Added sse2 instrinsic version of vp8_sad16x3

3.7% boost in decoder performance for the clip used.

Change-Id: I74f28486a9352b472b36e21b5eaf30eff35e9199

diff --git a/vp8/common/rtcd_defs.sh b/vp8/common/rtcd_defs.sh
index ea64c9682f3d697a47dd6a82c0c6db39fd335e70..90debf5cbcb0a39fc4b81f54f9712127c2db7b30 100644 (file)
--- a/vp8/common/rtcd_defs.sh
+++ b/vp8/common/rtcd_defs.sh
@@ -177,11 +177,13 @@ vp8_loop_filter_simple_bh_neon=vp8_loop_filter_bhs_neon
 #
 # sad 16x3, 3x16
 #
+if [ "$CONFIG_NEWBESTREFMV" = "yes" ]; then
 prototype unsigned int vp8_sad16x3 "const unsigned char *src_ptr, int  src_stride, const unsigned char *ref_ptr, int ref_stride, int max_sad"
-specialize vp8_sad16x3
+specialize vp8_sad16x3 sse2
 
 prototype unsigned int vp8_sad3x16 "const unsigned char *src_ptr, int  src_stride, const unsigned char *ref_ptr, int ref_stride, int max_sad"
 specialize vp8_sad3x16
+fi
 
 #
 # Encoder functions below this point.

diff --git a/vp8/common/x86/sadmxn_x86.c b/vp8/common/x86/sadmxn_x86.c
new file mode 100644 (file)
index 0000000..5057703
--- /dev/null
+++ b/vp8/common/x86/sadmxn_x86.c
@@ -0,0 +1,50 @@
+/*
+ *  Copyright (c) 2012 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <emmintrin.h>  // SSE2
+#include "./vpx_config.h"
+#include "./vpx_rtcd.h"
+
+
+#if CONFIG_NEWBESTREFMV
+
+
+#if HAVE_SSE2
+unsigned int vp8_sad16x3_sse2(
+  const unsigned char *src_ptr,
+  int  src_stride,
+  const unsigned char *ref_ptr,
+  int  ref_stride,
+  int max_sad) {
+  __m128i s0, s1, s2;
+  __m128i r0, r1, r2;
+  __m128i sad;
+
+  (void)max_sad;
+
+  s0 = _mm_loadu_si128((const __m128i *)(src_ptr + 0 * src_stride));
+  s1 = _mm_loadu_si128((const __m128i *)(src_ptr + 1 * src_stride));
+  s2 = _mm_loadu_si128((const __m128i *)(src_ptr + 2 * src_stride));
+
+  r0 = _mm_loadu_si128((const __m128i *)(ref_ptr + 0 * src_stride));
+  r1 = _mm_loadu_si128((const __m128i *)(ref_ptr + 1 * src_stride));
+  r2 = _mm_loadu_si128((const __m128i *)(ref_ptr + 2 * src_stride));
+
+  sad = _mm_sad_epu8(s0, r0);
+  sad = _mm_add_epi16(sad,  _mm_sad_epu8(s1, r1));
+  sad = _mm_add_epi16(sad,  _mm_sad_epu8(s2, r2));
+  sad = _mm_add_epi16(sad,  _mm_srli_si128(sad, 8));
+
+  return _mm_cvtsi128_si32(sad);
+}
+#endif
+
+
+#endif  // CONFIG_NEWBESTREFMV

diff --git a/vp8/vp8_common.mk b/vp8/vp8_common.mk
index fbbdec14592505d72fa463f333dbba5feeb372f3..d6f31ed7939856c7f8a891e69dacbd5021b94fb9 100644 (file)
--- a/vp8/vp8_common.mk
+++ b/vp8/vp8_common.mk
@@ -118,10 +118,11 @@ vp8/common/x86/filter_sse4.c.o: CFLAGS += -msse4
 endif
 
 VP8_COMMON_SRCS-$(HAVE_SSE2) += common/x86/filter_sse2.c
+VP8_COMMON_SRCS-$(HAVE_SSE2) += common/x86/sadmxn_x86.c
 ifeq ($(HAVE_SSE2),yes)
 vp8/common/x86/filter_sse2.c.o: CFLAGS += -msse2
 vp8/common/x86/loopfilter_x86.c.o: CFLAGS += -msse2
-vp8/common/loopfilter_filters.c.o: CFLAGS += -msse2
+vp8/common/x86/sadmxn_x86.c.o: CFLAGS += -msse2
 endif
 
 VP8_COMMON_SRCS-$(ARCH_ARM)  += common/arm/arm_systemdependent.c