reconinter.c

/*
 * 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 "./aom_scale_rtcd.h"
#include "./aom_dsp_rtcd.h"
#include "./aom_config.h"

#include "aom/aom_integer.h"
#include "aom_dsp/blend.h"

#include "av1/common/blockd.h"
#include "av1/common/reconinter.h"
#include "av1/common/reconintra.h"
#if CONFIG_MOTION_VAR
#include "av1/common/onyxc_int.h"
#endif  // CONFIG_MOTION_VAR

#if CONFIG_EXT_INTER

#define NSMOOTHERS 1
#define USE_SOFT_WEIGHTS_IN_WEDGE 1
static int get_masked_weight(int m, int smoothness) {
#define SMOOTHER_LEN 32
  static const uint8_t smoothfn[NSMOOTHERS][2 * SMOOTHER_LEN + 1] = { {
#if USE_SOFT_WEIGHTS_IN_WEDGE
      0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,
      0,  0,  0,  0,  0,  0,  0,  0,  0,  1,  2,  4,  7,  13, 21, 32, 43,
      51, 57, 60, 62, 63, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64,
      64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64,
#else
      0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,
      0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  64, 64, 32, 64,
      64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64,
      64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64,
#endif  // USE_SOFT_WEIGHTS_IN_WEDGE
  } };
  if (m < -SMOOTHER_LEN)
    return 0;
  else if (m > SMOOTHER_LEN)
    return (1 << WEDGE_WEIGHT_BITS);
  else
    return smoothfn[smoothness][m + SMOOTHER_LEN];
}

// [smoother][negative][direction]
DECLARE_ALIGNED(16, static uint8_t,
                wedge_mask_obl[NSMOOTHERS][2][WEDGE_DIRECTIONS]
                              [MASK_MASTER_SIZE * MASK_MASTER_SIZE]);

DECLARE_ALIGNED(16, static uint8_t,
                wedge_signflip_lookup[BLOCK_SIZES][MAX_WEDGE_TYPES]);

// 3 * MAX_WEDGE_SQUARE is an easy to compute and fairly tight upper bound
// on the sum of all mask sizes up to an including MAX_WEDGE_SQUARE.
DECLARE_ALIGNED(16, static uint8_t,
                wedge_mask_buf[2 * MAX_WEDGE_TYPES * 3 * MAX_WEDGE_SQUARE]);

static wedge_masks_type wedge_masks[BLOCK_SIZES][2];

// Some unused wedge codebooks left temporarily to facilitate experiments.
// To be removed when settled.
/*
static wedge_code_type wedge_codebook_8_hgtw[8] = {
  { WEDGE_OBLIQUE27, 4, 4 },  { WEDGE_OBLIQUE63, 4, 4 },
  { WEDGE_OBLIQUE117, 4, 4 }, { WEDGE_OBLIQUE153, 4, 4 },
  { WEDGE_OBLIQUE27, 4, 2 },  { WEDGE_OBLIQUE27, 4, 6 },
  { WEDGE_OBLIQUE153, 4, 2 }, { WEDGE_OBLIQUE153, 4, 6 },
};

static wedge_code_type wedge_codebook_8_hltw[8] = {
  { WEDGE_OBLIQUE27, 4, 4 },  { WEDGE_OBLIQUE63, 4, 4 },
  { WEDGE_OBLIQUE117, 4, 4 }, { WEDGE_OBLIQUE153, 4, 4 },
  { WEDGE_OBLIQUE63, 2, 4 },  { WEDGE_OBLIQUE63, 6, 4 },
  { WEDGE_OBLIQUE117, 2, 4 }, { WEDGE_OBLIQUE117, 6, 4 },
};

static wedge_code_type wedge_codebook_8_heqw[8] = {
  { WEDGE_OBLIQUE27, 4, 4 },  { WEDGE_OBLIQUE63, 4, 4 },
  { WEDGE_OBLIQUE117, 4, 4 }, { WEDGE_OBLIQUE153, 4, 4 },
  { WEDGE_HORIZONTAL, 4, 2 }, { WEDGE_HORIZONTAL, 4, 6 },
  { WEDGE_VERTICAL, 2, 4 },   { WEDGE_VERTICAL, 6, 4 },
};
*/

#if !USE_LARGE_WEDGE_CODEBOOK
static const wedge_code_type wedge_codebook_16_hgtw[16] = {
  { WEDGE_OBLIQUE27, 4, 4 },  { WEDGE_OBLIQUE63, 4, 4 },
  { WEDGE_OBLIQUE117, 4, 4 }, { WEDGE_OBLIQUE153, 4, 4 },
  { WEDGE_HORIZONTAL, 4, 2 }, { WEDGE_HORIZONTAL, 4, 4 },
  { WEDGE_HORIZONTAL, 4, 6 }, { WEDGE_VERTICAL, 4, 4 },
  { WEDGE_OBLIQUE27, 4, 2 },  { WEDGE_OBLIQUE27, 4, 6 },
  { WEDGE_OBLIQUE153, 4, 2 }, { WEDGE_OBLIQUE153, 4, 6 },
  { WEDGE_OBLIQUE63, 2, 4 },  { WEDGE_OBLIQUE63, 6, 4 },
  { WEDGE_OBLIQUE117, 2, 4 }, { WEDGE_OBLIQUE117, 6, 4 },
};

static const wedge_code_type wedge_codebook_16_hltw[16] = {
  { WEDGE_OBLIQUE27, 4, 4 },  { WEDGE_OBLIQUE63, 4, 4 },
  { WEDGE_OBLIQUE117, 4, 4 }, { WEDGE_OBLIQUE153, 4, 4 },
  { WEDGE_VERTICAL, 2, 4 },   { WEDGE_VERTICAL, 4, 4 },
  { WEDGE_VERTICAL, 6, 4 },   { WEDGE_HORIZONTAL, 4, 4 },
  { WEDGE_OBLIQUE27, 4, 2 },  { WEDGE_OBLIQUE27, 4, 6 },
  { WEDGE_OBLIQUE153, 4, 2 }, { WEDGE_OBLIQUE153, 4, 6 },
  { WEDGE_OBLIQUE63, 2, 4 },  { WEDGE_OBLIQUE63, 6, 4 },
  { WEDGE_OBLIQUE117, 2, 4 }, { WEDGE_OBLIQUE117, 6, 4 },
};

static const wedge_code_type wedge_codebook_16_heqw[16] = {
  { WEDGE_OBLIQUE27, 4, 4 },  { WEDGE_OBLIQUE63, 4, 4 },
  { WEDGE_OBLIQUE117, 4, 4 }, { WEDGE_OBLIQUE153, 4, 4 },
  { WEDGE_HORIZONTAL, 4, 2 }, { WEDGE_HORIZONTAL, 4, 6 },
  { WEDGE_VERTICAL, 2, 4 },   { WEDGE_VERTICAL, 6, 4 },
  { WEDGE_OBLIQUE27, 4, 2 },  { WEDGE_OBLIQUE27, 4, 6 },
  { WEDGE_OBLIQUE153, 4, 2 }, { WEDGE_OBLIQUE153, 4, 6 },
  { WEDGE_OBLIQUE63, 2, 4 },  { WEDGE_OBLIQUE63, 6, 4 },
  { WEDGE_OBLIQUE117, 2, 4 }, { WEDGE_OBLIQUE117, 6, 4 },
};

