1707> didn't you said it was 5d?
[18:47] <cbsrobot> well the software is called cinema 4d
[18:47] <cbsrobot> so you see the marketing concept !
[18:47] <durandal_1707> iirc that is not just software
[18:50] <durandal_1707> michaelni: block_align is required for demuxer that do not give proper packets to decoder
[18:50] <cbsrobot> just compare: http://www.vnix.com.sg/graphisoftsg/products/archicad/addons/cinema_4d/index.html vs http://imgur.com/Reb9WwP
[18:52] <durandal_1707> first have nothing in common with second
[18:59] <michaelni> durandal_1707, where can i find a file that my block alig change breaks ? (so i can fix it)
[19:02] <durandal_1707> and how would you fix that?
[19:03] <durandal_1707> the block_align is set depending on bitrate, etc...
[19:03] <durandal_1707> and in your commit that fix bug i see no explanation of problem it solves and why
[19:09] <michaelni> the packet size doesnt match block align in that file, the packet size also isnt constant in it IIRC
[19:09] <michaelni> the result where warnings/errors of packets being smaller than block align
[19:22] <durandal_1707> than it probably should return smaller one of two
[19:22] <durandal_1707> because if demuxer give big packet with multiple frames in it, decoder will most likely return only one
[19:22] <durandal_1707> all this sucks
[19:23] <durandal_1707> so either demuxer should be fixed, or decoder returns multiple frames....
[19:53] <ubitux> yay double idct working
[19:53] <ubitux> last step and i'm done
[19:55] <cone-822> ffmpeg.git 03Michael Niedermayer 07master:d04ec6ffc51c: avformat/rmdec: english typos
[19:55] <cone-822> ffmpeg.git 03Michael Niedermayer 07master:ba8716df7fb5: avformat/mp3dec: perform seek resync in the correct direction
[20:10] <cone-822> ffmpeg.git 03Michael Niedermayer 07master:08a89761964b: avcodec/h264: reduce noisiness of "mmco: unref short failure"
[20:19] <ubitux> can someone make me a summary of DEFINE_ARGS and STORE_DIFFx2 macros?
[20:19] <ubitux> BBB? Skyler_ maybe?
[20:37] <durandal_1707> i can't belive my eyes: [FFmpeg-user] Is "-q:v" the same as "-qscale:v"?
[21:25] <lukaszmluki> Hi, I have lame question, why there are 2 methods av_frame_alloc and avcodec_alloc_frame which seems to do the same?
[21:27] <wm4> lukaszmluki: because the ffmpeg API is freaking terrible
[21:28] <wm4> but use av_frame_alloc and ignore the other one
[21:28] <lukaszmluki> ok, thx
[21:36] <ubitux> lukaszmluki: avcodec_alloc_frame() if not ref counting frames
[21:36] <ubitux> av_frame_alloc otherwise
[21:37] <ubitux> lukaszmluki: see http://ffmpeg.org/pipermail/ffmpeg-devel/2013-September/147921.html
[21:37] <ubitux> lukaszmluki: nice to see you on irc btw :)
[21:38] <lukaszmluki> :)
[21:38] <wm4> ubitux: how is there a difference? because I don't think there is
[21:38] <lukaszmluki> I log in when need something :)
[21:38] <Daemon404> ...wait
[21:38] <wm4> and libavcodec is using libavutil incorrectly
[21:38] <Daemon404> we have alloc_frame and frame_alloc?
[21:38] Action: Daemon404 facedesks
[21:39] <lukaszmluki> thx for answer, make sense
[21:39] <ubitux> wm4: it was (and still is in the examples...) causing various memleak or invalid r/w
[21:39] <wm4> nice
[21:39] <ubitux> that's the reason i wrote that example
[21:40] <wm4> +    if (use_ref_counting)
[21:40] <wm4> +        frame = av_frame_alloc();
[21:40] <wm4> +    else
[21:40] <wm4> +        frame = avcodec_alloc_frame();
[21:40] <wm4> Daemon404 is going to get a heart attack
[21:40] <ubitux> :)
[21:40] <wm4> and me too
[21:40] <ubitux> feel free to propose better&
[21:40] <wm4> because what the flying fuck
[21:41] <ubitux> well, that's because of TEP
[21:41] <ubitux> and this is not really documented so far afaict
[21:42] <wm4> there doesn't seem to be much difference between these two functions
[21:42] <wm4> the frame defaults could be slightly different
[21:42] <Daemon404> ubitux, i luckily moved all my code over to TEP as soon as it was pushed
[21:43] <Daemon404> with a lot of yelling at its author for "docs"
[21:43] <Daemon404> <_<
[21:59] <ubitux> omg, reimar adding arm intrinsics to mplayer
[22:03] <wm4> so stupid, he could be using libavresample
[22:03] <wm4> or ffmpeg's nih'ed copy of it
[22:04] <ubitux> wm4: ... ?
[22:04] <wm4> like float<->int conversions for audio
[22:04] <wm4> pretty sure libavresample has asm for it, or would appreciate it
[22:04] <ubitux> (lavr is the nih of lsws)
[22:04] <ubitux> you're allowed to troll, but only when you are right
[22:05] <wm4> lavr was in the works, when mini ninja-nih'ed it
[22:05] <wm4> or something like this
[22:05] <ubitux> nih 1 year earlier? from a code he already wrote?
[22:05] <ubitux> come on...
[22:05] <ubitux> i'm pretty sure it was done before it was even started on the other side
[22:07] <wm4> *shrug*
[22:07] <wm4> libswresample is still less useful, because it's available only on one of the 2 projects
[22:08] <ubitux> i was just commenting on your nih comment
[22:08] <ubitux> feel free to escape the topic
[22:08] <ubitux> wm4: should we drop lavr from the project?
[22:09] <wm4> that would be way more sincere than keeping a disabled copy of it
[22:09] <wm4> but you won't do it because your chance you can force someone to use ffmpeg insteadof libav drops by 1%
[22:10] <ubitux> "force someone to use ffmpeg" lol
[22:10] <wm4> a patch removing libavresample from ffmpeg would make a good troll though
[22:10] <ubitux> yup
[22:11] <ubitux> let's add the macro wrapper instead
[22:12] <ubitux> ...and keep that wrapper disabled by default ;)
[22:12] <ubitux> -  --enable-avresample      enable libavresample build [no]
[22:12] <ubitux> +  --enable-avresample      enable libavresample wrapper build [no]
[22:54] <cone-822> ffmpeg.git 03lemonlinger at gmail.com 07master:2cf94485de9a: avformat/rtpproto: when local rtp port and local rtcp port are not specified, try to make two successive ports and retry 3 times when failed
[23:00] <Daemon404> line breaks, how do they work?
[23:12] <ubitux> BBB: i tried to stole vp8 code but it crashes for some reason (in STORE_DIFFx2); https://github.com/ubitux/FFmpeg/compare/vp9-asm
[23:12] <ubitux> for now the double idct is properly working (i'm getting lost in my own shuffling though)
[23:43] <ubitux> oh i think i get it
[23:45] <ubitux> doesn't crash anymore, almost there.
[00:00] --- Sat Oct 26 2013


More information about the Ffmpeg-devel-irc mailing list