[Ffmpeg-devel-irc] ffmpeg-devel.log.20160122

burek burek021 at gmail.com
Sat Jan 23 02:05:03 CET 2016

[00:32:51 CET] <cone-469> ffmpeg 03Andreas Cadhalpun 07master:064963bd2702: build: make out-of-tree builds bit-identical to in-tree builds
[00:41:32 CET] <Timothy_Gu> I really wish I could use some of the modern services for web development for FATE. The current file-based not-even-a-database is a pain to manipulate in large numbers
[04:39:01 CET] <cone-047> ffmpeg 03Neil Birkbeck 07master:17399f6a9f00: Add missing conversions from side data enum to name
[04:39:01 CET] <cone-047> ffmpeg 03John Cox 07master:48f80831bad8: cabac: Ensure 2-byte cabac loads are on 2-byte boundry
[04:39:01 CET] <cone-047> ffmpeg 03John Cox 07master:5115d8326e8e: cabac_functions: Count zeros with ctz if it is fast
[04:39:01 CET] <cone-047> ffmpeg 03John Cox 07master:74105fc9a3c4: cabac_functions: Allow more functions to be overridden
[10:57:18 CET] <durandal_1707> I have idea how to do conditional filter
[11:24:39 CET] <cone-711> ffmpeg 03Paul B Mahol 07master:b18467a9433c: avfilter/avf_showwaves: add logarithmic amplitude scaler
[12:08:42 CET] <cone-711> ffmpeg 03Paul B Mahol 07master:1c02af307a67: avfilter: add convolution filter
[13:12:21 CET] <durandal_1707> please comment streamselect patch
[13:14:19 CET] <ubitux> sorry, i will
[13:14:47 CET] <ubitux> as usual i'm in a rush on many things, which will calm down pretty soon
[13:20:01 CET] <cone-711> ffmpeg 03Michael Niedermayer 07master:70ece6ccd0c3: fate: add 1 and 15BPP rawvideo in avi tests
[13:26:46 CET] <Franciman> Hi
[13:26:57 CET] <Franciman> What's the difference between avcodec_decode_audio3 and avcodec_decode_audio4 ?
[13:28:25 CET] <jermy> The number?
[13:29:01 CET] <nevcairiel> 3 is deprecated, 4 is the replacement with a new interface
[13:29:09 CET] <Franciman> I see
[13:29:18 CET] <jermy> But seriously, it's largely that the former required passing in a pre-allocated buffer of the right size, and the latter returns a AVPacket instead
[13:29:56 CET] <jermy> (What size should the buffer have been? A very good question)
[13:30:05 CET] <Franciman> I see, ok thank you
[14:02:46 CET] <nevcairiel> A year later, still dealing with fallout from "refactoring"
[14:06:11 CET] <nevcairiel> I should setup some sort of unit tests to make FATE verify that the struct content of hwaccels doesnt change randomly
[14:07:23 CET] <wm4> which refactoring?
[14:08:15 CET] <nevcairiel> in this case, this commit http://git.videolan.org/?p=ffmpeg.git;a=commitdiff;h=a12d3188cbec15e22070e139fa5cc541da07e2c3
[14:09:11 CET] <wm4> oh
[14:09:34 CET] <nevcairiel> (its not the first dxva regression this particular patch series caused, but this one took ages to surface, apparenly)
[14:10:59 CET] <Franciman> thanks for your help, bye
[14:24:34 CET] <jkqxz> wm4:  I've updated the VAAPI encode patch to use refcounting on AVFrames/AVBuffers throughout.  Thank you for your thoughts on that; the libavutil parts in particular are much more sensible now.
[14:25:19 CET] <jkqxz> Unfortunately I seem to have broken something low-level in H.265 encode in the process (segfaults deep in the Intel driver, yay).  Hopefully I will be able to fix that and post a new version tomorrow.
[14:26:00 CET] <wm4> ok
[14:35:50 CET] <BtbN> One could argue that the intel driver is what is broken here.
