Commit Graph

172 Commits

Author SHA1 Message Date
Jason Garrett-Glaser 2a1f431d38 H.264/SVQ3: make chroma DC work the same way as luma DC
No speed improvement, but necessary for some future stuff.
Also opens up the possibility of asm chroma dc idct/dequant.

Originally committed as revision 26349 to svn://svn.ffmpeg.org/ffmpeg/trunk
2011-01-15 01:10:46 +00:00
Jason Garrett-Glaser 5657d14094 H.264: switch to x264-style tracking of luma/chroma DC NNZ
Useful so that we don't have to run the hierarchical DC iDCT if there aren't
any coefficients.  Opens up some future opportunities for optimization as well.

Originally committed as revision 26337 to svn://svn.ffmpeg.org/ffmpeg/trunk
2011-01-14 21:36:16 +00:00
Jason Garrett-Glaser 19fb234e4a H.264: split luma dc idct out and implement MMX/SSE2 versions
About 2.5x the speed.

NOTE: the way that the asm code handles large qmuls is a bit suboptimal.
If x264-style dequant was used (separate shift and qmul values), it might
be possible to get some extra speed.

Originally committed as revision 26336 to svn://svn.ffmpeg.org/ffmpeg/trunk
2011-01-14 21:34:25 +00:00
Eli Friedman 9049fa5479 Add av_unused to decode_mb_skip declaration to fix the following warning:
libavcodec/h264.h:1260: warning: ‘decode_mb_skip’ defined but not used
patch by Eli Friedman, eli.friedman gmail com

Originally committed as revision 24069 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-07-06 07:40:35 +00:00
Michael Niedermayer 733f5990d0 Factorize ff_generate_sliding_window_mmcos() out.
Originally committed as revision 24056 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-07-05 12:42:19 +00:00
Måns Rullgård 49bd8e4b84 Fix grammar errors in documentation
Originally committed as revision 23904 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-06-30 15:38:06 +00:00
Howard Chu 82f1ffc7ba Cleanup prev commit, flag variable should start with 0
Originally committed as revision 23364 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-05-28 20:14:14 +00:00
Howard Chu 23584bec87 Parse avctx->extradata if available.
Fixes many "non-existing PPS referenced" error messages

Originally committed as revision 23363 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-05-28 18:50:39 +00:00
Howard Chu 05e953193d Factorize ff_h264_decode_extradata().
Patch by Howard Chu, hyc highlandsun com

Originally committed as revision 23340 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-05-26 19:00:59 +00:00
Diego Biurrun ba87f0801d Remove explicit filename from Doxygen @file commands.
Passing an explicit filename to this command is only necessary if the
documentation in the @file block refers to a file different from the
one the block resides in.

Originally committed as revision 22921 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-04-20 14:45:34 +00:00
Diego Biurrun d02bb3ecf1 Move static function fill_filter_caches() from h264.h to h264.c.
The function is only used within that file, so it makes sense to place
it there. This fixes many warnings of the type:
h264.h:1170: warning: ‘fill_filter_caches’ defined but not used

Originally committed as revision 22876 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-04-13 22:15:49 +00:00
Michael Niedermayer 1052b76f0f Fix implicit weight for b frames in mbaff.
Originally committed as revision 22733 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-03-30 21:05:11 +00:00
Benoit Fouet 32e543f866 Replace @returns by @return.
Originally committed as revision 22729 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-03-30 15:50:57 +00:00
Måns Rullgård 4693b031a3 Move H264 dsputil functions into their own struct
This moves the H264-specific functions from DSPContext to the new
H264DSPContext.  The code is made conditional on CONFIG_H264DSP
which is set by the codecs requiring it.

The qpel and chroma MC functions are not moved as these are used by
non-h264 code.

Originally committed as revision 22565 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-03-16 01:17:00 +00:00
Måns Rullgård 404793f4ac H264: fix signed overflow in constant multiplication
This fixes libavcodec/h264.h:1100: warning: integer overflow in expression

Originally committed as revision 22558 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-03-15 23:00:53 +00:00
Måns Rullgård 84dc2d8afa Remove DECLARE_ALIGNED_{8,16} macros
These macros are redundant.  All uses are replaced with the generic
DECLARE_ALIGNED macro instead.

Originally committed as revision 22233 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-03-06 14:24:59 +00:00
Michael Niedermayer 38768cb70a Port Optimizations about *_type init from decode to filter code.
1 cpu cycle faster

Originally committed as revision 22193 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-03-04 02:00:05 +00:00
Michael Niedermayer b46b5ac9f8 Optimize *_type init, 1.5 cpu cycles faster.
Originally committed as revision 22192 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-03-04 01:03:15 +00:00
Michael Niedermayer 3d9137c883 Reorder indexes in weight tables.
5 cpu cycles faster.