const wedge_params_type wedge_params_lookup[BLOCK_SIZES] = {
#if CONFIG_CB4X4
  { 0, NULL, NULL, 0, NULL },
  { 0, NULL, NULL, 0, NULL },
  { 0, NULL, NULL, 0, NULL },
#endif  // CONFIG_CB4X4
  { 0, NULL, NULL, 0, NULL },
  { 0, NULL, NULL, 0, NULL },
  { 0, NULL, NULL, 0, NULL },
#if CONFIG_WEDGE
  { 4, wedge_codebook_16_heqw, wedge_signflip_lookup[BLOCK_8X8], 0,
    wedge_masks[BLOCK_8X8] },
  { 4, wedge_codebook_16_hgtw, wedge_signflip_lookup[BLOCK_8X16], 0,
    wedge_masks[BLOCK_8X16] },
  { 4, wedge_codebook_16_hltw, wedge_signflip_lookup[BLOCK_16X8], 0,
    wedge_masks[BLOCK_16X8] },
  { 4, wedge_codebook_16_heqw, wedge_signflip_lookup[BLOCK_16X16], 0,
    wedge_masks[BLOCK_16X16] },
  { 4, wedge_codebook_16_hgtw, wedge_signflip_lookup[BLOCK_16X32], 0,
    wedge_masks[BLOCK_16X32] },
  { 4, wedge_codebook_16_hltw, wedge_signflip_lookup[BLOCK_32X16], 0,
    wedge_masks[BLOCK_32X16] },
  { 4, wedge_codebook_16_heqw, wedge_signflip_lookup[BLOCK_32X32], 0,
    wedge_masks[BLOCK_32X32] },
  { 0, wedge_codebook_16_hgtw, wedge_signflip_lookup[BLOCK_32X64], 0,
    wedge_masks[BLOCK_32X64] },
  { 0, wedge_codebook_16_hltw, wedge_signflip_lookup[BLOCK_64X32], 0,
    wedge_masks[BLOCK_64X32] },
  { 0, wedge_codebook_16_heqw, wedge_signflip_lookup[BLOCK_64X64], 0,
    wedge_masks[BLOCK_64X64] },
#else
  { 0, wedge_codebook_16_heqw, wedge_signflip_lookup[BLOCK_8X8], 0,
    wedge_masks[BLOCK_8X8] },
  { 0, wedge_codebook_16_hgtw, wedge_signflip_lookup[BLOCK_8X16], 0,
    wedge_masks[BLOCK_8X16] },
  { 0, wedge_codebook_16_hltw, wedge_signflip_lookup[BLOCK_16X8], 0,
    wedge_masks[BLOCK_16X8] },
  { 0, wedge_codebook_16_heqw, wedge_signflip_lookup[BLOCK_16X16], 0,
    wedge_masks[BLOCK_16X16] },
  { 0, wedge_codebook_16_hgtw, wedge_signflip_lookup[BLOCK_16X32], 0,
    wedge_masks[BLOCK_16X32] },
  { 0, wedge_codebook_16_hltw, wedge_signflip_lookup[BLOCK_32X16], 0,
    wedge_masks[BLOCK_32X16] },
  { 0, wedge_codebook_16_heqw, wedge_signflip_lookup[BLOCK_32X32], 0,
    wedge_masks[BLOCK_32X32] },
  { 0, wedge_codebook_16_hgtw, wedge_signflip_lookup[BLOCK_32X64], 0,
    wedge_masks[BLOCK_32X64] },
  { 0, wedge_codebook_16_hltw, wedge_signflip_lookup[BLOCK_64X32], 0,
    wedge_masks[BLOCK_64X32] },
  { 0, wedge_codebook_16_heqw, wedge_signflip_lookup[BLOCK_64X64], 0,
    wedge_masks[BLOCK_64X64] },
#endif  // CONFIG_WEDGE
#if CONFIG_EXT_PARTITION
  { 0, NULL, NULL, 0, NULL },
  { 0, NULL, NULL, 0, NULL },
  { 0, NULL, NULL, 0, NULL },
#endif  // CONFIG_EXT_PARTITION
};

#else

static const wedge_code_type wedge_codebook_32_hgtw[32] = {
  { WEDGE_OBLIQUE27, 4, 4 },  { WEDGE_OBLIQUE63, 4, 4 },
  { WEDGE_OBLIQUE117, 4, 4 }, { WEDGE_OBLIQUE153, 4, 4 },
  { WEDGE_HORIZONTAL, 4, 2 }, { WEDGE_HORIZONTAL, 4, 4 },
  { WEDGE_HORIZONTAL, 4, 6 }, { WEDGE_VERTICAL, 4, 4 },
  { WEDGE_OBLIQUE27, 4, 1 },  { WEDGE_OBLIQUE27, 4, 2 },
  { WEDGE_OBLIQUE27, 4, 3 },  { WEDGE_OBLIQUE27, 4, 5 },
  { WEDGE_OBLIQUE27, 4, 6 },  { WEDGE_OBLIQUE27, 4, 7 },
  { WEDGE_OBLIQUE153, 4, 1 }, { WEDGE_OBLIQUE153, 4, 2 },
  { WEDGE_OBLIQUE153, 4, 3 }, { WEDGE_OBLIQUE153, 4, 5 },
  { WEDGE_OBLIQUE153, 4, 6 }, { WEDGE_OBLIQUE153, 4, 7 },
  { WEDGE_OBLIQUE63, 1, 4 },  { WEDGE_OBLIQUE63, 2, 4 },
  { WEDGE_OBLIQUE63, 3, 4 },  { WEDGE_OBLIQUE63, 5, 4 },
  { WEDGE_OBLIQUE63, 6, 4 },  { WEDGE_OBLIQUE63, 7, 4 },
  { WEDGE_OBLIQUE117, 1, 4 }, { WEDGE_OBLIQUE117, 2, 4 },
  { WEDGE_OBLIQUE117, 3, 4 }, { WEDGE_OBLIQUE117, 5, 4 },
  { WEDGE_OBLIQUE117, 6, 4 }, { WEDGE_OBLIQUE117, 7, 4 },
};

