Adds masked variance and sad functions for wedge

Adds masked variance and sad functions needed for wedge
prediction modes to come.

Change-Id: I25b231bbc345e6a494316abb0a7d5cd5586a3a54

test/masked_sad_test.cc

+/*
+ *  Copyright (c) 2015 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 <math.h>
+#include <stdlib.h>
+#include <string.h>
+
+#include "third_party/googletest/src/include/gtest/gtest.h"
+#include "test/acm_random.h"
+#include "test/clear_system_state.h"
+#include "test/register_state_check.h"
+#include "test/util.h"
+
+#include "./vpx_config.h"
+#include "./vpx_dsp_rtcd.h"
+#include "vpx/vpx_integer.h"
+
+using libvpx_test::ACMRandom;
+
+namespace {
+const int number_of_iterations = 500;
+
+typedef unsigned int (*MaskedSADFunc)(const uint8_t *a, int a_stride,
+                                      const uint8_t *b, int b_stride,
+                                      const uint8_t *m, int m_stride);
+typedef std::tr1::tuple<MaskedSADFunc, MaskedSADFunc> MaskedSADParam;
+
+class MaskedSADTest : public ::testing::TestWithParam<MaskedSADParam> {
+ public:
+  virtual ~MaskedSADTest() {}
+  virtual void SetUp() {
+    maskedSAD_op_   = GET_PARAM(0);
+    ref_maskedSAD_op_ = GET_PARAM(1);
+  }
+
+  virtual void TearDown() { libvpx_test::ClearSystemState(); }
+
+ protected:
+  MaskedSADFunc maskedSAD_op_;
+  MaskedSADFunc ref_maskedSAD_op_;
+};
+
+TEST_P(MaskedSADTest, OperationCheck) {
+  unsigned int ref_ret, ret;
+  ACMRandom rnd(ACMRandom::DeterministicSeed());
+  DECLARE_ALIGNED(16, uint8_t,  src_ptr[4096]);
+  DECLARE_ALIGNED(16, uint8_t,  ref_ptr[4096]);
+  DECLARE_ALIGNED(16, uint8_t,  msk_ptr[4096]);
+  int err_count = 0;
+  int first_failure = -1;
+  int src_stride = 64;
+  int ref_stride = 64;
+  int msk_stride = 64;
+  for (int i = 0; i < number_of_iterations; ++i) {
+    for (int j = 0; j < 4096; j++) {
+      src_ptr[j] = rnd.Rand8();
+      ref_ptr[j] = rnd.Rand8();
+      msk_ptr[j] = ((rnd.Rand8()&0x7f) > 64) ? rnd.Rand8()&0x3f : 64;
+      assert(msk_ptr[j] <= 64);
+    }
+
+    ref_ret = ref_maskedSAD_op_(src_ptr, src_stride, ref_ptr, ref_stride,
+                                msk_ptr, msk_stride);
+    ASM_REGISTER_STATE_CHECK(ret = maskedSAD_op_(src_ptr, src_stride,
+                                                 ref_ptr, ref_stride,
+                                                 msk_ptr, msk_stride));
+    if (ret != ref_ret) {
+      err_count++;
+      if (first_failure == -1)
+        first_failure = i;
+    }
+  }
+  EXPECT_EQ(0, err_count)
