[Ffmpeg-devel-irc] ffmpeg-devel.log.20120110

burek burek021 at gmail.com
Wed Jan 11 02:05:02 CET 2012

[00:13] <OanaStra_> michaelni: ping
[00:30] <CIA-46> ffmpeg: 03Michael Niedermayer 07master * r73f2cf4ed9 10ffmpeg/ffplay.c: 
[00:30] <CIA-46> ffmpeg: ffplay: Exit on ctrl-c.
[00:30] <CIA-46> ffmpeg: This allows to get out of ffplay if it or SDL got stuck.
[00:30] <CIA-46> ffmpeg: This for example happens when the audio driver is playing something
[00:30] <CIA-46> ffmpeg: else and doesnt support mixing multiple sources.
[00:30] <CIA-46> ffmpeg: Signed-off-by: Michael Niedermayer <michaelni at gmx.at>
[00:30] <CIA-46> ffmpeg: Signed-off-by: Marton Balint <cus at passwd.hu>
[00:30] <CIA-46> ffmpeg: 03Michael Niedermayer 07master * r4805a33043 10ffmpeg/: 
[00:30] <CIA-46> ffmpeg: Merge remote-tracking branch 'cus/stable'
[00:30] <CIA-46> ffmpeg: * cus/stable:
[00:30] <CIA-46> ffmpeg:  ffplay: Exit on ctrl-c.
[00:30] <CIA-46> ffmpeg: Merged-by: Michael Niedermayer <michaelni at gmx.at>
[01:32] <durandal_1707> when converting single image ffmpeg doesn't ask for permission to overwrite file
[01:39] <OanaStra_> michaelni: ping
[02:13] <michaelni> OanaStra_, pong
[02:14] <OanaStra_> are you busy ? i am working on a bug 
[02:14] <OanaStra_> and i have some questions
[02:14] <OanaStra_> https://ffmpeg.org/trac/ffmpeg/ticket/890
[02:15] <michaelni> iam always busy :(
[02:16] <OanaStra_> :(
[02:16] <michaelni> just ask the questions ...
[02:17] <OanaStra_> durandal_1707: said that c->cur and c->prev are not init and that`s why it`s hanging
[02:17] Action: durandal_1707 was wrong
[02:17] <OanaStra_> and i have init them in decode_init with NULL and check if they are init when using them 
[02:17] <OanaStra_> but afterwards we found out that the valgrind report is wrong
[02:18] <OanaStra_> have any idea about this ?
[02:20] <michaelni> so what kind of bug is this, OOM , hang, crash ?
[02:20] <OanaStra_> i think it`s combined OOM with a hang
[02:20] <OanaStra_> hanging and killing memory
[02:21] <michaelni> what is allocating all the memory ?
[02:22] <OanaStra_> i did not find out
[02:23] <OanaStratulat> moved to my linux machine, valgrind is buggy on osx
[02:23] <OanaStratulat> need the gdb report?
[02:28] <OanaStratulat> gdb : http://pastebin.com/64gqLSv4   valgrind : http://pastebin.com/2P1Fdeib
[02:32] <michaelni> OanaStra_, durandal_1707 the issue is simply that (probably due to timestamps) frames get duplicated 33554431 times
[02:32] <michaelni> -vsync 0 prevents this
[02:32] <OanaStra_> and what must be verified if the user does not use this option ?
[02:32] <michaelni> and for muxing these many frames are then buffered
[02:33] <michaelni> the timestamping code in ffmpeg is not trivial
[02:33] <OanaStra_> so no luck for a trivial solution ?
