Michael Niedermayer
d42fc4a8ca
Use the new VLC table for the first non trailing coeff too.
...
Sadly only 5 cycles faster here on pentium dual. So maybe the
complexity is not worth it and this should be reverted ...
Originally committed as revision 16295 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
Michael Niedermayer
593af7cdda
Optimize esc removal code.
...
Originally committed as revision 16294 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
Michael Niedermayer
2d76bf391a
Indent
...
Originally committed as revision 16292 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
Michael Niedermayer
8140955d39
unified CAVLC level decoding LUT.
...
Quite a bit faster (HPCVMOLQ_BRCM_B.264 was 3% faster here)
Originally committed as revision 16291 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
Michael Niedermayer
abb27cfb24
100l, I broke H.264 again, forgot one hunk.
...
Thanks to FATE for finding it.
Originally committed as revision 16285 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
Michael Niedermayer
e08715d391
Optimize 0 0 0-3 search, 45% faster on pentium dual.
...
Originally committed as revision 16284 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
Michael Niedermayer
ec3686e889
Simplify decode_cabac_mb_ref() a little bit, 2 cpu cycles faster on
...
pentium dual.
Originally committed as revision 16279 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
Michael Niedermayer
26695973c7
Indent
...
Originally committed as revision 16278 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
Michael Niedermayer
b68a455313
inline decode_cabac_mb_type for I & P frames, 9 cycles faster on pentium dual.
...
Originally committed as revision 16277 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
Michael Niedermayer
1952ac3713
Negate 2 more variables, 1 cpu cycle faster on pentium dual.
...
Originally committed as revision 16276 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
Michael Niedermayer
03a035e059
Simplify if/else, no speed change
...
Originally committed as revision 16275 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
Michael Niedermayer
6f3c50f2f9
Negate a few variables, this simplifies the code and makes it 5 cycles faster
...
on pentium dual.
Originally committed as revision 16274 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
Michael Niedermayer
60c6ba7aea
Simplify ifs(), 8 cpu cycles faster on pentium dual
...
Originally committed as revision 16273 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
Michael Niedermayer
127a20e3b8
Simplify if(), 3 cpu cycles faster in pentium dual.
...
Originally committed as revision 16272 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
Diego Biurrun
a6493a8fbd
Rename libavcodec/i386/ --> libavcodec/x86/.
...
It contains optimizations that are not specific to i386 and
libavutil uses this naming scheme already.
Originally committed as revision 16270 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
Diego Biurrun
bef05f05e4
Remove a bunch of unused variables.
...
Originally committed as revision 16263 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
Michael Niedermayer
befc8fe086
Remove useless code.
...
Originally committed as revision 16253 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
Michael Niedermayer
c212fb0cb1
Only execute clear_blocks() when needed.
...
+0.3% speedup for both aladin & cathedral.
Originally committed as revision 16252 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
Michael Niedermayer
66c07ca96f
Optimize get_dct8x8_allowed().
...
30 cpu cycles faster on pentium dual.
Originally committed as revision 16248 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
Jason Garrett-Glaser
aac8b76983
H.264 loopfilter speed tweaks
...
Originally committed as revision 16240 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
Michael Niedermayer
a5805aa9d1
Fix decoding with the plain C idcts of
...
FRExt/HPCAMOLQ_BRCM_B
FRExt/HPCAQ2LQ_BRCM_B
FRExt/HPCVMOLQ_BRCM_B
Originally committed as revision 16236 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
Michael Niedermayer
a5b807a6c1
Replace /2 by >>1 in decode_cabac_mb_dqp()
...
3 cpu cycles speed up on pentium dual.
Originally committed as revision 16233 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
Michael Niedermayer
1aea5d35e5
Simplify ctx update in decode_cabac_mb_dqp().
...
no speed change
Originally committed as revision 16232 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
Michael Niedermayer
7cfca0dfd8
Simplify ctx calculation in decode_cabac_mb_dqp()
...
no speed change
Originally committed as revision 16231 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
Alexander Strange
d43696309a
Clear FF_INPUT_BUFFER_PADDING_SIZE bytes at the end of NALs in rbsp_buffer.
...
Fixes valgrind uninitialized value warnings at the end of decoding H.264
frames.
Originally committed as revision 16230 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
Jason Garrett-Glaser
712ca84c21
Move filter_luma_intra into dsputil for later addition of asm.
