[Ffmpeg-devel-irc] ffmpeg-devel.log.20140106

[burek]

[01:14] <cone-766> ffmpeg.git 03Michael Niedermayer 07master:27b1e63f32a9: avcodec/huffyuv: Support more 8bit YUV formats
[01:28] <ubitux> michaelni: is there some ressembling a specs for the yuv formats you added or this is just v3 specific ff' thing?
[01:28] <ubitux> something*
[01:30] Action: ubitux will read answer tomorrow
[01:30] Action: ubitux &
[01:30] <michaelni> ubitux, ff specific
[01:31] <michaelni> and before someone asks, yes i will add more formats
[02:39] <cone-766> ffmpeg.git 03Michael Niedermayer 07master:1bef9bfc773b: fate: test version 3 of ffvhuff
[02:39] <cone-766> ffmpeg.git 03Michael Niedermayer 07master:4816d63fb7f6: avcodec/huffyuvdec: optimize decode_plane_bitstream(), use joint tables for new yuv formats
[04:15] <BBB2> bleh what a mess
[04:16] <BBB2> (netsplit still affecting my normal irc thing)
[04:33] <cone-766> ffmpeg.git 03Michael Niedermayer 07master:d66bab0a69ac: avcodec/hevc_ps: check that VPS referenced from SPS exists
[04:33] <cone-766> ffmpeg.git 03Michael Niedermayer 07master:4ced5d7780fe: avcodec/hevc: Fix modulo operations
[07:55] <ubitux> michaelni: ok
[08:02] <ubitux> it's fun how randomizing a few instructions order leads to various optim/slowdown
[08:02] <ubitux> like, i'm toggling two macro, -10 cycles, wtf
[08:06] <ubitux> 3899 decicycles in ff_vp9_loop_filter_v_16_16_ssse3, 4193912 runs, 392 skips
[08:06] <ubitux> I DID IT
[08:07] <ubitux> nevcairiel ^
[08:07] <ubitux> thx to various random
[08:08] <ubitux> (and luck, because not always, but hey)
[08:35] <kierank> did you stay up all night?
[09:12] <ubitux> kierank: no, just did that this morning :p
[09:12] <j-b> 'morning
[09:15] <ubitux> BBB: so, transpose16x16b(lpf_v_16_16(transpose16x16b(block))) is supposed to work for lpf_h_16_16, right?
[09:48] <ubitux> BBB: btw, i saved 10 reads since the patch, better review on github, but i'll resend a patch in a few days when i'm done with the transpose
[10:21] <nevcairiel> ubitux: congrats :D
[12:16] <BBB> ubitux: ok and yes
[13:16] <cone-311> ffmpeg.git 03Guillaume Martres 07master:98c0cd0b9925: hevc: add new conformance streams
[13:16] <cone-311> ffmpeg.git 03Michael Niedermayer 07master:0db6402d9116: Merge commit '98c0cd0b99254cd12f312bf6a4c24cd3ec5886b5'
[13:29] <cone-311> ffmpeg.git 03Guillaume Martres 07master:060667195e6e: hevc: rename some HEVC conformance streams
[13:29] <cone-311> ffmpeg.git 03Michael Niedermayer 07master:8d86e7b49589: Merge commit '060667195e6e6bc0fa63ce13c47e23aabe70d122'
[13:37] <cone-311> ffmpeg.git 03Anton Khirnov 07master:fffca3d278c2: 4xm: return a proper error code.
[13:37] <cone-311> ffmpeg.git 03Michael Niedermayer 07master:d701f5996daf: Merge commit 'fffca3d278c2a2422c2f61f21c5a9d5f690d328e'
[13:55] <cone-311> ffmpeg.git 03Anton Khirnov 07master:fe80fa53bc49: 4xm: replace forcing EMU_EDGE by a copy
[13:56] <cone-311> ffmpeg.git 03Michael Niedermayer 07master:d1134ababdbd: Merge commit 'fe80fa53bc4977ef6e8e749a9df8ca5590a09307'
[14:07] <BBB> michaelni: are you sure you like these force-emu-edge commits?
