X-Git-Url: http://review.tizen.org/git/?a=blobdiff_plain;f=src%2Fthird_party%2Flibvpx%2Fsource%2Flibvpx%2Fvp9%2Fencoder%2Fx86%2Fvp9_variance_sse2.c;h=b4d2b0ac408d41fcfab53db5f045a6fb68cdd2b3;hb=3545e9f2671f595d2a2f3ee75ca0393b01e35ef6;hp=e935a233a896f967a96c5632a6d7cef59c2c801b;hpb=7d210d4c7e9ba36e635eabc5b5780495f8a63292;p=platform%2Fframework%2Fweb%2Fcrosswalk.git

diff --git a/src/third_party/libvpx/source/libvpx/vp9/encoder/x86/vp9_variance_sse2.c b/src/third_party/libvpx/source/libvpx/vp9/encoder/x86/vp9_variance_sse2.c
index e935a23..b4d2b0a 100644
--- a/src/third_party/libvpx/source/libvpx/vp9/encoder/x86/vp9_variance_sse2.c
+++ b/src/third_party/libvpx/source/libvpx/vp9/encoder/x86/vp9_variance_sse2.c
@@ -8,6 +8,8 @@
  *  be found in the AUTHORS file in the root of the source tree.
  */
 
+#include <emmintrin.h>  // SSE2
+
 #include "./vpx_config.h"
 
 #include "vp9/encoder/vp9_variance.h"
@@ -17,18 +19,137 @@ typedef unsigned int (*variance_fn_t) (const unsigned char *src, int src_stride,
                                        const unsigned char *ref, int ref_stride,
                                        unsigned int *sse, int *sum);
 
-unsigned int vp9_get4x4var_mmx(const unsigned char *src, int src_stride,
-                               const unsigned char *ref, int ref_stride,
-                               unsigned int *sse, int *sum);
+unsigned int vp9_get_mb_ss_sse2(const int16_t *src) {
+  __m128i vsum = _mm_setzero_si128();
+  int i;
+
+  for (i = 0; i < 32; ++i) {
+    const __m128i v = _mm_loadu_si128((const __m128i *)src);
+    vsum = _mm_add_epi32(vsum, _mm_madd_epi16(v, v));
+    src += 8;
+  }
 
+  vsum = _mm_add_epi32(vsum, _mm_srli_si128(vsum, 8));
+  vsum = _mm_add_epi32(vsum, _mm_srli_si128(vsum, 4));
+  return  _mm_cvtsi128_si32(vsum);
+}
 
