/*

 * Copyright (c) 2020, 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 "av1/common/av1_common_int.h"

#include "av1/common/cfl.h"

#include "av1/common/reconintra.h"

#include "av1/encoder/intra_mode_search.h"

#include "av1/encoder/intra_mode_search_utils.h"

#include "av1/encoder/palette.h"

#include "av1/encoder/speed_features.h"

#include "av1/encoder/tx_search.h"

// Even though there are 7 delta angles, this macro is set to 9 to facilitate

// the rd threshold check to prune -3 and 3 delta angles.

#define SIZE_OF_ANGLE_DELTA_RD_COST_ARRAY (2 * MAX_ANGLE_DELTA + 3)

// The order for evaluating delta angles while processing the luma directional

// intra modes. Currently, this order of evaluation is applicable only when

// speed feature prune_luma_odd_delta_angles_in_intra is enabled. In this case,

// even angles are evaluated first in order to facilitate the pruning of odd

// delta angles based on the rd costs of the neighboring delta angles.

static const int8_t luma_delta_angles_order[2 * MAX_ANGLE_DELTA] = {

  -2, 2, -3, -1, 1, 3,

};

/*!\cond */

static const PREDICTION_MODE intra_rd_search_mode_order[INTRA_MODES] = {

  DC_PRED,       H_PRED,        V_PRED,    SMOOTH_PRED, PAETH_PRED,

  SMOOTH_V_PRED, SMOOTH_H_PRED, D135_PRED, D203_PRED,   D157_PRED,

  D67_PRED,      D113_PRED,     D45_PRED,

};

static const UV_PREDICTION_MODE uv_rd_search_mode_order[UV_INTRA_MODES] = {

  UV_DC_PRED,     UV_CFL_PRED,   UV_H_PRED,        UV_V_PRED,

  UV_SMOOTH_PRED, UV_PAETH_PRED, UV_SMOOTH_V_PRED, UV_SMOOTH_H_PRED,

  UV_D135_PRED,   UV_D203_PRED,  UV_D157_PRED,     UV_D67_PRED,

  UV_D113_PRED,   UV_D45_PRED,

};

// The bitmask corresponds to the filter intra modes as defined in enums.h

// FILTER_INTRA_MODE enumeration type. Setting a bit to 0 in the mask means to

// disable the evaluation of corresponding filter intra mode. The table

// av1_derived_filter_intra_mode_used_flag is used when speed feature

// prune_filter_intra_level is 1. The evaluated filter intra modes are union

// of the following:

// 1) FILTER_DC_PRED

// 2) mode that corresponds to best mode so far of DC_PRED, V_PRED, H_PRED,

// D157_PRED and PAETH_PRED. (Eg: FILTER_V_PRED if best mode so far is V_PRED).

static const uint8_t av1_derived_filter_intra_mode_used_flag[INTRA_MODES] = {

  0x01,  // DC_PRED:           0000 0001

  0x03,  // V_PRED:            0000 0011

  0x05,  // H_PRED:            0000 0101

  0x01,  // D45_PRED:          0000 0001

  0x01,  // D135_PRED:         0000 0001

  0x01,  // D113_PRED:         0000 0001

  0x09,  // D157_PRED:         0000 1001

  0x01,  // D203_PRED:         0000 0001

  0x01,  // D67_PRED:          0000 0001

  0x01,  // SMOOTH_PRED:       0000 0001

  0x01,  // SMOOTH_V_PRED:     0000 0001

  0x01,  // SMOOTH_H_PRED:     0000 0001

  0x11   // PAETH_PRED:        0001 0001

};

// The bitmask corresponds to the chroma intra modes as defined in enums.h

// UV_PREDICTION_MODE enumeration type. Setting a bit to 0 in the mask means to

// disable the evaluation of corresponding chroma intra mode. The table

// av1_derived_chroma_intra_mode_used_flag is used when speed feature

// prune_chroma_modes_using_luma_winner is enabled. The evaluated chroma

// intra modes are union of the following:

// 1) UV_DC_PRED

// 2) UV_SMOOTH_PRED

// 3) UV_CFL_PRED

// 4) mode that corresponds to luma intra mode winner (Eg : UV_V_PRED if luma

// intra mode winner is V_PRED).

static const uint16_t av1_derived_chroma_intra_mode_used_flag[INTRA_MODES] = {

  0x2201,  // DC_PRED:           0010 0010 0000 0001

  0x2203,  // V_PRED:            0010 0010 0000 0011

  0x2205,  // H_PRED:            0010 0010 0000 0101

  0x2209,  // D45_PRED:          0010 0010 0000 1001

  0x2211,  // D135_PRED:         0010 0010 0001 0001

  0x2221,  // D113_PRED:         0010 0010 0010 0001

  0x2241,  // D157_PRED:         0010 0010 0100 0001

  0x2281,  // D203_PRED:         0010 0010 1000 0001

  0x2301,  // D67_PRED:          0010 0011 0000 0001

  0x2201,  // SMOOTH_PRED:       0010 0010 0000 0001

  0x2601,  // SMOOTH_V_PRED:     0010 0110 0000 0001

  0x2a01,  // SMOOTH_H_PRED:     0010 1010 0000 0001

  0x3201   // PAETH_PRED:        0011 0010 0000 0001

};

DECLARE_ALIGNED(16, static const uint8_t, all_zeros[MAX_SB_SIZE]) = { 0 };

DECLARE_ALIGNED(16, static const uint16_t,

                highbd_all_zeros[MAX_SB_SIZE]) = { 0 };

int av1_calc_normalized_variance(aom_variance_fn_t vf, const uint8_t *const buf,

                                 const int stride, const int is_hbd) {

  unsigned int sse;

  if (is_hbd)

    return vf(buf, stride, CONVERT_TO_BYTEPTR(highbd_all_zeros), 0, &sse);

  else

    return vf(buf, stride, all_zeros, 0, &sse);