[14:20] <michaelni> no, iam not sure ...
[14:20] <michaelni> but they could be reverted or improved if needed
[14:21] <ubitux> michaelni: btw, the freetype fix from libav seems broken right now, so no need to merge it; it will make beastd happy too
[14:29] <cone-311> ffmpeg.git 03Anton Khirnov 07master:adb199d1bc7f: kgv1dec: replace forcing EMU_EDGE by a copy
[14:29] <cone-311> ffmpeg.git 03Michael Niedermayer 07master:80d44190c4d9: Merge commit 'adb199d1bc7f86aa6b85986f40190ec17f99d34f'
[14:48] <cone-311> ffmpeg.git 03Anton Khirnov 07master:f4a8a0080537: sgidec: fix buffer size check in expand_rle_row()
[14:48] <cone-311> ffmpeg.git 03Michael Niedermayer 07master:e86cd37a7026: Merge commit 'f4a8a0080537484154bb74e08ec76cbcbd25484b'
[15:12] <cone-311> ffmpeg.git 03Kostya Shishkov 07master:2f97094608cf: lagarith: do not call simd functions on unaligned lines
[15:12] <cone-311> ffmpeg.git 03Michael Niedermayer 07master:b7697d29ed77: Merge commit '2f97094608cfd2665660f7a26a3291559b186752'
[15:16] <relaxed> ubitux: ffmpeg fails at building statically too: http://www.johnvansickle.com/pub/ffmpeg_static_build-config.log.txt
[15:17] <ubitux> ah? i guess the --static patch will be welcome then
[15:17] <ubitux> thx
[15:18] <relaxed> you're welcome.
[15:19] <relaxed> be sure and get their next patch too
[15:21] <ubitux> we'll need to change it in require_libfreetype as well probably
[15:24] <cone-311> ffmpeg.git 03Anton Khirnov 07master:8058284ce090: lavc: add 422/444 YUV with alpha to align_dimensions()
[15:24] <cone-311> ffmpeg.git 03Michael Niedermayer 07master:4e276b84e697: Merge commit '8058284ce09030b47512746d726fb2ad3ae8a20f'
[15:33] <cone-311> ffmpeg.git 03Anton Khirnov 07master:75647dea6f7d: oggparseogm: check timing variables
[15:33] <cone-311> ffmpeg.git 03Michael Niedermayer 07master:59725592a97f: Merge commit '75647dea6f7db79b409bad66a119f5c73da730f3'
[15:39] <cone-311> ffmpeg.git 03Anton Khirnov 07master:3867f3718ba8: lavf: remove a pointless check
[15:39] <cone-311> ffmpeg.git 03Michael Niedermayer 07master:76c63bc5d856: Merge commit '3867f3718ba82ff11d3e24c6d84beb520d0b174f'
[15:49] <cone-311> ffmpeg.git 03Anton Khirnov 07master:50079a6aa932: lavc: do not leak the internal frame if opening the codec fails
[15:49] <cone-311> ffmpeg.git 03Michael Niedermayer 07master:3328d105f76c: Merge commit '50079a6aa93291e6dc9d9fb8d33da83f79e9311d'
[15:58] <cone-311> ffmpeg.git 03Anton Khirnov 07master:4736d003fac3: cavsdec: check ff_get_buffer() return value
[15:58] <cone-311> ffmpeg.git 03Michael Niedermayer 07master:a4f8a5649561: Merge commit '4736d003fac30af4afd8390182f12f480ad78801'
[16:24] <cone-311> ffmpeg.git 03Anton Khirnov 07master:9a026c72982f: h264: rebuild the default ref list if the reference count changes
[16:24] <cone-311> ffmpeg.git 03Michael Niedermayer 07master:7ee8a1c562e8: Merge commit '9a026c72982faf20e1c8dfbe48f0b312cdea69c8'
[16:24] <cone-311> ffmpeg.git 03Michael Niedermayer 07master:98dcbb47fa5f: avcodec/h264: reset list_count too in case of error in ff_set_ref_count()
[16:37] <cone-311> ffmpeg.git 03Anton Khirnov 07master:bfd26b7ce6ef: h264: reject mismatching luma/chroma bit depths during sps parsing