[14:46:50 CET] <jkqxz> The Intel driver is very fun.  Oh, you stupidly set some option to a value we don't support?  Have a GPU hang.  Hopefully the kernel will notice within a few seconds and reset everything.
[15:04:44 CET] <cone-711> ffmpeg 03Ganesh Ajjanagadde 07master:89189b1010de: lavc/wma: add av_warn_unused_result to ff_wma_init
[15:04:45 CET] <cone-711> ffmpeg 03Ganesh Ajjanagadde 07master:9254e6176c1a: lavc/atrac3plusdsp: change pow(2,x) to exp2f(x)
[15:04:46 CET] <cone-711> ffmpeg 03Ganesh Ajjanagadde 07master:cea529dd7f0d: lavc/opus_celt: replace pow(2,x) by exp2f(x)
[15:04:47 CET] <cone-711> ffmpeg 03Ganesh Ajjanagadde 07master:1fb5ae44c43d: lavc/wmadec: replace pow by faster functions
[15:15:11 CET] <esdwdftty>  Make decoding vp9, HEVC on OpenCl CPU + GPU offers a good trade-off http://www.arm.com/zh/files/event/B-2.2014_ARM_Multimedia_Seminar_Ittiam.pdf
[15:31:51 CET] <esdwdftty> <rektimus> vp9 @ 200kbps doesnt seem bad
[15:31:53 CET] <esdwdftty> <rektimus> for a 340p video
[15:32:33 CET] <esdwdftty>  i know, i only have on gpu, h264, vc1, mpg2, mpg4. the pdf talks about "Mali" gpu (tablets, smartphones) I showed the idea on the example of Mali: CPU + GPU / Open CL
[15:42:23 CET] <cone-711> ffmpeg 03Perette Barella 07master:84110f4f7760: libavformat/tcp.c : add send_buffer_size and recv_buffer_size options
[15:59:34 CET] <wm4> durandal_1707: I'm fairly sure libavfilter doesn't support dynamic reconfiguration
[15:59:38 CET] <wm4> even if vf_scale pretends to
[16:00:08 CET] <wm4> the buffersink filter for one doesn't update its parameters when a change happens
[16:00:28 CET] <wm4> and I bet buffersrc (or most other filters) don't support it either
[16:02:19 CET] <durandal_1707> w/h change is supported
[16:02:57 CET] <wm4> not really
[16:04:03 CET] <wm4> or why does buffersink not adjust the input pad parameters?
[16:04:16 CET] <durandal_1707> call avfilter_config_links
[16:05:55 CET] <wm4> who?
[16:07:04 CET] <wm4> bufferink's inputs[0]->w isn't update when I change the width in vf_scale before it
[16:08:50 CET] <durandal_1707> vf_scale doesn't call that function
[16:09:29 CET] <durandal_1707> but it probably should
[16:09:58 CET] <wm4> ???
[16:10:13 CET] <wm4> and such a garbage patch was applied? that's baffling
[16:10:29 CET] <wm4> hurrr let's add dynamic reconfiguration without taking any of the design into account?
[16:10:31 CET] <durandal_1707> what patch?
[16:10:49 CET] <wm4> the one which added size changes to vf_scale by commands
[16:11:12 CET] <durandal_1707> ah, you mean when input change
[16:13:28 CET] <durandal_1707> some filters are going to be reset, others are intra only so it doesn't matter for them
[16:20:07 CET] <durandal_1707> wm4: about design you can ask elenril
[16:20:56 CET] <wm4> the buffersink's filter link is not updated, and it returns a frame of the wrong configuration - this is a bug
[16:21:26 CET] <wm4> and it seems to be a rather fundamental bug, rather than some oversight in buffersink
[16:22:50 CET] <durandal_1707> you mean link w/h is not updated or also AVframe is not?
[16:23:05 CET] <wm4> they don't match
[16:24:45 CET] <durandal_1707> hmm, let me try something...
