[Ffmpeg-devel-irc] ffmpeg-devel.log.20140315

[burek]

[00:06] <llogan> shit. forgot a period. OCD OCD OCD.
[00:50] <gueb> how can I configure ffmpeg to use -pg?
[00:52] <gueb> -pg is a flag for profiling with gcc
[00:54] <BBB_> gueb: --extra-cflags="-pg"
[00:54] <gueb> thanks
[00:54] <BBB_> gueb: also, user questions in #ffmpeg-users please
[00:55] <gueb> profiling is developing, isn't it?
[00:55] <BBB_> passing a flag to configure isn't really
[00:55] <wm4> does gprof really have any worth?
[00:56] <gueb> wm4, what do you recommend for profiling?
[00:56] <wm4> on linux, maybe the perf utility
[00:56] <wm4> or counting cycles like ffmpeg devs seem to love it
[00:56] <wm4> BBB should know a lot about this
[00:56] <BBB_> I've used perf, it's pretty good
[00:57] <BBB_> kcachegrind was quite useful, in combination with valgrind, but that's long ago
[01:05] <iive> i've done gprof on mplayer on my previous athlon tbird. the speed up was quite miniature, about 1%.
[01:05] <iive> and I don't consider that cpu to be modern. :)
[01:09] <BBB_> gueb: anyway, perf, really - don't use gprof
[01:09] <gueb> ok
[01:09] <gueb> I have a performance problem
[01:09] <gueb> when I compile the code externally
[01:09] <gueb> it does some work in less than 5 ms
[01:10] <gueb> when doing inside ffmpeg it uses about 40 ms
[01:10] <gueb> :S
[01:11] <gueb> any advice?
[01:12] <BBB_> gueb: I don't know what code you're talking about so it's really hard to say
[01:12] <gueb> it is a part of my new codec
[01:12] <BBB_> also 5ms is a weird metric, we typically talk about cycles for performance-sensitive stuff (unless this is sth. like network io or so)
[01:13] <gueb> I know that there is a performance problem because the file is processed very slow
[01:13] <BBB_> I'd try perf and see where it's spending its time
[01:13] <gueb> with some other tests outside ffmpeg it was fast
[01:14] <BBB_> I don't have enough context to answer your question, I'd need things like code, compile flags, sample files, etc. - probably a spec of the codec would be useful although not necessary
[01:14] <BBB_> maybe you compiled ffmpeg with -O0?
[01:14] <gueb> I do not know, I just added the module to Makefile
[01:14] <gueb> OBJ-... += foo.c
[01:17] <iive> gueb: is the code complete codec, or just some routine?
[01:17] <gueb> some routine
[01:19] <iive> when you do synthetic benchmark, you usually totally ignore the cache pollution.
[01:29] <cone-797> ffmpeg.git 03Michael Niedermayer 07master:2b8d28439b6e: avcodec/h264_cabac: move the arm unchecked_bitstream reader special case closer to where the issue is
[01:29] <cone-797> ffmpeg.git 03Michael Niedermayer 07master:41d08ca57564: avcodec/arm/cabac: fix inline cabac reader with the UNCHECKED bitstream reader
[01:58] <gueb> solved, the problem was a spurious initialization that I had to put to avoid an absurd and annoying warning
[01:59] <gueb> warning that only appeared with -O3
[02:09] <cone-797> ffmpeg.git 03Peter Ross 07master:d1bb17940dd2: avcodec/adpcm: ADPCM_IMA_DK3 packets are padded to 16-bit packet boundary
[02:33] <rcombs> can I get lavf/lavc to give me MKV-format ASS instead of text-file-format?
[02:35] <wm4> rcombs: at what point? lavf should give you always the mkv format
[02:35] <wm4> but lavc output not
[02:35] <wm4> (and that's annoying)
[02:35] <rcombs> AFAICT matroskadec spits out the wrong thing
[02:35] <wm4> makes me wonder, how does ffmpeg.c do srt-to-ass and then mux into mkv
[02:36] <wm4> rcombs: used to
[02:36] <wm4> and Libav probably still does
[02:40] Action: rcombs stares upon Dialogue: 0,0:00:39.93,0:00:46.02,op trans,,0,0,0,,{}Both our infinite dreams and our sketched out future\r\n
[02:41] <rcombs> oh, damn, that's my fault
[02:44] <rcombs> erm, wait, no it's not
[02:45] <rcombs> how do I force AV_CODEC_ID_ASS over AV_CODEC_ID_SSA in lavf?
