[Ffmpeg-devel-irc] ffmpeg-devel.log.20150406

[burek]

[00:00:55 CEST] <himangi> however, what makes av_uninit silence compiler warnings?
[00:02:20 CEST] <michaelni> it silences uninitialized variable warnings with gcc, how that works exactly in gcc i dont know maybe gcc is silly and treats this as a valid initialization or they specifically check for this construct
[00:02:39 CEST] <michaelni> see libavutil/attributes.h if you havent yet
[00:03:24 CEST] <himangi> ok.. I did. but wasnt sure of what the #defines could do
[00:04:21 CEST] <michaelni> int x; // gcc, this is not initialized \n int x=x; // gcc ohh thats ok its initialized i wont complain
[00:04:59 CEST] <himangi> cool. thanks :)
[01:50:53 CEST] <cone-844> ffmpeg 03Anton Khirnov 07master:249796e2560c: svq3: initialize some required H264Context fields.
[01:50:54 CEST] <cone-844> ffmpeg 03Michael Niedermayer 07master:99a567cbe24a: Merge commit '249796e2560c2cce1add722c3e873de96b17be74'
[02:18:22 CEST] <cone-844> ffmpeg 03Mariusz SzczepaDczyk 07master:428675227908: lavf/libsmbclient: implement directory listing callbacks
[02:52:31 CEST] <cone-844> ffmpeg 03Michael Niedermayer 07master:ba54512a10f8: avfilter/vf_mpdecimate: Fix missing ()
[04:48:27 CEST] <cone-844> ffmpeg 03Michael Niedermayer 07master:ff6841c6bb2d: avformat/utils: avoid discarded streams in av_find_default_stream_index()
[05:48:10 CEST] <Prelude2004c> hey anyone know what the library was for the old cuda stuff ? for the new libnvenc i know but what was the old cuda version of it
[05:48:11 CEST] <Prelude2004c> ?
[05:49:20 CEST] <Prelude2004c> libxvid?
[05:55:22 CEST] <Compn> Prelude2004c : no
[05:56:29 CEST] <Compn> i mean, its not libxvid
[05:57:16 CEST] <Compn> there was vdpau but thats encoding only
[05:57:26 CEST] <Compn> er decoding only
[05:58:52 CEST] <Compn> theres opencl too
[06:00:59 CEST] <Compn> Prelude2004c : https://trac.ffmpeg.org/wiki/HWAccelIntro
[06:01:15 CEST] <Prelude2004c> there is no encoding cuda for FFMPEG ?
[06:01:55 CEST] <Prelude2004c> i have tried and used the nvenc for encoding and it works but limited to 2 sessions
[06:02:00 CEST] <Prelude2004c> i have 600 and 700 series cards
[06:02:01 CEST] <Compn> i cant remember anything before nvenc
[06:02:24 CEST] <Prelude2004c> so i wanted to test encoding on those cards and was told that the older drivers didn't have this limit with cuda support
[06:02:35 CEST] <Compn> 2 jobs was the limit for h265, how many jobs do the cards support ?
[06:02:54 CEST] <Prelude2004c> 2 jobs is the limit of the new drivers.. if not the super epensive cards
[06:02:57 CEST] <Compn> or what do you mean by sessions ?
[06:03:05 CEST] <Prelude2004c> the 900 series all have a 2 session limit
[06:03:11 CEST] <Prelude2004c> sorry session is what i mean
[06:03:43 CEST] <Compn> you mean 2 encoding threads simultaneously or 2 different sessions and then it quits ?
[06:03:59 CEST] <Prelude2004c> 2 encoding threads
[06:04:05 CEST] <Prelude2004c> i can only do 2 bit rates
[06:04:23 CEST] <Compn> how many threads do you want? :)
[06:05:32 CEST] <Compn> According to NVENC specs there is a limit for number of simultaneous encoding threads. GeForce cards are limited by 2 streams and Quadro are limited by 6 streams. 
[06:05:46 CEST] <Compn> Prelude2004c : got any quadro cards?
[06:06:12 CEST] <Compn> if no, then you are limited to 2 threads by nvenc, nothing to do with ffmpeg 
[06:06:17 CEST] <philipl> Yeah.