[16:37] <cone-311> ffmpeg.git 03Michael Niedermayer 07master:a60abb1ee01e: Merge commit 'bfd26b7ce6efea594f2b99441d900419df3af638'
[16:45] <cone-311> ffmpeg.git 03Anton Khirnov 07master:9eef9eb3014b: h264: check that execute_decode_slices() is not called too many times
[16:45] <cone-311> ffmpeg.git 03Michael Niedermayer 07master:64591f8f86f2: Merge commit '9eef9eb3014b2ed9c3ff4aac510a9f04edb555cf'
[16:59] <cone-311> ffmpeg.git 03Anton Khirnov 07master:a03a642d5ceb: h264: do not use 422 functions for monochrome
[16:59] <cone-311> ffmpeg.git 03Michael Niedermayer 07master:30056fd0be8a: Merge commit 'a03a642d5ceb5f2f7c6ebbf56ff365dfbcdb65eb'
[17:08] <cone-311> ffmpeg.git 03Anton Khirnov 07master:9cbf2d78f0a9: pthread_frame: unref decoded frames on failure
[17:08] <cone-311> ffmpeg.git 03Michael Niedermayer 07master:8367bc6be890: Merge commit '9cbf2d78f0a9c19129e7a70b2281a450d386c6d9'
[17:25] <cone-311> ffmpeg.git 03Anton Khirnov 07master:6892d145a0c8: segafilm: fix leaks if reading the header fails
[17:25] <cone-311> ffmpeg.git 03Michael Niedermayer 07master:2a58d5fc0e84: Merge commit '6892d145a0c80249bd61ee7dd31ec851c5076bcd'
[17:38] <ubitux> http://ubitux.fr/pub/pics/_perf-vp9lpf-psrlq.png
[17:38] <ubitux> huuh?
[17:40] <cone-311> ffmpeg.git 03Luca Barbato 07master:d68dc3c9446e: drawtext: Drop pointless header
[17:41] <cone-311> ffmpeg.git 03Michael Niedermayer 07master:d1dc0d218423: Merge commit 'd68dc3c9446e38b4d686cc0f55433c9e8d7c128b'
[17:43] <ubitux> BBB: btw, it seems mc 8tap will be the next bottleneck: http://ubitux.fr/pub/pics/_vp9-mc-8tap.png
[17:49] <cone-311> ffmpeg.git 03Luca Barbato 07master:e61b8fa5605b: configure: Update freetype check to follow upstream
[17:49] <cone-311> ffmpeg.git 03Michael Niedermayer 07master:49a5912440db: Merge remote-tracking branch 'qatar/master'
[18:51] <Compn> i am not sure how much of the concat demuxer vs playlist api was committed from this 
[18:51] <Compn> http://comments.gmane.org/gmane.comp.video.ffmpeg.devel/94761
[18:51] <Compn> back in 2009
[18:56] <JEEB> didn't elenril also do playlist work more lately?
[20:07] <cone-311> ffmpeg.git 03Michael Niedermayer 07master:22f59a8cbeb1: avcodec/huffyuv: add GBRP support
[20:07] <cone-311> ffmpeg.git 03Michael Niedermayer 07master:60ab582a4e87: avcodec/huffyuv: add GRAY8 support
[21:26] <cone-311> ffmpeg.git 03Michael Niedermayer 07master:13f3092b387f: avcodec/huffyuv: add YUVA420P, YUVA422P, YUVA444P, GBRAP, GRAY8A
[22:07] <Daemon404> this callback/interrupt thread is good popcorn
[22:10] <beastd> Daemon404: heh, i kind of thought the same when reading it few hours ago
[22:14] <beastd> callback interrupt is a no go for ffmpeg IMNSHO. but it is still fun to read the argument ;)
[22:24] <cone-311> ffmpeg.git 03Michael Niedermayer 07master:01d245ef4392: random_seed: Rewrite the generic clock() based seed code
[22:24] <cone-311> ffmpeg.git 03Michael Niedermayer 07master:00f25e0a990f: Merge commit '01d245ef4392152dbdc78a6ba4dfa0a6e8b08e6f'
[22:31] <beastd> michaelni: i would say no need to really merge any of the freetype build fix stuff from the fork. arch uses my patch since 2013-12-10 (and the AUR ffmpeg-full-git since even earlier). other distros use it too IIRC. i know of no problem reports.