+    << "Error: Masked SAD Test, C output doesn't match SSSE3 output. "
+    << "First failed at test case " << first_failure;
+}
+
+#if CONFIG_VP9_HIGHBITDEPTH
+typedef unsigned int (*HighbdMaskedSADFunc)(const uint8_t *a, int a_stride,
+                                            const uint8_t *b, int b_stride,
+                                            const uint8_t *m, int m_stride);
+typedef std::tr1::tuple<HighbdMaskedSADFunc, HighbdMaskedSADFunc>
+    HighbdMaskedSADParam;
+
+class HighbdMaskedSADTest : public ::testing::
+        TestWithParam<HighbdMaskedSADParam> {
+ public:
+  virtual ~HighbdMaskedSADTest() {}
+  virtual void SetUp() {
+    maskedSAD_op_   = GET_PARAM(0);
+    ref_maskedSAD_op_ = GET_PARAM(1);
+  }
+
+  virtual void TearDown() { libvpx_test::ClearSystemState(); }
+
+ protected:
+  HighbdMaskedSADFunc maskedSAD_op_;
+  HighbdMaskedSADFunc ref_maskedSAD_op_;
+};
+
+TEST_P(HighbdMaskedSADTest, OperationCheck) {
+  unsigned int ref_ret, ret;
+  ACMRandom rnd(ACMRandom::DeterministicSeed());
+  DECLARE_ALIGNED(16, uint16_t,  src_ptr[4096]);
+  DECLARE_ALIGNED(16, uint16_t,  ref_ptr[4096]);
+  DECLARE_ALIGNED(16, uint8_t,  msk_ptr[4096]);
+  uint8_t* src8_ptr = CONVERT_TO_BYTEPTR(src_ptr);
+  uint8_t* ref8_ptr = CONVERT_TO_BYTEPTR(ref_ptr);
+  int err_count = 0;
+  int first_failure = -1;
+  int src_stride = 64;
+  int ref_stride = 64;
+  int msk_stride = 64;
+  for (int i = 0; i < number_of_iterations; ++i) {
+    for (int j = 0; j < 4096; j++) {
+      src_ptr[j] = rnd.Rand16()&0xfff;
+      ref_ptr[j] = rnd.Rand16()&0xfff;
+      msk_ptr[j] = ((rnd.Rand8()&0x7f) > 64) ? rnd.Rand8()&0x3f : 64;
+    }
+
+    ref_ret = ref_maskedSAD_op_(src8_ptr, src_stride, ref8_ptr, ref_stride,
+                                msk_ptr, msk_stride);
+    ASM_REGISTER_STATE_CHECK(ret = maskedSAD_op_(src8_ptr, src_stride,
+                                                 ref8_ptr, ref_stride,
+                                                 msk_ptr, msk_stride));
+    if (ret != ref_ret) {
+      err_count++;
+      if (first_failure == -1)
+        first_failure = i;
+    }
+  }
+  EXPECT_EQ(0, err_count)
+    << "Error: High BD Masked SAD Test, C output doesn't match SSSE3 output. "
+    << "First failed at test case " << first_failure;
+}
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+
+using std::tr1::make_tuple;
+
+#if HAVE_SSSE3
+INSTANTIATE_TEST_CASE_P(
+  SSSE3_C_COMPARE, MaskedSADTest,
+  ::testing::Values(
+    make_tuple(&vpx_masked_sad64x64_ssse3,
