Lib/VPX/intrapred.c

/*
 *  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.
 */

#define INLINE __inline

#include <stdint.h>
#include <string.h>
#include "vpx_dsp_common.h"
#include "vpx_dsp_rtcd.h"

#define dst(x, y) dst[(x) + (y)*stride]
#define AVG3(a, b, c) (((a) + 2 * (b) + (c) + 2) >> 2)
#define AVG2(a, b) (((a) + (b) + 1) >> 1)

static INLINE void d207_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
                                  const uint8_t *above, const uint8_t *left) {
  int r, c;
  (void)above;
  // first column
  for (r = 0; r < bs - 1; ++r) dst[r * stride] = AVG2(left[r], left[r + 1]);
  dst[(bs - 1) * stride] = left[bs - 1];
  dst++;

  // second column
  for (r = 0; r < bs - 2; ++r)
    dst[r * stride] = AVG3(left[r], left[r + 1], left[r + 2]);
  dst[(bs - 2) * stride] = AVG3(left[bs - 2], left[bs - 1], left[bs - 1]);
  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 d63_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
                                 const uint8_t *above, const uint8_t *left) {
  int r, c;
  int size;
  (void)left;
  for (c = 0; c < bs; ++c) {
    dst[c] = AVG2(above[c], above[c + 1]);
    dst[stride + c] = AVG3(above[c], above[c + 1], above[c + 2]);
  }
  for (r = 2, size = bs - 2; r < bs; r += 2, --size) {
    memcpy(dst + (r + 0) * stride, dst + (r >> 1), size);
    memset(dst + (r + 0) * stride + size, above[bs - 1], bs - size);
    memcpy(dst + (r + 1) * stride, dst + stride + (r >> 1), size);
    memset(dst + (r + 1) * stride + size, above[bs - 1], bs - size);
  }
}

static INLINE void d45_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
                                 const uint8_t *above, const uint8_t *left) {
  const uint8_t above_right = above[bs - 1];
  const uint8_t *const dst_row0 = dst;
  int x, size;
  (void)left;

  for (x = 0; x < bs - 1; ++x) {
    dst[x] = AVG3(above[x], above[x + 1], above[x + 2]);
  }
  dst[bs - 1] = above_right;
  dst += stride;
  for (x = 1, size = bs - 2; x < bs; ++x, --size) {
    memcpy(dst, dst_row0 + x, size);
    memset(dst + size, above_right, x + 1);
    dst += stride;
  }
}

static INLINE void d117_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
                                  const uint8_t *above, const uint8_t *left) {
  int r, c;

  // first row
  for (c = 0; c < bs; c++) dst[c] = AVG2(above[c - 1], above[c]);
  dst += stride;

  // second row
  dst[0] = AVG3(left[0], above[-1], above[0]);
  for (c = 1; c < bs; c++) dst[c] = AVG3(above[c - 2], above[c - 1], above[c]);
  dst += stride;

  // the rest of first col
  dst[0] = AVG3(above[-1], left[0], left[1]);
  for (r = 3; r < bs; ++r)
    dst[(r - 2) * stride] = AVG3(left[r - 3], left[r - 2], left[r - 1]);

  // 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 d135_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
                                  const uint8_t *above, const uint8_t *left) {
  int i;
#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ > 7
  // silence a spurious -Warray-bounds warning, possibly related to:
  // https://gcc.gnu.org/bugzilla/show_bug.cgi?id=56273
  uint8_t border[69];
#else
  uint8_t border[32 + 32 - 1];  // outer border from bottom-left to top-right
#endif

  // dst(bs, bs - 2)[0], i.e., border starting at bottom-left
  for (i = 0; i < bs - 2; ++i) {
    border[i] = AVG3(left[bs - 3 - i], left[bs - 2 - i], left[bs - 1 - i]);
  }
  border[bs - 2] = AVG3(above[-1], left[0], left[1]);
  border[bs - 1] = AVG3(left[0], above[-1], above[0]);
  border[bs - 0] = AVG3(above[-1], above[0], above[1]);
  // dst[0][2, size), i.e., remaining top border ascending
  for (i = 0; i < bs - 2; ++i) {
    border[bs + 1 + i] = AVG3(above[i], above[i + 1], above[i + 2]);
  }

  for (i = 0; i < bs - 1; ++i) {
    memcpy(dst + i * stride, border + bs - 1 - i, bs);
  }
}

