[Ffmpeg-devel-irc] ffmpeg-devel.log.20150712

burek burek021 at gmail.com
Mon Jul 13 02:05:02 CEST 2015

[00:14:15 CEST] <jamrial> durandal_1707: what does libavutil/cpu-test print?
[00:14:53 CEST] <cone-029> ffmpeg 03Michael Niedermayer 07master:944a1aa3c594: avformat/yuv4mpegdec: Use 64bit for file offset
[00:17:00 CEST] <jamrial> judging by the stuff in libavutil/x86/cpu.c, your cpu seems to be among the ones that toggle the sse2slow flag rather than sse2 like nevcairiel, so that would explain it
[00:20:05 CEST] <durandal_1707> well its not that slow but gives different results
[00:26:05 CEST] <durandal_1707> michaelni: on how many cores you tested w3fdif?
[01:09:15 CEST] <michaelni> durandal_1707, 6core 
[01:10:44 CEST] <michaelni> that is 6 real cores 12 if you count hypethreading
[01:11:33 CEST] <michaelni> i tested using time ./ffmpeg ...
[01:16:02 CEST] <durandal_1707> michaelni: maybe lut and stereo3d will have bigger speed up..
[01:18:37 CEST] <durandal_1707> sorry for sending 3x times same thing, something was wrong with gmail
[01:33:06 CEST] <BBB> if I have a value in xmm0, and on 32bit I want to return that value from the function, and its float
[01:33:13 CEST] <BBB> the abi says it has to be in st0 or fp0'
[01:33:17 CEST] <BBB> how do I do that in yasm?
[01:36:20 CEST] <BBB> I guess theres no way and I need to use stack as in-between
[01:36:21 CEST] <BBB> blegh
[01:56:33 CEST] <BBB> durandal_1707: theres some more simd for you to review
[01:57:05 CEST] <J_Darnley> yes, as far as I
[01:57:25 CEST] <BBB> I guess having tests for vf_ssim/psnr would be useful at this point, anyone wanna write some?
[01:57:36 CEST] <J_Darnley> yes, as far as I know you cannot do xmm<->fp stack directly
[01:57:46 CEST] <BBB> ok, thats what it looked like
[01:57:47 CEST] <BBB> sad
[01:57:48 CEST] <BBB> ohwell
[02:17:49 CEST] <jamrial> BBB: check scalarproduct_float in lavu/x86/float_dsp.asm
[02:24:27 CEST] <BBB> oh that has a good horizontal add idea
[02:39:44 CEST] <BBB> is there proper sample code for lavfi integration in an application that provides its own input data?
[02:40:15 CEST] <BBB> like, lets say Id like to think of the ssim filter as a push or pull thing where I have two frames, I want to call a function, and I want to get the ssim value. how would I do that?
[02:40:28 CEST] <BBB> (ideally without massively redesigning the application)
[02:55:37 CEST] <BBB> apparently buffer+buffersink can do that
[05:15:50 CEST] <Compn> nice someone cracked audible eh
[05:15:51 CEST] <Compn> :P
[09:05:16 CEST] <pross> trac error: https://trac.ffmpeg.org/ticket/2252 -- "Genshi UnicodeEncodeError error while rendering template (unknown template location)"
[11:47:22 CEST] <michaelni> pross, is this the error described in http://trac.edgewall.org/ticket/11218 ? 
[12:08:41 CEST] <durandal_1707> michaelni: is 2nd w3fdif threading patch ok?
[12:31:23 CEST] <pross> michaelni: yes. the trac entry was recently updated, and now it does not render.
[12:48:52 CEST] <michaelni> durandal_1707, 0m0.593s ->0m0.423s for time make THREADS=12 fate-filter-w3fdif-complex fate-filter-w3fdif-simple without vframes limit
[12:49:42 CEST] <durandal_1707> gussing fate passes still
[12:49:57 CEST] <durandal_1707> *guessing
[12:51:03 CEST] <michaelni> iam testing atm
[13:10:13 CEST] <michaelni> durandal_1707, seems working
[14:20:20 CEST] <BBB> do we support float metadata?
[14:20:42 CEST] <BBB> like, lets say I wanted to export the metadata in ssim/psnr as floats instead of strings, so I can control the formatting myself, is that possible in the api?
[14:21:50 CEST] <J_Darnley> I thought it was string only.  Abuse it and write the bytes as a hex string?