+               &vpx_masked_sad64x64_c),
+    make_tuple(&vpx_masked_sad64x32_ssse3,
+               &vpx_masked_sad64x32_c),
+    make_tuple(&vpx_masked_sad32x64_ssse3,
+               &vpx_masked_sad32x64_c),
+    make_tuple(&vpx_masked_sad32x32_ssse3,
+               &vpx_masked_sad32x32_c),
+    make_tuple(&vpx_masked_sad32x16_ssse3,
+               &vpx_masked_sad32x16_c),
+    make_tuple(&vpx_masked_sad16x32_ssse3,
+               &vpx_masked_sad16x32_c),
+    make_tuple(&vpx_masked_sad16x16_ssse3,
+               &vpx_masked_sad16x16_c),
+    make_tuple(&vpx_masked_sad16x8_ssse3,
+               &vpx_masked_sad16x8_c),
+    make_tuple(&vpx_masked_sad8x16_ssse3,
+               &vpx_masked_sad8x16_c),
+    make_tuple(&vpx_masked_sad8x8_ssse3,
+               &vpx_masked_sad8x8_c),
+    make_tuple(&vpx_masked_sad8x4_ssse3,
+               &vpx_masked_sad8x4_c),
+    make_tuple(&vpx_masked_sad4x8_ssse3,
+               &vpx_masked_sad4x8_c),
+    make_tuple(&vpx_masked_sad4x4_ssse3,
+               &vpx_masked_sad4x4_c)));
+#if CONFIG_VP9_HIGHBITDEPTH
+INSTANTIATE_TEST_CASE_P(
+  SSSE3_C_COMPARE, HighbdMaskedSADTest,
+  ::testing::Values(
+    make_tuple(&vp9_highbd_masked_sad64x64_ssse3,
+               &vp9_highbd_masked_sad64x64_c),
+    make_tuple(&vp9_highbd_masked_sad64x32_ssse3,
+               &vp9_highbd_masked_sad64x32_c),
+    make_tuple(&vp9_highbd_masked_sad32x64_ssse3,
+               &vp9_highbd_masked_sad32x64_c),
+    make_tuple(&vp9_highbd_masked_sad32x32_ssse3,
+               &vp9_highbd_masked_sad32x32_c),
+    make_tuple(&vp9_highbd_masked_sad32x16_ssse3,
+               &vp9_highbd_masked_sad32x16_c),
+    make_tuple(&vp9_highbd_masked_sad16x32_ssse3,
+               &vp9_highbd_masked_sad16x32_c),
+    make_tuple(&vp9_highbd_masked_sad16x16_ssse3,
+               &vp9_highbd_masked_sad16x16_c),
+    make_tuple(&vp9_highbd_masked_sad16x8_ssse3,
+               &vp9_highbd_masked_sad16x8_c),
+    make_tuple(&vp9_highbd_masked_sad8x16_ssse3,
+               &vp9_highbd_masked_sad8x16_c),
+    make_tuple(&vp9_highbd_masked_sad8x8_ssse3,
+               &vp9_highbd_masked_sad8x8_c),
+    make_tuple(&vp9_highbd_masked_sad8x4_ssse3,
+               &vp9_highbd_masked_sad8x4_c),
+    make_tuple(&vp9_highbd_masked_sad4x8_ssse3,
+               &vp9_highbd_masked_sad4x8_c),
+    make_tuple(&vp9_highbd_masked_sad4x4_ssse3,
+               &vp9_highbd_masked_sad4x4_c)));
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+#endif  // HAVE_SSSE3
+}  // namespace

