Adding high-bitdepth intra prediction functions

Change-Id: I6f5cb101e2dc57c3d3f4d7e0ffb4ddbed027d111

test/test.mk

@@ -128,6 +128,7 @@ LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += fdct4x4_test.cc
 LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += fdct8x8_test.cc
 LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += variance_test.cc
 LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += vp9_subtract_test.cc
+LIBVPX_TEST_SRCS-$(CONFIG_VP9)         += vp9_intrapred_test.cc
 
 ifeq ($(CONFIG_VP9_ENCODER),yes)
 LIBVPX_TEST_SRCS-$(CONFIG_SPATIAL_SVC) += svc_test.cc

test/vp9_intrapred_test.cc

+/*
+ *  Copyright (c) 2014 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 <string>
+
+#include "test/acm_random.h"
+#include "test/clear_system_state.h"
+#include "test/register_state_check.h"
+#include "third_party/googletest/src/include/gtest/gtest.h"
+
+#include "./vpx_config.h"
+#include "./vp9_rtcd.h"
+#include "vp9/common/vp9_blockd.h"
+#include "vp9/common/vp9_pred_common.h"
+#include "vpx_mem/vpx_mem.h"
+#include "test/util.h"
+
+namespace {
+
+using libvpx_test::ACMRandom;
+
+const int count_test_block = 100000;
+
+// Base class for VP9 intra prediction tests.
+class VP9IntraPredBase {
+ public:
+  virtual ~VP9IntraPredBase() { libvpx_test::ClearSystemState(); }
+
+ protected:
+  virtual void Predict(PREDICTION_MODE mode) = 0;
+
+  void CheckPrediction(int test_case_number, int *error_count) const {
+    // For each pixel ensure that the calculated value is the same as reference.
+    for (int y = 0; y < block_size_; y++) {
+      for (int x = 0; x < block_size_; x++) {
+        *error_count += ref_dst_[x + y * stride_] != dst_[x + y * stride_];
+        if (*error_count == 1) {
+          ASSERT_EQ(ref_dst_[x + y * stride_], dst_[x + y * stride_])
+              << " Failed on Test Case Number "<< test_case_number;
+        }
+      }
+    }
+  }
+
+  void RunTest(uint16_t* left_col, uint16_t* above_data,
+               uint16_t* dst, uint16_t* ref_dst) {
+    ACMRandom rnd(ACMRandom::DeterministicSeed());
+    left_col_ = left_col;
+    dst_ = dst;
+    ref_dst_ = ref_dst;
+    above_row_ = above_data + 16;
+    int error_count = 0;
+    for (int i = 0; i < count_test_block; ++i) {
+      // Fill edges with random data, try first with saturated values.
+      for (int x = -1; x <= block_size_*2; x++) {
+        if (i == 0) {
+          above_row_[x] = mask_;
+        } else {
+          above_row_[x] = rnd.Rand16() & mask_;
+        }
+      }
+      for (int y = 0; y < block_size_; y++) {
+        if (i == 0) {
+          left_col_[y] = mask_;
+        } else {
+          left_col_[y] = rnd.Rand16() & mask_;
+        }
+      }
+      Predict(DC_PRED);
+      CheckPrediction(i, &error_count);
+    }
+    ASSERT_EQ(0, error_count);
+  }
+
+  int block_size_;
+  uint16_t *above_row_;
+  uint16_t *left_col_;
+  uint16_t *dst_;
+  uint16_t *ref_dst_;
+  ptrdiff_t stride_;
+  int mask_;
+};
+
+typedef void (*intra_pred_fn_t)(
+      uint16_t *dst, ptrdiff_t stride, const uint16_t *above,
+      const uint16_t *left, int bps);
+typedef std::tr1::tuple<intra_pred_fn_t,
+                        intra_pred_fn_t, int, int> intra_pred_params_t;
+class VP9IntraPredTest
+    : public VP9IntraPredBase,
+      public ::testing::TestWithParam<intra_pred_params_t> {
+
+  virtual void SetUp() {
+    pred_fn_    = GET_PARAM(0);
+    ref_fn_     = GET_PARAM(1);
+    block_size_ = GET_PARAM(2);
