[Ffmpeg-devel-irc] ffmpeg-devel.log.20140127

burek burek021 at gmail.com
Tue Jan 28 02:05:02 CET 2014


[00:15] <cone-907> ffmpeg.git 03Loren Merritt 07master:b7d0d10a1d54: x86inc: Speed up assembling with Yasm
[00:15] <cone-907> ffmpeg.git 03Michael Niedermayer 07master:17e7048d30e5: Merge commit 'b7d0d10a1d54073501b728dbe166a32e2b7b26f1'
[00:40] <Rodeo> cbsrobot: erm, pong
[02:22] <cone-907> ffmpeg.git 03Michael Niedermayer 07master:7cf8918b0df7: avcodec/huffyuv: update years in copyright
[02:22] <cone-907> ffmpeg.git 03Michael Niedermayer 07master:6369766f015f: avcodec/huffyuv: support gbrp9/10/12/14
[08:24] <Keestu> i m building ffmpeg in ubuntu 12.04, and  First i built yasm-1.2.0, and tring to build x264, and for which i get the error  ""No working C compiler found."".  To refer the yasm-1.2.0 library i have used --extra-cflags, and --extra-ldflags options.
[08:26] <JEEB> yasm is not a library, it's a binary that's needed
[08:26] <JEEB> or well, it could also have a library but that's not needed
[08:27] <JEEB> you only need to be able to call yasm --version and get the 1.2.0 one
[08:27] <JEEB> as for the no working c compiler thing, check config.log :P
[08:28] <Keestu> JEEB, thanks. yasm 1.1.0.2352 is installed version,  but   x264 requiers minimum of 1.2.0 of yasm. 
[08:29] <Keestu> hence wanted to refer to manually built yasm. :)
[08:29] <Keestu> i still can do make install into system path, but i don't want to do that. :)
[08:32] <JEEB> 1) configure yasm to some --prefix (for example --prefix=/home/myusername/ownapps/ ) 2) build 3) make install to copy the things to their places 4) export PATH=/your/prefix/bin:${PATH}
[08:32] <JEEB> this should lead you to be able to call the new yasm with just yasm
[08:33] <JEEB> if you don't want it to be in the PATH at all then I'm afraid you'll have to have some extra fun with that :P
[08:33] <JEEB> setting the PATH in a single terminal in my opinion is nothing bad, though
[08:34] <Keestu> JEEB,  that means x264 only expects the binary of yasm not the dev library ?
[08:34] <JEEB> yes
[08:34] <JEEB> only the actual assembler is needed
[08:35] <Keestu> I got the below error: Found yasm 1.1.0.2352
[08:35] <Keestu> Minimum version is yasm-1.2.0
[08:35] <JEEB> yes
[08:35] <JEEB> because the assembler is too old
[08:36] <Keestu> hence i used --extra-cflags, and --extra-ldflags to point to the  'prefix' location, where in i got the error no working c compiler found. 
[08:36] <JEEB> it most probably would tell you it needs libsomething if it wanted the library
[08:36] <Keestu> JEEB, true, i set  to PATH to  prefixed bin directory of yasm. 
[08:37] <Keestu> Looks like it need libyasm.a file. 
[08:37] <JEEB> no... it should not
[08:37] <JEEB> and in many cases that doesn't change the actual currently used yasm, if you do `which yasm` you should see which yasm is getting picked up :P
[08:37] <JEEB> (in that case you might have to set the PATH in bashrc or something and open a new terminal)
[08:45] <Keestu> JEEB, oops. shame on me.  it works now    i did  PATH=$PATH:<mypath> rather than PATH=<MYPATH>:PATH :). 
[08:45] <Keestu> :)
[08:46] <Keestu> thanks JEEB. 
[09:53] <kierank> "This is useful if you need only to read a video/samples buffer, without to fetch it" ?
[10:33] <Keestu> dear all, how can i map ffpmeg error code with message ?  for example return error code of  avformat_open_input,? Kindly suggest. 
[11:13] <Rodeo> cbsrobot: pong #2, if you're around
[12:40] <cbsrobot> Rodeo: still around ?