test/test.mk

@@ -168,6 +168,11 @@ LIBVPX_TEST_SRCS-$(CONFIG_VP10_ENCODER) += vp10_dct_test.cc
 LIBVPX_TEST_SRCS-$(CONFIG_ANS)          += vp10_ans_test.cc
 
 LIBVPX_TEST_SRCS-$(CONFIG_VP10_ENCODER) += sum_squares_test.cc
+
+ifeq ($(CONFIG_EXT_INTER),yes)
+LIBVPX_TEST_SRCS-$(HAVE_SSSE3) += masked_variance_test.cc
+LIBVPX_TEST_SRCS-$(HAVE_SSSE3) += masked_sad_test.cc
+endif
 endif # VP10
 
 ## Multi-codec / unconditional whitebox tests.

vpx_dsp/sad.c

@@ -316,3 +316,105 @@ highbd_sadMxNxK(4, 4, 8)
 highbd_sadMxNx4D(4, 4)
 
 #endif  // CONFIG_VP9_HIGHBITDEPTH
+
+#if CONFIG_VP10 && CONFIG_EXT_INTER
+static INLINE unsigned int masked_sad(const uint8_t *a, int a_stride,
+                                      const uint8_t *b, int b_stride,
+                                      const uint8_t *m, int m_stride,
+                                      int width, int height) {
+  int y, x;
+  unsigned int sad = 0;
+
+  for (y = 0; y < height; y++) {
+    for (x = 0; x < width; x++)
+      sad += m[x] * abs(a[x] - b[x]);
+
+    a += a_stride;
+    b += b_stride;
+    m += m_stride;
+  }
+  sad = (sad + 31) >> 6;
+
+  return sad;
+}
+
+#define MASKSADMxN(m, n) \
+unsigned int vpx_masked_sad##m##x##n##_c(const uint8_t *src, int src_stride, \
+                                         const uint8_t *ref, int ref_stride, \
+                                         const uint8_t *msk, int msk_stride) { \
+  return masked_sad(src, src_stride, ref, ref_stride, msk, msk_stride, m, n); \
+}
+
+#if CONFIG_EXT_PARTITION
+MASKSADMxN(128, 128)
+MASKSADMxN(128, 64)
+MASKSADMxN(64, 128)
+#endif  // CONFIG_EXT_PARTITION
+MASKSADMxN(64, 64)
+MASKSADMxN(64, 32)
+MASKSADMxN(32, 64)
+MASKSADMxN(32, 32)
+MASKSADMxN(32, 16)
+MASKSADMxN(16, 32)
+MASKSADMxN(16, 16)
+MASKSADMxN(16, 8)
+MASKSADMxN(8, 16)
+MASKSADMxN(8, 8)
+MASKSADMxN(8, 4)
+MASKSADMxN(4, 8)
+MASKSADMxN(4, 4)
+
+#if CONFIG_VP9_HIGHBITDEPTH
+static INLINE unsigned int highbd_masked_sad(const uint8_t *a8, int a_stride,
+                                             const uint8_t *b8, int b_stride,
+                                             const uint8_t *m, int m_stride,
+                                             int width, int height) {
+  int y, x;
+  unsigned int sad = 0;
+  const uint16_t *a = CONVERT_TO_SHORTPTR(a8);
+  const uint16_t *b = CONVERT_TO_SHORTPTR(b8);
+
+  for (y = 0; y < height; y++) {
+    for (x = 0; x < width; x++)
+      sad += m[x] * abs(a[x] - b[x]);
+
+    a += a_stride;
+    b += b_stride;
+    m += m_stride;
+  }
+  sad = (sad + 31) >> 6;
+
+  return sad;
+}
+
+#define HIGHBD_MASKSADMXN(m, n) \
+unsigned int vpx_highbd_masked_sad##m##x##n##_c(const uint8_t *src, \
+                                                int src_stride, \
+                                                const uint8_t *ref, \
+                                                int ref_stride, \
+                                                const uint8_t *msk, \
+                                                int msk_stride) { \
+  return highbd_masked_sad(src, src_stride, ref, ref_stride, \
+                           msk, msk_stride, m, n); \
+}
+
+#if CONFIG_EXT_PARTITION
+HIGHBD_MASKSADMXN(128, 128)
+HIGHBD_MASKSADMXN(128, 64)
+HIGHBD_MASKSADMXN(64, 128)
+#endif  // CONFIG_EXT_PARTITION
+HIGHBD_MASKSADMXN(64, 64)
+HIGHBD_MASKSADMXN(64, 32)
+HIGHBD_MASKSADMXN(32, 64)
+HIGHBD_MASKSADMXN(32, 32)
+HIGHBD_MASKSADMXN(32, 16)
+HIGHBD_MASKSADMXN(16, 32)
+HIGHBD_MASKSADMXN(16, 16)
+HIGHBD_MASKSADMXN(16, 8)
+HIGHBD_MASKSADMXN(8, 16)
+HIGHBD_MASKSADMXN(8, 8)
+HIGHBD_MASKSADMXN(8, 4)
+HIGHBD_MASKSADMXN(4, 8)
+HIGHBD_MASKSADMXN(4, 4)
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+#endif  // CONFIG_VP10 && CONFIG_EXT_INTER

vpx_dsp/variance.c

@@ -15,8 +15,9 @@
 #include "vpx/vpx_integer.h"
 
