[Ffmpeg-devel-irc] ffmpeg-devel.log.20140425

[burek]

[00:54] <cone-441> ffmpeg.git 03James Almer 07master:5ac10d40fb9b: x86/mpegaudiodsp: define apply_window_mp3 as SSE
[01:03] <cone-441> ffmpeg.git 03Janne Grunau 07master:a88e1d1c598e: lavu: add CHK_OFFS as AV_CHECK_OFFSET to check struct member offsets
[01:04] <cone-441> ffmpeg.git 03Michael Niedermayer 07master:06e664366a66: Merge commit 'a88e1d1c598e641eecd5d43730211d91c82787c6'
[01:18] <cone-441> ffmpeg.git 03Janne Grunau 07master:cae8df78759c: lavr: define ResampleContext in resample.h
[01:19] <cone-441> ffmpeg.git 03Michael Niedermayer 07master:26953ed2e39d: Merge commit 'cae8df78759c2e69257f7fe58842f34c0d98a7ec'
[01:34] <cone-441> ffmpeg.git 03Janne Grunau 07master:a24a252709dd: aarch64: NEON optimized FIR audio resampling
[01:34] <cone-441> ffmpeg.git 03Michael Niedermayer 07master:e32fc9b45b11: Merge commit 'a24a252709dd38f12aa4929ce4981f87091a5113'
[01:40] <cone-441> ffmpeg.git 03Derek Buitenhuis 07master:8de77b665d2a: fate: Add fic-in-avi test
[01:41] <cone-441> ffmpeg.git 03Michael Niedermayer 07master:817627d92538: Merge commit '8de77b665d2a2f1cd560d2183fd4664298b30715'
[02:12] <cone-441> ffmpeg.git 03Carl Eugen Hoyos 07master:eeee59ba4d48: Never write 0 as maximum bitrate for asf files.
[02:12] <cone-441> ffmpeg.git 03Carl Eugen Hoyos 07master:a19bcf4ee8d9: Fix libpostproc compilation with !HAVE_6REGS.
[02:12] <cone-441> ffmpeg.git 03Carl Eugen Hoyos 07master:9cc4bc973c20: Fix vf_eq.c and vf_eq2.c compilation with !HAVE_6REGS.
[02:12] <cone-441> ffmpeg.git 03Carl Eugen Hoyos 07master:ced0d6c14d1a: Use correct msvc type specifiers for ptrdiff_t and size_t.
[02:12] <cone-441> ffmpeg.git 03Carl Eugen Hoyos 07master:18e7e21e2fb1: Enable muxing ac-3 in caf.
[02:13] <cone-441> ffmpeg.git 03Michael Niedermayer 07master:e148a5820dc5: Merge remote-tracking branch 'cehoyos/master'
[02:47] <cone-441> ffmpeg.git 03Ben Avison 07master:270cede3f377: h264: Move search code search functions into separate source files.
[02:47] <cone-441> ffmpeg.git 03Ben Avison 07master:9d8ecdd8ca6d: vc-1: Add platform-specific start code search routine to VC1DSPContext.
[02:47] <cone-441> ffmpeg.git 03Ben Avison 07master:a0d7f9ec9a10: vc-1: Optimise parser (with special attention to ARM)
[05:15] <cone-441> ffmpeg.git 03Michael Niedermayer 07master:6e5cce1cbe54: configure: allow overriding ranlib
[05:30] <cone-441> ffmpeg.git 03Y.C. Liu 07master:cebe06a0bf19: avutil/opencl: fix a segmentfault in libavutil/opencl.c
[07:07] <cone-441> ffmpeg.git 03James Almer 07master:25d5ea6d5a08: lavu: add LOCAL_ALIGNED_32
[07:07] <cone-441> ffmpeg.git 03James Almer 07master:c7b089048dbe: vp9: use LOCAL_ALIGNED_32 for left/top intra_pred pointers
[07:33] <ubitux> jamrial: you're going to do some avx2 for vp9?
[07:53] <jamrial> Wrote a couple trivial intra_pred 32x32 functions, which is how i found out the alignment problem
[07:57] <jamrial> Although even after the above patches a (top) is still unaligned for some reason
[07:57] <jamrial> l (left) is ok now
[08:00] <jamrial> dst also seems to be unaligned
[08:00] <nevcairiel> sounds like it may be inherently unaligned
[08:01] <jamrial> It works wonders with movdqu, but if that's needed then one might as well just use the xmm version
[08:02] <nevcairiel> some of the xmm versions use movu in some places because its just accessing odd offsets
[08:05] <wm4> can ffprobe hexdump packets yet?
