commit ce110cc563d89885d2d8e20c2f844c57642a78d3
author David Barker <david.barker@argondesign.com> Wed Feb 22 10:38:59 2017 +0000
committer Debargha Mukherjee <debargha@google.com> Mon Mar 06 20:23:16 2017 +0000
tree 5f4f3acd867388aaea213a739aba9135779fd0fb
parent 4d5bbbd907a1eb7acdafe2a4a7594721ec33a850

Vectorize self-guided filter

Add an SSE4.1 lowbd version of the self-guided filter for
loop-restoration, and apply some optimizations to the C
version.

Approximate times per 128x128 / 256x256 tile on the machine
this was developed on:
Previous C:  620us / 2800us
Optimized C: 500us / 2200us ( 24% /  27% faster)
SSE4.1:      147us / 600us  (320% / 370% faster)

Change-Id: I23ff5a5482a191aeb06f9d1f767a9f036bb357fe

test/test.mk

diff --git a/test/test.mk b/test/test.mk
index 01014e6..8ffa87a 100644
--- a/test/test.mk
+++ b/test/test.mk
@@ -209,6 +209,9 @@
 ifneq ($(findstring yes,$(CONFIG_GLOBAL_MOTION) $(CONFIG_WARPED_MOTION)),)
 LIBAOM_TEST_SRCS-$(HAVE_SSE2) += warp_filter_test.cc
 endif
+ifeq ($(CONFIG_LOOP_RESTORATION),yes)
+LIBAOM_TEST_SRCS-$(HAVE_SSE4_1) += selfguided_filter_test.cc
+endif
 
 TEST_INTRA_PRED_SPEED_SRCS-yes := test_intra_pred_speed.cc
 TEST_INTRA_PRED_SPEED_SRCS-yes += ../md5_utils.h ../md5_utils.c