static const wedge_code_type wedge_codebook_32_hltw[32] = {
  { WEDGE_OBLIQUE27, 4, 4 },  { WEDGE_OBLIQUE63, 4, 4 },
  { WEDGE_OBLIQUE117, 4, 4 }, { WEDGE_OBLIQUE153, 4, 4 },
  { WEDGE_VERTICAL, 2, 4 },   { WEDGE_VERTICAL, 4, 4 },
  { WEDGE_VERTICAL, 6, 4 },   { WEDGE_HORIZONTAL, 4, 4 },
  { WEDGE_OBLIQUE27, 4, 1 },  { WEDGE_OBLIQUE27, 4, 2 },
  { WEDGE_OBLIQUE27, 4, 3 },  { WEDGE_OBLIQUE27, 4, 5 },
  { WEDGE_OBLIQUE27, 4, 6 },  { WEDGE_OBLIQUE27, 4, 7 },
  { WEDGE_OBLIQUE153, 4, 1 }, { WEDGE_OBLIQUE153, 4, 2 },
  { WEDGE_OBLIQUE153, 4, 3 }, { WEDGE_OBLIQUE153, 4, 5 },
  { WEDGE_OBLIQUE153, 4, 6 }, { WEDGE_OBLIQUE153, 4, 7 },
  { WEDGE_OBLIQUE63, 1, 4 },  { WEDGE_OBLIQUE63, 2, 4 },
  { WEDGE_OBLIQUE63, 3, 4 },  { WEDGE_OBLIQUE63, 5, 4 },
  { WEDGE_OBLIQUE63, 6, 4 },  { WEDGE_OBLIQUE63, 7, 4 },
  { WEDGE_OBLIQUE117, 1, 4 }, { WEDGE_OBLIQUE117, 2, 4 },
  { WEDGE_OBLIQUE117, 3, 4 }, { WEDGE_OBLIQUE117, 5, 4 },
  { WEDGE_OBLIQUE117, 6, 4 }, { WEDGE_OBLIQUE117, 7, 4 },
};

static const wedge_code_type wedge_codebook_32_heqw[32] = {
  { WEDGE_OBLIQUE27, 4, 4 },  { WEDGE_OBLIQUE63, 4, 4 },
  { WEDGE_OBLIQUE117, 4, 4 }, { WEDGE_OBLIQUE153, 4, 4 },
  { WEDGE_HORIZONTAL, 4, 2 }, { WEDGE_HORIZONTAL, 4, 6 },
  { WEDGE_VERTICAL, 2, 4 },   { WEDGE_VERTICAL, 6, 4 },
  { WEDGE_OBLIQUE27, 4, 1 },  { WEDGE_OBLIQUE27, 4, 2 },
  { WEDGE_OBLIQUE27, 4, 3 },  { WEDGE_OBLIQUE27, 4, 5 },
  { WEDGE_OBLIQUE27, 4, 6 },  { WEDGE_OBLIQUE27, 4, 7 },
  { WEDGE_OBLIQUE153, 4, 1 }, { WEDGE_OBLIQUE153, 4, 2 },
  { WEDGE_OBLIQUE153, 4, 3 }, { WEDGE_OBLIQUE153, 4, 5 },
  { WEDGE_OBLIQUE153, 4, 6 }, { WEDGE_OBLIQUE153, 4, 7 },
  { WEDGE_OBLIQUE63, 1, 4 },  { WEDGE_OBLIQUE63, 2, 4 },
  { WEDGE_OBLIQUE63, 3, 4 },  { WEDGE_OBLIQUE63, 5, 4 },
  { WEDGE_OBLIQUE63, 6, 4 },  { WEDGE_OBLIQUE63, 7, 4 },
  { WEDGE_OBLIQUE117, 1, 4 }, { WEDGE_OBLIQUE117, 2, 4 },
  { WEDGE_OBLIQUE117, 3, 4 }, { WEDGE_OBLIQUE117, 5, 4 },
  { WEDGE_OBLIQUE117, 6, 4 }, { WEDGE_OBLIQUE117, 7, 4 },
};

const wedge_params_type wedge_params_lookup[BLOCK_SIZES] = {
#if CONFIG_CB4X4
  { 0, NULL, NULL, 0, NULL },
  { 0, NULL, NULL, 0, NULL },
  { 0, NULL, NULL, 0, NULL },
#endif
  { 0, NULL, NULL, 0, NULL },
  { 0, NULL, NULL, 0, NULL },
  { 0, NULL, NULL, 0, NULL },
#if CONFIG_WEDGE
  { 5, wedge_codebook_32_heqw, wedge_signflip_lookup[BLOCK_8X8], 0,
    wedge_masks[BLOCK_8X8] },
  { 5, wedge_codebook_32_hgtw, wedge_signflip_lookup[BLOCK_8X16], 0,
    wedge_masks[BLOCK_8X16] },
  { 5, wedge_codebook_32_hltw, wedge_signflip_lookup[BLOCK_16X8], 0,
    wedge_masks[BLOCK_16X8] },
  { 5, wedge_codebook_32_heqw, wedge_signflip_lookup[BLOCK_16X16], 0,
    wedge_masks[BLOCK_16X16] },
  { 5, wedge_codebook_32_hgtw, wedge_signflip_lookup[BLOCK_16X32], 0,
    wedge_masks[BLOCK_16X32] },
  { 5, wedge_codebook_32_hltw, wedge_signflip_lookup[BLOCK_32X16], 0,
    wedge_masks[BLOCK_32X16] },
  { 5, wedge_codebook_32_heqw, wedge_signflip_lookup[BLOCK_32X32], 0,
    wedge_masks[BLOCK_32X32] },
  { 0, wedge_codebook_32_hgtw, wedge_signflip_lookup[BLOCK_32X64], 0,
    wedge_masks[BLOCK_32X64] },
  { 0, wedge_codebook_32_hltw, wedge_signflip_lookup[BLOCK_64X32], 0,
    wedge_masks[BLOCK_64X32] },
  { 0, wedge_codebook_32_heqw, wedge_signflip_lookup[BLOCK_64X64], 0,
    wedge_masks[BLOCK_64X64] },
#else
  { 0, wedge_codebook_32_heqw, wedge_signflip_lookup[BLOCK_8X8], 0,
    wedge_masks[BLOCK_8X8] },
  { 0, wedge_codebook_32_hgtw, wedge_signflip_lookup[BLOCK_8X16], 0,
    wedge_masks[BLOCK_8X16] },
  { 0, wedge_codebook_32_hltw, wedge_signflip_lookup[BLOCK_16X8], 0,
    wedge_masks[BLOCK_16X8] },
  { 0, wedge_codebook_32_heqw, wedge_signflip_lookup[BLOCK_16X16], 0,
    wedge_masks[BLOCK_16X16] },
  { 0, wedge_codebook_32_hgtw, wedge_signflip_lookup[BLOCK_16X32], 0,
    wedge_masks[BLOCK_16X32] },
  { 0, wedge_codebook_32_hltw, wedge_signflip_lookup[BLOCK_32X16], 0,
    wedge_masks[BLOCK_32X16] },
  { 0, wedge_codebook_32_heqw, wedge_signflip_lookup[BLOCK_32X32], 0,
    wedge_masks[BLOCK_32X32] },
  { 0, wedge_codebook_32_hgtw, wedge_signflip_lookup[BLOCK_32X64], 0,
    wedge_masks[BLOCK_32X64] },
  { 0, wedge_codebook_32_hltw, wedge_signflip_lookup[BLOCK_64X32], 0,
    wedge_masks[BLOCK_64X32] },
  { 0, wedge_codebook_32_heqw, wedge_signflip_lookup[BLOCK_64X64], 0,
    wedge_masks[BLOCK_64X64] },
#endif  // CONFIG_WEDGE
#if CONFIG_EXT_PARTITION
  { 0, NULL, NULL, 0, NULL },
  { 0, NULL, NULL, 0, NULL },
  { 0, NULL, NULL, 0, NULL },
#endif  // CONFIG_EXT_PARTITION
};
#endif  // USE_LARGE_WEDGE_CODEBOOK

