/*
 * 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 <stdio.h>
#include <limits.h>

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

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

#include "av1/common/av1_common_int.h"
#include "av1/common/blockd.h"
#include "av1/common/mvref_common.h"
#include "av1/common/obmc.h"
#include "av1/common/reconinter.h"
#include "av1/common/reconintra.h"

// This function will determine whether or not to create a warped
// prediction.
static int allow_warp(const MB_MODE_INFO *const mbmi,
                      const WarpTypesAllowed *const warp_types,
                      const WarpedMotionParams *const gm_params,
                      int build_for_obmc, const struct scale_factors *const sf,
                      WarpedMotionParams *final_warp_params) {
  // Note: As per the spec, we must test the fixed point scales here, which are
  // at a higher precision (1 << 14) than the xs and ys in subpel_params (that
  // have 1 << 10 precision).
  if (av1_is_scaled(sf)) return 0;

  if (final_warp_params != NULL) *final_warp_params = default_warp_params;

  if (build_for_obmc) return 0;

  if (warp_types->local_warp_allowed && !mbmi->wm_params.invalid) {
    if (final_warp_params != NULL) *final_warp_params = mbmi->wm_params;
    return 1;
  } else if (warp_types->global_warp_allowed && !gm_params->invalid) {
    if (final_warp_params != NULL) *final_warp_params = *gm_params;
    return 1;
  }

  return 0;
}

void av1_init_warp_params(InterPredParams *inter_pred_params,
                          const WarpTypesAllowed *warp_types, int ref,
                          const MACROBLOCKD *xd, const MB_MODE_INFO *mi) {
  if (inter_pred_params->block_height < 8 || inter_pred_params->block_width < 8)
    return;

  if (xd->cur_frame_force_integer_mv) return;

  if (allow_warp(mi, warp_types, &xd->global_motion[mi->ref_frame[ref]], 0,
                 inter_pred_params->scale_factors,
                 &inter_pred_params->warp_params)) {
#if CONFIG_REALTIME_ONLY && !CONFIG_AV1_DECODER
    aom_internal_error(xd->error_info, AOM_CODEC_UNSUP_FEATURE,
                       "Warped motion is disabled in realtime only build.");
#endif  // CONFIG_REALTIME_ONLY && !CONFIG_AV1_DECODER
    inter_pred_params->mode = WARP_PRED;
  }
}

void av1_make_inter_predictor(const uint8_t *src, int src_stride, uint8_t *dst,
                              int dst_stride,
                              InterPredParams *inter_pred_params,
                              const SubpelParams *subpel_params) {
  assert(IMPLIES(inter_pred_params->conv_params.is_compound,
                 inter_pred_params->conv_params.dst != NULL));

  if (inter_pred_params->mode == TRANSLATION_PRED) {
#if CONFIG_AV1_HIGHBITDEPTH
    if (inter_pred_params->use_hbd_buf) {
      highbd_inter_predictor(src, src_stride, dst, dst_stride, subpel_params,
                             inter_pred_params->block_width,
                             inter_pred_params->block_height,
                             &inter_pred_params->conv_params,
                             inter_pred_params->interp_filter_params,
                             inter_pred_params->bit_depth);
    } else {
      inter_predictor(src, src_stride, dst, dst_stride, subpel_params,
                      inter_pred_params->block_width,
                      inter_pred_params->block_height,
                      &inter_pred_params->conv_params,
                      inter_pred_params->interp_filter_params);
    }
#else
    inter_predictor(src, src_stride, dst, dst_stride, subpel_params,
                    inter_pred_params->block_width,
                    inter_pred_params->block_height,
                    &inter_pred_params->conv_params,
                    inter_pred_params->interp_filter_params);
#endif
  }
#if !CONFIG_REALTIME_ONLY || CONFIG_AV1_DECODER
  // TODO(jingning): av1_warp_plane() can be further cleaned up.
  else if (inter_pred_params->mode == WARP_PRED) {
    av1_warp_plane(
        &inter_pred_params->warp_params, inter_pred_params->use_hbd_buf,
        inter_pred_params->bit_depth, inter_pred_params->ref_frame_buf.buf0,
        inter_pred_params->ref_frame_buf.width,
        inter_pred_params->ref_frame_buf.height,
        inter_pred_params->ref_frame_buf.stride, dst,
        inter_pred_params->pix_col, inter_pred_params->pix_row,
        inter_pred_params->block_width, inter_pred_params->block_height,
        dst_stride, inter_pred_params->subsampling_x,
        inter_pred_params->subsampling_y, &inter_pred_params->conv_params);
  }
#endif  // !CONFIG_REALTIME_ONLY || CONFIG_AV1_DECODER
  else {
    assert(0 && "Unsupported inter_pred_params->mode");
  }
}

/* clang-format off */
DECLARE_ALIGNED(16, static const uint8_t,
                wedge_signflip_lookup[BLOCK_SIZES_ALL][MAX_WEDGE_TYPES]) = {
  { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, },  // not used
  { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, },  // not used
  { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, },  // not used
  { 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 1, 1, 1, 0, 1, },
  { 1, 1, 1, 1, 0, 1, 1, 1, 1, 1, 0, 1, 1, 1, 0, 1, },
  { 1, 1, 1, 1, 0, 1, 1, 1, 1, 1, 0, 1, 1, 1, 0, 1, },
  { 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 1, 1, 1, 0, 1, },
  { 1, 1, 1, 1, 0, 1, 1, 1, 1, 1, 0, 1, 1, 1, 0, 1, },
  { 1, 1, 1, 1, 0, 1, 1, 1, 1, 1, 0, 1, 1, 1, 0, 1, },
  { 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 1, 1, 1, 0, 1, },
  { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, },  // not used
  { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, },  // not used
  { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, },  // not used
  { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, },  // not used
  { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, },  // not used
  { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, },  // not used
  { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, },  // not used
  { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, },  // not used
  { 1, 1, 1, 1, 0, 1, 1, 1, 0, 1, 0, 1, 1, 1, 0, 1, },
  { 1, 1, 1, 1, 0, 1, 1, 1, 1, 1, 0, 1, 0, 1, 0, 1, },
  { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, },  // not used
  { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, },  // not used
};
/* clang-format on */