 #include "vpx_dsp/variance.h"
+#include "vpx_dsp/vpx_filter.h"
 
-static const uint8_t bilinear_filters[8][2] = {
+const uint8_t vpx_bilinear_filters[BIL_SUBPEL_SHIFTS][2] = {
   { 128,   0  },
   { 112,  16  },
   {  96,  32  },
@@ -175,9 +176,9 @@ uint32_t vpx_sub_pixel_variance##W##x##H##_c(const uint8_t *a, int a_stride, \
   uint8_t temp2[H * W]; \
 \
   var_filter_block2d_bil_first_pass(a, fdata3, a_stride, 1, H + 1, W, \
-                                    bilinear_filters[xoffset]); \
+                                    vpx_bilinear_filters[xoffset]); \
   var_filter_block2d_bil_second_pass(fdata3, temp2, W, W, H, W, \
-                                     bilinear_filters[yoffset]); \
+                                     vpx_bilinear_filters[yoffset]); \
 \
   return vpx_variance##W##x##H##_c(temp2, W, b, b_stride, sse); \
 }
@@ -195,9 +196,9 @@ uint32_t vpx_sub_pixel_avg_variance##W##x##H##_c(const uint8_t *a, \
   DECLARE_ALIGNED(16, uint8_t, temp3[H * W]); \
 \
   var_filter_block2d_bil_first_pass(a, fdata3, a_stride, 1, H + 1, W, \
-                                    bilinear_filters[xoffset]); \
+                                    vpx_bilinear_filters[xoffset]); \
   var_filter_block2d_bil_second_pass(fdata3, temp2, W, W, H, W, \
-                                     bilinear_filters[yoffset]); \
+                                     vpx_bilinear_filters[yoffset]); \
 \
   vpx_comp_avg_pred(temp3, second_pred, W, H, temp2, W); \
 \
@@ -500,9 +501,9 @@ uint32_t vpx_highbd_8_sub_pixel_variance##W##x##H##_c( \
   uint16_t temp2[H * W]; \
 \
   highbd_var_filter_block2d_bil_first_pass(src, fdata3, src_stride, 1, H + 1, \
-                                           W, bilinear_filters[xoffset]); \
+                                           W, vpx_bilinear_filters[xoffset]); \
   highbd_var_filter_block2d_bil_second_pass(fdata3, temp2, W, W, H, W, \
-                                            bilinear_filters[yoffset]); \
+                                            vpx_bilinear_filters[yoffset]); \
 \
   return vpx_highbd_8_variance##W##x##H##_c(CONVERT_TO_BYTEPTR(temp2), W, dst, \
                                           dst_stride, sse); \
@@ -517,9 +518,9 @@ uint32_t vpx_highbd_10_sub_pixel_variance##W##x##H##_c( \
   uint16_t temp2[H * W]; \
 \
   highbd_var_filter_block2d_bil_first_pass(src, fdata3, src_stride, 1, H + 1, \
-                                           W, bilinear_filters[xoffset]); \
+                                           W, vpx_bilinear_filters[xoffset]); \
   highbd_var_filter_block2d_bil_second_pass(fdata3, temp2, W, W, H, W, \
-                                            bilinear_filters[yoffset]); \
+                                            vpx_bilinear_filters[yoffset]); \
 \
   return vpx_highbd_10_variance##W##x##H##_c(CONVERT_TO_BYTEPTR(temp2), \
                                              W, dst, dst_stride, sse); \
@@ -534,9 +535,9 @@ uint32_t vpx_highbd_12_sub_pixel_variance##W##x##H##_c( \
   uint16_t temp2[H * W]; \
 \
   highbd_var_filter_block2d_bil_first_pass(src, fdata3, src_stride, 1, H + 1, \
-                                           W, bilinear_filters[xoffset]); \
+                                           W, vpx_bilinear_filters[xoffset]); \
