A pointer should never be assigned a value which may have less than the
required alignment of the target type. Compilers may assume pointer
values have the required alignment, and emit normal load/store instructions.
Unaligned pointers should use a character type or compiler-specific
type modifiers.
Originally committed as revision 19318 to svn://svn.ffmpeg.org/ffmpeg/trunk
Most muxers in ffmpeg ignore the SEI if it is placed in extradata, so instead
it has to be catted to the front of the first video frame.
Originally committed as revision 19310 to svn://svn.ffmpeg.org/ffmpeg/trunk
Too often it ends up decoding random data into noise without detecting
it (for example after seeking of some MP3 data with oddly often occurring
startcode emulation).
Fixes issue1154.
Originally committed as revision 19302 to svn://svn.ffmpeg.org/ffmpeg/trunk
First, reverted one was r19239.
Patch by Haruhiko Yamagata, h D yamagata A nifty D com
Originally committed as revision 19258 to svn://svn.ffmpeg.org/ffmpeg/trunk
next_frame_offset are not set, because second field has next frame offset set
but not frame_offset, otherwise this wrongly fetch timestamp for the
second field.
Originally committed as revision 19244 to svn://svn.ffmpeg.org/ffmpeg/trunk
- remove unneeded vlc code < 0 check
- reorder vlc code handling so that the unlikely escape decoding
part comes last
- move overflow check out of the decode loop
- branchless sign conversion
Originally committed as revision 19234 to svn://svn.ffmpeg.org/ffmpeg/trunk
and change default type to float so that the run level
decoding functionality can be shared with wmapro
Originally committed as revision 19231 to svn://svn.ffmpeg.org/ffmpeg/trunk
Based on code by DivX, Inc., heavily changed by Björn Axelsson [gecko acc.umu.se] and me.
Originally committed as revision 19222 to svn://svn.ffmpeg.org/ffmpeg/trunk
These are only supposed to be called once per row, not once per macroblock.
~1.5% faster according to oprofile.
Originally committed as revision 19213 to svn://svn.ffmpeg.org/ffmpeg/trunk
The threads' contexts and rbsp_buffers were not freed at the end
of decoding.
Fixes issue 1581
Originally committed as revision 19207 to svn://svn.ffmpeg.org/ffmpeg/trunk
~8% faster VC-1 decoding.
Possible future optimization: clear blocks after use instead of before, and for
DC-only blocks, only clear the DC coefficient.
Originally committed as revision 19205 to svn://svn.ffmpeg.org/ffmpeg/trunk
Includes mmx2 asm for the various functions.
Note that the actual idct still does not have an x86 SIMD implemtation.
For wmv3 files using regular idct, the decoder just falls back to simple_idct,
since simple_idct_dc doesn't exist (yet).
Originally committed as revision 19204 to svn://svn.ffmpeg.org/ffmpeg/trunk
Baptiste Coudurier
Michael Niedermayer
Måns Rullgård
Kostya Shishkov
Jason Garrett-Glaser
Reimar Döffinger
Stefan Gehrer
Ronald S. Bultje
David Conrad
Carl Eugen Hoyos
Vladimir Voroshilov
Stefano Sabatini
Haruhiko Yamagata
Brian Crowell
Yoshihisa Uchida
Sascha Sommer
Alexander Strange