floating point.
Should fix different results between PPC and x86 for the idroq-video-encode
FATE test.
Originally committed as revision 21745 to svn://svn.ffmpeg.org/ffmpeg/trunk
start and to avoid initializing codecs with missing headers is set for all streams.
Fixes issue 1723.
Originally committed as revision 21693 to svn://svn.ffmpeg.org/ffmpeg/trunk
which has a up to 200% overhead.
Also allows to use memcpy instead of strcpy.
Originally committed as revision 21560 to svn://svn.ffmpeg.org/ffmpeg/trunk
This allows to get rid of the macho64 specific hack that moves them
to rodata (with worse cache behaviour) and avoids textrels which
e.g. Gentoo does not allow for x86_64 libraries.
Originally committed as revision 21551 to svn://svn.ffmpeg.org/ffmpeg/trunk
Avoids a (incorrect) "possibly used uninitialized" warning, no measurable
speed difference.
Originally committed as revision 21518 to svn://svn.ffmpeg.org/ffmpeg/trunk
5% faster on Intel Atom with gcc 4.4.1 (performance is unchanged
if using av_always_inline).
Originally committed as revision 21516 to svn://svn.ffmpeg.org/ffmpeg/trunk
finding the 4-byte signature.
Add a check that ignores the signature if we do not have enough previous
data to go back at least 7 bytes.
Originally committed as revision 21487 to svn://svn.ffmpeg.org/ffmpeg/trunk
Only retry 2 - 5 times in quick succession and afterwards sleep a bit
to avoid creating high CPU load without any progress.
Originally committed as revision 21427 to svn://svn.ffmpeg.org/ffmpeg/trunk
Probably only DoS, init_get_bits sets buffer to NULL, thus causing a
NULL-dereference directly after.
Originally committed as revision 21426 to svn://svn.ffmpeg.org/ffmpeg/trunk
This is consistent with other codecs and will also avoid a crash on the
memcpy to data[1] if AVPALETTE_SIZE ever increases.
Originally committed as revision 21399 to svn://svn.ffmpeg.org/ffmpeg/trunk
inv_zigzag_direct16 16-byte aligned, so mark it appropriately.
Fixes encoder crashes e.g. with MPlayer's -vf lavc.
Originally committed as revision 21389 to svn://svn.ffmpeg.org/ffmpeg/trunk
About 1% faster ff_ac3_bit_alloc_calc_psd on Intel Atom, overall speedup
not measurable though.
Should have a bigger effect on systems without cmov or with very slow cmov.
Originally committed as revision 21214 to svn://svn.ffmpeg.org/ffmpeg/trunk
cause libswscale to assume the incorrect slice order and complain about
slices starting in the middle.
Originally committed as revision 30246 to svn://svn.mplayerhq.hu/mplayer/trunk/libswscale
causes no significant speed difference and can avoid compilation issues with
--enable-pic.
Originally committed as revision 21003 to svn://svn.ffmpeg.org/ffmpeg/trunk
enough other registers available that are less likely to cause issues.
Originally committed as revision 30160 to svn://svn.mplayerhq.hu/mplayer/trunk/libswscale
reference must be 3 because both "fields" are used as reference,
and buffer_hints must include readable since we might memcpy from the old frame.
Originally committed as revision 20923 to svn://svn.ffmpeg.org/ffmpeg/trunk
This fixes playback when lavf and lavc are less tightly couples as in e.g. MPlayer.
Originally committed as revision 20626 to svn://svn.ffmpeg.org/ffmpeg/trunk
instead of either aborting on the assert or crashing due to writing beyond
the array due to insufficient padding.
Originally committed as revision 20552 to svn://svn.ffmpeg.org/ffmpeg/trunk
This in particular ensures that -DPIC is set appropriately for YASM, which avoids
linking failures on x86_64 for compilers that have PIE enabled by default
(e.g. hardened Gentoo).
It also makes the macho64 special-case hack adding -DPIC to YASMFLAGS obsolete.