   highbd_var_filter_block2d_bil_second_pass(fdata3, temp2, W, W, H, W, \
-                                            bilinear_filters[yoffset]); \
+                                            vpx_bilinear_filters[yoffset]); \
 \
   return vpx_highbd_12_variance##W##x##H##_c(CONVERT_TO_BYTEPTR(temp2), \
                                              W, dst, dst_stride, sse); \
@@ -554,9 +555,9 @@ uint32_t vpx_highbd_8_sub_pixel_avg_variance##W##x##H##_c( \
   DECLARE_ALIGNED(16, uint16_t, temp3[H * W]); \
 \
   highbd_var_filter_block2d_bil_first_pass(src, fdata3, src_stride, 1, H + 1, \
-                                           W, bilinear_filters[xoffset]); \
+                                           W, vpx_bilinear_filters[xoffset]); \
   highbd_var_filter_block2d_bil_second_pass(fdata3, temp2, W, W, H, W, \
-                                            bilinear_filters[yoffset]); \
+                                            vpx_bilinear_filters[yoffset]); \
 \
   vpx_highbd_comp_avg_pred(temp3, second_pred, W, H, \
                            CONVERT_TO_BYTEPTR(temp2), W); \
@@ -576,9 +577,9 @@ uint32_t vpx_highbd_10_sub_pixel_avg_variance##W##x##H##_c( \
   DECLARE_ALIGNED(16, uint16_t, temp3[H * W]); \
 \
   highbd_var_filter_block2d_bil_first_pass(src, fdata3, src_stride, 1, H + 1, \
-                                           W, bilinear_filters[xoffset]); \
+                                           W, vpx_bilinear_filters[xoffset]); \
   highbd_var_filter_block2d_bil_second_pass(fdata3, temp2, W, W, H, W, \
-                                            bilinear_filters[yoffset]); \
+                                            vpx_bilinear_filters[yoffset]); \
 \
   vpx_highbd_comp_avg_pred(temp3, second_pred, W, H, \
                            CONVERT_TO_BYTEPTR(temp2), W); \
@@ -598,9 +599,9 @@ uint32_t vpx_highbd_12_sub_pixel_avg_variance##W##x##H##_c( \
   DECLARE_ALIGNED(16, uint16_t, temp3[H * W]); \
 \
   highbd_var_filter_block2d_bil_first_pass(src, fdata3, src_stride, 1, H + 1, \
-                                           W, bilinear_filters[xoffset]); \
+                                           W, vpx_bilinear_filters[xoffset]); \
   highbd_var_filter_block2d_bil_second_pass(fdata3, temp2, W, W, H, W, \
-                                            bilinear_filters[yoffset]); \
+                                            vpx_bilinear_filters[yoffset]); \
 \
   vpx_highbd_comp_avg_pred(temp3, second_pred, W, H, \
                            CONVERT_TO_BYTEPTR(temp2), W); \
@@ -654,3 +655,323 @@ void vpx_highbd_comp_avg_pred(uint16_t *comp_pred, const uint8_t *pred8,
   }
 }
 #endif  // CONFIG_VP9_HIGHBITDEPTH
+
+#if CONFIG_VP10 && CONFIG_EXT_INTER
+void masked_variance(const uint8_t *a, int  a_stride,
+                     const uint8_t *b, int  b_stride,
+                     const uint8_t *m, int  m_stride,
+                     int  w, int  h, unsigned int *sse, int *sum) {
+  int i, j;
+
+  int64_t sum64 = 0;
+  uint64_t sse64 = 0;
+
+  for (i = 0; i < h; i++) {
+    for (j = 0; j < w; j++) {
+      const int diff = (a[j] - b[j]) * (m[j]);
+      sum64 += diff;
+      sse64 += diff * diff;
+    }
+
+    a += a_stride;
+    b += b_stride;
+    m += m_stride;
+  }
+  *sum = (sum64 >= 0) ? ((sum64 + 31) >> 6) : -((-sum64 + 31) >> 6);