-unsigned int vp9_get8x8var_sse2(const unsigned char *src, int src_stride,
-                                const unsigned char *ref, int ref_stride,
-                                unsigned int *sse, int *sum);
+#define READ64(p, stride, i) \
+  _mm_unpacklo_epi8(_mm_cvtsi32_si128(*(const uint32_t *)(p + i * stride)), \
+      _mm_cvtsi32_si128(*(const uint32_t *)(p + (i + 1) * stride)))
+
+unsigned int vp9_get4x4var_sse2(const uint8_t *src, int src_stride,
+                                const uint8_t *ref, int ref_stride,
+                                unsigned int *sse, int *sum) {
+  const __m128i zero = _mm_setzero_si128();
+  const __m128i src0 = _mm_unpacklo_epi8(READ64(src, src_stride, 0), zero);
+  const __m128i src1 = _mm_unpacklo_epi8(READ64(src, src_stride, 2), zero);
+  const __m128i ref0 = _mm_unpacklo_epi8(READ64(ref, ref_stride, 0), zero);
+  const __m128i ref1 = _mm_unpacklo_epi8(READ64(ref, ref_stride, 2), zero);
+  const __m128i diff0 = _mm_sub_epi16(src0, ref0);
+  const __m128i diff1 = _mm_sub_epi16(src1, ref1);
+
+  // sum
+  __m128i vsum = _mm_add_epi16(diff0, diff1);
+  vsum = _mm_add_epi16(vsum, _mm_srli_si128(vsum, 8));
+  vsum = _mm_add_epi16(vsum, _mm_srli_si128(vsum, 4));
+  vsum = _mm_add_epi16(vsum, _mm_srli_si128(vsum, 2));
+  *sum = (int16_t)_mm_extract_epi16(vsum, 0);
+
+  // sse
+  vsum = _mm_add_epi32(_mm_madd_epi16(diff0, diff0),
+                       _mm_madd_epi16(diff1, diff1));
+  vsum = _mm_add_epi32(vsum, _mm_srli_si128(vsum, 8));
+  vsum = _mm_add_epi32(vsum, _mm_srli_si128(vsum, 4));
+  *sse = _mm_cvtsi128_si32(vsum);
+
+  return 0;
+}
+
+unsigned int vp9_get8x8var_sse2(const uint8_t *src, int src_stride,
+                                const uint8_t *ref, int ref_stride,
+                                unsigned int *sse, int *sum) {
+  const __m128i zero = _mm_setzero_si128();
+  __m128i vsum = _mm_setzero_si128();
+  __m128i vsse = _mm_setzero_si128();
+  int i;
+
+  for (i = 0; i < 8; i += 2) {
+    const __m128i src0 = _mm_unpacklo_epi8(_mm_loadl_epi64(
+        (const __m128i *)(src + i * src_stride)), zero);
+    const __m128i ref0 = _mm_unpacklo_epi8(_mm_loadl_epi64(
+        (const __m128i *)(ref + i * ref_stride)), zero);
+    const __m128i diff0 = _mm_sub_epi16(src0, ref0);
+
+    const __m128i src1 = _mm_unpacklo_epi8(_mm_loadl_epi64(
+        (const __m128i *)(src + (i + 1) * src_stride)), zero);
+    const __m128i ref1 = _mm_unpacklo_epi8(_mm_loadl_epi64(
+        (const __m128i *)(ref + (i + 1) * ref_stride)), zero);
+    const __m128i diff1 = _mm_sub_epi16(src1, ref1);
+
+    vsum = _mm_add_epi16(vsum, diff0);
+    vsum = _mm_add_epi16(vsum, diff1);
+    vsse = _mm_add_epi32(vsse, _mm_madd_epi16(diff0, diff0));
+    vsse = _mm_add_epi32(vsse, _mm_madd_epi16(diff1, diff1));
+  }
+
+  // sum
+  vsum = _mm_add_epi16(vsum, _mm_srli_si128(vsum, 8));
+  vsum = _mm_add_epi16(vsum, _mm_srli_si128(vsum, 4));
+  vsum = _mm_add_epi16(vsum, _mm_srli_si128(vsum, 2));
+  *sum = (int16_t)_mm_extract_epi16(vsum, 0);
+
+  // sse
+  vsse = _mm_add_epi32(vsse, _mm_srli_si128(vsse, 8));
+  vsse = _mm_add_epi32(vsse, _mm_srli_si128(vsse, 4));
+  *sse = _mm_cvtsi128_si32(vsse);
+
+  return 0;
+}
+
+unsigned int vp9_get16x16var_sse2(const uint8_t *src, int src_stride,
+                                  const uint8_t *ref, int ref_stride,
+                                  unsigned int *sse, int *sum) {
+  const __m128i zero = _mm_setzero_si128();
+  __m128i vsum = _mm_setzero_si128();
+  __m128i vsse = _mm_setzero_si128();
+  int i;
+
+  for (i = 0; i < 16; ++i) {
+    const __m128i s = _mm_loadu_si128((const __m128i *)src);
+    const __m128i r = _mm_loadu_si128((const __m128i *)ref);
+
+    const __m128i src0 = _mm_unpacklo_epi8(s, zero);
+    const __m128i ref0 = _mm_unpacklo_epi8(r, zero);
+    const __m128i diff0 = _mm_sub_epi16(src0, ref0);
+
+    const __m128i src1 = _mm_unpackhi_epi8(s, zero);
+    const __m128i ref1 = _mm_unpackhi_epi8(r, zero);
+    const __m128i diff1 = _mm_sub_epi16(src1, ref1);
+
+    vsum = _mm_add_epi16(vsum, diff0);
+    vsum = _mm_add_epi16(vsum, diff1);
+    vsse = _mm_add_epi32(vsse, _mm_madd_epi16(diff0, diff0));
+    vsse = _mm_add_epi32(vsse, _mm_madd_epi16(diff1, diff1));
+
+    src += src_stride;
+    ref += ref_stride;
+  }
+
+  // sum
+  vsum = _mm_add_epi16(vsum, _mm_srli_si128(vsum, 8));
+  vsum = _mm_add_epi16(vsum, _mm_srli_si128(vsum, 4));
+  *sum = (int16_t)_mm_extract_epi16(vsum, 0) +
+             (int16_t)_mm_extract_epi16(vsum, 1);
+
+  // sse
+  vsse = _mm_add_epi32(vsse, _mm_srli_si128(vsse, 8));
+  vsse = _mm_add_epi32(vsse, _mm_srli_si128(vsse, 4));
+  *sse = _mm_cvtsi128_si32(vsse);
+
+  return 0;
+}
 