+    bit_depth_  = GET_PARAM(3);
+    stride_     = block_size_ * 3;
+    mask_       = (1 << bit_depth_) - 1;
+  }
+
+  virtual void Predict(PREDICTION_MODE mode) {
+    const uint16_t *const_above_row = above_row_;
+    const uint16_t *const_left_col = left_col_;
+    ref_fn_(ref_dst_, stride_, const_above_row, const_left_col, bit_depth_);
+    ASM_REGISTER_STATE_CHECK(pred_fn_(dst_, stride_, const_above_row,
+                                      const_left_col, bit_depth_));
+  }
+  intra_pred_fn_t pred_fn_;
+  intra_pred_fn_t ref_fn_;
+  int bit_depth_;
+};
+
+TEST_P(VP9IntraPredTest, IntraPredTests) {
+  // max block size is 32
+  DECLARE_ALIGNED_ARRAY(16, uint16_t, left_col, 2*32);
+  DECLARE_ALIGNED_ARRAY(16, uint16_t, above_data, 2*32+32);
+  DECLARE_ALIGNED_ARRAY(16, uint16_t, dst, 3 * 32 * 32);
+  DECLARE_ALIGNED_ARRAY(16, uint16_t, ref_dst, 3 * 32 * 32);
+  RunTest(left_col, above_data, dst, ref_dst);
+}
+
+using std::tr1::make_tuple;
+
+#if HAVE_SSE2
+#if CONFIG_VP9_HIGHBITDEPTH
+#if ARCH_X86_64
+INSTANTIATE_TEST_CASE_P(SSE2_TO_C_8, VP9IntraPredTest,
+                        ::testing::Values(
+                            make_tuple(&vp9_high_dc_predictor_32x32_sse2,
+                                       &vp9_high_dc_predictor_32x32_c, 32, 8),
+                            make_tuple(&vp9_high_tm_predictor_16x16_sse2,
+                                       &vp9_high_tm_predictor_16x16_c, 16, 8),
+                            make_tuple(&vp9_high_tm_predictor_32x32_sse2,
+                                       &vp9_high_tm_predictor_32x32_c, 32, 8),
+                            make_tuple(&vp9_high_dc_predictor_4x4_sse,
+                                       &vp9_high_dc_predictor_4x4_c, 4, 8),
+                            make_tuple(&vp9_high_dc_predictor_8x8_sse2,
+                                       &vp9_high_dc_predictor_8x8_c, 8, 8),
+                            make_tuple(&vp9_high_dc_predictor_16x16_sse2,
+                                       &vp9_high_dc_predictor_16x16_c, 16, 8),
+                            make_tuple(&vp9_high_v_predictor_4x4_sse,
+                                       &vp9_high_v_predictor_4x4_c, 4, 8),
+                            make_tuple(&vp9_high_v_predictor_8x8_sse2,
+                                       &vp9_high_v_predictor_8x8_c, 8, 8),
+                            make_tuple(&vp9_high_v_predictor_16x16_sse2,
+                                       &vp9_high_v_predictor_16x16_c, 16, 8),
+                            make_tuple(&vp9_high_v_predictor_32x32_sse2,
+                                       &vp9_high_v_predictor_32x32_c, 32, 8),
+                            make_tuple(&vp9_high_tm_predictor_4x4_sse,
+                                       &vp9_high_tm_predictor_4x4_c, 4, 8),
+                            make_tuple(&vp9_high_tm_predictor_8x8_sse2,
+                                       &vp9_high_tm_predictor_8x8_c, 8, 8)));
+#else
+INSTANTIATE_TEST_CASE_P(SSE2_TO_C_8, VP9IntraPredTest,
+                        ::testing::Values(
+                            make_tuple(&vp9_high_dc_predictor_4x4_sse,
+                                       &vp9_high_dc_predictor_4x4_c, 4, 8),
+                            make_tuple(&vp9_high_dc_predictor_8x8_sse2,
+                                       &vp9_high_dc_predictor_8x8_c, 8, 8),
+                            make_tuple(&vp9_high_dc_predictor_16x16_sse2,