// The wedge / inter-intra mask buffers used to be writable .data populated
// by init_all_wedge_masks() at runtime, which gave every process its own
// private COW copy. They are now precomputed at codegen time (see
// tools/gen_wedge_masks_data.py) and included here as const data so they
// live in .rdata and are shared across processes.
//
// `wedge_masks` stores byte offsets into `wedge_mask_buf` (not pointers) to
// avoid loader relocations that would otherwise dirty the .rdata pages.
#include "av1/common/wedge_masks_data.inc"

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 av1_wedge_params_lookup[BLOCK_SIZES_ALL] = {
  { 0, NULL, NULL, NULL },
  { 0, NULL, NULL, NULL },
  { 0, NULL, NULL, NULL },
  { MAX_WEDGE_TYPES, wedge_codebook_16_heqw, wedge_signflip_lookup[BLOCK_8X8],
    wedge_masks[BLOCK_8X8] },
  { MAX_WEDGE_TYPES, wedge_codebook_16_hgtw, wedge_signflip_lookup[BLOCK_8X16],
    wedge_masks[BLOCK_8X16] },
  { MAX_WEDGE_TYPES, wedge_codebook_16_hltw, wedge_signflip_lookup[BLOCK_16X8],
    wedge_masks[BLOCK_16X8] },
  { MAX_WEDGE_TYPES, wedge_codebook_16_heqw, wedge_signflip_lookup[BLOCK_16X16],
    wedge_masks[BLOCK_16X16] },
  { MAX_WEDGE_TYPES, wedge_codebook_16_hgtw, wedge_signflip_lookup[BLOCK_16X32],
    wedge_masks[BLOCK_16X32] },
  { MAX_WEDGE_TYPES, wedge_codebook_16_hltw, wedge_signflip_lookup[BLOCK_32X16],
    wedge_masks[BLOCK_32X16] },
  { MAX_WEDGE_TYPES, wedge_codebook_16_heqw, wedge_signflip_lookup[BLOCK_32X32],
    wedge_masks[BLOCK_32X32] },
  { 0, NULL, NULL, NULL },
  { 0, NULL, NULL, NULL },
  { 0, NULL, NULL, NULL },
  { 0, NULL, NULL, NULL },
  { 0, NULL, NULL, NULL },
  { 0, NULL, NULL, NULL },
  { 0, NULL, NULL, NULL },
  { 0, NULL, NULL, NULL },
  { MAX_WEDGE_TYPES, wedge_codebook_16_hgtw, wedge_signflip_lookup[BLOCK_8X32],
    wedge_masks[BLOCK_8X32] },
  { MAX_WEDGE_TYPES, wedge_codebook_16_hltw, wedge_signflip_lookup[BLOCK_32X8],
    wedge_masks[BLOCK_32X8] },
  { 0, NULL, NULL, NULL },
  { 0, NULL, NULL, NULL },
};

const uint8_t *av1_get_contiguous_soft_mask(int8_t wedge_index,
                                            int8_t wedge_sign,
                                            BLOCK_SIZE sb_type) {
  return wedge_mask_buf +
         av1_wedge_params_lookup[sb_type].masks[wedge_sign][wedge_index];
}

const uint8_t *av1_get_compound_type_mask(
    const INTERINTER_COMPOUND_DATA *const comp_data, BLOCK_SIZE sb_type) {
  (void)sb_type;
  switch (comp_data->type) {
    case COMPOUND_WEDGE:
      return av1_get_contiguous_soft_mask(comp_data->wedge_index,
                                          comp_data->wedge_sign, sb_type);
    default: return comp_data->seg_mask;
  }
}

static inline void diffwtd_mask_d16(uint8_t *mask, int which_inverse,
                                    int mask_base, const CONV_BUF_TYPE *src0,
                                    int src0_stride, const CONV_BUF_TYPE *src1,
                                    int src1_stride, int h, int w,
                                    ConvolveParams *conv_params, int bd) {
  int round =
      2 * FILTER_BITS - conv_params->round_0 - conv_params->round_1 + (bd - 8);
  int i, j, m, diff;
  for (i = 0; i < h; ++i) {
    for (j = 0; j < w; ++j) {
      diff = abs(src0[i * src0_stride + j] - src1[i * src1_stride + j]);
      diff = ROUND_POWER_OF_TWO(diff, round);
      m = clamp(mask_base + (diff / DIFF_FACTOR), 0, AOM_BLEND_A64_MAX_ALPHA);
      mask[i * w + j] = which_inverse ? AOM_BLEND_A64_MAX_ALPHA - m : m;
    }
  }
}

void av1_build_compound_diffwtd_mask_d16_c(
    uint8_t *mask, DIFFWTD_MASK_TYPE mask_type, const CONV_BUF_TYPE *src0,
    int src0_stride, const CONV_BUF_TYPE *src1, int src1_stride, int h, int w,
    ConvolveParams *conv_params, int bd) {
  switch (mask_type) {
    case DIFFWTD_38:
      diffwtd_mask_d16(mask, 0, 38, src0, src0_stride, src1, src1_stride, h, w,
                       conv_params, bd);
      break;
    case DIFFWTD_38_INV:
      diffwtd_mask_d16(mask, 1, 38, src0, src0_stride, src1, src1_stride, h, w,
                       conv_params, bd);
      break;
    default: assert(0);
  }
}

static inline 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, int h,
                                int w) {
  int i, j, m, diff;
  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 * w + j] = which_inverse ? AOM_BLEND_A64_MAX_ALPHA - m : m;
    }
  }
}

void av1_build_compound_diffwtd_mask_c(uint8_t *mask,
                                       DIFFWTD_MASK_TYPE mask_type,
                                       const uint8_t *src0, int src0_stride,
                                       const uint8_t *src1, int src1_stride,
                                       int h, int w) {
  switch (mask_type) {
    case DIFFWTD_38:
      diffwtd_mask(mask, 0, 38, src0, src0_stride, src1, src1_stride, h, w);
      break;
    case DIFFWTD_38_INV:
      diffwtd_mask(mask, 1, 38, src0, src0_stride, src1, src1_stride, h, w);
      break;
    default: assert(0);
  }
}

