FFmpeg

Commit Graph

Author	SHA1	Message	Date
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	15 years ago
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	15 years ago
Michael Niedermayer	24006f59cb	10l ;; gcc 2.95 compilation fix by mike Originally committed as revision 21830 to svn://svn.ffmpeg.org/ffmpeg/trunk	15 years ago
Michael Niedermayer	cbfcde9f52	10l use pack16to32(), should fix big endian h264 failures. Originally committed as revision 21829 to svn://svn.ffmpeg.org/ffmpeg/trunk	15 years ago
Michael Niedermayer	6e7c397b5c	Cosmetics, vertical align. Originally committed as revision 21817 to svn://svn.ffmpeg.org/ffmpeg/trunk	15 years ago
Michael Niedermayer	5ca43c25f6	Move setting MB_TYPE_L0L1 for direct MBs up, this is simpler. Originally committed as revision 21794 to svn://svn.ffmpeg.org/ffmpeg/trunk	15 years ago
Michael Niedermayer	5134ad53ed	Simplify if() Originally committed as revision 21791 to svn://svn.ffmpeg.org/ffmpeg/trunk	15 years ago
Michael Niedermayer	752cc0dcb2	Init sub_mb_type cleanly. Originally committed as revision 21790 to svn://svn.ffmpeg.org/ffmpeg/trunk	15 years ago
Michael Niedermayer	923bcb4048	100l Forgoten hunk.unbreak compile before someone notices. Originally committed as revision 21789 to svn://svn.ffmpeg.org/ffmpeg/trunk	15 years ago
Michael Niedermayer	af0d6581e1	Restructure spatial direct MV generation so that the zero predictior shortcut is closer at the top. 50-130 cpu cycles faster depending on which kind of direct MBs are meassured. Originally committed as revision 21788 to svn://svn.ffmpeg.org/ffmpeg/trunk	15 years ago
Michael Niedermayer	edb4df436f	Split spatial and temporal direct MV generation. A little faster and needed for future optimizations. This sadly leads to some code duplication (which i hope i can factor out again after the optimizations on the direcr mv code are done) Originally committed as revision 21787 to svn://svn.ffmpeg.org/ffmpeg/trunk	15 years ago
Michael Niedermayer	812bfd7b44	Special case for spatial direct MV predictor being 0. a little less than 200 cpu cycles faster with the cathedral sample. Originally committed as revision 21785 to svn://svn.ffmpeg.org/ffmpeg/trunk	15 years ago
Michael Niedermayer	828cfff63d	Pack MVs together from the begin for spatial direct, this simplifies the code and is a bit faster (5-10 cpu cycles depending on what is meassured). Originally committed as revision 21784 to svn://svn.ffmpeg.org/ffmpeg/trunk	15 years ago
Michael Niedermayer	daaf6e3e9e	Reorder and factorize mb_type ifs, 1 cpu cycle faster and simpler. Originally committed as revision 21694 to svn://svn.ffmpeg.org/ffmpeg/trunk	15 years ago
Michael Niedermayer	1be49ce54b	Set partitioning to 16x16 for spatial direct MBs with mixed interlacing. 11cylcles slower MV generation 98cycles faster MC Originally committed as revision 21691 to svn://svn.ffmpeg.org/ffmpeg/trunk	15 years ago
Michael Niedermayer	3babc711d8	Detect spatial direct MBs partitioned smaller than 16x16 that can be partitioned as 16x16 (except ones changing interlacing relative to the colocated MB). 20 cycles slower during MV generation 175 cycles faster during MC Originally committed as revision 21690 to svn://svn.ffmpeg.org/ffmpeg/trunk	15 years ago
Michael Niedermayer	8b9eb2dfec	Detect equal 4x4 blocks in spatial direct MBs. 19 cycles slower MV generation 575 cycles faster MC Originally committed as revision 21689 to svn://svn.ffmpeg.org/ffmpeg/trunk	15 years ago
Michael Niedermayer	542d4a756a	Branchless calculation of ref_offset. 7 cpu cycles faster. Originally committed as revision 21685 to svn://svn.ffmpeg.org/ffmpeg/trunk	15 years ago
