/*

 * Copyright (c) 2016, Alliance for Open Media. All rights reserved.

 *

 * This source code is subject to the terms of the BSD 2 Clause License and

 * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License

 * was not distributed with this source code in the LICENSE file, you can

 * obtain it at www.aomedia.org/license/software. If the Alliance for Open

 * Media Patent License 1.0 was not distributed with this source code in the

 * PATENTS file, you can obtain it at www.aomedia.org/license/patent.

 */

#include <assert.h>

#include <math.h>

#include "config/aom_config.h"

#include "config/aom_dsp_rtcd.h"

#include "aom_dsp/aom_dsp_common.h"

#include "aom_dsp/intrapred_common.h"

#include "aom_mem/aom_mem.h"

#include "aom_ports/bitops.h"

static inline void v_predictor(uint8_t *dst, ptrdiff_t stride, int bw, int bh,

                               const uint8_t *above, const uint8_t *left) {

  int r;

  (void)left;

  for (r = 0; r < bh; r++) {

    memcpy(dst, above, bw);

    dst += stride;

  }

}

static inline void h_predictor(uint8_t *dst, ptrdiff_t stride, int bw, int bh,

                               const uint8_t *above, const uint8_t *left) {

  int r;

  (void)above;

  for (r = 0; r < bh; r++) {

    memset(dst, left[r], bw);

    dst += stride;

  }

}

static inline int abs_diff(int a, int b) { return (a > b) ? a - b : b - a; }

static inline uint16_t paeth_predictor_single(uint16_t left, uint16_t top,

                                              uint16_t top_left) {

  const int base = top + left - top_left;

  const int p_left = abs_diff(base, left);

  const int p_top = abs_diff(base, top);

  const int p_top_left = abs_diff(base, top_left);

  // Return nearest to base of left, top and top_left.

  return (p_left <= p_top && p_left <= p_top_left) ? left

         : (p_top <= p_top_left)                   ? top

                                                   : top_left;

}

static inline void paeth_predictor(uint8_t *dst, ptrdiff_t stride, int bw,

                                   int bh, const uint8_t *above,

                                   const uint8_t *left) {

  int r, c;

  const uint8_t ytop_left = above[-1];

  for (r = 0; r < bh; r++) {

    for (c = 0; c < bw; c++)

      dst[c] = (uint8_t)paeth_predictor_single(left[r], above[c], ytop_left);

    dst += stride;

  }

}

// Some basic checks on weights for smooth predictor.

#define sm_weights_sanity_checks(weights_w, weights_h, weights_scale, \

                                 pred_scale)                          \

  assert(weights_w[0] < weights_scale);                               \

  assert(weights_h[0] < weights_scale);                               \

  assert(weights_scale - weights_w[bw - 1] < weights_scale);          \

  assert(weights_scale - weights_h[bh - 1] < weights_scale);          \

  assert(pred_scale < 31)  // ensures no overflow when calculating predictor.

#define divide_round(value, bits) (((value) + (1 << ((bits)-1))) >> (bits))

static inline void smooth_predictor(uint8_t *dst, ptrdiff_t stride, int bw,

                                    int bh, const uint8_t *above,

                                    const uint8_t *left) {

  const uint8_t below_pred = left[bh - 1];   // estimated by bottom-left pixel

  const uint8_t right_pred = above[bw - 1];  // estimated by top-right pixel

  const uint8_t *const sm_weights_w = smooth_weights + bw - 4;

  const uint8_t *const sm_weights_h = smooth_weights + bh - 4;

  // scale = 2 * 2^SMOOTH_WEIGHT_LOG2_SCALE

  const int log2_scale = 1 + SMOOTH_WEIGHT_LOG2_SCALE;

  const uint16_t scale = (1 << SMOOTH_WEIGHT_LOG2_SCALE);

  sm_weights_sanity_checks(sm_weights_w, sm_weights_h, scale,

                           log2_scale + sizeof(*dst));

  int r;

  for (r = 0; r < bh; ++r) {

    int c;

    for (c = 0; c < bw; ++c) {

      const uint8_t pixels[] = { above[c], below_pred, left[r], right_pred };

      const uint8_t weights[] = { sm_weights_h[r], scale - sm_weights_h[r],

                                  sm_weights_w[c], scale - sm_weights_w[c] };

      uint32_t this_pred = 0;

      int i;

      assert(scale >= sm_weights_h[r] && scale >= sm_weights_w[c]);

      for (i = 0; i < 4; ++i) {

        this_pred += weights[i] * pixels[i];

      }

      dst[c] = divide_round(this_pred, log2_scale);

    }

    dst += stride;

  }

}

static inline void smooth_v_predictor(uint8_t *dst, ptrdiff_t stride, int bw,

                                      int bh, const uint8_t *above,

                                      const uint8_t *left) {

  const uint8_t below_pred = left[bh - 1];  // estimated by bottom-left pixel

  const uint8_t *const sm_weights = smooth_weights + bh - 4;

  // scale = 2^SMOOTH_WEIGHT_LOG2_SCALE

  const int log2_scale = SMOOTH_WEIGHT_LOG2_SCALE;

  const uint16_t scale = (1 << SMOOTH_WEIGHT_LOG2_SCALE);

  sm_weights_sanity_checks(sm_weights, sm_weights, scale,

                           log2_scale + sizeof(*dst));

  int r;

  for (r = 0; r < bh; r++) {

    int c;

    for (c = 0; c < bw; ++c) {

      const uint8_t pixels[] = { above[c], below_pred };

      const uint8_t weights[] = { sm_weights[r], scale - sm_weights[r] };

      uint32_t this_pred = 0;

      assert(scale >= sm_weights[r]);

      int i;

      for (i = 0; i < 2; ++i) {

        this_pred += weights[i] * pixels[i];

