[Ffmpeg-devel-irc] ffmpeg-devel.log.20121117

[burek]

[00:01] Action: ubitux is adding a runtime cpu detect fate instance
[00:17] <ubitux> yepee pp honoring cpu flags
[00:42] <ubitux> michaelni: crash in sws: ./ffmpeg -i tests/lena.pnm -vf format=yuv444p -cpuflags sse2 -f null -
[00:42] <ubitux> do you want me to open a trac?
[00:42] <ubitux> +issue
[01:13] <saste> uhm... ffmpeg abhorret vacuum
[01:14] <saste> it always destroys initial offset in input
[01:14] <saste> i find it a bit annoying
[01:14] <saste> ffmpeg -f lavfi -i "testsrc=s=qcif:d=3, setpts=PTS+5/TB"  -y out.nut
[01:15] <saste> ^ this will remove the initial offset
[01:15] <saste> ffmpeg -f lavfi -i "testsrc=s=qcif:d=3" -vf  setpts=PTS+5/TB -y out.nut
[01:15] <saste> ^ this won't
[01:16] <michaelni> ubitux, works with -cpuflags mmx+mmx2+sse2
[01:16] <michaelni> and theres no cpu with sse2 and no mmx/mmx2
[01:17] <saste> michaelni, can you say why this is the default ffmpeg behavior?
[01:18] <michaelni> saste, i think some containers have issues with not starting from 0
[01:18] <michaelni> or close to 0
[01:18] <michaelni> or maybe better said some players do when reading such files
[01:18] <saste> michaelni, also, can you provide any hint about how should I set start_time in the avi muxer?
[01:19] <michaelni> theres a per stream field that should be set once the first packet arrived for that stream
[01:19] <saste> michaelni, ok, then I'll try to add an option to skip the offset deletion
[01:20] <michaelni> ok
[02:15] <cone-480> ffmpeg.git 03Michael Niedermayer 0731fce399425b: tm2: check for invalid vlcs, fix out of array read
[02:15] <cone-480> ffmpeg.git 03Michael Niedermayer 0765e57975b9e7: cpu: add cmov ro the table, which appears to have been forgotten
[02:15] <cone-480> ffmpeg.git 03Michael Niedermayer 077b723ed4c0e5: cpu: add a few cpu names as aliases for their cpu flags
[02:43] <cone-480> ffmpeg.git 03Michael Niedermayer 07caedd51e56e2: mxfdec: fix potential integer overflow in mxf_compute_sample_count()
[02:43] <cone-480> ffmpeg.git 03Michael Niedermayer 0784281d111c95: ff_mpeg4videodec_static_init: fix return type
[04:50] <cone-480> ffmpeg.git 03Michael Niedermayer 07774830050aff: cook: check subbands more completely, fix out of array read.
[04:50] <cone-480> ffmpeg.git 03Michael Niedermayer 078ad9b48c9b5b: xxan: check ybuf index before use.
[10:40] <ubitux> michaelni: ok, so it's expecting correct input? shouldn't we detect "invalid" cpu flags sets?
[11:12] <ubitux> Skyler_: ok so fyi, my bench was wrong, when using the START/STOP timer macro sse2 version is faster
[11:13] <ubitux> also, using movq+movhps seems slower than movlps+movhps
[11:13] <ubitux> and i see no diff with xorps/pxor
[11:16] <ubitux> at least at the beginning, maybe i should rely on the latest results instead? http://pastie.org/5391497
[11:16] <cone-131> ffmpeg.git 03Stefano Sabatini 07a0f2d3fff534: lavf/avienc: return proper error codes, and provide some more feedback
[11:27] <pengvado> you should look at the last decicycles line. it includes all of the previous results. whereas the first line is just 1 sample, and doubly wrong because its time is dominated by icache misses that don't happen in steady state.
[11:28] <ubitux> mmh ok
[11:28] <ubitux> thanks
[12:03] <cone-131> ffmpeg.git 03Nicolas George 07c551c3b18bab: configure: warn when en/disabling has no effect.
[12:03] <cone-131> ffmpeg.git 03Nicolas George 071ba2e589cc33: configure: add --fatal-warnings option.
[12:06] <Skyler_> yeah, it's about 12 cycles faster
[12:06] <Skyler_> if you have an updated link I can give more comments, ubitux 
[12:06] <ubitux> i'm going to try the 4 reg version you proposed, and i'll keep you up to date
[12:06] <ubitux> thank you :)
[12:08] <Skyler_> okies
[12:41] <ubitux> it's indeed a little faster
[12:43] <ubitux> http://pastie.org/5391774
[12:53] <cone-131> ffmpeg.git 03Nicolas George 07cb8c5e531c8d: cmdutils: free the report file name.
[13:02] <ubitux> what does the "skips" mean?
