[Ffmpeg-devel-irc] ffmpeg-devel.log.20160705

burek burek021 at gmail.com
Wed Jul 6 02:05:03 CEST 2016


[00:11:37 CEST] <kode54> bp0, correct
[00:11:47 CEST] <kode54> don't know what to do with that information, if anything
[00:11:56 CEST] <kode54> this filter was derived from a closed source library
[00:12:04 CEST] <kode54> which I assume was derived from Microsoft's code
[00:12:13 CEST] <kode54> it's been rewritten a few times
[01:51:00 CEST] <cone-032> ffmpeg 03Martin Vignali 07master:9cc114fa5428: fate/png : add test for rgba64 and interleaved rgb
[01:51:00 CEST] <cone-032> ffmpeg 03Martin Vignali 07master:ce466d0a1558: fate/apng : add test for apng decoding
[03:23:18 CEST] <atomnuker> grr, dnxhr
[03:26:35 CEST] <atomnuker> I'd have a yacht and a gold plated rolls royce if we had a separate codec id for every single bitstream variant of every single codec
[03:27:22 CEST] <atomnuker> *if I had a dollar for each
[03:27:57 CEST] <atomnuker> lacking sleep, should go
[03:42:41 CEST] <TD-Linux> why would you buy a yacht with that money?
[06:11:54 CEST] <Joker> hello?
[06:15:29 CEST] <Joker> I am an undergraduate student. Does anyone know how to implement H.264 standard in C?
[06:18:16 CEST] <Joker> Can anybody answer my question?
[06:22:52 CEST] <smarter> you read the specification, and then you implement it
[09:58:48 CEST] <cone-907> ffmpeg 03smitbose 07master:d915b6e59230: Added Quadrox format
[15:48:17 CEST] <BBB> so on ppc64...
[15:48:35 CEST] <BBB> going from 30k units to 20k
[15:48:39 CEST] <BBB> in going from c to simd
[15:48:50 CEST] <BBB> isnt that very & unexciting?
[15:49:04 CEST] <BBB> I mean, at a dsp function level, isnt simd supposed to give Nx improvement in speed?
[15:49:08 CEST] <BBB> where N is like 2, 3, 4 or so
[15:57:28 CEST] <jkqxz> Yes.  For the one significant case I looked at (RGB -> YUV, with N = 4) the code looked sensible, so I think there must be something else going on (it might just be intrinsics sucking, of course).
[15:59:40 CEST] <BBB> how much speed improvement do we get in that same function using sse2?
[15:59:49 CEST] <BBB> (Im blindly assuming we already have an sse2 implementation)
[16:00:30 CEST] <BBB> right, nm shows _ff_rgb24ToY_sse2/ssse3/avx implementations
[16:01:46 CEST] <BBB> especially given that this is probably one of our first ppc64 simd implementations, I Think it might make sense spending some time getting the fundamentals right here, sort of as groundwork to make future ppc64 simd work very trivial
[16:20:01 CEST] <jkqxz> It's the same as other PPC stuff (AltiVec), so actually there is quite a bit of it around already in lavc.
[17:12:51 CEST] <Raiz> is mister bellard here?
[17:13:59 CEST] <BBB> Raiz: no
[17:16:03 CEST] <Raiz> :( okay
[17:16:39 CEST] <Compn> dont think i've ever seen fabrice online
[17:23:15 CEST] <durandal_1707> Compn: why you haven't updated mplayer codecs.conf, because of you mplayer is really dead
[17:25:08 CEST] <haasn> https://ffmpeg.org/doxygen/trunk/pixfmt_8h.html#aff71a069509a1ad3ff54d53a1c894c85 seems to be missing a definition for ICtCp
[17:28:57 CEST] <nevcairiel> ICtCp should probably just be implemented like YCgCo as a color matrix, instead of duplicating all kinds of pixel formats
[17:29:24 CEST] <nevcairiel> oh right thats the thing you linked
[17:29:36 CEST] <nevcairiel> didnt load fast enough :D
[17:30:29 CEST] <haasn> (can HEVC actually signal ICtCp yet, and does a sample file exist?)
[17:31:14 CEST] <nevcairiel> there is no flag for ICtCp in the latest official spec
[17:31:55 CEST] <haasn> Okay
[17:32:04 CEST] <haasn> Then I'll wait on implementing until HEVC and ffmpeg support it :)
[17:32:06 CEST] <Compn> durandal_1707 : had 3 hd failures in a row. i'm limping online atm
[17:33:00 CEST] <Compn> almost lost my private ssh key lol
[17:33:10 CEST] <nevcairiel> haasn: there are various drafts adding all sorts of color  spaces, YDzDx, ICtCp and two others that I don't understand :D
[17:33:48 CEST] <haasn> nevcairiel: ICtCp is the HDR analog of BT.2020's CL system
[17:33:57 CEST] <haasn> never heard of YDzDx
[17:34:05 CEST] <nevcairiel> more specifically its dolbys idea for HDR
[17:34:11 CEST] <nevcairiel> HDR10 uses ordinary BT2020
[17:34:31 CEST] <haasn> BT.2020-NCL yeah
[17:34:34 CEST] <BBB> I bet they have a patent on it
[17:34:42 CEST] <BBB> and other than the patent theres nothing interesting to it?
[17:34:57 CEST] <nevcairiel> i have no idea if there are practical advantages
[17:35:20 CEST] <nevcairiel> unfortunately tech development is often driven by petty financial needs
[17:35:40 CEST] <BBB> hevc shows us how well thats going
[17:39:02 CEST] <haasn> nevcairiel: CL has practical advantages, and I imagine ICtCp has too
[17:39:18 CEST] <haasn> Only reason NCL caught on is was because it was backwards compatible with legacy systems and easier for people to understand
[17:39:23 CEST] <haasn> (IMO)
[17:39:51 CEST] <haasn> CL avoids the effect where subsampled chroma causes dark fringes around color transitions
[17:41:47 CEST] <nevcairiel> haasn: YDzDx uses the XYZ space
[17:41:53 CEST] <haasn> ah okay
[17:42:02 CEST] <haasn> yeah, and ICtCp uses LMS
[17:42:17 CEST] <haasn> both are a sort of weird departure from using RGB though. I guess calling it a BT.2020-CL analog was not entirely true
[17:42:30 CEST] <haasn> How long until some color space comes around that is based on L*ab? :^)
[17:42:35 CEST] <nevcairiel> there is also YDuDv which is based on CIE Luv
[17:42:39 CEST] <haasn> or that
[17:43:36 CEST] <haasn> somebody should compare their image qualities in a simulated scenario with a reference implementation of YCbCr, ICtCp, YCbcCrc, YDzDx and YDuDv
[17:44:07 CEST] <haasn> that somebody should not be me
[17:45:17 CEST] <atomnuker> don't omit YCoCg, only a matter of time before someone does an HDR spin off of it
[20:11:14 CEST] <BtbN> andrey_turkin, did nvidia drop yuv420p support on windows?
[20:11:25 CEST] <BtbN> Only nv12 works for me, yuv420p errors on EncodeVideo
[22:50:13 CEST] <cone-253> ffmpeg 03James Almer 07master:645489cf90b0: x86/dcadsp: optimize lfe_fir0_float_fma3 on x86_32
[23:48:57 CEST] <cone-253> ffmpeg 03Carl Eugen Hoyos 07master:420ba222c615: lavc/mjpeg_parser: Allow jpegls parsing.
[00:00:00 CEST] --- Wed Jul  6 2016


More information about the Ffmpeg-devel-irc mailing list