+                                       &vp9_high_dc_predictor_16x16_c, 16, 8),
+                            make_tuple(&vp9_high_v_predictor_4x4_sse,
+                                       &vp9_high_v_predictor_4x4_c, 4, 8),
+                            make_tuple(&vp9_high_v_predictor_8x8_sse2,
+                                       &vp9_high_v_predictor_8x8_c, 8, 8),
+                            make_tuple(&vp9_high_v_predictor_16x16_sse2,
+                                       &vp9_high_v_predictor_16x16_c, 16, 8),
+                            make_tuple(&vp9_high_v_predictor_32x32_sse2,
+                                       &vp9_high_v_predictor_32x32_c, 32, 8),
+                            make_tuple(&vp9_high_tm_predictor_4x4_sse,
+                                       &vp9_high_tm_predictor_4x4_c, 4, 8),
+                            make_tuple(&vp9_high_tm_predictor_8x8_sse2,
+                                       &vp9_high_tm_predictor_8x8_c, 8, 8)));
+#endif
+#if ARCH_X86_64
+INSTANTIATE_TEST_CASE_P(SSE2_TO_C_10, VP9IntraPredTest,
+                        ::testing::Values(
+                            make_tuple(&vp9_high_dc_predictor_32x32_sse2,
+                                       &vp9_high_dc_predictor_32x32_c, 32, 10),
+                            make_tuple(&vp9_high_tm_predictor_16x16_sse2,
+                                       &vp9_high_tm_predictor_16x16_c, 16, 10),
+                            make_tuple(&vp9_high_tm_predictor_32x32_sse2,
+                                       &vp9_high_tm_predictor_32x32_c, 32, 10),
+                            make_tuple(&vp9_high_dc_predictor_4x4_sse,
+                                       &vp9_high_dc_predictor_4x4_c, 4, 10),
+                            make_tuple(&vp9_high_dc_predictor_8x8_sse2,
+                                       &vp9_high_dc_predictor_8x8_c, 8, 10),
+                            make_tuple(&vp9_high_dc_predictor_16x16_sse2,
+                                   &vp9_high_dc_predictor_16x16_c, 16, 10),
+                            make_tuple(&vp9_high_v_predictor_4x4_sse,
+                                       &vp9_high_v_predictor_4x4_c, 4, 10),
+                            make_tuple(&vp9_high_v_predictor_8x8_sse2,
+                                       &vp9_high_v_predictor_8x8_c, 8, 10),
+                            make_tuple(&vp9_high_v_predictor_16x16_sse2,
+                                       &vp9_high_v_predictor_16x16_c, 16, 10),
+                            make_tuple(&vp9_high_v_predictor_32x32_sse2,
+                                       &vp9_high_v_predictor_32x32_c, 32, 10),
+                            make_tuple(&vp9_high_tm_predictor_4x4_sse,
+                                       &vp9_high_tm_predictor_4x4_c, 4, 10),
+                            make_tuple(&vp9_high_tm_predictor_8x8_sse2,
+                                       &vp9_high_tm_predictor_8x8_c, 8, 10)));
+#else
+INSTANTIATE_TEST_CASE_P(SSE2_TO_C_10, VP9IntraPredTest,
+                        ::testing::Values(
+                            make_tuple(&vp9_high_dc_predictor_4x4_sse,
+                                       &vp9_high_dc_predictor_4x4_c, 4, 10),
+                            make_tuple(&vp9_high_dc_predictor_8x8_sse2,
+                                       &vp9_high_dc_predictor_8x8_c, 8, 10),
+                            make_tuple(&vp9_high_dc_predictor_16x16_sse2,
+                                       &vp9_high_dc_predictor_16x16_c, 16, 10),
