]> granicus.if.org Git - libvpx/commit
projects / libvpx / commit
? search:
summary | shortlog | log | commit | commitdiff | tree
(parent: 3967435) | patch
AVX2 SAD Optimization:
authorlevytamar82 <levytamar82@gmail.com>
Thu, 13 Mar 2014 21:47:30 +0000 (14:47 -0700)
committerYunqing Wang <yunqingwang@google.com>
Fri, 21 Mar 2014 20:53:32 +0000 (13:53 -0700)
commit0fa8b668c1019005030b3f584c048d8ddf3012d8
tree48c8a34b3064b30ee471bef980f10afe49612cc0tree | snapshot
parent3967435f655fec96dba08051a58c5f66a91b1c5bcommit | diff
AVX2 SAD Optimization:

2 functions were optimized for avx2 by using full 256 bit register
In order to handle 32 elements in parallel instead of only 16 in parallel:
1. vp9_sad32x32x4d
2. vp9_sad64x64x4d

The function level gain is 66% and the user level gain is ~1%.

Change-Id: I4efbb3bc7d8bc03b64b6c98f5cd5c4a9dd3212cb
vp9/common/vp9_rtcd_defs.pl diff | blob | history
vp9/encoder/x86/vp9_sad4d_intrin_avx2.c [new file with mode: 0644] blob
vp9/vp9cx.mk diff | blob | history
Unnamed repository; edit this file 'description' to name the repository.