      }

      dst[c] = divide_round(this_pred, log2_scale);

    }

    dst += stride;

  }

}

static inline void smooth_h_predictor(uint8_t *dst, ptrdiff_t stride, int bw,

                                      int bh, const uint8_t *above,

                                      const uint8_t *left) {

  const uint8_t right_pred = above[bw - 1];  // estimated by top-right pixel

  const uint8_t *const sm_weights = smooth_weights + bw - 4;

  // scale = 2^SMOOTH_WEIGHT_LOG2_SCALE

  const int log2_scale = SMOOTH_WEIGHT_LOG2_SCALE;

  const uint16_t scale = (1 << SMOOTH_WEIGHT_LOG2_SCALE);

  sm_weights_sanity_checks(sm_weights, sm_weights, scale,

                           log2_scale + sizeof(*dst));

  int r;

  for (r = 0; r < bh; r++) {

    int c;

    for (c = 0; c < bw; ++c) {

      const uint8_t pixels[] = { left[r], right_pred };

      const uint8_t weights[] = { sm_weights[c], scale - sm_weights[c] };

      uint32_t this_pred = 0;

      assert(scale >= sm_weights[c]);

      int i;

      for (i = 0; i < 2; ++i) {

        this_pred += weights[i] * pixels[i];

      }

      dst[c] = divide_round(this_pred, log2_scale);

    }

    dst += stride;

  }

}

static inline void dc_128_predictor(uint8_t *dst, ptrdiff_t stride, int bw,

                                    int bh, const uint8_t *above,

                                    const uint8_t *left) {

  int r;

  (void)above;

  (void)left;

  for (r = 0; r < bh; r++) {

    memset(dst, 128, bw);

    dst += stride;

  }

}

static inline void dc_left_predictor(uint8_t *dst, ptrdiff_t stride, int bw,

                                     int bh, const uint8_t *above,

                                     const uint8_t *left) {

  int i, r, expected_dc, sum = 0;

  (void)above;

  for (i = 0; i < bh; i++) sum += left[i];

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

  for (r = 0; r < bh; r++) {

    memset(dst, expected_dc, bw);

    dst += stride;

  }

}

static inline void dc_top_predictor(uint8_t *dst, ptrdiff_t stride, int bw,

                                    int bh, const uint8_t *above,

                                    const uint8_t *left) {

  int i, r, expected_dc, sum = 0;

  (void)left;

  for (i = 0; i < bw; i++) sum += above[i];

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

  for (r = 0; r < bh; r++) {

    memset(dst, expected_dc, bw);

    dst += stride;

  }

}

static inline void dc_predictor(uint8_t *dst, ptrdiff_t stride, int bw, int bh,

                                const uint8_t *above, const uint8_t *left) {

  int i, r, expected_dc, sum = 0;

  const int count = bw + bh;

  for (i = 0; i < bw; i++) {

    sum += above[i];

  }

  for (i = 0; i < bh; i++) {

    sum += left[i];

  }

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

  for (r = 0; r < bh; r++) {

    memset(dst, expected_dc, bw);

    dst += stride;

  }

}

static inline int divide_using_multiply_shift(int num, int shift1,

                                              int multiplier, int shift2) {

  const int interm = num >> shift1;

  return interm * multiplier >> shift2;

}

// The constants (multiplier and shifts) for a given block size are obtained

// as follows:

// - Let sum_w_h =  block width + block height.

// - Shift 'sum_w_h' right until we reach an odd number. Let the number of

// shifts for that block size be called 'shift1' (see the parameter in

// dc_predictor_rect() function), and let the odd number be 'd'. [d has only 2

// possible values: d = 3 for a 1:2 rect block and d = 5 for a 1:4 rect

// block].

// - Find multipliers for (i) dividing by 3, and (ii) dividing by 5,

// using the "Algorithm 1" in:

// http://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=1467632

// by ensuring that m + n = 16 (in that algorithm). This ensures that our 2nd

// shift will be 16, regardless of the block size.

// Note: For low bitdepth, assembly code may be optimized by using smaller

// constants for smaller block sizes, where the range of the 'sum' is

// restricted to fewer bits.

#define DC_MULTIPLIER_1X2 0x5556

#define DC_MULTIPLIER_1X4 0x3334

#define DC_SHIFT2 16

static inline void dc_predictor_rect(uint8_t *dst, ptrdiff_t stride, int bw,

                                     int bh, const uint8_t *above,

                                     const uint8_t *left, int shift1,

                                     int multiplier) {

  int sum = 0;

  for (int i = 0; i < bw; i++) {

    sum += above[i];

  }

  for (int i = 0; i < bh; i++) {

    sum += left[i];

  }

  const int expected_dc = divide_using_multiply_shift(

      sum + ((bw + bh) >> 1), shift1, multiplier, DC_SHIFT2);

  assert(expected_dc < (1 << 8));

  for (int r = 0; r < bh; r++) {

    memset(dst, expected_dc, bw);

    dst += stride;

  }

}

#undef DC_SHIFT2

void aom_dc_predictor_4x8_c(uint8_t *dst, ptrdiff_t stride,

                            const uint8_t *above, const uint8_t *left) {

  dc_predictor_rect(dst, stride, 4, 8, above, left, 2, DC_MULTIPLIER_1X2);