static INLINE void d153_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
                                  const uint8_t *above, const uint8_t *left) {
  int r, c;
  dst[0] = AVG2(above[-1], left[0]);
  for (r = 1; r < bs; r++) dst[r * stride] = AVG2(left[r - 1], left[r]);
  dst++;

  dst[0] = AVG3(left[0], above[-1], above[0]);
  dst[stride] = AVG3(above[-1], left[0], left[1]);
  for (r = 2; r < bs; r++)
    dst[r * stride] = AVG3(left[r - 2], left[r - 1], left[r]);
  dst++;

  for (c = 0; c < bs - 2; c++)
    dst[c] = AVG3(above[c - 1], above[c], above[c + 1]);
  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 v_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
                               const uint8_t *above, const uint8_t *left) {
  int r;
  (void)left;

  for (r = 0; r < bs; r++) {
    memcpy(dst, above, bs);
    dst += stride;
  }
}

static INLINE void h_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
                               const uint8_t *above, const uint8_t *left) {
  int r;
  (void)above;

  for (r = 0; r < bs; r++) {
    memset(dst, left[r], bs);
    dst += stride;
  }
}

static INLINE void tm_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
                                const uint8_t *above, const uint8_t *left) {
  int r, c;
  int ytop_left = above[-1];

  for (r = 0; r < bs; r++) {
    for (c = 0; c < bs; c++)
      dst[c] = (uint8_t)clip_pixel(left[r] + above[c] - ytop_left);
    dst += stride;
  }
}

static INLINE void dc_128_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
                                    const uint8_t *above, const uint8_t *left) {
  int r;
  (void)above;
  (void)left;

  for (r = 0; r < bs; r++) {
    memset(dst, 128, bs);
    dst += stride;
  }
}

static INLINE void dc_left_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
                                     const uint8_t *above,
                                     const uint8_t *left) {
  int i, r, expected_dc, sum = 0;
  (void)above;

  for (i = 0; i < bs; i++) sum += left[i];
  expected_dc = (sum + (bs >> 1)) / bs;

  for (r = 0; r < bs; r++) {
    memset(dst, expected_dc, bs);
    dst += stride;
  }
}

static INLINE void dc_top_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
                                    const uint8_t *above, const uint8_t *left) {
  int i, r, expected_dc, sum = 0;
  (void)left;

  for (i = 0; i < bs; i++) sum += above[i];
  expected_dc = (sum + (bs >> 1)) / bs;

  for (r = 0; r < bs; r++) {
    memset(dst, expected_dc, bs);
    dst += stride;
  }
}

static INLINE void dc_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
                                const uint8_t *above, const uint8_t *left) {
  int i, r, expected_dc, sum = 0;
  const int count = 2 * bs;

  for (i = 0; i < bs; i++) {
    sum += above[i];
    sum += left[i];
  }

  expected_dc = (sum + (count >> 1)) / count;

  for (r = 0; r < bs; r++) {
    memset(dst, expected_dc, bs);
    dst += stride;
  }
}

void eb_vp9_he_predictor_4x4_c(uint8_t *dst, ptrdiff_t stride,
                            const uint8_t *above, const uint8_t *left) {
  const int H = above[-1];
  const int I = left[0];
  const int J = left[1];
  const int K = left[2];
  const int L = left[3];

  memset(dst + stride * 0, AVG3(H, I, J), 4);
  memset(dst + stride * 1, AVG3(I, J, K), 4);
  memset(dst + stride * 2, AVG3(J, K, L), 4);
  memset(dst + stride * 3, AVG3(K, L, L), 4);
}

void eb_vp9_ve_predictor_4x4_c(uint8_t *dst, ptrdiff_t stride,
                            const uint8_t *above, const uint8_t *left) {
  const int H = above[-1];
  const int I = above[0];
  const int J = above[1];
  const int K = above[2];
  const int L = above[3];
  const int M = above[4];
  (void)left;

  dst[0] = (uint8_t)AVG3(H, I, J);
  dst[1] = (uint8_t)AVG3(I, J, K);
  dst[2] = (uint8_t)AVG3(J, K, L);
  dst[3] = (uint8_t)AVG3(K, L, M);
  memcpy(dst + stride * 1, dst, 4);
  memcpy(dst + stride * 2, dst, 4);
  memcpy(dst + stride * 3, dst, 4);
}