[02:33] <michaelni> faulty timestamps should be detected somehow and ignored
[02:34] <michaelni> a threashold above which a timestamp is considered faulty if the source doesnt support ts discontinuities should work
[02:35] <OanaStra_> please be more explicit
[02:37] <michaelni> check AVFMT_TS_DISCONT isnt set for the source && FFABS(last valid ts - current ts) > 1 hour && last ts known && current ts known then replace current ts by "interpolated value" or unknown 
[02:38] <OanaStra_> pff, if you have time to guide me. this is too advance for me
[02:41] <michaelni> maybe tomorrow
[02:42] <OanaStra_> can you extend my deadline for this
[02:42] <OanaStra_> http://google-melange.appspot.com/gci/task/view/google/gci2011/7180239
[02:42] <michaelni> done
[02:42] <OanaStra_> thx, i will ping you 2morrow
[02:43] <michaelni> ok
[03:52] <CIA-46> ffmpeg: 03Diego Biurrun 07master * r83890c66c4 10ffmpeg/libswscale/ (ppc/yuv2rgb_altivec.h rgb2rgb.h swscale.h swscale_internal.h): swscale: K&R reformatting cosmetics for header files
[03:52] <CIA-46> ffmpeg: 03Daniel Huang 07master * r8549cf073b 10ffmpeg/libavcodec/mjpegdec.c: 
[03:52] <CIA-46> ffmpeg: mjpegdec: K&R formatting cosmetics
[03:52] <CIA-46> ffmpeg: Signed-off-by: Diego Biurrun <diego at biurrun.de>
[03:52] <CIA-46> ffmpeg: 03Diego Biurrun 07master * r999484c9fb 10ffmpeg/libswscale/ (colorspace-test.c swscale-test.c): swscale: K&R formatting cosmetics for code examples
[03:52] <CIA-46> ffmpeg: 03Mike Melanson 07master * ra83f5b8a96 10ffmpeg/tests/ (fate/video.mak ref/fate/cvid-grayscale): 
[03:52] <CIA-46> ffmpeg: FATE test: cvid-grayscale; ensures that the grayscale Cinepak variant is exercised.
[03:52] <CIA-46> ffmpeg: Signed-off-by: Ronald S. Bultje <rsbultje at gmail.com>
[03:52] <CIA-46> ffmpeg: 03Mike Melanson 07master * r599855748a 10ffmpeg/tests/ (fate/demux.mak ref/fate/wtv-demux): 
[03:52] <CIA-46> ffmpeg: FATE: test to exercise WTV demuxer.
[03:52] <CIA-46> ffmpeg: Does not attempt to decode perceptual codecs inside.
[03:52] <CIA-46> ffmpeg: Code coverage: libavformat/wtv.c: 0% -> 71%
[03:52] <CIA-46> ffmpeg: Signed-off-by: Ronald S. Bultje <rsbultje at gmail.com>
[03:52] <CIA-46> ffmpeg: 03Vitor Sessak 07master * r421c99a4a7 10ffmpeg/libavcodec/mpegaudiodec.c: 
[03:52] <CIA-46> ffmpeg: mpegaudiodec: interleave iMDCT buffer to simplify future SIMD implementations
[03:52] <CIA-46> ffmpeg: Signed-off-by: Ronald S. Bultje <rsbultje at gmail.com>
[03:52] <CIA-46> ffmpeg: 03Vitor Sessak 07master * r6dfcf53092 10ffmpeg/libavcodec/ (4 files): 
[03:52] <CIA-46> ffmpeg: mpegaudiodec: move imdct and windowing function to mpegaudiodsp
[03:52] <CIA-46> ffmpeg: Signed-off-by: Ronald S. Bultje <rsbultje at gmail.com>
[03:52] <CIA-46> ffmpeg: 03Vitor Sessak 07master * r06677d0dd9 10ffmpeg/libavcodec/ (mpegaudiodsp.h mpegaudiodsp_template.c): 
[03:52] <CIA-46> ffmpeg: mpegaudiodec: change imdct window arrangment for better pointer alignment
[03:52] <CIA-46> ffmpeg: Signed-off-by: Ronald S. Bultje <rsbultje at gmail.com>
[03:53] <CIA-46> ffmpeg: mpegaudiodec: optimized iMDCT transform
[03:53] <CIA-46> ffmpeg: Signed-off-by: Ronald S. Bultje <rsbultje at gmail.com>
[03:53] <CIA-46> ffmpeg: 03Mike Melanson 07master * r2dee0cdb8e 10ffmpeg/tests/ (fate/demux.mak ref/fate/smjpeg): 
[03:53] <CIA-46> ffmpeg: FATE: add test for SMJPEG demuxer and associated IMA ADPCM audio decoder.
[03:53] <CIA-46> ffmpeg: (Don't attempt to decode JPEG data.)
[03:53] <CIA-46> ffmpeg: Code coverage: libavformat/smjpeg.c: 0% -> 69%
[03:53] <CIA-46> (36 lines omitted)
[04:10] <durandal_1707> hmm why is x86_64-linux...gcc4.6 failing for v410dec?
[04:10] <durandal_1707> that codes should be pretty trivial
[04:11] <durandal_1707> or perhaps it is still missing samples?