[14:22:16 CEST] <ubitux> BBB: snprintf with "%f"
[14:22:31 CEST] <BBB> J_Darnley: omg that is hacky
[14:22:33 CEST] <BBB> I love it
[14:22:33 CEST] <ubitux> like we do in every other filter
[14:22:50 CEST] <BBB> ubitux: maybe this would be an interesting future extension?
[14:22:54 CEST] <BBB> generic value system or so
[14:23:12 CEST] <ubitux> sure, but have fun redesigning the AVDictionary API
[14:24:38 CEST] <ubitux> git grep SET_META will give you a bunch of hints in other filters
[14:24:46 CEST] <ubitux> -i even
[14:27:03 CEST] <cone-725> ffmpeg 03Michael Niedermayer 07master:cdb0d7e43993: avformat/brstm: Remove unused variable
[14:27:04 CEST] <cone-725> ffmpeg 03Ronald S. Bultje 07master:3a18d3fa0475: ivfenc: write duration at end-of-stream.
[14:52:26 CEST] <wm4> BBB: you could use hex floats (%a)
[14:52:52 CEST] <wm4> IMO a good compromise between being lossless and readable
[14:55:56 CEST] <BBB> ok let me try that
[15:10:08 CEST] <kurosu> %a is C99 - how much is that portable?
[15:10:46 CEST] <kurosu> (actually: supported by mingw / visual studio)
[15:12:07 CEST] <kurosu> BBB: regarding ssim, the code could go in (haven't looked at it much), but next step could be "what did the x264 guys think about it"
[15:12:29 CEST] <BBB> about what? the ssim code?
[15:12:40 CEST] <BBB> or the simd code?
[15:12:42 CEST] <BBB> or & ?
[15:13:02 CEST] <kurosu> simd ssim code
[15:13:30 CEST] <BBB> they usually dont care much about our simd code, I think (and they shouldnt)
[15:13:33 CEST] <kurosu> sure we may have another use from them, but maybe it's worth it
[15:13:49 CEST] <kurosu> I mean the reverse: what tricks did they use
[15:13:56 CEST] <kurosu> and let's use that
[15:14:45 CEST] <kurosu> but as it's negligable gain, I didn't mean that to stop the current code to be commited
[15:15:31 CEST] <cone-725> ffmpeg 03Luca Barbato 07master:355864ef7a95: g726: Do not crash on user mistake
[15:15:32 CEST] <cone-725> ffmpeg 03Michael Niedermayer 07master:e72988d3f804: Merge commit '355864ef7a9548ee6491a25de1e0650bd983a667'
[15:16:08 CEST] <BBB> oh I see waht you mean
[15:16:15 CEST] <BBB> yeah we could look at what they did
[15:30:03 CEST] <cone-725> ffmpeg 03Luca Barbato 07master:d80811c94e06: riff: Use the correct logging context
[15:30:04 CEST] <cone-725> ffmpeg 03Michael Niedermayer 07master:ba77fb61f741: Merge commit 'd80811c94e068085aab797f9ba35790529126f85'
[16:09:54 CEST] <cone-725> ffmpeg 03Luca Barbato 07master:c1d647b15afa: mp3: Make the seek more robust
[16:09:55 CEST] <cone-725> ffmpeg 03Michael Niedermayer 07master:7da1f00e79eb: Merge commit 'c1d647b15afa69fa70f999a9ddcb98346409fb4d'
[16:20:10 CEST] <cone-725> ffmpeg 03Andreas Cadhalpun 07master:189420cb5619: riffdec: prevent negative bit rate
[16:33:54 CEST] <cone-725> ffmpeg 03Andreas Cadhalpun 07master:f1bdc2343704: riff: Validate bitrate
[16:33:55 CEST] <cone-725> ffmpeg 03Michael Niedermayer 07master:cd326377414d: Merge commit 'f1bdc234370401c032cd85184e93c7c155eb6d62'
[16:33:56 CEST] <cone-725> ffmpeg 03Michael Niedermayer 07master:839d6bc192f7: avformat/riffde: Fix integer overflow in bitrate
[17:10:07 CEST] <cone-725> ffmpeg 03Hendrik Leppkes 07master:796268654c78: asfdec: always reset packet state after seeking
[17:10:08 CEST] <cone-725> ffmpeg 03Michael Niedermayer 07master:d2e5297e9369: Merge commit '796268654c7807c9a1cfb322c838383e2b900d60'
[17:25:48 CEST] <cone-725> ffmpeg 03Ronald S. Bultje 07master:f353b851b925: vf_ssim: fix s->coefs for yuv with non-4:2:0 subsampling.
