Combine fdct8x8 and quantization process
authorJingning Han <jingning@google.com>
Tue, 18 Nov 2014 19:53:14 +0000 (11:53 -0800)
committerJingning Han <jingning@google.com>
Wed, 19 Nov 2014 02:10:56 +0000 (18:10 -0800)
commitc6908fd5f744552db044dd034f88967e5fd980b4
tree725cdee3a2651da747eeb39c45d30a89cc34c6f8
parent2d3cc8ea2b163fd7275ed0c44102659ff0acd68b
Combine fdct8x8 and quantization process

This commit reworks the forward transform and quantization process
for 8x8 block coding. It combines the two operations in a single
function to save a store/load stage of the original transform
coefficients. Overall the speed -6 is slightly faster (around 1%
range). The compression performance of speed -6 is improved by
3.4%.

Change-Id: Id6628daef123f3e4649248735ec2ad7423629387
vp9/common/vp9_rtcd_defs.pl
vp9/encoder/vp9_dct.c
vp9/encoder/vp9_encodemb.c
vp9/encoder/x86/vp9_dct_sse2.c