#if CONFIG_AV1_HIGHBITDEPTH
static AOM_FORCE_INLINE 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, int h, int w,
    const unsigned int bd) {
  assert(bd >= 8);
  if (bd == 8) {
    if (which_inverse) {
      for (int i = 0; i < h; ++i) {
        for (int j = 0; j < w; ++j) {
          int diff = abs((int)src0[j] - (int)src1[j]) / DIFF_FACTOR;
          unsigned int m = negative_to_zero(mask_base + diff);
          m = AOMMIN(m, AOM_BLEND_A64_MAX_ALPHA);
          mask[j] = AOM_BLEND_A64_MAX_ALPHA - m;
        }
        src0 += src0_stride;
        src1 += src1_stride;
        mask += w;
      }
    } else {
      for (int i = 0; i < h; ++i) {
        for (int j = 0; j < w; ++j) {
          int diff = abs((int)src0[j] - (int)src1[j]) / DIFF_FACTOR;
          unsigned int m = negative_to_zero(mask_base + diff);
          m = AOMMIN(m, AOM_BLEND_A64_MAX_ALPHA);
          mask[j] = m;
        }
        src0 += src0_stride;
        src1 += src1_stride;
        mask += w;
      }
    }
  } else {
    const unsigned int bd_shift = bd - 8;
    if (which_inverse) {
      for (int i = 0; i < h; ++i) {
        for (int j = 0; j < w; ++j) {
          int diff =
              (abs((int)src0[j] - (int)src1[j]) >> bd_shift) / DIFF_FACTOR;
          unsigned int m = negative_to_zero(mask_base + diff);
          m = AOMMIN(m, AOM_BLEND_A64_MAX_ALPHA);
          mask[j] = AOM_BLEND_A64_MAX_ALPHA - m;
        }
        src0 += src0_stride;
        src1 += src1_stride;
        mask += w;
      }
    } else {
      for (int i = 0; i < h; ++i) {
        for (int j = 0; j < w; ++j) {
          int diff =
              (abs((int)src0[j] - (int)src1[j]) >> bd_shift) / DIFF_FACTOR;
          unsigned int m = negative_to_zero(mask_base + diff);
          m = AOMMIN(m, AOM_BLEND_A64_MAX_ALPHA);
          mask[j] = m;
        }
        src0 += src0_stride;
        src1 += src1_stride;
        mask += w;
      }
    }
  }
}

void av1_build_compound_diffwtd_mask_highbd_c(
    uint8_t *mask, DIFFWTD_MASK_TYPE mask_type, const uint8_t *src0,
    int src0_stride, const uint8_t *src1, int src1_stride, int h, int w,
    int bd) {
  switch (mask_type) {
    case DIFFWTD_38:
      diffwtd_mask_highbd(mask, 0, 38, CONVERT_TO_SHORTPTR(src0), src0_stride,
                          CONVERT_TO_SHORTPTR(src1), src1_stride, h, w, bd);
      break;
    case DIFFWTD_38_INV:
      diffwtd_mask_highbd(mask, 1, 38, CONVERT_TO_SHORTPTR(src0), src0_stride,
                          CONVERT_TO_SHORTPTR(src1), src1_stride, h, w, bd);
      break;
    default: assert(0);
  }
}
#endif  // CONFIG_AV1_HIGHBITDEPTH

/* clang-format off */
#if CONFIG_AV1_HIGHBITDEPTH
static const uint8_t ii_weights1d[MAX_SB_SIZE] = {
  60, 58, 56, 54, 52, 50, 48, 47, 45, 44, 42, 41, 39, 38, 37, 35, 34, 33, 32,
  31, 30, 29, 28, 27, 26, 25, 24, 23, 22, 22, 21, 20, 19, 19, 18, 18, 17, 16,
  16, 15, 15, 14, 14, 13, 13, 12, 12, 12, 11, 11, 10, 10, 10,  9,  9,  9,  8,
  8,  8,  8,  7,  7,  7,  7,  6,  6,  6,  6,  6,  5,  5,  5,  5,  5,  4,  4,
  4,  4,  4,  4,  4,  4,  3,  3,  3,  3,  3,  3,  3,  3,  3,  2,  2,  2,  2,
  2,  2,  2,  2,  2,  2,  2,  2,  2,  2,  2,  1,  1,  1,  1,  1,  1,  1,  1,
  1,  1,  1,  1,  1,  1,  1,  1,  1,  1,  1,  1,  1,  1
};
static const uint8_t ii_size_scales[BLOCK_SIZES_ALL] = {
    32, 16, 16, 16, 8, 8, 8, 4,
    4,  4,  2,  2,  2, 1, 1, 1,
    8,  8,  4,  4,  2, 2
};
#endif  // CONFIG_AV1_HIGHBITDEPTH
/* clang-format on */

#if CONFIG_AV1_HIGHBITDEPTH
// Used at runtime by combine_interintra_highbd() below for a per-call
// stack-allocated mask. The block-size-major precomputed
// `smooth_interintra_mask_buf` only covers block sizes with bw,bh <=
// MAX_WEDGE_SIZE; the highbd path may need larger sizes.
static inline void build_smooth_interintra_mask(uint8_t *mask, int stride,
                                                BLOCK_SIZE plane_bsize,
                                                INTERINTRA_MODE mode) {
  int i, j;
  const int bw = block_size_wide[plane_bsize];
  const int bh = block_size_high[plane_bsize];
  const int size_scale = ii_size_scales[plane_bsize];

  switch (mode) {
    case II_V_PRED:
      for (i = 0; i < bh; ++i) {
        memset(mask, ii_weights1d[i * size_scale], bw * sizeof(mask[0]));
        mask += stride;
      }
      break;

    case II_H_PRED:
      for (i = 0; i < bh; ++i) {
        for (j = 0; j < bw; ++j) mask[j] = ii_weights1d[j * size_scale];
        mask += stride;
      }
      break;

    case II_SMOOTH_PRED:
      for (i = 0; i < bh; ++i) {
        for (j = 0; j < bw; ++j)
          mask[j] = ii_weights1d[(i < j ? i : j) * size_scale];
        mask += stride;
      }
      break;

    case II_DC_PRED:
    default:
      for (i = 0; i < bh; ++i) {
        memset(mask, 32, bw * sizeof(mask[0]));
        mask += stride;
      }
      break;
  }
}
#endif  // CONFIG_AV1_HIGHBITDEPTH

