[Ffmpeg-devel-irc] ffmpeg-devel.log.20120302

[burek]

[01:55] <Compn> hmm
[01:55] <Compn> that is a strange error
[01:55] <Compn> someone want to make a bugreport ?
[01:55] <Compn> http://video.in64.in/8182.mp4
[01:55] <Compn> ffmpeg -i 8182.mp4 out.avi
[02:28] <Daemon404> http://ffmpeg.org/libavfilter.html#asplit <-- can anyone explain how exactly to use this
[02:28] <Daemon404> the docs dont make it clear at all
[02:34] <Compn> saste can :P
[02:34] <Compn> when hes here
[03:09] <CIA-17> ffmpeg: 03Michael Niedermayer 07master * r7a7b1f5c4d 10ffmpeg/libavcodec/roqvideodec.c: 
[03:09] <CIA-17> ffmpeg: roqvideodec: improve end of input buffer check
[03:09] <CIA-17> ffmpeg: This fixes a out of array read.
[03:09] <CIA-17> ffmpeg: Found-by: Mateusz "j00ru" Jurczyk and Gynvael Coldwind
[03:09] <CIA-17> ffmpeg: Signed-off-by: Michael Niedermayer <michaelni at gmx.at>
[03:09] <CIA-17> ffmpeg: 03Derek Buitenhuis 07master * rd91912effa 10ffmpeg/tests/ (lavf-regression.sh ref/lavf/sun): 
[03:09] <CIA-17> ffmpeg: fate: Add sunrast regression test
[03:09] <CIA-17> ffmpeg: Signed-off-by: Derek Buitenhuis <derek.buitenhuis at gmail.com>
[03:09] <CIA-17> ffmpeg: Signed-off-by: Michael Niedermayer <michaelni at gmx.at>
[03:09] <CIA-17> ffmpeg: 03Derek Buitenhuis 07master * r83c418e68e 10ffmpeg/libavcodec/wavpack.c: 
[03:09] <CIA-17> ffmpeg: wavpack: Fix an integer overflow
[03:09] <CIA-17> ffmpeg: Integer Overflow Checker detected an integer
[03:09] <CIA-17> ffmpeg: overflow while FATE was running.
[03:09] <CIA-17> ffmpeg: Signed-off-by: Derek Buitenhuis <derek.buitenhuis at gmail.com>
[03:09] <CIA-17> ffmpeg: Signed-off-by: Michael Niedermayer <michaelni at gmx.at>
[03:55] <funman> michaelni: is there a way i can have almost lossless h264 encoding using lossy profiles?
[03:56] <funman> i am reverse engineering a pixel format and my idea is to have one macroblock in the top left corner using distinct luma values for each pixels (trying to figure only luma first)
[03:56] <funman> it's a so-called 'tiled' format
[03:57] <funman> i guess i should try to encode+decode a frame and see how it goes out
[04:00] <michaelni> funman, simply using a low enough quantizer should get you close to lossless
[04:00] <michaelni> one also could force a MB to be coded as PCM 
[04:00] <michaelni> then that mb would be lossless
[04:45] <funman> http://pastie.org/3501447 :)
[04:48] <Compn> reverse engineering a pixel format
[04:48] <Compn> by having near lossless h264 encoding
[04:48] <Compn> i have no idea how this works , but ok :)
[04:48] <funman> Compn: my tablet h264 decoder spits out some unknown format
[04:49] <funman> usually it's drawn directly in a hardware surface but that prevents e.g. subs
[04:49] <Compn> oh
[04:49] <Compn> you want the colorspace it outputs ?
[04:49] <funman> yeah
[04:49] <Compn> were you able to dump it ?
[04:50] <Compn> i mean, how did you dump the pixelformat ?
[04:50] <funman> didn't do it yet
[04:50] <Compn> ...
[04:50] <funman> vlc gets the decoded picture
[04:50] <funman> so it's as simple as add a open()+read()
[04:50] <funman> write8
[04:50] <funman> write*
[04:50] <funman> not so simple in fact :(
[04:51] <Compn> is your tablet open source ? :P
[04:51] <funman> yeah it runs CM9
[04:51] <Compn> you could just look at the source to find out what pixelformat it is
[04:51] <Compn> unless i'm misunderstanding something
[04:51] <funman> unfortunately the DSP is not opensource
[04:52] <Compn> ohh
[04:52] <Compn> so its an h264 pixel format ?
[04:52] <Compn> like h264 4:4:4 ?
