Much faster CAVLC residual coding
commit77028cd3671de855affb02ffefe6bbd99ac7816e
authorJason Garrett-Glaser <darkshikari@gmail.com>
Mon, 8 Dec 2008 21:44:23 +0000 (8 13:44 -0800)
committerJason Garrett-Glaser <darkshikari@gmail.com>
Thu, 11 Dec 2008 06:48:18 +0000 (10 22:48 -0800)
tree75d17b5844cdbfc98a36e07cd7ead42036714f9f
parentf773bf06256a467f6b18418d97ce2c7ddbe5728c
Much faster CAVLC residual coding
Use a VLC table for common levelcodes instead of constructing them on-the-spot
Branchless version of i_trailing calculation (2x faster on Nehalem)
Completely remove array_non_zero_count and instead use the count calculated in level/run coding.  Note: this slightly changes output with subme > 7 due to different nonzero counts being stored during qpel RD.
common/bs.h
common/common.h
common/macroblock.h
common/vlc.c
common/x86/util.h
encoder/cabac.c
encoder/cavlc.c
encoder/encoder.c
encoder/macroblock.c