[Ffmpeg-devel-irc] ffmpeg-devel.log.20140131

[burek]

[00:22] <ramiro> in rgb16to24(), while expanding a value from 5 to 8 bits, the last 3 bits are made up from the top 3 bits from the original value (so abcde becomes abcdeabc). why is this better than just adding zeroes (abcde000) or half (abcde100) to the end?
[00:26] <michaelni> it ensures white stays white
[00:27] <michaelni> consider 1bit -> 8bit, white would end as 50% gray, for 5->7 its less extreem
[00:29] <ramiro> michaelni: ah, that makes sense. thanks...
[00:33] <nevcairiel> its a bit tricky though if you deal with a limited range format, where 235 is perfect white in 8-bit, and 940 is perfect white in 10-bit, simply adding two zeros ensures this, and i dont think swscale cares about this difference
[00:34] <nevcairiel> (not in this particular function, but the same expansion code is used all around)
[00:49] <cone-455> ffmpeg.git 03Stefano Sabatini 07master:411185654577: doc/APIchanges: apply minor formatting fixes
[01:31] <drv> j-b: hello
[01:41] <cone-455> ffmpeg.git 03Michael Niedermayer 07master:2e02d7123788: ffmpeg: print data size too in statistic
[01:53] <llogan> michaelni: "Your work will be to add support for P frames, that is for motion compensation and estimation". the ", that is" is confusing to me. are you saying the p frames are used for motion comp and est?
[02:00] <michaelni> P frames use motion compeansation in the decoder and encoder, and motion estimation in the encoder
[02:01] <michaelni> i meant that to add support for P frames one has to add support for MC/ME
[02:01] <michaelni> feel free to reword/rewrite as you feel is best
[02:01] <michaelni> (that applied to all i wrote)
[02:02] <michaelni> no need to ask me
[02:24] <llogan> michaelni: do you have an image suggestion for ffv1 p frames?
[02:28] <michaelni> hmm, maybe a screenshot from a random video with "-vismv 7" could be used
[02:29] <llogan> LGTM. I'll add it later.
[02:29] <michaelni> llogan,  or maybe -vismv 7 -debug +vis_mb_type
[02:31] <michaelni> also all other projects should proably be copied from the 2013 page as a startpoint, but i can do that later if noone else did
[02:31] <iive> GSoC?
[02:31] <michaelni> iive, yes 
[02:32] <iive> good luck. :)
[02:33] <iive> btw, do we have proper concatenation in ffmpeg already? 
[02:36] <iive> see ya
[03:14] <cone-455> ffmpeg.git 03Michael Niedermayer 07master:c89f8f80cc83: avformat/hdsenc: check mkdir() return code
[03:34] <cone-455> ffmpeg.git 03Michael Niedermayer 07master:a85a718f26ba: doc/examples/decoding_encoding: Check all av_samples_get_buffer_size() returns
[10:05] <ubitux> https://github.com/sympy/sympy #math
[10:22] <ubitux> i'll probably try to implement in asm a dct/idct (float DCT-II/DCT-III) 8x8 and 16x16
[10:22] <ubitux> the current C implementation in dctdnoiz is extremely naive and slow
[10:23] <ubitux> it's also relying on some of our rdft utils, but actually most of it is actually in C
[10:23] <ubitux> (only ARM actually has some code for this)
[10:24] <ubitux> (http://pastie.org/pastes/8684988/text)
[10:25] <ubitux> any hint where i could look for writing a fast "simd compliant" C code for this?
[10:25] <ubitux> (without all the various slow dct layers we have curently)
[10:28] <ubitux> BBB: you mentioned libvpx last time i raised that topic, but libvpx has integer only
[10:28] <ubitux> i could probably look at making the filter working with integer though
[10:30] <ubitux> (and yes i know i must finish the misc. lpf functions :P)
[11:57] <j-b> drv: your patch does not work for me. What sample?
[12:54] <BBB> ubitux: yeah libvpx is int only, if you look in the git history you'll find float implementations (c only though) of 16x16/32x32
[14:36] <cone-488> ffmpeg.git 03Ronald S. Bultje 07master:51daafb02eaf: x86: videodsp: Properly mark sse2 instructions in emulated_edge_mc as such.