[04:52] <funman> what is open sourced is 'put encoded picture here', 'set bit here', 'get decoded picture here'
[04:52] <Compn> right
[04:52] <funman> and: 'set picture size here', 'set bit here', 'put decoded picture here'
[04:52] <Compn> they never open source the dsp :\
[04:52] <funman> then it magically appears on the screen
[04:54] <funman> wbs started working on this format (it's from qualcomm)
[05:06] <funman> -qp 0 gives losses, i suppose i need to make a yuv picture
[05:11] <funman> yeah that works better
[05:43] <Daemon404> funman, what format?
[05:45] <funman> QOMX_COLOR_FormatYUV420PackedSemiPlanar64x32Tile2m8ka
[05:45] <Daemon404> god thats a ugly name
[05:46] <Daemon404> where's t even used?
[05:47] <funman> on QualComm OMX
[05:48] Action: funman -> bed
[06:41] <CIA-17> ffmpeg: 03Michael Niedermayer 07master * r0b90db01b5 10ffmpeg/ (libavformat/utils.c tests/ref/fate/pva-demux): 
[06:41] <CIA-17> ffmpeg: lavf: fix update_initial_durations() so it handles missing durations with the initial timestamp being known.
[06:41] <CIA-17> ffmpeg: This fixes duplicate timestamps on mp2 in ts with non seekable input.
[06:41] <CIA-17> ffmpeg: It also fixed the fate pva demux timestamps.
[06:41] <CIA-17> ffmpeg: Signed-off-by: Michael Niedermayer <michaelni at gmx.at>
[07:11] <bparker> I'm trying to use libswscale (it is working, but a little too slow) and I get this message:
[07:11] <bparker> [swscaler @ 0x1bc74a0] Warning: data is not aligned! This can lead to a speedloss
[07:11] <bparker> any idea how I can fix this issue?
[07:13] <nevcairiel> ensure the data is at least mod16 aligned, most importantly use an aligned malloc (av_malloc will do the job), and also pad the video to be mod16 on all planes (so for YUV420, pad Y to mod32, U/V to mod16)
[07:15] <bparker> my video is UYVY422 (packed instead of planar) and I'm giving sws_scale an offset into an existing buffer, and that offset is what is giving me the message
[07:16] <bparker> if I give it a pointer to the actual beginning of the data, there is no complaint
[07:16] <nevcairiel> whatever memory pointer you give it, it needs to be mod16 aligned
[07:17] <bparker> hmm
[07:18] <bparker> I'm wondering how I would do that without re-copying the image somewhere else
[07:24] <bparker> apparently using av_malloc and copying the data there, it still complains
[07:24] <bparker> and using av_freep afterwards (instead of av_free) segfaults inside of libc for some strange reason...
[07:31] <bparker> nevcairiel: simply calling sws_scale twice with the same exact options produces the alignment warning... how is that possible?
[07:31] <bparker> if I remove the 2nd call, the warning goes away
[08:44] <Haferstroh> Hello, nobody on the user-channel could help answering my question so I give it a try here...
[08:45] <Haferstroh> my command for screencasting is this:
[08:45] <Haferstroh> ffmpeg -f x11grab -r 25 -s 1280x800 -i :0.0 -vcodec mszh -sameq  -f alsa -i hw:0 -acodec libvorbis bildschirmvideo2.mkv
[08:46] <Haferstroh> If I delete the option -sameq, then the videoquality is not so good than with this option enabled, allthough mszh is a lossless videcodec
[08:46] <Haferstroh> why is this?
[12:42] <kriegerod> hey, is there a known issue with running transcoding (both using API and utility) under Valgrind, on amd64 box? Getting this with both ffmpeg and libav git HEADs: http://pastebin.com/T6Z1VZT0
[13:57] <Tjoppen> ah, that mxfdec was a couple of days ago and mostly applied :)
[13:57] <Tjoppen> +thread
[14:04] <cbsrobot> lucky you :)
[15:34] <michaelni> kriegerod, from a quicklook it looks like a x264 vs. valgrind issue
[15:36] <kriegerod> michaelni, yep, thanks, i was already told it can be x264 code. Switched to encoding with codec from inside ffmpeg (built with debug), works under valgrind.
[16:19] <kierank> kriegerod: happens with some swscale asm as well
[16:22] <CIA-17> ffmpeg: 03Michael Niedermayer 07master * rec3cd74f2d 10ffmpeg/libavcodec/h261dec.c: 
[16:22] <CIA-17> ffmpeg: h261: check mtype.
[16:22] <CIA-17> ffmpeg: Fixes out of array read
[16:22] <CIA-17> ffmpeg: Found-by: Mateusz "j00ru" Jurczyk and Gynvael Coldwind
[16:22] <CIA-17> ffmpeg: Signed-off-by: Michael Niedermayer <michaelni at gmx.at>
[16:22] <CIA-17> ffmpeg: 03Michael Niedermayer 07master * ra5bf9b351c 10ffmpeg/libavdevice/timefilter.c: 
[16:22] <CIA-17> ffmpeg: timefilter: better selftest code.