// No-op now that the wedge / inter-intra mask buffers are precomputed at
// codegen time and stored in `.rdata`. The symbol is kept exported because
// decoder.c, encoder.c, and unit tests still call it.
void av1_init_wedge_masks(void) {}

static inline void build_masked_compound_no_round(
    uint8_t *dst, int dst_stride, const CONV_BUF_TYPE *src0, int src0_stride,
    const CONV_BUF_TYPE *src1, int src1_stride,
    const INTERINTER_COMPOUND_DATA *const comp_data, BLOCK_SIZE sb_type, int h,
    int w, InterPredParams *inter_pred_params) {
  const int ssy = inter_pred_params->subsampling_y;
  const int ssx = inter_pred_params->subsampling_x;
  const uint8_t *mask = av1_get_compound_type_mask(comp_data, sb_type);
  const int mask_stride = block_size_wide[sb_type];
#if CONFIG_AV1_HIGHBITDEPTH
  if (inter_pred_params->use_hbd_buf) {
    aom_highbd_blend_a64_d16_mask(dst, dst_stride, src0, src0_stride, src1,
                                  src1_stride, mask, mask_stride, w, h, ssx,
                                  ssy, &inter_pred_params->conv_params,
                                  inter_pred_params->bit_depth);
  } else {
    aom_lowbd_blend_a64_d16_mask(dst, dst_stride, src0, src0_stride, src1,
                                 src1_stride, mask, mask_stride, w, h, ssx, ssy,
                                 &inter_pred_params->conv_params);
  }
#else
  aom_lowbd_blend_a64_d16_mask(dst, dst_stride, src0, src0_stride, src1,
                               src1_stride, mask, mask_stride, w, h, ssx, ssy,
                               &inter_pred_params->conv_params);
#endif
}

void av1_make_masked_inter_predictor(const uint8_t *pre, int pre_stride,
                                     uint8_t *dst, int dst_stride,
                                     InterPredParams *inter_pred_params,
                                     const SubpelParams *subpel_params) {
  const INTERINTER_COMPOUND_DATA *comp_data = &inter_pred_params->mask_comp;
  BLOCK_SIZE sb_type = inter_pred_params->sb_type;

  // We're going to call av1_make_inter_predictor to generate a prediction into
  // a temporary buffer, then will blend that temporary buffer with that from
  // the other reference.
  DECLARE_ALIGNED(32, uint8_t, tmp_buf[2 * MAX_SB_SQUARE]);
  uint8_t *tmp_dst =
      inter_pred_params->use_hbd_buf ? CONVERT_TO_BYTEPTR(tmp_buf) : tmp_buf;

  const int tmp_buf_stride = MAX_SB_SIZE;
  CONV_BUF_TYPE *org_dst = inter_pred_params->conv_params.dst;
  int org_dst_stride = inter_pred_params->conv_params.dst_stride;
  CONV_BUF_TYPE *tmp_buf16 = (CONV_BUF_TYPE *)tmp_buf;
  inter_pred_params->conv_params.dst = tmp_buf16;
  inter_pred_params->conv_params.dst_stride = tmp_buf_stride;
  assert(inter_pred_params->conv_params.do_average == 0);

  // This will generate a prediction in tmp_buf for the second reference
  av1_make_inter_predictor(pre, pre_stride, tmp_dst, MAX_SB_SIZE,
                           inter_pred_params, subpel_params);

  if (!inter_pred_params->conv_params.plane &&
      comp_data->type == COMPOUND_DIFFWTD) {
    av1_build_compound_diffwtd_mask_d16(
        comp_data->seg_mask, comp_data->mask_type, org_dst, org_dst_stride,
        tmp_buf16, tmp_buf_stride, inter_pred_params->block_height,
        inter_pred_params->block_width, &inter_pred_params->conv_params,
        inter_pred_params->bit_depth);
  }
  build_masked_compound_no_round(
      dst, dst_stride, org_dst, org_dst_stride, tmp_buf16, tmp_buf_stride,
      comp_data, sb_type, inter_pred_params->block_height,
      inter_pred_params->block_width, inter_pred_params);
}

void av1_dist_wtd_comp_weight_assign(const AV1_COMMON *cm,
                                     const MB_MODE_INFO *mbmi, int *fwd_offset,
                                     int *bck_offset,
                                     int *use_dist_wtd_comp_avg,
                                     int is_compound) {
  assert(fwd_offset != NULL && bck_offset != NULL);
  if (!is_compound || mbmi->compound_idx) {
    *fwd_offset = 8;
    *bck_offset = 8;
    *use_dist_wtd_comp_avg = 0;
    return;
  }

  *use_dist_wtd_comp_avg = 1;
  const RefCntBuffer *const bck_buf = get_ref_frame_buf(cm, mbmi->ref_frame[0]);
  const RefCntBuffer *const fwd_buf = get_ref_frame_buf(cm, mbmi->ref_frame[1]);
  const int cur_frame_index = cm->cur_frame->order_hint;
  int bck_frame_index = 0, fwd_frame_index = 0;

  if (bck_buf != NULL) bck_frame_index = bck_buf->order_hint;
  if (fwd_buf != NULL) fwd_frame_index = fwd_buf->order_hint;

  int d0 = clamp(abs(get_relative_dist(&cm->seq_params->order_hint_info,
                                       fwd_frame_index, cur_frame_index)),
                 0, MAX_FRAME_DISTANCE);
  int d1 = clamp(abs(get_relative_dist(&cm->seq_params->order_hint_info,
                                       cur_frame_index, bck_frame_index)),
                 0, MAX_FRAME_DISTANCE);

  const int order = d0 <= d1;

  if (d0 == 0 || d1 == 0) {
    *fwd_offset = quant_dist_lookup_table[3][order];
    *bck_offset = quant_dist_lookup_table[3][1 - order];
    return;
  }

  int i;
  for (i = 0; i < 3; ++i) {
    int c0 = quant_dist_weight[i][order];
    int c1 = quant_dist_weight[i][!order];
    int d0_c0 = d0 * c0;
    int d1_c1 = d1 * c1;
    if ((d0 > d1 && d0_c0 < d1_c1) || (d0 <= d1 && d0_c0 > d1_c1)) break;
  }