[16:46:19 CET] <cone-711> ffmpeg 03Michael Niedermayer 07master:ee3eac59df10: fate: Add a small number of rawvideo in mov tests
[17:07:16 CET] <Daemon404> almost have it build with clang-cl... 
[17:11:05 CET] <durandal_1707> wm4: not all links are updated with avfilter_config_links
[17:13:04 CET] <durandal_1707> michaelni: crash if input size changes with bunch of video encoders still apply
[17:13:37 CET] <wm4> the situation with encoders is similar
[17:13:52 CET] <wm4> normally encoders are configured with a width and height before sending the first frame, right?
[17:14:07 CET] <wm4> so you just trigger undefined behavior by sending a frame with different configuration
[17:24:46 CET] <nevcairiel> Daemon404: i didnt bother yet, its still too limited todaa
[17:24:48 CET] <nevcairiel> today*
[17:26:58 CET] <wm4> is clang-cl the windows thing?
[17:27:38 CET] <nevcairiel> technically its called clang/c2, but i assume thats what he meant
[17:28:28 CET] <Daemon404> limited?
[17:28:46 CET] <nevcairiel> didnt you read the long list of limitations
[17:29:03 CET] <Daemon404> most of that is either: a) debug info b) C++ related
[17:29:10 CET] <Daemon404> i dont recall a ton else
[17:29:50 CET] <nevcairiel> although there was a new release just two days ago, maybe they fixed a bunch
[17:30:13 CET] <michaelni> durandal_1707, how to reproduce that crash ? but most encoders dont support that currently
[17:30:53 CET] <nevcairiel> Daemon404: no inline asm yet either =p
[17:31:45 CET] <Daemon404> eh?
[17:31:50 CET] <Daemon404> i am building with inline asm right now.
[17:32:06 CET] <Daemon404> unless it secretly ignores it
[17:32:14 CET] <Daemon404> this is SVN trunk
[17:33:05 CET] <Daemon404> granted i patched configure and mem.h
[17:34:58 CET] <nevcairiel> Daemon404: it said no inline asm in their announcement as a known limitation right now,  either it was fixed, or your configure just disables it
[17:35:12 CET] <nevcairiel> configure has a check that tries to use it and d isables it if it fails, fwiw
[17:35:23 CET] <durandal_1707> michaelni: see patch I just posted
[17:35:46 CET] <Daemon404> i checkec config.h
[17:35:47 CET] <Daemon404> its enabled
[17:37:40 CET] <Daemon404> ah damn...
[17:37:41 CET] <Daemon404> libavcodec.a(lpc.o) : error LNK2019: unresolved external symbol pd_1 referenced in function lpc_apply_welch_window_sse2
[17:37:44 CET] <Daemon404> libavcodec.a(lpc.o) : error LNK2019: unresolved external symbol pd_2 referenced in function lpc_apply_welch_window_sse2
[17:37:49 CET] <Daemon404> it looks like it fails at the asm constanta.
[17:56:25 CET] <cone-711> ffmpeg 03Paul B Mahol 07master:8a5d4a51fff7: avfilter/vf_convolution: add timeline support
[18:15:11 CET] <cone-711> ffmpeg 03Rostislav Pehlivanov 07master:ce855bf43bcb: diracdec: fix full range videos
[18:16:27 CET] <atomnuker> well, this opens up a nice niche use case for the encoder: near lossless desktop recording with a 1:5 compression ratio
[18:42:20 CET] <Daemon404> nevcairiel, i got inline asm to build.
[21:44:42 CET] <cone-711> ffmpeg 03Paul B Mahol 07master:805563c814bf: avfilter/af_ladspa: accepts ' ' as command separator too
[22:15:20 CET] <durandal_1707> why there is no API to check number of available frames in sink?
[00:00:00 CET] --- Sat Jan 23 2016

More information about the Ffmpeg-devel-irc mailing list