[22:32] <cone-311> ffmpeg.git 03Martin Storsjö 07master:9409c9bdbfd8: configure: Disable networking if winsock2.h is available but winsock functions aren't
[22:32] <cone-311> ffmpeg.git 03Michael Niedermayer 07master:523a803b7352: Merge commit '9409c9bdbfd829353473ee6cc3e91c726481c069'
[22:35] <llogan> beastd: lol. ffmpeg-full-git.
[22:36] <beastd> :)
[23:01] <ubitux> fun, with a transpose_c(lpfvp9_16_16_v(transpose_c(block))), i'm going from 16372 to 9041 decicycles
[23:03] <ubitux> smarter: where are the bench for hevc? :p
[23:03] <smarter> ubitux: decoding benchs?
[23:03] <kurosu_> ubitux: how long is the longest deblocking filter ? I would naively assume you don't need to transpose all of the block
[23:03] <ubitux> smarter: what you just pushed on libav
[23:03] <smarter> ubitux: it should not have been pushed
[23:04] <smarter> ubitux: I think DonDiego slipped on his keyboard :)
[23:04] <nevcairiel> sounds like diego did it :p
[23:04] <JEEBsv> I know that current ffmpeg gives me 1080p25 playback on a C2D P8400
[23:04] <JEEBsv> which is already pretty gut :3
[23:04] <ubitux> strange, i see you as comitter smarter 
[23:04] <smarter> yeah, no idea why
[23:04] <smarter> I don't have commit rights afaik
[23:04] <ubitux> kurosu_: i don't understand the question
[23:06] <ubitux> kurosu_: what i'm saying is that the horizontal c code is slower than a c code doing twice a transpose and calling the vertical asm
[23:06] <ubitux> "almost" twice as fast
[23:06] <ubitux> (more like 1.8)
[23:06] <kurosu_> what I mean is that the transpose is transposing pixels not needed by the filter
[23:07] <ubitux> aah, yeah probably
[23:07] <ubitux> but the simd is branchless, whatever the data i don't think it will change the performance much
[23:07] <ubitux> it will do all the path anyway
[23:08] <ubitux> so the bench is somehow accurate
[23:08] <ubitux> anyway, how am i going to do that transpose...
[23:08] <kurosu_> it's just that I would expect a full 16x16 transpose to be troublesome using simd, while not neeeded
[23:08] <nevcairiel> it was a C transpose for testing fun :)
[23:09] <SevereOverfl0w> Is this the place to report a libavcodec stacktrace, I found in relation to VLC? The VLC guys seem to be hinting you are the blame.
[23:09] <kurosu_> yeah I understood that
[23:10] <ubitux> SevereOverfl0w: generally it's on the trac, unless you want to help debugging
[23:11] <ubitux> if you can reproduce with a small code or better with the ff* tools, that's better
[23:11] <Zeranoe> Is there any way to compile just one part of ffmpeg, like avcodec?
[23:12] <SevereOverfl0w> ubitux: http://bpaste.net/show/t2wIkiw9WPTHpik6mFZX/ This is what I have. I'm assuming I'm going to have to recompile libavcodec without stripping the binaries? 