}

// Computes average of log(1 + variance) across 4x4 sub-blocks for source and

// reconstructed blocks.

static void compute_avg_log_variance(const AV1_COMP *const cpi, MACROBLOCK *x,

                                     const BLOCK_SIZE bs,

                                     double *avg_log_src_variance,

                                     double *avg_log_recon_variance) {

  const MACROBLOCKD *const xd = &x->e_mbd;

  const BLOCK_SIZE sb_size = cpi->common.seq_params->sb_size;

  const int mi_row_in_sb = x->e_mbd.mi_row & (mi_size_high[sb_size] - 1);

  const int mi_col_in_sb = x->e_mbd.mi_col & (mi_size_wide[sb_size] - 1);

  const int right_overflow =

      (xd->mb_to_right_edge < 0) ? ((-xd->mb_to_right_edge) >> 3) : 0;

  const int bottom_overflow =

      (xd->mb_to_bottom_edge < 0) ? ((-xd->mb_to_bottom_edge) >> 3) : 0;

  const int bw = (MI_SIZE * mi_size_wide[bs] - right_overflow);

  const int bh = (MI_SIZE * mi_size_high[bs] - bottom_overflow);

  const int is_hbd = is_cur_buf_hbd(xd);

  aom_variance_fn_t vf = cpi->ppi->fn_ptr[BLOCK_4X4].vf;

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

    const int r = mi_row_in_sb + (i >> MI_SIZE_LOG2);

    for (int j = 0; j < bw; j += MI_SIZE) {

      const int c = mi_col_in_sb + (j >> MI_SIZE_LOG2);

      const int mi_offset = r * mi_size_wide[sb_size] + c;

      Block4x4VarInfo *block_4x4_var_info =

          &x->src_var_info_of_4x4_sub_blocks[mi_offset];

      int src_var = block_4x4_var_info->var;

      double log_src_var = block_4x4_var_info->log_var;

      // Compute average of log(1 + variance) for the source block from 4x4

      // sub-block variance values. Calculate and store 4x4 sub-block variance

      // and log(1 + variance), if the values present in

      // src_var_of_4x4_sub_blocks are invalid. Reuse the same if it is readily

      // available with valid values.

      if (src_var < 0) {

        src_var = av1_calc_normalized_variance(

            vf, x->plane[0].src.buf + i * x->plane[0].src.stride + j,

            x->plane[0].src.stride, is_hbd);

        block_4x4_var_info->var = src_var;

        log_src_var = log1p(src_var / 16.0);

        block_4x4_var_info->log_var = log_src_var;

      } else {

        // When source variance is already calculated and available for

        // retrieval, check if log(1 + variance) is also available. If it is

        // available, then retrieve from buffer. Else, calculate the same and

        // store to the buffer.

        if (log_src_var < 0) {

          log_src_var = log1p(src_var / 16.0);

          block_4x4_var_info->log_var = log_src_var;

        }

      }

      *avg_log_src_variance += log_src_var;

      const int recon_var = av1_calc_normalized_variance(

          vf, xd->plane[0].dst.buf + i * xd->plane[0].dst.stride + j,

          xd->plane[0].dst.stride, is_hbd);

      *avg_log_recon_variance += log1p(recon_var / 16.0);

    }

  }

  const int blocks = (bw * bh) / 16;

  *avg_log_src_variance /= (double)blocks;

  *avg_log_recon_variance /= (double)blocks;

}

// Returns a factor to be applied to the RD value based on how well the

// reconstructed block variance matches the source variance.

static double intra_rd_variance_factor(const AV1_COMP *cpi, MACROBLOCK *x,

                                       BLOCK_SIZE bs) {

  double threshold = INTRA_RD_VAR_THRESH(cpi->oxcf.speed);

  // For non-positive threshold values, the comparison of source and

  // reconstructed variances with threshold evaluates to false

  // (src_var < threshold/rec_var < threshold) as these metrics are greater than

  // than 0. Hence further calculations are skipped.

  if (threshold <= 0) return 1.0;

  double variance_rd_factor = 1.0;

  double avg_log_src_variance = 0.0;

  double avg_log_recon_variance = 0.0;

  double var_diff = 0.0;

  compute_avg_log_variance(cpi, x, bs, &avg_log_src_variance,

                           &avg_log_recon_variance);

  // Dont allow 0 to prevent / 0 below.

  avg_log_src_variance += 0.000001;

  avg_log_recon_variance += 0.000001;

  if (avg_log_src_variance >= avg_log_recon_variance) {

    var_diff = (avg_log_src_variance - avg_log_recon_variance);

    if ((var_diff > 0.5) && (avg_log_recon_variance < threshold)) {

      variance_rd_factor = 1.0 + ((var_diff * 2) / avg_log_src_variance);

    }

  } else {

    var_diff = (avg_log_recon_variance - avg_log_src_variance);

    if ((var_diff > 0.5) && (avg_log_src_variance < threshold)) {

      variance_rd_factor = 1.0 + (var_diff / (2 * avg_log_src_variance));

    }

  }

  // Limit adjustment;

  variance_rd_factor = AOMMIN(3.0, variance_rd_factor);

  return variance_rd_factor;

}

/*!\endcond */

/*!\brief Search for the best filter_intra mode when coding intra frame.

 *

 * \ingroup intra_mode_search

 * \callergraph

 * This function loops through all filter_intra modes to find the best one.

 *

 * \return Returns 1 if a new filter_intra mode is selected; 0 otherwise.