  *fwd_offset = quant_dist_lookup_table[i][order];
  *bck_offset = quant_dist_lookup_table[i][1 - order];
}

void av1_setup_dst_planes(struct macroblockd_plane *planes, BLOCK_SIZE bsize,
                          const YV12_BUFFER_CONFIG *src, int mi_row, int mi_col,
                          const int plane_start, const int plane_end) {
  // We use AOMMIN(num_planes, MAX_MB_PLANE) instead of num_planes to quiet
  // the static analysis warnings.
  for (int i = plane_start; i < AOMMIN(plane_end, MAX_MB_PLANE); ++i) {
    struct macroblockd_plane *const pd = &planes[i];
    const int is_uv = i > 0;
    setup_pred_plane(&pd->dst, bsize, src->buffers[i], src->crop_widths[is_uv],
                     src->crop_heights[is_uv], src->strides[is_uv], mi_row,
                     mi_col, NULL, pd->subsampling_x, pd->subsampling_y);
  }
}

void av1_setup_pre_planes(MACROBLOCKD *xd, int idx,
                          const YV12_BUFFER_CONFIG *src, int mi_row, int mi_col,
                          const struct scale_factors *sf,
                          const int num_planes) {
  if (src != NULL) {
    // We use AOMMIN(num_planes, MAX_MB_PLANE) instead of num_planes to quiet
    // the static analysis warnings.
    for (int i = 0; i < AOMMIN(num_planes, MAX_MB_PLANE); ++i) {
      struct macroblockd_plane *const pd = &xd->plane[i];
      const int is_uv = i > 0;
      setup_pred_plane(&pd->pre[idx], xd->mi[0]->bsize, src->buffers[i],
                       src->crop_widths[is_uv], src->crop_heights[is_uv],
                       src->strides[is_uv], mi_row, mi_col, sf,
                       pd->subsampling_x, pd->subsampling_y);
    }
  }
}

// obmc_mask_N[overlap_position]
static const uint8_t obmc_mask_1[1] = { 64 };
DECLARE_ALIGNED(2, static const uint8_t, obmc_mask_2[2]) = { 45, 64 };

DECLARE_ALIGNED(4, static const uint8_t, obmc_mask_4[4]) = { 39, 50, 59, 64 };

static const uint8_t obmc_mask_8[8] = { 36, 42, 48, 53, 57, 61, 64, 64 };

static const uint8_t obmc_mask_16[16] = { 34, 37, 40, 43, 46, 49, 52, 54,
                                          56, 58, 60, 61, 64, 64, 64, 64 };

static const uint8_t obmc_mask_32[32] = { 33, 35, 36, 38, 40, 41, 43, 44,
                                          45, 47, 48, 50, 51, 52, 53, 55,
                                          56, 57, 58, 59, 60, 60, 61, 62,
                                          64, 64, 64, 64, 64, 64, 64, 64 };

static const uint8_t obmc_mask_64[64] = {
  33, 34, 35, 35, 36, 37, 38, 39, 40, 40, 41, 42, 43, 44, 44, 44,
  45, 46, 47, 47, 48, 49, 50, 51, 51, 51, 52, 52, 53, 54, 55, 56,
  56, 56, 57, 57, 58, 58, 59, 60, 60, 60, 60, 60, 61, 62, 62, 62,
  62, 62, 63, 63, 63, 63, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64,
};

const uint8_t *av1_get_obmc_mask(int length) {
  switch (length) {
    case 1: return obmc_mask_1;
    case 2: return obmc_mask_2;
    case 4: return obmc_mask_4;
    case 8: return obmc_mask_8;
    case 16: return obmc_mask_16;
    case 32: return obmc_mask_32;
    case 64: return obmc_mask_64;
    default: assert(0); return NULL;
  }
}

static inline void increment_int_ptr(MACROBLOCKD *xd, int rel_mi_row,
                                     int rel_mi_col, uint8_t op_mi_size,
                                     int dir, MB_MODE_INFO *mi, void *fun_ctxt,
                                     const int num_planes) {
  (void)xd;
  (void)rel_mi_row;
  (void)rel_mi_col;
  (void)op_mi_size;
  (void)dir;
  (void)mi;
  ++*(uint8_t *)fun_ctxt;
  (void)num_planes;
}

void av1_count_overlappable_neighbors(const AV1_COMMON *cm, MACROBLOCKD *xd) {
  MB_MODE_INFO *mbmi = xd->mi[0];

  mbmi->overlappable_neighbors = 0;

  if (!is_motion_variation_allowed_bsize(mbmi->bsize)) return;

  foreach_overlappable_nb_above(cm, xd, INT_MAX, increment_int_ptr,
                                &mbmi->overlappable_neighbors);
  if (mbmi->overlappable_neighbors) return;
  foreach_overlappable_nb_left(cm, xd, INT_MAX, increment_int_ptr,
                               &mbmi->overlappable_neighbors);
}

// HW does not support < 4x4 prediction. To limit the bandwidth requirement, if
// block-size of current plane is smaller than 8x8, always only blend with the
// left neighbor(s) (skip blending with the above side).
#define DISABLE_CHROMA_U8X8_OBMC 0  // 0: one-sided obmc; 1: disable

int av1_skip_u4x4_pred_in_obmc(BLOCK_SIZE bsize,
                               const struct macroblockd_plane *pd, int dir) {
  assert(is_motion_variation_allowed_bsize(bsize));

  const BLOCK_SIZE bsize_plane =
      get_plane_block_size(bsize, pd->subsampling_x, pd->subsampling_y);
  switch (bsize_plane) {
#if DISABLE_CHROMA_U8X8_OBMC
    case BLOCK_4X4:
    case BLOCK_8X4:
    case BLOCK_4X8: return 1;
#else
    case BLOCK_4X4:
    case BLOCK_8X4:
    case BLOCK_4X8: return dir == 0;
#endif
    default: return 0;
  }
}