[23:12] <SevereOverfl0w> If I can avoid it, I'd like to though :p
[23:12] <ubitux> dunno if you can get the symbols only
[23:13] <ubitux> Zeranoe: --disable-everything (-all?) --enable-avcodec?
[23:13] <nevcairiel> Zeranoe: you can write "make libavcodec/avcodec-55.so" to only build one of the libs directly
[23:13] <ubitux> nevcairiel is always smarter :(
[23:14] <nevcairiel> (adjust file name to match your environment of course)
[23:14] <nevcairiel> ubitux: he didnt really specify what he wanted, but i'm sure yours would build only avcodec and its deps too!
[23:15] <smarter> SevereOverfl0w: does that libavcodec comes from your distribution? If so they may have a debug package you can install
[23:15] <SevereOverfl0w> smarter: Unfortunately Arch Linux does not seem to have a debug package :( 
[23:15] <SevereOverfl0w> It's something that is in planning currently
[23:15] <ubitux> SevereOverfl0w: how do you reproduce?
[23:16] <Zeranoe> I'm basically looking to compile FFmpeg so pdb files (Windows debugging files) are statically generated for each of the libs. I can generate the pdb files with a static build, but there is only one pdb file, and I would like to have a separate one for each part
[23:16] <nevcairiel> how does it make sense to have split pdbs for a static build, you have one binary in the end, should have one debug file =p
[23:17] <Zeranoe> So something like libavcodec.pdb, which is generated with a shared build, but not with a static build
[23:17] <Zeranoe> nevcairiel: I agree... It isn't for me, but I'm curious if it's possible.
[23:17] <nevcairiel> does a debugger even accept that
[23:18] <SevereOverfl0w> ubitux: I have a movie file, that when I run vlc against it. 9 seconds in, it crashes. 
[23:18] <SevereOverfl0w> I am currently recompiling ffmpeg with -O0 -fbuiltin and -g, and not stripping the binary.
[23:19] <Zeranoe> nevcairiel: I'm guessing so... you can generate static lib files for Windows, static debug files should be possible IMO
[23:19] <ubitux> SevereOverfl0w: can you reproduce with ffmpeg -i input -f null - ?
[23:19] <nevcairiel> hm, does ffmpeg even build with O0? i thought that didnt work =p
[23:19] <SevereOverfl0w> nevcairiel: Well, we're about to find out...
[23:19] <ubitux> < SevereOverfl0w> I am currently recompiling ffmpeg with -O0 -fbuiltin and -g, and not stripping the binary. // you don't need to.
[23:19] <SevereOverfl0w> ubitux: I do not? 
[23:20] <ubitux> build it normally, and then run ./ffmpeg_g
[23:20] <ubitux> instead of ./ffmpeg
[23:20] <ubitux> but anyway, can you reproduce with ffmpeg?
[23:21] <beastd> SevereOverfl0w: when you make ffmpeg it is usually sufficient. you always get an additional binary ffmpeg_g which has debug symbols. (and of course ubitux was way faster to say that ;)
[23:22] <SevereOverfl0w> I am running a modified version of my distros automated build script. And installing debug version globally.
[23:22] <SevereOverfl0w> But it would seem ubitux, that no, I am not seeing the same issue with ffmpeg -i myfile.avi -f null -
[23:23] <ubitux> so if you play the file with ffplay it works?
[23:23] <SevereOverfl0w> ubitux: Exact pastebin http://bpaste.net/show/BbLVzlLqFCGAdKAfMqza/
[23:24] <Zeranoe> I'm guessing the pdb files would be generated if it was possible to compile just one lib in static. Which goes back to my original question. I'll try --disable-all --enable-avcodec
[23:24] <ubitux> sigint mh.
[23:24] <nevcairiel> "make libavcodec/avcodec-55.a" should work to just create one of the static libs
[23:24] <SevereOverfl0w> ubitux: It would seem that yes, it does work in ffplay :) 
[23:26] <Zeranoe> nevcairiel: make: *** No rule to make target `libavcodec/avcodec-55.a'.  Stop.