 */

static int rd_pick_filter_intra_sby(const AV1_COMP *const cpi, MACROBLOCK *x,

                                    int *rate, int *rate_tokenonly,

                                    int64_t *distortion, uint8_t *skippable,

                                    BLOCK_SIZE bsize, int mode_cost,

                                    PREDICTION_MODE best_mode_so_far,

                                    int64_t *best_rd, int64_t *best_model_rd,

                                    PICK_MODE_CONTEXT *ctx) {

  // Skip the evaluation of filter intra modes.

  if (cpi->sf.intra_sf.prune_filter_intra_level == 2) return 0;

  MACROBLOCKD *const xd = &x->e_mbd;

  MB_MODE_INFO *mbmi = xd->mi[0];

  int filter_intra_selected_flag = 0;

  FILTER_INTRA_MODE mode;

  TX_SIZE best_tx_size = TX_8X8;

  FILTER_INTRA_MODE_INFO filter_intra_mode_info;

  uint8_t best_tx_type_map[MAX_MIB_SIZE * MAX_MIB_SIZE];

  av1_zero(filter_intra_mode_info);

  mbmi->filter_intra_mode_info.use_filter_intra = 1;

  mbmi->mode = DC_PRED;

  mbmi->palette_mode_info.palette_size[0] = 0;

  // Skip the evaluation of filter-intra if cached MB_MODE_INFO does not have

  // filter-intra as winner.

  if (x->use_mb_mode_cache &&

      !x->mb_mode_cache->filter_intra_mode_info.use_filter_intra)

    return 0;

  for (mode = 0; mode < FILTER_INTRA_MODES; ++mode) {

    int64_t this_rd;

    RD_STATS tokenonly_rd_stats;

    mbmi->filter_intra_mode_info.filter_intra_mode = mode;

    if ((cpi->sf.intra_sf.prune_filter_intra_level == 1) &&

        !(av1_derived_filter_intra_mode_used_flag[best_mode_so_far] &

          (1 << mode)))

      continue;

    // Skip the evaluation of modes that do not match with the winner mode in

    // x->mb_mode_cache.

    if (x->use_mb_mode_cache &&

        mode != x->mb_mode_cache->filter_intra_mode_info.filter_intra_mode)

      continue;

    if (model_intra_yrd_and_prune(cpi, x, bsize, best_model_rd)) {

      continue;

    }

    av1_pick_uniform_tx_size_type_yrd(cpi, x, &tokenonly_rd_stats, bsize,

                                      *best_rd);

    if (tokenonly_rd_stats.rate == INT_MAX) continue;

    const int this_rate =

        tokenonly_rd_stats.rate +

        intra_mode_info_cost_y(cpi, x, mbmi, bsize, mode_cost, 0);

    this_rd = RDCOST(x->rdmult, this_rate, tokenonly_rd_stats.dist);

    // Visual quality adjustment based on recon vs source variance.

    if ((cpi->oxcf.mode == ALLINTRA) && (this_rd != INT64_MAX)) {

      this_rd = (int64_t)(this_rd * intra_rd_variance_factor(cpi, x, bsize));

    }

    // Collect mode stats for multiwinner mode processing

    const int txfm_search_done = 1;

    store_winner_mode_stats(

        &cpi->common, x, mbmi, NULL, NULL, NULL, 0, NULL, bsize, this_rd,

        cpi->sf.winner_mode_sf.multi_winner_mode_type, txfm_search_done);

    if (this_rd < *best_rd) {

      *best_rd = this_rd;

      best_tx_size = mbmi->tx_size;

      filter_intra_mode_info = mbmi->filter_intra_mode_info;

      av1_copy_array(best_tx_type_map, xd->tx_type_map, ctx->num_4x4_blk);

      memcpy(ctx->blk_skip, x->txfm_search_info.blk_skip,

             sizeof(x->txfm_search_info.blk_skip[0]) * ctx->num_4x4_blk);

      *rate = this_rate;

      *rate_tokenonly = tokenonly_rd_stats.rate;

      *distortion = tokenonly_rd_stats.dist;

      *skippable = tokenonly_rd_stats.skip_txfm;

      filter_intra_selected_flag = 1;

    }

  }

  if (filter_intra_selected_flag) {

    mbmi->mode = DC_PRED;

    mbmi->tx_size = best_tx_size;

    mbmi->filter_intra_mode_info = filter_intra_mode_info;

    av1_copy_array(ctx->tx_type_map, best_tx_type_map, ctx->num_4x4_blk);

    return 1;

  } else {

    return 0;

  }

}

void av1_count_colors(const uint8_t *src, int stride, int rows, int cols,

                      int *val_count, int *num_colors) {

  const int max_pix_val = 1 << 8;

  memset(val_count, 0, max_pix_val * sizeof(val_count[0]));

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

    for (int c = 0; c < cols; ++c) {

      const int this_val = src[r * stride + c];

      assert(this_val < max_pix_val);

      ++val_count[this_val];

    }

  }

  int n = 0;

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

    if (val_count[i]) ++n;

  }

  *num_colors = n;

}

void av1_count_colors_highbd(const uint8_t *src8, int stride, int rows,

                             int cols, int bit_depth, int *val_count,

                             int *bin_val_count, int *num_color_bins,

                             int *num_colors) {

  assert(bit_depth <= 12);

  const int max_bin_val = 1 << 8;

  const int max_pix_val = 1 << bit_depth;

  const uint16_t *src = CONVERT_TO_SHORTPTR(src8);

  memset(bin_val_count, 0, max_bin_val * sizeof(val_count[0]));

  if (val_count != NULL)

    memset(val_count, 0, max_pix_val * sizeof(val_count[0]));

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

    for (int c = 0; c < cols; ++c) {

      /*

       * Down-convert the pixels to 8-bit domain before counting.