[16:22] <CIA-17> ffmpeg: Add support for divergence detection and variable period.
[16:22] <CIA-17> ffmpeg: Signed-off-by: Michael Niedermayer <michaelni at gmx.at>
[17:12] <CIA-17> ffmpeg: 03Michael Niedermayer 07master * re93d911e48 10ffmpeg/libavcodec/ituh263dec.c: 
[17:12] <CIA-17> ffmpeg: h263: fix zygo debug printing overreading.
[17:12] <CIA-17> ffmpeg: Found-by: Mateusz "j00ru" Jurczyk and Gynvael Coldwind
[17:12] <CIA-17> ffmpeg: Signed-off-by: Michael Niedermayer <michaelni at gmx.at>
[17:13] <CIA-17> ffmpeg: 03Michael Niedermayer 07master * rd964db5742 10ffmpeg/libavcodec/wmadec.c: 
[17:13] <CIA-17> ffmpeg: wmadec: fix off by 1 error on the pow_tab index check.
[17:13] <CIA-17> ffmpeg: Fixes global out of array read.
[17:13] <CIA-17> ffmpeg: Found-by: Mateusz "j00ru" Jurczyk and Gynvael Coldwind
[17:13] <CIA-17> ffmpeg: Signed-off-by: Michael Niedermayer <michaelni at gmx.at>
[17:59] <Daemon404> michaelni, ping
[18:00] <michaelni> Daemon404, pong
[18:01] <Daemon404> michaelni, for teh ra144 and aac tests
[18:01] <Daemon404> after talking to ruggles, i dont think they should go in
[18:01] <Daemon404> sicne theyre float-based
[18:01] <Daemon404> and can differ on platforms
[18:01] <Daemon404> (we dont have a fuzzy audio testing methid afaik)
[18:01] <michaelni> i realized this too and thought try and revert
[18:02] <Daemon404> OK
[18:02] <michaelni> also at least the decoded data could be fuzzy compared
[18:03] <michaelni> the encoded files, if they differ cant really be compared
[18:03] <Daemon404> i dont think fate can do this at the moment
[18:03] <Daemon404> at least not codec-regressionsh
[18:03] <Daemon404> /sh/.sh/
[18:03] <michaelni> maybe
[18:05] <michaelni> but if not, it could be added, i suspect this should not be hard
[18:06] <Daemon404> yea
[18:06] <michaelni> also tiny_psnr is fuzzy in the sense that it does not show infinite precission
[18:07] <Daemon404> thats not something greta to rely on for fuzzy testing though
[18:08] <michaelni> no but if it happens to work, no need to do further work :)
[18:08] <Daemon404> lol.
[18:08] <Daemon404> BTW, i did test ra144 and aac on ARM as well as x86_64
[18:08] <Daemon404> but i might have just gotten lucky
[18:26] <michaelni> does anyone have a recent version of S421m ? (mine is from 2005)
[18:29] <Daemon404> ... oh internet....
[18:29] <Daemon404> google for s421m
[18:29] <Daemon404> find porn
[18:29] <Daemon404> -_-
[18:30] <michaelni> google for anything
[18:30] <michaelni> find porn
[18:30] <Daemon404> indeed
[18:32] <cbsrobot> search for porn and find google
[18:35] <MP4_maniac> michaelni: do you want 2006?
[18:38] <michaelni> MP4_maniac, yes
[18:41] <MP4_maniac> you can amd1 and amd2 from smpte site
[18:42] <MP4_maniac> http://www.smpte.org/s?as_q=smpte%20421m
[18:42] <MP4_maniac> s/can/can download/
[18:42] <michaelni> MP4_maniac, Thanks!
[18:46] <nevcairiel> stupid vc1, reminds me i should figure out how the mv prediction in interlaced b-frames works and finish those patches
[18:46] <nevcairiel> the whole vc1 mv code seems like voodoo to me
[18:51] <JEEB> not to mention that there seem to be older VC-1 streams that only the official decoder can decode correctly (mainconcept and friends can't handle them)
[18:52] <nevcairiel> i only know about that for wmv3, where they changed the syntax last-minute, so streams encoded with the beta version don't work anymore
[18:57] <michaelni> JEEB, where can i find such file ?
