path='C:\Downloads\samples\FastVDO\FVDO_Freeway_720p.264'main debug: looking for demux2 module: 42 candidatesps warning: this does not look like an MPEG PS stream, continuing anywaymain debug: using demux2 module "ps"main debug: looking for a subtitle file in C:\Downloads\samples\FastVDO\main debug: `C:\Downloads\samples\FastVDO\FVDO_Freeway_720p.264' successfully openedps warning: garbage at input, trying to resync...
Works in mplayer (20050928-cvs) with -fps 25 (FPS not specified in header or invalid)
100% CPU for performance hungry H.264/AVC clips cause ffmpeg crash using svn-20051129. Tested with HD AVC sample from trailer/demo page with P4 3Ghz with ATI 9800XT.
main warning: late picture skipped (172967)main warning: late picture skipped (178256)ffmpeg error: more than 5 seconds of late video -> dropping frame (computer too slow ?)main warning: late picture skipped (230545)main warning: late picture skipped (188834)ffmpeg warning: illegal short term buffer state detected (h264@09A5D5E0)main warning: late picture skipped (4547834)
Program received signal SIGSEGV, Segmentation fault.[Switching to thread 5188.0x1e28]0x006f8f18 in put_h264_qpel16_mc00_mmx2 ( dst=0x14977a40 '\020' <repeats 200 times>..., src=0x0, stride=1952) at i386/dsputil_mmx.c:419419 i386/dsputil_mmx.c: No such file or directory. in i386/dsputil_mmx.c
elecard The tomsky AVC/AAC elecard clips (tomsk_sif.mp4 and video_d1.mp4) decode fine in 20060228 svn with ffmpeg cvs but also in 0.8.4a win32. Perhaps previous samples were "broken"?
Segfault in 100% CPU clips is caused by inability of VLC to drop frames under high load and this triggers the H.264 assertion bug/crash. ffmpeg in contrib adds resilience for this, having VLC drop frames would be even better.
That leaves the fastvdo raw h264 samples that don't work and the i8x8 decoding problem in analyse all.
Encoding with 8x8dct (which is a seperate option now in 0.8.5 svn) and analyse=none already causes the problem where decoding only shows one frame and debug complains about late picture skipped repeatedly. Tested against vlc-0.8.5-test2-win32-15180-contrib-20060330-ffmpeg-cvs-20060411.
8x8dct is not enabled by default in 0.8.5 so this problem will only affect "advanced" usage (commandline). The initial reported problem with decoding (raw) H.264 files is already fixed.
Trax: have you tested this with multiple file formats ????
Because it might be specific to a demux. In which case it is a packetization issue, causing frames to be in an order ffmpeg doesn't expect.
12:14:31 < Trax> zorglub: yes postpone. it's a raw vs annexB issue according to thedj and is related to incorrect ts packetizing (so both udp/http) for avc.. should be some extra info in the bitstream (NAL etc) and that seems a bit more complicated .. or it's the decoding part that screws up with reordering.. but anyway postpone
Still a problem with 0.8.6 final as encoder and svn-0.9.0-20061227-0000 as decoder. Strangely enough introducing bframes to the encode does make some pictures appear, though with the same warning messages and partially garbled video (blocks) as if there was continous packet loss.