static const uint8_t *get_wedge_mask_inplace(int wedge_index, int neg,
                                             BLOCK_SIZE sb_type) {
  const uint8_t *master;
  const int bh = block_size_high[sb_type];
  const int bw = block_size_wide[sb_type];
  const wedge_code_type *a =
      wedge_params_lookup[sb_type].codebook + wedge_index;
  const int smoother = wedge_params_lookup[sb_type].smoother;
  int woff, hoff;
  const uint8_t wsignflip = wedge_params_lookup[sb_type].signflip[wedge_index];

  assert(wedge_index >= 0 &&
         wedge_index < (1 << get_wedge_bits_lookup(sb_type)));
  woff = (a->x_offset * bw) >> 3;
  hoff = (a->y_offset * bh) >> 3;
  master = wedge_mask_obl[smoother][neg ^ wsignflip][a->direction] +
           MASK_MASTER_STRIDE * (MASK_MASTER_SIZE / 2 - hoff) +
           MASK_MASTER_SIZE / 2 - woff;
  return master;
}

const uint8_t *av1_get_soft_mask(int wedge_index, int wedge_sign,
                                 BLOCK_SIZE sb_type, int offset_x,
                                 int offset_y) {
  const uint8_t *mask =
      get_wedge_mask_inplace(wedge_index, wedge_sign, sb_type);
  if (mask) mask -= (offset_x + offset_y * MASK_MASTER_STRIDE);
  return mask;
}

#if CONFIG_COMPOUND_SEGMENT
static uint8_t *invert_mask(uint8_t *mask_inv_buffer, const uint8_t *const mask,
                            int h, int w, int stride) {
  int i, j;

  for (i = 0; i < h; ++i)
    for (j = 0; j < w; ++j) {
      mask_inv_buffer[i * stride + j] =
          AOM_BLEND_A64_MAX_ALPHA - mask[i * stride + j];
    }
  return mask_inv_buffer;
}
#endif  // CONFIG_COMPOUND_SEGMENT

const uint8_t *av1_get_compound_type_mask_inverse(
    const INTERINTER_COMPOUND_DATA *const comp_data,
#if CONFIG_COMPOUND_SEGMENT
    uint8_t *mask_buffer, int h, int w, int stride,
#endif
    BLOCK_SIZE sb_type) {
  assert(is_masked_compound_type(comp_data->type));
  switch (comp_data->type) {
    case COMPOUND_WEDGE:
      return av1_get_contiguous_soft_mask(comp_data->wedge_index,
                                          !comp_data->wedge_sign, sb_type);
#if CONFIG_COMPOUND_SEGMENT
    case COMPOUND_SEG:
      return invert_mask(mask_buffer, comp_data->seg_mask, h, w, stride);
#endif  // CONFIG_COMPOUND_SEGMENT
    default: assert(0); return NULL;
  }
}

const uint8_t *av1_get_compound_type_mask(
    const INTERINTER_COMPOUND_DATA *const comp_data, BLOCK_SIZE sb_type) {
  assert(is_masked_compound_type(comp_data->type));
  switch (comp_data->type) {
    case COMPOUND_WEDGE:
      return av1_get_contiguous_soft_mask(comp_data->wedge_index,
                                          comp_data->wedge_sign, sb_type);
#if CONFIG_COMPOUND_SEGMENT
    case COMPOUND_SEG: return comp_data->seg_mask;
#endif  // CONFIG_COMPOUND_SEGMENT
    default: assert(0); return NULL;
  }
}

#if CONFIG_COMPOUND_SEGMENT
#if COMPOUND_SEGMENT_TYPE == 0
static void uniform_mask(uint8_t *mask, int which_inverse, BLOCK_SIZE sb_type,
                         int h, int w, int mask_val) {
  int i, j;
  int block_stride = block_size_wide[sb_type];
  for (i = 0; i < h; ++i)
    for (j = 0; j < w; ++j) {
      mask[i * block_stride + j] =
          which_inverse ? AOM_BLEND_A64_MAX_ALPHA - mask_val : mask_val;
    }
}

void build_compound_seg_mask(uint8_t *mask, SEG_MASK_TYPE mask_type,
                             const uint8_t *src0, int src0_stride,
                             const uint8_t *src1, int src1_stride,
                             BLOCK_SIZE sb_type, int h, int w) {
  (void)src0;
  (void)src1;
  (void)src0_stride;
  (void)src1_stride;
  switch (mask_type) {
    case UNIFORM_45: uniform_mask(mask, 0, sb_type, h, w, 45); break;
    case UNIFORM_45_INV: uniform_mask(mask, 1, sb_type, h, w, 45); break;
    default: assert(0);
  }
}

#if CONFIG_AOM_HIGHBITDEPTH
void build_compound_seg_mask_highbd(uint8_t *mask, SEG_MASK_TYPE mask_type,
                                    const uint8_t *src0, int src0_stride,
                                    const uint8_t *src1, int src1_stride,
                                    BLOCK_SIZE sb_type, int h, int w, int bd) {
  (void)src0;
  (void)src1;
  (void)src0_stride;
  (void)src1_stride;
  (void)bd;
  switch (mask_type) {
    case UNIFORM_45: uniform_mask(mask, 0, sb_type, h, w, 45); break;
    case UNIFORM_45_INV: uniform_mask(mask, 1, sb_type, h, w, 45); break;
    default: assert(0);
  }
}
#endif  // CONFIG_AOM_HIGHBITDEPTH