#if CONFIG_AV1_DECODER
static void modify_neighbor_predictor_for_obmc(MB_MODE_INFO *mbmi) {
  mbmi->ref_frame[1] = NONE_FRAME;
  mbmi->interinter_comp.type = COMPOUND_AVERAGE;
}
#endif  // CONFIG_AV1_DECODER

struct obmc_inter_pred_ctxt {
  uint8_t **adjacent;
  int *adjacent_stride;
};

static inline void build_obmc_inter_pred_above(
    MACROBLOCKD *xd, int rel_mi_row, int rel_mi_col, uint8_t op_mi_size,
    int dir, MB_MODE_INFO *above_mi, void *fun_ctxt, const int num_planes) {
  (void)above_mi;
  (void)rel_mi_row;
  (void)dir;
  struct obmc_inter_pred_ctxt *ctxt = (struct obmc_inter_pred_ctxt *)fun_ctxt;
  const BLOCK_SIZE bsize = xd->mi[0]->bsize;
  const int overlap =
      AOMMIN(block_size_high[bsize], block_size_high[BLOCK_64X64]) >> 1;

  for (int plane = 0; plane < num_planes; ++plane) {
    const struct macroblockd_plane *pd = &xd->plane[plane];
    const int bw = (op_mi_size * MI_SIZE) >> pd->subsampling_x;
    const int bh = overlap >> pd->subsampling_y;
    const int plane_col = (rel_mi_col * MI_SIZE) >> pd->subsampling_x;

    if (av1_skip_u4x4_pred_in_obmc(bsize, pd, 0)) continue;

    const int dst_stride = pd->dst.stride;
    uint8_t *const dst = &pd->dst.buf[plane_col];
    const int tmp_stride = ctxt->adjacent_stride[plane];
    const uint8_t *const tmp = &ctxt->adjacent[plane][plane_col];
    const uint8_t *const mask = av1_get_obmc_mask(bh);
#if CONFIG_AV1_HIGHBITDEPTH
    const int is_hbd = is_cur_buf_hbd(xd);
    if (is_hbd)
      aom_highbd_blend_a64_vmask(dst, dst_stride, dst, dst_stride, tmp,
                                 tmp_stride, mask, bw, bh, xd->bd);
    else
      aom_blend_a64_vmask(dst, dst_stride, dst, dst_stride, tmp, tmp_stride,
                          mask, bw, bh);
#else
    aom_blend_a64_vmask(dst, dst_stride, dst, dst_stride, tmp, tmp_stride, mask,
                        bw, bh);
#endif
  }
}

static inline void build_obmc_inter_pred_left(
    MACROBLOCKD *xd, int rel_mi_row, int rel_mi_col, uint8_t op_mi_size,
    int dir, MB_MODE_INFO *left_mi, void *fun_ctxt, const int num_planes) {
  (void)left_mi;
  (void)rel_mi_col;
  (void)dir;
  struct obmc_inter_pred_ctxt *ctxt = (struct obmc_inter_pred_ctxt *)fun_ctxt;
  const BLOCK_SIZE bsize = xd->mi[0]->bsize;
  const int overlap =
      AOMMIN(block_size_wide[bsize], block_size_wide[BLOCK_64X64]) >> 1;

  for (int plane = 0; plane < num_planes; ++plane) {
    const struct macroblockd_plane *pd = &xd->plane[plane];
    const int bw = overlap >> pd->subsampling_x;
    const int bh = (op_mi_size * MI_SIZE) >> pd->subsampling_y;
    const int plane_row = (rel_mi_row * MI_SIZE) >> pd->subsampling_y;

    if (av1_skip_u4x4_pred_in_obmc(bsize, pd, 1)) continue;

    const int dst_stride = pd->dst.stride;
    uint8_t *const dst = &pd->dst.buf[plane_row * dst_stride];
    const int tmp_stride = ctxt->adjacent_stride[plane];
    const uint8_t *const tmp = &ctxt->adjacent[plane][plane_row * tmp_stride];
    const uint8_t *const mask = av1_get_obmc_mask(bw);

#if CONFIG_AV1_HIGHBITDEPTH
    const int is_hbd = is_cur_buf_hbd(xd);
    if (is_hbd)
      aom_highbd_blend_a64_hmask(dst, dst_stride, dst, dst_stride, tmp,
                                 tmp_stride, mask, bw, bh, xd->bd);
    else
      aom_blend_a64_hmask(dst, dst_stride, dst, dst_stride, tmp, tmp_stride,
                          mask, bw, bh);
#else
    aom_blend_a64_hmask(dst, dst_stride, dst, dst_stride, tmp, tmp_stride, mask,
                        bw, bh);
#endif
  }
}

// This function combines motion compensated predictions that are generated by
// top/left neighboring blocks' inter predictors with the regular inter
// prediction. We assume the original prediction (bmc) is stored in
// xd->plane[].dst.buf
void av1_build_obmc_inter_prediction(const AV1_COMMON *cm, MACROBLOCKD *xd,
                                     uint8_t *above[MAX_MB_PLANE],
                                     int above_stride[MAX_MB_PLANE],
                                     uint8_t *left[MAX_MB_PLANE],
                                     int left_stride[MAX_MB_PLANE]) {
  const BLOCK_SIZE bsize = xd->mi[0]->bsize;

  // handle above row
  struct obmc_inter_pred_ctxt ctxt_above = { above, above_stride };
  foreach_overlappable_nb_above(cm, xd,
                                max_neighbor_obmc[mi_size_wide_log2[bsize]],
                                build_obmc_inter_pred_above, &ctxt_above);