[08:05] <jamrial> The ones i ported to avx2 don't. A couple others do, though
[08:05] <ubitux> wm4: iirc yes but i may be wrong
[08:06] <wm4> oh, -show_data
[08:06] <wm4> I made the mistake trying to read the ffprobe help output
[08:25] <jamrial> ubitux: here's what did if you're curious https://github.com/jamrial/FFmpeg/commit/400445b0ab1efc933995852354269972ac4dee44
[08:27] <jamrial> Seems to work as is, but the movdqu galore is probably far from good
[08:41] <wm4> why does ffmpeg still have libpostproc in its repo?
[08:42] <wm4> isn't there a separate repo for libpostproc now?
[08:42] <wm4> and why does the separate repo not have all features the ffmpeg libpostproc has? (at least CPU autodetection is missing)
[09:28] <j-b> good mornin
[09:30] <ubitux> michaelni: it seems my ip is blacklisted from gmx, that's the reason i can't send you an email on your address
[09:56] <ubitux> ok that's probably because i don't have a valid reverse dns... fuck that shit
[09:56] <nevcairiel> some mail servers are rather picky about that
[09:57] <ubitux> well i have an spf record, it should be enough
[13:22] <BBB> ubitux: so far I've found that it's coefficient-coding related, chroma only, only affects 32x32 luma-coded blocks (so 16x16 chroma-predicted blocks with a 8x8 transform, I think)
[13:22] <BBB> ubitux: not a race condition, so even if frame 1 completes entirely before I start frame 2, it's an issue, but indeed goes away when you use -threads 1
[13:22] <BBB> ubitux: but that's as far as I got... weird stuff
[13:45] <BBB> oh, sorry, 16x16 transform obviously
[13:45] <BBB> anyway
[13:48] <BBB> hum, looks like uveob is wrong
[13:50] <BBB> ah there it is
[13:50] <BBB> that's stupid
[13:52] <BBB> ubitux: https://github.com/rbultje/ffmpeg/commit/6d69f9f37689c999815a65a2d99999fad3a41705
[13:52] <BBB> michaelni: for you to merge also ^^
[13:53] <BBB> my git send-email isn't working so can't send to ML right now, I probably broke something when upgrading my macports
[14:12] <BBB> ah fixed
[14:12] <BBB> yay
[14:12] <ubitux> BBB: i don't get why threading is relevant; can you elaborate?
[14:13] <BBB> neither do I
[14:13] <BBB> I think what happens is that in the single-threaded case, the upper byte from the previous run is retained
[14:13] <BBB> i.e. we never write the value, but we read it
[14:13] <BBB> so we take it from the previous thread
[14:13] <BBB> where it may or may not be correct
[14:13] <ubitux> ah, ok
[14:13] <BBB> apparently it is correct here o_o
[14:14] <BBB> in the multi-threaded cases, we start from uninitialized memor
[14:14] <ubitux> does it fix -02 as well?
[14:14] <BBB> probably
[14:14] <BBB> this should affect all low-q (non-zero) tests
[14:14] <BBB> possibly all the way up to 5 or 6 or so
[14:14] <ubitux> i mean fate-vp9-00-quantizer-02
[14:14] <BBB> yeah, I know, it should fix it
[14:14] <ubitux> ok ok :)
[14:15] <BBB> works here now, yes
[14:15] <BBB> don't know if it broke before, but it probably did
[14:15] <ubitux> cool, nice
[14:15] <michaelni> BBB, should i merge 6d69f9f37689c999815a65a2d99999fad3a41705 or apply the patch from the ML ?
[14:15] <BBB> oh actually single-threaded is easier
[14:15] <BBB> because it doesn't use 2-pass decoding
[14:15] <BBB> so it takes the rob values from the previous block
[14:16] <BBB> which for low-q values is the upper bits of the previous block, and that should always be "high"
[14:16] <BBB> so that's why that works in single-threaded mode
[14:16] <BBB> michaelni: they should be identical, so up to you
[14:16] <ubitux> ok :)
[14:16] <michaelni> i already have the merge locally so ill go with that then
[14:17] <cone-743> ffmpeg.git 03Ronald S. Bultje 07master:6d69f9f37689: vp9: write uveob as 16-bit value for 16x16/32x32 transforms.