[07:08] <CIA-46> ffmpeg: 03anonymous 07master * r7e2ba2df28 10ffmpeg/libavcodec/mjpegdec.c: mjpeg: update whitespaces to qatar
[07:37] <ubitux> anonymous?
[08:20] <CIA-46> ffmpeg: 03Michael Niedermayer 07master * r2169f971ad 10ffmpeg/tests/fate/microsoft.mak: 
[08:20] <CIA-46> ffmpeg: fate: Disable fate-wmv8-x8intra
[08:20] <CIA-46> ffmpeg: This test does not work on all platforms and until it does
[08:20] <CIA-46> ffmpeg: it just hides new failures, which is really bad.
[08:20] <CIA-46> ffmpeg: Signed-off-by: Michael Niedermayer <michaelni at gmx.at>
[12:05] <NapHtaKeRoSene> hi
[12:05] <NapHtaKeRoSene> anybody know how to call a pointer to function in linux C ?
[12:06] <Tjoppen> fnptr(args);
[12:06] <Tjoppen> works pretty much like any other function
[13:35] <cbsrobot> Tjoppen: any news on the mxf bug ?
[13:35] <cbsrobot> [mxf @ 0x10700a400] failed to find absolute offset of 5b0dd in BodySID 1 - partial file?
[13:51] <Tjoppen> nope. I didn't manage to reproduce it earlier
[13:55] <Tjoppen> ehm, avcodec_align_dimensions2() sets stride_align = 8 for some codecs even though MMX is enabled, leading swscale to complain if such a frame is passed to it
[14:21] <cbsrobot> Tjoppen: you can reproduce it with the DCPTEST.V.mxf file you downloaded a few weeks ago
[14:33] <Tjoppen> I tried with that file. it didn't play with the old version either
[14:34] <Tjoppen> michaelni: since f44ad92c libswscale complains about frames given by the mpeg2 decoder, due to STRIDE_ALIGN = 8
[14:44] <michaelni> Tjoppen, ill look into it
[14:45] <Tjoppen> avcodec_align_dimensions2() should probably set stride_align = 16 if SSE2 is enabled. BBB seems to agree the current behavior is a bug
[14:47] <michaelni> yes, but last i tried setting it to 16 broke something ...
[14:47] <michaelni> that is probably a fate test
[14:48] <michaelni> but i dont remember precisely
[14:49] <Tjoppen> I'm also a bit confused why ffmpeg.c doesn't trigger that nag. my guess it has to do with the scaling being done via lavfi nowadays
[14:51] <michaelni> btw, there are 2 mxf patches on the ML
[14:51] <michaelni> [PATCH] mxfdec: do not fail on error parsing footer. & [FFmpeg-devel] [PATCH] mxfdec: support AVFMT_FLAG_IGNIDX.
[14:52] <Tjoppen> ah, missed that
[14:52] <Tjoppen> three actually
[14:55] <ubitux> ergonomy question: i'd like to add an option to amerge, allowing to merge only a subset of channels, what syntax would you prefer?: amerge=0|0:1 (ch0 for source0, ch0 and ch1 for source1), amerge=0:0|1 (same), amerge=a=0:b=0,1 (same), something else?
[14:57] <cbsrobot> Tjoppen: there was a bug in openjpegdec aswell
[14:57] <cbsrobot> actually it should b
[14:57] <cbsrobot> play, but with wrong colors
[14:59] <cbsrobot> and it plays only with a few fps
[15:06] <michaelni> ubitux, a:a+b, this way one also could do a:0.3*a+0.7*b
[15:06] <ubitux> well then, better prefix with a af pan filter
[15:35] <Tjoppen> wait a minute.. that file claims to be opatom but it's not clip wrapped
[15:42] <ubitux> (mmh i think i'll just port the "filtering" of mapchannel into a filter)
[15:47] <Tjoppen> there we go. I get all frames up to 01:02:04:23
[15:56] <ubitux> michaelni: does libswr lack some features of af_aconvert, af_aformat and af_aresample?
[15:56] <Tjoppen> cbsrobot: posting a patch to the ML
[16:10] <cbsrobot> Tjoppen: compiling ...