void eb_vp9_d207_predictor_4x4_c(uint8_t *dst, ptrdiff_t stride,
                              const uint8_t *above, const uint8_t *left) {
  const int I = left[0];
  const int J = left[1];
  const int K = left[2];
  const int L = left[3];
  (void)above;
  dst(0, 0) = (uint8_t)AVG2(I, J);
  dst(2, 0) = dst(0, 1) = (uint8_t)AVG2(J, K);
  dst(2, 1) = dst(0, 2) = (uint8_t)AVG2(K, L);
  dst(1, 0) = (uint8_t)AVG3(I, J, K);
  dst(3, 0) = dst(1, 1) = (uint8_t)AVG3(J, K, L);
  dst(3, 1) = dst(1, 2) = (uint8_t)AVG3(K, L, L);
  dst(3, 2) = dst(2, 2) = dst(0, 3) = dst(1, 3) = dst(2, 3) = dst(3, 3) = (uint8_t)L;
}

void eb_vp9_d63_predictor_4x4_c(uint8_t *dst, ptrdiff_t stride,
                             const uint8_t *above, const uint8_t *left) {
  const int A = above[0];
  const int B = above[1];
  const int C = above[2];
  const int D = above[3];
  const int E = above[4];
  const int F = above[5];
  const int G = above[6];
  (void)left;
  dst(0, 0) = (uint8_t)AVG2(A, B);
  dst(1, 0) = dst(0, 2) = (uint8_t)AVG2(B, C);
  dst(2, 0) = dst(1, 2) = (uint8_t)AVG2(C, D);
  dst(3, 0) = dst(2, 2) = (uint8_t)AVG2(D, E);
  dst(3, 2) = (uint8_t)AVG2(E, F);  // differs from vp8

  dst(0, 1) = (uint8_t)AVG3(A, B, C);
  dst(1, 1) = dst(0, 3) = (uint8_t)AVG3(B, C, D);
  dst(2, 1) = dst(1, 3) = (uint8_t)AVG3(C, D, E);
  dst(3, 1) = dst(2, 3) = (uint8_t)AVG3(D, E, F);
  dst(3, 3) = (uint8_t)AVG3(E, F, G);  // differs from vp8
}

void eb_vp9_d63e_predictor_4x4_c(uint8_t *dst, ptrdiff_t stride,
                              const uint8_t *above, const uint8_t *left) {
  const int A = above[0];
  const int B = above[1];
  const int C = above[2];
  const int D = above[3];
  const int E = above[4];
  const int F = above[5];
  const int G = above[6];
  const int H = above[7];
  (void)left;
  dst(0, 0) = (uint8_t)AVG2(A, B);
  dst(1, 0) = dst(0, 2) = (uint8_t)AVG2(B, C);
  dst(2, 0) = dst(1, 2) = (uint8_t)AVG2(C, D);
  dst(3, 0) = dst(2, 2) = (uint8_t)AVG2(D, E);
  dst(3, 2) = (uint8_t)AVG3(E, F, G);

  dst(0, 1) = (uint8_t)AVG3(A, B, C);
  dst(1, 1) = dst(0, 3) = (uint8_t)AVG3(B, C, D);
  dst(2, 1) = dst(1, 3) = (uint8_t)AVG3(C, D, E);
  dst(3, 1) = dst(2, 3) = (uint8_t)AVG3(D, E, F);
  dst(3, 3) = (uint8_t)AVG3(F, G, H);
}

void eb_vp9_d45_predictor_4x4_c(uint8_t *dst, ptrdiff_t stride,
                             const uint8_t *above, const uint8_t *left) {
  const int A = above[0];
  const int B = above[1];
  const int C = above[2];
  const int D = above[3];
  const int E = above[4];
  const int F = above[5];
  const int G = above[6];
  const int H = above[7];
  (void)stride;
  (void)left;
  dst(0, 0) = (uint8_t)AVG3(A, B, C);
  dst(1, 0) = dst(0, 1) = (uint8_t)AVG3(B, C, D);
  dst(2, 0) = dst(1, 1) = dst(0, 2) = (uint8_t)AVG3(C, D, E);
  dst(3, 0) = dst(2, 1) = dst(1, 2) = dst(0, 3) = (uint8_t)AVG3(D, E, F);
  dst(3, 1) = dst(2, 2) = dst(1, 3) = (uint8_t)AVG3(E, F, G);
  dst(3, 2) = dst(2, 3) = (uint8_t)AVG3(F, G, H);
  dst(3, 3) = (uint8_t)H;  // differs from vp8
}