+                            make_tuple(&vp9_high_v_predictor_4x4_sse,
+                                       &vp9_high_v_predictor_4x4_c, 4, 10),
+                            make_tuple(&vp9_high_v_predictor_8x8_sse2,
+                                       &vp9_high_v_predictor_8x8_c, 8, 10),
+                            make_tuple(&vp9_high_v_predictor_16x16_sse2,
+                                       &vp9_high_v_predictor_16x16_c, 16, 10),
+                            make_tuple(&vp9_high_v_predictor_32x32_sse2,
+                                       &vp9_high_v_predictor_32x32_c, 32, 10),
+                            make_tuple(&vp9_high_tm_predictor_4x4_sse,
+                                   &vp9_high_tm_predictor_4x4_c, 4, 10),
+                            make_tuple(&vp9_high_tm_predictor_8x8_sse2,
+                                       &vp9_high_tm_predictor_8x8_c, 8, 10)));
+#endif
+
+#if ARCH_X86_64
+INSTANTIATE_TEST_CASE_P(SSE2_TO_C_12, VP9IntraPredTest,
+                        ::testing::Values(
+                            make_tuple(&vp9_high_dc_predictor_32x32_sse2,
+                                       &vp9_high_dc_predictor_32x32_c, 32, 12),
+                            make_tuple(&vp9_high_tm_predictor_16x16_sse2,
+                                       &vp9_high_tm_predictor_16x16_c, 16, 12),
+                            make_tuple(&vp9_high_tm_predictor_32x32_sse2,
+                                       &vp9_high_tm_predictor_32x32_c, 32, 12),
+                            make_tuple(&vp9_high_dc_predictor_4x4_sse,
+                                       &vp9_high_dc_predictor_4x4_c, 4, 12),
+                            make_tuple(&vp9_high_dc_predictor_8x8_sse2,
+                                       &vp9_high_dc_predictor_8x8_c, 8, 12),
+                            make_tuple(&vp9_high_dc_predictor_16x16_sse2,
+                                       &vp9_high_dc_predictor_16x16_c, 16, 12),
+                            make_tuple(&vp9_high_v_predictor_4x4_sse,
+                                       &vp9_high_v_predictor_4x4_c, 4, 12),
+                            make_tuple(&vp9_high_v_predictor_8x8_sse2,
+                                       &vp9_high_v_predictor_8x8_c, 8, 12),
+                            make_tuple(&vp9_high_v_predictor_16x16_sse2,
+                                       &vp9_high_v_predictor_16x16_c, 16, 12),
+                            make_tuple(&vp9_high_v_predictor_32x32_sse2,
+                                       &vp9_high_v_predictor_32x32_c, 32, 12),
+                            make_tuple(&vp9_high_tm_predictor_4x4_sse,
+                                       &vp9_high_tm_predictor_4x4_c, 4, 12),
+                            make_tuple(&vp9_high_tm_predictor_8x8_sse2,
+                                       &vp9_high_tm_predictor_8x8_c, 8, 12)));
+#else
+INSTANTIATE_TEST_CASE_P(SSE2_TO_C_12, VP9IntraPredTest,
+                        ::testing::Values(
+                            make_tuple(&vp9_high_dc_predictor_4x4_sse,
+                                       &vp9_high_dc_predictor_4x4_c, 4, 12),
+                            make_tuple(&vp9_high_dc_predictor_8x8_sse2,
+                                       &vp9_high_dc_predictor_8x8_c, 8, 12),
+                            make_tuple(&vp9_high_dc_predictor_16x16_sse2,
+                                       &vp9_high_dc_predictor_16x16_c, 16, 12),
