[Ffmpeg-devel-irc] ffmpeg-devel.log.20140115

burek burek021 at gmail.com
Thu Jan 16 02:05:02 CET 2014


[01:10] <BBB> ubitux: very nice, lgtm
[01:39] <cone-768> ffmpeg.git 03Michael Niedermayer 07master:785dc146544a: avcodec/adpcm: fix sample count for stereo SBPRO3
[03:04] <BBB> ubitux: iadst16 simd on my github, working now
[03:04] <BBB> ubitux: I'll do some measurements and then submit a patch
[04:16] <BBB> and email sent
[04:57] <cone-722> ffmpeg.git 03James Almer 07master:8bfa5f7fab3d: webp: fix capitalization of the word Exif
[04:57] <cone-722> ffmpeg.git 03Michael Niedermayer 07master:ab296c7a9fe5: avcodec/hevc_ps: Override max_dec_pic_buffering when its inconsistent with num_reorder_pics
[05:16] <cone-722> ffmpeg.git 03Janne Grunau 07master:8b4119187b62: cabac: remove leftovers from the cabac encoder
[05:16] <cone-722> ffmpeg.git 03Michael Niedermayer 07master:2df53e0f7cea: Merge commit '8b4119187b62d6932e07aded11d33d3b24e1b42f'
[05:22] <cone-722> ffmpeg.git 03Janne Grunau 07master:032d4da8af35: aarch64: configure: set arch dependent features
[05:22] <cone-722> ffmpeg.git 03Michael Niedermayer 07master:9d2286761329: Merge commit '032d4da8af353e32ab5c86ee3a210f6065921a4a'
[05:31] <cone-722> ffmpeg.git 03Janne Grunau 07master:d0cd2a8c461f: aarch64: bswap inline assembly
[05:31] <cone-722> ffmpeg.git 03Michael Niedermayer 07master:53e6977c07a8: Merge remote-tracking branch 'qatar/master'
[13:46] <ubitux> BBB: should i wait for your mc and iadst funcs before adding avx?
[13:46] <BBB> whichever goes in first
[13:46] <BBB> if yours goes in first, I'll simply add INIT_XMM avx to my patch
[13:46] <BBB> else you add it to yours
[13:46] <ubitux> well i can push my little patch right now
[13:46] <BBB> go for it
[13:46] <BBB> I'll amend mind
[13:46] <ubitux> i didn't add anything for mc though
[13:47] <BBB> mc won't be helped much
[13:47] <BBB> it doesn't use any avx instructions
[13:47] <BBB> for mc, we should probably just use axv2 for 16/32 wide blocks
[13:47] <ubitux> ok
[13:47] <BBB> and then ignore avx
[13:48] <ubitux> i'll push in a moment then
[13:48] <BBB> ok
[13:48] <BBB> feel free to review mine after that, I'll rebase and add the relevant INIT statements
[13:49] <BBB> and you said you'll do lf_8 right?
[13:49] <BBB> it's very much on top of the "my performance sucks" list right now
[13:53] <cone-432> ffmpeg.git 03Clément BSsch 07master:8b4190da9382: vp9/x86: add AVX for itxfm and lpf.
[13:55] <ubitux> BBB: the patch seems not rebased on the recent itxfm changes
[13:56] <ubitux> ah i'm looking at the wrong change..
[13:57] <ubitux> mmh no it was correct
[13:59] <ubitux> lea  sstride3q, [sstrideq*3]
[13:59] <ubitux> oh?
[14:00] <ubitux> you can do that? :o
[14:06] <michaelni> with the right assembler you can, but they turn it into [reg+2*reg]
[14:08] <ubitux> ooh
[14:08] <ubitux> that's the trick :D
[14:08] <nevcairiel> yasm is smart like that
[14:11] <cone-432> ffmpeg.git 03Martin Storsjö 07master:d51f09962d5b: mov: Free intermediate arrays in the normal cleanup function
[14:11] <cone-432> ffmpeg.git 03Michael Niedermayer 07master:04cc753d0744: Merge commit 'd51f09962d5b4bc999fb70c040f330dd1873212e'
[14:13] <BBB> ubitux: yeah yasm is cool like that :)
[14:13] <BBB> ubitux: you can also do *5 or *9 like that
[14:13] <ubitux> i didn't even though about that trick
[14:13] <ubitux> i wonder if i could use that for the vertical version of lpf
[14:15] <BBB> usually not
[14:16] <BBB> it takes an extra lea to generate the *3, so *3 is the same as lea address, add+stride*2 and then [adress+stride]
[14:16] <BBB> it works if you do it many times
[14:16] <ubitux> BBB: i do it many times :)
[14:16] <BBB> then it may help yes
[14:16] <BBB> becasue you can read groups of 4 pixel rows instead of 2
[14:19] <cone-432> ffmpeg.git 03Martin Storsjö 07master:2620df13104d: mov: Free an earlier allocated array if allocating a new one
[14:19] <cone-432> ffmpeg.git 03Michael Niedermayer 07master:fb37d03a8708: Merge commit '2620df13104ddaa136158eb6bb1195adbf9d7692'
[14:20] <ubitux> with stride, mstride, stride3 and mstride3 you can the [-8;8] range without ±7 and ±5
[14:20] <ubitux> can *do*
[14:21] <ubitux> but with an extra dst you could cover everything i suppose
[14:21] <ubitux> and with the 2 extra dst that's even better
[14:25] <ubitux> BBB: why do you interleave 3 sub       srcq, sstrideq?