+  *sse = (sse64 + 2047) >> 12;
+}
+
+#define MASK_VAR(W, H) \
+unsigned int vpx_masked_variance##W##x##H##_c(const uint8_t *a, int a_stride, \
+                                              const uint8_t *b, int b_stride, \
+                                              const uint8_t *m, int m_stride, \
+                                              unsigned int *sse) { \
+  int sum; \
+  masked_variance(a, a_stride, b, b_stride, m, m_stride, W, H, sse, &sum); \
+  return *sse - (((int64_t)sum * sum) / (W * H)); \
+}
+
+#define MASK_SUBPIX_VAR(W, H) \
+unsigned int vpx_masked_sub_pixel_variance##W##x##H##_c( \
+  const uint8_t *src, int  src_stride, \
+  int xoffset, int  yoffset, \
+  const uint8_t *dst, int dst_stride, \
+  const uint8_t *msk, int msk_stride, \
+  unsigned int *sse) { \
+  uint16_t fdata3[(H + 1) * W]; \
+  uint8_t temp2[H * W]; \
+\
+  var_filter_block2d_bil_first_pass(src, fdata3, src_stride, 1, H + 1, W, \
+                                    vpx_bilinear_filters[xoffset]); \
+  var_filter_block2d_bil_second_pass(fdata3, temp2, W, W, H, W, \
+                                     vpx_bilinear_filters[yoffset]); \
+\
+  return vpx_masked_variance##W##x##H##_c(temp2, W, dst, dst_stride, \
+                                          msk, msk_stride, sse); \
+}
+
+MASK_VAR(4, 4)
+MASK_SUBPIX_VAR(4, 4)
+
+MASK_VAR(4, 8)
+MASK_SUBPIX_VAR(4, 8)
+
+MASK_VAR(8, 4)
+MASK_SUBPIX_VAR(8, 4)
+
+MASK_VAR(8, 8)
+MASK_SUBPIX_VAR(8, 8)
+
+MASK_VAR(8, 16)
+MASK_SUBPIX_VAR(8, 16)
+
+MASK_VAR(16, 8)
+MASK_SUBPIX_VAR(16, 8)
+
+MASK_VAR(16, 16)
+MASK_SUBPIX_VAR(16, 16)
+
+MASK_VAR(16, 32)
+MASK_SUBPIX_VAR(16, 32)
+
+MASK_VAR(32, 16)
+MASK_SUBPIX_VAR(32, 16)
+
+MASK_VAR(32, 32)
+MASK_SUBPIX_VAR(32, 32)
+
+MASK_VAR(32, 64)
+MASK_SUBPIX_VAR(32, 64)
+
+MASK_VAR(64, 32)
+MASK_SUBPIX_VAR(64, 32)
+
+MASK_VAR(64, 64)
+MASK_SUBPIX_VAR(64, 64)
+
+#if CONFIG_EXT_PARTITION
+MASK_VAR(64, 128)
+MASK_SUBPIX_VAR(64, 128)
+
+MASK_VAR(128, 64)
+MASK_SUBPIX_VAR(128, 64)
+
+MASK_VAR(128, 128)
+MASK_SUBPIX_VAR(128, 128)
+#endif  // CONFIG_EXT_PARTITION
+
+#if CONFIG_VP9_HIGHBITDEPTH
+void highbd_masked_variance64(const uint8_t *a8, int  a_stride,
+                              const uint8_t *b8, int  b_stride,
+                              const uint8_t *m, int  m_stride,
+                              int  w, int  h,
+                              uint64_t *sse64, int *sum) {
+  int i, j;
+  uint16_t *a = CONVERT_TO_SHORTPTR(a8);
+  uint16_t *b = CONVERT_TO_SHORTPTR(b8);
+
+  int64_t sum64 = 0;
+  *sse64 = 0;
+
+  for (i = 0; i < h; i++) {
+    for (j = 0; j < w; j++) {
+      const int diff = (a[j] - b[j]) * (m[j]);
+      sum64 += diff;
+      *sse64 += (int64_t)diff * diff;
+    }
+
+    a += a_stride;
+    b += b_stride;
+    m += m_stride;
+  }
+  *sum = (sum64 >= 0) ? ((sum64 + 31) >> 6) : -((-sum64 + 31) >> 6);
+  *sse64 = (*sse64 + 2047) >> 12;
+}
+
+void highbd_masked_variance(const uint8_t *a8, int  a_stride,
+                            const uint8_t *b8, int  b_stride,
+                            const uint8_t *m, int  m_stride,
+                            int  w, int  h,
+                            unsigned int *sse, int *sum) {
+  uint64_t sse64;
+  highbd_masked_variance64(a8, a_stride, b8, b_stride, m, m_stride,
+                           w, h, &sse64, sum);