       * This provides consistency of behavior for palette search

       * between lbd and hbd encodes. This down-converted pixels

       * are only used for calculating the threshold (n).

       */

      const int this_val = ((src[r * stride + c]) >> (bit_depth - 8));

      assert(this_val < max_bin_val);

      if (this_val >= max_bin_val) continue;

      ++bin_val_count[this_val];

      if (val_count != NULL) ++val_count[(src[r * stride + c])];

    }

  }

  int n = 0;

  // Count the colors based on 8-bit domain used to gate the palette path

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

    if (bin_val_count[i]) ++n;

  }

  *num_color_bins = n;

  // Count the actual hbd colors used to create top_colors

  n = 0;

  if (val_count != NULL) {

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

      if (val_count[i]) ++n;

    }

    *num_colors = n;

  }

}

void set_y_mode_and_delta_angle(const int mode_idx, MB_MODE_INFO *const mbmi,

                                int reorder_delta_angle_eval) {

  if (mode_idx < INTRA_MODE_END) {

    mbmi->mode = intra_rd_search_mode_order[mode_idx];

    mbmi->angle_delta[PLANE_TYPE_Y] = 0;

  } else {

    mbmi->mode = (mode_idx - INTRA_MODE_END) / (MAX_ANGLE_DELTA * 2) + V_PRED;

    int delta_angle_eval_idx =

        (mode_idx - INTRA_MODE_END) % (MAX_ANGLE_DELTA * 2);

    if (reorder_delta_angle_eval) {

      mbmi->angle_delta[PLANE_TYPE_Y] =

          luma_delta_angles_order[delta_angle_eval_idx];

    } else {

      mbmi->angle_delta[PLANE_TYPE_Y] =

          (delta_angle_eval_idx < 3 ? (delta_angle_eval_idx - 3)

                                    : (delta_angle_eval_idx - 2));

    }

  }

}

static inline int get_model_rd_index_for_pruning(

    const MACROBLOCK *const x,

    const INTRA_MODE_SPEED_FEATURES *const intra_sf) {

  const int top_intra_model_count_allowed =

      intra_sf->top_intra_model_count_allowed;

  if (!intra_sf->adapt_top_model_rd_count_using_neighbors)

    return top_intra_model_count_allowed - 1;

  const MACROBLOCKD *const xd = &x->e_mbd;

  const PREDICTION_MODE mode = xd->mi[0]->mode;

  int model_rd_index_for_pruning = top_intra_model_count_allowed - 1;

  int is_left_mode_neq_cur_mode = 0, is_above_mode_neq_cur_mode = 0;

  if (xd->left_available)

    is_left_mode_neq_cur_mode = xd->left_mbmi->mode != mode;

  if (xd->up_available)

    is_above_mode_neq_cur_mode = xd->above_mbmi->mode != mode;

  // The pruning of luma intra modes is made more aggressive at lower quantizers

  // and vice versa. The value for model_rd_index_for_pruning is derived as

  // follows.

  // qidx 0 to 127: Reduce the index of a candidate used for comparison only if

  // the current mode does not match either of the available neighboring modes.

  // qidx 128 to 255: Reduce the index of a candidate used for comparison only

  // if the current mode does not match both the available neighboring modes.

  if (x->qindex <= 127) {

    if (is_left_mode_neq_cur_mode || is_above_mode_neq_cur_mode)

      model_rd_index_for_pruning = AOMMAX(model_rd_index_for_pruning - 1, 0);

  } else {

    if (is_left_mode_neq_cur_mode && is_above_mode_neq_cur_mode)

      model_rd_index_for_pruning = AOMMAX(model_rd_index_for_pruning - 1, 0);

  }

  return model_rd_index_for_pruning;

}

/*! \brief prune luma intra mode based on the model rd.

 * \param[in]    this_model_rd              model rd for current mode.

 * \param[in]    best_model_rd              Best model RD seen for this block so

 *                                          far.

 * \param[in]    top_intra_model_rd         Top intra model RD seen for this

 *                                          block so far.

 * \param[in]    max_model_cnt_allowed      The maximum number of top intra

 *                                          model RD allowed.

 * \param[in]    model_rd_index_for_pruning Index of the candidate used for

 *                                          pruning based on model rd.

 */

static int prune_intra_y_mode(int64_t this_model_rd, int64_t *best_model_rd,

                              int64_t top_intra_model_rd[],

                              int max_model_cnt_allowed,

                              int model_rd_index_for_pruning) {

  const double thresh_best = 1.50;

  const double thresh_top = 1.00;

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

    if (this_model_rd < top_intra_model_rd[i]) {

      for (int j = max_model_cnt_allowed - 1; j > i; j--) {

        top_intra_model_rd[j] = top_intra_model_rd[j - 1];

      }

      top_intra_model_rd[i] = this_model_rd;

      break;

    }

  }

  if (top_intra_model_rd[model_rd_index_for_pruning] != INT64_MAX &&

      this_model_rd >

          thresh_top * top_intra_model_rd[model_rd_index_for_pruning])

    return 1;

  if (this_model_rd != INT64_MAX &&

      this_model_rd > thresh_best * (*best_model_rd))

    return 1;

  if (this_model_rd < *best_model_rd) *best_model_rd = this_model_rd;

  return 0;

}

// Run RD calculation with given chroma intra prediction angle., and return

// the RD cost. Update the best mode info. if the RD cost is the best so far.