[06:06:30 CEST] <philipl> That's how they segment the market
[06:06:48 CEST] <Prelude2004c> ope no quadro cards
[06:07:08 CEST] <Prelude2004c> so i wanted to use the cuda based stuff on the 600 series
[06:07:36 CEST] <Compn> dont think you'll find any thing about cuda and ffmpeg ... good luck
[06:07:36 CEST] <Prelude2004c> the maxwell gen 2 stuff does't have a limit?
[06:07:54 CEST] <philipl> It's the same. 2 streams on consumer maxwell
[06:08:19 CEST] <philipl> What are you trying to achieve? nvenc quality is comparatively garbage. It's only useful for realtime stuff.
[06:08:49 CEST] <Prelude2004c> live channels > multi HLS streams
[06:08:56 CEST] <Prelude2004c> so like 1 channel pushing 4 or 5 bit rates
[06:09:01 CEST] <Prelude2004c> and its live and realtime
[06:09:11 CEST] <philipl> nvidia want you to buy a quadro.
[06:09:23 CEST] <Prelude2004c> 6 is not enough
[06:09:32 CEST] <philipl> Or maybe do the low bit rate streams with CPU? x264 in real-time mode?
[06:09:37 CEST] <Prelude2004c> i am hopping to put like 10 channels X 3 bit rates each.. in one seysm
[06:09:49 CEST] <Prelude2004c> tried CPU.. better quality really kills the system
[06:09:52 CEST] <philipl> nvidia would be delighted to sell you a multi-gpu solution :-)
[06:11:30 CEST] <philipl> I don't know what the commercial solutions are for this kind of thing, but I suspect you cannot achieve it with nvenc. As you say, even 6 streams isn't enough for you. You'd still be looking at 5 quadro GPUs in one system.
[06:12:24 CEST] <Prelude2004c> but its not per GPU
[06:12:26 CEST] <Prelude2004c> its per system
[06:12:53 CEST] <Prelude2004c> i tried.. running 2 simultanious cards ( 2 x 690 [4 gpus [) and still it only allows 2 sessions
[06:12:58 CEST] <Prelude2004c> i wish it were per GPU
[06:13:01 CEST] <Prelude2004c> that would solve my problem
[06:13:10 CEST] <Prelude2004c> well not really but..
[06:14:31 CEST] <Prelude2004c> you sure the quadro only has 6 sessions?
[06:14:33 CEST] <philipl> oh dear.
[06:14:35 CEST] <Prelude2004c> even the M6000 ?
[06:14:45 CEST] <philipl> I believe the 6 session limit is an actual hardware limit
[06:15:04 CEST] <philipl> I don't know if the library will then let you do 6x per GPU or still limit you to 6x total.
[06:15:10 CEST] <philipl> You'd need to ask nvidia
[06:16:38 CEST] <Prelude2004c> i have been playingw ith quality all day but 3 streams at 720p basically kill my system.. runing at nearly 1500% .. 720p.. ( 5Mbit/s ) in fast preset
[06:17:45 CEST] <philipl> I couldn't say. There's probably a good body of knowledge online on how to use x264's low latency mode for this kind of stuff, but even then, you're going to be CPU constrained quickly.
[06:19:16 CEST] <Prelude2004c> i know.. can't seem to find the right solution
[06:29:47 CEST] <philipl> Anyway. You'd asked OpenCL based encoding. x264 has an OpenCL mode, but it's not explicitly exposed by ffmpeg. If x264 is built with the feature enabled and you use the option pass-through mechanism, it might be possible to turn on.
[06:30:35 CEST] <philipl> It doesn't make things magically faster but I guess it should reduce CPU load. I have no idea how many concurrent cuda sessions you could manage before the GPU bottlenecks, mind you.
[06:33:33 CEST] <Prelude2004c> hey.. make: *** [libavcodec/nvenc.o] Error 1 anyone know why that is when i try to compile ffmpeg?
[06:36:43 CEST] <Prelude2004c> did it before and it worked
[06:36:43 CEST] <Prelude2004c> ./configure --prefix=/usr/local/bin/ --extra-cflags=-I/usr/local/cuda-6.5/bin --enable-gpl --enable-nonfree --enable-libfdk_aac --enable-libmp3lame --enable-libvorbis --enable-libvpx --enable-libx264 --enable-nvenc
[06:36:48 CEST] <Prelude2004c> nots ure what i am missing
[06:42:13 CEST] <philipl> you need the rest of the error
[06:49:12 CEST] <Prelude2004c> does't say anything
[06:49:14 CEST] <Prelude2004c> just gives me that error
[06:51:08 CEST] <Prelude2004c> http://pastebin.com/iNxVavxG < --- error
[06:51:14 CEST] <Prelude2004c> that is all it tells me
[07:35:31 CEST] <philipl> Prelude2004c: that's very weird. Your source code might be corrupt to give that.