[13:03] <cone-131> ffmpeg.git 03Janne Grunau 0760b6b8c01972: h264: always check ref_count for validity
[13:03] <cone-131> ffmpeg.git 03Janne Grunau 07e6160bda9864: h264: Fix parameters to ff_er_add_slice() call
[13:03] <cone-131> ffmpeg.git 03Janne Grunau 078f5587c3d0bc: cpu.h: define AV_CPU_FLAG_MMX2 for libavutil major 52
[13:03] <cone-131> ffmpeg.git 03Michael Niedermayer 07abf91f7fee32: Merge commit '8f5587c3d0bc4b5f075e4282215bda91a21fc12e'
[13:09] <ubitux> ok, seems it's the ignored runs ok
[13:12] <cone-131> ffmpeg.git 03Janne Grunau 0780716953c227: libswscale: remove unnecessary direct #if LIBSWSCALE_VERSION_MAJOR
[13:12] <cone-131> ffmpeg.git 03Anton Khirnov 07108bfe4e43c7: FATE: add a bink version 'i' test
[13:12] <cone-131> ffmpeg.git 03Anton Khirnov 07da5408ede972: FATE: add a bink version 'b' test
[13:12] <cone-131> ffmpeg.git 03Michael Niedermayer 07130dc9518779: Merge remote-tracking branch 'qatar/master'
[13:15] <michaelni> ubitux, yes, you can ignore these normally, just in cases where a function has a very variable runtime like needing 10cycles at average and rarely 1000 cycles should you pay attention to them as in such case they may indicate incorrectly skiped cases where the function took exceptionally long
[13:16] <ubitux> ok
[13:16] <ubitux> thx
[13:32] <nevcairiel> ew, missing fate samples
[15:56] <j-b> SSE2 for pp ? nice
[18:45] <durandal_1707> why are my patches ignored?
[18:49] <michaelni> durandal_1707, which patches ?
[18:55] <durandal_1707> michaelni: there are 5 patches in queue without any comment
[19:00] <michaelni> durandal_1707, i just commented on 3, i need to reread the jpeg one, i had the feeling it could be simplified
[19:06] <durandal_1707> michaelni: thanks
[19:13] <durandal_1707> michaelni: the 3 patches you replied depends on (are imho more useful) if lxf one is approved, than I would just push 4 patches at once
[19:18] <durandal_1707> in other words adding 2 more planar decoder have more sense if lxf demuxer use them
[19:28] <cone-131> ffmpeg.git 03Michael Niedermayer 0763ac64864c6e: eamad: fix out of array accesses
[19:28] <cone-131> ffmpeg.git 03Michael Niedermayer 07f96a653184e6: flvdec: disable hack that attempts to parse aac bitstream in the flv demuxer.
[19:34] <durandal_1707> michaelni: you added dead code for coverity
[19:35] <durandal_1707> and I'm for removing such hacks
[19:36] <durandal_1707> because there should be parser anyway
[19:38] <ubitux> zimbatm: what are you doing?
[19:43] <durandal_1707> michaelni: and why lxf patch may not produce same output?
[19:45] <michaelni> durandal_1707, if lxf produces the same (audio) output before and after the patch then is ok otherwise it may be buggy
[19:46] <durandal_1707> michaelni: but I see no reason why it could
[19:47] <durandal_1707> michaelni: about mjpeg last hunks: i did this to make adding 444 support for inter more easy otherwise it could be nightmare to debug code
[19:47] <michaelni> ok
[19:56] <cone-131> ffmpeg.git 03Nicolas George 074cd74c814359: lavu/bprint: implement av_bprint_strftime().
[19:56] <cone-131> ffmpeg.git 03Nicolas George 0729e388abcbc4: lavfi/drawtext: use bprint for the expanded text.
[19:56] <cone-131> ffmpeg.git 03Nicolas George 07ee2d6fbc92e6: doc/texi2pod: recognize @table @command.
[20:01] <durandal_1707> michaelni: there is no coverage for lxf demuxer
[20:14] <michaelni> it would be good to add a test ...
[20:15] <michaelni> for lxf that is ...
[20:15] <durandal_1707> michaelni: i'm short on time these days, feel free to beat me
[20:33] <cone-131> ffmpeg.git 03Paul B Mahol 07467dfd5dfae7: PCM signed 24-bit/32-bit little-endian planar decoder
[20:33] <cone-131> ffmpeg.git 03Paul B Mahol 0737e2a9783fd0: lxfdec: remove deplanarization hack
[20:33] <cone-131> ffmpeg.git 03Paul B Mahol 079ba41ae63e07: PCM signed 16-bit big-endian planar decoder
[20:33] <cone-131> ffmpeg.git 03Paul B Mahol 07461ecea06806: AST demuxer
[20:50] <cone-131> ffmpeg.git 03Michael Niedermayer 07fb1ea777b3a0: electronicarts: check size before reading duration out of a chunk.
[20:50] <cone-131> ffmpeg.git 03Michael Niedermayer 07845724c82c1a: vcr1: check if dimensions are supported, fix out of array accesses.
[21:04] <cone-131> ffmpeg.git 03Paul B Mahol 07d98364edcedb: smush: check audio packet size
[21:31] <cone-131> ffmpeg.git 03Paul B Mahol 07e3edee6d2fcf: eamad: return meaningful error codes
[22:23] <juanmabc> durandal_1707: just to be in touch, how it is going that adpcm_ima_amv and others ? patched yet?
[00:00] --- Sun Nov 18 2012