[16:12] <cbsrobot> yay - it works - thaks alot
[16:35] <michaelni> ubitux, not that i know of
[16:36] <ubitux> it would be nice to port them to libswr&
[16:36] <ubitux> i'm atm working on porting the map channel feature into a filter
[16:36] <ubitux> (filtering/reordering)
[16:36] <ubitux> and i'll certainly use libswr for this
[16:37] <ubitux> i wonder how simple libswr will integrates itself with lavfi
[18:00] <ubitux> michaelni: swr can deal with planar only internally, right?
[18:00] <ubitux> like i can ask non planar input & output
[18:01] <durandal_1707> planar what?
[18:01] <ubitux> i have a strange "Assertion a->planar failed at libswresample/swresample.c:284" in swr_convert when trying to do non-planar 
[18:01] <ubitux> durandal_1707: planar audio
[18:06] <durandal_1707> so you are doing conversion from planar to non-planar?
[18:06] <durandal_1707> full ffmpeg output please
[18:06] <ubitux> non-planar to non-planar, packed to packed
[18:07] <ubitux> there is no particular output, i'm using the api
[18:07] <ubitux> output: chl=5.1 | fmt=00000001 | srate=44100
[18:07] <ubitux> input:  chl=5.1 | fmt=00000001 | srate=44100
[18:07] <ubitux> this are the swr_alloc_set_opts main parameters
[18:08] <ubitux> and then just a call to swr_convert
[18:08] <ubitux> with the buffers & nb samples
[18:10] <durandal_1707> my hunch planar output is not supported, i'm now busy with swscale
[18:10] <ubitux> swr is using planar internally
[18:11] <ubitux> what i'm wondering is if there is some packed limitation about the input/output
[18:11] <ubitux> which sounds weird
[18:14] <ubitux> (fmt->s16)
[18:24] <michaelni> ubitux, planar in/out should worl
[18:24] <michaelni> work
[18:24] <michaelni> durandal_1707, swscale?
[18:24] <michaelni> what exactly ?
[18:25] <durandal_1707> writing functions
[18:25] <ubitux> michaelni: what about packed in/out ?
[18:26] <CIA-46> ffmpeg: 03Stefano Sabatini 07master * rd3e435164b 10ffmpeg/ffprobe.c: 
[18:26] <CIA-46> ffmpeg: ffprobe: make upcase_string() ignore non-ASCII characters
[18:26] <CIA-46> ffmpeg: This is required as some section names may contain non-ASCII characters
[18:26] <CIA-46> ffmpeg: (e.g. '_').
[18:26] <CIA-46> ffmpeg: 03Stefano Sabatini 07master * r5226be0dd5 10ffmpeg/ (Changelog doc/ffprobe.texi doc/ffprobe.xsd ffprobe.c): ffprobe: add -show_program_version and -show_library_versions options
[18:26] <CIA-46> ffmpeg: 03Stefano Sabatini 07master * r93d49cba7f 10ffmpeg/ (doc/ffprobe.texi ffprobe.c): ffprobe: add compact option to JSON writer
[18:26] <michaelni> ubitux, packed in out should work too
[18:27] <michaelni> durandal_1707, what functions ?
[18:27] <durandal_1707> michaelni: rgb2rgb stuff
[18:28] <ubitux> michaelni: then any idea what could go wrong in: http://pastie.org/3161321 ? (output on top of the paste)
[18:28] <michaelni> durandal_1707, can you submit patches against ffmpegs swscale ?
[18:28] <michaelni> iam asking because ours is in a little better shape
[18:29] <ubitux> (i'm using in sample rate in both case because out sample rate isn't defined, but that should cause any issue)
[18:30] <michaelni> i know swresample_test.c worked with planar & packed
[18:31] <durandal_1707> michaelni: what is better?, i can create swscale branch on github, and you can do with it whatever you want (beware of rebases)
[18:33] <michaelni> our swscale has many more bugfixes
[18:33] <michaelni> which reminds me i have to push one :)
[18:34] <michaelni> ubitux, which of the realloc_audio() calls is failing thea assert =
[18:34] <michaelni> ?
[18:53] <durandal_1707> michaelni: but my swscale changes depends on one of commits in qatar which is not yet merged into ffmpeg master
[18:54] <michaelni> why do you do the work based on the qatar version ?
[18:54] <michaelni> i mean why dont you submit sws patches to ffmpeg-dev ?