static int64_t pick_intra_angle_routine_sbuv(

    const AV1_COMP *const cpi, MACROBLOCK *x, BLOCK_SIZE bsize,

    int rate_overhead, int64_t best_rd_in, int *rate, RD_STATS *rd_stats,

    int *best_angle_delta, int64_t *best_rd) {

  MB_MODE_INFO *mbmi = x->e_mbd.mi[0];

  assert(!is_inter_block(mbmi));

  int this_rate;

  int64_t this_rd;

  RD_STATS tokenonly_rd_stats;

  if (!av1_txfm_uvrd(cpi, x, &tokenonly_rd_stats, bsize, best_rd_in))

    return INT64_MAX;

  this_rate = tokenonly_rd_stats.rate +

              intra_mode_info_cost_uv(cpi, x, mbmi, bsize, rate_overhead);

  this_rd = RDCOST(x->rdmult, this_rate, tokenonly_rd_stats.dist);

  if (this_rd < *best_rd) {

    *best_rd = this_rd;

    *best_angle_delta = mbmi->angle_delta[PLANE_TYPE_UV];

    *rate = this_rate;

    rd_stats->rate = tokenonly_rd_stats.rate;

    rd_stats->dist = tokenonly_rd_stats.dist;

    rd_stats->skip_txfm = tokenonly_rd_stats.skip_txfm;

  }

  return this_rd;

}

/*!\brief Search for the best angle delta for chroma prediction

 *

 * \ingroup intra_mode_search

 * \callergraph

 * Given a chroma directional intra prediction mode, this function will try to

 * estimate the best delta_angle.

 *

 * \returns Return if there is a new mode with smaller rdcost than best_rd.

 */

static int rd_pick_intra_angle_sbuv(const AV1_COMP *const cpi, MACROBLOCK *x,

                                    BLOCK_SIZE bsize, int rate_overhead,

                                    int64_t best_rd, int *rate,

                                    RD_STATS *rd_stats) {

  MACROBLOCKD *const xd = &x->e_mbd;

  MB_MODE_INFO *mbmi = xd->mi[0];

  assert(!is_inter_block(mbmi));

  int i, angle_delta, best_angle_delta = 0;

  int64_t this_rd, best_rd_in, rd_cost[2 * (MAX_ANGLE_DELTA + 2)];

  rd_stats->rate = INT_MAX;

  rd_stats->skip_txfm = 0;

  rd_stats->dist = INT64_MAX;

  for (i = 0; i < 2 * (MAX_ANGLE_DELTA + 2); ++i) rd_cost[i] = INT64_MAX;

  for (angle_delta = 0; angle_delta <= MAX_ANGLE_DELTA; angle_delta += 2) {

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

      best_rd_in = (best_rd == INT64_MAX)

                       ? INT64_MAX

                       : (best_rd + (best_rd >> ((angle_delta == 0) ? 3 : 5)));

      mbmi->angle_delta[PLANE_TYPE_UV] = (1 - 2 * i) * angle_delta;

      this_rd = pick_intra_angle_routine_sbuv(cpi, x, bsize, rate_overhead,

                                              best_rd_in, rate, rd_stats,

                                              &best_angle_delta, &best_rd);

      rd_cost[2 * angle_delta + i] = this_rd;

      if (angle_delta == 0) {

        if (this_rd == INT64_MAX) return 0;

        rd_cost[1] = this_rd;

        break;

      }

    }

  }

  assert(best_rd != INT64_MAX);

  for (angle_delta = 1; angle_delta <= MAX_ANGLE_DELTA; angle_delta += 2) {

    int64_t rd_thresh;

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

      int skip_search = 0;

      rd_thresh = best_rd + (best_rd >> 5);

      if (rd_cost[2 * (angle_delta + 1) + i] > rd_thresh &&

          rd_cost[2 * (angle_delta - 1) + i] > rd_thresh)

        skip_search = 1;

      if (!skip_search) {

        mbmi->angle_delta[PLANE_TYPE_UV] = (1 - 2 * i) * angle_delta;

        pick_intra_angle_routine_sbuv(cpi, x, bsize, rate_overhead, best_rd,

                                      rate, rd_stats, &best_angle_delta,

                                      &best_rd);

      }

    }

  }

  mbmi->angle_delta[PLANE_TYPE_UV] = best_angle_delta;

  return rd_stats->rate != INT_MAX;

}

#define PLANE_SIGN_TO_JOINT_SIGN(plane, a, b) \

  (plane == CFL_PRED_U ? a * CFL_SIGNS + b - 1 : b * CFL_SIGNS + a - 1)

static void cfl_idx_to_sign_and_alpha(int cfl_idx, CFL_SIGN_TYPE *cfl_sign,

                                      int *cfl_alpha) {

  int cfl_linear_idx = cfl_idx - CFL_INDEX_ZERO;

  if (cfl_linear_idx == 0) {

    *cfl_sign = CFL_SIGN_ZERO;

    *cfl_alpha = 0;

  } else {

    *cfl_sign = cfl_linear_idx > 0 ? CFL_SIGN_POS : CFL_SIGN_NEG;

    *cfl_alpha = abs(cfl_linear_idx) - 1;

  }

}

static int64_t cfl_compute_rd(const AV1_COMP *const cpi, MACROBLOCK *x,

                              int plane, TX_SIZE tx_size,

                              BLOCK_SIZE plane_bsize, int cfl_idx,

                              int fast_mode, RD_STATS *rd_stats) {

  assert(IMPLIES(fast_mode, rd_stats == NULL));

  const AV1_COMMON *const cm = &cpi->common;

  MACROBLOCKD *const xd = &x->e_mbd;

  MB_MODE_INFO *const mbmi = xd->mi[0];

  int cfl_plane = get_cfl_pred_type(plane);