void eb_vp9_d45e_predictor_4x4_c(uint8_t *dst, ptrdiff_t stride,
                              const uint8_t *above, const uint8_t *left) {
  const int A = above[0];
  const int B = above[1];
  const int C = above[2];
  const int D = above[3];
  const int E = above[4];
  const int F = above[5];
  const int G = above[6];
  const int H = above[7];
  (void)stride;
  (void)left;
  dst(0, 0) = (uint8_t)AVG3(A, B, C);
  dst(1, 0) = dst(0, 1) = (uint8_t)AVG3(B, C, D);
  dst(2, 0) = dst(1, 1) = dst(0, 2) = (uint8_t)AVG3(C, D, E);
  dst(3, 0) = dst(2, 1) = dst(1, 2) = dst(0, 3) = (uint8_t)AVG3(D, E, F);
  dst(3, 1) = dst(2, 2) = dst(1, 3) = (uint8_t)AVG3(E, F, G);
  dst(3, 2) = dst(2, 3) = (uint8_t)AVG3(F, G, H);
  dst(3, 3) = (uint8_t)AVG3(G, H, H);
}

void eb_vp9_d117_predictor_4x4_c(uint8_t *dst, ptrdiff_t stride,
                              const uint8_t *above, const uint8_t *left) {
  const int I = left[0];
  const int J = left[1];
  const int K = left[2];
  const int X = above[-1];
  const int A = above[0];
  const int B = above[1];
  const int C = above[2];
  const int D = above[3];
  dst(0, 0) = dst(1, 2) = (uint8_t)AVG2(X, A);
  dst(1, 0) = dst(2, 2) = (uint8_t)AVG2(A, B);
  dst(2, 0) = dst(3, 2) = (uint8_t)AVG2(B, C);
  dst(3, 0) = (uint8_t)AVG2(C, D);

  dst(0, 3) = (uint8_t)AVG3(K, J, I);
  dst(0, 2) = (uint8_t)AVG3(J, I, X);
  dst(0, 1) = dst(1, 3) = (uint8_t)AVG3(I, X, A);
  dst(1, 1) = dst(2, 3) = (uint8_t)AVG3(X, A, B);
  dst(2, 1) = dst(3, 3) = (uint8_t)AVG3(A, B, C);
  dst(3, 1) = (uint8_t)AVG3(B, C, D);
}

void eb_vp9_d135_predictor_4x4_c(uint8_t *dst, ptrdiff_t stride,
                              const uint8_t *above, const uint8_t *left) {
  const int I = left[0];
  const int J = left[1];
  const int K = left[2];
  const int L = left[3];
  const int X = above[-1];
  const int A = above[0];
  const int B = above[1];
  const int C = above[2];
  const int D = above[3];
  (void)stride;
  dst(0, 3) = (uint8_t)AVG3(J, K, L);
  dst(1, 3) = dst(0, 2) = (uint8_t)AVG3(I, J, K);
  dst(2, 3) = dst(1, 2) = dst(0, 1) = (uint8_t)AVG3(X, I, J);
  dst(3, 3) = dst(2, 2) = dst(1, 1) = dst(0, 0) = (uint8_t)AVG3(A, X, I);
  dst(3, 2) = dst(2, 1) = dst(1, 0) = (uint8_t)AVG3(B, A, X);
  dst(3, 1) = dst(2, 0) = (uint8_t)AVG3(C, B, A);
  dst(3, 0) = (uint8_t)AVG3(D, C, B);
}

void eb_vp9_d153_predictor_4x4_c(uint8_t *dst, ptrdiff_t stride,
                              const uint8_t *above, const uint8_t *left) {
  const int I = left[0];
  const int J = left[1];
  const int K = left[2];
  const int L = left[3];
  const int X = above[-1];
  const int A = above[0];
  const int B = above[1];
  const int C = above[2];

  dst(0, 0) = dst(2, 1) = (uint8_t)AVG2(I, X);
  dst(0, 1) = dst(2, 2) = (uint8_t)AVG2(J, I);
  dst(0, 2) = dst(2, 3) = (uint8_t)AVG2(K, J);
  dst(0, 3) = (uint8_t)AVG2(L, K);

  dst(3, 0) = (uint8_t)AVG3(A, B, C);
  dst(2, 0) = (uint8_t)AVG3(X, A, B);
  dst(1, 0) = dst(3, 1) = (uint8_t)AVG3(I, X, A);
  dst(1, 1) = dst(3, 2) = (uint8_t)AVG3(J, I, X);
  dst(1, 2) = dst(3, 3) = (uint8_t)AVG3(K, J, I);
  dst(1, 3) = (uint8_t)AVG3(L, K, J);
}