[18:56] <michaelni> IMHO it would be better if everyone would work on the same swscale
[18:56] <michaelni> and that should be the version that is technically in better shape
[18:57] <durandal_1707> michaelni: i do work on ffmpeg swscale, and that port that to libav and the submit to them
[18:57] <durandal_1707> s/that/than
[18:57] <michaelni> that sounds kinda hard
[18:57] <durandal_1707> indeed
[18:58] <durandal_1707> just bunch of extra git fp, git am, git fp, git sm
[18:58] <michaelni> i mean i really think we all (inclusing ronald) should somehow work together on sws
[18:58] <durandal_1707> fp,am,sm are aliases ...
[18:59] <michaelni> i exchanged some emails about this already with ronald but got no reply to my last :(
[19:00] Action: michaelni hates politics
[19:01] <gnafu> My official life motto is, "Politics suck."
[19:02] <gnafu> michaelni: Were you going to try to meet up at FOSDEM?
[19:02] <durandal_1707> michaelni: pushed sws branch
[19:03] Action: gnafu remembers a conversation about FOSDEM the other day, but doesn't remember where it left off.
[19:03] <av500> gnafu: that would imply actually going
[19:03] <gnafu> av500: Of course.
[19:10] <ubitux> michaelni: http://pastebin.com/5FV1tMNX
[19:11] <ubitux> (4096 samples this time)
[19:13] <michaelni> it asserts on a->ch_count not planar unless my line numbers are off
[19:14] <ubitux> -Assertion a->planar failed at libswresample/swresample.c:284
[19:14] <ubitux> i have this
[19:14] <michaelni> s->postin.planar = s->midbuf.planar = s->preout.planar =  1;
[19:15] <ubitux> (the gdb output looks strange indeed)
[19:15] <michaelni> maybe some av_log() would clear things up ...
[19:15] <michaelni> that is what is wrong and where it does become wrong
[19:16] <ubitux> but i guess the multiline macro is doing crazy things
[19:16] <ubitux> ok, going to add some
[19:17] <ubitux> realloc_audio: planar=0 bps=0 ch_count=0
[19:17] <ubitux> Assertion a->planar failed at libswresample/swresample.c:286
[19:17] <ubitux> heh :p
[19:18] <michaelni> which realloc_audio() is this ?
[19:19] <ubitux> seems to be the one line 463 according to the backtrace
[19:19] <michaelni> your lines dont match mine
[19:19] <CIA-46> ffmpeg: 03Dominique Leuenberger 07master * r5aac76159e 10ffmpeg/RELEASE: RELEASE: We're now at 0.9.1.git
[19:20] <ubitux> michaelni: sorry, -2 because of the av_log
[19:20] <ubitux> 461
[19:21] <michaelni> s->postin.planar = s->midbuf.planar = s->preout.planar =  1;
[19:21] <michaelni> is that executed ?
[19:21] <michaelni> If no, why not, if yes what turns it to 0 later ?
[19:22] <ubitux> mmmh oh i think i get it
[19:22] <ubitux> the swr_init might have failed for some reasons
[19:23] <ubitux> 'going to invinstigate
[19:23] <ubitux> (no, it's not executed)
[19:23] <ubitux> thank you
[19:23] <michaelni> np
[19:25] <ubitux> ok i'm stupid, i assumed the alloc_set_opts was actually calling the init
[19:26] <ubitux> great, works now :)
[20:19] <CIA-46> ffmpeg: 03Michael Niedermayer 07master * r5ab6f0fe5a 10ffmpeg/libswscale/swscale.c: (log message trimmed)
[20:19] <CIA-46> ffmpeg: Revert "swscale: fix integer overflows in RGB pixel writing."
[20:19] <CIA-46> ffmpeg: This reverts commit 77d88b872d8cbb42738ede2d4fc098c16f204236.
[20:19] <CIA-46> ffmpeg: The revert fixes actual overflows and a segfault as the variables
[20:19] <CIA-46> ffmpeg: are signed and can be negative.
[20:19] <CIA-46> ffmpeg: Conflicts:
[20:19] <CIA-46> ffmpeg:  libswscale/swscale.c
[21:05] <ubitux> michaelni: what is the cliping into tables for?