-unsigned int vp9_get16x16var_sse2(const unsigned char *src, int src_stride,
-                                  const unsigned char *ref, int ref_stride,
-                                  unsigned int *sse, int *sum);
 
 static void variance_sse2(const unsigned char *src, int src_stride,
                           const unsigned char *ref, int ref_stride,
@@ -55,8 +176,7 @@ unsigned int vp9_variance4x4_sse2(const unsigned char *src, int src_stride,
                                   const unsigned char *ref, int ref_stride,
                                   unsigned int *sse) {
   int sum;
-  variance_sse2(src, src_stride, ref, ref_stride, 4, 4,
-                sse, &sum, vp9_get4x4var_mmx, 4);
+  vp9_get4x4var_sse2(src, src_stride, ref, ref_stride, sse, &sum);
   return *sse - (((unsigned int)sum * sum) >> 4);
 }
 
@@ -65,7 +185,7 @@ unsigned int vp9_variance8x4_sse2(const uint8_t *src, int src_stride,
                                   unsigned int *sse) {
   int sum;
   variance_sse2(src, src_stride, ref, ref_stride, 8, 4,
-                sse, &sum, vp9_get4x4var_mmx, 4);
+                sse, &sum, vp9_get4x4var_sse2, 4);
   return *sse - (((unsigned int)sum * sum) >> 5);
 }
 
@@ -74,7 +194,7 @@ unsigned int vp9_variance4x8_sse2(const uint8_t *src, int src_stride,
                                   unsigned int *sse) {
   int sum;
   variance_sse2(src, src_stride, ref, ref_stride, 4, 8,
-                sse, &sum, vp9_get4x4var_mmx, 4);
+                sse, &sum, vp9_get4x4var_sse2, 4);
   return *sse - (((unsigned int)sum * sum) >> 5);
 }
 
@@ -82,8 +202,7 @@ unsigned int vp9_variance8x8_sse2(const unsigned char *src, int src_stride,
                                   const unsigned char *ref, int ref_stride,
                                   unsigned int *sse) {
   int sum;
-  variance_sse2(src, src_stride, ref, ref_stride, 8, 8,
-                sse, &sum, vp9_get8x8var_sse2, 8);
+  vp9_get8x8var_sse2(src, src_stride, ref, ref_stride, sse, &sum);
   return *sse - (((unsigned int)sum * sum) >> 6);
 }
 
@@ -109,17 +228,8 @@ unsigned int vp9_variance16x16_sse2(const unsigned char *src, int src_stride,
                                     const unsigned char *ref, int ref_stride,
                                     unsigned int *sse) {
   int sum;
-  variance_sse2(src, src_stride, ref, ref_stride, 16, 16,
-                sse, &sum, vp9_get16x16var_sse2, 16);
-  return *sse - (((unsigned int)sum * sum) >> 8);
-}
-
-unsigned int vp9_mse16x16_sse2(const unsigned char *src, int src_stride,
-                               const unsigned char *ref, int ref_stride,
-                               unsigned int *sse) {
-  int sum;
   vp9_get16x16var_sse2(src, src_stride, ref, ref_stride, sse, &sum);
-  return *sse;
+  return *sse - (((unsigned int)sum * sum) >> 8);
 }
 
 unsigned int vp9_variance32x32_sse2(const uint8_t *src, int src_stride,
@@ -176,6 +286,34 @@ unsigned int vp9_variance32x64_sse2(const uint8_t *src, int src_stride,
   return *sse - (((int64_t)sum * sum) >> 11);
 }
 
+unsigned int vp9_mse8x8_sse2(const uint8_t *src, int src_stride,
+                             const uint8_t *ref, int ref_stride,
+                             unsigned int *sse) {
+  vp9_variance8x8_sse2(src, src_stride, ref, ref_stride, sse);
+  return *sse;
+}
+
+unsigned int vp9_mse8x16_sse2(const uint8_t *src, int src_stride,
+                              const uint8_t *ref, int ref_stride,
+                              unsigned int *sse) {
+  vp9_variance8x16_sse2(src, src_stride, ref, ref_stride, sse);
+  return *sse;
+}
+
+unsigned int vp9_mse16x8_sse2(const uint8_t *src, int src_stride,
+                              const uint8_t *ref, int ref_stride,
+                              unsigned int *sse) {
+  vp9_variance16x8_sse2(src, src_stride, ref, ref_stride, sse);
+  return *sse;
+}
+
+unsigned int vp9_mse16x16_sse2(const uint8_t *src, int src_stride,
+                               const uint8_t *ref, int ref_stride,
+                               unsigned int *sse) {
+  vp9_variance16x16_sse2(src, src_stride, ref, ref_stride, sse);
+  return *sse;
+}
+
 #define DECL(w, opt) \
 int vp9_sub_pixel_variance##w##xh_##opt(const uint8_t *src, \
                                         ptrdiff_t src_stride, \