[08:18:40 CEST] <markus___> hmm does anyone know why on mpeg 1 layer II audio some frames are 590 bytes big while most of them are 576 bytes?
[08:18:56 CEST] <markus___> ffmpeg complains that the 590byte frames are too big
[08:19:24 CEST] <markus___> I'm still debugging and trying to figure out what's happening with those 590byte frames
[08:21:05 CEST] <markus___> ah
[08:21:09 CEST] <markus___> Note: Skipped 14 bytes in input.
[08:21:10 CEST] <markus___> Note: Illegal Audio-MPEG-Header 0x000001c0 at offset 1361968.
[08:25:05 CEST] <markus___> ok I'm too far ... it's mpeg audio the audio frames should be within 0x1c0
[10:28:28 CEST] <Prelude2004c> hi everyone
[10:28:32 CEST] <Prelude2004c> anyone around?
[10:29:05 CEST] <durandal_1707>  Me
[10:29:26 CEST] <Prelude2004c> hey, what is the proper library to be used for encoding the old NVCU ?
[10:29:33 CEST] <Prelude2004c> i know vdpau is decode
[10:29:44 CEST] <Prelude2004c> but i want to encode the old way ... without the session limit on the old 600 series cards
[10:29:52 CEST] <Prelude2004c> i can't be using nvenc right now
[10:35:23 CEST] <Prelude2004c> can someone please build an API interface to to do session control with NVIDIA encoding cards ?
[10:35:31 CEST] <nevcairiel> the old encoder sucks terribly in quality though
[10:35:44 CEST] <nevcairiel> and is pretty slow
[10:35:46 CEST] <Prelude2004c> basically some mechanism that receives the work for all the processes ( sessions ) and then only sends it as one single process or sessino to the cards
[10:35:55 CEST] <Prelude2004c> so the cards think its one .. but inf act its a stream of data
[12:20:07 CEST] <markus___> is there any change introduced with the latest ffmpeg library that http://dranger.com/ffmpeg/tutorial06.html this tutorial doesn't work anymore?
[12:20:58 CEST] <markus___> something's weird with the decoded mpeg1 layer 2 audio seems like data is missing and it sounds like chipmunk
[12:48:48 CEST] <markus___> this is getting boring... what the heck is wrong with this mp2 decoder the simplest task is made that difficult
[12:49:40 CEST] <nevcairiel> those dranger tutorials are way old, i dont think they have been updates for years
[12:50:56 CEST] <markus___> they got updated this year
[12:54:02 CEST] <markus___> I guess some stupid parameter must be passed another way now
[13:00:52 CEST] <markus___> something's wrong with the second channel
[13:01:55 CEST] <markus___> if I divide the number of decoded samples by 2 audio plays fine mono
[13:05:33 CEST] <markus___> ok the audio is not interleaved anymore
[13:05:44 CEST] <markus___> line[0] / line[1] contain audio channels...
[13:08:32 CEST] <markus___> ok.. go tit
[13:08:37 CEST] <markus___> got it
[13:17:11 CEST] <loki_> markus___: hi.
[13:17:20 CEST] <loki_> you can try to read api changes.
[13:17:27 CEST] <loki_> there is file in docs/
[13:17:48 CEST] <loki_> i think it is not answer that you wait, but maybe it can help.
[13:23:30 CEST] <markus___> ast->codec->sample_fmt =  AV_SAMPLE_FMT_S16P;
[13:23:35 CEST] <markus___> this seems to have no effect at all
[13:23:40 CEST] <markus___> S16 / S16P ..
[13:23:49 CEST] <markus___> even with S16P data[1] is filled with data
[13:26:43 CEST] <markus___> I think that's not right why is data[0] and data[1] filled with PCM data when the sample_fmt is set to S16P .. 
[13:27:08 CEST] <markus___> I would only expect data[0] to be filled of the returned AVFrame
[13:27:33 CEST] <markus___> that's probably the reason why the dranger manual also fails now
[13:34:05 CEST] <nevcairiel> the P formats specifically mean that [0] and [1] are filled for stereo
[13:34:11 CEST] <nevcairiel> P stands for planar
[13:35:21 CEST] <markus___> well even without P both are filled?