Originally committed as revision 20497 to svn://svn.ffmpeg.org/ffmpeg/trunk
and -DPIC makes little sense without -fPIC or similar - if -fPIC is
enabled by default, libavutil/internal.h defines PICi anyway.
Originally committed as revision 20486 to svn://svn.ffmpeg.org/ffmpeg/trunk
This covers the previous case of x86_64 and shared, but also
works with --enable-pic.
For 32 bit x86 it currently makes no difference since x86inc.asm
ignores -DPIC for non-x86_64.
Originally committed as revision 20476 to svn://svn.ffmpeg.org/ffmpeg/trunk
necessary ff_cos_tabs tables are initialized.
Fixes issue 1507 (QDM2 broken since r20237 without hardcoded tables).
Originally committed as revision 20464 to svn://svn.ffmpeg.org/ffmpeg/trunk
While this "wastes" up to 2x32 bytes it makes the code slightly simpler and
less confusing.
Originally committed as revision 20449 to svn://svn.ffmpeg.org/ffmpeg/trunk
At least for cross-compilation with Gentoo gcc 4.4.2 it fixes the build errors
due to GPREL 22 being insufficient for the .bss size of > 4 MB without it.
Why these errors do not appear with static builds is unclear, it is not due
to PIC (shared builds with PIC disabled show the same issue).
Originally committed as revision 20409 to svn://svn.ffmpeg.org/ffmpeg/trunk
This makes the output nicely aligned even in for those cases, and might
also avoid some issues with printf implementations that can not handle well
very large float values (e.g. msvcrt without MinGW wrapper IIRC).
Originally committed as revision 20334 to svn://svn.ffmpeg.org/ffmpeg/trunk
Hopefully this fixes regression tests on DOS.
It breaks alignment of the output a bit more, but this should be easy to
fix with a second patch.
Originally committed as revision 20282 to svn://svn.ffmpeg.org/ffmpeg/trunk
work_chunks_* and dv_idct_factor_* variables appear only once in the binary
instead of 3 times.
Saves 3264 bytes in .rodata and 312416 bytes in .bss on x86_64.
Originally committed as revision 20246 to svn://svn.ffmpeg.org/ffmpeg/trunk
initialized by ff_fft_init and using different code can result in slightly
different values, in addition it crashes when the tables are hardcoded.
On amd64 this slightly changes qdm2 output.
Originally committed as revision 20237 to svn://svn.ffmpeg.org/ffmpeg/trunk
This hopefully is fast enough so that it is reasonable to use the same formula
directly instead of the table for CONFIG_SMALL.
Originally committed as revision 20233 to svn://svn.ffmpeg.org/ffmpeg/trunk
--enable-hardcoded-tables was used.
Due to the size, the code for the tables is generated at compile time.
Originally committed as revision 20232 to svn://svn.ffmpeg.org/ffmpeg/trunk
buffers, this hopefully should help detect codecs that do not release all
buffers e.g. in the decode_end function.
Originally committed as revision 20230 to svn://svn.ffmpeg.org/ffmpeg/trunk
part instead of failing completely.
This partly fixes issue 1470 (broken files created by BBC ingex recorder).
Originally committed as revision 20221 to svn://svn.ffmpeg.org/ffmpeg/trunk
This allows for some simplifications like removing some outer loops
and gives much better performance for thread_count > number of idle CPUs.
Originally committed as revision 20211 to svn://svn.ffmpeg.org/ffmpeg/trunk
processing with jobs > threads without wasting too much memory.
It also avoids needing a separate int array when the only additional data
the jobs needs is a single int running from 0 to count-1.
Originally committed as revision 20210 to svn://svn.ffmpeg.org/ffmpeg/trunk
This is not correct, but libtheora does not seem to provide a way
to get the correct value, and this is necessary to make encoding
produce sensible time stamps when encoded content is variable
FPS or the time base is simply different from FPS.
Somewhat fixes issue 1197.
Originally committed as revision 20199 to svn://svn.ffmpeg.org/ffmpeg/trunk
Reimar Döffinger
