[Ffmpeg-devel-irc] ffmpeg-devel.log.20190104

[burek]

[01:44:38 CET] <jamrial> michaelni: valgrind complains about a recent change to rangecoder by you: http://fate.ffmpeg.org/report.cgi?time=20190103234808&slot=x86_64-archlinux-gcc-valgrind
[03:37:42 CET] <cone-757> ffmpeg 03James Almer 07master:ef67af31ffec: x86/af_afir: use three operand form forat some instructions
[04:16:43 CET] <cone-757> ffmpeg 03Steven Liu 07master:2780cd33d938: fate: add tests/fate/hlsenc.mak for hls FATE
[04:23:48 CET] <cone-757> ffmpeg 03Steven Liu 07master:eb81fd792fa8: avcodec/fft_template: improve performance of the ff_fft_init in fft_template
[06:30:00 CET] <cone-757> ffmpeg 03Steven Liu 07master:b9523f2fd316: avformat/hlsenc: free variant streams after write all variant streams info
[10:53:14 CET] <cone-899> ffmpeg 03Gyan Doshi 07master:d7ef10f16d56: doc/ffmpeg: update entry for stream disable options
[11:20:15 CET] <cone-899> ffmpeg 03Gyan Doshi 07master:1087b792c376: doc/ffmpeg: clarify behaviour of -discard
[13:05:11 CET] <Easyfab> Trying to build with msys2 I got this error for af_sofalizer https://pastebin.com/jgrr731p . Is it my build system ?
[13:07:00 CET] <nevcairiel> your sofa library is probably  incompatible, possibly out of date
[13:20:25 CET] <Easyfab> Msys2 latest version is mingw64/mingw-w64-x86_64-libmysofa 0.6-1
[13:20:25 CET] <Easyfab>  . So I need to build  a more recent dev version ?
[13:31:51 CET] <Easyfab> It's Ok now, thanks
[14:18:15 CET] <cone-899> ffmpeg 03Jun Zhao 07master:807b57a31b4e: lavc/libdavs2: Cosmetics: Fix indentation for switch statement
[14:18:16 CET] <cone-899> ffmpeg 03Jun Zhao 07master:796da4f3e083: lavc/libxavs2: Cosmetics: Fix indentation for switch statement
[14:18:17 CET] <cone-899> ffmpeg 03Jun Zhao 07master:4fd6df6934d1: doc/encoders: Update docs for libxavs2
[16:01:25 CET] <cone-899> ffmpeg 03nu774 07master:e0686318ddd7: avformat/vapoursynth: properly initialize err variable in read_header_vs()
[22:19:34 CET] <durandal_1707> i wrote almost perfect audio denoiser but it is extremly slow
[22:20:47 CET] <atomnuker> how so?
[22:20:52 CET] <jamrial> rewrite it in opencl
[22:21:14 CET] <cone-297> ffmpeg 03Michael Niedermayer 07master:5faa1b8b5a04: tools/target_dec_fate: Add entries from around issue 500 to 700
[22:22:00 CET] <durandal_1707> atomnuker: ~0.06 realtime for 48000 sample rate
[22:23:08 CET] <atomnuker> what's it using as a basis? ffts or iir filters?
[22:23:30 CET] <durandal_1707> non-local mean
[22:26:18 CET] <atomnuker> ah, yeah, that'd be slow
[22:26:37 CET] <atomnuker> you can slice thread it though, can't you?
[22:32:12 CET] <durandal_1707> i could, but i do only channels separately
[22:32:55 CET] <durandal_1707> this thing only works rt with 8k rate, at least as described in scientific papers
[22:38:13 CET] <atomnuker> can it be simd'd?
[22:38:36 CET] <atomnuker> assuming floats and avx, 0.06*8 = 0.48, halfway there
[22:41:18 CET] <durandal_1707> sad/ssd could be simdable, but big slowup is caused by expf * (Search window size) - 600 for 48000 rate
[22:45:33 CET] <atomnuker> got code up somewhere?
[22:56:35 CET] <durandal_1707> atomnuker: https://github.com/richardpl/FFmpeg/tree/anlmdn
[23:01:18 CET] <durandal_1707> actually >90% time is spend on ssd
[23:01:33 CET] <durandal_1707> so no point in touching expf
[23:02:20 CET] <durandal_1707> also volume needs to be renormalized somehow
[23:04:44 CET] <atomnuker> mov m0, [src1] ; subps m0, [src2] ; mulps m0, m0 ; addps m1, m0 ; jmp start ; <horizontal add left as an exercise for the reader>
[23:05:49 CET] <atomnuker> I can see plenty of overlap during ssd computation so further on with some reordering you could probably reduce it
[00:00:00 CET] --- Sat Jan  5 2019