[14:26] <ubitux> (instead of doing a single sub srcq, sstride3q)
[14:28] <ubitux> also, srq4q is just the original srcq+sstrideq
[14:29] <ubitux> no idea if that makes any difference to not have the *4
[14:30] <BBB> what code where?
[14:30] <ubitux> gonna send a mail
[14:30] <ubitux> p
[14:30] <ubitux> just a min
[14:33] <ubitux> BBB: mail sent
[14:33] <BBB> ty
[14:33] <BBB> I'll read it at work
[14:33] <BBB> gotta go now
[14:33] <BBB> bye
[14:33] <ubitux> cya
[14:34] <BBB> oh that patch was merged a while ago wasn't it?
[14:35] <BBB> anyway will check sure, if it's faster, feel free to merge it
[14:35] <BBB> bye
[14:36] <ubitux> ah, i guess i'm lost in the wrong mails
[14:36] <ubitux> :D
[14:36] <ubitux> i must have archived the wrong mails, silly me
[14:41] <cone-432> ffmpeg.git 03Janne Grunau 07master:1e9265cd8f08: dct-test: test ff_simple_idct_neon only on ARM for now
[14:41] <cone-432> ffmpeg.git 03Michael Niedermayer 07master:d01f8790d93a: Merge commit '1e9265cd8f0821acbeca1db437be1361a3976b85'
[14:59] <cone-432> ffmpeg.git 03Janne Grunau 07master:b7b17ed66e19: aarch64: add cpuflags support for NEON and VFP
[14:59] <cone-432> ffmpeg.git 03Michael Niedermayer 07master:d5560f1fecde: Merge commit 'b7b17ed66e199afc7246e642bf3b35c3f8eca217'
[15:06] <cone-432> ffmpeg.git 03Janne Grunau 07master:71617884a2a6: aarch64: h264 chroma motion compensation NEON optimizations
[15:06] <cone-432> ffmpeg.git 03Michael Niedermayer 07master:6f001d87ff01: Merge commit '71617884a2a673908bd5c0f73d4f91fdca3da82a'
[15:13] <cone-432> ffmpeg.git 03Janne Grunau 07master:8438b3f09f6b: aarch64: h264 idct NEON assembler optimizations
[15:13] <cone-432> ffmpeg.git 03Michael Niedermayer 07master:fb1c786a9dc6: Merge commit '8438b3f09f6b225d0886cc385117c38eb44ca0c1'
[15:19] <cone-432> ffmpeg.git 03Janne Grunau 07master:d5dd8c7bf0f0: aarch64: h264 qpel NEON optimizations
[15:19] <cone-432> ffmpeg.git 03Michael Niedermayer 07master:19fc3c0122eb: Merge commit 'd5dd8c7bf0f0d77c581db3236e0d938f06fd5591'
[15:24] <ubitux> BBB: just replied to the correct mail
[15:24] <ubitux> (hopefully)
[15:26] <cone-432> ffmpeg.git 03Janne Grunau 07master:c65d67ef5024: aarch64: hpeldsp NEON optimizations
[15:26] <cone-432> ffmpeg.git 03Michael Niedermayer 07master:28207bfafb90: Merge commit 'c65d67ef50241886b9749e88fd333c52090f70c7'
[15:31] <ubitux> BBB: anyway, for the 8x8, should i write mmx code?