Originally committed as revision 22183 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-03-03 21:10:08 +00:00
Michael Niedermayer bd8868e092 Move all context fields that are not used in the mb and block layers
to the end of the structure.
4 cpu cycles faster in 3k cpu cycles

Originally committed as revision 22181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-03-03 20:36:56 +00:00
Michael Niedermayer 65f3c029b9 remove unused left_border field from context.
Originally committed as revision 22179 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-03-03 19:44:27 +00:00
Michael Niedermayer af2b0df40f Note about luma/chroma_weight tables and their datatype.
Originally committed as revision 22177 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-03-03 19:31:58 +00:00
Michael Niedermayer d7f5e520bf move svq3 specific fields to the end of the context
Originally committed as revision 22171 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-03-03 16:47:40 +00:00
Michael Niedermayer 70118abd68 Merge weight & offset tables, 15 cpu cycles faster.
Originally committed as revision 22169 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-03-03 14:41:43 +00:00
Michael Niedermayer f57880d244 Another 3 useless zeroing instructions.
Originally committed as revision 22162 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-03-03 02:20:48 +00:00
Michael Niedermayer 16b802fe93 Load the whole left side of mv&ref only when needed.
30 cpu cycles faster

Originally committed as revision 22161 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-03-03 01:38:27 +00:00
Michael Niedermayer ce9c691616 Merge h->slice_table[left_xy[0/1] ] checks, 4 cpu cycles speedup
Originally committed as revision 22086 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-27 04:09:48 +00:00
Michael Niedermayer 82fb5bb2ee Split *_type setting up, 4 cpu cycles faster.
Originally committed as revision 22085 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-27 03:46:16 +00:00
Michael Niedermayer cf41a02b1b Only load the topleft mv/ref when the topright is unavailable.
8 cpu cycles faster.

Originally committed as revision 22079 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-26 15:26:11 +00:00
Michael Niedermayer cf7b67bc40 Remove some useless operations from the code setting left_cbp.
maybe 0.5 cpu cycles faster

Originally committed as revision 22078 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-26 15:03:00 +00:00
Michael Niedermayer 59b5370f02 Simplify code to set cbp_*
this seems 1 cpu cycle slower even though we practically just remove code.
Speed loss seems caused by the merge of if(left_type), iam commiting this
anyway as i cant imagine this to be anything but compiler messup.

Originally committed as revision 22073 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-26 09:13:40 +00:00
Michael Niedermayer 747db4e31a Move init of right side of ref_cache from fill_caches() to init_the_darn_decoder().
Originally committed as revision 22071 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-26 03:27:52 +00:00
Michael Niedermayer 77c6edb846 Remove 3 mv_cache zeroing instructions that zeroed the right side.
This seems unneeded as nothing seems to ever set it to non zero values.

Originally committed as revision 22070 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-26 02:54:03 +00:00
Michael Niedermayer 8f8497ae78 Remove useless check of the 2 left MBs of a pair being in the same slice.
Originally committed as revision 22069 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-26 01:38:12 +00:00
Michael Niedermayer 6e2fe0f20a Remove unneeded line of code from the neighbor setting code in h264.
Originally committed as revision 22067 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-26 00:10:35 +00:00
Michael Niedermayer 358b5b1a59 Get rid of mb2b8_xy and b8_stride, change arrays organized based on b8_stride to
ones based on mb_stride in h264.
about 20 cpu cycles faster overall per MB

Originally committed as revision 22065 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-25 23:44:42 +00:00
Michael Niedermayer 5e350863cc Store data in direct_table interleaved.
seems 20cpu cycles faster

Originally committed as revision 22055 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-25 15:27:55 +00:00
Michael Niedermayer 013202d720 Simplify intra4x4_pred_mode_cache init.
Originally committed as revision 22054 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-25 14:54:31 +00:00
Michael Niedermayer 662a5b2370 Reorder intra4x4_pred_mode so that we can read/write 4 values at once.
3-7 cpu cycles faster

Originally committed as revision 22053 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-25 14:26:12 +00:00
Michael Niedermayer 5b0fb5244d Store intra4x4_pred_mode per row only.
about 5 cpu cycles slower in the local code but should be overall faster
due to reduced cache use. (my sample though has too few intra4x4 blocks
for this to be meassureable easily either way)

Originally committed as revision 22052 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-25 14:02:39 +00:00
Michael Niedermayer c2186cbddc unroll tiny and trivial loop. Same speed but clearer.
Originally committed as revision 22051 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-25 12:51:32 +00:00
Michael Niedermayer e1c88a2138 Cut the size of mvd_table by yet another factor of 2.
The code read/write code itself was 1 cycle faster, overall its
likely more due to cache effects

Originally committed as revision 22048 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-25 04:11:33 +00:00
Michael Niedermayer d43c192236 Keep mvd_table values of only 2 mb rows.
Originally committed as revision 22047 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-25 02:42:25 +00:00
Michael Niedermayer b5bd070029 Change mvd_cache & mvd_table to 8bit, this is overall a bit faster
for high resolution videos.
about 20cycles faster per MB for cathederal.

Originally committed as revision 22038 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-24 20:43:06 +00:00
Michael Niedermayer 9127a369ad Replace /2 by faster >>1 as the mvd values are now all positive.
Originally committed as revision 22013 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-24 01:57:31 +00:00
Michael Niedermayer 5c34e36a23 Remove unused variable. Seems i forgot to commit this.
Originally committed as revision 22012 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-24 01:56:27 +00:00
Diego Biurrun dd3475682e Remove unused variable, fixes warnings of the type:
libavcodec/h264.h:816: warning: unused variable `mb_xy'

Originally committed as revision 21941 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-21 15:29:17 +00:00
Måns Rullgård 19769ece3b H264: use alias-safe macros
This eliminates all aliasing violation warnings in h264 code.
No measurable speed difference with gcc-4.4.3 on i7.

Originally committed as revision 21881 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-18 16:24:31 +00:00
Michael Niedermayer 69a28f3e2b Move predict_field_decoding_flag() from h264.h to .c as its only used there and belongs
there as well.

Originally committed as revision 21861 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-17 02:25:05 +00:00
Michael Niedermayer 69cc31832f Move check for and call of predict_field_decoding_flag() from the mb code to
the row code. This function would only be needed on a MB basis for MBAFF+FMO

Originally committed as revision 21860 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-17 02:14:02 +00:00