...
Originally committed as revision 16228 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
Jason Garrett-Glaser
b9fe706305
Simplify chroma AC in CABAC residual decoding.
...
Originally committed as revision 16227 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
Michael Niedermayer
8955b66950
Optimize ctx calculation in decode_cabac_mb_mvd(), code by dark shikari.
...
The case for 16x16 blocks becomes 10 cpu cycles faster on pentium dual,
i could not find a speed difference in the case of subblocks though.
Originally committed as revision 16226 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
Michael Niedermayer
17779f39b6
Remove unacceptable NULL pointer hack from mc code.
...
Originally committed as revision 16225 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
Michael Niedermayer
04618b98e3
Check ref values in CABAC H.264 for validity.
...
Originally committed as revision 16224 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
Michael Niedermayer
c25ac15a07
Move idct_(dc)add closer to where it is needed.
...
Originally committed as revision 16223 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
Michael Niedermayer
aebb5d6d96
indent
...
Originally committed as revision 16222 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
Michael Niedermayer
96465b90a1
Reorder ifs in chroma hl_decode_mb to avoid a duplicate transform_bypass
...
check.
14 cpu cycles speedup on Pentium Dual
Originally committed as revision 16221 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
Michael Niedermayer
6456d6d87c
s/h->cbp_table[mb_xy]/h->cbp/
...
Originally committed as revision 16220 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
Michael Niedermayer
04824298a9
Faster CAVLC decoding of trailing_ones. Based on a patch by dark shikari.
...
decode_residual is about 3.3% faster.
Originally committed as revision 16219 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
Jason Garrett-Glaser
93445d1617
Replace i by trailing_ones, part of a patch by dark shikari.
...
No speed change meassureable with START/STOP_TIMER, but this is needed
for future optimizations.
Originally committed as revision 16218 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
Michael Niedermayer
c375d87f6f
Remove if() surrounding decode_cabac_mb_type() that can never be true.
...
Originally committed as revision 16217 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
Michael Niedermayer
c325b5054f
Remove unreachable else clause, found by dark shikari.
...
Originally committed as revision 16216 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
Michael Niedermayer
dae006d7d7
Remove useless IS_8x8DCT check i forgot, spotted by dark shikari.
...
Originally committed as revision 16215 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
Michael Niedermayer
1eb960352b
Do not calculate idct_dc_add/idct_add when the variables are unused.
...
Originally committed as revision 16210 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
Michael Niedermayer
62bc966f8f
Remove redundant nnz variable.
...
Originally committed as revision 16209 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
Michael Niedermayer
0a8ca22f4e
indent
...
Originally committed as revision 16208 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
Michael Niedermayer
2fd1f0e026
Use the new idct functions (except chroma as it was slower in benchmarks)
...
cathedral +0.5% speed
aladin +0.6% speed [note aladin has been cat-ed 10 times to reduce the influence
of init time]
Speedup also verified via START/STOP_TIMER (difference was very significant
for the changed parts)
Originally committed as revision 16207 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
Michael Niedermayer
49c084a745
Skip non intra luma code when there is no coded luma.
...
0.7% speedup for the cathedral sample.
Originally committed as revision 16203 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
Michael Niedermayer
621561cdf3
Skip chroma handling when there is no coded chroma.
...
0.5% overall speedup for the cathedral sample.
Originally committed as revision 16201 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
Michael Niedermayer
4080e67c8e
Replace != 0 || check by |
...
3 cpu cycles faster
Originally committed as revision 16183 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
Michael Niedermayer
ad9ca7e720
Split filter_mb_dir() out of filter_mb().
...
1% overall decoding speed up for cathedral-beta2-400extra-crop-avc.mp4
no speed change for Aladin.mpg
Benchmarks done on Pentium dual
Originally committed as revision 16182 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
Michael Niedermayer
ac0623b23c
Fix indention, also do a little vertical alignment of changed lines.
...
Originally committed as revision 16176 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
Michael Niedermayer
6120a343aa
Factorize 3 multiplications out, code becomes 3 cpu cycles faster.
...
(not significant as thats just per MB)
Originally committed as revision 16174 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
Michael Niedermayer
1dd488e955
Move ENABLE_SMALL back to the per MB check, as otherwise gcc wont remove
...
the code.
Originally committed as revision 16173 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago