[Ffmpeg-devel-irc] ffmpeg-devel.log.20131226

[burek]

[00:13] <cone-705> ffmpeg.git 03Michael Niedermayer 07master:e33b6ccfa782: avformat/mpc8: clear buffer padding area
[00:13] <cone-705> ffmpeg.git 03Michael Niedermayer 07master:26ce266e3df8: avformat/mpc8: check avio_read() return in mpc8_parse_seektable()
[00:21] <cone-705> ffmpeg.git 03Diego Biurrun 07master:bd0fba87972b: configure: Explicitly disable w32threads if the test for it fails
[00:21] <cone-705> ffmpeg.git 03Michael Niedermayer 07master:f371a4a57d48: Merge remote-tracking branch 'qatar/master'
[02:00] <cone-705> ffmpeg.git 03Michael Niedermayer 07master:1486ed0815a1: avcodec/shorten: clear bitstream buffer
[04:36] <cone-705> ffmpeg.git 03Michael Niedermayer 07master:4e394a98f2ea: avformat/rmdec: check against mismatching int4 interleaver parameters which would leave uninitialized holes
[04:36] <cone-705> ffmpeg.git 03Michael Niedermayer 07master:165f96cd2d68: avformat/rmdec: move packet allocation down
[07:54] <ubitux> any simd trick to >> 3 signed bytes?
[07:58] <ubitux> 3 avg per zero?
[11:47] <cone-149> ffmpeg.git 03Stefano Sabatini 07master:e2b54464c6a9: lavu/opt: fix range check logic in set_format()
[11:47] <cone-149> ffmpeg.git 03Stefano Sabatini 07master:1575a96b3af5: lavu/opt: factorize setting of format values from string
[11:47] <cone-149> ffmpeg.git 03Stefano Sabatini 07master:55f046be1193: lavu/opt: apply range checks also when setting format string value
[11:47] <cone-149> ffmpeg.git 03Stefano Sabatini 07master:3b8c7da7a36d: lavu,lavfi,lavd: do not hardcode AV_PIX_FMT_NB value when setting pixel format max value
[11:47] <cone-149> ffmpeg.git 03Stefano Sabatini 07master:334e2e236383: lavu,lavc,lswr: do not hardcode AV_SAMPLE_FMT_NB value when setting sample format max value
[11:47] <cone-149> ffmpeg.git 03Stefano Sabatini 07master:cd355d4d5966: lavfi/abuffersrc: use AV_OPT_TYPE_SAMPLE_FMT for sample_fmt option
[13:36] <cone-149> ffmpeg.git 03Luca Barbato 07master:1716b4c7b888: mms: Remove non-utf8 characters
[13:36] <cone-149> ffmpeg.git 03Michael Niedermayer 07master:6b36f577a12a: Merge remote-tracking branch 'qatar/master'
[13:49] <cone-149> ffmpeg.git 03Clément BSsch 07master:ec73bd1fe876: avfilter/avectorscope: fix {} mistake in alloc check.
[13:52] <ubitux> avectorscope is really beautiful&
[14:01] <cone-149> ffmpeg.git 03Michael Niedermayer 07master:14bec7dcf829: avcodec/wnv1: clear padding area of rbuf
[16:20] <cone-149> ffmpeg.git 03Michael Niedermayer 07master:d164ad3298c1: avcodec/ivi_common: use av_mallocz() to allocate mbs array
[16:20] <cone-149> ffmpeg.git 03Michael Niedermayer 07master:635987287726: avformat/oggparseogm: check input size before reading parameters
[16:22] <ubitux> is there anyone from ffmpeg going to fosdem this year?
[16:22] <ubitux> i mean, is the project represented?
[19:34] <Daemon404> hmm cool
[19:34] <Daemon404> filter-volume is crashing on icl
[19:35] <nevcairiel> its icl
[19:35] <nevcairiel> color me surprised
[19:35] <Daemon404> nevcairiel: miscompilation?
[19:35] <Daemon404> i can maybe move to icl14 and see
[19:35] Action: Daemon404 has it too
[19:36] <nevcairiel> over-optimization is the term i would use
[19:36] <nevcairiel> but sure
[19:41] <saste> Daemon404, can you provide a gdb backtrace
[19:42] <Daemon404> uh
[19:42] <Daemon404> icl != gdb
[19:42] <nevcairiel> can you provide a msvc backtrace?
[19:42] <nevcairiel> :D
[19:42] <Daemon404> ;)
[19:42] <Daemon404> its a miscompilation im sure, so it will be asm
[19:44] <Daemon404> i will poke it in a bit
[19:44] <Daemon404> too busy playying snes.
[21:42] <ubitux> BBB: any trick for a simd signed >> 3? (and >>4)
[21:42] <ubitux> byte-wise
[21:42] <ubitux> doing with avg is going to be painful
[21:42] <ubitux> (or maybe not?)
[21:45] <iive> ubitux: is there signed byte multiply that returns only the high part?
[21:46] <iive> the high part would be like >>8, so you do <<5 by using a constant.
[21:47] <ubitux> mul & shift are word and larger
[21:48] <iive> not useful then. are you doing sse2 or 3,4...
[21:49] <ubitux> no constraint so far
[21:49] <ubitux> sse2 & 3 are fine, dunno if i need 4
[21:49] <ubitux> i might not even need 3
[21:51] <ubitux> right now i was doing the pavgb(pavgb(pavgb(x+1)+1)+1)
[21:51] <ubitux> but i realized it will only work for unsigned
[21:52] <ubitux> i was considering doing the unpack, doing it word wise, and the repack
[21:52] <ubitux> but hey, i'm short in registers :(
[21:53] <iive> so, no register to keep the sign?
[21:53] <ubitux> well i guess i can do it
[21:53] <ubitux> ideally i'd like to use only 2 registers
[21:54] <ubitux> which was enough for the pavgb thing
[21:54] <ubitux> i guess i can go up to 3 though
[21:54] <iive> yes, cmp for the sign, shift >>3 for the data shift <<5 for the sign and or.
[21:55] <iive> should be faster than doing 3 interdependent instructions.
[21:55] <ubitux> well i'll see later
[21:55] Action: ubitux &
[21:56] <Daemon404> fg
[22:00] <michaelni> ubitux, add 128 to each byte, do a shift, mask and subtract 16 from each byte
[22:32] <iive> that was the first thing i thought, but that would need 2 different constants.
[22:33] <iive> i also wasn't sure if there might be small difference in rounding.
[00:00] --- Fri Dec 27 2013