}

void aom_dc_predictor_8x4_c(uint8_t *dst, ptrdiff_t stride,

                            const uint8_t *above, const uint8_t *left) {

  dc_predictor_rect(dst, stride, 8, 4, above, left, 2, DC_MULTIPLIER_1X2);

}

#if !CONFIG_REALTIME_ONLY || CONFIG_AV1_DECODER

void aom_dc_predictor_4x16_c(uint8_t *dst, ptrdiff_t stride,

                             const uint8_t *above, const uint8_t *left) {

  dc_predictor_rect(dst, stride, 4, 16, above, left, 2, DC_MULTIPLIER_1X4);

}

void aom_dc_predictor_16x4_c(uint8_t *dst, ptrdiff_t stride,

                             const uint8_t *above, const uint8_t *left) {

  dc_predictor_rect(dst, stride, 16, 4, above, left, 2, DC_MULTIPLIER_1X4);

}

#endif  // !CONFIG_REALTIME_ONLY || CONFIG_AV1_DECODER

void aom_dc_predictor_8x16_c(uint8_t *dst, ptrdiff_t stride,

                             const uint8_t *above, const uint8_t *left) {

  dc_predictor_rect(dst, stride, 8, 16, above, left, 3, DC_MULTIPLIER_1X2);

}

void aom_dc_predictor_16x8_c(uint8_t *dst, ptrdiff_t stride,

                             const uint8_t *above, const uint8_t *left) {

  dc_predictor_rect(dst, stride, 16, 8, above, left, 3, DC_MULTIPLIER_1X2);

}

#if !CONFIG_REALTIME_ONLY || CONFIG_AV1_DECODER

void aom_dc_predictor_8x32_c(uint8_t *dst, ptrdiff_t stride,

                             const uint8_t *above, const uint8_t *left) {

  dc_predictor_rect(dst, stride, 8, 32, above, left, 3, DC_MULTIPLIER_1X4);

}

void aom_dc_predictor_32x8_c(uint8_t *dst, ptrdiff_t stride,

                             const uint8_t *above, const uint8_t *left) {

  dc_predictor_rect(dst, stride, 32, 8, above, left, 3, DC_MULTIPLIER_1X4);

}

#endif  // !CONFIG_REALTIME_ONLY || CONFIG_AV1_DECODER

void aom_dc_predictor_16x32_c(uint8_t *dst, ptrdiff_t stride,

                              const uint8_t *above, const uint8_t *left) {

  dc_predictor_rect(dst, stride, 16, 32, above, left, 4, DC_MULTIPLIER_1X2);

}

void aom_dc_predictor_32x16_c(uint8_t *dst, ptrdiff_t stride,

                              const uint8_t *above, const uint8_t *left) {

  dc_predictor_rect(dst, stride, 32, 16, above, left, 4, DC_MULTIPLIER_1X2);

}

#if !CONFIG_REALTIME_ONLY || CONFIG_AV1_DECODER

void aom_dc_predictor_16x64_c(uint8_t *dst, ptrdiff_t stride,

                              const uint8_t *above, const uint8_t *left) {

  dc_predictor_rect(dst, stride, 16, 64, above, left, 4, DC_MULTIPLIER_1X4);

}

void aom_dc_predictor_64x16_c(uint8_t *dst, ptrdiff_t stride,

                              const uint8_t *above, const uint8_t *left) {

  dc_predictor_rect(dst, stride, 64, 16, above, left, 4, DC_MULTIPLIER_1X4);

}

#endif  // !CONFIG_REALTIME_ONLY || CONFIG_AV1_DECODER

void aom_dc_predictor_32x64_c(uint8_t *dst, ptrdiff_t stride,

                              const uint8_t *above, const uint8_t *left) {

  dc_predictor_rect(dst, stride, 32, 64, above, left, 5, DC_MULTIPLIER_1X2);

}

void aom_dc_predictor_64x32_c(uint8_t *dst, ptrdiff_t stride,

                              const uint8_t *above, const uint8_t *left) {

  dc_predictor_rect(dst, stride, 64, 32, above, left, 5, DC_MULTIPLIER_1X2);

}

#undef DC_MULTIPLIER_1X2

#undef DC_MULTIPLIER_1X4

#if CONFIG_AV1_HIGHBITDEPTH

static inline void highbd_v_predictor(uint16_t *dst, ptrdiff_t stride, int bw,

                                      int bh, const uint16_t *above,

                                      const uint16_t *left, int bd) {

  int r;

  (void)left;

  (void)bd;

  for (r = 0; r < bh; r++) {

    memcpy(dst, above, bw * sizeof(uint16_t));

    dst += stride;

  }

}

static inline void highbd_h_predictor(uint16_t *dst, ptrdiff_t stride, int bw,

                                      int bh, const uint16_t *above,

                                      const uint16_t *left, int bd) {

  int r;

  (void)above;

  (void)bd;

  for (r = 0; r < bh; r++) {

    aom_memset16(dst, left[r], bw);

    dst += stride;

  }

}

static inline void highbd_paeth_predictor(uint16_t *dst, ptrdiff_t stride,

                                          int bw, int bh, const uint16_t *above,

                                          const uint16_t *left, int bd) {

  int r, c;

  const uint16_t ytop_left = above[-1];

  (void)bd;

  for (r = 0; r < bh; r++) {

    for (c = 0; c < bw; c++)

      dst[c] = paeth_predictor_single(left[r], above[c], ytop_left);

    dst += stride;