#elif COMPOUND_SEGMENT_TYPE == 1
#define DIFF_FACTOR 16
static void diffwtd_mask(uint8_t *mask, int which_inverse, int mask_base,
                         const uint8_t *src0, int src0_stride,
                         const uint8_t *src1, int src1_stride,
                         BLOCK_SIZE sb_type, int h, int w) {
  int i, j, m, diff;
  int block_stride = block_size_wide[sb_type];
  for (i = 0; i < h; ++i) {
    for (j = 0; j < w; ++j) {
      diff =
          abs((int)src0[i * src0_stride + j] - (int)src1[i * src1_stride + j]);
      m = clamp(mask_base + (diff / DIFF_FACTOR), 0, AOM_BLEND_A64_MAX_ALPHA);
      mask[i * block_stride + j] =
          which_inverse ? AOM_BLEND_A64_MAX_ALPHA - m : m;
    }
  }
}

void build_compound_seg_mask(uint8_t *mask, SEG_MASK_TYPE mask_type,
                             const uint8_t *src0, int src0_stride,
                             const uint8_t *src1, int src1_stride,
                             BLOCK_SIZE sb_type, int h, int w) {
  switch (mask_type) {
    case DIFFWTD_42:
      diffwtd_mask(mask, 0, 42, src0, src0_stride, src1, src1_stride, sb_type,
                   h, w);
      break;
    case DIFFWTD_42_INV:
      diffwtd_mask(mask, 1, 42, src0, src0_stride, src1, src1_stride, sb_type,
                   h, w);
      break;
    default: assert(0);
  }
}

#if CONFIG_AOM_HIGHBITDEPTH
static void diffwtd_mask_highbd(uint8_t *mask, int which_inverse, int mask_base,
                                const uint16_t *src0, int src0_stride,
                                const uint16_t *src1, int src1_stride,
                                BLOCK_SIZE sb_type, int h, int w, int bd) {
  int i, j, m, diff;
  int block_stride = block_size_wide[sb_type];
  for (i = 0; i < h; ++i) {
    for (j = 0; j < w; ++j) {
      diff = abs((int)src0[i * src0_stride + j] -
                 (int)src1[i * src1_stride + j]) >>
             (bd - 8);
      m = clamp(mask_base + (diff / DIFF_FACTOR), 0, AOM_BLEND_A64_MAX_ALPHA);
      mask[i * block_stride + j] =
          which_inverse ? AOM_BLEND_A64_MAX_ALPHA - m : m;
    }
  }
}

void build_compound_seg_mask_highbd(uint8_t *mask, SEG_MASK_TYPE mask_type,
                                    const uint8_t *src0, int src0_stride,
                                    const uint8_t *src1, int src1_stride,
                                    BLOCK_SIZE sb_type, int h, int w, int bd) {
  switch (mask_type) {
    case DIFFWTD_42:
      diffwtd_mask_highbd(mask, 0, 42, CONVERT_TO_SHORTPTR(src0), src0_stride,
                          CONVERT_TO_SHORTPTR(src1), src1_stride, sb_type, h, w,
                          bd);
      break;
    case DIFFWTD_42_INV:
      diffwtd_mask_highbd(mask, 1, 42, CONVERT_TO_SHORTPTR(src0), src0_stride,
                          CONVERT_TO_SHORTPTR(src1), src1_stride, sb_type, h, w,
                          bd);
      break;
    default: assert(0);
  }
}
#endif  // CONFIG_AOM_HIGHBITDEPTH
#endif  // COMPOUND_SEGMENT_TYPE
#endif  // CONFIG_COMPOUND_SEGMENT

static void init_wedge_master_masks() {
  int i, j, s;
  const int w = MASK_MASTER_SIZE;
  const int h = MASK_MASTER_SIZE;
  const int stride = MASK_MASTER_STRIDE;
  const int a[2] = { 2, 1 };
  const double asqrt = sqrt(a[0] * a[0] + a[1] * a[1]);
  for (s = 0; s < NSMOOTHERS; s++) {
    for (i = 0; i < h; ++i)
      for (j = 0; j < w; ++j) {
        int x = (2 * j + 1 - w);
        int y = (2 * i + 1 - h);
        int m = (int)rint((a[0] * x + a[1] * y) / asqrt);
        wedge_mask_obl[s][1][WEDGE_OBLIQUE63][i * stride + j] =
            wedge_mask_obl[s][1][WEDGE_OBLIQUE27][j * stride + i] =
                get_masked_weight(m, s);
        wedge_mask_obl[s][1][WEDGE_OBLIQUE117][i * stride + w - 1 - j] =
            wedge_mask_obl[s][1][WEDGE_OBLIQUE153][(w - 1 - j) * stride + i] =
                (1 << WEDGE_WEIGHT_BITS) - get_masked_weight(m, s);
        wedge_mask_obl[s][0][WEDGE_OBLIQUE63][i * stride + j] =
            wedge_mask_obl[s][0][WEDGE_OBLIQUE27][j * stride + i] =
                (1 << WEDGE_WEIGHT_BITS) - get_masked_weight(m, s);
        wedge_mask_obl[s][0][WEDGE_OBLIQUE117][i * stride + w - 1 - j] =
            wedge_mask_obl[s][0][WEDGE_OBLIQUE153][(w - 1 - j) * stride + i] =
                get_masked_weight(m, s);
        wedge_mask_obl[s][1][WEDGE_VERTICAL][i * stride + j] =
            wedge_mask_obl[s][1][WEDGE_HORIZONTAL][j * stride + i] =
                get_masked_weight(x, s);
        wedge_mask_obl[s][0][WEDGE_VERTICAL][i * stride + j] =
            wedge_mask_obl[s][0][WEDGE_HORIZONTAL][j * stride + i] =
                (1 << WEDGE_WEIGHT_BITS) - get_masked_weight(x, s);
      }
  }
}

// If the signs for the wedges for various blocksizes are
// inconsistent flip the sign flag. Do it only once for every
// wedge codebook.
static void init_wedge_signs() {
  BLOCK_SIZE sb_type;
  memset(wedge_signflip_lookup, 0, sizeof(wedge_signflip_lookup));
  for (sb_type = BLOCK_4X4; sb_type < BLOCK_SIZES; ++sb_type) {
    const int bw = block_size_wide[sb_type];
    const int bh = block_size_high[sb_type];
    const wedge_params_type wedge_params = wedge_params_lookup[sb_type];
    const int wbits = wedge_params.bits;
    const int wtypes = 1 << wbits;
    int i, w;
    if (wbits == 0) continue;
    for (w = 0; w < wtypes; ++w) {
      const uint8_t *mask = get_wedge_mask_inplace(w, 0, sb_type);
      int sum = 0;
      for (i = 0; i < bw; ++i) sum += mask[i];
      for (i = 0; i < bh; ++i) sum += mask[i * MASK_MASTER_STRIDE];
      sum = (sum + (bw + bh) / 2) / (bw + bh);
      wedge_params.signflip[w] = (sum < 32);
    }
  }
}

static void init_wedge_masks() {
  uint8_t *dst = wedge_mask_buf;
  BLOCK_SIZE bsize;
  memset(wedge_masks, 0, sizeof(wedge_masks));
  for (bsize = BLOCK_4X4; bsize < BLOCK_SIZES; ++bsize) {
    const uint8_t *mask;
    const int bw = block_size_wide[bsize];
    const int bh = block_size_high[bsize];
    const wedge_params_type *wedge_params = &wedge_params_lookup[bsize];
    const int wbits = wedge_params->bits;
    const int wtypes = 1 << wbits;
    int w;
    if (wbits == 0) continue;
    for (w = 0; w < wtypes; ++w) {
      mask = get_wedge_mask_inplace(w, 0, bsize);
      aom_convolve_copy(mask, MASK_MASTER_STRIDE, dst, bw, NULL, 0, NULL, 0, bw,
                        bh);
      wedge_params->masks[0][w] = dst;
      dst += bw * bh;

      mask = get_wedge_mask_inplace(w, 1, bsize);
      aom_convolve_copy(mask, MASK_MASTER_STRIDE, dst, bw, NULL, 0, NULL, 0, bw,
                        bh);
      wedge_params->masks[1][w] = dst;
      dst += bw * bh;
    }
    assert(sizeof(wedge_mask_buf) >= (size_t)(dst - wedge_mask_buf));
  }
}