#if CONFIG_VP9_HIGHBITDEPTH
static INLINE void highbd_d207_predictor(uint16_t *dst, ptrdiff_t stride,
                                         int bs, const uint16_t *above,
                                         const uint16_t *left, int bd) {
  int r, c;
  (void)above;
  (void)bd;

  // First column.
  for (r = 0; r < bs - 1; ++r) {
    dst[r * stride] = AVG2(left[r], left[r + 1]);
  }
  dst[(bs - 1) * stride] = left[bs - 1];
  dst++;

  // Second column.
  for (r = 0; r < bs - 2; ++r) {
    dst[r * stride] = AVG3(left[r], left[r + 1], left[r + 2]);
  }
  dst[(bs - 2) * stride] = AVG3(left[bs - 2], left[bs - 1], left[bs - 1]);
  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 highbd_d63_predictor(uint16_t *dst, ptrdiff_t stride, int bs,
                                        const uint16_t *above,
                                        const uint16_t *left, int bd) {
  int r, c;
  int size;
  (void)left;
  (void)bd;
  for (c = 0; c < bs; ++c) {
    dst[c] = AVG2(above[c], above[c + 1]);
    dst[stride + c] = AVG3(above[c], above[c + 1], above[c + 2]);
  }
  for (r = 2, size = bs - 2; r < bs; r += 2, --size) {
    memcpy(dst + (r + 0) * stride, dst + (r >> 1), size * sizeof(*dst));
    vpx_memset16(dst + (r + 0) * stride + size, above[bs - 1], bs - size);
    memcpy(dst + (r + 1) * stride, dst + stride + (r >> 1),
           size * sizeof(*dst));
    vpx_memset16(dst + (r + 1) * stride + size, above[bs - 1], bs - size);
  }
}

static INLINE void highbd_d45_predictor(uint16_t *dst, ptrdiff_t stride, int bs,
                                        const uint16_t *above,
                                        const uint16_t *left, int bd) {
  const uint16_t above_right = above[bs - 1];
  const uint16_t *const dst_row0 = dst;
  int x, size;
  (void)left;
  (void)bd;

  for (x = 0; x < bs - 1; ++x) {
    dst[x] = AVG3(above[x], above[x + 1], above[x + 2]);
  }
  dst[bs - 1] = above_right;
  dst += stride;
  for (x = 1, size = bs - 2; x < bs; ++x, --size) {
    memcpy(dst, dst_row0 + x, size * sizeof(*dst));
    vpx_memset16(dst + size, above_right, x + 1);
    dst += stride;
  }
}

static INLINE void highbd_d117_predictor(uint16_t *dst, ptrdiff_t stride,
                                         int bs, const uint16_t *above,
                                         const uint16_t *left, int bd) {
  int r, c;
  (void)bd;

  // first row
  for (c = 0; c < bs; c++) dst[c] = AVG2(above[c - 1], above[c]);
  dst += stride;

  // second row
  dst[0] = AVG3(left[0], above[-1], above[0]);
  for (c = 1; c < bs; c++) dst[c] = AVG3(above[c - 2], above[c - 1], above[c]);
  dst += stride;

  // the rest of first col
  dst[0] = AVG3(above[-1], left[0], left[1]);
  for (r = 3; r < bs; ++r)
    dst[(r - 2) * stride] = AVG3(left[r - 3], left[r - 2], left[r - 1]);

  // 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 highbd_d135_predictor(uint16_t *dst, ptrdiff_t stride,
                                         int bs, const uint16_t *above,
                                         const uint16_t *left, int bd) {
  int i;
#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ > 7
  // silence a spurious -Warray-bounds warning, possibly related to:
  // https://gcc.gnu.org/bugzilla/show_bug.cgi?id=56273
  uint16_t border[69];
#else
  uint16_t border[32 + 32 - 1];  // outer border from bottom-left to top-right
#endif
  (void)bd;