+                            make_tuple(&vp9_high_v_predictor_4x4_sse,
+                                       &vp9_high_v_predictor_4x4_c, 4, 12),
+                            make_tuple(&vp9_high_v_predictor_8x8_sse2,
+                                       &vp9_high_v_predictor_8x8_c, 8, 12),
+                            make_tuple(&vp9_high_v_predictor_16x16_sse2,
+                                       &vp9_high_v_predictor_16x16_c, 16, 12),
+                            make_tuple(&vp9_high_v_predictor_32x32_sse2,
+                                       &vp9_high_v_predictor_32x32_c, 32, 12),
+                            make_tuple(&vp9_high_tm_predictor_4x4_sse,
+                                       &vp9_high_tm_predictor_4x4_c, 4, 12),
+                            make_tuple(&vp9_high_tm_predictor_8x8_sse2,
+                                       &vp9_high_tm_predictor_8x8_c, 8, 12)));
+#endif
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+#endif  // HAVE_SSE2
+}  // namespace

vp9/common/vp9_common.h

@@ -65,6 +65,18 @@ static INLINE int get_unsigned_bits(unsigned int num_values) {
 }
 
 #if CONFIG_VP9_HIGHBITDEPTH
+static INLINE uint16_t clip_pixel_high(int val, int bd) {
+  switch (bd) {
+    case 8:
+    default:
+      return (uint16_t)clamp(val, 0, 255);
+    case 10:
+      return (uint16_t)clamp(val, 0, 1023);
+    case 12:
+      return (uint16_t)clamp(val, 0, 4095);
+  }
+}
+
 #define CONVERT_TO_SHORTPTR(x) ((uint16_t*)(((uintptr_t)x) << 1))
 #define CONVERT_TO_BYTEPTR(x) ((uint8_t*)(((uintptr_t)x) >> 1 ))
 #endif  // CONFIG_VP9_HIGHBITDEPTH

vp9/common/vp9_reconintra.c

@@ -40,11 +40,291 @@ const TX_TYPE intra_mode_to_tx_type_lookup[INTRA_MODES] = {
     type##_predictor(dst, stride, size, above, left); \
   }
 
+#if CONFIG_VP9_HIGHBITDEPTH
+#define intra_pred_high_sized(type, size) \
+  void vp9_high_##type##_predictor_##size##x##size##_c( \
+      uint16_t *dst, ptrdiff_t stride, const uint16_t *above, \
+      const uint16_t *left, int bd) { \
+    high_##type##_predictor(dst, stride, size, above, left, bd); \
+  }
+
+#define intra_pred_allsizes(type) \
+  intra_pred_sized(type, 4) \
+  intra_pred_sized(type, 8) \
+  intra_pred_sized(type, 16) \
+  intra_pred_sized(type, 32) \
+  intra_pred_high_sized(type, 4) \
+  intra_pred_high_sized(type, 8) \
+  intra_pred_high_sized(type, 16) \
+  intra_pred_high_sized(type, 32)
+
+#else
+
 #define intra_pred_allsizes(type) \
   intra_pred_sized(type, 4) \
   intra_pred_sized(type, 8) \
   intra_pred_sized(type, 16) \
   intra_pred_sized(type, 32)
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+
+#if CONFIG_VP9_HIGHBITDEPTH
+static INLINE void high_d207_predictor(uint16_t *dst, ptrdiff_t stride, int bs,
+                                       const uint16_t *above,
+                                       const uint16_t *left, int bd) {
+  (void) above;
+  (void) bd;
+  int r, c;
+
+  // First column.
+  for (r = 0; r < bs - 1; ++r) {
+    dst[r * stride] = ROUND_POWER_OF_TWO(left[r] + left[r + 1], 1);
+  }
+  dst[(bs - 1) * stride] = left[bs - 1];
+  dst++;
+
+  // Second column.
+  for (r = 0; r < bs - 2; ++r) {
+    dst[r * stride] = ROUND_POWER_OF_TWO(left[r] + left[r + 1] * 2 +
+                                         left[r + 2], 2);
+  }
+  dst[(bs - 2) * stride] = ROUND_POWER_OF_TWO(left[bs - 2] +
+                                              left[bs - 1] * 3, 2);
+  dst[(bs - 1) * stride] = left[bs - 1];
+  dst++;
+
+  // Rest of last row.