// Equation of line: f(x, y) = a[0]*(x - a[2]*w/8) + a[1]*(y - a[3]*h/8) = 0
void av1_init_wedge_masks() {
  init_wedge_master_masks();
  init_wedge_signs();
  init_wedge_masks();
}

#if CONFIG_SUPERTX
static void build_masked_compound_wedge_extend(
    uint8_t *dst, int dst_stride, const uint8_t *src0, int src0_stride,
    const uint8_t *src1, int src1_stride,
    const INTERINTER_COMPOUND_DATA *const comp_data, BLOCK_SIZE sb_type,
    int wedge_offset_x, int wedge_offset_y, int h, int w) {
  const int subh = (2 << b_height_log2_lookup[sb_type]) == h;
  const int subw = (2 << b_width_log2_lookup[sb_type]) == w;
  const uint8_t *mask;
  size_t mask_stride;
  switch (comp_data->type) {
    case COMPOUND_WEDGE:
      mask = av1_get_soft_mask(comp_data->wedge_index, comp_data->wedge_sign,
                               sb_type, wedge_offset_x, wedge_offset_y);
      mask_stride = MASK_MASTER_STRIDE;
      break;
#if CONFIG_COMPOUND_SEGMENT
    case COMPOUND_SEG:
      mask = comp_data->seg_mask;
      mask_stride = block_size_wide[sb_type];
      break;
#endif
    default: assert(0); return;
  }
  aom_blend_a64_mask(dst, dst_stride, src0, src0_stride, src1, src1_stride,
                     mask, mask_stride, h, w, subh, subw);
}

#if CONFIG_AOM_HIGHBITDEPTH
static void build_masked_compound_wedge_extend_highbd(
    uint8_t *dst_8, int dst_stride, const uint8_t *src0_8, int src0_stride,
    const uint8_t *src1_8, int src1_stride,
    const INTERINTER_COMPOUND_DATA *const comp_data, BLOCK_SIZE sb_type,
    int wedge_offset_x, int wedge_offset_y, int h, int w, int bd) {
  const int subh = (2 << b_height_log2_lookup[sb_type]) == h;
  const int subw = (2 << b_width_log2_lookup[sb_type]) == w;
  const uint8_t *mask;
  size_t mask_stride;
  switch (comp_data->type) {
    case COMPOUND_WEDGE:
      mask = av1_get_soft_mask(comp_data->wedge_index, comp_data->wedge_sign,
                               sb_type, wedge_offset_x, wedge_offset_y);
      mask_stride = MASK_MASTER_STRIDE;
      break;
#if CONFIG_COMPOUND_SEGMENT
    case COMPOUND_SEG:
      mask = comp_data->seg_mask;
      mask_stride = block_size_wide[sb_type];
      break;
#endif
    default: assert(0); return;
  }
  aom_highbd_blend_a64_mask(dst_8, dst_stride, src0_8, src0_stride, src1_8,
                            src1_stride, mask, mask_stride, h, w, subh, subw,
                            bd);
}
#endif  // CONFIG_AOM_HIGHBITDEPTH
#else
static void build_masked_compound(
    uint8_t *dst, int dst_stride, const uint8_t *src0, int src0_stride,
    const uint8_t *src1, int src1_stride,
    const INTERINTER_COMPOUND_DATA *const comp_data, BLOCK_SIZE sb_type, int h,
    int w) {
  // Derive subsampling from h and w passed in. May be refactored to
  // pass in subsampling factors directly.
  const int subh = (2 << b_height_log2_lookup[sb_type]) == h;
  const int subw = (2 << b_width_log2_lookup[sb_type]) == w;
  const uint8_t *mask = av1_get_compound_type_mask(comp_data, sb_type);
  aom_blend_a64_mask(dst, dst_stride, src0, src0_stride, src1, src1_stride,
                     mask, block_size_wide[sb_type], h, w, subh, subw);
}

#if CONFIG_AOM_HIGHBITDEPTH
static void build_masked_compound_highbd(
    uint8_t *dst_8, int dst_stride, const uint8_t *src0_8, int src0_stride,
    const uint8_t *src1_8, int src1_stride,
    const INTERINTER_COMPOUND_DATA *const comp_data, BLOCK_SIZE sb_type, int h,
    int w, int bd) {
  // Derive subsampling from h and w passed in. May be refactored to
  // pass in subsampling factors directly.
  const int subh = (2 << b_height_log2_lookup[sb_type]) == h;
  const int subw = (2 << b_width_log2_lookup[sb_type]) == w;
  const uint8_t *mask = av1_get_compound_type_mask(comp_data, sb_type);
  // const uint8_t *mask =
  //     av1_get_contiguous_soft_mask(wedge_index, wedge_sign, sb_type);
  aom_highbd_blend_a64_mask(dst_8, dst_stride, src0_8, src0_stride, src1_8,
                            src1_stride, mask, block_size_wide[sb_type], h, w,
                            subh, subw, bd);
}
#endif  // CONFIG_AOM_HIGHBITDEPTH
#endif  // CONFIG_SUPERTX

void av1_make_masked_inter_predictor(const uint8_t *pre, int pre_stride,
                                     uint8_t *dst, int dst_stride,
                                     const int subpel_x, const int subpel_y,
                                     const struct scale_factors *sf, int w,
                                     int h,
#if CONFIG_DUAL_FILTER
                                     const InterpFilter *interp_filter,
#else
                                     const InterpFilter interp_filter,
#endif
                                     int xs, int ys,
#if CONFIG_SUPERTX
                                     int wedge_offset_x, int wedge_offset_y,
#endif  // CONFIG_SUPERTX
                                     int plane,
#if CONFIG_GLOBAL_MOTION
                                     int is_global, int p_col, int p_row,
                                     int ref,
#endif  // CONFIG_GLOBAL_MOTION
                                     MACROBLOCKD *xd) {
  MODE_INFO *mi = xd->mi[0];
  INTERINTER_COMPOUND_DATA *comp_data = &mi->mbmi.interinter_compound_data;
// The prediction filter types used here should be those for
// the second reference block.
#if CONFIG_DUAL_FILTER
  InterpFilter tmp_ipf[4] = {
    interp_filter[2], interp_filter[3], interp_filter[2], interp_filter[3],
  };
#else
  InterpFilter tmp_ipf = interp_filter;
#endif  // CONFIG_DUAL_FILTER
  ConvolveParams conv_params = get_conv_params(0, plane);