[14:17] <cone-743> ffmpeg.git 03Michael Niedermayer 07master:2f2629c8700b: Merge commit '6d69f9f37689c999815a65a2d99999fad3a41705'
[15:12] <cone-743> ffmpeg.git 03Michael Niedermayer 07master:abbcc6b26b4a: avcodec/utils: use av_malloc(z)_array()
[15:12] <cone-743> ffmpeg.git 03Michael Niedermayer 07master:681a5b8d6f35: avcodec/ttaenc: use av_malloc_array()
[15:12] <cone-743> ffmpeg.git 03Michael Niedermayer 07master:92cc6d5163cd: avcodec/mdct_template: Use av_malloc_array()
[16:54] <cone-743> ffmpeg.git 03James Almer 07master:cdac3ab59f3c: swresample: add swri_resample_double_sse2
[19:58] <cone-743> ffmpeg.git 03Don Moir 07master:62056d09b13c: avformat/avidec: skip len=0 entries from the index
[20:15] <Daemon404> does anyone have an account on the gcc bugtracker
[20:15] <Daemon404> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=60902 is fixed
[20:15] <Daemon404> but it hasnt been backported to the 4.9 branch
[20:16] <Daemon404> it might nt be backported at all to 4.9 (making 4.9 useless for ffmpeg) if someone doesnt poke it
[20:16] <Daemon404> and ask
[20:17] <Compnn> michael does
[20:17] <Daemon404> ubitux might
[20:17] <Daemon404> maybe jamrial 
[20:17] <Daemon404> or reimar
[20:17] <Compnn> whoever else commented on that gcc bug ffmpeg triggered
[20:18] <jamrial> "Target Milestone: 4.9.1"
[20:18] <jamrial> i want to believe it's going to be backported sooner or later
[20:19] <Daemon404> ok there mulder
[20:19] <ubitux> i don't have an account on that bugzilla
[20:19] <Daemon404> i only have one on llvm's
[20:19] <jamrial> neither do i
[20:20] <jamrial> cehoyos does, though
[20:20] <ubitux> carl certainly has one, and since he's often reading the irc logs he might do it
[20:20] <jamrial> he reported the duplicates of this bug
[20:20] Action: Compnn waves to carl
[20:20] <Daemon404> did someone update our trac with the link to the commit/bug?
[20:20] <ubitux> iirc yes
[20:21] <ubitux> by carl himself
[20:21] <ubitux> https://trac.ffmpeg.org/ticket/3559#comment:15
[20:34] <Compnn> "The function looks suspicious and I'm not sure if the code is valid."
[20:34] <Compnn> :D
[20:34] <michaelni> i might have an account but if so i didnt use it for many years and dont remember the pw
[20:37] Action: Compnn wonders how many ffmpeg testcases in gcc testsuite
[21:02] <cone-743> ffmpeg.git 03Aidan Skinner 07master:802385dbc2c5: mov: Write prof section of tapt tag
[21:02] <cone-743> ffmpeg.git 03Michael Niedermayer 07master:944a744bf585: Merge commit '802385dbc2c57abd76f6a00e32f3df35e9526c08'
[21:09] <llogan> hell, i can make an account, and regurgitate whatever you want if evetyone else is too lazy to do it
[21:12] <Daemon404> :P
[21:13] <Daemon404> i have too many bug tracker accounts
[21:13] <Daemon404> one-off ones
[21:22] <jamrial> there
[21:23] <jamrial> posted a comment
[21:32] <llogan> thanks, jamrial 
[21:52] <jamrial> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=60902#c33 and the reply
[21:57] <jamrial> so yeah, wasted time. but at least now i have a gcc bugzilla account in case i need to actually report something :P
[21:59] <Daemon404> jamrial, kind of a dickish response from gcc guy
[21:59] <Daemon404> but :foss:
[22:06] <cone-743> ffmpeg.git 03Miles Gould 07master:99e22b785917: mov: Emit the correct tags for clcp tracks
[22:06] <cone-743> ffmpeg.git 03Michael Niedermayer 07master:c9e0f7a080c1: Merge commit '99e22b7859177f6d3ed6121040924b337dce5497'
[22:06] <cone-743> ffmpeg.git 03Michael Niedermayer 07master:af165acefacd: avformat/movenc: dont store invalid tapt atom
[23:18] <llogan> Compnn: how many beard inches (Bi) ago was that?
[23:32] <Compnn> llogan : it was ... PRE BEARD!
[23:33] <Compnn> or 5 P.B. 
[23:47] <cone-743> ffmpeg.git 03Michael Niedermayer 07master:e0e60c921133: avcodec/dpx: extract frame rate
[00:00] --- Sat Apr 26 2014