[23:27] <nevcairiel> on windows it might be .lib, what do i know
[23:27] <nevcairiel> whatever name the static library gets
[23:27] <Zeranoe> Should be .a... where should i look for it? libavcodec/Makefile perhaps
[23:28] <nevcairiel> look at the build dir which file exists :p
[23:28] <nevcairiel> i havent build static in ages
[23:28] <Zeranoe> So i need to build through first to find out?
[23:28] <SevereOverfl0w> ubitux: So, where do I go from here? :) 
[23:28] <Zeranoe> then recompile with just that
[23:29] <ubitux> SevereOverfl0w: i don't know the vlc build, but i'd look into it to see how to build it with a custom ffmpeg version/flags
[23:29] <ubitux> ask j-b 
[23:29] <ubitux> or you can also upload the sample
[23:30] <SevereOverfl0w> ubitux: Certain er.. legal restrictions on doing that
[23:30] <ubitux> extract 10 seconds with ffmpeg :p
[23:30] <ubitux> hopefully it will still crash
[23:30] <ubitux> (or use dd depending on the format)
[23:31] <nevcairiel> its an avi, dd might screw with it a bit, tbh no clue how well you can just cut avi
[23:31] <SevereOverfl0w> ffmpeg is probably best. You know off the top of your head on how to do that?
[23:32] <smarter> ffmpeg -i foo.foo -vframes 250 out.foo
[23:33] <ubitux> with -c copy and -map 0
[23:33] <ubitux> and -t 10 instead
[23:33] <ubitux> ffmpeg -i in.foo -c copy -map 0 -t 10 out.foo
[23:36] <ubitux> so, there is a TRANSPOSE8x8W
[23:37] <ubitux> TRANSPOSE16x16B should be doable then
[23:37] <SevereOverfl0w> ubitux: Okay. Confirmed still crashing vlc with the 20s sample.
[23:37] <SevereOverfl0w> I added a little extra, just in case.
[23:37] <ubitux> https://trac.ffmpeg.org/
[23:37] <ubitux> see upload on http://ffmpeg.org/bugreports.html
[23:38] <SevereOverfl0w> Okay. 
[23:39] <SevereOverfl0w> ubitux: ffmpeg doesn't cause an issue, what stack trace should I give?
[23:39] <SevereOverfl0w> Well, not directly.
[23:39] <SevereOverfl0w> Should I install the debug versions I have compiled, re-run VLC, provide the gdb core information
[23:40] <ubitux> you need a vlc with using the libs with debug symbols
[23:40] <Zeranoe> With --disable-all --enable-avcodec only avutil is built... whatr
[23:40] <ubitux> i don't know how to achieve that, ask #videolan
[23:41] <nevcairiel> avcodec probably has some dependency which is still disabled, diable-all is a bit weird
[23:41] <nevcairiel> note that disable-all also disables all components, so all decoders, encoders, parsers, whathaveyou
[23:42] <SevereOverfl0w> ubitux: I believe I have vlc in debug mode.
[23:42] <SevereOverfl0w> I'll try ffmpeg in debug mode.
[23:44] <Zeranoe> nevcairiel: make libavcodec/avcodec.a (no version) seems to be working.
[23:44] <SevereOverfl0w> Hmm, appears to not have compiled in debug mode
[23:45] <wm4> SevereOverfl0w: ffmpeg is by default stripped or so, but it also dumps ffmpeg_g into the build dir
[23:45] <SevereOverfl0w> wm4: I specifically added !strip to the build script.
[23:56] <beastd> SevereOverfl0w: did you configure with --enable-debug=1
[23:57] <SevereOverfl0w> beastd: I don't think so 
[23:58] <beastd> SevereOverfl0w: and --disable-stripping would be a good idea probably
[23:59] <SevereOverfl0w> beastd: !strip should prevent any stripping.
[23:59] <SevereOverfl0w> The --enable-debug=1, where do I add that? :) 
[23:59] <SevereOverfl0w> To the configure?
[00:00] --- Tue Jan  7 2014