#if CONFIG_AOM_HIGHBITDEPTH
  DECLARE_ALIGNED(16, uint8_t, tmp_dst_[2 * MAX_SB_SQUARE]);
  uint8_t *tmp_dst = (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH)
                         ? CONVERT_TO_BYTEPTR(tmp_dst_)
                         : tmp_dst_;
  av1_make_inter_predictor(pre, pre_stride, tmp_dst, MAX_SB_SIZE, subpel_x,
                           subpel_y, sf, w, h, &conv_params, tmp_ipf,
#if CONFIG_GLOBAL_MOTION
                           is_global, p_col, p_row, plane, ref,
#if CONFIG_MOTION_VAR
                           0, 0,
#endif
#endif  // CONFIG_GLOBAL_MOTION
                           xs, ys, xd);
#if CONFIG_COMPOUND_SEGMENT
  if (!plane && comp_data->type == COMPOUND_SEG) {
    if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH)
      build_compound_seg_mask_highbd(comp_data->seg_mask, comp_data->mask_type,
                                     dst, dst_stride, tmp_dst, MAX_SB_SIZE,
                                     mi->mbmi.sb_type, h, w, xd->bd);
    else
      build_compound_seg_mask(comp_data->seg_mask, comp_data->mask_type, dst,
                              dst_stride, tmp_dst, MAX_SB_SIZE,
                              mi->mbmi.sb_type, h, w);
  }
#endif  // CONFIG_COMPOUND_SEGMENT

#if CONFIG_SUPERTX
  if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH)
    build_masked_compound_wedge_extend_highbd(
        dst, dst_stride, dst, dst_stride, tmp_dst, MAX_SB_SIZE, comp_data,
        mi->mbmi.sb_type, wedge_offset_x, wedge_offset_y, h, w, xd->bd);
  else
    build_masked_compound_wedge_extend(
        dst, dst_stride, dst, dst_stride, tmp_dst, MAX_SB_SIZE, comp_data,
        mi->mbmi.sb_type, wedge_offset_x, wedge_offset_y, h, w);
#else
  if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH)
    build_masked_compound_highbd(dst, dst_stride, dst, dst_stride, tmp_dst,
                                 MAX_SB_SIZE, comp_data, mi->mbmi.sb_type, h, w,
                                 xd->bd);
  else
    build_masked_compound(dst, dst_stride, dst, dst_stride, tmp_dst,
                          MAX_SB_SIZE, comp_data, mi->mbmi.sb_type, h, w);
#endif  // CONFIG_SUPERTX

#else  // CONFIG_AOM_HIGHBITDEPTH
  DECLARE_ALIGNED(16, uint8_t, tmp_dst[MAX_SB_SQUARE]);
  av1_make_inter_predictor(pre, pre_stride, tmp_dst, MAX_SB_SIZE, subpel_x,
                           subpel_y, sf, w, h, &conv_params, tmp_ipf,
#if CONFIG_GLOBAL_MOTION
                           is_global, p_col, p_row, plane, ref,
#if CONFIG_MOTION_VAR
                           0, 0,
#endif
#endif  // CONFIG_GLOBAL_MOTION
                           xs, ys, xd);
#if CONFIG_COMPOUND_SEGMENT
  if (!plane && comp_data->type == COMPOUND_SEG)
    build_compound_seg_mask(comp_data->seg_mask, comp_data->mask_type, dst,
                            dst_stride, tmp_dst, MAX_SB_SIZE, mi->mbmi.sb_type,
                            h, w);
#endif  // CONFIG_COMPOUND_SEGMENT
#if CONFIG_SUPERTX
  build_masked_compound_wedge_extend(dst, dst_stride, dst, dst_stride, tmp_dst,
                                     MAX_SB_SIZE, comp_data, mi->mbmi.sb_type,
                                     wedge_offset_x, wedge_offset_y, h, w);
#else
  build_masked_compound(dst, dst_stride, dst, dst_stride, tmp_dst, MAX_SB_SIZE,
                        comp_data, mi->mbmi.sb_type, h, w);
#endif  // CONFIG_SUPERTX
#endif  // CONFIG_AOM_HIGHBITDEPTH
#if CONFIG_COMPOUND_SEGMENT
  (void)plane;
#endif  // CONFIG_COMPOUND_SEGMENT
}
#endif  // CONFIG_EXT_INTER

// TODO(sarahparker) av1_highbd_build_inter_predictor and
// av1_build_inter_predictor should be combined with
// av1_make_inter_predictor
#if CONFIG_AOM_HIGHBITDEPTH
void av1_highbd_build_inter_predictor(const uint8_t *src, int src_stride,
                                      uint8_t *dst, int dst_stride,
                                      const MV *src_mv,
                                      const struct scale_factors *sf, int w,
                                      int h, int ref,
#if CONFIG_DUAL_FILTER
                                      const InterpFilter *interp_filter,
#else
                                      const InterpFilter interp_filter,
#endif
#if CONFIG_GLOBAL_MOTION
                                      int is_global, int p_col, int p_row,
#endif  // CONFIG_GLOBAL_MOTION
                                      int plane, enum mv_precision precision,
                                      int x, int y, const MACROBLOCKD *xd) {
  const int is_q4 = precision == MV_PRECISION_Q4;
  const MV mv_q4 = { is_q4 ? src_mv->row : src_mv->row * 2,
                     is_q4 ? src_mv->col : src_mv->col * 2 };
  MV32 mv = av1_scale_mv(&mv_q4, x, y, sf);
  const int subpel_x = mv.col & SUBPEL_MASK;
  const int subpel_y = mv.row & SUBPEL_MASK;
  ConvolveParams conv_params = get_conv_params(ref, plane);

  src += (mv.row >> SUBPEL_BITS) * src_stride + (mv.col >> SUBPEL_BITS);

  av1_make_inter_predictor(src, src_stride, dst, dst_stride, subpel_x, subpel_y,
                           sf, w, h, &conv_params, interp_filter,
#if CONFIG_GLOBAL_MOTION
                           is_global, p_col, p_row, plane, ref,
#if CONFIG_MOTION_VAR
                           0, 0,
#endif
#endif  // CONFIG_GLOBAL_MOTION
                           sf->x_step_q4, sf->y_step_q4, xd);
}
#endif  // CONFIG_AOM_HIGHBITDEPTH