+  for (c = 0; c < bs - 2; ++c)
+    dst[(bs - 1) * stride + c] = left[bs - 1];
+
+  for (r = bs - 2; r >= 0; --r) {
+    for (c = 0; c < bs - 2; ++c)
+      dst[r * stride + c] = dst[(r + 1) * stride + c - 2];
+  }
+}
+
+static INLINE void high_d63_predictor(uint16_t *dst, ptrdiff_t stride, int bs,
+                                      const uint16_t *above,
+                                      const uint16_t *left, int bd) {
+  (void) left;
+  (void) bd;
+  int r, c;
+  for (r = 0; r < bs; ++r) {
+    for (c = 0; c < bs; ++c) {
+      dst[c] = r & 1 ? ROUND_POWER_OF_TWO(above[r/2 + c] +
+                                          above[r/2 + c + 1] * 2 +
+                                          above[r/2 + c + 2], 2)
+                     : ROUND_POWER_OF_TWO(above[r/2 + c] +
+                                          above[r/2 + c + 1], 1);
+    }
+    dst += stride;
+  }
+}
+
+static INLINE void high_d45_predictor(uint16_t *dst, ptrdiff_t stride, int bs,
+                                      const uint16_t *above,
+                                      const uint16_t *left, int bd) {
+  (void) left;
+  (void) bd;
+  int r, c;
+  for (r = 0; r < bs; ++r) {
+    for (c = 0; c < bs; ++c) {
+      dst[c] = r + c + 2 < bs * 2 ?  ROUND_POWER_OF_TWO(above[r + c] +
+                                                        above[r + c + 1] * 2 +
+                                                        above[r + c + 2], 2)
+                                  : above[bs * 2 - 1];
+    }
+    dst += stride;
+  }
+}
+
+static INLINE void high_d117_predictor(uint16_t *dst, ptrdiff_t stride,
+                                       int bs, const uint16_t *above,
+                                       const uint16_t *left, int bd) {
+  (void) bd;
+  int r, c;
+
+  // first row
+  for (c = 0; c < bs; c++)
+    dst[c] = ROUND_POWER_OF_TWO(above[c - 1] + above[c], 1);
+  dst += stride;
+
+  // second row
+  dst[0] = ROUND_POWER_OF_TWO(left[0] + above[-1] * 2 + above[0], 2);
+  for (c = 1; c < bs; c++)
+    dst[c] = ROUND_POWER_OF_TWO(above[c - 2] + above[c - 1] * 2 + above[c], 2);
+  dst += stride;
+
+  // the rest of first col
+  dst[0] = ROUND_POWER_OF_TWO(above[-1] + left[0] * 2 + left[1], 2);
+  for (r = 3; r < bs; ++r)
+    dst[(r - 2) * stride] = ROUND_POWER_OF_TWO(left[r - 3] + left[r - 2] * 2 +
+                                               left[r - 1], 2);
+
+  // the rest of the block
+  for (r = 2; r < bs; ++r) {
+    for (c = 1; c < bs; c++)
+      dst[c] = dst[-2 * stride + c - 1];
+    dst += stride;
+  }
+}
+
+static INLINE void high_d135_predictor(uint16_t *dst, ptrdiff_t stride, int bs,
+                                       const uint16_t *above,
+                                       const uint16_t *left, int bd) {
+  (void) bd;
+  int r, c;
+  dst[0] = ROUND_POWER_OF_TWO(left[0] + above[-1] * 2 + above[0], 2);
+  for (c = 1; c < bs; c++)
+    dst[c] = ROUND_POWER_OF_TWO(above[c - 2] + above[c - 1] * 2 + above[c], 2);
+
+  dst[stride] = ROUND_POWER_OF_TWO(above[-1] + left[0] * 2 + left[1], 2);
+  for (r = 2; r < bs; ++r)
+    dst[r * stride] = ROUND_POWER_OF_TWO(left[r - 2] + left[r - 1] * 2 +
+                                         left[r], 2);
+
+  dst += stride;
+  for (r = 1; r < bs; ++r) {
+    for (c = 1; c < bs; c++)
+      dst[c] = dst[-stride + c - 1];
+    dst += stride;
+  }
+}
+
+static INLINE void high_d153_predictor(uint16_t *dst, ptrdiff_t stride, int bs,
+                                       const uint16_t *above,
+                                       const uint16_t *left, int bd) {
+  (void) bd;
