This change is due to r14590.
The AVPacket position now points to the first byte of the actual
packet data in the file. It previously pointed to the EBML element
ID preceding packet data.
Originally committed as revision 14612 to svn://svn.ffmpeg.org/ffmpeg/trunk
contains the first picture startcode that commences in the PES
packet, instead of the first access unit that commences in the
PES packet. Fix the parser to
handle that properly. This was a very long standing bug ...
The change to the seek regressions is because the mpeg ts muxer
stores too many invalid and randomized timestamps which overflow
the 4 entry buffer we use in the parser.
Originally committed as revision 13643 to svn://svn.ffmpeg.org/ffmpeg/trunk
The PSP MP4 format requires an AAC audio stream, so until
we have an AAC encoder we cannot test this format.
The existing test is broken and does not actually use the
PSP format.
Originally committed as revision 12359 to svn://svn.ffmpeg.org/ffmpeg/trunk
through lrintf(), that is gcc put the 32bit int flags in a 32bit float
which caused some to be lost ...).
I wonder why FATE did not pick this up?
Originally committed as revision 12329 to svn://svn.ffmpeg.org/ffmpeg/trunk
and the information needed to guess the duration only becomes known at a later packet.
Originally committed as revision 11963 to svn://svn.ffmpeg.org/ffmpeg/trunk
This only removes 2 bytes from MP3 and MP2 currently.
Up to 4 could be removed from MP3/MP2 though this might need a 2pass muxer.
Primitive code to remove headers from MPEG-1/2/4 is there too but for the
single file I tried it on (the one in the regression tests), it was a loss
because all video frames were >4096 byte, so that it is disabled ATM.
Originally committed as revision 11936 to svn://svn.ffmpeg.org/ffmpeg/trunk
these were wrong (in pts vs dts sense) when b frames were in use
they were also wrong if the average framerate was smaller than 1/timebase
resulting in totally wrong final bitrate
Originally committed as revision 10477 to svn://svn.ffmpeg.org/ffmpeg/trunk
flag for diagonal interpolation
the primary reason for this change is that previously MC up to 1/4 pel
matched H.264 exactly and that increases the risk of stumbling over
patents
secondly this allows 0.10 db or more quality gain by choosing a longer
filter and the filter could also be chosen optimally for each frame
though that of course would cause speed loss at the decoder and encoder
side ...
Originally committed as revision 10436 to svn://svn.ffmpeg.org/ffmpeg/trunk
perform interpolation steps in such an order that halfpel interpolation
could be done per picture
this also makes mc_block() match h.264 for the 1/4 pel cases so that the
use of the h264 functions for some cases does not introduce a fantastic mess
Originally committed as revision 10433 to svn://svn.ffmpeg.org/ffmpeg/trunk
the old 32bit code)
disable mmx/sse2 optimizations as they need a rewrite now
Originally committed as revision 10218 to svn://svn.ffmpeg.org/ffmpeg/trunk
This allows some simplifications and optimizations and should
not have any effect on quality.
Originally committed as revision 10172 to svn://svn.ffmpeg.org/ffmpeg/trunk
(correct initial timestamps which have AV_NOPTS_VALUE)
sorry, i will rm tests/seek_test ; make ; make seektest from now on
Before r9917 lavf replaced the initial dts which were unknown (AV_NOPTS_VALUE)
by values starting from -frame_duration or another guessed value.
After r9917 lavf buffers packets until it finds a known timestamp or it buffered
too much, it will then correct the previous unknown timestamps in the buffer exactly.
For the seek tests that meant that a few initial timestamps would get changed.
If no packet in the whole stream contains a timestamp then all timestamps
will be changed by first_frame_duration or so as 0 is used instead of
-frame_duration as startpoint ...
Originally committed as revision 9949 to svn://svn.ffmpeg.org/ffmpeg/trunk