  }

}

static inline void highbd_smooth_predictor(uint16_t *dst, ptrdiff_t stride,

                                           int bw, int bh,

                                           const uint16_t *above,

                                           const uint16_t *left, int bd) {

  (void)bd;

  const uint16_t below_pred = left[bh - 1];   // estimated by bottom-left pixel

  const uint16_t right_pred = above[bw - 1];  // estimated by top-right pixel

  const uint8_t *const sm_weights_w = smooth_weights + bw - 4;

  const uint8_t *const sm_weights_h = smooth_weights + bh - 4;

  // scale = 2 * 2^SMOOTH_WEIGHT_LOG2_SCALE

  const int log2_scale = 1 + SMOOTH_WEIGHT_LOG2_SCALE;

  const uint16_t scale = (1 << SMOOTH_WEIGHT_LOG2_SCALE);

  sm_weights_sanity_checks(sm_weights_w, sm_weights_h, scale,

                           log2_scale + sizeof(*dst));

  int r;

  for (r = 0; r < bh; ++r) {

    int c;

    for (c = 0; c < bw; ++c) {

      const uint16_t pixels[] = { above[c], below_pred, left[r], right_pred };

      const uint8_t weights[] = { sm_weights_h[r], scale - sm_weights_h[r],

                                  sm_weights_w[c], scale - sm_weights_w[c] };

      uint32_t this_pred = 0;

      int i;

      assert(scale >= sm_weights_h[r] && scale >= sm_weights_w[c]);

      for (i = 0; i < 4; ++i) {

        this_pred += weights[i] * pixels[i];

      }

      dst[c] = divide_round(this_pred, log2_scale);

    }

    dst += stride;

  }

}

static inline void highbd_smooth_v_predictor(uint16_t *dst, ptrdiff_t stride,

                                             int bw, int bh,

                                             const uint16_t *above,

                                             const uint16_t *left, int bd) {

  (void)bd;

  const uint16_t below_pred = left[bh - 1];  // estimated by bottom-left pixel

  const uint8_t *const sm_weights = smooth_weights + bh - 4;

  // scale = 2^SMOOTH_WEIGHT_LOG2_SCALE

  const int log2_scale = SMOOTH_WEIGHT_LOG2_SCALE;

  const uint16_t scale = (1 << SMOOTH_WEIGHT_LOG2_SCALE);

  sm_weights_sanity_checks(sm_weights, sm_weights, scale,

                           log2_scale + sizeof(*dst));

  int r;

  for (r = 0; r < bh; r++) {

    int c;

    for (c = 0; c < bw; ++c) {

      const uint16_t pixels[] = { above[c], below_pred };

      const uint8_t weights[] = { sm_weights[r], scale - sm_weights[r] };

      uint32_t this_pred = 0;

      assert(scale >= sm_weights[r]);

      int i;

      for (i = 0; i < 2; ++i) {

        this_pred += weights[i] * pixels[i];

      }

      dst[c] = divide_round(this_pred, log2_scale);

    }

    dst += stride;

  }

}

static inline void highbd_smooth_h_predictor(uint16_t *dst, ptrdiff_t stride,

                                             int bw, int bh,

                                             const uint16_t *above,

                                             const uint16_t *left, int bd) {

  (void)bd;

  const uint16_t right_pred = above[bw - 1];  // estimated by top-right pixel

  const uint8_t *const sm_weights = smooth_weights + bw - 4;

  // scale = 2^SMOOTH_WEIGHT_LOG2_SCALE

  const int log2_scale = SMOOTH_WEIGHT_LOG2_SCALE;

  const uint16_t scale = (1 << SMOOTH_WEIGHT_LOG2_SCALE);

  sm_weights_sanity_checks(sm_weights, sm_weights, scale,

                           log2_scale + sizeof(*dst));

  int r;

  for (r = 0; r < bh; r++) {

    int c;

    for (c = 0; c < bw; ++c) {

      const uint16_t pixels[] = { left[r], right_pred };

      const uint8_t weights[] = { sm_weights[c], scale - sm_weights[c] };

      uint32_t this_pred = 0;

      assert(scale >= sm_weights[c]);

      int i;

      for (i = 0; i < 2; ++i) {

        this_pred += weights[i] * pixels[i];

      }

      dst[c] = divide_round(this_pred, log2_scale);

    }

    dst += stride;

  }

}

static inline void highbd_dc_128_predictor(uint16_t *dst, ptrdiff_t stride,

                                           int bw, int bh,

                                           const uint16_t *above,

                                           const uint16_t *left, int bd) {

  int r;

  (void)above;

  (void)left;

  for (r = 0; r < bh; r++) {

    aom_memset16(dst, 128 << (bd - 8), bw);

    dst += stride;

  }

}

static inline void highbd_dc_left_predictor(uint16_t *dst, ptrdiff_t stride,

                                            int bw, int bh,

                                            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 < bh; i++) sum += left[i];

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

  for (r = 0; r < bh; r++) {

    aom_memset16(dst, expected_dc, bw);

    dst += stride;

  }

}

static inline void highbd_dc_top_predictor(uint16_t *dst, ptrdiff_t stride,

                                           int bw, int bh,

                                           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 < bw; i++) sum += above[i];

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

  for (r = 0; r < bh; r++) {

    aom_memset16(dst, expected_dc, bw);

    dst += stride;

  }

}

static inline void highbd_dc_predictor(uint16_t *dst, ptrdiff_t stride, int bw,

                                       int bh, const uint16_t *above,

                                       const uint16_t *left, int bd) {

  int i, r, expected_dc, sum = 0;

  const int count = bw + bh;

  (void)bd;

  for (i = 0; i < bw; i++) {

    sum += above[i];

  }

  for (i = 0; i < bh; i++) {

    sum += left[i];

  }

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

  for (r = 0; r < bh; r++) {

    aom_memset16(dst, expected_dc, bw);

    dst += stride;

  }

}

// Obtained similarly as DC_MULTIPLIER_1X2 and DC_MULTIPLIER_1X4 above, but

// assume 2nd shift of 17 bits instead of 16.