[21:06] <CIA-46> ffmpeg: 03Michael Niedermayer 07master * r31801ded9a 10ffmpeg/: 
[21:06] <CIA-46> ffmpeg: Merge remote-tracking branch 'hexene/stagefright'
[21:06] <CIA-46> ffmpeg: * hexene/stagefright:
[21:06] <CIA-46> ffmpeg:  libstagefright: fix build
[21:06] <CIA-46> ffmpeg: Merged-by: Michael Niedermayer <michaelni at gmx.at>
[21:06] <CIA-46> ffmpeg: 03Dmitry Monakhov 07master * rd7b786e9a1 10ffmpeg/tools/build_libstagefright: 
[21:06] <CIA-46> ffmpeg: libstagefright: fix build
[21:06] <CIA-46> ffmpeg: - Use headers from gingerbread-release branch instead of the current default ice cream sandwich
[21:06] <CIA-46> ffmpeg: - Fix path mistype
[21:07] <michaelni> ubitux, after the scaling the samples are not guranteed to be within 0..255 so cliping is needed
[21:07] <michaelni> if the tables are a bit bigger and do the cliping it should be faster then doing it excplicitly
[21:08] <ubitux> ok
[21:08] <ubitux> thank you
[21:18] <OanaStratulat> michaelni: ping
[21:26] <michaelni> OanaStratulat, pong
[21:26] <OanaStratulat> michaelni: do you have time to work on that bug ?
[21:42] <michaelni> OanaStratulat, well, yes  a bit
[21:42] <OanaStratulat> ok in 1 min i will be on my linux machine
[21:42] <OanaStratulat> and ping you then
[21:45] <Oana> michaelni: ping
[21:45] <michaelni> pong
[21:45] <Oana> i`m OanaStratulat but on my linux machine
[21:56] <Oana> michaelni: https://ffmpeg.org/trac/ffmpeg/ticket/890 this is the bug
[21:57] <michaelni> i know
[21:57] <michaelni> i can fix it but iam not sure how i can explain it
[21:57] <Oana> can you try a bit
[21:57] <Oana> and guide me
[22:01] <Oana> this it the only bug i have found lately in ffmpeg
[22:01] <michaelni> well, i dont know exactly where to best fix it ,yself
[22:04] <Oana> michaelni: where it is simplest to fix 
[22:05] <michaelni> maybe to detect the outrageous value in the demuxer
[22:09] <michaelni> some check somewhere around line 550 in avidec.c should be possible
[22:10] <Oana> to check for what?
[22:11] <michaelni> crazy starttime
[22:12] <michaelni> something like > 1 hour is likely broken
[22:12] <Oana> ast->start is the starttime ?
[22:15] <michaelni> theres no ast->start
[22:15] <michaelni> ast->cum_len=avio_rl32(pb); /* start */
[22:16] <michaelni> durandal_1707, is your sws branch ok to be merged ?
[22:16] <michaelni> or should i wait for something? reviewes? 
[22:16] <durandal_1707> mru had some objection regarding your pointers dance
[22:16] <michaelni> pointers dance ?
[22:17] <durandal_1707> i used 15to15 as template
[22:17] <durandal_1707> look at libav ML
[22:17] <Oana> how do i avlog an int64_t var ? what format do i use ?
[22:18] <durandal_1707> read manual for printf
[22:18] <durandal_1707> ll or something like that
[22:19] <Oana> lld
[22:20] <Oana> the value for ast->cum_len is in seconds ?
[22:21] <michaelni> Oana, in scale/rate units
[22:21] <Oana> so how do i check if it is bigger than 1 h?
[22:22] <michaelni> val*scale/rate > 3600
[22:23] <Oana> michaelni: ok so my file has start time bigger than 1 h
[22:24] <Oana> if this happens what should i do ?
[22:24] <michaelni> durandal_1707, you can do 2 pixels at once using uint32_t
[22:25] <durandal_1707> wouldn't that break 7x7 case?
[22:25] <michaelni> Oana, av_log + something like return error
[22:25] <Oana> av log with what message ?
[22:25] <michaelni> durandal_1707, its messy but odd width would need the last pixel handled seperately
[22:26] <michaelni> so probably not worth
[22:26] <michaelni> but theoretically doing 2 pixels at once should be faster
[22:26] <michaelni> Oana, "crazy start time, iam scared, giving up"
[22:27] <durandal_1707> i gonna stay with uint16_t
[22:28] <Oana> and return ?
[22:29] <Oana> michaelni: AVERROR_INVALIDDATA ?
[22:30] <michaelni> that or just set the start to 0
[22:31] <Oana> michaelni: http://pastebin.com/RMWf7KYU this is my diff
[22:31] <Oana> looks good?