[12:40] <Rodeo> yes
[12:41] <cbsrobot> about the fdk_aac channel layout
[12:41] <cbsrobot> see: http://pastebin.com/XpNVDCJQ
[12:42] <cbsrobot> does that seem correct to you ?
[12:42] <cbsrobot> I guess 7.1 is really MODE_7_1_REAR_SURROUND
[12:42] <cbsrobot> but MODE_1_2_2_2_1 / MODE_7_1_FRONT_CENTER seem to be 7.1wide
[12:43] <Rodeo> I would say so, except for consistency
[12:43] <Rodeo> look at 5.0 and 5.1, they already use back channels, not sides
[12:43] <cbsrobot> I compared channel_layout from ffmeg with: https://github.com/mstorsjo/fdk-aac/commit/fa3eba16446cc8f2f5e2dfc20d86a49dbd37299e#diff-49dd6137ce6bd569361d4e99baca1434L541
[12:43] <j-b> don't use 7.1 :)
[12:45] <Rodeo> either is fine though, but MODE_7_1_REAR_SURROUND should definitely be 7.1, not wide or wide back
[12:45] <Rodeo> also, did you see my comment about MODE_1_2_2_2_1 vs. MODE_7_1_FRONT_CENTER?
[12:49] <cbsrobot> sure, but how to deal with it ?
[12:51] <Rodeo> make a choice
[12:52] <Rodeo> i.e. do you want to use the official AAC channel config 7, or do you want channel_config = 0 and write a PCE on purpose
[12:53] <nevcairiel> the official 7.1 channel config is stupid, it has two sets of front speakers, instead of the more common front, surround, back model
[12:55] <cbsrobot> I have no specific requirement
[12:56] <cbsrobot> I don't own a 7.1 setup - I just needed to test some 7.1 files with a soundcard
[12:57] <Rodeo> nevcairiel: that's not the point, we're talking about only one layout
[12:57] <Rodeo> but two ways to indicate it, one using a channel_config defined in the spec., vs using a PCE (can't remember what it means just now, but basically it describes the layout instead of using a predefined config)
[12:58] <Rodeo> Program Config Element, maybe
[13:08] <cbsrobot> so a new private option could be added instead of a new channel_layout
[13:19] <Rodeo> not sure how that would work, TBH
[13:26] <cbsrobot> where can I find information about pce ?
[13:34] <av500> ask peloverde
[13:36] <cone-319> ffmpeg.git 03Lukasz Marek 07master:9d087ab5ef15: ffplay: remove redundant prototype
[14:01] <Rodeo> cbsrobot: AAC spec would me my first guess
[14:28] <cone-319> ffmpeg.git 03Rainer Hochecker 07master:bceeccc648ba: dxva2: bump maximum number of slieces for mpeg2
[15:49] <cone-319> ffmpeg.git 03Martin Storsjö 07release/1.2:89c917fcd92c: rtpdec_asf: Copy the need_parsing field from the chained demuxer
[15:49] <cone-319> ffmpeg.git 03Martin Storsjö 07release/2.1:d0e0329e9d3d: rtpdec_asf: Copy the need_parsing field from the chained demuxer
[16:20] <cone-319> ffmpeg.git 03Carl Eugen Hoyos 07master:05e5bb6107b9: Fix decoding of some 8 < bpc < 16 signed j2k samples with libopenjpeg.
[19:10] <kurosu_> BBB / ubitux: regarding vp9 loop filter: I guess there are different filters of different lengths. What are these lengths ?
[19:11] <ubitux> 2,4,6,14
[19:13] <kurosu_> ubitux: 14 !? that must be a pain... Did you have to implement some kind of transpose or is it limited to 8xN blocks ?
[19:14] <ubitux> there is a 16x16 transpose
[19:15] <ubitux> filter6() and filter14() are actually pretty fun to write
[19:15] <kurosu_> ubitux: from vp9dsp.c it seems the filter size is decided per border, right? if such is the case, do you think it would be possible to use a transpose smaller than 16x16? already done maybe?
[19:15] <ubitux> filter2/4 are same as vp8 and honestly that was fucking painful because of the weird signess all over
[19:15] <kurosu_> I guess you are calling twice this function for 32xN partitions
[19:15] <ubitux> i don't really know about the caller
[19:16] <kurosu_> actually the dsp function works on chunk of 8 pixels, so I do have my reply
[19:18] <ubitux> mmh i think i'll push the simplifications before adding 88...