// Note: Strictly speaking, 2nd shift needs to be 17 only when:

// - bit depth == 12, and

// - bw + bh is divisible by 5 (as opposed to divisible by 3).

// All other cases can use half the multipliers with a shift of 16 instead.

// This special optimization can be used when writing assembly code.

#define HIGHBD_DC_MULTIPLIER_1X2 0xAAAB

// Note: This constant is odd, but a smaller even constant (0x199a) with the

// appropriate shift should work for neon in 8/10-bit.

#define HIGHBD_DC_MULTIPLIER_1X4 0x6667

#define HIGHBD_DC_SHIFT2 17

static inline void highbd_dc_predictor_rect(uint16_t *dst, ptrdiff_t stride,

                                            int bw, int bh,

                                            const uint16_t *above,

                                            const uint16_t *left, int bd,

                                            int shift1, uint32_t multiplier) {

  int sum = 0;

  (void)bd;

  for (int i = 0; i < bw; i++) {

    sum += above[i];

  }

  for (int i = 0; i < bh; i++) {

    sum += left[i];

  }

  const int expected_dc = divide_using_multiply_shift(

      sum + ((bw + bh) >> 1), shift1, multiplier, HIGHBD_DC_SHIFT2);

  assert(expected_dc < (1 << bd));

  for (int r = 0; r < bh; r++) {

    aom_memset16(dst, expected_dc, bw);

    dst += stride;

  }

}

#undef HIGHBD_DC_SHIFT2

void aom_highbd_dc_predictor_4x8_c(uint16_t *dst, ptrdiff_t stride,

                                   const uint16_t *above, const uint16_t *left,

                                   int bd) {

  highbd_dc_predictor_rect(dst, stride, 4, 8, above, left, bd, 2,

                           HIGHBD_DC_MULTIPLIER_1X2);

}

void aom_highbd_dc_predictor_8x4_c(uint16_t *dst, ptrdiff_t stride,

                                   const uint16_t *above, const uint16_t *left,

                                   int bd) {

  highbd_dc_predictor_rect(dst, stride, 8, 4, above, left, bd, 2,

                           HIGHBD_DC_MULTIPLIER_1X2);

}

#if !CONFIG_REALTIME_ONLY || CONFIG_AV1_DECODER

void aom_highbd_dc_predictor_4x16_c(uint16_t *dst, ptrdiff_t stride,

                                    const uint16_t *above, const uint16_t *left,

                                    int bd) {

  highbd_dc_predictor_rect(dst, stride, 4, 16, above, left, bd, 2,

                           HIGHBD_DC_MULTIPLIER_1X4);

}

void aom_highbd_dc_predictor_16x4_c(uint16_t *dst, ptrdiff_t stride,

                                    const uint16_t *above, const uint16_t *left,

                                    int bd) {

  highbd_dc_predictor_rect(dst, stride, 16, 4, above, left, bd, 2,

                           HIGHBD_DC_MULTIPLIER_1X4);

}

#endif  // !CONFIG_REALTIME_ONLY || CONFIG_AV1_DECODER

void aom_highbd_dc_predictor_8x16_c(uint16_t *dst, ptrdiff_t stride,

                                    const uint16_t *above, const uint16_t *left,

                                    int bd) {

  highbd_dc_predictor_rect(dst, stride, 8, 16, above, left, bd, 3,

                           HIGHBD_DC_MULTIPLIER_1X2);

}

void aom_highbd_dc_predictor_16x8_c(uint16_t *dst, ptrdiff_t stride,

                                    const uint16_t *above, const uint16_t *left,

                                    int bd) {

  highbd_dc_predictor_rect(dst, stride, 16, 8, above, left, bd, 3,

                           HIGHBD_DC_MULTIPLIER_1X2);

}

#if !CONFIG_REALTIME_ONLY || CONFIG_AV1_DECODER

void aom_highbd_dc_predictor_8x32_c(uint16_t *dst, ptrdiff_t stride,

                                    const uint16_t *above, const uint16_t *left,

                                    int bd) {

  highbd_dc_predictor_rect(dst, stride, 8, 32, above, left, bd, 3,

                           HIGHBD_DC_MULTIPLIER_1X4);

}

void aom_highbd_dc_predictor_32x8_c(uint16_t *dst, ptrdiff_t stride,

                                    const uint16_t *above, const uint16_t *left,

                                    int bd) {

  highbd_dc_predictor_rect(dst, stride, 32, 8, above, left, bd, 3,

                           HIGHBD_DC_MULTIPLIER_1X4);

}

#endif  // !CONFIG_REALTIME_ONLY || CONFIG_AV1_DECODER

void aom_highbd_dc_predictor_16x32_c(uint16_t *dst, ptrdiff_t stride,

                                     const uint16_t *above,

                                     const uint16_t *left, int bd) {

  highbd_dc_predictor_rect(dst, stride, 16, 32, above, left, bd, 4,

                           HIGHBD_DC_MULTIPLIER_1X2);