  // dst(bs, bs - 2)[0], i.e., border starting at bottom-left
  for (i = 0; i < bs - 2; ++i) {
    border[i] = AVG3(left[bs - 3 - i], left[bs - 2 - i], left[bs - 1 - i]);
  }
  border[bs - 2] = AVG3(above[-1], left[0], left[1]);
  border[bs - 1] = AVG3(left[0], above[-1], above[0]);
  border[bs - 0] = AVG3(above[-1], above[0], above[1]);
  // dst[0][2, size), i.e., remaining top border ascending
  for (i = 0; i < bs - 2; ++i) {
    border[bs + 1 + i] = AVG3(above[i], above[i + 1], above[i + 2]);
  }

  for (i = 0; i < bs; ++i) {
    memcpy(dst + i * stride, border + bs - 1 - i, bs * sizeof(dst[0]));
  }
}

static INLINE void highbd_d153_predictor(uint16_t *dst, ptrdiff_t stride,
                                         int bs, const uint16_t *above,
                                         const uint16_t *left, int bd) {
  int r, c;
  (void)bd;
  dst[0] = AVG2(above[-1], left[0]);
  for (r = 1; r < bs; r++) dst[r * stride] = AVG2(left[r - 1], left[r]);
  dst++;

  dst[0] = AVG3(left[0], above[-1], above[0]);
  dst[stride] = AVG3(above[-1], left[0], left[1]);
  for (r = 2; r < bs; r++)
    dst[r * stride] = AVG3(left[r - 2], left[r - 1], left[r]);
  dst++;

  for (c = 0; c < bs - 2; c++)
    dst[c] = AVG3(above[c - 1], above[c], above[c + 1]);
  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 highbd_v_predictor(uint16_t *dst, ptrdiff_t stride, int bs,
                                      const uint16_t *above,
                                      const uint16_t *left, int bd) {
  int r;
  (void)left;
  (void)bd;
  for (r = 0; r < bs; r++) {
    memcpy(dst, above, bs * sizeof(uint16_t));
    dst += stride;
  }
}

static INLINE void highbd_h_predictor(uint16_t *dst, ptrdiff_t stride, int bs,
                                      const uint16_t *above,
                                      const uint16_t *left, int bd) {
  int r;
  (void)above;
  (void)bd;
  for (r = 0; r < bs; r++) {
    vpx_memset16(dst, left[r], bs);
    dst += stride;
  }
}

static INLINE void highbd_tm_predictor(uint16_t *dst, ptrdiff_t stride, int bs,
                                       const uint16_t *above,
                                       const uint16_t *left, int bd) {
  int r, c;
  int ytop_left = above[-1];
  (void)bd;

  for (r = 0; r < bs; r++) {
    for (c = 0; c < bs; c++)
      dst[c] = clip_pixel_highbd(left[r] + above[c] - ytop_left, bd);
    dst += stride;
  }
}

static INLINE void highbd_dc_128_predictor(uint16_t *dst, ptrdiff_t stride,
                                           int bs, const uint16_t *above,
                                           const uint16_t *left, int bd) {
  int r;
  (void)above;
  (void)left;

  for (r = 0; r < bs; r++) {
    vpx_memset16(dst, 128 << (bd - 8), bs);
    dst += stride;
  }
}

static INLINE void highbd_dc_left_predictor(uint16_t *dst, ptrdiff_t stride,
                                            int bs, const uint16_t *above,
                                            const uint16_t *left, int bd) {
  int i, r, expected_dc, sum = 0;
  (void)above;
  (void)bd;

  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 highbd_dc_top_predictor(uint16_t *dst, ptrdiff_t stride,
                                           int bs, const uint16_t *above,
                                           const uint16_t *left, int bd) {
  int i, r, expected_dc, sum = 0;
  (void)left;
  (void)bd;