  // handle left column
  struct obmc_inter_pred_ctxt ctxt_left = { left, left_stride };
  foreach_overlappable_nb_left(cm, xd,
                               max_neighbor_obmc[mi_size_high_log2[bsize]],
                               build_obmc_inter_pred_left, &ctxt_left);
}

void av1_setup_obmc_dst_bufs(MACROBLOCKD *xd, uint8_t **dst_buf1,
                             uint8_t **dst_buf2) {
  if (is_cur_buf_hbd(xd)) {
    int len = sizeof(uint16_t);
    dst_buf1[0] = CONVERT_TO_BYTEPTR(xd->tmp_obmc_bufs[0]);
    dst_buf1[1] =
        CONVERT_TO_BYTEPTR(xd->tmp_obmc_bufs[0] + MAX_SB_SQUARE * len);
    dst_buf1[2] =
        CONVERT_TO_BYTEPTR(xd->tmp_obmc_bufs[0] + MAX_SB_SQUARE * 2 * len);
    dst_buf2[0] = CONVERT_TO_BYTEPTR(xd->tmp_obmc_bufs[1]);
    dst_buf2[1] =
        CONVERT_TO_BYTEPTR(xd->tmp_obmc_bufs[1] + MAX_SB_SQUARE * len);
    dst_buf2[2] =
        CONVERT_TO_BYTEPTR(xd->tmp_obmc_bufs[1] + MAX_SB_SQUARE * 2 * len);
  } else {
    dst_buf1[0] = xd->tmp_obmc_bufs[0];
    dst_buf1[1] = xd->tmp_obmc_bufs[0] + MAX_SB_SQUARE;
    dst_buf1[2] = xd->tmp_obmc_bufs[0] + MAX_SB_SQUARE * 2;
    dst_buf2[0] = xd->tmp_obmc_bufs[1];
    dst_buf2[1] = xd->tmp_obmc_bufs[1] + MAX_SB_SQUARE;
    dst_buf2[2] = xd->tmp_obmc_bufs[1] + MAX_SB_SQUARE * 2;
  }
}

#if CONFIG_AV1_DECODER
void av1_setup_build_prediction_by_above_pred(
    MACROBLOCKD *xd, int rel_mi_col, uint8_t above_mi_width,
    MB_MODE_INFO *above_mbmi, struct build_prediction_ctxt *ctxt,
    const int num_planes) {
  const BLOCK_SIZE a_bsize = AOMMAX(BLOCK_8X8, above_mbmi->bsize);
  const int above_mi_col = xd->mi_col + rel_mi_col;

  modify_neighbor_predictor_for_obmc(above_mbmi);

  for (int j = 0; j < num_planes; ++j) {
    struct macroblockd_plane *const pd = &xd->plane[j];
    setup_pred_plane(&pd->dst, a_bsize, ctxt->tmp_buf[j], ctxt->tmp_width[j],
                     ctxt->tmp_height[j], ctxt->tmp_stride[j], 0, rel_mi_col,
                     NULL, pd->subsampling_x, pd->subsampling_y);
  }

  const int num_refs = 1 + has_second_ref(above_mbmi);

  for (int ref = 0; ref < num_refs; ++ref) {
    const MV_REFERENCE_FRAME frame = above_mbmi->ref_frame[ref];

    const RefCntBuffer *const ref_buf = get_ref_frame_buf(ctxt->cm, frame);
    const struct scale_factors *const sf =
        get_ref_scale_factors_const(ctxt->cm, frame);
    xd->block_ref_scale_factors[ref] = sf;
    if ((!av1_is_valid_scale(sf)))
      aom_internal_error(xd->error_info, AOM_CODEC_UNSUP_BITSTREAM,
                         "Reference frame has invalid dimensions");
    av1_setup_pre_planes(xd, ref, &ref_buf->buf, xd->mi_row, above_mi_col, sf,
                         num_planes);
  }

  xd->mb_to_left_edge = 8 * MI_SIZE * (-above_mi_col);
  xd->mb_to_right_edge =
      ctxt->mb_to_far_edge +
      (xd->width - rel_mi_col - above_mi_width) * MI_SIZE * 8;
}

void av1_setup_build_prediction_by_left_pred(MACROBLOCKD *xd, int rel_mi_row,
                                             uint8_t left_mi_height,
                                             MB_MODE_INFO *left_mbmi,
                                             struct build_prediction_ctxt *ctxt,
                                             const int num_planes) {
  const BLOCK_SIZE l_bsize = AOMMAX(BLOCK_8X8, left_mbmi->bsize);
  const int left_mi_row = xd->mi_row + rel_mi_row;

  modify_neighbor_predictor_for_obmc(left_mbmi);

  for (int j = 0; j < num_planes; ++j) {
    struct macroblockd_plane *const pd = &xd->plane[j];
    setup_pred_plane(&pd->dst, l_bsize, ctxt->tmp_buf[j], ctxt->tmp_width[j],
                     ctxt->tmp_height[j], ctxt->tmp_stride[j], rel_mi_row, 0,
                     NULL, pd->subsampling_x, pd->subsampling_y);
  }

  const int num_refs = 1 + has_second_ref(left_mbmi);

  for (int ref = 0; ref < num_refs; ++ref) {
    const MV_REFERENCE_FRAME frame = left_mbmi->ref_frame[ref];

    const RefCntBuffer *const ref_buf = get_ref_frame_buf(ctxt->cm, frame);
    const struct scale_factors *const ref_scale_factors =
        get_ref_scale_factors_const(ctxt->cm, frame);

    xd->block_ref_scale_factors[ref] = ref_scale_factors;
    if ((!av1_is_valid_scale(ref_scale_factors)))
      aom_internal_error(xd->error_info, AOM_CODEC_UNSUP_BITSTREAM,
                         "Reference frame has invalid dimensions");
    av1_setup_pre_planes(xd, ref, &ref_buf->buf, left_mi_row, xd->mi_col,
                         ref_scale_factors, num_planes);
  }

  xd->mb_to_top_edge = GET_MV_SUBPEL(MI_SIZE * (-left_mi_row));
  xd->mb_to_bottom_edge =
      ctxt->mb_to_far_edge +
      GET_MV_SUBPEL((xd->height - rel_mi_row - left_mi_height) * MI_SIZE);
}
#endif  // CONFIG_AV1_DECODER

static inline void combine_interintra(
    INTERINTRA_MODE mode, int8_t use_wedge_interintra, int8_t wedge_index,
    int8_t wedge_sign, BLOCK_SIZE bsize, BLOCK_SIZE plane_bsize,
    uint8_t *comppred, int compstride, const uint8_t *interpred,
    int interstride, const uint8_t *intrapred, int intrastride) {
  const int bw = block_size_wide[plane_bsize];
  const int bh = block_size_high[plane_bsize];

  if (use_wedge_interintra) {
    if (av1_is_wedge_used(bsize)) {
      const uint8_t *mask =
          av1_get_contiguous_soft_mask(wedge_index, wedge_sign, bsize);
      const int subw = 2 * mi_size_wide[bsize] == bw;
      const int subh = 2 * mi_size_high[bsize] == bh;
      aom_blend_a64_mask(comppred, compstride, intrapred, intrastride,
                         interpred, interstride, mask, block_size_wide[bsize],
                         bw, bh, subw, subh);
    }
    return;
  }

  const uint8_t *mask = smooth_interintra_mask_buf[mode][plane_bsize];
  aom_blend_a64_mask(comppred, compstride, intrapred, intrastride, interpred,
                     interstride, mask, bw, bw, bh, 0, 0);
}