[15:36] <cone-432> ffmpeg.git 03Janne Grunau 07master:36e3b1f2fd26: aarch64: h264 loop filter NEON optimizations
[15:36] <cone-432> ffmpeg.git 03Michael Niedermayer 07master:bf0470a5be1c: Merge commit '36e3b1f2fd262028834a9d7b1eb533c1218ee6c2'
[15:49] <cone-432> ffmpeg.git 03Janne Grunau 07master:f896bca03fc6: aarch64: h264 (bi)weight NEON optimizations
[15:49] <cone-432> ffmpeg.git 03Michael Niedermayer 07master:92d07ea4b592: Merge commit 'f896bca03fc63b93851c1c14c9321c20b3cd44a6'
[15:54] <cone-432> ffmpeg.git 03Janne Grunau 07master:fe96769bed82: aarch64: port neon clobber test from arm
[15:54] <cone-432> ffmpeg.git 03Michael Niedermayer 07master:949adce12516: Merge remote-tracking branch 'qatar/master'
[18:07] <cone-668> ffmpeg.git 03Michael Niedermayer 07master:fd334b9b6e89: avformat/mxfdec: fix typo in error check of mxf_edit_unit_absolute_offset()
[18:35] <cone-668> ffmpeg.git 03Michael Niedermayer 07release/2.1:eda6effcabcf: Update for 2.1.3
[20:00] <cone-668> ffmpeg.git 03Michael Niedermayer 07fatal: ambiguous argument 'refs/tags/n2.1.3': unknown revision or path not in the working tree.
[20:00] <cone-668> Use '--' to separate paths from revisions
[20:00] <cone-668> refs/tags/n2.1.3:HEAD: avformat/mxfdec: fix typo in error check of mxf_edit_unit_absolute_offset()
[20:30] <cone-668> ffmpeg.git 03Diego Biurrun 07master:dc4d726bab59: build: Add separate hidden config option for the intrax8 code
[20:30] <cone-668> ffmpeg.git 03Michael Niedermayer 07master:221dc52d03e1: Merge commit 'dc4d726bab59f349d366916ba01df16289e5c81f'
[21:17] <cone-668> ffmpeg.git 03Diego Biurrun 07master:aab40bbfd5a7: x86: dsputil: Simplify xvmc deprecation conditional
[21:17] <cone-668> ffmpeg.git 03Michael Niedermayer 07master:6391dec82a97: Merge remote-tracking branch 'qatar/master'
[21:23] <llogan> michaelni: for twitter - "FFmpeg 2.1.3 is now available! This fixes a libxavs compilation issue." does that LGTyou?
[21:23] <michaelni> llogan, yes assuming ive not added a new bug ;)
[21:23] <llogan> we'll fix it in 2.1.4
[21:36] <cone-668> ffmpeg.git 03James Almer 07master:b90c71cc59ed: fate: add WebP Exif metadata test
[21:53] <llogan> fateserver...perl. maybe i won't add the header.
[22:08] <funman> http://streams.videolan.org/issues/raw-h264/test.h264
[22:08] <funman> I can play it with ffplay or mplayer
[22:08] <funman> using lavf from VLC however, av_probe_input_format fails
[22:29] <michaelni> funman, vlc calls av_probe_input_format3() with 2261 bytes, ffprobe calls it with 2048 and then re-calls it with more and more until it succeeds at 1mb
[22:30] <funman> love that 2048 + 213
[22:31] <funman> https://mailman.videolan.org/pipermail/vlc-devel/2009-September/066721.html
[22:31] <nevcairiel> that sure is a hack
[22:33] <j-b> no shit?
[22:35] <ubitux> BBB: http://b.pkh.me/0001-vp9-x86-simplify-a-few-mc-inits.patch
[22:36] <ubitux> i guess i have to do the benchmark?
[22:36] <ubitux> i doubt it will make anything visible though
[22:43] <ubitux> meh i'm too lazy to add the timings
[22:58] <jedir0x> i'm having a problem with avlib where when creating an mp4 with h264 (no audio stream) the file is corrupt, but there were no errors or problems during encoding (just drawing bitmaps to the AVPicture) - mplayer (and other forgiving players) play it just fine - but it wont play on OSX or on my Android device.  Any ideas on how to go about debugging this?
[23:31] <cone-668> ffmpeg.git 03Michael Niedermayer 07master:1a01dc83434f: avcodec/pgssubdec: Use av_fast_padded_malloc() for rle buffer
[23:31] <cone-668> ffmpeg.git 03Michael Niedermayer 07master:b948ab8132e2: avformat/jvdec: clear packet padding after downsize
[00:00] --- Thu Jan 16 2014


More information about the Ffmpeg-devel-irc mailing list