[14:36] <cone-488> ffmpeg.git 03Michael Niedermayer 07master:de17ccc774ea: Merge commit '51daafb02eaf96e0743a37ce95a7f5d02c1fa3c2'
[14:53] <cone-488> ffmpeg.git 03Ronald S. Bultje 07master:9ee9c679a7d4: x86: videodsp: Fix a bug in a %if statement where we used '%%' instead of '&&'.
[14:53] <cone-488> ffmpeg.git 03Michael Niedermayer 07master:8eaa66212c31: Merge commit '9ee9c679a7d444db223cf932e89cd39351f54f2d'
[14:58] <cone-488> ffmpeg.git 03Jan Ekström 07master:531281852448: atrac3plus: Make initialization dependant on channel count rather than channel map
[14:58] <cone-488> ffmpeg.git 03Michael Niedermayer 07master:dd4237f87d84: Merge commit '5312818524484a995433b986a2a7a6602572d4db'
[15:00] <michaelni> btw, ive increased the weight of tracs bayesian and regex spam filters
[15:01] <beastd> michaelni: yeah, so much spam :(
[15:05] <cone-488> ffmpeg.git 03Janne Grunau 07master:78987a88a88b: lavf: include 60 fps in guessed standard frame rates
[15:05] <cone-488> ffmpeg.git 03Michael Niedermayer 07master:e53ec2fe3829: Merge commit '78987a88a88b28d93d03ed6c228bcb33f178444f'
[15:24] <cone-488> ffmpeg.git 03Janne Grunau 07master:7716dd9fbbce: xvid: switch to xvid mmx idct as soon as possible
[15:24] <cone-488> ffmpeg.git 03Michael Niedermayer 07master:3f9e2aad3451: Merge commit '7716dd9fbbcea2700a100c995df967d5547c2eb9'
[15:32] <cone-488> ffmpeg.git 03Janne Grunau 07master:04ffd2203065: fate: add xvid test for custom matrices
[15:32] <cone-488> ffmpeg.git 03Michael Niedermayer 07master:aa36dcf53441: Merge remote-tracking branch 'qatar/master'
[16:41] <drv> j-b: all the samples I have work; do you have a sample that fails?
[16:41] <drv> e.g. for a small sample http://samples.mplayerhq.hu/game-formats/bink/logo_legal.bik works
[17:29] <cone-488> ffmpeg.git 03Michael Niedermayer 07master:fff2953163ff: avformat/flac_picture: allocate buffer padding for picture
[18:35] <cone-488> ffmpeg.git 03Michael Niedermayer 07master:13aa82bbbb71: avformat/flac_picture: clear padding area
[18:35] <cone-488> ffmpeg.git 03Michael Niedermayer 07master:2884688bd51a: avcodec/mjpegdec: pass into ff_mjpeg_decode_sos() and check bitmask size
[18:35] <cone-488> ffmpeg.git 03Michael Niedermayer 07master:0de9d72b7c00: mxpeg_decode_end: zero bitmask size
[19:55] <cone-488> ffmpeg.git 03Michael Niedermayer 07master:f58eab151214: avcodec/takdec: always check bits_per_raw_sample
[22:23] <J_Darnley> Does anyone know if, I think it's called this, a jump table exists in ffmpeg's x86 assembly?
[22:24] <ubitux> array of asm callbacks are handled in C code, if that's what you mean
[22:29] <J_Darnley> I want to jump to different locations in some assembly, the C has a switch-case block without breaks, so rather than a chain of if(value==1)...if(value==2) I wondered if there was an existing example.
[22:37] <ubitux> change the code to call functions
[22:37] <ubitux> and change the array in an init function for in the asm init C code
[22:38] <nevcairiel> yeah such branches are usually better kept out of the asm
[22:49] <J_Darnley> You're probably right...
[22:49] <J_Darnley> I should get it working first then try to clean up
[22:50] <cone-488> ffmpeg.git 03Michael Niedermayer 07master:19b41f86457d: avcodec/mjpegdec: use the correct linesize in the flipping code
[23:09] <cone-488> ffmpeg.git 03Ståle Kristoffersen 07master:16097efb8caa: Auto-detect hevc in mpeg-ts.
[23:09] <cone-488> ffmpeg.git 03Michael Niedermayer 07master:dce93218d7bb: Merge remote-tracking branch 'cehoyos/master'
[00:00] --- Sat Feb  1 2014