[22:32] <michaelni> yes but it should be something else than 0,0
[22:32] <Oana> the context and what else?
[22:33] <michaelni> AV_LOG_ERROR
[22:34] <Oana> michaelni: http://pastebin.com/N607ybaG good to submit?
[22:35] <michaelni> Oana, does it work ?
[22:35] <Oana> yes it does not crash anymore
[22:35] <Oana> what good files should i test from the fate-suite ?
[22:36] <michaelni> i dont know pick maybe 10 random avi files or so 
[22:38] <Oana> michaelni: it works, have tested some .avi files and they are good
[22:41] <Oana> so good for submit on melange ?
[22:41] <Oana> and trac/
[22:43] <michaelni> durandal_1707, can you use AV_RN16(A)() and AV_WN16(A)() to access the values ? that should fix the aliassing issue
[22:43] <michaelni> though the issue is pretty much theoretical
[22:43] <michaelni> id be happy to merge it as it is if you want ...
[22:44] <michaelni> Oana,  it looks like theres something missing in the pastebin: av_log(s, AV_LOG_ERROR, "crazy start time, iam scared, giving u
[22:45] <Oana> michaelni: http://pastebin.com/9ii7iSNN
[22:45] <Oana> the correct one
[22:47] <michaelni> Oana,  i think the check should be before * sample_size
[22:48] <durandal_1707> michaelni: what is difference between rgb2rgb.c and rgb2rgb_template.c?
[22:48] <Oana> michaelni: http://pastebin.com/9EbmTvkJ good ?
[22:49] <CIA-46> ffmpeg: 03Stefano Sabatini 07master * r0cc8864687 10ffmpeg/ffprobe.c: ffprobe: use av_toupper() in upcase_string()
[22:50] <durandal_1707> i could move function to template, and use uint8_t as 24 variant do
[22:50] <michaelni> beware of big vs little endian if you do that
[22:50] <michaelni> the difference of the template vs normal one is that
[22:51] <michaelni> in the past there was one template that contained mmx/mmx2/3dnow code unde ifdef and then C code to handle the last 0-7 pixels
[22:52] <michaelni> that has been split up so asm and C is seperate
[22:52] <michaelni> now
[22:53] <michaelni> some people said it was a mess in the past, iam not so sure if its not worse now but it doesnt really matter
[22:54] <durandal_1707> i think i can get rid of br
[22:54] <durandal_1707> but compiler should already know how to optimize this
[22:54] <michaelni> yes
[22:54] <Oana> michaelni: http://pastebin.com/9EbmTvkJ is this good ?
[22:55] <michaelni> does it fix the bug and not break other files ?
[22:55] <Oana> yes, it does not crash and i have tested 10 avi files and all worked out good
[22:56] <michaelni> with the latest version ?
[22:56] <durandal_1707> btw, i get bunch of this redefining multi-line macro 'ovhlps'
[22:56] <Oana> yes, the latest version of ffmpeg
[22:56] <michaelni> Oana, latest version of your patch, i mean the one where its before sample_size ?
[22:57] <Oana> michaelni: workes good with the latest version
[22:57] <Oana> of the patch
[22:58] <michaelni> durandal_1707, thx minor bug, ill fix it
[22:59] <michaelni> Oana, then the patch should be good
[22:59] <Oana> ok uploading it to menalge and traC
[23:02] <Oana> michaelni: http://www.google-melange.com/gci/task/view/google/gci2011/7180239 marked as completed
[23:03] <durandal_1707> michaelni: i'm fine if you want to merge it as is, if someone doesn't like how it looks he can send "fix"
[23:09] <Oana> michaelni: is the patch good ?
[23:11] <michaelni> yes, good
[23:13] <Oana> michaelni: http://www.google-melange.com/gci/task/view/google/gci2011/7179244 please accept the claim for
[23:15] <CIA-46> ffmpeg: 03Paul B Mahol 07master * ra6839c4e60 10ffmpeg/libavcodec/bmp.c: 
[23:15] <CIA-46> ffmpeg: bmpdec: support for rgb444 with bitfields compression
[23:15] <CIA-46> ffmpeg: Do not display garbage for invalid/unsupported bitfields values.