[13:37:34 CEST] <markus___> http://pastebin.com/MVLqYkFa
[13:37:38 CEST] <markus___> this is the current setup I'm testing
[13:38:04 CEST] <markus___> seems like there's no difference between S16 and S16P
[13:39:18 CEST] <markus___> the input file is fine, it plays with ffplay and mplayer; aplay --format=S16_LE --rate=48000 -c 1 test.pcm this works with the decoded output
[13:43:22 CEST] <markus___> I'm going to build an older ffmpeg version to doublecheck this issue
[13:44:00 CEST] <nevcairiel> you cannot overwrite these properties for a decoder like that
[13:44:09 CEST] <nevcairiel> the decoder decides which format its going to give you
[13:44:44 CEST] <nevcairiel> some support request_sample_fmt, but very few
[13:45:12 CEST] <markus___> I set the 3rd argument to NULL
[13:45:24 CEST] <markus___> I was just testing the properties but nothing happens..
[13:47:53 CEST] <markus___> ok, but this worked in the past
[13:48:22 CEST] <nevcairiel> you could never tell the decoder which format to give you, it has always decided
[13:48:34 CEST] <nevcairiel> also, this channel is about development of ffmpeg, not development with ffmpeg, you want #ffmpeg
[14:06:48 CEST] <loki_> nevcairiel: there no chance to find answers on #ffmpeg about any development
[14:14:10 CEST] <markus___> thanks I'm interleaving it by myself now
[14:14:14 CEST] <markus___> loki_: that's what I thought too..
[14:14:34 CEST] <markus___> this clearly changed in the past it has been different one or two years ago
[14:58:21 CEST] <cone-841> ffmpeg 03Michael Niedermayer 07master:d2277aa7e2d2: avformat/httpauth: Change enum to int, which is accessed via AVOption as int
[14:58:22 CEST] <cone-841> ffmpeg 03Michael Niedermayer 07master:8a8a1cce6c3a: avfilter/vsrc_mptestsrc: Change enum to int, which is accessed via AVOption as int
[15:03:48 CEST] <BBB> markus___: use swresample or avresample for interleaving
[15:04:02 CEST] <BBB> markus___: they are optimized and can do any-to-any, regardless of what format the decoder outputs today or tomorrow
[15:14:30 CEST] <cone-841> ffmpeg 03Michael Niedermayer 07master:34bddf443c6f: avfilter/af_amix: Use av_mallocz_array()
[15:14:31 CEST] <cone-841> ffmpeg 03Michael Niedermayer 07master:a66dcfeedc68: avcodec/vc1dec: Check buf2 for malloc failure
[17:14:44 CEST] <Daemon404> wat
[17:14:52 CEST] <Daemon404> how can you be "below" 0x1
[17:14:59 CEST] <Daemon404> (i didnt open the bug page)
[17:15:59 CEST] <phh> ok I had the same question for a while (not the first time I see this bug)
[17:16:02 CEST] <phh> and I'm fairly disappointed
[17:16:05 CEST] <phh> it's only for 0
[17:16:59 CEST] <Daemon404> it sounds more ominous with hex
[17:22:25 CEST] <cone-841> ffmpeg 03Lukasz Marek 07master:f79cd98caf36: examples/avio_list_dir: init/deinit network
[17:22:38 CEST] <kierank> Daemon404: 0x00
[18:12:01 CEST] <Prelude2004c> hey, anyone here interested in doing some work and getting paid for it ?
[18:12:33 CEST] <Prelude2004c> 1. I want to port NVENC's AMD version to work with AMD cards and ffmpeg
[18:13:13 CEST] <loki_> can somebody explain me how i can build old x11grab device?
[18:13:21 CEST] <Prelude2004c> 2. I want to build a process handler that will take advantage of the NVIDIA 2 session limit. Basically a process that will handle the flow of data and from the NVIDIA perspective it will only use 1 or 2 sessions however really it si work from multiple session bunched up in to one
[18:13:59 CEST] <Prelude2004c> anyone interested?
[18:14:19 CEST] <loki_> are you in nvenc?
[18:15:20 CEST] <loki_> ah you searching for dev, ok nevermind.