[18:59] <JEEB> not sure if it was the LAV Filters' thread or the MPC-HC thread or what
[18:59] <nevcairiel> i dont remember
[19:00] <nevcairiel> cruncher probably posted it
[19:00] <JEEB> aye
[19:00] <nevcairiel> he always digs up the broken files. :p
[19:01] <nevcairiel> http://www.mediafire.com/?lh785fl24aukxja
[19:02] <nevcairiel> found it ^
[19:02] <nevcairiel> kostya has some posts about that specific problem on his blog, apparently the encoding of p-frames is different somehow
[19:02] <nevcairiel> but he never figured out how exactly
[19:03] <CIA-17> ffmpeg: 03Michael Niedermayer 07master * r8fdd93eaad 10ffmpeg/libavcodec/huffyuv.c: 
[19:03] <CIA-17> ffmpeg: huffyuv: pad classic huffman tables so as to avoid bitreader overread.
[19:03] <CIA-17> ffmpeg: Found-by: Mateusz "j00ru" Jurczyk and Gynvael Coldwind
[19:03] <CIA-17> ffmpeg: Signed-off-by: Michael Niedermayer <michaelni at gmx.at>
[19:03] <CIA-17> ffmpeg: 03Michael Niedermayer 07master * r64c58f1436 10ffmpeg/libavcodec/vc1dec.c: 
[19:03] <CIA-17> ffmpeg: vc1: mquant is not allowed to be 0
[19:03] <CIA-17> ffmpeg: Fixes out of bounds read.
[19:03] <CIA-17> ffmpeg: Checked against SMPTE 421M-2006
[19:03] <CIA-17> ffmpeg: Found-by: Mateusz "j00ru" Jurczyk and Gynvael Coldwind
[19:03] <CIA-17> ffmpeg: Signed-off-by: Michael Niedermayer <michaelni at gmx.at>
[19:07] <Daemon404> who is this j00ru person
[19:07] <Daemon404> his name has been appearing in tons of commits
[19:07] <funman> google security guy i htink
[19:08] <Daemon404> oic
[19:17] <michaelni> JEEB, thx for the file, it doesnt look easy to fix from a very quick look
[21:14] <CIA-17> ffmpeg: 03Michael Niedermayer 07master * r0af48e29f5 10ffmpeg/libavcodec/snowdec.c: 
[21:15] <CIA-17> ffmpeg: snow: check reference frame indices.
[21:15] <CIA-17> ffmpeg: Fixes NULL ptr dereference
[21:15] <CIA-17> ffmpeg: Found-by: Mateusz "j00ru" Jurczyk and Gynvael Coldwind
[21:15] <CIA-17> ffmpeg: Signed-off-by: Michael Niedermayer <michaelni at gmx.at>
[21:50] <CIA-17> ffmpeg: 03Michael Niedermayer 07master * ra3f5ee297a 10ffmpeg/libavcodec/mjpegdec.c: 
[21:50] <CIA-17> ffmpeg: mjpeg: Check for interlaced progressive frames
[21:50] <CIA-17> ffmpeg: Fixes null pointer dereference.
[21:50] <CIA-17> ffmpeg: Found-by: Mateusz "j00ru" Jurczyk and Gynvael Coldwind
[21:50] <CIA-17> ffmpeg: Signed-off-by: Michael Niedermayer <michaelni at gmx.at>
[21:50] <CIA-17> ffmpeg: 03Michael Niedermayer 07master * re7b43e8e84 10ffmpeg/libavcodec/truemotion1.c: 
[21:50] <CIA-17> ffmpeg: truemotion1: Check input buffer size against header size.
[21:50] <CIA-17> ffmpeg: Fixes overread.
[21:50] <CIA-17> ffmpeg: Found-by: Mateusz "j00ru" Jurczyk and Gynvael Coldwind
[21:50] <CIA-17> ffmpeg: Signed-off-by: Michael Niedermayer <michaelni at gmx.at>
[22:42] <CIA-17> ffmpeg: 03Michael Niedermayer 07master * r422e3a74b9 10ffmpeg/libavcodec/rawdec.c: 
[22:42] <CIA-17> ffmpeg: rawdec: fix input overread.
[22:42] <CIA-17> ffmpeg: Found-by: Mateusz "j00ru" Jurczyk and Gynvael Coldwind
[22:42] <CIA-17> ffmpeg: Signed-off-by: Michael Niedermayer <michaelni at gmx.at>
[22:43] <CIA-17> ffmpeg: 03Michael Niedermayer 07master * r689f65126b 10ffmpeg/libavcodec/simple_idct.c: 
[22:43] <CIA-17> ffmpeg: simple_idct: idct_4col_put: Fix out of array reads.
[22:43] <CIA-17> ffmpeg: Found-by: Mateusz "j00ru" Jurczyk and Gynvael Coldwind
[22:43] <CIA-17> ffmpeg: Signed-off-by: Michael Niedermayer <michaelni at gmx.at>
[00:00] --- Sat Mar  3 2012