}

void aom_highbd_dc_predictor_32x16_c(uint16_t *dst, ptrdiff_t stride,

                                     const uint16_t *above,

                                     const uint16_t *left, int bd) {

  highbd_dc_predictor_rect(dst, stride, 32, 16, above, left, bd, 4,

                           HIGHBD_DC_MULTIPLIER_1X2);

}

#if !CONFIG_REALTIME_ONLY || CONFIG_AV1_DECODER

void aom_highbd_dc_predictor_16x64_c(uint16_t *dst, ptrdiff_t stride,

                                     const uint16_t *above,

                                     const uint16_t *left, int bd) {

  highbd_dc_predictor_rect(dst, stride, 16, 64, above, left, bd, 4,

                           HIGHBD_DC_MULTIPLIER_1X4);

}

void aom_highbd_dc_predictor_64x16_c(uint16_t *dst, ptrdiff_t stride,

                                     const uint16_t *above,

                                     const uint16_t *left, int bd) {

  highbd_dc_predictor_rect(dst, stride, 64, 16, above, left, bd, 4,

                           HIGHBD_DC_MULTIPLIER_1X4);

}

#endif  // !CONFIG_REALTIME_ONLY || CONFIG_AV1_DECODER

void aom_highbd_dc_predictor_32x64_c(uint16_t *dst, ptrdiff_t stride,

                                     const uint16_t *above,

                                     const uint16_t *left, int bd) {

  highbd_dc_predictor_rect(dst, stride, 32, 64, above, left, bd, 5,

                           HIGHBD_DC_MULTIPLIER_1X2);

}

void aom_highbd_dc_predictor_64x32_c(uint16_t *dst, ptrdiff_t stride,

                                     const uint16_t *above,

                                     const uint16_t *left, int bd) {

  highbd_dc_predictor_rect(dst, stride, 64, 32, above, left, bd, 5,

                           HIGHBD_DC_MULTIPLIER_1X2);

}

#undef HIGHBD_DC_MULTIPLIER_1X2

#undef HIGHBD_DC_MULTIPLIER_1X4

#endif  // CONFIG_AV1_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, width, height)                  \