[02:45] <wm4> oh maybe it's still not bumped
[02:46] <wm4> it's connected to an api bump
[02:46] <rcombs> yeah, it is
[02:46] <rcombs> but that seems to imply it'd be overrideable at runtime
[02:56] <cone-797> ffmpeg.git 03Stephen Hutchinson 07release/2.2:2b9ee7d5b901: doc: Add section about AviSynth support
[02:56] <cone-797> ffmpeg.git 03Michael Niedermayer 07release/2.2:29df24252aa9: Merge commit '2b9ee7d5b901e0d7ba617511e4ed31d3043894d3' into release/2.2
[02:56] <cone-797> ffmpeg.git 03Stephen Hutchinson 07release/2.2:c4f5f4dbd303: doc/general.texi: Adjust the notes on AviSynth
[02:56] <cone-797> ffmpeg.git 03Matt Oliver 07release/2.2:bf08665e2ede: Fix modplug linkage on Windows.
[02:56] <cone-797> ffmpeg.git 03Carl Eugen Hoyos 07release/2.2:3caa6a5a5744: Revert "Allow stream-copying grayscale mov files."
[02:56] <cone-797> ffmpeg.git 03Michael Niedermayer 07release/2.2:46c2dba20e09: doc/texi2pod: fix encoding type
[02:56] <cone-797> ffmpeg.git 03Michael Niedermayer 07release/2.2:242df26b4440: mvformat/movenc: fix IMX
[02:56] <cone-797> ffmpeg.git 03Michael Niedermayer 07release/2.2:70e3cc282bee: avutil/timestamp: Warn about missing __STDC_FORMAT_MACROS for C++ use
[02:56] <cone-797> ffmpeg.git 03Michael Niedermayer 07release/2.2:125bea15d1f6: avcodec/libx265: fill headers in extradata
[03:22] <cone-797> ffmpeg.git 03Michael Niedermayer 07release/2.2:6ba07e9948d5: update for 2.2-rc2
[04:26] <cone-797> ffmpeg.git 03Peter Ross 07fatal: ambiguous argument 'refs/tags/n2.2-rc2': unknown revision or path not in the working tree.
[04:26] <cone-797> Use '--' to separate paths from revisions
[04:26] <cone-797> refs/tags/n2.2-rc2:HEAD: avcodec/adpcm: ADPCM_IMA_DK3 packets are padded to 16-bit packet boundary
[04:58] <cone-797> ffmpeg.git 03Peter Ross 07master:dadf668df20e: avformat/mov: decode compilation metadata
[04:58] <cone-797> ffmpeg.git 03Peter Ross 07master:ac9bfb1fafde: avformat/movenc: encode compilation metadata
[04:58] <cone-797> ffmpeg.git 03Peter Ross 07master:847d8af59af5: avformat/id3v2: decode compilation metadata
[05:32] <gueb> good night
[06:58] <yanrong> Is there anyone know the difference between ff_fft_calc_interleave_xx and ff_fft_calc_xx? When to use respectively, thanks.
[15:52] <cone-678> ffmpeg.git 03Michael Niedermayer 07master:da89572004f6: avcodec: Move STRIDE_ALIGN to internal.h
[15:52] <cone-678> ffmpeg.git 03Michael Niedermayer 07master:06a3185e3861: avcodec/mpegvideo_enc: dont use direct mode for unaligned input
[17:27] <jnvsor> With regards to bug 3192, the summary is wrong, can I change it or should I leave that to a dev?
[17:30] <Compn> jnvsor : change it :)
[17:30] <Compn> if you have a dev account on the bug
[17:30] <Compn> not sure if users can change it
[17:30] <Compn> you might face carl in mortal kombat, but thats always fun to watch
[17:31] <jnvsor> Compn: Cool :)
[18:15] <wm4> interplayvideo is broken
[18:16] <wm4> block soup on descent 2 movies
[19:58] <kierank> these guys are hillarious
[19:59] <kierank> they put  all their software in a hidden folder
[19:59] <JEEB> lol
[20:02] <kierank> it's literally every media software under the sun in that hidden folder
[20:02] <kierank> from jm to realmedia
[20:03] <JEEB> sounds like a payware version of SUPER, with a hidden directory :D
[20:09] <kierank> basically yes
[20:41] <J_Darnley> kierank: Thet should make it more secret by making it a system file too
[20:41] <J_Darnley> uh folder in this case
[22:23] <cone-678> ffmpeg.git 03Michael Niedermayer 07master:8803b970ef98: ffmpeg_opt: Check ist is valid before using it in audio_channels_map init
[22:58] <cone-678> ffmpeg.git 03Michael Niedermayer 07master:6c47a4e97248: swscale/x86/swscale: fix missing xmm clobbers in yuv2yuvX_sse3()
[00:00] --- Sun Mar 16 2014