[18:22:23 CEST] <livingBEEF> If I understand it correctly, paletteuse can't do transparent stuff. Any idea how hard it would be to rewrite it? I did some graphics stuff and C stuff, but I'm not really familiar with ffmpeg's codebase... 
[18:33:35 CEST] <cone-725> ffmpeg 03James Almer 07master:a3b721d10ddf: avcodec/dcadec: silence request_channels deprecation warnings
[18:56:42 CEST] <durandal_1707> livingBEEF: what you mean by transparent stuff?
[18:59:35 CEST] <livingBEEF> durandal_1707: alpha-channel is dropped https://trac.ffmpeg.org/ticket/4443
[19:03:44 CEST] <livingBEEF> from looking at vf_paletteuse.c it seems that the whole thing drops alpha channel in the beginning and then works only with rgb
[19:16:33 CEST] <durandal_1707> ask ubitux he worked on that filter
[19:20:00 CEST] <ubitux> livingBEEF: store the first transparent color id found in the palette, and reuse it whenever the input has transparency 
[19:20:12 CEST] <ubitux> paletteuse should be patchable that way
[19:20:54 CEST] <ubitux> i need to think a bit about how it's going to behave with the transparency optimization... maybe you should pick the 2nd transparent color, dunno
[19:24:36 CEST] <ubitux> livingBEEF: http://pastie.org/10288291 100% untested but you get the idea
[19:25:19 CEST] <ubitux> try to play along these lines
[19:26:06 CEST] <livingBEEF> thanks. Looks less complicated than I expected
[19:26:58 CEST] <ubitux> it will only handle 0% or 100% opacity
[19:27:37 CEST] <ubitux> that's just a guess on how to handle the issue, but i don't have time to look deeper currently
[19:28:10 CEST] <livingBEEF> yeah... still better than nothing; and it looks simple enough
[19:29:00 CEST] <cone-725> ffmpeg 03Niklesh 07master:774ceee17ed9: movtextdec.c: Add support for highlight and hilightcolor box
[19:30:55 CEST] <ubitux> what need to be check is animated gif (with transparent optimization) and transparent background
[19:31:13 CEST] <ubitux> they must not overlap
[19:31:53 CEST] <ubitux> so you need to make sure the transparent color that will be picked in the gif encoding will be different than the transparent "background" color 
[19:32:54 CEST] <ubitux> i didn't want to think about that mess originally (shit was complex enough already) so i just said fuck it to source transparency
[19:35:00 CEST] <wm4> apropos gif, does anyone have an idea how to determine the duration of gifs?
[19:35:46 CEST] <durandal_1707> full parsing feel free to write parser
[19:35:47 CEST] <rcombs> sum the frames' delay fields?
[19:37:00 CEST] <ubitux> duration = -1
[19:45:37 CEST] <Zeranoe> Probably an unrealistic requirement, but if you remove all the GPL licensed source files and try to compile FFmpeg as LGPL it will not compile without some minor modifications.
[19:47:30 CEST] <durandal_1707> Zeranoe: what modifications?
[20:34:47 CEST] <cone-725> ffmpeg 03Ganesh Ajjanagadde 07master:4547cf68a0d2: avformat/isom: utilize bitrate hints
[21:12:31 CEST] <durandal_1707> michaelni: does it make sense to push stereo3d slice threading?