  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 highbd_dc_predictor(uint16_t *dst, ptrdiff_t stride, int bs,
                                       const uint16_t *above,
                                       const uint16_t *left, int bd) {
  int i, r, expected_dc, sum = 0;
  const int count = 2 * bs;
  (void)bd;

  for (i = 0; i < bs; i++) {
    sum += above[i];
    sum += left[i];
  }

  expected_dc = (sum + (count >> 1)) / count;

  for (r = 0; r < bs; r++) {
    vpx_memset16(dst, expected_dc, bs);
    dst += stride;
  }
}

void vpx_highbd_d207_predictor_4x4_c(uint16_t *dst, ptrdiff_t stride,
                                     const uint16_t *above,
                                     const uint16_t *left, int bd) {
  const int I = left[0];
  const int J = left[1];
  const int K = left[2];
  const int L = left[3];
  (void)above;
  (void)bd;
  dst(0, 0) = AVG2(I, J);
  dst(2, 0) = dst(0, 1) = AVG2(J, K);
  dst(2, 1) = dst(0, 2) = AVG2(K, L);
  dst(1, 0) = AVG3(I, J, K);
  dst(3, 0) = dst(1, 1) = AVG3(J, K, L);
  dst(3, 1) = dst(1, 2) = AVG3(K, L, L);
  dst(3, 2) = dst(2, 2) = dst(0, 3) = dst(1, 3) = dst(2, 3) = dst(3, 3) = L;
}

void vpx_highbd_d63_predictor_4x4_c(uint16_t *dst, ptrdiff_t stride,
                                    const uint16_t *above, const uint16_t *left,
                                    int bd) {
  const int A = above[0];
  const int B = above[1];
  const int C = above[2];
  const int D = above[3];
  const int E = above[4];
  const int F = above[5];
  const int G = above[6];
  (void)left;
  (void)bd;
  dst(0, 0) = AVG2(A, B);
  dst(1, 0) = dst(0, 2) = AVG2(B, C);
  dst(2, 0) = dst(1, 2) = AVG2(C, D);
  dst(3, 0) = dst(2, 2) = AVG2(D, E);
  dst(3, 2) = AVG2(E, F);  // differs from vp8

  dst(0, 1) = AVG3(A, B, C);
  dst(1, 1) = dst(0, 3) = AVG3(B, C, D);
  dst(2, 1) = dst(1, 3) = AVG3(C, D, E);
  dst(3, 1) = dst(2, 3) = AVG3(D, E, F);
  dst(3, 3) = AVG3(E, F, G);  // differs from vp8
}

void vpx_highbd_d45_predictor_4x4_c(uint16_t *dst, ptrdiff_t stride,
                                    const uint16_t *above, const uint16_t *left,
                                    int bd) {
  const int A = above[0];
  const int B = above[1];
  const int C = above[2];
  const int D = above[3];
  const int E = above[4];
  const int F = above[5];
  const int G = above[6];
  const int H = above[7];
  (void)left;
  (void)bd;
  dst(0, 0) = AVG3(A, B, C);
  dst(1, 0) = dst(0, 1) = AVG3(B, C, D);
  dst(2, 0) = dst(1, 1) = dst(0, 2) = AVG3(C, D, E);
  dst(3, 0) = dst(2, 1) = dst(1, 2) = dst(0, 3) = AVG3(D, E, F);
  dst(3, 1) = dst(2, 2) = dst(1, 3) = AVG3(E, F, G);
  dst(3, 2) = dst(2, 3) = AVG3(F, G, H);
  dst(3, 3) = H;  // differs from vp8
}

void vpx_highbd_d117_predictor_4x4_c(uint16_t *dst, ptrdiff_t stride,
                                     const uint16_t *above,
                                     const uint16_t *left, int bd) {
  const int I = left[0];
  const int J = left[1];
  const int K = left[2];
  const int X = above[-1];
  const int A = above[0];
  const int B = above[1];
  const int C = above[2];
  const int D = above[3];
  (void)bd;
  dst(0, 0) = dst(1, 2) = AVG2(X, A);
  dst(1, 0) = dst(2, 2) = AVG2(A, B);
  dst(2, 0) = dst(3, 2) = AVG2(B, C);
  dst(3, 0) = AVG2(C, D);

  dst(0, 3) = AVG3(K, J, I);
  dst(0, 2) = AVG3(J, I, X);
  dst(0, 1) = dst(1, 3) = AVG3(I, X, A);
  dst(1, 1) = dst(2, 3) = AVG3(X, A, B);
  dst(2, 1) = dst(3, 3) = AVG3(A, B, C);
  dst(3, 1) = AVG3(B, C, D);
}