  void aom_##type##_predictor_##width##x##height##_c(          \

      uint8_t *dst, ptrdiff_t stride, const uint8_t *above,    \

      const uint8_t *left) {                                   \

    type##_predictor(dst, stride, width, height, above, left); \

  }

aom_v_predictor_4x4_c

Line

Count

Source

729

257k

      const uint8_t *left) {                                   \

730

257k

    type##_predictor(dst, stride, width, height, above, left); \

731

257k

  }

aom_v_predictor_8x8_c

Line

Count

Source

729

16.8k

      const uint8_t *left) {                                   \

730

16.8k

    type##_predictor(dst, stride, width, height, above, left); \

731

16.8k

  }

aom_v_predictor_16x16_c

Line

Count

Source

729

10.6k

      const uint8_t *left) {                                   \

730

10.6k

    type##_predictor(dst, stride, width, height, above, left); \

731

10.6k

  }

aom_v_predictor_32x32_c

Line

Count

Source

729

21.6k

      const uint8_t *left) {                                   \

730

21.6k

    type##_predictor(dst, stride, width, height, above, left); \

731

21.6k

  }

aom_v_predictor_64x64_c

Line

Count

Source

729

1.66k

      const uint8_t *left) {                                   \

730

1.66k

    type##_predictor(dst, stride, width, height, above, left); \

731

1.66k

  }

aom_v_predictor_4x8_c

Line

Count

Source

729

9.12k

      const uint8_t *left) {                                   \

730

9.12k

    type##_predictor(dst, stride, width, height, above, left); \

731

9.12k

  }

aom_v_predictor_8x4_c

Line

Count

Source

729

13.2k

      const uint8_t *left) {                                   \

730

13.2k

    type##_predictor(dst, stride, width, height, above, left); \

731

13.2k

  }

aom_v_predictor_8x16_c

Line

Count

Source

729

5.83k

      const uint8_t *left) {                                   \

730

5.83k

    type##_predictor(dst, stride, width, height, above, left); \

731

5.83k

  }

aom_v_predictor_16x8_c

Line

Count

Source

729

7.83k

      const uint8_t *left) {                                   \

730

7.83k

    type##_predictor(dst, stride, width, height, above, left); \

731

7.83k

  }

aom_v_predictor_16x32_c

Line

Count

Source

729

2.38k

      const uint8_t *left) {                                   \

730

2.38k

    type##_predictor(dst, stride, width, height, above, left); \

731

2.38k

  }

aom_v_predictor_32x16_c

Line

Count

Source

729

2.13k

      const uint8_t *left) {                                   \

730

2.13k

    type##_predictor(dst, stride, width, height, above, left); \

731

2.13k

  }

aom_v_predictor_32x64_c

Line

Count

Source

729

140

      const uint8_t *left) {                                   \

730

140

    type##_predictor(dst, stride, width, height, above, left); \

731

140

  }

aom_v_predictor_64x32_c

Line

Count

Source

729

149

      const uint8_t *left) {                                   \

730

149

    type##_predictor(dst, stride, width, height, above, left); \

731

149

  }

aom_v_predictor_4x16_c

Line

Count

Source

729

3.70k

      const uint8_t *left) {                                   \

730

3.70k

    type##_predictor(dst, stride, width, height, above, left); \

731

3.70k

  }

aom_v_predictor_16x4_c

Line

Count

Source

729

6.83k

      const uint8_t *left) {                                   \

730

6.83k

    type##_predictor(dst, stride, width, height, above, left); \

731

6.83k

  }

aom_v_predictor_8x32_c

Line

Count

Source

729

1.95k

      const uint8_t *left) {                                   \

730

1.95k

    type##_predictor(dst, stride, width, height, above, left); \

731

1.95k

  }

aom_v_predictor_32x8_c

Line

Count

Source

729

2.88k

      const uint8_t *left) {                                   \

730

2.88k

    type##_predictor(dst, stride, width, height, above, left); \

731

2.88k

  }

aom_v_predictor_16x64_c

Line

Count

Source

729

250

      const uint8_t *left) {                                   \

730

250

    type##_predictor(dst, stride, width, height, above, left); \

731

250

  }

aom_v_predictor_64x16_c

Line

Count

Source

729

187

      const uint8_t *left) {                                   \

730

187

    type##_predictor(dst, stride, width, height, above, left); \

731

187

  }

aom_h_predictor_4x4_c

Line

Count

Source

729

187k

      const uint8_t *left) {                                   \

730

187k

    type##_predictor(dst, stride, width, height, above, left); \

731

187k

  }

aom_h_predictor_8x8_c

Line

Count

Source

729

21.3k

      const uint8_t *left) {                                   \

730

21.3k

    type##_predictor(dst, stride, width, height, above, left); \

731

21.3k

  }

aom_h_predictor_16x16_c

Line

Count

Source

729

21.2k

      const uint8_t *left) {                                   \

730

21.2k

    type##_predictor(dst, stride, width, height, above, left); \

731

21.2k

  }

aom_h_predictor_32x32_c

Line

Count

Source

729

20.7k

      const uint8_t *left) {                                   \

730

20.7k

    type##_predictor(dst, stride, width, height, above, left); \

731

20.7k

  }

aom_h_predictor_64x64_c

Line

Count

Source

729

1.87k

      const uint8_t *left) {                                   \

730

1.87k

    type##_predictor(dst, stride, width, height, above, left); \

731

1.87k

  }

aom_h_predictor_4x8_c

Line

Count

Source

729

13.7k

      const uint8_t *left) {                                   \

730

13.7k

    type##_predictor(dst, stride, width, height, above, left); \

731

13.7k

  }

aom_h_predictor_8x4_c

Line

Count

Source

729

19.4k

      const uint8_t *left) {                                   \

730

19.4k

    type##_predictor(dst, stride, width, height, above, left); \

731

19.4k

  }

aom_h_predictor_8x16_c

Line

Count

Source

729

7.23k

      const uint8_t *left) {                                   \

730

7.23k

    type##_predictor(dst, stride, width, height, above, left); \

731

7.23k

  }

aom_h_predictor_16x8_c

Line

Count

Source

729

9.03k

      const uint8_t *left) {                                   \

730

9.03k

    type##_predictor(dst, stride, width, height, above, left); \

731

9.03k

  }

aom_h_predictor_16x32_c

Line

Count

Source

729

2.38k

      const uint8_t *left) {                                   \

730

2.38k

    type##_predictor(dst, stride, width, height, above, left); \

731

2.38k

  }

aom_h_predictor_32x16_c

Line

Count

Source

729

2.88k

      const uint8_t *left) {                                   \

730

2.88k

    type##_predictor(dst, stride, width, height, above, left); \

731

2.88k

  }

aom_h_predictor_32x64_c

Line

Count

Source

729

173

      const uint8_t *left) {                                   \

730

173

    type##_predictor(dst, stride, width, height, above, left); \

731

173

  }

aom_h_predictor_64x32_c

Line

Count

Source

729

200

      const uint8_t *left) {                                   \

730

200

    type##_predictor(dst, stride, width, height, above, left); \

731

200

  }

aom_h_predictor_4x16_c

Line

Count

Source

729