  CFL_SIGN_TYPE cfl_sign;

  int cfl_alpha;

  cfl_idx_to_sign_and_alpha(cfl_idx, &cfl_sign, &cfl_alpha);

  // We conly build CFL for a given plane, the other plane's sign is dummy

  int dummy_sign = CFL_SIGN_NEG;

  const int8_t orig_cfl_alpha_signs = mbmi->cfl_alpha_signs;

  const uint8_t orig_cfl_alpha_idx = mbmi->cfl_alpha_idx;

  mbmi->cfl_alpha_signs =

      PLANE_SIGN_TO_JOINT_SIGN(cfl_plane, cfl_sign, dummy_sign);

  mbmi->cfl_alpha_idx = (cfl_alpha << CFL_ALPHABET_SIZE_LOG2) + cfl_alpha;

  int64_t cfl_cost;

  if (fast_mode) {

    cfl_cost =

        intra_model_rd(cm, x, plane, plane_bsize, tx_size, /*use_hadamard=*/0);

  } else {

    av1_init_rd_stats(rd_stats);

    av1_txfm_rd_in_plane(x, cpi, rd_stats, INT64_MAX, 0, plane, plane_bsize,

                         tx_size, FTXS_NONE, 0);

    av1_rd_cost_update(x->rdmult, rd_stats);

    cfl_cost = rd_stats->rdcost;

  }

  mbmi->cfl_alpha_signs = orig_cfl_alpha_signs;

  mbmi->cfl_alpha_idx = orig_cfl_alpha_idx;

  return cfl_cost;

}

static const int cfl_dir_ls[2] = { 1, -1 };

// If cfl_search_range is CFL_MAGS_SIZE, return zero. Otherwise return the index

// of the best alpha found using intra_model_rd().

static int cfl_pick_plane_parameter(const AV1_COMP *const cpi, MACROBLOCK *x,

                                    int plane, TX_SIZE tx_size,

                                    int cfl_search_range) {

  assert(cfl_search_range >= 1 && cfl_search_range <= CFL_MAGS_SIZE);

  if (cfl_search_range == CFL_MAGS_SIZE) return CFL_INDEX_ZERO;

  const MACROBLOCKD *const xd = &x->e_mbd;

  const MB_MODE_INFO *const mbmi = xd->mi[0];

  assert(mbmi->uv_mode == UV_CFL_PRED);

  const MACROBLOCKD_PLANE *pd = &xd->plane[plane];

  const BLOCK_SIZE plane_bsize =

      get_plane_block_size(mbmi->bsize, pd->subsampling_x, pd->subsampling_y);

  int est_best_cfl_idx = CFL_INDEX_ZERO;

  int fast_mode = 1;

  int start_cfl_idx = CFL_INDEX_ZERO;

  int64_t best_cfl_cost = cfl_compute_rd(cpi, x, plane, tx_size, plane_bsize,

                                         start_cfl_idx, fast_mode, NULL);

  for (int si = 0; si < 2; ++si) {

    const int dir = cfl_dir_ls[si];

    for (int i = 1; i < CFL_MAGS_SIZE; ++i) {

      int cfl_idx = start_cfl_idx + dir * i;

      if (cfl_idx < 0 || cfl_idx >= CFL_MAGS_SIZE) break;

      int64_t cfl_cost = cfl_compute_rd(cpi, x, plane, tx_size, plane_bsize,

                                        cfl_idx, fast_mode, NULL);

      if (cfl_cost < best_cfl_cost) {

        best_cfl_cost = cfl_cost;

        est_best_cfl_idx = cfl_idx;

      } else {

        break;

      }

    }

  }

  return est_best_cfl_idx;

}

static inline void set_invalid_cfl_parameters(uint8_t *best_cfl_alpha_idx,

                                              int8_t *best_cfl_alpha_signs) {

  *best_cfl_alpha_idx = 0;

  *best_cfl_alpha_signs = 0;

}

static void cfl_pick_plane_rd(const AV1_COMP *const cpi, MACROBLOCK *x,

                              int plane, TX_SIZE tx_size, int cfl_search_range,

                              RD_STATS cfl_rd_arr[CFL_MAGS_SIZE],

                              int est_best_cfl_idx) {

  assert(cfl_search_range >= 1 && cfl_search_range <= CFL_MAGS_SIZE);

  const MACROBLOCKD *const xd = &x->e_mbd;

  const MB_MODE_INFO *const mbmi = xd->mi[0];

  assert(mbmi->uv_mode == UV_CFL_PRED);

  const MACROBLOCKD_PLANE *pd = &xd->plane[plane];

  const BLOCK_SIZE plane_bsize =

      get_plane_block_size(mbmi->bsize, pd->subsampling_x, pd->subsampling_y);

  for (int cfl_idx = 0; cfl_idx < CFL_MAGS_SIZE; ++cfl_idx) {

    av1_invalid_rd_stats(&cfl_rd_arr[cfl_idx]);

  }

  int fast_mode = 0;

  int start_cfl_idx = est_best_cfl_idx;

  cfl_compute_rd(cpi, x, plane, tx_size, plane_bsize, start_cfl_idx, fast_mode,

                 &cfl_rd_arr[start_cfl_idx]);

  if (cfl_search_range == 1) return;

  for (int si = 0; si < 2; ++si) {

    const int dir = cfl_dir_ls[si];

    for (int i = 1; i < cfl_search_range; ++i) {

      int cfl_idx = start_cfl_idx + dir * i;

      if (cfl_idx < 0 || cfl_idx >= CFL_MAGS_SIZE) break;

      cfl_compute_rd(cpi, x, plane, tx_size, plane_bsize, cfl_idx, fast_mode,

                     &cfl_rd_arr[cfl_idx]);

    }

  }

}

/*!\brief Pick the optimal parameters for Chroma to Luma (CFL) component

 *

 * \ingroup intra_mode_search

 * \callergraph

 *

 * This function will use DCT_DCT followed by computing SATD (sum of absolute

 * transformed differences) to estimate the RD score and find the best possible

 * CFL parameter.