[19:19] <kurosu_> BBB: do you know the gain of this 14taps filter over a 6-long one ?
[19:19] <ubitux> kurosu_: https://github.com/ubitux/FFmpeg/blob/vp9-simd/libavcodec/x86/vp9lpf.asm#L591
[19:19] <ubitux> filter14() ^
[19:19] <ubitux> it's basically just the FILTER_{INIT,UPDATE}
[19:20] <ubitux> the tricky part is the re-use of registers
[19:20] <kurosu_> ubitux: do you know how long your asm implementation of filter14 and filter6 are taking respectively ?
[19:20] <ubitux> mmh
[19:22] <ubitux> full block is ~3000 decicycles, without filter14 it's about ~1900 decicycles
[19:22] <ubitux> feel free to bench :p
[19:22] <kurosu_> I'm just trying to figure the complexity vs gain tradeoff here, which might not be the best
[19:23] <kurosu_> ubitux: it's been so long since my last pull, don't think I will any time soon
[20:30] <llogan> beastd: can you take a look at https://trac.ffmpeg.org/ticket/3352
[20:31] <beastd> llogan: ah. i am having a look.
[20:38] <llogan> i wish roger would proof read first...
[22:04] <ubitux> BBB: ping
[22:05] <ubitux> BBB: so i did the transpose differently: https://github.com/ubitux/FFmpeg/commit/4d126ce878b1a33455de9b38397473168bc778fc
[22:06] <ubitux> it's basically a 8x8 transpose, but it's working on full width (16 instead of 8)
[22:06] <ubitux> i don't have less instructions though
[22:06] <ubitux> what did i miss? :p
[22:11] <Skyler_> that looks like a nice solution
[22:12] <ubitux> it doesn't look better though
[22:12] <ubitux> it's probably slower actually but i didn't bench
[22:15] <ubitux> well i guess i'll start pushing the beginning of the patchset meanwhile
[22:15] <llogan> now everyone can watch me flail around with regex shenanigans
[22:16] Action: llogan starts playing "Yackety Sax"
[22:17] <ubitux> llogan: https://pbs.twimg.com/media/Be72GYUCYAAM7Vq.png:large ?
[22:19] <llogan> ubitux: one clue is simply ".*"
[22:20] <llogan> you'll make a soduku player cry
[22:22] <nevcairiel> there is way too many stars in that thing!
[22:43] <cone-319> ffmpeg.git 03Clément BSsch 07master:5d144086cc55: x86/vp9lpf: faster P7..Q7 accesses.
[22:43] <cone-319> ffmpeg.git 03Clément BSsch 07master:315b4775adf8: x86/vp9lpf: refactor v/h using common macros for P7 to Q7.
[22:43] <cone-319> ffmpeg.git 03Clément BSsch 07master:822385d77598: x86/vp9lpf: add a preload system in FILTER_UPDATE.
[23:06] <ubitux> heh, strangely it's a bit faster
[23:06] <ubitux> i guess i'll go for this
[23:07] <ubitux> ah my bad, it's slower, switched the 2 results
[23:08] <ubitux> (2738 -> 2808)
[23:13] <ubitux> BBB: i'll push https://github.com/ubitux/FFmpeg/commit/699b25ce73514da89c0b168347eeecf5d38aea0d soon
[23:14] <ubitux> https://github.com/ubitux/FFmpeg/commit/c156d99c16fa1b39f8c88ba038bbd14df16cf6ac make it a bit slower 
[23:44] <cone-319> ffmpeg.git 03Michael Niedermayer 07master:a38842120add: avcodec/libfdk-aacenc: change MODE_7_1_REAR_SURROUND to map to AV_CH_LAYOUT_7POINT1
[23:44] <cone-319> ffmpeg.git 03Michael Niedermayer 07master:673ce8e46ae2: avcodec/libfdk-aacenc: change MODE_7_1_FRONT_CENTER to map to AV_CH_LAYOUT_7POINT1_WIDE_BACK
[00:00] --- Tue Jan 28 2014


More information about the Ffmpeg-devel-irc mailing list