5.34k

      const uint8_t *left) {                                   \

730

5.34k

    type##_predictor(dst, stride, width, height, above, left); \

731

5.34k

  }

aom_h_predictor_16x4_c

Line

Count

Source

729

8.56k

      const uint8_t *left) {                                   \

730

8.56k

    type##_predictor(dst, stride, width, height, above, left); \

731

8.56k

  }

aom_h_predictor_8x32_c

Line

Count

Source

729

2.45k

      const uint8_t *left) {                                   \

730

2.45k

    type##_predictor(dst, stride, width, height, above, left); \

731

2.45k

  }

aom_h_predictor_32x8_c

Line

Count

Source

729

3.19k

      const uint8_t *left) {                                   \

730

3.19k

    type##_predictor(dst, stride, width, height, above, left); \

731

3.19k

  }

aom_h_predictor_16x64_c

Line

Count

Source

729

322

      const uint8_t *left) {                                   \

730

322

    type##_predictor(dst, stride, width, height, above, left); \

731

322

  }

aom_h_predictor_64x16_c

Line

Count

Source

729

287

      const uint8_t *left) {                                   \

730

287

    type##_predictor(dst, stride, width, height, above, left); \

731

287

  }

aom_smooth_predictor_4x4_c

Line

Count

Source

729

161k

      const uint8_t *left) {                                   \

730

161k

    type##_predictor(dst, stride, width, height, above, left); \

731

161k

  }

aom_smooth_predictor_8x8_c

Line

Count

Source

729

67.1k

      const uint8_t *left) {                                   \

730

67.1k

    type##_predictor(dst, stride, width, height, above, left); \

731

67.1k

  }

aom_smooth_predictor_16x16_c

Line

Count

Source

729

49.1k

      const uint8_t *left) {                                   \

730

49.1k

    type##_predictor(dst, stride, width, height, above, left); \

731

49.1k

  }

aom_smooth_predictor_32x32_c

Line

Count

Source

729

38.7k

      const uint8_t *left) {                                   \

730

38.7k

    type##_predictor(dst, stride, width, height, above, left); \

731

38.7k

  }

aom_smooth_predictor_64x64_c

Line

Count

Source

729

7.30k

      const uint8_t *left) {                                   \

730

7.30k

    type##_predictor(dst, stride, width, height, above, left); \

731

7.30k

  }

aom_smooth_predictor_4x8_c

Line

Count

Source

729

24.7k

      const uint8_t *left) {                                   \

730

24.7k

    type##_predictor(dst, stride, width, height, above, left); \

731

24.7k

  }

aom_smooth_predictor_8x4_c

Line

Count

Source

729

31.6k

      const uint8_t *left) {                                   \

730

31.6k

    type##_predictor(dst, stride, width, height, above, left); \

731

31.6k

  }

aom_smooth_predictor_8x16_c

Line

Count

Source

729

22.1k

      const uint8_t *left) {                                   \

730

22.1k

    type##_predictor(dst, stride, width, height, above, left); \

731

22.1k

  }

aom_smooth_predictor_16x8_c

Line

Count

Source

729

33.5k

      const uint8_t *left) {                                   \

730

33.5k

    type##_predictor(dst, stride, width, height, above, left); \

731

33.5k

  }

aom_smooth_predictor_16x32_c

Line

Count

Source

729

7.29k

      const uint8_t *left) {                                   \

730

7.29k

    type##_predictor(dst, stride, width, height, above, left); \

731

7.29k

  }

aom_smooth_predictor_32x16_c

Line

Count

Source

729

8.91k

      const uint8_t *left) {                                   \

730

8.91k

    type##_predictor(dst, stride, width, height, above, left); \

731

8.91k

  }

aom_smooth_predictor_32x64_c

Line

Count

Source

729

570

      const uint8_t *left) {                                   \

730

570

    type##_predictor(dst, stride, width, height, above, left); \

731

570

  }

aom_smooth_predictor_64x32_c

Line

Count

Source

729

526

      const uint8_t *left) {                                   \

730

526

    type##_predictor(dst, stride, width, height, above, left); \

731

526

  }

aom_smooth_predictor_4x16_c

Line

Count

Source

729

22.4k

      const uint8_t *left) {                                   \

730

22.4k

    type##_predictor(dst, stride, width, height, above, left); \

731

22.4k

  }

aom_smooth_predictor_16x4_c

Line

Count

Source

729

30.0k

      const uint8_t *left) {                                   \

730

30.0k

    type##_predictor(dst, stride, width, height, above, left); \

731

30.0k

  }

aom_smooth_predictor_8x32_c

Line

Count

Source

729

7.79k

      const uint8_t *left) {                                   \

730

7.79k

    type##_predictor(dst, stride, width, height, above, left); \

731

7.79k

  }

aom_smooth_predictor_32x8_c

Line

Count

Source

729

9.54k

      const uint8_t *left) {                                   \

730

9.54k

    type##_predictor(dst, stride, width, height, above, left); \

731

9.54k

  }

aom_smooth_predictor_16x64_c

Line

Count

Source

729

731

      const uint8_t *left) {                                   \

730

731

    type##_predictor(dst, stride, width, height, above, left); \

731

731

  }

aom_smooth_predictor_64x16_c

Line

Count

Source

729

672

      const uint8_t *left) {                                   \

730

672

    type##_predictor(dst, stride, width, height, above, left); \

731

672

  }

aom_smooth_v_predictor_4x4_c

Line

Count

Source

729

47.7k

      const uint8_t *left) {                                   \

730

47.7k

    type##_predictor(dst, stride, width, height, above, left); \

731

47.7k

  }

aom_smooth_v_predictor_8x8_c

Line

Count

Source

729

23.5k

      const uint8_t *left) {                                   \

730

23.5k

    type##_predictor(dst, stride, width, height, above, left); \

731

23.5k

  }

aom_smooth_v_predictor_16x16_c

Line

Count

Source

729

11.4k

      const uint8_t *left) {                                   \

730

11.4k

    type##_predictor(dst, stride, width, height, above, left); \

731

11.4k

  }

aom_smooth_v_predictor_32x32_c

Line

Count

Source

729

17.5k

      const uint8_t *left) {                                   \

730

17.5k

    type##_predictor(dst, stride, width, height, above, left); \

731

17.5k

  }

aom_smooth_v_predictor_64x64_c

Line

Count

Source

729

2.32k

      const uint8_t *left) {                                   \

730

2.32k

    type##_predictor(dst, stride, width, height, above, left); \

731

2.32k

  }

aom_smooth_v_predictor_4x8_c

Line

Count

Source

729

7.83k

      const uint8_t *left) {                                   \

730

7.83k

    type##_predictor(dst, stride, width, height, above, left); \

731

7.83k

  }

aom_smooth_v_predictor_8x4_c

Line

Count

Source

729

11.7k

      const uint8_t *left) {                                   \

730

11.7k

    type##_predictor(dst, stride, width, height, above, left); \

731

11.7k

  }

aom_smooth_v_predictor_8x16_c

Line

Count

Source

729

7.70k

      const uint8_t *left) {                                   \

730

7.70k

    type##_predictor(dst, stride, width, height, above, left); \

731

7.70k

  }