void vpx_highbd_d135_predictor_4x4_c(uint16_t *dst, ptrdiff_t stride,
                                     const uint16_t *above,
                                     const uint16_t *left, int bd) {
  const int I = left[0];
  const int J = left[1];
  const int K = left[2];
  const int L = left[3];
  const int X = above[-1];
  const int A = above[0];
  const int B = above[1];
  const int C = above[2];
  const int D = above[3];
  (void)bd;
  dst(0, 3) = AVG3(J, K, L);
  dst(1, 3) = dst(0, 2) = AVG3(I, J, K);
  dst(2, 3) = dst(1, 2) = dst(0, 1) = AVG3(X, I, J);
  dst(3, 3) = dst(2, 2) = dst(1, 1) = dst(0, 0) = AVG3(A, X, I);
  dst(3, 2) = dst(2, 1) = dst(1, 0) = AVG3(B, A, X);
  dst(3, 1) = dst(2, 0) = AVG3(C, B, A);
  dst(3, 0) = AVG3(D, C, B);
}

void vpx_highbd_d153_predictor_4x4_c(uint16_t *dst, ptrdiff_t stride,
                                     const uint16_t *above,
                                     const uint16_t *left, int bd) {
  const int I = left[0];
  const int J = left[1];
  const int K = left[2];
  const int L = left[3];
  const int X = above[-1];
  const int A = above[0];
  const int B = above[1];
  const int C = above[2];
  (void)bd;

  dst(0, 0) = dst(2, 1) = AVG2(I, X);
  dst(0, 1) = dst(2, 2) = AVG2(J, I);
  dst(0, 2) = dst(2, 3) = AVG2(K, J);
  dst(0, 3) = AVG2(L, K);

  dst(3, 0) = AVG3(A, B, C);
  dst(2, 0) = AVG3(X, A, B);
  dst(1, 0) = dst(3, 1) = AVG3(I, X, A);
  dst(1, 1) = dst(3, 2) = AVG3(J, I, X);
  dst(1, 2) = dst(3, 3) = AVG3(K, J, I);
  dst(1, 3) = AVG3(L, K, J);
}
#endif  // CONFIG_VP9_HIGHBITDEPTH

// This serves as a wrapper function, so that all the prediction functions
// can be unified and accessed as a pointer array. Note that the boundary
// above and left are not necessarily used all the time.
#define intra_pred_sized(type, size)                        \
  void eb_vp9_##type##_predictor_##size##x##size##_c(          \
      uint8_t *dst, ptrdiff_t stride, const uint8_t *above, \
      const uint8_t *left) {                                \
    type##_predictor(dst, stride, size, above, left);       \
  }

#if CONFIG_VP9_HIGHBITDEPTH
#define intra_pred_highbd_sized(type, size)                        \
  void vpx_highbd_##type##_predictor_##size##x##size##_c(          \
      uint16_t *dst, ptrdiff_t stride, const uint16_t *above,      \
      const uint16_t *left, int bd) {                              \
    highbd_##type##_predictor(dst, stride, size, above, left, bd); \
  }

/* clang-format off */
#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_highbd_sized(type, 4) \
  intra_pred_highbd_sized(type, 8) \
  intra_pred_highbd_sized(type, 16) \
  intra_pred_highbd_sized(type, 32)

#define intra_pred_no_4x4(type) \
  intra_pred_sized(type, 8) \
  intra_pred_sized(type, 16) \
  intra_pred_sized(type, 32) \
  intra_pred_highbd_sized(type, 8) \
  intra_pred_highbd_sized(type, 16) \
  intra_pred_highbd_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)

#define intra_pred_no_4x4(type) \
  intra_pred_sized(type, 8) \
  intra_pred_sized(type, 16) \
  intra_pred_sized(type, 32)
#endif  // CONFIG_VP9_HIGHBITDEPTH

intra_pred_no_4x4(d207)
intra_pred_no_4x4(d63)
intra_pred_no_4x4(d45)
intra_pred_no_4x4(d117)
intra_pred_no_4x4(d135)
intra_pred_no_4x4(d153)
intra_pred_allsizes(v)
intra_pred_allsizes(h)
intra_pred_allsizes(tm)
intra_pred_allsizes(dc_128)
intra_pred_allsizes(dc_left)
intra_pred_allsizes(dc_top)
intra_pred_allsizes(dc)
/* clang-format on */
#undef intra_pred_allsizes