[23:15] <CIA-46> ffmpeg: 03Paul B Mahol 07master * r047a28f759 10ffmpeg/libavcodec/bmpenc.c: bmpenc: support for PIX_FMT_RGB444
[23:15] <CIA-46> ffmpeg: 03Paul B Mahol 07master * r277030627d 10ffmpeg/libswscale/swscale_unscaled.c: rgb2rgb: allow conversion for <15 bpp
[23:15] <CIA-46> ffmpeg: 03Paul B Mahol 07master * r4ad40d6d9f 10ffmpeg/libswscale/ (rgb2rgb.c rgb2rgb.h swscale_unscaled.c): rgb2rgb: rgb12tobgr12()
[23:15] <CIA-46> ffmpeg: 03Michael Niedermayer 07master * r24be30de64 10ffmpeg/libavcodec/mjpegdec.c: 
[23:15] <CIA-46> ffmpeg: ljpegdec: Fix the left column of rgb24 decoding
[23:15] <CIA-46> ffmpeg: Fixes Ticket894
[23:15] <CIA-46> ffmpeg: Signed-off-by: Michael Niedermayer <michaelni at gmx.at>
[23:15] <CIA-46> ffmpeg: 03Michael Niedermayer 07master * rf4a01d7788 10ffmpeg/libavutil/x86/x86inc.asm: 
[23:15] <CIA-46> ffmpeg: x86inc: remove duplicate movlhps
[23:15] <CIA-46> ffmpeg: Signed-off-by: Michael Niedermayer <michaelni at gmx.at>
[23:15] <CIA-46> ffmpeg: 03Michael Niedermayer 07master * ra9bd29e15e 10ffmpeg/: (log message trimmed)
[23:15] <CIA-46> ffmpeg: Merge remote-tracking branch 'richardpl/sws'
[23:15] <CIA-46> ffmpeg: * richardpl/sws:
[23:15] <CIA-46> ffmpeg:  rgb2rgb: remove unused bgr8torgb8()
[23:15] <CIA-46> ffmpeg:  rgb2rgb: rgb12tobgr12()
[23:15] <CIA-46> ffmpeg:  rgb2rgb: allow conversion for <15 bpp
[23:15] <CIA-46> ffmpeg:  bmpenc: support for PIX_FMT_RGB444
[23:15] <CIA-46> ffmpeg: 03Paul B Mahol 07master * r296a338261 10ffmpeg/libswscale/ (rgb2rgb.c rgb2rgb.h): 
[23:15] <CIA-46> ffmpeg: rgb2rgb: remove unused bgr8torgb8()
[23:15] <CIA-46> ffmpeg: RGB8 and BGR8 are paletted formats now.
[23:16] <michaelni> durandal_1707, merged, thanks
[23:19] <michaelni> Oana, have you found a easy bug to fix ?
[23:20] <michaelni> i am not sure if i will have time to help ...
[23:20] <OanaStratulat> searching for one this night
[23:20] <OanaStratulat> hoping for an easy bug..
[23:20] <OanaStratulat> anyway i`ll try my luck
[23:30] <OanaStratulat> michaelni: https://ffmpeg.org/trac/ffmpeg/ticket/296 is this a hard bug to fix?
[23:35] <durandal_1707> OanaStratulat: define hard
[23:36] <OanaStratulat> durandal_1707: needs a lot of work on it
[23:39] <durandal_1707> needs lot of experience
[23:40] <OanaStratulat> durandal_1707: i have exactly none...
[23:40] <durandal_1707> i guess there is something missing so seeking fails, like duration
[23:41] <durandal_1707> is there option to rebuild index?
[23:42] <durandal_1707> mplayer -forceidx fails
[23:42] <durandal_1707> OanaStratulat: pick another bug
[23:42] <OanaStratulat> durandal_1707: ok
[23:42] <OanaStratulat> durandal_1707: https://ffmpeg.org/trac/ffmpeg/ticket/599 this one what do you say?
[23:43] <durandal_1707> probably would involve to recalculate duration on available data or something like that
[23:44] <durandal_1707> you would need to locate and explore relevant demuxer
[23:50] <OanaStratulat> michaelni: ^
[23:55] <OanaStratulat> durandal_1707: what is the demuxer used here?
[23:56] <durandal_1707> that should be pretty obvious mov
[23:56] <durandal_1707> mov is container format
[23:57] <OanaStratulat> durandal_1707: what is avc1 ?
[00:00] --- Wed Jan 11 2012

More information about the Ffmpeg-devel-irc mailing list