Re-design quantization process for 32x32 transform block
authorJingning Han <jingning@google.com>
Mon, 7 Jul 2014 19:08:40 +0000 (12:08 -0700)
committerJingning Han <jingning@google.com>
Tue, 8 Jul 2014 23:55:28 +0000 (16:55 -0700)
commit9ad1b9fc67921cfaf58ed85732ed139530edab4a
tree66a4e1abb320a5b3ff3b174904b3a05ade433ef0
parent00fc0e3ff5e34472efc652686703493387fe36ee
Re-design quantization process for 32x32 transform block

This commit enables a new quantization process for 32x32 2D-DCT
transform coefficient blocks. It improves the compression
performance of speed 5 by 1.4%. The overall compression gains of
speed 5 due to the new quantization scheme is 4.7%. It also includes
the SSSE3 implementation of the 32x32 quantization process.

Change-Id: I0855b124fd6462418683f783f5bcb44255c9993b
vp9/common/vp9_rtcd_defs.pl
vp9/encoder/vp9_encodemb.c
vp9/encoder/vp9_quantize.c
vp9/encoder/vp9_speed_features.c
vp9/encoder/x86/vp9_quantize_ssse3_x86_64.asm