Michael Niedermayer	fb70c63f70	Remove incorrect fixme, i see no case that is missing. Originally committed as revision 21684 to svn://svn.ffmpeg.org/ffmpeg/trunk	15 years ago
Michael Niedermayer	35c8b9c094	Replace call to pred_motion() in direct spatial mv pred by code and simplify cases that cannot happen away. 8 cpu cycles faster Originally committed as revision 21683 to svn://svn.ffmpeg.org/ffmpeg/trunk	15 years ago
Michael Niedermayer	055a6aa76a	Set x264_build so that checks are simpler. Originally committed as revision 21681 to svn://svn.ffmpeg.org/ffmpeg/trunk	15 years ago
Michael Niedermayer	3b9e61e45f	Zero a/b only in the branch where they need to be zeroed. Originally committed as revision 21679 to svn://svn.ffmpeg.org/ffmpeg/trunk	15 years ago
Michael Niedermayer	f28a2656c3	Merge mv&ref related code for spatial direct MV code. a bit more than 10 cpu cycles faster. Originally committed as revision 21678 to svn://svn.ffmpeg.org/ffmpeg/trunk	15 years ago
Michael Niedermayer	0a8dca8765	Precalculate a few variables for direct mv prediction for interlaced MBs. Originally committed as revision 21677 to svn://svn.ffmpeg.org/ffmpeg/trunk	15 years ago
Michael Niedermayer	464baa6650	Set direct MB partitioning for 16x8 and 8x16 colocated MBs to the respective true partitioning. Originally committed as revision 21675 to svn://svn.ffmpeg.org/ffmpeg/trunk	15 years ago
Michael Niedermayer	e8d964fa52	Factorize setting sub_mb_type out. Originally committed as revision 21674 to svn://svn.ffmpeg.org/ffmpeg/trunk	15 years ago
Michael Niedermayer	668fd14287	Remove FIXMEs for cases that are disallowed by the spec. Originally committed as revision 21669 to svn://svn.ffmpeg.org/ffmpeg/trunk	15 years ago
Michael Niedermayer	33a6c0c9fe	Skip the fill_colmap() case thats for MBAFF if we dont have an MBAFF frame. Originally committed as revision 21667 to svn://svn.ffmpeg.org/ffmpeg/trunk	15 years ago
Michael Niedermayer	488dba7000	Fix colocated map. Originally committed as revision 21664 to svn://svn.ffmpeg.org/ffmpeg/trunk	15 years ago
Michael Niedermayer	d88ea35463	Fix reference selection for colocated MBs from frames to fields. Originally committed as revision 21663 to svn://svn.ffmpeg.org/ffmpeg/trunk	15 years ago
Michael Niedermayer	5aa1d7598f	Add assert(sps.direct_8x8_inference_flag) to FIXME comment. Originally committed as revision 21662 to svn://svn.ffmpeg.org/ffmpeg/trunk	15 years ago
Michael Niedermayer	50f8e98507	ref_offset must be added after checking references validity. Originally committed as revision 21661 to svn://svn.ffmpeg.org/ffmpeg/trunk	15 years ago
Michael Niedermayer	8dcbf4c7aa	Long term references behave different from short term in spatial direct MV calculation. Fixes FRExt_MMCO4_Sony_B / issue1251 Originally committed as revision 21651 to svn://svn.ffmpeg.org/ffmpeg/trunk	15 years ago
Michael Niedermayer	2b3649f656	Fix compilation with -O0. Originally committed as revision 21308 to svn://svn.ffmpeg.org/ffmpeg/trunk	15 years ago
Michael Niedermayer	188d3c510d	Split motion vector prediction off h264.c/h. Originally committed as revision 21174 to svn://svn.ffmpeg.org/ffmpeg/trunk	15 years ago
Michael Niedermayer	943f69a6ea	Split direct mode (macro)block decoding off h264.c. No speedloss meassured (its slightly faster here but that may be random fluctuations) Originally committed as revision 21165 to svn://svn.ffmpeg.org/ffmpeg/trunk	15 years ago

36 Commits (8eecdc8b789e3fa5b640ddd26e29e9254a1dffe3)