void av1_build_inter_predictor(const uint8_t *src, int src_stride, uint8_t *dst,
                               int dst_stride, const MV *src_mv,
                               const struct scale_factors *sf, int w, int h,
                               ConvolveParams *conv_params,
#if CONFIG_DUAL_FILTER
                               const InterpFilter *interp_filter,
#else
                               const InterpFilter interp_filter,
#endif
#if CONFIG_GLOBAL_MOTION
                               int is_global, int p_col, int p_row, int plane,
                               int ref,
#endif  // CONFIG_GLOBAL_MOTION
                               enum mv_precision precision, int x, int y,
                               const MACROBLOCKD *xd) {
  const int is_q4 = precision == MV_PRECISION_Q4;
  const MV mv_q4 = { is_q4 ? src_mv->row : src_mv->row * 2,
                     is_q4 ? src_mv->col : src_mv->col * 2 };
  MV32 mv = av1_scale_mv(&mv_q4, x, y, sf);
  const int subpel_x = mv.col & SUBPEL_MASK;
  const int subpel_y = mv.row & SUBPEL_MASK;

  src += (mv.row >> SUBPEL_BITS) * src_stride + (mv.col >> SUBPEL_BITS);

  av1_make_inter_predictor(src, src_stride, dst, dst_stride, subpel_x, subpel_y,
                           sf, w, h, conv_params, interp_filter,
#if CONFIG_GLOBAL_MOTION
                           is_global, p_col, p_row, plane, ref,
#if CONFIG_MOTION_VAR
                           0, 0,
#endif
#endif  // CONFIG_GLOBAL_MOTION
                           sf->x_step_q4, sf->y_step_q4, xd);
}

typedef struct SubpelParams {
  int xs;
  int ys;
  int subpel_x;
  int subpel_y;
} SubpelParams;

void build_inter_predictors(MACROBLOCKD *xd, int plane,
#if CONFIG_MOTION_VAR
                            int mi_col_offset, int mi_row_offset,
#endif  // CONFIG_MOTION_VAR
                            int block, int bw, int bh, int x, int y, int w,
                            int h,
#if CONFIG_SUPERTX && CONFIG_EXT_INTER
                            int wedge_offset_x, int wedge_offset_y,
#endif  // CONFIG_SUPERTX && CONFIG_EXT_INTER
                            int mi_x, int mi_y) {
  struct macroblockd_plane *const pd = &xd->plane[plane];
#if CONFIG_MOTION_VAR
  const MODE_INFO *mi = xd->mi[mi_col_offset + xd->mi_stride * mi_row_offset];
#if !CONFIG_CB4X4
  const int build_for_obmc = !(mi_col_offset == 0 && mi_row_offset == 0);
#endif
#else
  const MODE_INFO *mi = xd->mi[0];
#endif  // CONFIG_MOTION_VAR
  const int is_compound = has_second_ref(&mi->mbmi);
  int ref;
#if CONFIG_GLOBAL_MOTION
  int is_global[2];
  for (ref = 0; ref < 1 + is_compound; ++ref) {
    WarpedMotionParams *const wm = &xd->global_motion[mi->mbmi.ref_frame[ref]];
    is_global[ref] = is_global_mv_block(mi, block, wm->wmtype);
  }
#endif  // CONFIG_GLOBAL_MOTION

#if CONFIG_CB4X4
  (void)block;
#endif

#if CONFIG_SUB8X8_MC
#if CONFIG_MOTION_VAR
  if (mi->mbmi.sb_type < BLOCK_8X8 && plane > 0 && !build_for_obmc) {
#else
  if (mi->mbmi.sb_type < BLOCK_8X8 && plane > 0) {
#endif  // CONFIG_MOTION_VAR
    // block size in log2
    const int b4_wl = b_width_log2_lookup[mi->mbmi.sb_type];
    const int b4_hl = b_height_log2_lookup[mi->mbmi.sb_type];
    const int b8_sl = b_width_log2_lookup[BLOCK_8X8];

    // block size
    const int b4_w = 1 << b4_wl;
    const int b4_h = 1 << b4_hl;
    const int b8_s = 1 << b8_sl;
    int idx, idy;

    const int x_base = x;
    const int y_base = y;

    // processing unit size
    const int x_step = w >> (b8_sl - b4_wl);
    const int y_step = h >> (b8_sl - b4_hl);

    for (idy = 0; idy < b8_s; idy += b4_h) {
      for (idx = 0; idx < b8_s; idx += b4_w) {
        const int chr_idx = (idy * 2) + idx;
        for (ref = 0; ref < 1 + is_compound; ++ref) {
          const struct scale_factors *const sf = &xd->block_refs[ref]->sf;
          struct buf_2d *const pre_buf = &pd->pre[ref];
          struct buf_2d *const dst_buf = &pd->dst;
          uint8_t *dst = dst_buf->buf;
          const MV mv = mi->bmi[chr_idx].as_mv[ref].as_mv;
          const MV mv_q4 = clamp_mv_to_umv_border_sb(
              xd, &mv, bw, bh, pd->subsampling_x, pd->subsampling_y);
          uint8_t *pre;
          MV32 scaled_mv;
          int xs, ys, subpel_x, subpel_y;
          const int is_scaled = av1_is_scaled(sf);
          ConvolveParams conv_params = get_conv_params(ref, plane);

          x = x_base + idx * x_step;
          y = y_base + idy * y_step;

          dst += dst_buf->stride * y + x;

          if (is_scaled) {
            pre =
                pre_buf->buf + scaled_buffer_offset(x, y, pre_buf->stride, sf);
            scaled_mv = av1_scale_mv(&mv_q4, mi_x + x, mi_y + y, sf);
            xs = sf->x_step_q4;
            ys = sf->y_step_q4;
          } else {
            pre = pre_buf->buf + y * pre_buf->stride + x;
            scaled_mv.row = mv_q4.row;
            scaled_mv.col = mv_q4.col;
            xs = ys = 16;
          }

          subpel_x = scaled_mv.col & SUBPEL_MASK;
          subpel_y = scaled_mv.row & SUBPEL_MASK;
          pre += (scaled_mv.row >> SUBPEL_BITS) * pre_buf->stride +
                 (scaled_mv.col >> SUBPEL_BITS);

#if CONFIG_EXT_INTER
          if (ref &&
              is_masked_compound_type(mi->mbmi.interinter_compound_data.type))
            av1_make_masked_inter_predictor(
                pre, pre_buf->stride, dst, dst_buf->stride, subpel_x, subpel_y,
                sf, w, h, mi->mbmi.interp_filter, xs, ys,
#if CONFIG_SUPERTX
                wedge_offset_x, wedge_offset_y,
#endif  // CONFIG_SUPERTX
                plane,
#if CONFIG_GLOBAL_MOTION
                is_global[ref], (mi_x >> pd->subsampling_x) + x,
                (mi_y >> pd->subsampling_y) + y, ref
#endif  // CONFIG_GLOBAL_MOTION