+  int r, c;
+  dst[0] = ROUND_POWER_OF_TWO(above[-1] + left[0], 1);
+  for (r = 1; r < bs; r++)
+    dst[r * stride] = ROUND_POWER_OF_TWO(left[r - 1] + left[r], 1);
+  dst++;
+
+  dst[0] = ROUND_POWER_OF_TWO(left[0] + above[-1] * 2 + above[0], 2);
+  dst[stride] = ROUND_POWER_OF_TWO(above[-1] + left[0] * 2 + left[1], 2);
+  for (r = 2; r < bs; r++)
+    dst[r * stride] = ROUND_POWER_OF_TWO(left[r - 2] + left[r - 1] * 2 +
+                                         left[r], 2);
+  dst++;
+
+  for (c = 0; c < bs - 2; c++)
+    dst[c] = ROUND_POWER_OF_TWO(above[c - 1] + above[c] * 2 + above[c + 1], 2);
+  dst += stride;
+
+  for (r = 1; r < bs; ++r) {
+    for (c = 0; c < bs - 2; c++)
+      dst[c] = dst[-stride + c - 2];
+    dst += stride;
+  }
+}
+
+static INLINE void high_v_predictor(uint16_t *dst, ptrdiff_t stride, int bs,
+                                    const uint16_t *above,
+                                    const uint16_t *left, int bd) {
+  (void) left;
+  (void) bd;
+  int r;
+
+  for (r = 0; r < bs; r++) {
+    vpx_memcpy(dst, above, bs * sizeof(uint16_t));
+    dst += stride;
+  }
+}
+
+static INLINE void high_h_predictor(uint16_t *dst, ptrdiff_t stride, int bs,
+                                    const uint16_t *above, const uint16_t *left,
+                                    int bd) {
+  (void) above;
+  (void) bd;
+  int r;
+
+  for (r = 0; r < bs; r++) {
+    vpx_memset16(dst, left[r], bs);
+    dst += stride;
+  }
+}
+
+static INLINE void high_tm_predictor(uint16_t *dst, ptrdiff_t stride, int bs,
+                                     const uint16_t *above,
+                                     const uint16_t *left, int bd) {
+  (void) bd;
+  int r, c;
+  int ytop_left = above[-1];
+
+  for (r = 0; r < bs; r++) {
+    for (c = 0; c < bs; c++)
+      dst[c] = clip_pixel_high(left[r] + above[c] - ytop_left, bd);
+    dst += stride;
+  }
+}
+
+static INLINE void high_dc_128_predictor(uint16_t *dst, ptrdiff_t stride,
+                                         int bs, const uint16_t *above,
+                                         const uint16_t *left, int bd) {
+  (void) above;
+  (void) left;
+  int r;
+
+  for (r = 0; r < bs; r++) {
+    vpx_memset16(dst, 128 << (bd - 8), bs);
+    dst += stride;
+  }
+}
+
+static INLINE void high_dc_left_predictor(uint16_t *dst, ptrdiff_t stride,
+                                          int bs, const uint16_t *above,
+                                          const uint16_t *left, int bd) {
+  (void) above;
+  (void) bd;
+  int i, r, expected_dc, sum = 0;
+
+  for (i = 0; i < bs; i++)
+    sum += left[i];
+  expected_dc = (sum + (bs >> 1)) / bs;
+
+  for (r = 0; r < bs; r++) {
+    vpx_memset16(dst, expected_dc, bs);
+    dst += stride;
+  }
+}
+
+static INLINE void high_dc_top_predictor(uint16_t *dst, ptrdiff_t stride,
+                                         int bs, const uint16_t *above,
+                                         const uint16_t *left, int bd) {
+  (void) left;
+  (void) bd;
+  int i, r, expected_dc, sum = 0;
+
+  for (i = 0; i < bs; i++)
+    sum += above[i];
+  expected_dc = (sum + (bs >> 1)) / bs;
+
+  for (r = 0; r < bs; r++) {
+    vpx_memset16(dst, expected_dc, bs);
+    dst += stride;
+  }
+}
+
+static INLINE void high_dc_predictor(uint16_t *dst, ptrdiff_t stride,
+                                     int bs, const uint16_t *above,
+                                     const uint16_t *left, int bd) {
+  (void) bd;
+  int i, r, expected_dc, sum = 0;
+  const int count = 2 * bs;
+
+  for (i = 0; i < bs; i++) {
+    sum += above[i];
+    sum += left[i];
+  }
+