#if CONFIG_AV1_HIGHBITDEPTH
static inline void combine_interintra_highbd(
    INTERINTRA_MODE mode, int8_t use_wedge_interintra, int8_t wedge_index,
    int8_t wedge_sign, BLOCK_SIZE bsize, BLOCK_SIZE plane_bsize,
    uint8_t *comppred8, int compstride, const uint8_t *interpred8,
    int interstride, const uint8_t *intrapred8, int intrastride, int bd) {
  const int bw = block_size_wide[plane_bsize];
  const int bh = block_size_high[plane_bsize];

  if (use_wedge_interintra) {
    if (av1_is_wedge_used(bsize)) {
      const uint8_t *mask =
          av1_get_contiguous_soft_mask(wedge_index, wedge_sign, bsize);
      const int subh = 2 * mi_size_high[bsize] == bh;
      const int subw = 2 * mi_size_wide[bsize] == bw;
      aom_highbd_blend_a64_mask(comppred8, compstride, intrapred8, intrastride,
                                interpred8, interstride, mask,
                                block_size_wide[bsize], bw, bh, subw, subh, bd);
    }
    return;
  }

  uint8_t mask[MAX_SB_SQUARE];
  build_smooth_interintra_mask(mask, bw, plane_bsize, mode);
  aom_highbd_blend_a64_mask(comppred8, compstride, intrapred8, intrastride,
                            interpred8, interstride, mask, bw, bw, bh, 0, 0,
                            bd);
}
#endif

void av1_build_intra_predictors_for_interintra(const AV1_COMMON *cm,
                                               MACROBLOCKD *xd,
                                               BLOCK_SIZE bsize, int plane,
                                               const BUFFER_SET *ctx,
                                               uint8_t *dst, int dst_stride) {
  struct macroblockd_plane *const pd = &xd->plane[plane];
  const int ssx = xd->plane[plane].subsampling_x;
  const int ssy = xd->plane[plane].subsampling_y;
  BLOCK_SIZE plane_bsize = get_plane_block_size(bsize, ssx, ssy);
  PREDICTION_MODE mode = interintra_to_intra_mode[xd->mi[0]->interintra_mode];
  assert(xd->mi[0]->angle_delta[PLANE_TYPE_Y] == 0);
  assert(xd->mi[0]->angle_delta[PLANE_TYPE_UV] == 0);
  assert(xd->mi[0]->filter_intra_mode_info.use_filter_intra == 0);
  assert(xd->mi[0]->use_intrabc == 0);
  const SequenceHeader *seq_params = cm->seq_params;

  av1_predict_intra_block(xd, seq_params->sb_size,
                          seq_params->enable_intra_edge_filter, pd->width,
                          pd->height, max_txsize_rect_lookup[plane_bsize], mode,
                          0, 0, FILTER_INTRA_MODES, ctx->plane[plane],
                          ctx->stride[plane], dst, dst_stride, 0, 0, plane);
}

void av1_combine_interintra(MACROBLOCKD *xd, BLOCK_SIZE bsize, int plane,
                            const uint8_t *inter_pred, int inter_stride,
                            const uint8_t *intra_pred, int intra_stride) {
  const int ssx = xd->plane[plane].subsampling_x;
  const int ssy = xd->plane[plane].subsampling_y;
  const BLOCK_SIZE plane_bsize = get_plane_block_size(bsize, ssx, ssy);
#if CONFIG_AV1_HIGHBITDEPTH
  if (is_cur_buf_hbd(xd)) {
    combine_interintra_highbd(
        xd->mi[0]->interintra_mode, xd->mi[0]->use_wedge_interintra,
        xd->mi[0]->interintra_wedge_index, INTERINTRA_WEDGE_SIGN, bsize,
        plane_bsize, xd->plane[plane].dst.buf, xd->plane[plane].dst.stride,
        inter_pred, inter_stride, intra_pred, intra_stride, xd->bd);
    return;
  }
#endif
  combine_interintra(
      xd->mi[0]->interintra_mode, xd->mi[0]->use_wedge_interintra,
      xd->mi[0]->interintra_wedge_index, INTERINTRA_WEDGE_SIGN, bsize,
      plane_bsize, xd->plane[plane].dst.buf, xd->plane[plane].dst.stride,
      inter_pred, inter_stride, intra_pred, intra_stride);
}

// build interintra_predictors for one plane
void av1_build_interintra_predictor(const AV1_COMMON *cm, MACROBLOCKD *xd,
                                    uint8_t *pred, int stride,
                                    const BUFFER_SET *ctx, int plane,
                                    BLOCK_SIZE bsize) {
  assert(bsize < BLOCK_SIZES_ALL);
  if (is_cur_buf_hbd(xd)) {
    DECLARE_ALIGNED(16, uint16_t, intrapredictor[MAX_SB_SQUARE]);
    av1_build_intra_predictors_for_interintra(
        cm, xd, bsize, plane, ctx, CONVERT_TO_BYTEPTR(intrapredictor),
        MAX_SB_SIZE);
    av1_combine_interintra(xd, bsize, plane, pred, stride,
                           CONVERT_TO_BYTEPTR(intrapredictor), MAX_SB_SIZE);
  } else {
    DECLARE_ALIGNED(16, uint8_t, intrapredictor[MAX_SB_SQUARE]);
    av1_build_intra_predictors_for_interintra(cm, xd, bsize, plane, ctx,
                                              intrapredictor, MAX_SB_SIZE);
    av1_combine_interintra(xd, bsize, plane, pred, stride, intrapredictor,
                           MAX_SB_SIZE);
  }
}