 *

 * Then the function will apply a full RD search near the best possible CFL

 * parameter to find the best actual CFL parameter.

 *

 * Side effect:

 * We use ths buffers in x->plane[] and xd->plane[] as throw-away buffers for RD

 * search.

 *

 * \param[in] x                Encoder prediction block structure.

 * \param[in] cpi              Top-level encoder instance structure.

 * \param[in] tx_size          Transform size.

 * \param[in] ref_best_rd      Reference best RD.

 * \param[in] cfl_search_range The search range of full RD search near the

 *                             estimated best CFL parameter.

 *

 * \param[out]   best_rd_stats          RD stats of the best CFL parameter

 * \param[out]   best_cfl_alpha_idx     Best CFL alpha index

 * \param[out]   best_cfl_alpha_signs   Best CFL joint signs

 *

 */

static int cfl_rd_pick_alpha(MACROBLOCK *const x, const AV1_COMP *const cpi,

                             TX_SIZE tx_size, int64_t ref_best_rd,

                             int cfl_search_range, RD_STATS *best_rd_stats,

                             uint8_t *best_cfl_alpha_idx,

                             int8_t *best_cfl_alpha_signs) {

  assert(cfl_search_range >= 1 && cfl_search_range <= CFL_MAGS_SIZE);

  const ModeCosts *mode_costs = &x->mode_costs;

  RD_STATS cfl_rd_arr_u[CFL_MAGS_SIZE];

  RD_STATS cfl_rd_arr_v[CFL_MAGS_SIZE];

  MACROBLOCKD *const xd = &x->e_mbd;

  int est_best_cfl_idx_u, est_best_cfl_idx_v;

  av1_invalid_rd_stats(best_rd_stats);

  // As the dc pred data is same for different values of alpha, enable the

  // caching of dc pred data. Call clear_cfl_dc_pred_cache_flags() before

  // returning to avoid the unintentional usage of cached dc pred data.

  xd->cfl.use_dc_pred_cache = true;

  // Evaluate alpha parameter of each chroma plane.

  est_best_cfl_idx_u =

      cfl_pick_plane_parameter(cpi, x, 1, tx_size, cfl_search_range);

  est_best_cfl_idx_v =

      cfl_pick_plane_parameter(cpi, x, 2, tx_size, cfl_search_range);

  if (cfl_search_range == 1) {

    // For cfl_search_range=1, further refinement of alpha is not enabled. Hence

    // CfL index=0 for both the chroma planes implies invalid CfL mode.

    if (est_best_cfl_idx_u == CFL_INDEX_ZERO &&

        est_best_cfl_idx_v == CFL_INDEX_ZERO) {

      set_invalid_cfl_parameters(best_cfl_alpha_idx, best_cfl_alpha_signs);

      clear_cfl_dc_pred_cache_flags(&xd->cfl);

      return 0;

    }

    int cfl_alpha_u, cfl_alpha_v;

    CFL_SIGN_TYPE cfl_sign_u, cfl_sign_v;

    const MB_MODE_INFO *mbmi = xd->mi[0];

    cfl_idx_to_sign_and_alpha(est_best_cfl_idx_u, &cfl_sign_u, &cfl_alpha_u);

    cfl_idx_to_sign_and_alpha(est_best_cfl_idx_v, &cfl_sign_v, &cfl_alpha_v);

    const int joint_sign = cfl_sign_u * CFL_SIGNS + cfl_sign_v - 1;

    // Compute alpha and mode signaling rate.

    const int rate_overhead =

        mode_costs->cfl_cost[joint_sign][CFL_PRED_U][cfl_alpha_u] +

        mode_costs->cfl_cost[joint_sign][CFL_PRED_V][cfl_alpha_v] +

        mode_costs

            ->intra_uv_mode_cost[is_cfl_allowed(xd)][mbmi->mode][UV_CFL_PRED];

    // Skip the CfL mode evaluation if the RD cost derived using the rate needed

    // to signal the CfL mode and alpha parameter exceeds the ref_best_rd.

    if (RDCOST(x->rdmult, rate_overhead, 0) > ref_best_rd) {

      set_invalid_cfl_parameters(best_cfl_alpha_idx, best_cfl_alpha_signs);

      clear_cfl_dc_pred_cache_flags(&xd->cfl);

      return 0;

    }

  }

  // Compute the rd cost of each chroma plane using the alpha parameters which

  // were already evaluated.

  cfl_pick_plane_rd(cpi, x, 1, tx_size, cfl_search_range, cfl_rd_arr_u,

                    est_best_cfl_idx_u);

  cfl_pick_plane_rd(cpi, x, 2, tx_size, cfl_search_range, cfl_rd_arr_v,

                    est_best_cfl_idx_v);

  clear_cfl_dc_pred_cache_flags(&xd->cfl);

  for (int ui = 0; ui < CFL_MAGS_SIZE; ++ui) {

    if (cfl_rd_arr_u[ui].rate == INT_MAX) continue;

    int cfl_alpha_u;

    CFL_SIGN_TYPE cfl_sign_u;

    cfl_idx_to_sign_and_alpha(ui, &cfl_sign_u, &cfl_alpha_u);

    for (int vi = 0; vi < CFL_MAGS_SIZE; ++vi) {

      if (cfl_rd_arr_v[vi].rate == INT_MAX) continue;

      int cfl_alpha_v;