[21:12:35 CEST] <cone-725> ffmpeg 03Henrik Gramner 07master:8bc67ec2c0d2: Checkasm: assembly testing and benchmarking tool
[21:12:36 CEST] <cone-725> ffmpeg 03Michael Niedermayer 07master:f14fc55969d8: Merge commit '8bc67ec2c0d2b5444d51a1bed1d50f0e10d92717'
[21:15:33 CEST] <michaelni> durandal_1707, i dont know, it was faster in the one test, so based on just that it would be a yes, you could do more tests to see if its always faster that is in more real world test case with multitreaded encoder and decoder
[21:21:44 CEST] <cone-725> ffmpeg 03Anton Khirnov 07master:86eee85daddb: bytestream2: set the reader to the end when reading more than available
[21:21:45 CEST] <cone-725> ffmpeg 03Michael Niedermayer 07master:98c7f9741edd: Merge commit '86eee85daddb682fa072c2e2657c90a514b855e3'
[21:31:15 CEST] <cone-725> ffmpeg 03Anton Khirnov 07master:a9a2f3613040: doc: add a section about building with libmfx
[21:31:16 CEST] <cone-725> ffmpeg 03Michael Niedermayer 07master:6727380d0527: Merge commit 'a9a2f3613040c4f90bf15cbd76f8671252ecc043'
[21:32:59 CEST] <cone-725> ffmpeg 03Andreas Cadhalpun 07master:9f4c7397a296: hevc: check slice address length
[21:33:00 CEST] <cone-725> ffmpeg 03Michael Niedermayer 07master:547848f71184: Merge commit '9f4c7397a296e6d11b3c6c121a6896163577dc7c'
[21:42:01 CEST] <cone-725> ffmpeg 03Anton Khirnov 07master:ab05ed4c322e: mpegvideo_parser: export pixel format and dimensions
[21:42:02 CEST] <cone-725> ffmpeg 03Michael Niedermayer 07master:ddd86e4747f2: Merge commit 'ab05ed4c322ed0488ac9b5d2ef5d4ffa55a946a7'
[22:07:57 CEST] <cone-725> ffmpeg 03Paul B Mahol 07master:c02ee69be9c5: fate: add tests for stereo3d anaglyph modes
[22:07:58 CEST] <cone-725> ffmpeg 03Paul B Mahol 07master:7b07e72abc03: avfilter/vf_stereo3d: implement slice threading for anaglyph output
[22:07:59 CEST] <cone-725> ffmpeg 03Paul B Mahol 07master:885afff01c72: avfilter/vf_w3fdif: implement slice threading
[22:27:16 CEST] <Kaedenn> 1) Hi! 2) <3 and 3) :D
[22:27:26 CEST] Action: Kaedenn hops back over to the other channel
[22:45:03 CEST] <cone-725> ffmpeg 03Anton Khirnov 07master:b11acd57326d: hevc: remove HEVCContext usage from hevc_ps
[22:45:04 CEST] <cone-725> ffmpeg 03Michael Niedermayer 07master:d5fcca83b915: Merge commit 'b11acd57326db6c2cc1475dd0bea2a06fbc85aa2'
[22:53:42 CEST] <cone-725> ffmpeg 03Anton Khirnov 07master:d7bebe480519: hevc: skip invalid/ignored NALUs when splitting the packet
[22:53:43 CEST] <cone-725> ffmpeg 03Michael Niedermayer 07master:760304e4c318: Merge commit 'd7bebe4805193783f0b6f292f9127a75709fb7d9'
[23:03:37 CEST] <cone-725> ffmpeg 03Anton Khirnov 07master:ae05b4865514: hevc: eliminate the second call to hls_nal_unit()
[23:03:38 CEST] <cone-725> ffmpeg 03Michael Niedermayer 07master:4c42c66935a7: Merge commit 'ae05b4865514fd71b5e9431e93aa0d03d7ba7751'
[23:45:39 CEST] <cone-725> ffmpeg 03Anton Khirnov 07master:d82e1adc2019: hevc: move splitting the packet into NALUs into a separate function
[23:45:40 CEST] <cone-725> ffmpeg 03Michael Niedermayer 07master:4690a6363246: Merge commit 'd82e1adc2019135a2fc45372e0ed0b5ef107cdd0'
[23:52:30 CEST] <cone-725> ffmpeg 03Anton Khirnov 07master:7f78155dc6b6: hevc: improve a comment
[23:52:31 CEST] <cone-725> ffmpeg 03Michael Niedermayer 07master:93aa84c1ea9a: Merge commit '7f78155dc6b65be55efb5309b6dd2bb33925a8c4'
[23:54:58 CEST] <Guest78083> hi, I am wondering about how to activate "audio_chunk_duration" AVOption in webm_chunk.c? While I can pass the AVOption correctly, the corresponding demuxer doesn't seem to get it. is there magic involved here? :)
[23:59:27 CEST] <jamrial> make sure you put the demuxer options before the input file. and in any case you can use -loglevel debug to see how it's being parsed
[00:00:00 CEST] --- Mon Jul 13 2015

More information about the Ffmpeg-devel-irc mailing list