[18:53:15 CEST] <cone-841> ffmpeg 03Michael Niedermayer 07master:7ddbb4e23a4d: avcodec/pthread_frame: Fix memleak with frame threads and allocated AVOptions
[19:15:48 CEST] <durandal_1707> some filters just set sample formats in query_formats function
[19:23:51 CEST] <durandal_1707> and nothing else just sample formats see af_volumedetect
[19:43:23 CEST] <llogan> michaelni: how often do you do the security update to the server(s)? i can probably start doing that if you tell me any specific I need to know.
[19:45:37 CEST] <michaelni> llogan, when i see a mail on the distros security announce lists, whatever they are called exactly i update
[19:47:26 CEST] <michaelni> llogan, you certainly can update the boxes if you see a security update and ive not applied it yet
[20:25:17 CEST] <Prelude2004c> anyone interested in the project?
[20:30:58 CEST] <cone-841> ffmpeg 03Michael Niedermayer 07master:eca39e29a3af: MAINTAINERS: Add Lou Logan as server admin as he is root too
[20:48:17 CEST] <llogan> wot project?
[21:32:20 CEST] Action: Compn throws congrats at llogan for being promoted :)
[21:32:35 CEST] <kierank> michaelni: is there anything I can do with regards to h264 crashes? bisects perhaps?
[21:33:29 CEST] <Compn> llogan : Prelude2004c wants to make nvenc work with amd cards anf ffmpeg
[21:33:45 CEST] <Compn> bisect never hurts :)
[21:34:08 CEST] <Compn> Prelude2004c : please also ask videolan , as they have a lot of devels who also work on ffmpeg libs
[21:34:24 CEST] <Compn> Prelude2004c : and ask on ffmpeg-devel mailing list, as there are more devs on the list than are on irc i think
[21:36:25 CEST] <llogan> does anyone know anyone who uses Haiku?
[21:36:36 CEST] <kierank> french people
[22:24:21 CEST] <cone-841> ffmpeg 03Stephan Holljes 07master:6d8c27f9512a: libavformat/http.c: Fix missing parenthesis in http_listen()
[22:34:27 CEST] <cone-841> ffmpeg 03Andreas Cadhalpun 07master:1b8d071734d6: fate: add mpeg4_unpack_bframes bsf test
[23:23:18 CEST] <kierank> any idea how to make function arguments appear in gdb
[23:23:27 CEST] <kierank> #0  ff_put_pixels16_sse2.loop () at libavcodec/x86/fpel.asm:106
[23:23:28 CEST] <kierank> #1  0x00000000005f0673 in put_h264_qpel16_mc00_sse2 ()
[23:23:28 CEST] <kierank> #2  0x00000000004d3675 in mc_part_422_simple_8 ()
[23:23:28 CEST] <kierank> #3  0x00000000004d4ed6 in hl_motion_422_simple_8 ()
[23:23:28 CEST] <kierank> #4  0x00000000004db5d0 in hl_decode_mb_simple_8 ()
[23:23:28 CEST] <kierank> #5  0x000000000051865e in ff_h264_hl_decode_mb ()
[23:23:29 CEST] <kierank> is not useful
[23:27:52 CEST] <kepstin-laptop> well, if you go to a particular stack frame, you can print out the current contents of any variable that wasn't optimized away
[23:28:05 CEST] <kepstin-laptop> you can't see the original values that the function was called with, because those aren't stored anywhere
[23:29:20 CEST] <Daemon404> bt full
[23:29:21 CEST] <Daemon404> ?
[23:29:24 CEST] <Daemon404> oh asm
[23:31:12 CEST] <kierank> asm and crazy macros
[23:31:21 CEST] <kierank> oh and nondeterministic crash
[23:31:57 CEST] <kierank> adding printf makes it go away
[23:32:37 CEST] <jamrial> yay heisenbugs
[23:32:57 CEST] <kierank> yay multiple h264 rewrites with a lack of coordination
[23:34:29 CEST] <kepstin-laptop> if it used to work, can it be git bisected maybe?
[23:34:47 CEST] <kierank> difficult to tell if it used to work
[23:34:52 CEST] <kierank> because you get different crashes
[23:38:18 CEST] <cone-841> ffmpeg 03Michael Niedermayer 07master:4ec5762e253c: tests/fate/mpeg4: add bitexact flag so as to avoid the version number from being included in the checksum
[23:51:55 CEST] <kierank> http://stackoverflow.com/questions/19229882/gdb-says-no-symbol-table-but-nm-shows-file-has-debug-symbols
[23:51:59 CEST] <kierank> this is the reason apparently
[00:00:00 CEST] --- Tue Apr  7 2015