      CFL_SIGN_TYPE cfl_sign_v;

      cfl_idx_to_sign_and_alpha(vi, &cfl_sign_v, &cfl_alpha_v);

      // cfl_sign_u == CFL_SIGN_ZERO && cfl_sign_v == CFL_SIGN_ZERO is not a

      // valid parameter for CFL

      if (cfl_sign_u == CFL_SIGN_ZERO && cfl_sign_v == CFL_SIGN_ZERO) continue;

      int joint_sign = cfl_sign_u * CFL_SIGNS + cfl_sign_v - 1;

      RD_STATS rd_stats = cfl_rd_arr_u[ui];

      av1_merge_rd_stats(&rd_stats, &cfl_rd_arr_v[vi]);

      if (rd_stats.rate != INT_MAX) {

        rd_stats.rate +=

            mode_costs->cfl_cost[joint_sign][CFL_PRED_U][cfl_alpha_u];

        rd_stats.rate +=

            mode_costs->cfl_cost[joint_sign][CFL_PRED_V][cfl_alpha_v];

      }

      av1_rd_cost_update(x->rdmult, &rd_stats);

      if (rd_stats.rdcost < best_rd_stats->rdcost) {

        *best_rd_stats = rd_stats;

        *best_cfl_alpha_idx =

            (cfl_alpha_u << CFL_ALPHABET_SIZE_LOG2) + cfl_alpha_v;

        *best_cfl_alpha_signs = joint_sign;

      }

    }

  }

  if (best_rd_stats->rdcost >= ref_best_rd) {

    av1_invalid_rd_stats(best_rd_stats);

    // Set invalid CFL parameters here since the rdcost is not better than

    // ref_best_rd.

    set_invalid_cfl_parameters(best_cfl_alpha_idx, best_cfl_alpha_signs);

    return 0;

  }

  return 1;

}

static bool should_prune_chroma_smooth_pred_based_on_source_variance(

    const AV1_COMP *cpi, const MACROBLOCK *x, BLOCK_SIZE bsize) {

  if (!cpi->sf.intra_sf.prune_smooth_intra_mode_for_chroma) return false;

  // If the source variance of both chroma planes is less than 20 (empirically

  // derived), prune UV_SMOOTH_PRED.

  for (int i = AOM_PLANE_U; i < av1_num_planes(&cpi->common); i++) {

    const unsigned int variance = av1_get_perpixel_variance_facade(

        cpi, &x->e_mbd, &x->plane[i].src, bsize, i);

    if (variance >= 20) return false;

  }

  return true;

}

int64_t av1_rd_pick_intra_sbuv_mode(const AV1_COMP *const cpi, MACROBLOCK *x,

                                    int *rate, int *rate_tokenonly,

                                    int64_t *distortion, uint8_t *skippable,

                                    BLOCK_SIZE bsize, TX_SIZE max_tx_size) {

  const AV1_COMMON *const cm = &cpi->common;

  MACROBLOCKD *xd = &x->e_mbd;

  MB_MODE_INFO *mbmi = xd->mi[0];

  assert(!is_inter_block(mbmi));

  MB_MODE_INFO best_mbmi = *mbmi;

  int64_t best_rd = INT64_MAX, this_rd;

  const ModeCosts *mode_costs = &x->mode_costs;

  const IntraModeCfg *const intra_mode_cfg = &cpi->oxcf.intra_mode_cfg;

  init_sbuv_mode(mbmi);

  // Return if the current block does not correspond to a chroma block.

  if (!xd->is_chroma_ref) {

    *rate = 0;

    *rate_tokenonly = 0;

    *distortion = 0;

    *skippable = 1;

    return INT64_MAX;

  }

  // Only store reconstructed luma when there's chroma RDO. When there's no

  // chroma RDO, the reconstructed luma will be stored in encode_superblock().

  xd->cfl.store_y = store_cfl_required_rdo(cm, x);

  if (xd->cfl.store_y) {

    // Restore reconstructed luma values.

    // TODO(chiyotsai@google.com): right now we are re-computing the txfm in

    // this function everytime we search through uv modes. There is some

    // potential speed up here if we cache the result to avoid redundant

    // computation.

    av1_encode_intra_block_plane(cpi, x, mbmi->bsize, AOM_PLANE_Y,

                                 DRY_RUN_NORMAL,

                                 cpi->optimize_seg_arr[mbmi->segment_id]);

    xd->cfl.store_y = 0;

  }

  IntraModeSearchState intra_search_state;

  init_intra_mode_search_state(&intra_search_state);

  const CFL_ALLOWED_TYPE cfl_allowed = is_cfl_allowed(xd);

  // Search through all non-palette modes.

  for (int mode_idx = 0; mode_idx < UV_INTRA_MODES; ++mode_idx) {

    int this_rate;

    RD_STATS tokenonly_rd_stats;

    UV_PREDICTION_MODE uv_mode = uv_rd_search_mode_order[mode_idx];

    // Skip the current mode evaluation if the RD cost derived using the mode

    // signaling rate exceeds the best_rd so far.

    const int mode_rate =

        mode_costs->intra_uv_mode_cost[cfl_allowed][mbmi->mode][uv_mode];

    if (RDCOST(x->rdmult, mode_rate, 0) > best_rd) continue;

    PREDICTION_MODE intra_mode = get_uv_mode(uv_mode);

    const int is_diagonal_mode = av1_is_diagonal_mode(intra_mode);

    const int is_directional_mode = av1_is_directional_mode(intra_mode);

    if (is_diagonal_mode && !cpi->oxcf.intra_mode_cfg.enable_diagonal_intra)