+  *sse = (unsigned int)sse64;
+}
+
+void highbd_10_masked_variance(const uint8_t *a8, int  a_stride,
+                               const uint8_t *b8, int  b_stride,
+                               const uint8_t *m, int  m_stride,
+                               int  w, int  h,
+                               unsigned int *sse, int *sum) {
+  uint64_t sse64;
+  highbd_masked_variance64(a8, a_stride, b8, b_stride, m, m_stride,
+                           w, h, &sse64, sum);
+  *sum = ROUND_POWER_OF_TWO(*sum, 2);
+  *sse = (unsigned int)ROUND_POWER_OF_TWO(sse64, 4);
+}
+
+void highbd_12_masked_variance(const uint8_t *a8, int  a_stride,
+                               const uint8_t *b8, int  b_stride,
+                               const uint8_t *m, int  m_stride,
+                               int  w, int  h,
+                               unsigned int *sse, int *sum) {
+  uint64_t sse64;
+  highbd_masked_variance64(a8, a_stride, b8, b_stride, m, m_stride,
+                           w, h, &sse64, sum);
+  *sum = ROUND_POWER_OF_TWO(*sum, 4);
+  *sse = (unsigned int)ROUND_POWER_OF_TWO(sse64, 8);
+}
+
+#define HIGHBD_MASK_VAR(W, H) \
+unsigned int vpx_highbd_masked_variance##W##x##H##_c(const uint8_t *a, \
+                                                     int a_stride, \
+                                                     const uint8_t *b, \
+                                                     int b_stride, \
+                                                     const uint8_t *m, \
+                                                     int m_stride, \
+                                                     unsigned int *sse) { \
+  int sum; \
+  highbd_masked_variance(a, a_stride, b, b_stride, m, m_stride, \
+                         W, H, sse, &sum); \
+  return *sse - (((int64_t)sum * sum) / (W * H)); \
+} \
+\
+unsigned int vpx_highbd_10_masked_variance##W##x##H##_c(const uint8_t *a, \
+                                                        int a_stride, \
+                                                        const uint8_t *b, \
+                                                        int b_stride, \
+                                                        const uint8_t *m, \
+                                                        int m_stride, \
+                                                        unsigned int *sse) { \
+  int sum; \
+  highbd_10_masked_variance(a, a_stride, b, b_stride, m, m_stride, \
+                            W, H, sse, &sum); \
+  return *sse - (((int64_t)sum * sum) / (W * H)); \
+} \
+\
+unsigned int vpx_highbd_12_masked_variance##W##x##H##_c(const uint8_t *a, \
+                                                        int a_stride, \
+                                                        const uint8_t *b, \
+                                                        int b_stride, \
+                                                        const uint8_t *m, \
+                                                        int m_stride, \
+                                                        unsigned int *sse) { \
+  int sum; \
+  highbd_12_masked_variance(a, a_stride, b, b_stride, m, m_stride, \
+                            W, H, sse, &sum); \
+  return *sse - (((int64_t)sum * sum) / (W * H)); \
+}